1700 words-Tuition, Jobs, or Housing: What is keeping Millennials at Home?

Instructions1 xFinalPaperInstructions xarticle1-economics Article2-Harvard

III. Guidelines for the Paper. Due in both hard-copy and electronic form in class on December 4

A. Basic Reference: The basic reference for all aspects of the paper, especially acknowledging sources, will be either A Student’s Guide for Writing College Papers or A Manual for Writers of Term Papers, Theses and Dissertations. Both are by Kate Turabian and published by the University of Chicago Press. Any comparable conventional manual will also be acceptable.

B. Recommended Structure of the 3F03 Final Paper (Length of paper: 8-9 pages). Section 1. Economic questions and policy relevance of your papers as a group (1-2 pages).

THERE IS NO NEED TO LIST THE PAPERS AT THE BEGINNING.

Section 2. First Paper (approximately 2-3 pages)

Economic questions Policy Relevance Data and estimation methods Main Results Policy Implications Internal and External strengths and weaknesses (instructions for Summary above)

Section 3. Second Paper (same topics as for first paper), 2-3 pages

There is no need to repeat if, for example, two papers use the same data source. Just refer to the description in the earlier paper.

Final Section. (1-2 pages) Sum conclusions regarding questions and policy

Compare findings if questions are similar. Which one do you believe more?

Unanswered questions and agenda for future research C. A Few Basic Rules

o Each paper should have a title page and a list of references. Please use the short form for referencing, e.g., (Mulroney 1992 p. 12) and then give the full citation in the List of references. Footnotes and end notes are discouraged.

o The body should contain approximately 10 pages with double spacing, 1″ margins and 10 or 12 characters per inch (an 11 or 12 point font).

o Number your pages and make sure all pages are legible. o Points will be deducted for poor spelling and grammar. Use a good spell check and grammar check. o Please avoid a large number of quotations. o A partial list of practices to avoid

???? Longparagraphs.(Paragraphsareusedtoindicateachangeintopic.) ???? Contractions(won’t,can’t,etc.) ???? Frequentuseofthefirstperson(Iwilldiscuss,Iwillshow,etc.) ???? Informalor“chatty”styleofwriting

D. Papers for other courses. A paper may be done for this course and another course but only when the two faculty members involved have cleared the topic in advance. In general, more will be expected of a joint paper.

E. Acknowledging Sources. Please be VERY careful to acknowledge all sources which you consulted during the preparation of your paper. You should reference not just published work, but also unpublished papers, including those of other students, and previous papers of your own. Each source should be acknowledged by author and date in the text. THE PAPER WILL BE SUBMITTED TO TURNITIN.COM. This tool provides a very effective report on the extent to which your text matches the text in journal articles, working papers and other student papers. NOT ONLY OUTRIGHT PLAGIARISM WILL BE PENALIZED; POINTS WILL BE DEDUCTED FOR FAILURE TO ACKNOWLEDGE DIRECT QUOTES TAKEN FROM OTHER SOURCES.

F. Consulting the Instructor. Please be sure to consult me early on if you are having difficulty with getting going on your paper. It is my job to help you get started, not just to grade the paper. Also see me for help with references at any time.

G. Key Goal for the Paper. Your main job is to demonstrate your understanding of what constitutes strong and weak economic research, that is, to evaluate the quality of the methods rather than your personal agreement (or not) with any given policy position. I want your paper to be a balanced review of evidence by a judge rather than the one-sided summation to the jury by a lawyer. Keep that analogy in mind. Your goal is to write a good (balanced) judicial review and to avoid writing a good (one-sided) advocate’s brief.

H. Writing Help. McMaster students now have free access to an on-line program to help with writing. The user copies and pastes text into the program and receives notification of errors and suggestions for corrections. This

program is called Grammarly (formerly Sentence Works). Registration is at https://ed.grammarly.com/register/signup/features/?edu=true. I encourage you to try this and give me feedback.

The Writing Clinic is in the Centre for Student Development (basement of the University Centre) to help students with writing problems. You can have a one-on-one session with a Writing Clinic Peer by making an appointment. See http://csd.mcmaster.ca/academic/. Note that your assignments will be graded on both economic content and writing style.

This is what you need to do:

General notes:

First 3 pages:

Section 1:

In order to complete 3 pages, I want you to write the follow about the articles:

For each paper, what kind of data set was used and the methods used to conclude their research and findings. This means what experiments did they do, what happened and what were the results?? Also include the null-hypothesis from every paper that is written.

Each time you talk about this, make sure you say which paper you are talking about and include in text citations!

Policy Implications

For each paper, talk about the policy implications of the paper. Write about each paper separately. This means different paragraph for each paper when you are talking about the papers.

NOTE: Write about each paper separately, when you are talking about the data, methods used, policy. I want everything about the papers written separately. For example: 1.5 pages about paper 1 and 1.5 pages about paper two!

Final Section (1-2 pages) 

Sum conclusions regarding questions and policy

Compare findings if questions are similar. Which one do you believe more?

Unanswered questions and agenda for future research

General notes:

When you are referring to the papers, please do in text citations.

If you do not know what is, look it up! I do not want the references in the bottom.

Compare the articles together and say which one you are talking about! Use in text citation and say which one you agree with more and why

III. Guidelines for the Paper. Due in both hard-copy and electronic form in class on December

B. Recommended Structure of the 3F03 Final Paper (Length of paper: 8-9 pages). Section 1. Economic questions and policy relevance of your papers as a group (1-2 pages).

THERE IS NO NEED TO LIST THE PAPERS AT THE BEGINNING.

Section 2. First Paper (approximately 2-3 pages)

Economic questions Policy Relevance Data and estimation methods Main Results Policy Implications Internal and External strengths and weaknesses (instructions for Summary above)

Section 3. Second Paper (same topics as for first paper), 2-3 pages

There is no need to repeat if, for example, two papers use the same data source. Just refer to the description in the earlier paper.

Final Section. (1-2 pages) Sum conclusions regarding questions and policy

Compare findings if questions are similar. Which one do you believe more?

Unanswered questions and agenda for future research C. A Few Basic Rules

o The body should contain approximately 10 pages with double spacing, 1″ margins and 10 or 12 characters per inch (an 11 or 12 point font).

o Number your pages and make sure all pages are legible. o Points will be deducted for poor spelling and grammar. Use a good spell check and grammar check. o Please avoid a large number of quotations. o A partial list of practices to avoid

􏰀Longparagraphs.(Paragraphsareusedtoindicateachangeintopic.) 􏰀Contractions(won’t,can’t,etc.) 􏰀Frequentuseofthefirstperson(Iwilldiscuss,Iwillshow,etc.) 􏰀Informalor“chatty”styleofwriting

program is called Grammarly (formerly Sentence Works). Registration is at https://ed.grammarly.com/register/signup/features/?edu=true. I encourage you to try this and give me feedback.

This paper presents preliminary findings and is being distributed to economists

and other interested readers solely to stimulate discussion and elicit comments.

The views expressed in this paper are those of the authors and do not necessarily

reflect the position of the Federal Reserve Bank of New York or the Federal

Reserve System. Any errors or omissions are the responsibility of the authors.

Federal Reserve Bank of New York

Staff Reports

Tuition, Jobs, or Housing:

What’s Keeping Millennials at Home?

Zachary Bleemer

Meta Brown

Donghoon Lee

Wilbert van der Klaauw

Staff Report No. 700

November 2014

Revised July 2017

Tuition, Jobs, or Housing: What’s Keeping Millennials at Home?

Zachary Bleemer, Meta Brown, Donghoon Lee, and Wilbert van der Klaauw

Federal Reserve Bank of New York Staff Reports, no. 700

November 2014; revised July 2017

JEL classification: D14, E24, R21

Abstract

This paper documents marked changes in young Americans’ residence choices over the past

fifteen years, with recent cohorts decreasingly living with roommates and instead lingering much

longer in parents’ households. To understand the sources and implications of this decline in

independence, we estimate the contributions of local economic circumstances to the decision to

live with parents or independently. Transition models, local aggregates, and state-cohort tuition

patterns are used to address the likely presence of individual- and neighborhood-level unobserved

heterogeneity. In regions where many students are exposed to college costs, we find that

increased tuition is associated with more coresidence with parents and less living with

roommates. Where fewer youth confront college tuition, however, local job market conditions are

paramount in shaping the decision of whether to live with parents.

Key words: household formation, mobility, student loans

_________________

Lee and van der Klaauw: Federal Reserve Bank of New York (emails: donghoon.lee@ny.frb.org,

wilbert.vanderklaauw@ny.frb.org). Bleemer: University of California at Berkeley (email:

bleemer@berkeley.edu). Brown: Stony Brook University (email: meta.brown@stonybrook.edu).

This paper was previously distributed under the title “Debt, Jobs, or Housing: What’s Keeping

Millennials at Home?” The authors thank their referees, Andrew Haughwout and Henry

Korytkowski; seminar participants at the Federal Reserve Banks of Boston and New York, the

University of Virginia, and the Urban Institute; and conference participants at the Federal Reserve

Bank of St. Louis, the Family Economics Workshop of the Barcelona GSE 2015 Summer Forum,

the FDIC Consumer Research Symposium, Goldman Sachs’ Millennials and Housing Day, New

York University, and the University of California at Los Angeles’ Ziman Center for Real Estate

for valuable comments. The views expressed in this paper are those of the authors and do not

necessarily reflect the position of the Federal Reserve Bank of New York or the Federal Reserve

System.

Recent cohorts of young Americans have, on average, extended their stays in their

parents’ households. After eventually moving out, their members return home to parents at

higher rates. Climbing U.S. trends toward co-residence with parents are displayed in Duca

(2014) for 1962-2012 and Matsudaira (2016) for 1960-2007. In this paper, we use millions of

credit records from the Equifax-sourced Federal Reserve Bank of New York Consumer Credit

Panel (CCP) to describe young Americans’ residential arrangements from 2004-2015. Among

25-year-olds, we report an 11.4 percentage point increase in living with parents or similar elders

and a 12.8 percentage point decline in living with groups of (similarly-aged) roommates. For

these same years, several researchers document first stability and then a steep decline in young

Americans’ rate of homeownership.1 The dual trends of extended co-residence with parents and

decreasing participation in rental and housing markets may contribute to slowed growth in both

consumption and the housing market, as young people living “at home” delay major purchases

and general entry into economic life.

This paper investigates the residence choices of young people in the Consumer Credit

Panel, and their relationship to evolving local house prices, local employment conditions, and the

cost of college for local students. We document upward trends in aggregate rates of co-residence

with parents and other elders among 25-year-olds that are not only persistent but also wide-

spread, with substantial increases in co-residence with parents in all 48 contiguous states. In

addition, we discuss a range of co-residence measurement concerns and cite outside evidence

suggesting a steep upward trend.2,3

What are the likely consequences of lingering at home for young people’s economic

lives? Relatedly, what consequences might these trends have for the duration of the ongoing U.S.

economic recovery? In order to answer these questions, we must understand the origins of the

decline in independent living among American youth. Recent work on household formation,

such as Dyrda, Kaplan, and Rios-Rull (2012), Duca (2013), and Matsudaira (forthcoming), has

analyzed the link between an observed decline in household formation and changes in

1 Agarwal, Hu, and Huang (2013) , Brown and Caldwell (2013), Brown, Caldwell, and Sutherland (2014), and
Brown et al (2015), the CFPB (2013), and Mezza, Sherlund, and Sommer (2014) report substantial declines in youth
homeownership since 2007.
2 See, for example, Mykyta and Macartney (2011).
3 The link between homeownership and student debt has been examined in the PSID and the NELS88 by Cooper and
Wang (2014), in the SCF by Gicheva and Thompson (2014), in the 1997 cohort of the NLSY by Houle and Berger
(2014), and in the CCP by Bleemer et al. (2017). Kurz and Li (2015) address the link between student debt and auto
purchase.

employment and poverty. Along with Matsudaira, Duca, and Dyrda, Kaplan, and Rios-Rull, our

first candidate explanation for youths’ increasing reliance on parents might be labor market

difficulties. Following Agarwal, Hu, and Huang, we might next suspect that youth residence

choices respond to local house prices. Finally, we and others have studied the unprecedented

U.S. student debt climb that coincides with the trend toward living with parents.4 Figure 1

depicts the enrollment-weighted mean of public and private college and university tuition for

each U.S. state and the District of Columbia from 2004 to 2015. College costs have risen steeply,

with mean tuition across the states rising by $6576, or 76 percent, over the period. If the financial

burden of college is increasingly borne by students and families, we might also expect this to

influence students’ ability to live independently following school. To what extent, then, can we

say that the observed climb in intergenerational co-residence coincides with economic challenges

to youth such as weak job markets, costly housing, or high tuition?

Worth noting is that national labor and housing market trends display pronounced

cyclicality, while co-residence with parents and college costs follow comparatively acyclical

upward trends. Of course, these aggregates mask evolving local relationships among housing

cost, labor markets, and youth residence choices. The fine geographic data, vast sample size, and

long panel of the CCP allow us to observe the residence choices of large numbers of 25-year-

olds at fine geographic levels, and to compare them over many birth cohorts. All of this allows

us to study youth residence choices under a rich variety of economic circumstances. Our hope is

that the resulting understanding of the origins of the decline in independent living yields insights

regarding its relationship to existing policy, its potential consequences for young Americans’

consumption and welfare, and its likely persistence or development in the future. While labor

and housing market origins may suggest a cyclicality in the rate of co-residence with parents,

college cost origins, given longstanding U.S. college tuition trends, may suggest an ongoing

decline in youth independence.

In an approach that builds on Ermisch (1999) and Kaplan (2012), we model the fraction

of young consumers who live with their parents, as well as the flows of young consumers into

and out of parents’ households over time, as a function of patterns in local unemployment, youth

unemployment, house prices, wages, and the enrollment-weighted average college tuition

4 See Brown et al. (2015a) and Dettling and Hsu (2014).

confronting a given cohort in a given state.5 We address endogeneity concerns regarding the

college spending of an individual student or family by estimating in state and county aggregates,

and by providing direct estimates of the relationship between state-cohort tuition levels and

subsequent co-residence. Our aim is to report intent-to-treat estimates of (an approximation of)

the gross tuition amount a student would be charged for college should she attend, given her state

and birth cohort.

Our estimates of the relationship among intergenerational co-residence choices and the

economic conditions under which they are made reveal markedly different decision processes in

regions where college is more and less relevant. In high college graduation rate regions, many

young students are exposed to college costs, and jumps in the tuition facing a college cohort are

met with extended periods of co-residence with parents, or more moving “home” in early

adulthood. For example, a $1000 increase in a state-cohort’s mean college tuition is associated

with a 0.72 percentage point increase in co-residence with parents.6 Given a $6202 increase in

mean tuition for the higher graduation rate states, tuition growth is able to account for 4.5

percentage points of their observed 10.5 percentage point growth in co-residence with parents at

age 25 from 2004 to 2015.

In low college graduation rate regions, fewer students confront college tuition, and local

job market conditions, with particular emphasis on wages, are the dominant economic factor in

young people’s decision to live with parents. An increase of $100 in average weekly wages is

associated with a 3.9 percentage point decline in co-residence with parents among the lower

college graduation rate states.7 What the high and low college graduation regions share in

common is a secular trend toward intergenerational co-residence: in all cases, and with or

without controlling for the levels or progress of economic circumstances, the coefficients we

estimate on indicators for each calendar year reflect steep and approximately monotonic growth

in intergenerational co-residence from 2004 to 2015. This growth is particularly pronounced

among the lower college graduation rate states.

5 Ermisch poses the question in the context of survey data on British youth of the 1990s, who made co-residence
choices under very different economic and social conditions, and for whom college cost was of little relevance.
Kaplan’s study emphasized high frequency job shocks and residential transitions, and also did not address college
costs.
6 This coefficient estimate differs significantly from zero at the five percent level.
7 This coefficient also differs significantly from zero at the five percent level.

Analysis of the geographic sources of the estimated tuition-co-residence relationship

indicates that their positive association arises primarily in the Northeast and Midwest, or,

alternatively, in more urban settings. All of these estimates align with the broader claim that we

observe a meaningful tuition-co-residence association where a greater share of youth attends

college, and hence where a greater share of youth is exposed to college costs.

If young Americans are increasingly living with parents, one might ask what residential

arrangements have fallen out of favor. Our estimates reveal declining secular trends in living

with two or more roommates that approximately offset the climb in living with parents. Rates of

living in couples or alone, controlling for economic factors, are comparatively stable. Estimates

relating state-cohort college costs to residential arrangements show that the afore-mentioned

$1000 hike in tuition for youth in high graduation states is associated not only with a 0.72

percentage point increase in co-residence with parents, but also with a 0.86 percentage point drop

in living with roommates, and little change in living alone or in couples. Hence youth in high

graduation states facing greater college costs appear to be foregoing rooming with peers for the

cost-saving option of living with parents.

Our transition model estimates of the rate at which American 23-to-25-year-olds move

home to parents suggest a protective effect of strengthening local labor markets on the

independence of youth. Here a one percentage point increase in local employment in the youth

location is associated with a 0.2 to 0.3 percentage point decline in the rate at which youth move

home to parents over the two years, and a $100 increase in average weekly wages for the county

over two years is associated with a 0.12 to 0.15 percentage point decline in the rate of moving

home to parents.8 By and large, the economic determinants of moving away from parents, if they

exist, are far less obvious. In contrast to the rate of moving home, we do not find a substantial or

significant association between employment or house prices and the rate at which youth in the

county who were living with parents at 23 achieve independence by 25. The one meaningful

association between economic growth and moving out is this: as wages rise in the parent

location, youth living with parents are considerably less likely to move out.

Section I of the paper discusses the emerging literature on the recent changes in youth

residence, and positions our paper within it. Section II describes the various data sources that we

employ, particularly the CCP, which is comparatively novel. In Section III, we present new

8 The employment and wage coefficient estimates are each significant at the one percent level.

evidence from the CCP on the (rapidly evolving) residential circumstances of 25-year-olds from

2004 to 2015. Section IV places this descriptive analysis of the data, and our subsequent

empirical models of intergenerational co-residence patterns and the decision to move home or

away, in the context of the existing theory of co-residence with parents developed by Kaplan

(2012). Section V lays out an empirical model of the stock of youth living with parents, along

with transition models of the flow to independence for youth living with parents and the flow

back home to parents for youth living independently. Section VI reports estimation results, and

Section VII offers some concluding thoughts.

I. Related literature

The existing literature emphasizes the relationship between employment conditions, or

poverty, and intergenerational co-residence. Earlier work, including Goldscheider and DaVanzo

(1985, 1989), Haurin et al. (1993), and Whittington and Peters (1996) establishes a longstanding

pattern of greater youth co-residence with parents when economic circumstances are poor. More

recently, Card and Lemieux (2000) demonstrate a noteworthy retreat home for Canadian youth in

response to the prolonged Canadian recession of the 1990s. Dyrda, Kaplan, and Rios-Rull (2012)

demonstrate a substantial influence of household formation responses to the business cycle on

the Frisch elasticity of labor supply. Duca (2013) finds a close relationship between 1979-2013

time series on U.S. 18-64-year-olds’ rate of co-residence with parents and U.S. poverty rates.

Two recent papers are particularly relevant to this study. Matsudaira (forthcoming) predates

the analysis here, and considers the influence of local economic conditions on co-residence with

parents in the U.S. over both an earlier and a longer window. He uses cross-sectional Census and

American Community Survey (ACS) data, available at varying intervals from 1960 to 2007, to

examine the relationship between state-level employment and housing conditions and co-

residence with parents over a 47-year period. Unlike our analysis, the paper is able to describe

patterns in the relationship between state economic conditions over a very long period of time,

which offers a better-informed picture of the potential role of social trends in living with parents.

Further, it paints a rich picture of the (considerable) degree of demographic heterogeneity in co-

residence patterns, a type of analysis which our administrative credit bureau data cannot support.

Our study also offers some unique evidence, owing to its panel approach, fine geography, recent

measures, range of residential outcomes, and introduction of college costs. By comparison,

Matsudaira’s data are cross-sectional, which restricts his analysis to coarse geography, for

reasons discussed below. His estimation window stops short of the largest downturn confronted

by U.S. youth in recent memory, and, finally, Matsudaira omits the cost of college.9

Student and other consumer debts, and their contribution to rising co-residence with parents,

are the primary interest of Dettling and Hsu (2014). Their study and ours have developed

concurrently and independently, using the same primary data source. The focus of the two

papers, however, is quite different; one might see them as complementary. Dettling and Hsu are

interested in the role of current individual debt levels and repayment, both in the decision to

move home and in the youth’s ability to recover from an economic shock and eventually regain

independence. As such, their estimation provides new insights into the relationship between debt

struggles—including, importantly, repayment struggles—and youth residence circumstances.

These findings are of clear independent value when compared with the results reported in this

paper (and vice versa, we believe). At the same time, current individual debt levels and

delinquency or default status contain the accumulated history of employment, housing, and other

shocks, and reflect past education choices. As such, estimated effects of, for example, realized

student loan delinquency on the move home or the move away do not isolate student loan effects

from, for example, the effects of recent job market fluctuations. Therefore Dettling and Hsu’s

estimates do not answer the specific question we pose in this paper regarding the separate

contributions of debt, jobs, and housing to 25-year-olds’ delayed independence, despite

providing many novel and policy-relevant insights on debt and co-residence.

II. Data

a. The FRBNY Consumer Credit Panel

The FRBNY Consumer Credit Panel (CCP) is a longitudinal dataset on consumer

liabilities and repayment. It is built from quarterly consumer credit report data collected and

provided by Equifax Inc. Data are collected quarterly since 1999Q1, and the panel is ongoing.10

Sample members have Social Security numbers ending in one of five arbitrarily selected,

randomly assigned pairs of digits. Therefore the sample comprises 5 percent of U.S. individuals

with credit reports (and Social Security numbers). The CCP sample design automatically

9 This is sensible, one might argue, given that the prevalence and balances of student loans were substantially lower
over much of his estimation window.
10 Student debt data are only available in the CCP starting in 2003.

refreshes the panel by including all new reports with Social Security numbers ending in the

above-mentioned digit pairs. Therefore the panel remains representative for any given quarter,

and includes both representative attrition, as the deceased and emigrants leave the sample, and

representative entry of new consumers, as young borrowers and immigrants enter the sample.11

While the sample is representative only of those individuals with Equifax credit reports, the

coverage of credit reports (that is, the share of individuals with at least one type of loan or

account) is fairly complete for American adults. Aggregates extrapolated from the data match

those based on the American Community Survey, Flow of Funds Accounts of the United States,

and SCF well.12 However, because we focus on young people’s co-residence decisions, we

restrict our dataset to 23- and 25-year-olds, which have lower coverage than later ages; coverage

over 2003-2013, the era used in our Section IV-V estimates, ranges between 83.4 and 93.9% for

25-year-olds, increasing from 2003 to 2007 and decreasing from 2007 to 2013 (compared to

estimates from the US Census).13 Nevertheless, we do have some information about individuals

not covered in the CCP; we know how many live in each state (based on Census figures), and we

know that, in nearly all cases, they do not have conventional consumer debt or credit (in which

case they would be covered by Equifax). We use this information to analyze and bound our

estimates below.14

We construct a cohort-level dataset from the CCP by extracting a panel of all individuals

who turn 23 or 25 years old in each year between 2003 and 2013. Because the time-series aspect

of our study drastically increases the number of observations, we only pull a random 1% sample

of the covered U.S. population, instead of the full CCP 5%. There are 546,824 25-year-olds in

the dataset, of whom we have 1.01 million observations.15

11 See Lee and van der Klaauw (2010) for details on the sample design.
12 See Lee and van der Klaauw (2010) and Brown et al. (2015b) for details.
13 We use the 2008 Census population projections as ‘true’ population data from 1999 to 2011 and the 2012 Census
year-age population projections for 2012 and 2013. In each case, these are the most accurate available data on
population size by age, year, and state.
14 Lee and van der Klaauw (2010) extrapolate similar populations of U.S. residents aged 18 and over, overall and by
age groups, using the CCP and the ACS, suggesting that the vast majority of US individuals at younger ages have
credit reports. Jacob and Schneider (2006) find that 10 percent of U.S. adults had no credit reports in 2006, and
Brown et al. (2015b) estimate that 8.33 percent of the (representative) Survey of Consumer Finances (SCF)
households in 2007 include no member with a credit report. They also find a proportion of household heads under
age 35 of 21.7 percent in the 2007 SCF, 20.64 in the 2007Q3 CCP, and 20.70 from Census 2007 projections,
suggesting good representation of younger households in the CCP.
15 Note that the panel data used in constructing some of the variables used in estimation contain roughly 10.1 million
person-year observations.

b. Other data sources

Table 1 summarizes the additional data that we use in our aggregate analysis of parental

co-residence. All financial variables in the paper are measured in 2013 dollars. The first

empirical model of co-residence we estimate below relies on data aggregated to the level of the

state by year by cohort. The descriptive statistics in Table 1 report means across these cells in the

pooled sample of aggregates. These descriptive measures of the pooled aggregates are

noteworthy for their variability. While the average state-year-cohort has a ratio of employed

residents to overall adult population of 57.54 percent, we also observe a standard deviation in

employment among the states of 13.27 percentage points. State-year-cohort means for house

price index, urban population, graduation rate, and tuition also vary widely in the pooled data.

Our next empirical specification estimates the relationship between the rate at which

youth transition away from or home to their parents’ households, and these are estimated using

county-cohort-year aggregates. Columns (2)-(4) provide average two-year changes in each

characteristic across county-cohort-year cells for three groups: all relevant youth, those living

independently, and those living with parents. The most noteworthy differences between the

locations of those living with parents and living independently are the more rapidly improving

labor markets characterizing the counties of the independent youth, and the more rapidly

growing house prices of the parent locations. This various measures taken from these data

remind us of the problems that arise when we measure local economic conditions in parent

neighborhoods for co-residing youth and youth neighborhoods for independent youth.

The annual county-level employment data are drawn from the Bureau of Labor Statistics’

(BLS) Quarterly Census of Employment and Wages (QCEW) program. The unemployment data

are reported on a quarterly basis, and they cover a total of 3,197 counties. In order to measure the

employment-to-population ratio, we also draw annual county-level population data from the US

Census’s Population Estimates.16 We calculate the youth unemployment rate at the state level

using employment data from 18- to 30-year-old individuals in the CPS, aggregated from months

to quarters.17 Average weekly county-level wage data for 3,197 counties are drawn from the

BLS’s QCEW program.

16 Data are from the 1990s Postcensal Estimates and the Vintage 2009 and 2014 estimates.
17 This aggregated sample of the CPS (over all months from 2003 to 2014) includes 3.2 million respondents between
age 18 and 30—19,333 of whom are missing labor force status information—though due to the sampling

House price appreciation values are calculated at the county level using data from the

CoreLogic home price index (HPI). The CoreLogic HPI uses repeat sales transactions to track

changes in sale prices for homes over time, with the January 2000 baseline receiving a value of

100. We aggregate an annual index to avoid seasonal variation. The CoreLogic data cover 1,266

counties (covering 89% of the 25-year-olds observed in our sample) in all 50 states and the

District of Columbia, but fail to cover some parts of rural America.18 In our regression analysis

below, we include an indicator for whether the individual lives in a county populous enough to

be covered by the CoreLogic HPI series. Its coefficient estimate goes largely unreported, but in

each case is available from the authors.19

Several of our estimates require a measurement of the college graduation rate for a given

cohort in a given state. We calculate the total number of graduates using the Integrated

Postsecondary Education Data System (IPEDS), summing over the number of graduates of four-

year and two-year institutions who receive degrees within 150% of the normal completion time

in that state-year. We calculate the average graduation rate as the ratio of the total number of

graduates to the total number of 24-year-olds in the state, as estimated by the US Census. The

mean graduation rate across states over our sample is 34.1 percent.

The analysis below also includes a set of college tuition measures. Ideally, we would like

to track the cost of college in a given location for a given cohort while the cohort is between ages

18 and 22, and then attach it to the cohort going forward as an indication of the cost of college

relevant to the cohort’s decision-making. However, given our objective of estimating outcomes

measured at ages 23 and 25, an age 18 to 22 tuition measure would require us to observe tuition

seven years before the last-observed outcome for each cohort. The IPEDS data on which we rely

for state tuition levels in each year are complete from 2000 forward. The CCP fourth-quarter

outcomes we track are available through the end of 2015. A seven-year look-back for tuition

would reduce our sample of 25-year-olds to the 2007-2015 window. In order to extend the

sample backward into the pre-recession window, we have chosen to shorten the ages at which we

measure tuition to ages 20 to 22. As a result, we estimate with a sample of 23-year-olds observed

methodology of the CPS, some people appear in the dataset twice (in two different quarters). Data are aggregated
using individual weights.
18 In our regression analysis below, we include an indicator variable for whether the individual lives in a county
covered by the CoreLogic HPI series, though we do not report the corresponding estimated coefficients.
19 We also draw the median value of owner-occupied housing units by county in 2000 from the US Census,
estimating county-level median house prices as the product of that and the CoreLogic series. These are employed in
some footnoted analysis of the relationship of intergenerational co-residence to the level of house prices.

from 2003 to 2013, and 25-year-olds from 2005 to 2015. We construct a series of state-cohort

average sticker and net costs of public and private colleges by pulling cost data from IPEDS.20

We define sticker cost as the sum of tuition and fees (excluding room and board) at US colleges

and universities, where net cost is sticker cost minus grant aid. Costs are averaged across

postsecondary institutions by state, sector, and year, where weighting in the averages is based on

each institution’s share of undergraduate enrollment.

Our decision to study sticker, and not net, prices of college in estimation sections V and

VI is guided primarily by one concern. Grant aid in a given year in a given state is shaped by

many factors, among them the financial need of the students. Hence any estimate of the

association between net tuition and youth residence outcomes would conflate true college cost

effects with effects of the economic conditions of college students’ families. Perhaps in part for

this reason, we find that estimating with net tuition leads, in many though not all cases, to large

and precise estimates of tuition-residence relationships. Net tuition results are available from the

authors.

III. Aggregate trends in young consumers’ residence choices

a. Co-residence with parents: measurement and trends

Each observation in the CCP includes the (anonymized) information in an individual’s

credit report at the end of that quarter (e.g. zip code, birth year, total balances of 10 types of

consumer debt, etc.) as well as the information in the credit reports of all members of that

individual’s household, where households are defined by street address (down to an apartment

number).21 These data lead us to define co-residence (with parents) to be the circumstance in

which a young person (here a 23- or 25-year-old) resides at the same street address as at least one

(Equifax-covered) individual who is between 15 and 45 years her senior, without regard to

household head status or the relationship between the household members.22 Data from the

Center for Disease Control and Prevention’s (CDC) National Vital Statistics System show that,

for the 1978 and 1988 birth cohorts (the early and late ends of our sample), almost all mothers

20 IPEDS covers all 7,255 postsecondary schools in the United States, 5,126 of which provide enrollment and tuition
data, accounting for 97.8 percent of enrollment in the dataset.
21 See Avery et al. (2003) for a detailed discussion of the contents, sources, and quality of credit report data.
22 We exclude household members with empty credit files, as those individuals’ addresses may no longer be
accurately recorded by their creditors, or thereby by Equifax itself.

and the vast majority of fathers were between the ages of 15 and 45 at the child’s birth.23

Moreover, we define individuals who live in households of more than 10 people (3.7% of 25-

year-olds and 3.6% of 30-year-olds) as not co-residing, because most situations in which one

would live in such a large household (prison, military, mobile home community) are not such

that the individual is in her parents’ household.24 Note that our definition might overestimate the

aggregate rate of co-residence with parents due to a possible lag between a young person’s

switching his home address and updating his credit report address (as reported by financial

institutions), which might bias the aggregate co-residence rate upwards.25

In order to evaluate the success of our measure of parental co-residence, we use the 2003-

2012 Current Population Surveys to estimate the fraction of individuals who would fall under our

definition of “living with parents” who are actually co-residing with parents (or other older

relatives).26 We find that, in 2010, 92.6 percent of 25-year-olds whom we designate as ‘living

with their parents’ either certainly or most likely co-reside, suggesting that we slightly

overestimate co-residence in our analysis below. First, 84.0 percent live with their parents or

similar elders: most commonly their parents themselves, but also their spouse’s or partner’s

parents, or the parents of a sibling or in-law, their foster parents, their grandparents, or a parent’s

unmarried partner. Another 8.6 percent of 25-year-olds that meet our CCP definition of living

with parents in 2010 most likely co-reside with elder relatives, but the CPS leaves their

designation unclear; they may live with an older sibling, older relatives from outside of the

nuclear family, or with a friend and the friend’s parents. The remaining 7.3 percent of 25-year-

olds in 2010 who meet our CCP criteria for co-residing with parents actually do not co-reside

with parents or elder relatives. 1.8 percent of 25-year-olds who “live with their parents” actually

live with older spouses; other, smaller groups are observed to live at the same address as an

older landlord, an older roommate, or an older roomer. Roughly one percent of cases are either

miscodes or exceedingly complex scenarios.

23 The birth rate for women aged 45-49 in 1978 and 1988 was 0.2 live births per 1000 women. The birth rate for
women aged 10-14 in 1978 (1988) was 1.2 (1.3) per 1000. The birth rate for men aged 45-49 in 1978 (1988) was
larger, at 5.8 (7.1) per 1000. However, this remains quite small relative to the men’s age 25-29 birth rate of 120.0
(111.1) per 1000.
24 We also assume that individuals whose address is listed as a post office box do not co-reside (4% of 25-year-olds,
and 5% of 30-year-olds).
25 Transition model estimates of the probability that independent youth move home in Sections V-VI are less
susceptible to lagged address updating concerns.
26 Our total sample size is 207,928 25-year-olds and 210,711 30-year-olds across the ten years of our analysis. We
use sample weights in order that our analysis is nationally representative.

Importantly, our CPS analysis shows that the rate at which we overestimate 25-year-old

co-residence is unchanging over time. The fraction of 25-year-olds that we categorize as “living

with their parents” who co-reside with a parent or elder relative was bounded between 91.6 and

92.6 percent from 2003 to 2012, and we find no evidence of either a linear or quadratic time

trend at the 10% level of significance. This provides evidence that our trend analysis below, in

both the stocks and flows, is unbiased despite slightly overestimating the fraction of young

people who live with their parents (or in similar living arrangements) at any fixed point in time.

Figure 2 depicts the proportion of U.S. 25-year-olds living with “parents” in the CCP

from 2004-2015.2728 For 25-year-old CCP sample members, we observe an increase in the rate of

co-residence with parents or similar elders from 33.5 percent in 2004 to 44.9 percent in 2015.29

Note that this pattern is free of life-cycle effects, as we measure co-residence with parents for the

cross-section of CCP sample members who are 25 years old in each year. This substantial growth

in living with parents is approximately monotonic over the period, and proceeds at a steady pace.

Overall, the rate of co-residence with parents observed in the CCP grows by 11.4 percentage

points for 25-year-olds over our 2004 to 2015 window.

Figure 3 extends these results by examining the increased prevalence of parental co-

residence at the state level.30 We find that parental co-residence among 25-year-olds increased in

all 48 contiguous states in the decade between 2003 and 2013 (though it slightly decreased in

Alaska), with a median increase of 13.8 percentage points. Heterogeneity in parental co-

residence is quite large across states, with state-level co-residence rates for 25-year-olds ranging

from 30 percent to over 50 percent in 2012-2013. States in the center of the country (Rocky

Mountain and Great Plains states) experienced the least growth in parental co-residence, while

27 In the CCP, we observe individuals’ birth years, but not their birth months. The median individual born in a year
turns 25 around July 1st 25 years later. In order to capture the average characteristics of 25-year-olds in a year, then,
we use the observations of those born 25 years earlier from the first quarter of the following year, allowing for a six
month lag in order to measure characteristics, on average, in the middle of the year in which the individual is 25, and
a one-quarter lag from the median time at which those individuals would be 25.5 years old to account for delays in
Equifax data updating, in which loans typically first appear in the data about one quarter later than the origination
date.
28 Importantly, intergenerational co-residence is a viable means of responding to labor or housing market shocks, or
general financial strain, mainly for the subset of families living close enough to each other for the move not to be
exceedingly disruptive. We would prefer to estimate in this population, but are unable to identify parent locations in
families in which youth remain independent. We thank a referee for this observation.
29 From this point we adopt the phrase “living with parents” to describe youth living with parents or with one of the
variety of responsible elders captured by our co-residence measure.
30 This analysis is enabled by the massive size of the CCP data set; our analysis includes at least 166 25-year-olds in
each year-state presented in Figure 3, with a median of 1,282 individuals per state-year.

states in the Northeast and West Coast experienced the sharpest increases, some by more than 20

percentage points between 2003 and 2013. Overall, a striking change appears to have occurred

since 2004 in the living arrangements of young consumers.31

Of course, others have documented a large and ongoing change in intergenerational co-

residence in the U.S., either over earlier periods of time or for very recent years and in parallel to

this study. Their findings using long-standing surveys provide us with an opportunity to validate

our measures of intergenerational co-residence in the comparatively novel, entirely

administrative Consumer Credit Panel. Briefly: our estimates of the national rate of

intergenerational co-residence are quite close to rates reported by Matsudaira (2016) using the

2000 Census, and by Paciorek (2014) and by Dettling and Hsu (2014, 2015) using CPS data for

some or all of 2000-2013. In the appendix to this paper, we give a detailed description of the co-

residence rate estimates in Matsudaira, Paciorek, and Dettling and Hsu, and compare them to our

own co-residence trend estimates.

In sum, we observe a steady growth in co-residence with parents among U.S. youth.

While the level of co-residence rates may be sensitive to measurement choices, the levels and

trends we obtain are similar enough to the findings of other recent researchers using alternative

methods and established survey, rather than administrative, sources to suggest that our CCP

measures are informative. All sources and methods discussed in this paper point to two empirical

facts: co-residence with parents was common in 2004, and it is substantially more common

today.

b. Trends in other living arrangements

Given general agreement that young Americans are staying home with parents at an

increasing rate, what living arrangements are they casting aside? Popular speculation suggests

declining rates of first marriage among young people in the wake of the recession. After the

release of the 2009 American Community Survey, Mather and Lavery (2010) noted a recession-

31 This trend could be determined in part by social or demographic phenomena, rather than economic pressures.
However, while the number of Americans aged 45-64 increased by 24 percent from 2002 to 2012 (according to the
U.S. Department of Health and Human Services’ Administration on Aging), the lifetime number of children per
woman remained near two and, if anything, was very slightly increasing from 1970 to 2010 (Population Reference
Bureau 2012). We thank a referee for this observation. It is unclear, then, the extent to which changing
demographics on their own can be expected to generate large changes in the rate of co-residence with parents.
Nevertheless, in the interest of accounting for possible social and demographic changes, we allow for a flexible time
trend as we model the stock of co-residence below.

era decline in the share of young people who had ever been married. Shortly after, Wolfers

(2010) countered that this data artifact represented not a meaningful decline in stable

relationships, but an ongoing increase in the age at first marriage in the U.S., coupled with an

increase in cohabitation during the recession, which may have been motivated by a desire to cut

living expenses. One relevant question for the current study, then, may be whether young

Americans are choosing extended adolescence at home with parents in place of independent

adulthood and marriage.

We categorize individuals who are not co-residing with parents into three types. An

individual is defined as living alone if he is the only (Equifax-covered adult) resident at his street

address. We then divide the remaining individuals into those who live with only one other person

and those who live with more than one other person, excluding households with more than 10

people and individuals whose report lists a post office box address.32

Figure 4 shows CCP trends from 2004 to 2015 in the rates at which 25-year-olds appear

alone, with parents, with one adult of similar age, and with two or more adults of similar age.33

The latter category we interpret as living with roommates.

The growth we observe in co-residence with parents appears to come at the cost of fewer

young people living with roommates. We calculate a decline in the rate at which 25-year-olds

live in groups of roommates from 32.5 percent in 2004 to 19.7 percent in 2015. Meanwhile, the

rates of living in couples and living alone are comparatively stable. Hence the evidence in

estimation Table 4, below, will not particularly fall in support of a claim that American youth

prolonging their stays with parents are postponing traditional milestones of adulthood. They

may, however, be giving up independent years of living with groups of friends for the cost-

saving benefit of years at home with parents.

c. Trends in tuition

32 Our CCP data do not allow us to measure the rates at which CCP sample members are marrying before

and after the recession. They do not even allow us to measure cohabiting relationships, whether or not they involve
marriage. What we can do, however, is look at trends in the rate at which young Americans co-reside with one other
adult of a similar age. The benefit of this approach is that it includes marriage along with both opposite sex and
same sex cohabitation, yielding a broader picture of trends in co-residing relationships over the period. The obvious
drawback, however, is that it includes roommate pairs whose relationships are platonic. Our analysis of CPS
household characteristics suggests that residing with one adult of similar age is a reliable predictor of romantic
cohabitation. (These results are available from the authors.) Interpretation of trends in living with a single adult
roommate of comparable age should, however, bear this inclusion in mind.
33 By similar age, we mean 14 or fewer years older, or any amount younger, than the 25-year-old file holder. This
cutoff is chosen to create mutually exclusive and exhaustive living arrangement categories.

College costs in the U.S. have been rising steadily for decades. As described previously,

Figure 1 depicts the enrollment-weighted public and private college and university mean tuition

and fees for each state in every year from 2001 to 2013. Each dot represents a state-year. Each

color of dot represents a single state. The figure demonstrates a climb in state tuitions: in our

estimation sample of state-cohort-years, we observe a mean tuition growth of 75.9 percent

between 2004 and 2015, from $8659 to $15,235. Figure 1 further demonstrates a meaningful

expansion in the cross-state dispersion of states’ mean tuitions from early to late in the

estimation window. And, finally, tracking states’ color-coded dots from year to year generates an

understanding of the intermittent nature of jumps in real college tuitions, with the year-to-year

tuition-setting decisions of state boards of regents leading several states to leap-frog each other in

the cross-state tuition ranking each year.

Our empirical approach in Sections V and VI relies on these unpredictable jumps in real

tuition within states for informative variation in the college costs facing individual students.

Further, our approach identifies tuition effects based on within-state dynamics while also

removing the overall positive trend in the national mean of tuition, as well as its nonlinear

movements from year to year. We discuss our identifying assumptions, and states’ tuition-

setting, in more detail in Subsection b of the Appendix.

As a first pass at local analysis of the relationship between college costs and

intergenerational co-residence, Figure 4 presents suggestive evidence of an association between

student debt and parental co-residence in a simple state-level scatter plot that relates the 2008-

2013 change in the rate of parental co-residence among 25-year-olds in a state to its 2008-2013

change in student debt per graduate.34 The regression line in this simple scatter plot reflects a

positive 2.9 percentage point increase in co-residence with a $10,000 increase in student debt per

graduate.

IV. Job market, house price, and tuition effects on intergenerational co-residence from a

theoretical perspective

We rely on Kaplan (2012) for a theoretical description of the dynamic game played

between a parent and child over the choice of separate or shared residence and the child’s labor

34 The chart looks qualitatively similar when constructed from 2003-2013. We isolate the post-recession window as
it is characterized by particularly active changes in states’ public college tuitions.

market activities. The child chooses consumption and saving, labor supply (or, where relevant,

job search), and whether to live independently or with his parent. The altruistic parent, in turn,

allocates a fixed income stream among private consumption, public household consumption, and

a financial transfer to her child.

In Kaplan, the manners in which the child benefits from co-residence with the parent are

the following: (i) a child who co-resides with his parent pays no rent or mortgage, and (ii) when

the parent and child co-reside, they each enjoy the public household good. The cost to the child

of co-residence enters as a preference shifter. Hence the benefits of co-residence appear through

the child’s consumption, and decline along with the marginal utility of consumption, while the

cost of co-residence remains fixed as consumption grows. It is this mechanism that generates a

tendency toward independence as the earnings and assets of the child rise.

Comparative statics for co-residence arising from the model for the relationships at issue

in this paper are fairly straightforward. Kaplan describes them as follows on pages 472-473:

“Youths are more likely to live away from home when earnings, assets, or the value of

independence is higher. However, the probability of living away from home is ambiguous with

respect to parental income. On the one hand, higher parental income generates higher parental

transfers and hence a lower earnings/assets threshold for the youth to live away. On the other

hand, higher parental income means higher consumption in the parental home, making living at

home a more attractive option for the youth.” These predictions for the relationship between

child earnings and co-residence with a parent, and between parent income (or assets) and co-

residence with a child, are helpful in interpreting our estimates of the association between local

labor market characteristics and co-residence below.35

Further, as housing costs in the model are paid in any case by the parent but by the child

only in the event that he child lives independently, the model easily predicts that, all else equal, a

35 Kaplan (2012) estimates the relationship between state-level job market conditions and the share of youth living
with their parents in early adulthood. Our empirical analysis may be differentiated from Kaplan’s own (earlier and
influential) estimates for its coverage of youth choices from 2011 to 2015, in addition to the 2000s, and for its
addition of college tuition to the set of factors that may shape co-residence. Further, Kaplan estimates co-residence
transition responses to individual youths’ realized job losses using the NLSY-97, but must cope with the potential
endogeneity of job loss to moving home. Owing to data that are unusual in that they permit panel tracking of the fine
geographic locations of millions of consumers, we are able to estimate the response of co-residence transitions to
arguably exogenous shocks in local labor and housing market conditions. Of course, our analysis lacks Kaplan’s
careful treatment of the insurance role of the option of moving home to parents, and his structural estimation of the
high-frequency residential dynamics of the family as the child weathers positive and negative job market shocks.

higher cost of housing may induce the family to co-reside. We relate this prediction to our house

price estimates below.

While the child’s current assets enter the Kaplan model, the model does not address

human capital or past educational investments or costs. One simple way of adapting the

framework for our purposes is to assume that the influence of past college tuition appears in the

child’s current assets and wages. To fix ideas, consider an extension of the model in which,

before entering the labor market, the child makes the simple binary decision to attend and

complete college, at a price equal to the prevailing college tuition, or not to attend college.

Following schooling the child enters the Kaplan model as written. Consider the co-residence

effect of a tuition increase under these assumptions. For an increase small enough that the child’s

college entry choice is unchanged, the effect of the tuition increase on the post schooling

circumstances is simply a reduction in the child’s assets. For a tuition increase large enough to

change the child’s college entry decision, the effect is both an increase in assets, as the child

avoids paying tuition, and any decrease in wages that results from reduced post-schooling human

capital.

What are the predictions of this modified Kaplan model for the relationship between

tuition and the child’s post-schooling decision to live with parents or independently? As written,

the direction of the effect is ambiguous. However, Bleemer et al. (2017) precisely estimate a

small and insignificant response of college enrollment and graduation, and of years of schooling,

to college tuition for a population comparable to the one we study in this paper. If the demand

for college is truly price inelastic, then, in the context of our modified Kaplan model, the effect

of a tuition increase will simply be to decrease the assets of a child who chooses college.

Moreover, the tuition increase will have no effect on a child who does not choose college. Hence

our simplistic modification of the Kaplan (2012) model predicts a negative effect of past college

tuition on present co-residence with parents for youth who attended college, and no effect of past

tuition on co-residence with parents for youth who did not attend college. Guided by this

prediction, we estimate the tuition-co-residence relationship in groups with more and less

exposure to college tuition in Section VI.

V. Empirical model

a. Stock of young people living with parents

Next, the fine geographic data and long panel of the CCP allow us to exploit time

variation in local economic conditions and student debt reliance to learn far more about the

contributions of jobs, housing costs, and the cost of college at the local level to the decisive

aggregate trend toward parents, and away from economic independence, that we observe for

recent cohorts of young adults. This section presents three empirical models of parental co-

residence. First, we describe a lagged stock model explaining the co-residence decisions of 23-

and 25-year-olds as a function of local unemployment, youth unemployment, house prices, and

the enrollment-weighted average college tuition that applies to the state-cohort pair in question.

This approach provides an informative description of the times and places in which parental co-

residence is and is not common. Further, heterogeneity analysis allows us to examine the factors

associated with co-residence across the country, and for more and less highly educated, and more

and less urban, areas.

We estimate a state-level model of the share of young residents who are living with parents

as a function of local socioeconomic conditions. In anticipation of the flow model to come, we

consider individuals at two ages, 23 and 25. Define cstY as the share of cohort c, state s, time t

youth who co-reside with parents. We model the share of co-resident youth as a function of the

conditions in the state one year earlier, as well as state fixed effects to control for unobserved

differences in culture and policy that do not vary over time. 36 We thus estimate the simple fixed

effects model:

, (1)

where cstX represents a vector of cohort c, state s, period t characteristics, the levels of which

may influence the residence choices of the youth of state s at t+1. This vector includes state-level

QCEW wages and employment to population ratios, state-level youth unemployment based on

our calculations in the CPS, and state-level CoreLogic home price indices. The vector csZ

represents characteristics of cohort c and state s that do not vary over time, which in this case are

the enrollment-weighted average public and private college and university tuition prevailing in

36 Hence the lagged regressors are observed when the estimation sample youth are 22 and 24.

state s when cohort c was between the ages of 20 and 22, as described in Section II above.37

Realized individual educational spending may contain confounding individual (observed and

unobserved) characteristics.38 We address this concern by estimating using a measure of the cost

of college for the state-cohort that is comparatively free of contributions from individual student

ability, diligence, and other factors that shape realized educational attainment. Further, we

include a vector of state fixed effects, denoted s , and time fixed effects, denoted .39

Idiosyncratic error cst is clustered at the state level.

In implementing empirical model (1), we weight each observation by the age-25 population

of the state. Though estimating in state aggregates without the population weight changes some

of the many stock model estimates described below, it is generally true that the qualitative results

are typically similar in weighted and unweighted specifications. Regarding the estimated

coefficients on average tuition, one broad observation we have made in estimating a variety of

weighted and unweighted specifications is the following: Some of the largest tuition swings over

our estimation period belong to the smallest states. This may be one reason that the estimated

tuition coefficients tend to be considerably more pronounced under the unweighted approach.

Throughout Section III, we report population-weighted estimates. Unweighted estimates are

available from the authors.

State-cohort-year average tuition is included in expression (1) as an exogenous measure of

the cost of college for the population whose residential outcome we are estimating. In subsection

b of the appendix to this paper, we describe the identifying assumption we make regarding the

37 The assumption underlying this approach is that the cost of college when a student reaches college age affects
decisions at that point and far into her adulthood. However, the current cost of college in the state is likely to have
little influence on the current residence choices of a 25-year-old who has long since left school.
38 Though we estimate the relationship between college costs and post-schooling co-residence with parents, much of
the relevant literature relates student debt to post-schooling economic circumstances. Gicheva and Thompson (2014)
discuss a student debt endogeneity concern in a related context. Heterogeneity in family generosity tends to bias the
student debt coefficient in a co-residence regression downward, as generous families both impose less student debt
and tolerate more co-residence. Lochner, Stinebrickner, and Sulemanoglu (2013) demonstrate a strong positive
relationship between family support and student debt repayment in recent data on Canada’s student loan system. On
the other hand, individual-level student debt is closely tied to the student’s level of educational investment. Lochner
and Monge-Naranjo (2012) model the relationship between the nature of the schooling investment and the credit
extended to students, with implications for the individual-level association between student debt and post-college
labor market success that are of obvious relevance here.
39 One concern the estimation confronts is the possibility of a non-economic, social trend in the acceptability of
intergenerational co-residence that may, in part, drive the growth in co-residence. However, the 2014 wave of the
Federal Reserve Board of Governors’ Survey of Household Economics and Decision-making (SHED) shows that
the 64 percent of 2014 SHED families living in “doubled-up” households do so primarily for financial reasons. Only
one in seven are doing so, at least in part, for the purpose of caregiving (Board of Governors 2015).

exogeneity of state-cohort tuition shocks, and we marshal supporting evidence for this

assumption.

We choose to estimate the stock relationship between local economic conditions and co-

residence choices at a high level of geographic aggregation, the state, for a variety of reasons.

Most important to us is the fact that aggregate analysis minimizes any endogeneity that might

arise from young people’s mobility. If there are systematic economic differences between the

regions where young people live independently and the regions where their parents live, then the

economic characteristics we measure in the young person’s observed location may be

endogenously determined by her co-residence decision. Assume, for example, that housing

prices are higher in parents’ neighborhoods than in children’s neighborhoods, in keeping with

typical life-cycle patterns of consumption in the U.S. Then the problem with the location of

measurement generates a spurious positive relationship between local house prices and living

with parents. By aggregating to the state level, we average across smaller geographic areas (like

zip codes and counties), abstracting away from most mobility concerns.40

b. Flow home to parents from independent living

Nevertheless, one would certainly prefer to exploit the extensive variation in economic

conditions that occurs below the state level in order to generate the most informative picture

possible of youth residence choices. With a detailed panel on the repeated location choices of

millions of early twentysomethings, in this study we are able to push the analysis to a finer

geographic level.

To do so, we model the flows of children into and out of parents’ households. We separate

our baseline sample into youth who live independently in the initial period and youth who live

with parents. This allows us to estimate the effect of more finely measured local economic

conditions on co-residence transitions for samples in which the measure of local conditions is

uniform: we estimate the effect of (changes in) local economic conditions in the parents’ location

on the rate at which dependent youth move out, and, separately, the effect of local economic

40 Most moves to and from parent households occur within a state. According to Molloy et al. (2014), while 19.1
percent of CPS 20-24 year-olds moved between counties over the course of a year, based on pooled data for 2000-
2012, only 3.3 percent of the 20-24 year-olds crossed state lines. Note further that the cross-state move rate declined
over the period. Brown, Grigsby, van der Klaauw, Wen, and Zafar (2015) find that, among CCP individuals
observed at age 18, only 12.82 percent had moved across state lines seven years later. Matsudaira (forthcoming)
employs a similar state-level strategy, presumably owing to similar measurement concerns.

conditions in the independent youth location on the rate at which independent youth move home.

This approach also allows us to ask whether the effects of local economic conditions on whether

a child moves away from home differ from the effects of those same conditions on whether a

child moves back home.

This solution is appealing for economic circumstances that evolve over time and are

relevant to youth residence choices in an ongoing manner. It is less helpful for the question of the

relationship between the cost of college and later co-residence with parents for two reasons.

First, for most young consumers, college costs are borne at only one point in the life-cycle.

Tuition increases that develop after a youth leaves school are, by and large, irrelevant to the

student. Second, tuition averages in our data vary primarily at the level of the state-cohort, as

most college students attend public colleges and universities, and public college and university

tuitions tend to move in response to a shared state legislative and budgeting process. Hence the

CCP’s panel dimension, and our transition models, may illuminate the relationship between

(more) local economic conditions and co-residence with parents or independence and yet add

little to our understanding of the relationship between college costs and co-residence. For the

latter, we rely on stock model estimates.

Since we model two-year flows of parental co-residence between the ages of 23 and 25, we

no longer lag the geographic characteristics by a year in identifying their effect on parental co-

residence. Instead, in most instances, we estimate the dependence of the rate of moving home or

away on the change in conditions over the two-year estimation window in the youth’s initial

location. However, college tuition in the youth’s location from age 20 to 22 is a time-fixed

characteristic. Hence we adopt two approaches to the estimation: in one specification, we

estimate models of the dependence of transitions into and out of co-residence on the changes in

local conditions over time, leaving out the time-fixed factor of interest, college cost. In addition,

we report estimates based on an alternative specification, in which we allow co-residence

transitions to depend also on the time-fixed tuition level relevant to the state-cohort, permitting

these characteristics to influence the transition probability through their level at t rather than

through their (null) flow from t to t + 1.41

41 An alternative specification of our model including level measurements of all covariates is available from the
authors. Given that the level of co-residence at time t, and hence the size and nature of the co-residing population, is
shaped by past tuition, the influence of past tuition on co-residence transitions from t to t+1 is unclear. Our Section

Consider first the decision to move home to parents. From this point, we estimate

intergenerational co-residence decisions aggregated at the county level.42 Maintaining the

definitions above, we estimate a model of the share of independent youth in county l (for

“location”) who move home between the ages of 23 and 25, 1
H

ltY  , in a sample of CCP youth who

lived independently at age 23, which we denote as time t.43

, (2)

Here superscript H denotes factors influencing the probability of moving “home”. Time-varying

local regressors include county-level average wages, employment to population ratio, and

house price index. As mentioned above, we estimate the flow models with and without time-

fixed tuition, . In addition, in these flow equations, our baseline includes a linear national

time trend, which is a constant when differenced. Further, we allow a vector of state-level fixed

effects in the probability of moving home, ( )
H
s l , which may be interpreted as state-level time

trends in co-residence. Importantly, in the flows home, location (county) l is defined as the

child’s location away from home at time t, and all local characteristics at t and t + 1 are

measured for location l.

Finally, though the modified Kaplan model described in Section IV does not predict a time

dependence of co-residence transitions, apart from a response to the evolution of wages,

employment, and housing costs, social developments may shift the level of the preference for co-

residence over time, generating movements in the national flows home or away. Therefore we

estimate under three separate approaches to the dependence of co-residence growth on the

calendar year. First, we estimate under the assumption that 0. This approach adds nothing

IV modified Kaplan model, for example, would generate an ambiguous prediction for the direction of this
relationship.
42 Note that expression (2), below, does not push the data to the individual level. All of the regressors we consider
are measured in local aggregates. While these local aggregates will have substantial predictive power for locally
aggregated cross-sectional co-residence measures and co-residence transitions, without relevant individual-level
exogenous covariates, which for this problem are limited in general and especially limited in the CCP, we have little
hope of identifying which residents of a local area will move. We find that the magnitude and significance of labor
market, housing, and student debt coefficients estimated based on the analogous individual-level co-residence
transition models are very similar to those of the coefficients estimated using aggregate expression (2). The
shortcoming of the individual-level analysis is evident in its modest R-squared values, as we identify the local
moving rate rather reliably, but have little information with which to predict exactly who moves.
43 The elapsed time from t to t + 1 is two years.

to the above-mentioned state-level trends in co-residence with parents. Second, we allow a linear

national time trend in the growth in co-residence, so that . Third, we estimate the

model adding a full set of year-specific fixed effects, so that . In this final

specification, in addition to the above state trends, we allow a separate intercept for the rate of

growth of parental co-residence in each year. This last specification might seem to involve

excessive flexibility in the time dependence of the transition path. Though the first and second

approaches are far easier to interpret in the context of the model discussed in Section IV, we

include it in the interest of thoroughness.

c. Flow away from parents to independent living

We estimate a similar model for the share of county youth living with parents who move out

between periods t and t + 1. The expression for the state fixed effects model estimated, using

county-level aggregates is

, (3)

where all arguments are defined analogously to those in expression (2). In this case, all location

characteristics are measured for location l, the parent’s location in period t. Superscript A denotes

factors influencing the probability of moving “away”.44

As in the case of moving in with parents, we estimate expression (3) first including only the

growth regressors, as well as the specified state and time effects. We then report estimates that

add time-fixed tuition averages relevant to the state-cohort youth while college aged. In doing so,

we note that the predictions of the modified Kaplan model discussed in Section IV for the

direction of the relationship between college costs and the rate of moving away from parents

among the resulting group of co-resident youth some years after college are ambiguous. As in the

transition home case above, we estimate the transition to independence model using three

specifications of time dependence. We estimate first assuming 0, then ,

and, finally, .

Standard endogeneity concerns deriving from observable and unobservable individual and

local characteristics that are fixed over the two-year window are accounted for by the transition

approach we take to estimation. Obvious examples include child ability, parent generosity, and

44 Owing to the unobservability of locations not chosen, what we will not be able to explore is the dependence of the
youth’s decision to move home on the characteristics of the parent’s location, and the dependence of the youth’s
decision to move out on the characteristics of the youth’s preferred independent location.

persistent regional characteristics. Some remaining endogeneity concerns arise from the

association between youth mobility and local house price and employment aggregates. By and

large, they work against the main results described below. They are discussed in the appendix.

V. Results

a. Stock of young people living with parents

Table 2 reports the coefficient estimates for the stock parental co-residence model in

expression (1) for 23- and 25-year-olds. Our baseline specification is shown in column (1),

which includes each of the covariates listed in section Va as well as state dummies to control for

unobserved permanent cross-state differences in culture and policy and year dummies to account

for time-varying aggregate conditions, such as credit market conditions.45 We find, as expected,

that geographic areas with higher average wages are characterized by substantially lower rates of

intergenerational co-residence: a $100 greater mean weekly wage observed for the state and year

in question is associated with a 2.1 percentage point decline in the share of 23- and 25-year-olds

living with parents, and this coefficient is significant at the one percent level.46 Despite this

substantial wage result, we estimate little association between either the overall employment to

population share or the youth unemployment rate and the rate of co-residence in the state for the

year. A once percentage point increase in employment is associated with a 0.03 percentage point

decline in co-residence, and a one percentage point increase in youth unemployment is

associated with a 0.06 percentage point increase in co-residence. The direction of each

coefficient estimate matches our expectations, but neither is significant or of economically

relevant magnitude. In sum, the job market results for the full population of state-years indicate

that co-residence is particularly prevalent in lower-earning regions, and yet not closely tied to

employment rates.

At the same time, we find a modest negative association between house prices and co-

residence. The coefficient on CoreLogic house price index is -0.013, and is significant at the ten

percent level. This indicates that a one standard deviation difference in house price index for the

pooled 2003-2015 sample of states is associated with a 0.53 percentage point lower rate of co-

residence with parents among 23- and 25-year-olds. One might consider this a precisely

45 We thank a referee for the observation that credit access common across states varies widely over this period and
may affect co-residence choices.
46 The association between national trends in intergenerational co-residence and poverty is described by Duca
(2013).

estimated very small association between house price and co-residence. Though the Section IV

model implies a positive association between housing cost and co-residence, this relies on a

measure of housing cost that sets aside family wealth. In our application, the regional house price

indices bring information both about housing costs in the region and about family wealth,

conditional on the region’s wages. Where high housing costs are predicted to encourage co-

residence, high wealth may help families support separate residences. This prediction is

reinforced empirically by our finding that co-residence is associated so closely with lower

earning power. On net, the two competing forces embodied in house prices appear to generate a

near-zero association between the house price index and the prevalence of intergenerational co-

residence, at least in the cross section.

Finally, we estimate a positive, moderate, and insignificant association between

enrollment-weighted mean tuition and co-residence for the full sample of states and years. A

$1000 increase in mean tuition is associated with a 0.14 percentage point increase in co-

residence, and this estimate has a t-statistic of less than one. In the full sample, college tuition

does not appear to be a strong determinant of co-residence for young consumers.

The parameters arising from year controls show the surprising strength of the time

dependence of co-residence with parents. The coefficients on our set of year effects reveal a

steep time path, even when conditioning on state labor and housing market conditions. The 2015

observations, for example, average 21 percentage points more co-residence than the 2003-2004

observations, all else equal.

Given a mean of 34.1 and standard deviation of 13.3 for our measure of the rate of

college graduation by age 24 among the sample state-cohort cells, our data signal substantial

variation in the relevance of college tuition across state-cohort populations. Therefore we are

also interested in differences in the relationship between the cost of college and later co-

residence with parents for regions with more and fewer college-going youth. To investigate

regional heterogeneity by educational attainment, we divide the sample of states into an upper

and a lower half, and eventually quartiles, based on the share of current 24-year-olds with

associate’s or bachelor’s degrees. Our choice of graduation rate as the educational criterion used

to group the states arose in part from the stability of graduation rates over the sample period.

While college enrollment rates for students of traditional college ages rose meaningfully between

2003 and 2015, the age-24 graduation rates we derive from IPEDS and Census measures are

approximately flat and close to a third throughout the period.47

Dividing the sample by college graduation rate reveals distinct patterns in higher- and

lower-graduation states. Columns (2) and (3) report estimates for the lower and the upper halves

of states in terms of age-24 graduation rate. For lower graduation rate states, we see a small and

insignificant negative association between state-cohort tuition and the age 23 and 25 rate of co-

residence with parents. The higher graduation rate states, however, behave differently. There we

see a positive and significant association between state-cohort tuition and co-residence, with a

state-cohort facing a $1000 higher mean tuition level realizing a 0.72 percentage point higher

rate of co-residence by ages 23 and 25, on average. Given a rise in mean tuition from 2003 to

2015 of roughly $6202 for this group, the estimate implies that tuition may explain as much as

4.5 percentage points of the roughly 10.5 percentage point rise in co-residence observed for high

graduation states over this period. The fact that the association between co-residence and tuition

appears specifically for highly educated states suggests that this relationship may be driven by

college costs themselves, and not by some additional local factor shaping both college budgets

and subsequent youth residence choices.

The co-residence pattern is quite different for lower education states. In addition to no

clear association between tuition and co-residence, in states with graduation rates in the bottom

half of the distribution, we see a meaningful association between employment and co-residence.

While the estimated association between the state-year’s QCEW employment to population ratio

and intergenerational co-residence is negative and insignificant for both the pooled sample and

the higher graduation states, the lower graduations states show a positive, large, and highly

significant relationship between employment and co-residence. Among lower education states, a

one percentage point higher employed share of the population is associated with a 0.16

percentage point higher rate of co-residence with parents at ages 23 and 25. This coefficient is

significant at the one percent level.48

47 Hence our heterogeneity analysis is insensitive to dividing states according to the graduation rate of current 24-
year-olds or of 24-year-olds from a fixed, pre-sample year. Note that the climbing enrollments and relatively
stagnant graduation rates we observe are symptomatic of a concern shared by higher education researchers. See, for
example, Athreya and Eberly (2013) and Looney and Yannelis (2015).
48 This association is consistent with positive social interaction effects with employed parents who may be better
able to help their children with residence, finding jobs, and other needs.

In sum, while intergenerational co-residence for higher education states is closely tied to

college costs, for lower education states we see co-residence instead associated meaningfully

with state employment. This pattern persists in finer graduation rate categories: when we

estimate in quartiles of state graduation rates, reported in columns (4) through (7), we find a

negative, insignificant coefficient on tuition for the lowest quartile, followed by positive and

economically meaningful coefficients on tuition for the upper three quartiles, with the largest

coefficient appearing for the highest graduation rate quartile. Here we find a $1000 increase in

mean tuition associated with a 0.94 percentage point increase in co-residence, and this estimate is

significant at the five percent level. Turning to employment, however, we find a positive

association between employment and co-residence for the lower two quartiles of states in terms

of graduation rates, with a significant coefficient of 0.327 on employment for the second quartile

indicating a 0.327 percentage point increase in co-residence associated with one percentage point

more employment there. And yet, in the fourth quartile, we see a negative and marginally

significant coefficient on employment. Our heterogeneity analysis reveals distinct co-residence

reactions to college costs and employment conditions for more and less educated regions.

Turning to the fixed effect coefficients, this division of the sample into higher and lower

education regions also reveals interesting differences in the time pattern of intergenerational co-

residence. While the year dummy coefficients climb from 4.506 in 2005 to 27.965 for the first

quartile of states by graduation rate, they reach only 1.449 by 2005 and 9.487 by 2015 for the

fourth graduation rate quartile.49 By and large, we see a substantially steeper upward trend in co-

residence for lower education states, and one closely associated with job market conditions. At

the same time, we see a lesser but still quite substantial upward trend in co-residence for higher

education states, and one more closely associated with the cost of college.

b. Stock model estimates of other residence choices

Given the steady climb in intergenerational co-residence that we have measured and estimated

above, what residential arrangements are younger Americans now forsaking, and why? Next we

estimate empirical specifications equivalent to those described by expression (1), but we replace

the outcome of the share of youth co-residing with parents with three alternative residential

49 Each of the first quartile coefficients is significant at the one percent level. The fourth quartile coefficients are
significant at the ten and five percent levels, respectively.

arrangements. The additional arrangements we model are the share of youth living with groups

of roommates, in couples, and the share living alone. These four categories – living with parents,

with groups of similarly-aged roommates, in a couple, or alone – are defined to be exhaustive.

Each observation in our sample can be categorized as one of the four. The measurement of each

living arrangement follows the definitions laid out in Section III on youth residence trends. We

report estimates of our models of other living arrangements in Table 3.

As intergenerational co-residence rates ascend from 2003 to 2015, the estimated year

effects in Table 3 reveal an offsetting decline in the rate of living with roommates. This is

particularly true for lower education states. The rate of living in couples is estimated to be

comparatively flat, though this masks a steep increase for lower-education states. Finally, in the

pooled sample and in high and low graduation groups, we observe a modest decline in the rate of

living alone.

Turning to college costs, the higher education states that showed a large positive

association between tuition and intergenerational co-residence also show an approximately

offsetting decline in living with groups of roommates as tuition rises. A $1000 increase in tuition

is associated with a 0.72 percentage point decline in co-residence with parents and a 0.86

percentage point decline in living with roommates.50 This same group shows little association

between tuition and living either in couples or alone. Our evidence suggests that young

Americans living in higher education states responded to the tuition increases that characterized

the 2003-2015 era by increasingly living with parents at the cost of living independently with

roommates.

The picture for lower education states appears to be quite different. Though these states

have lower graduation rates, college enrollment is quite common in all states, and we expect

substantial exposure to college costs for both groups. In the lower education states, our estimates

indicate no significant or substantial response to tuition in the rates of living with parents, with

roommates, or alone, but we do find a negative association between tuition and living in couples.

Specifically, $1000 increase in tuition is associated with a 0.28 percentage point decrease in the

rate of living in couples.

The employment effects on other residential arrangements that we estimate are mixed

and, for the most part, small. Wages show little effect on residential arrangements, aside from a

50 This coefficient is significant at the five percent level.

large positive association with living with roommates, particularly for lower education areas.

House prices show a modest positive and significant association with living in a couple; a one

standard deviation increase in the state-year house price index is associated with a 0.45

percentage point increase in coupledom, and this is true for both high and low education states.

c. Geographic heterogeneity in the response to tuition

One finds extensive variation across the United States in both population density and the

prevalence of intergenerational co-residence, whether in 2003 or 2015. Further, these correlate

with graduation rates. For example, the western states feature both lower graduation rates and

lower rates of intergenerational co-residence. It is useful, therefore, to examine the geographic

sources of our estimated tuition-co-residence relationship.51

First, we estimate our baseline model, as described by expression (1) and reported in

Table 2, interacting the tuition regressor with indicators for each of the four U.S. Census regions.

Though sample size begins to limit the precision of the estimates in finer categories, we also

perform this exercise separately for higher and lower graduation rate states. Table 4 reports the

estimates with region-specific tuition estimates for the full sample and top and bottom halves of

the graduation rate distribution. In the pooled data, we estimate no significant or sizable tuition

effects on co-residence in any of the four Census regions. Separating the tuition effect across the

four Census regions demonstrates that the tuition association with intergenerational co-residence

that we observe for the higher graduation rate sample arises from a positive relationship in the

Northeast and Midwest, and, to a degree, in the South Atlantic. (The precision of some of these

estimates, however, is reduced by the smaller sample sizes of the graduation rate by Census

region subcategories.) In particular, we observe a coefficient indicating that a $1000 increase in

tuition is associated with a 0.66 percentage point increase in intergenerational co-residence in the

Northeast.52

However, the relationship of co-residence to tuition appears to be quite different in the

West. In the West for both the low and high graduation rate states we estimate a negative, and in

one case substantial (though statistically insignificant), association between tuition and

intergenerational co-residence.

51 We thank a referee and the editor for this observation.
52 This coefficient is significant at the ten percent level.

Given the differing availability of living space in urban and rural environments, as well

as the difference in educational attainment in the two, we are also interested in whether the

estimated tuition-co-residence relationship arises in urban or rural locations. Table 5 reports

estimates of expression (1) with the addition of a regressor representing the percent of each state

living in urban areas, along with the interaction of the percent urban and the mean tuition for the

state.

The estimates indicate that intergenerational co-residence is considerably less common as

the population becomes more urban. (Note, of course, that the effects of time-fixed heterogeneity

are absorbed by the full set of state fixed effects.) At the average tuition level for the sample, the

point estimates in Table 5 column (1) indicate that a one percentage point increase in the share of

the state population living in urban areas decreases intergenerational co-residence by roughly

1.16 percentage points. Further, the estimates reflect a more positive association between tuition

and intergenerational co-residence in more urban states. In the pooled sample of state-years, and

evaluating at the US overall percent of urban population for 2010 of 80.7 percent, the estimates

indicate that a $1000 increase in tuition is associated with a small 0.045 percentage point decline

in intergenerational co-residence. However, in California, the state with the highest urban

population percentage, at 95 percent, a $1000 increase in tuition is associated with a 0.24

percentage point increase in intergenerational co-residence. This relationship arises from a

positive and highly significant coefficient on the interaction between percent urban and tuition.

The Table 5 estimates suggest that the positive tuition-co-residence estimate we observe

elsewhere is a more urban than rural phenomenon. Moreover, the column (2) and (3) estimates

reflect the now-familiar difference between the tuition-co-residence relationship in the lower and

higher graduation rate states. The estimates in column (3) produce a 0.59 percentage point

increase in co-residence with a $1000 increase in tuition when evaluated at the US average

percent urban of 80.7, or a 0.74 percentage point co-residence increase with $1000 higher mean

tuition for a state with the urban percentage of California.

Overall, the evidence we have gathered on the geographic heterogeneity in the tuition-co-

residence relationship reveals that the positive association between tuition and intergenerational

co-residence that we find for higher education states is driven by the strength of this relationship

in the Northeast and the Midwest, or, not unrelatedly, the strength of this relationship in states

with more urban populations. The tuition-co-residence connection is much weaker, and is even

estimated to be negative in some cases, in the Western states and in more rural areas.

d. Flow home to parents from independent living

Table 6 reports the coefficient estimates for the moving home model in expression (2).53 As

discussed above, we move to the transition approach in order to estimate parental co-residence

choices in two populations, one in which youth uniformly live on their own at the start of the

two-year period, and one in which youth uniformly live with parents. In doing so, we are using

the panel dimension of our data to resolve the problem of measuring location characteristics only

for youth locations when sample members live independently, and only for parent locations

when sample youth live with parents. This method is appealing in that it allows us to identify the

effects of local conditions operating below the state level on the choice of whether to co-reside

with parents. It is less effective, however, in relating college costs to co-residence, as college

costs do not vary over the life-cycle, and the bulk of their variation in our IPEDS-derived

measures occurs at the state level. However, to the extent that having experienced higher costs of

college potentiates co-residence reactions to unobserved shocks, such as individual job loss or

partnership dissolution, as might arise in the modified Kaplan model in Section IV, we may be

able to capture this effect by estimating the growth model with the addition of a state-cohort

tuition level.

The estimation sample consists of county-level observations of two-year changes in

employment, wage, and housing price conditions, as well as the share of 23-year-olds living

independently who have moved home to parents two years later. The full sample period is 2003-

2015, with the first set of 23-year-olds observed living independently in 2003 and the last two-

year transition window closing for 25-year-olds in 2015. Columns (1), (3), and (5) reflect

estimates of the model exclusively relating the growth in employment, wages, and housing costs

to transitions home. Columns (2), (4), and (6) add the level of the age 20-22 enrollment-weighted

state tuition mean faced by the state-cohort to what is otherwise a growth model. This latter

modification is not ideal, and our preferred specification is the simple growth model. However,

the reader may wonder about the role of college costs in shaping co-residence transitions.

Finally, note that columns (1) and (2) include state fixed effects in the rate of transition to parents

53 For the relevant scales of measurement, as well as average values of the flow variables, see Table 1.

but no time dependence in co-residence changes, columns (3) and (4) add a linear time trend to

this model of changes in co-residence, and columns (5) and (6) allow a full set of year-specific

effects in the rate of change in co-residence, arguably an extreme approach to the specification of

time dependence in a model of growth.

Estimates in Table 6 most prominently describe a protective effect of strengthening local

labor markets on the independence of youth. Across all specifications, a one percentage point

increase in local employment is associated with a 0.2 to 0.3 percentage point decline in the rate

at which youth move home to parents over the two years. The employment share coefficient is

significant at the one percent level in all cases. Increasing local wages behave similarly. A $100

increase in average weekly wages for the county over two years is associated with a 0.12 to 0.15

percentage point decline in the rate of moving home to parents, and these estimates are also each

significant at the one percent level. Strengthening local labor markets, reflected in both rising

employment rates and rising wages, stabilize youth independence.54

Increasing house prices, also often associated with strengthening local economies, appear to

have the opposite effect. Specifications (1) through (4) reflect a positive association between

house price growth and the rate at which independent youth move home, with a one standard

deviation larger growth in the house price index associated with a 1.21 to 1.92 percentage point

greater rate of transition home to parents; again each coefficient estimate is significant at the one

percent level. The direction and magnitude of these estimates stand to reason: as the cost of

housing in the county increases, families respond with a fair amount more “doubling up”. It must

be noted, however, that the result is not robust to our most flexible specification of the dynamics

of co-residence transitions.

Turning last to college costs, our transition model estimates uncover a positive association

between the tuition faced by a state-cohort and its later rate of transition home (among those who

first achieve independent living) only in the column (2) specification, which permits no time

dependence in the rate of transition home between ages 23 and 25. The estimate indicates a 1.26

percentage point increase in the two-year rate of transition home for a cohort that faced $1000

greater mean college tuition. Given the state fixed effects in this framework, the effect of state-

cohort tuition on transitions back to the parents’ home is identified using variation between

54 These decisive results are consistent with the findings in Kaplan (2012) that the job market exerts its greatest
influence on the timing of moves home.

cohorts in a given state in college costs, and hence accounts for state-specific levels and linear

trends in the influence of the degree of support for youth and education on the two-year

transitions.

Yet adding linear or fully flexible year controls leaves a negative and insignificant tuition

coefficient for the rate of moving home. Of course, the cost of college in part shapes the

population living independently by age 23, and so, recalling the Section IV modified Kaplan

model, theory generates an ambiguous prediction for association between college costs and the

rate of return for post-schooling youth who have achieved independence. For this and other

reasons described above, the tuition coefficient is only limitedly informative.

e. Flow away from parents to independent living

Table 7 reports the coefficient estimates for the model of the rate at which dependent youth

move away from home. Analogous to the Table 6 move-in estimates, the sample consists of

county-level two-year changes in employment, mean wages, and house price indices. These are

related to the rate at which 23-year-old county residents living with parents move out over the

subsequent two years. As before, columns (1), (3), and (5) estimate using only growth regressors

and columns (2), (4), and (6) add the levels of the relevant state-cohort tuition means, which

precede the estimation window. And as before, columns (1) and (2) impose no time dependence

beyond state fixed effects in the transition rates, columns (3) and (4) include a linear time trend

in the rate of change in co-residence, and columns (5) and (6) include free yearly effects in the

rate of residence changes.

By and large, the economic determinants of moving away from parents are less obvious

based on the estimates than the economic determinants of moving home. We find no significant

association between county employment growth and the rate of moving out. The estimates show

a small, positive and in some cases marginally significant relationship between house prices and

the rate at which youth move away from their parents, and yet a negative, smaller, and

insignificant association once one moves to the most flexible specification of time dependence.

The wage coefficients, however, are decisive. A $100 increase in the county’s mean weekly

wage over two years is associated with approximately a 0.4 percentage point decline in the rate

at which county youth move away from their parents.55 Weighing the wage coefficients for both

55 The wage coefficient is significant at the one percent level in all cases.

the move in and the move out, it appears both that local wage growth in the youth location

stabilizes youth independence and that local wage growth in the parent location stabilizes co-

residence with parents.

As in the model of transitions home, transitions away are closely associated with higher

state-cohort college costs only where the model omits a time trend in co-residence transitions. In

column (2), we estimate a 1.5 percentage point decline in the rate of transitions away from

parents in response to a $1000 greater state-cohort college tuition level. However, adding either a

linear or fully flexible time dependence of the rate of residence changes leaves a smaller positive

and insignificant tuition coefficient. As with the rate of moving home, the population of youth

co-residing with parents by the age of 23 is shaped in part by college costs, and college costs

affect a cohort of youth at one point over the life-cycle. As a result, the predictions of our Section

IV model for the direction of the association between past tuition and subsequent co-residence

transitions are ambiguous. Moreover, it is not clear that our approach could be expected to have

enough identifying power under such an extremely flexible model of time effects on

intergenerational co-residence to estimate this direction informatively.

VI. Discussion and Conclusions

This paper investigates young people’s parental co-residence rates in the CCP, and the

relationship among co-residence decisions and local house prices, local employment conditions,

and the cost of college. Evidence from the CCP shows that co-residence with parents has been

persistently increasing for 25-year-olds since 2004, while the number of 25-year-olds living with

more than one roommate has declined steadily. The co-residence trend corroborates similar

findings using the Census, CPS and CCP in Matsudaira (2016), Paciorek (2014), and Dettling

and Hsu (2014). Simultaneously, homeownership has decreased for both age groups, and college

tuitions have continued to grow.

In a pooled sample of state-cohort-year cells measuring the mean rate of co-residence

with parents among youth at ages 23 and 25, we find that co-residence arises most in the

presence of low or declining wages. Accounting for time-fixed heterogeneity in intergenerational

co-residence across states, as well as freely varying yearly shifts in overall co-residence, we find

that when states face average weekly wages that are $100 lower, their rate of intergenerational

co-residence is, on average, 2.2 percentage points greater. High rates of living with parents

appear to be associated with periods of weaker earnings in the region.

By splitting our sample of state-cohort-year cells into those characterized by higher and

lower college graduation rates, we uncover some interesting distinctions between the recent

growth in intergenerational co-residence for states whose populations have higher and lower

educational attainment. Among states with more educated populations, youth cohorts’ rates of

co-residence with parents following schooling are particularly responsive to changes in the

college tuition that students confronted when they were of college age. Presumably this is also,

in part, as a greater share the population is affected by higher tuition. Among the more highly

educated half of states, a $1000 increase in the tuition facing a state-cohort is associated with a

0.72 percentage point higher rate of co-residence with parents. The analogous co-residence

increase for the top quartile of states, in terms of graduation rate, is 0.94 percentage points.

Hence for the higher graduation rate half of states, the observed $6202 tuition climb over the

period is able to account for 4.5 of the observed 10.5 percentage point rise in co-residence with

parents. The influences on intergenerational co-residence of more conventional economic

factors, such as jobs and housing, for the more highly educated states are estimated to be

comparatively modest.

Among lower graduation rate states, the picture is quite different. We find no clear

response of co-residence with parents to college tuition movements from cohort to cohort, and

decisive evidence of a response to job market conditions. States with lower educational

attainment, then, make co-residence choices that appear to be closer to the jobs-based model of

residential independence of Kaplan (2012), for example. For the lower graduation rate states, we

estimate a 3.9 percentage point greater rate of co-residence with parents, on average, in times

when mean weekly wages in the state are $100 lower. In addition, the residual year effects in our

tuition model, after controlling for time-varying state economic conditions and time-fixed state

factors, increase far more steeply from 2003 to 2015 for the half of states with lower educational

attainment. While our estimates indicate a 13 percentage point secular increase in

intergenerational co-residence for the high graduation rate half of states (controlling for

economic conditions and college costs), the analogous increase for the lower graduation rate half

of states is 25 percentage points.

Analysis of the geographic heterogeneity in tuition effects on co-residence reveals that

the tuition and co-residence association we observe for the higher college graduation rate states

arises primarily in the Northeast and Midwest, or, relatedly, among states with a higher

percentage of population living in urban areas. All estimates in the paper point in the direction of

a tighter positive association between tuition and co-residence in regions of the U.S. where youth

are more likely to enroll in, or graduate from, college. In addition, while the states of the West

have also experienced a pronounced secular climb in intergenerational co-residence, the

description of a young worker lingering longer with parents in response to a large positive shock

in the cost of college turns out to be a particularly poor fit for the intergenerational co-residence

patterns that we observe in the West.

Estimates of other residential choices demonstrate that the move home to parents in

response to climbing tuition among youth in higher graduation rate states comes at the expense

of living with roommates, and not at the expense of coupledom. On the other hand, we observe a

significant and fairly substantial negative association between tuition and later residence in

couples for lower graduation rate states. Further, the rate at which area youth choose to live alone

is estimated to be surprisingly unresponsive to college costs, jobs, or housing markets.

Transition models allow us to relate developments in local economic conditions below

the state level to the decisions of young consumers to live with parents or independently. Our

estimates provide decisive evidence for the role of strengthening labor markets, in terms of both

wage growth and increasing employment rates, in reinforcing the independence of youth. Young

consumers living independently in improving local labor markets are substantially less likely to

move home. The protective effect of improving local labor markets may be weakened somewhat

by strengthening local housing markets, as rising house prices are estimated to increase the rate

at which youth move home to parents.56 Economic explanations for the rate at which youth move

away from parents between ages 23 and 25, however, find considerably less traction. Our only

clear insight regarding the economic determinants of the move out is that increasing local wages

stabilize living arrangements of both kinds, with fewer youth moving away when wages increase

in parent locations, and fewer youth moving home when wages increase in youth locations.

56 This latter result, however, is sensitive to specifications allowing extremely flexible time patterns in the rate of
residence changes.

To paint with a broad brush, our estimates of the relationship among intergenerational co-

residence choices and the economic conditions under which they are made describe two distinct

regional circumstances. In the first, many young students are exposed to college costs, and jumps

in the tuition facing a college cohort are met with extended periods of co-residence with parents,

or more moving “home” in early adulthood. In the second, fewer students confront college

tuition, and local job market conditions, with particular emphasis on wages, are paramount in

shaping the decision whether to live with parents or independently. Both types of regions,

nevertheless, are characterized by a steep secular climb between 2003 and 2015 in the rate of

intergenerational co-residence, one that operates over and above any pressure to co-reside arising

from weakening local labor markets or rising costs of housing and college.

Finally, it is important to qualify our estimates relating local tuition to population

averages of youth residential outcomes. Cross-sectional and transition estimates of parental co-

residence patterns that rely on local aggregates may be useful in addressing the specific

estimation challenges described by this paper, including both the confounding link between the

measurement of local conditions and whether youth have chosen to live with parents, and any

underlying heterogeneity in individual youth ability and family generosity. However, working

with aggregates implies that all statements made based on the estimates reported in this paper

describe local average levels and changes in the tuition cost of a degree. Higher education

investments, and their returns, may be quite heterogeneous, as described by Avery and Turner

(2012). While these results represent the association between the average cost of higher

education in a location and the rate of remaining or returning home, they may, for example,

reflect a mix of strong homeward pressure in response to low-return educational investments and

a pressure toward independence in response to high-return investments.

References

Agarwal, S., L. Hu, & X. Huang. 2013. Rushing into American Dream? House Prices, Timing of
Homeownership, and Adjustment of Consumer Credit. Federal Reserve Bank of Chicago Working Paper
2013-13.

Athreya, Kartik and Janice Eberly. 2013. “The supply of college-educated workers: the roles of
college premia, college costs, and risk,” Federal Reserve Bank of Richmond working paper,
http://www.richmondfed.org/publications/research/working_papers/2013/pdf/wp13-02 .

Autor, D., D. Dorn, and G. Hanson. 2013. The China Syndrome: Local Labor Market Effects of Import
Competition in the United States. American Economic Review 103(6): 2121-2168.

Avery, R., P. Calem, & G. Canner. 2003. An Overview of Consumer Data and Credit Reporting. Federal
Reserve Bulletin 01/2003.

Avery, C. and S. Turner. 2012. Student Loans: Do College Students Borrow Too Much–Or Not Enough?
Journal of Economic Perspectives, 26(1): 165–92.

Barr, A. and S. Turner. 2015. Out of Work nad Into School: Labor Market Policies and College
Enrollment During the Great Recession. Journal of Public Economics 124: 63-73.

Becker, S.O., S. Bentolila, A. Fernandes, & A. Ichino. 2010. Youth emancipation and perceived job
insecurity of parents and children. Journal of Population Economics 23: 1175-1199.

Bleemer, Z., M. Brown, D. Lee, K. Strair, & W. van der Klaauw. Echoes of Rising Tuition in Students’
Borrowing, Educational Attainment, and Homeownership in Post-Recession America. Manuscript.

Board of Governors of the Federal Reserve System. 2015. Report on the Economic Well-Being of U.S.
Households in 2014.

Brown, M., J. Scholz, & A. Seshadri. 2012. A New Test of Borrowing Constraints for Education. Review
of Economic Studies 79(2): 511-538.

Brown, M. & S. Caldwell. 2013. Young Student Loan Borrowers Retreat from Housing and Auto
Markets. Federal Reserve Bank of New York Liberty Street Economics Blog.

Brown, M., S. Caldwell, & S. Sutherland. 2014. Just Released: Young Student Loan Borrowers Remained
on the Sidelines of the Housing Market in 2013. Federal Reserve Bank of New York Liberty Street
Economics Blog.

Brown, M., J. Grigsby, W. van der Klaauw, J. Wen, and B. Zafar. 2014. Financial Education and the Debt
Behavior of the Young. Federal Reserve Bank of New York Staff Reports 634.

Brown, M., A. Haughwout, D. Lee, J. Scally, & W. van der Klaauw. 2015a. The Student Loan Landscape.
Federal Reserve Bank of New York Liberty Street Economics Blog.

Brown, M., A. Haughwout, D. Lee, & W. van der Klaauw. 2015b. Do We Know What We Owe? A
Comparison of Borrower- and Lender-Reported Consumer Debt. Federal Reserve Bank of New York
Economic Policy Review.

Card, D. & T. Lemieux. 2000. Adapting to Circumstances: The Evolution of Work, School, and Living
Arrangements among North American Youth. In Youth Employment and Joblessness in Advanced
Countries, ed. Blanchflower and Freeman. Chicago: University of Chicago Press.

Cooper, D. & J.C. Wang. 2014. Student Loan Debt and Economic Outcomes. Federal Reserve Bank of
Boston Current Policy Perspectives 14-7.

Consumer Financial Protection Bureau (CFPB). 2013. Student Loan Affordability: Analysis of Public
Input on Impact and Solutions. Washington, DC.

Dettling, L. & J. Hsu. 2014. Returning to the Nest: Debt and Parental Co-Residence Among Young
Adults. Federal Reserve Board Finance and Economics Discussion Series 2014-80.

Dettling, L. & J. Hsu. 2015. Why Boomerang? Debt, Access to Credit, and Parental Co-residence among
Young Adults. FEDS Notes, https://www.federalreserve.gov/econresdata/notes/feds-notes/2015/why-
boomerang-debt-access-to-credit-and-parental-co-residence-among-young-adults-20151001.html

Duca, J. 2014. When will the Kids Ever Move Out? Manuscript.

Dustmann, Christian. 2003. Return migration and the optimal migration duration. European Economic
Review 47, 353–67.

Dyrda, S., G. Kaplan, & J. Ríos-Rull. 2012. Business Cycles and Household Formation: The Micro vs the
Macro Labor Elasticity. NBER Working Paper 17880.

Ermisch, J. 1999. Prices, Parents, and Young People’s Household Formation. Journal of Urban
Economics 45: 47-71.

Federal Reserve Bank of New York (FRBNY). 2014. Quarterly Report on Household Debt and Credit,
February 2014. New York.

Gicheva, D. & J. Thompson. 2014. The Effects of Student Loans on Long-Term Household Financial
Stability. University of North Carolina at Greensboro Department of Economics Working Papers 14-02.

Goldscheider, F. K. & J. DaVanzo. 1985. Living Arrangements and the Transition to Adulthood.
Demography 22(4): 545-563.

Goldscheider, F. K. & J. DaVanzo. 1989. Pathways to Independent Living in Early Adulthood: Marriage,
Semiautonomy, and Premarital Residential Independence. Demography 26(4): 597-614.

Haurin, D., P. H. Hendershott, & D. Kim. 1993. The Impact of Real Rents and Wages on Household
Formation. The Review of Economics and Statistics 75(2): 284-293.

Hershbein, B & K. Hollenbeck. 2014. The Distribution of College Graduate Debt, 1990 to 2008: A
Decomposition Approach. Upjohn Institute Working Papers 14-204.

Houle, J. & L. Berger. 2014. Is Student Loan Debt Discouraging Home Buying Among Young Adults?.
Association for Public Policy and Management. Manuscript.

Hunt, J. 2004. Are migrants more skilled than non-migrants? Repeat, return, and same-employer
migrants. Canadian Journal of Economics/Revue canadienne d’économique, 37: 830–849.

Jacob, K., & R. Schneider. 2006. Market Interest in Alternative Data Sources and Credit Scoring. Center
for Financial Services Innovation Articles 2817.

Kaplan, Greg. 2012. Moving Back Home: Insurance Against Labor Market Risk. Journal of Political
Economy 120(3).

Klasen, S. & I. Woolard. 2008. Surviving Unemployment without State Support: Unemployment and
Household Formation in South Africa. Journal of African Economies 18(1): 1-51.

Kurz, C. & G. Li. 2015. How Does Student Loan Debt Affect Light Vehicle Purchases?. FEDS Notes,
February 2, 2015.
Lautz, J. 2011, March 7. Median Age of Home Buyers: 2001-2010. National Association of Realtors
Economists’ Outlook Blog.

Lawrence Journal-World. 2013. Republican budget writerspropose 4 percent across-the-board cut to
higher education. http://www2.ljworld.com/news/2013/mar/12/republican-budget-writers-propose-4-
percent-across/?print

Lee, D. & W. van der Klaaw. 2010. An Introduction to the FRBNY Consumer Credit Panel. Federal
Reserve Bank of New York Staff Reports 479.

Lochner, L. and A. Monge-Naranjo. 2012. The Nature of Credit Constraints and Human Capital.
American Economic Review, 101(6): 2487-2529.

Lochner, L., T. Stinebrickner, and U. Sulemanoglu. 2013. The Importance of Financial Resources for
Student Loan Repayment: Evidence from the Canada Student Loans Program. CIBC Working Paper
#2013-7, University of Western Ontario.

Looney, Adam and Constantine Yannelis. 2015. “A Crisis in Student Loans?” Brookings Institution,
Washington, D.C. http://www.brookings.edu/about/projects/bpea/papers/2015/looney-yannelis-student-
loan-defaults

Mather, M. & D. Lavery. 2010, September. In U.S., Proportion Married at Lowest Recorded Levels.
Population Reference Bureau Articles.

Matsudaira, J. 2016. Economic Conditions and the Living Arrangements of Young Adults. Journal of
Population Economics.

McGarry, K. & R. Schoeni. 1995. Transfer Behavior in the Health and Retirement Study; Measurement
and the Redistribution of Resources within the Family. Journal of Human Resources 30: S185-S226.

MeasureOne. 2013. The MeasureOne Private Student Loan Report 2013. San Francisco, CA: D.
Arvidson, D. Feshbach, R. Parikh, & J. Weinstein.

Mezza, Alvaro, Kamila Sommer, and Shane Sherlund. 2014. Student Loans and Homeownership Trends.
FEDS Notes, October 15, 2014.

Molloy, R., C. Smith, & A. Wozniak. 2014. Declining Migration within the U.S.: The Role of the Labor
Market. National Bureau of Economic Research Working Papers 20065.

Mykyta, L. & S. Macartney. 2011. The effects of recession on household composition: “doubling up” and
economic well-being. SEHSD Working Paper 2011-4.

National Association of Realtors (NAR). 2014, March 20. February Existing-Home Sales Remain
Subdued. NAR News Releases.

National Defense Authorization Act for Fiscal Year 2013, H.R. 4310.

Population Reference Bureau, 2012, Fact Sheet: The Decline in U.S. Fertility. World Population Data
Sheet 2012.

State Higher Education Executive Officers. 2013. State Higher Education Finance – FY2012.
http://www.sheeo.org/sites/default/files/publications/SHEF-FY12

Stock, J. H. & M. Yogo. 2005. Testing for Weak Instruments in Linear IV Regression. Identification and
Inference for Econometric Models: Essays in Honor of Thomas Rothenberg, ed. D. W. K. Andrews & J.
H. Stock, 80-108. New York: Cambridge University Press.

Whittington, L. A. & H. E. Peters. 1996. Economic Incentives for Financial and Residential
Independence. Demography 33(1): 82-97.

Wolfers, J. 2010, October 13. How Marriage Survives. The New York Times: A25.

Appendix

a. Comparison of our estimates of the national co-residence trend to those of prior or concurrent
studies using large and (arguably) reliable surveys

Of course, others have documented this large and ongoing change in intergenerational co-

residence in the U.S., either over earlier periods of time or for very recent years, in parallel to

this study. Their findings using large existing surveys and other sources provide us with an

opportunity to validate our measures of intergenerational co-residence in the comparatively

novel, entirely administrative Consumer Credit Panel. Matsudaira (2016) studies co-residence

with parents among young adult Americans using decennial Census data from 1960 through

2000 (as well as some ACS data thereafter). Matsudaira, therefore, provides a valuable point of

comparison between our year 2000 intergenerational co-residence rate for 25-year-olds and a rate

calculated in the large and reliable decennial Census as a part of careful prior research. Where

Matsudaira measures a 25 percent rate of co-residence with parents among 25-year-old men, and

a 21 percent rate among 25-year-old women during 2000, we find a 29 percent rate of co-

residence with parents at the end of 2000 among all 25-year-olds in the CCP. Our analysis of our

co-residence criterion applied to the CPS suggests that 2 percentage points out of the 29 are

likely miscategorized. The distance between the Matsudaira’s 23 percent co-residence rate and

our 29 percent co-residence rate may be accounted for by a combination of this

miscategorization and the difference generated by our broad definition of parent or similar elder

and Matsudaira’s parent-only criterion.

Paciorek (2014) is particularly relevant for our purposes, as he employs CPS data, an

additional, detailed, and arguably quite reliable source, and he applies a measure of co-residence

with parents that is quite close to our own. His broad measure of intergenerational co-residence

status for 18- to 31-year-olds in the 2000-2012 March CPS includes youth who are “living with

older relatives”, and excludes those living in group quarters. Hence Paciorek’s co-residence

criterion closely resembles an implementation of our CCP co-residence criterion for the survey

data of the CPS. Using this approach, he traces an increase in intergenerational co-residence

among 18- to 31-year-olds in the CPS from 39 to 46 percent between 2000 and 2012. By

comparison, we measure an increase from roughly 29 percent to more than 44 percent over the

same period for 25-year-olds in the CCP. This comparison demonstrates two things: the height of

co-residence at the end of our panel is not evidence of anomalous measures or poor data quality,

but instead aligns with the surprising height of intergenerational co-residence rates measured by

other researchers using reliable sources. Second, we chose to study ages in the mid-twenties

because, in our data, this seemed to be the age at which co-residence was changing most rapidly.

Where the vast majority of 18-year-olds co-reside with elders and thirtysomethings have largely

gained independence, the residential circumstances of Americans’ mid-twenties appear to be

undergoing rapid reform. Hence we are unsurprised to find a flatter co-residence trend for the

average among 18- to 31-year olds than the trend we measure for 25-year-olds, who stand at the

precise point of the greatest residential upheaval.

Finally, Dettling and Hsu (2014, 2015) measure intergenerational co-residence in the

CPS over many years, including very recent waves. This final point of comparison with the

literature allows us to consider the accuracy of our CCP co-residence measure through 2013.

Dettling and Hsu adopt a narrower definition of intergenerational co-residence: living with a

parent, which they can measure reasonably reliably given the detail available in the CPS.57 They

track co-residence for CPS youth who are ages 18 to 31. The prevalence of co-residence with

parents in their data is first flat near 31 percent over the early- and mid-2000s, and then climbs

from roughly 31 to 37 percent by 2013.58 Despite the narrower co-residence criterion that

Dettling and Hsu apply, the rate they measure reaches a height that approaches the 43 percent

that we observe in the CCP for 2013. Like the Dettling and Hsu CPS series, our CCP co-

residence trend includes a comparatively flat region in the mid-2000s. An important point of

distinction is the wide age 18 to 31 band used in each of the CPS papers, as compared with our

age 25 trend. Like Paciorek, Dettling and Hsu’s age 18 to 31 co-residence slope is considerably

flatter than the slope of intergenerational co-residence among 25-year-olds that we measure in

the CCP.

57 As an interesting but likely inconsequential, side note, we have been surprised to learn of the difficulty inherent in
measuring co-residence with parents in both survey and administrative data. In survey data, in order to infer co-
residence with parents, one must either ask about living with parents outright, which is a somewhat narrow survey
item to be applied to every household member and is thus unpopular, or one must document the structure of
relationships among all household members. The CPS, for example, tracks up to 15 household members. The exact
relationship between each possible pair among the 15 is not covered in its entirety, as of course this would be a
prohibitively costly data gathering exercise. However, the relationship accounting available in the CPS remains
unusually detailed.
58 Relatedly, a 2013 report from the Pew Research Center, based on their own analysis of the March Current
Population Survey (CPS), reported that 32 percent of 18-31-year-olds in 2007, 34 percent in 2009, and 36 percent in
2012 live with parents. The Pew analysis defines an individual as living with a parent only if she lives with a parent
or step-parent, not a parent-in-law or the partner of a parent, and only if she is not herself a head of household.

b. On the exogeneity of state-cohort mean tuition to post-schooling residential outcomes

State-cohort average tuition is included in expression (1) as an exogenous measure of the

cost of college for the population whose residential outcome we are estimating. The assumption

we impose is that variation in this tuition measure, once we condition on the calendar year, the

state, and the employment, youth unemployment, mean wages, and house price index for the

state-cohort-year, is exogenous to the state-cohort-year’s rate of intergenerational co-residence

(or living alone, with roommates, or in a couple). Practically speaking, we are all aware of

extensive time-fixed, cross-state variation in college and university tuition. Further, the broader

business cycle in the U.S. can be expected to affect state budgeting, and therefore state

appropriations for higher education, in a non-linear fashion over time. Local economic

conditions may deviate from the national cycle, and these are accounted for in the estimation

using four standard measures of prevailing economic conditions. The relationship that we

estimate between college costs and residential outcomes is identified using variation in average

tuition beyond the variation associated with each of these aforementioned factors.

Examples of the sort of tuition variation we have in mind often arise from legislative shocks.

In one example, large portions of the higher education budgets of 43 states in 2010, and 31 states

in 2011, were heavily dependent on federal American Recovery and Reinvestment Act (ARRA)

funds. In 2012, federal ARRA funding dried up for the states. This constituted a meaningful

negative shock to the higher education budgets of 31 states, and yet not a meaningful shock to

the higher education budgets of 19 states. Any difference across states in tuition growth from

2011 to 2012 or even 2013 generated by the differential impact on states of the withdrawal of

ARRA funding, over and above its effects on local labor and housing markets, will serve to

identify our tuition coefficient.59 Another example appears in Lawrence Journal-World (2013).

In 2012, Republican and Democratic Kansas state legislators engaged in negotiations regarding

the percent of funds to cut from the state’s higher education budget appropriations. To the extent

that realizations of such negotiations in state legislatures from year to year deviate from the

state’s usual budgeting outcomes, from national trends in states’ higher education budgeting, and

from regional business cycles as represented by state employment, youth unemployment, wages,

59 On state appropriations, state board of regents tuition-setting, and the process described in this paragraph, see, for
example, State Higher Education Executive Officers (2013).

and house prices, they will contribute to the identification of the tuition association with youth

residence outcomes that we estimate.

Other researchers have turned to state-year tuition means for arguably exogenous variation

in the cost of college. Kane (1994), Rouse (1995), and Souleles (2000), for example, have relied

on state college tuition averages to estimate, respectively, the influence of college costs on the

enrollment rates of African American students, the influence of community colleges on final

educational attainment, and the influence of college costs on the consumption and saving

decisions of students’ families.

Several empirical findings argue for the validity of the exogeneity assumption for our

college cost proxy. In Bleemer et al. (2017), we find that estimates of the tuition effect on

educational attainment, student borrowing, and homeownership are insensitive to the inclusion

(or exclusion) of controls for local employment, youth unemployment, wages, and house prices.

These results go some distance toward addressing the possibility of endogeneity of state-cohort

tuition shocks to youth residential outcomes via local economic conditions. Further, we estimate

precise and inconsequential associations of educational attainment (college enrollment, college

graduation, years of schooling) with state-cohort tuition levels. These results call into question

any claims that the tuition association with residential outcomes that we estimate arises solely

from schooling attainments that decline as tuition rises. (Though, if present, we would interpret

such schooling effects as simply one mechanism by which college costs affect young Americans’

post-college outcomes.)

Of course, tuition and residential outcomes could both be driven by aspects of local

economic conditions not represented in employment, youth unemployment, wage, and house

price measures. However, if this were the case, we might expect such factors to affect co-

residence outcomes for youth who do and do not attend college. The results in section VI,

however, indicate a substantial and precisely estimated tuition-co-residence relationship only

among higher college graduation states.60 Much stronger estimated tuition effects for populations

60 The notion that youth in higher education states should be more responsive to within-state tuition changes in their
post-schooling living arrangements is akin to the assumption underlying U.S.-China trade shock studies of the U.S.
labor market. There, the effect of manufacturing trade with China on U.S. workers is expected to be greater in
regions in which workers participate more heavily in the affected sector. See, for example, Autor, Dorn, and Hanson
(2013).

more exposed to college tuition would seem to argue against broader local economic factors that

drive both tuition and co-residence.61

In related research, Mezza et al. (2016) instrument student debt using state-cohort mean

tuition and demonstrate that their estimated effect is unresponsive to the inclusion (or exclusion)

of measures of local economic conditions. Further, they find that homeownership is closely

associated with state-cohort mean tuition only for youth who attend college, and unrelated with

tuition among youth who do not attend college.

61 Of course, our local aggregates must each include some college-educated youth, and so a finding of some tuition
effect on residential outcomes among lower education regions would not necessarily signal endogeneity of tuition to
residential outcomes via unmeasured local conditions.

Figure 1: Enrollment‐weighted mean public and private tuition by state, 2001‐2013

2003 2005 2007 2009 2011 2013 2015

Pe
rc
en

t
o
f
Po

p
u
la
ti
o
n

Figure 2: Co‐residence with parents among 25‐year‐olds in the CCP, 1999‐
2016

Figure 3a: Percentage of 25-year-olds living with parents in 2002-2003

Figure 3b: Percentage of 25-year-olds living with parents in 2012-2013

ALAR

AZCA
CO

FL GA

IA
ID

MA
MD

NY
OH

ORPA

SD
TN

TX
UT

WI
WV

‐10

‐5

‐5 0 5 10 15 20

P
e
rc
e
n
ta
ge

In
cr
e
as
e
in

2
5
‐Y
e
ar
‐O

ld
P
ar
e
n
ta
l C

o
‐R
e
si
d
e
n
ce

Increase in Average Student Debt per Graduate

Figure 4: Changes in Student Debt and Parental Co‐Residence, 2008 to 2013

Table 1: Descriptive statistics, pooled locations and parent v. independent youth locations

(1) (2) (3) (4)

Stock Values

All All No Parents Parents

Employment to Population, QCEW 57.54 ‐0.860 0.069 ‐0.745

(13.27) (2.6) (13.66) (9.85)

Youth unemployment, CPS 10.62 ‐0.021 0.130 ‐0.252

(3.48) (3.76) (3.82) (3.72)

Average Weekly Wage, $100s, QCEW 8.6 0.201 0.247 0.134

(1.7) (1.69) (1.89) (1.36)

House Price Index, 100 in 2000, CoreLogic 149.76 3.359 2.803 3.720

(33.79) (30.27) (30.88) (28.78)

State‐level urban pct of population, Census 74.1 ‐ ‐ ‐

(14.7)

Graduation rate, state‐cohort, IPEDS 34.1 28.90 29.37 29.26

(levels in all columns) (13.3) (10.81) (10.81) (10.86)

Average Total Sticker Cost, $1000s, IPEDS 11.8 12.47 12.02 12.99

(levels in all columns) (4.7) (4.41) (4.27) (4.50)

Living with Parents, CCP 39.8 41.85 ‐ ‐

(9.2) (49.33)

Move in / out over two years, CCP ‐ ‐ 18.23 33.73

(25.43) (33.46)

N = 1,020 546,824 273,574 196,864

Notes: Column (1) data are state‐year‐cohort aggregates. Column (2) ‐ (4) data are county‐year‐

cohort aggregates. Standard deviations in parentheses. Tuitions are state‐cohort averages

of age 20‐22 tuition for each cohort.

Flow Values

Table 2: Fixed effects model of the share of 23 and 25 year‐olds in the state who are living with parents

(1) (2) (3) (4) (5) (6) (7)

All Lower half grad rate Upper half grad rate First quartile Second quartile Third quartile Fourth quartile

Tuition mean, $1000s, IPEDS 0.140 ‐0.051 0.718** ‐0.217 0.696 0.652 0.935**

(0.198) (0.229) (0.265) (0.341) (0.413) (0.781) (0.380)

Employment to Population, QCEW ‐0.034 0.160*** ‐0.072 0.108 0.327*** 0.046 ‐0.146*

(0.062) (0.049) (0.067) (0.185) (0.080) (0.132) (0.082)

Youth unemployment, state, CPS 0.060 0.065 0.053 0.110 0.020 ‐0.013 0.073

(0.062) (0.072) (0.079) (0.144) (0.069) (0.109) (0.075)

Avg. Weekly Wage, $100s, QCEW ‐2.119*** ‐3.884** ‐1.577 ‐5.689* ‐2.886 ‐1.588 ‐1.616

(0.709) (1.669) (0.940) (2.942) (2.823) (2.157) (1.077)

House Price Index, 100 in 2000, ‐0.013* ‐0.005 ‐0.012 0.014 ‐0.007 ‐0.022 ‐0.005

CoreLogic (0.007) (0.010) (0.026) (0.019) (0.022) (0.040) (0.026)

Age = 23 6.768*** 6.272*** 6.946*** 5.705*** 6.328*** 6.889*** 6.839***

(0.394) (0.418) (0.647) (0.451) (0.862) (0.769) (1.063)

Year = 2005 3.671*** 4.210*** 2.052*** 4.506*** 3.268*** 2.699** 1.449*

(0.409) (0.414) (0.546) (0.699) (0.557) (1.007) (0.807)

Year = 2006 6.224*** 6.241*** 4.278*** 6.282*** 4.618*** 4.690*** 3.840***

(0.412) (0.488) (0.834) (0.862) (1.117) (1.095) (0.976)

Year = 2007 7.642*** 8.579*** 4.416*** 8.736*** 6.354*** 5.871** 2.641

(0.736) (0.687) (1.139) (1.271) (1.697) (2.164) (1.892)

Year = 2008 9.092*** 10.290*** 5.315*** 11.431*** 6.396*** 5.806* 4.588***

(0.949) (1.233) (1.197) (2.236) (2.023) (2.817) (1.566)

Year = 2009 8.939*** 10.346*** 4.716*** 11.368*** 6.564*** 5.534* 3.186

(0.965) (1.006) (1.327) (1.821) (2.218) (2.988) (2.203)

Year = 2010 13.453*** 15.545*** 8.933*** 17.421*** 10.561*** 9.994*** 7.459***

(1.164) (1.443) (1.429) (2.243) (2.830) (2.995) (2.420)

Year = 2011 13.803*** 16.394*** 8.622*** 17.662*** 11.624*** 9.738*** 7.323**

(1.280) (1.499) (1.518) (2.407) (2.953) (3.341) (2.574)

Year = 2012 14.903*** 18.229*** 9.045*** 19.915*** 12.442*** 10.623** 6.126*

(1.522) (1.842) (1.984) (3.166) (3.651) (4.262) (3.039)

Year = 2013 15.556*** 19.386*** 9.064*** 21.823*** 12.808*** 10.257* 6.767**

(1.716) (2.398) (2.076) (4.463) (3.823) (5.325) (3.020)

Year = 2014 19.930*** 24.174*** 12.721*** 25.945*** 17.864*** 13.923** 10.456**

(1.804) (2.213) (2.200) (3.791) (4.617) (5.357) (3.582)

Year = 2015 21.069*** 25.714*** 13.045*** 27.965*** 18.154*** 14.752** 9.487**

(2.164) (2.698) (2.600) (4.786) (4.885) (6.628) (3.446)

State FE Yes Yes Yes Yes Yes Yes Yes

R‐squared 0.881 0.904 0.863 0.897 0.928 0.837 0.909

N = 1019 475 544 227 248 282 262

Notes: *, **, and *** indicate significance at the ten, five, and one percent level, respectively. The sample covers 2003‐2015.

Ta
b
le
3
:
Fi
xe
d
e
ff
ec
ts
m

o
d
e
l o

f
th
e
s
h
a
re
o
f
23

a
n
d
2
5
y
ea

r‐
o
ld
s
in
t
h
e
s
ta
te
w
h
o
a
re
li
vi
n
g
in

c
o
u
p
le
s,
a
lo
n
e
,
an

d
w
it
h
g
ro
u
p
s
o
f
ro
o
m
m
at
e
s

(1
)

(2
)

(3
)

(4
)

(5
)

(6
)

(7
)

(8
)

(9
)

R
o
o
m
m
a
te
s

R
. ‐
L
o
w
g
ra
d
R
. ‐
H
ig
h
g
ra
d

C
o
up

le
C
o
u
pl
e
‐
Lo

w
g
ra
d
C
o
u
p
le
‐
H
ig
h
gr
ad

A
lo
ne

‐
L
o
w
g
ra
d

A
lo
n
e
‐
H
ig
h
gr
a
d

T
ui
ti
o
n
m
ea

n,
$
10

0
0s
, I
P
ED

S
‐0
.0
4
9

0.
20

3
‐0
.8
60

*
**

‐0
.2
24

**
*

‐0
.2
84

**
*

0
.0
59

‐0
.0
1
4

0.
00

3
‐0
.1
7
2

(0
.2
21

)
(0
.2
5
2)

(0
.2
8
1)

(0
.0
7
7)

(0
.0
86

)
(0
.1
53

)
(0
.0
70

)
(0
.0
7
8)

(0
.1
39

)

E
m
pl
o
ym

en
t
to
P
o
p
ul
at
io
n
, Q

C
E
W

0
.0
20

‐0
.2
01

*
**

0.
11

2
0.
0
46

*
0.
0
55

0
.0
22

‐0
.0
0
5

0.
00

8
‐0
.0
4
1

(0
.0
78

)
(0
.0
5
5)

(0
.0
8
7)

(0
.0
2
6)

(0
.0
37

)
(0
.0
43

)
(0
.0
24

)
(0
.0
2
4)

(0
.0
26

)

Y
o
u
th
u
ne

m
pl
o
ym

en
t,
s
ta
te
, C

P
S

‐0
.1
3
2*

‐0
.0
87

‐0
.1
50

*
0.
05

5
0.
0
09

0
.0
53

0
.0
18

‐0
.0
21

0.
0
99

*
*

(0
.0
73

)
(0
.0
9
0)

(0
.0
8
5)

(0
.0
3
4)

(0
.0
47

)
(0
.0
50

)
(0
.0
29

)
(0
.0
4
5)

(0
.0
37

)

A
vg
. W

ee
kl
y
W
ag
e
, $
1
00

s,
Q
C
E
W

2.
38

4*
*

4
.8
23

**
1.
70

6
0.
10

8
‐0
.4
68

0
.0
92

0
.0
07

‐0
.0
36

0
.3
15

(0
.9
07

)
(1
.9
9
5)

(1
.7
4
3)

(0
.4
0
2)

(0
.5
58

)
(0
.5
42

)
(0
.3
22

)
(0
.5
1
6)

(0
.6
48

)

H
o
u
se
P
ri
ce
I
nd

ex
, 1
00

in
2
0
00

,
0
.0
08

0.
00

4
‐0
.0
13

0
.0
10

**
0.
0
11

*
**

0
.0
11

0
.0
01

‐0
.0
05

0
.0
10

C
o
re
Lo

gi
c

(0
.0
08

)
(0
.0
1
1)

(0
.0
2
5)

(0
.0
0
4)

(0
.0
04

)
(0
.0
16

)
(0
.0
04

)
(0
.0
0
6)

(0
.0
11

)

A
ge
=
2
3

1.
9
64

*
**

1.
82

7
**

*
3.
13

4
**

*
‐5
.7
44

**
*

‐5
.2
57

**
*

‐6
.9
6
5*

**
‐2
.3
6
4*

*
*

‐2
.3
01

**
*

‐2
.2
4
1*

(0
.3
34

)
(0
.4
0
1)

(0
.5
5
2)

(0
.2
9
2)

(0
.2
76

)
(0
.4
21

)
(0
.1
95

)
(0
.2
1
7)

(0
.3
41

)

Ye
ar
=
2
00

5
‐3
.5
24

**
*

‐4
.1
44

*
**

‐1
.7
40

*
**

0.
87

2*
*
*

4.
2
10

*
**

0
.8
67

**
*

‐0
.9
6
8*

*
*

‐0
.8
9
2*

*
‐1
.0
08

(0
.3
60

)
(0
.3
2
2)

(0
.5
0
4)

(0
.2
1
2)

(0
.4
14

)
(0
.2
42

)
(0
.2
53

)
(0
.3
4
5)

(0
.3
88

)

Ye
ar
=
2
00

6
‐5
.1
04

**
*

‐5
.5
95

*
**

‐1
.8
3
5*

*
0.
82

7*
*
*

6.
2
41

*
**

0
.9
93

**
*

‐1
.5
2
0*

*
*

‐1
.3
25

**
*

‐1
.7
14

(0
.4
76

)
(0
.4
3
5)

(0
.8
8
8)

(0
.2
6
4)

(0
.4
88

)
(0
.2
77

)
(0
.3
02

)
(0
.4
1
4)

(0
.6
75

)

Ye
ar
=
2
00

7
‐7
.2
09

**
*

‐8
.8
60

*
**

‐2
.8
65

*
**

0.
85

7*
*
*

8.
5
79

*
**

1
.1
86

**
*

‐1
.1
2
2*

*
*

‐0
.8
44

‐1
.1
5
3

(0
.8
78

)
(0
.9
4
3)

(1
.0
3
9)

(0
.2
7
3)

(0
.6
87

)
(0
.3
42

)
(0
.3
66

)
(0
.5
0
2)

(0
.7
26

)

Ye
ar
=
2
00

8
‐8
.1
75

**
*

‐1
0.
27

2*
*
*

‐2
.5
38

*
0.
89

4*
*
*

1
0.
29

0*
*
*

1
.2
18

**
*

‐1
.2
7
7*

*
*

‐0
.9
4
6*

‐1
.4
88

(1
.1
81

)
(1
.4
3
1)

(1
.3
0
2)

(0
.3
2
6)

(1
.2
33

)
(0
.4
31

)
(0
.3
75

)
(0
.5
0
2)

(0
.8
44

)

Ye
ar
=
2
00

9
‐7
.1
68

**
*

‐9
.5
65

*
**

‐1
.0
92

0.
5
98

*
1
0.
34

6*
*
*

1.
11

0*
*

‐1
.3
7
9*

*
*

‐1
.1
2
7*

*
‐1
.3
7
6

(1
.1
29

)
(1
.2
2
2)

(1
.2
6
0)

(0
.3
0
8)

(1
.0
06

)
(0
.4
38

)
(0
.4
05

)
(0
.4
6
2)

(0
.9
00

)

Ye
ar
=
2
01

0
‐1
1
.6
37

**
*

‐1
4.
77

5*
*
*

‐5
.0
52

*
**

0.
40

0
1
5.
54

5*
*
*

0.
98

8*
*

‐0
.9
13

**
‐0
.6
64

‐0
.9
7
1

(1
.3
98

)
(1
.6
2
4)

(1
.4
5
9)

(0
.4
3
5)

(1
.4
43

)
(0
.4
51

)
(0
.4
46

)
(0
.4
8
4)

(0
.9
03

)

Ye
ar
=
2
01

1
‐1
0
.7
84

**
*

‐1
4.
37

6*
*
*

‐3
.3
6
2*

*
‐0
.1
06

1
6.
39

4*
*
*

0
.6
41

‐1
.0
56

**
‐0
.9
06

‐0
.8
9
5

(1
.4
12

)
(1
.5
7
5)

(1
.4
9
7)

(0
.4
9
4)

(1
.4
99

)
(0
.5
94

)
(0
.4
74

)
(0
.5
6
9)

(0
.9
90

)

Ye
ar
=
2
01

2
‐1
1
.8
00

**
*

‐1
6.
44

4*
*
*

‐3
.3
96

0.
29

3
1
8.
22

9*
*
*

1.
28

2*
*

‐1
.3
24

**
‐1
.1
7
1*

‐1
.0
4
9

(1
.7
37

)
(2
.1
4
6)

(2
.0
0
9)

(0
.5
4
7)

(1
.8
42

)
(0
.6
18

)
(0
.5
59

)
(0
.6
4
1)

(1
.1
11

)

Ye
ar
=
2
01

3
‐1
0
.3
64

**
*

‐1
5.
71

8*
*
*

‐1
.0
28

‐0
.3
84

1
9.
38

6*
*
*

0
.8
33

‐2
.2
9
9*

*
*

‐2
.2
58

**
*

‐1
.8
4
5

(2
.0
70

)
(2
.9
2
1)

(2
.3
6
8)

(0
.5
6
0)

(2
.3
98

)
(0
.7
25

)
(0
.5
77

)
(0
.6
5
7)

(1
.2
36

)

Ye
ar
=
2
01

4
‐1
3
.1
36

**
*

‐1
8.
46

9*
*
*

‐3
.1
64

‐0
.9
08

2
4.
17

4*
*
*

0
.0
11

‐3
.2
7
3*

*
*

‐3
.4
55

**
*

‐2
.4
68

(2
.1
84

)
(2
.8
2
2)

(2
.3
3
6)

(0
.5
9
6)

(2
.2
13

)
(0
.6
58

)
(0
.7
30

)
(0
.8
3
6)

(1
.4
50

)

Ye
ar
=
2
01

5
‐1
4
.1
05

**
*

‐2
0.
27

6*
*
*

‐2
.6
88

‐0
.9
26

2
5.
71

4*
*
*

0
.0
45

‐3
.0
6
1*

*
*

‐3
.0
51

**
*

‐2
.1
6
4

(2
.6
37

)
(3
.3
9
9)

(2
.8
4
9)

(0
.6
4
0)

(2
.6
98

)
(0
.8
61

)
(0
.7
69

)
(0
.8
6
6)

(1
.5
89

)

St
at
e
an

d
A
ge
F
E

Y
es

Ye
s

Y
es

Ye
s

R
‐s
q
ua

re
d

0
.6
18

0.
65

5
0.
48

5
0.
88

4
0.
9
08

0
.8
75

0
.8
00

0.
85

1
0
.7
01

N
=

1
01

9
47

5
54

4
1
01

9
4
75

5
44

10
19

47
5

5
44

N
o
te
s:
*
, *
*
, a
n
d
**

*
in
d
ic
at
e
si
gn
if
ic
an

ce
a
t
th
e
te
n
, f
iv
e,
a
nd

o
ne

p
er
ce
nt
le
ve
l,
re
sp
ec
ti
ve
ly
.
Th

e
sa
m
p
le
c
o
ve
rs
2
00

3‐
20

15
.

Table 4: Fixed effects model of the share of 23 and 25 year‐olds in the state who are living with parents by region

(1) (2) (3)

All Lower half grad rate Upper half grad rate

State public college tuition mean, $1000s, IPEDS 0.185 0.283 0.653*

* Northeast (0.232) (0.284) (0.344)

State public college tuition mean, $1000s, IPEDS ‐0.023 ‐0.078 0.541

* Midwest (0.283) (0.227) (0.579)

State public college tuition mean, $1000s, IPEDS ‐0.139 ‐0.418 0.246

* South Atlantic (0.326) (0.448) (0.480)

State public college tuition mean, $1000s, IPEDS 0.054 ‐0.231 ‐0.619

* West (0.197) (0.242) (0.991)

Employment to Population, QCEW ‐0.045 0.128** ‐0.077

(0.056) (0.056) (0.066)

Youth unemployment, state, CPS 0.051 0.064 0.064

(0.060) (0.078) (0.071)

Average Weekly Wage, $100s, QCEW ‐1.848** ‐3.218** ‐1.368

(0.709) (1.468) (0.853)

House Price Index, 100 in 2000, CoreLogic ‐0.011 ‐0.002 ‐0.004

(0.008) (0.011) (0.020)

Age = 23 6.946*** 6.538*** 7.258***

(0.396) (0.438) (0.656)

Year = 2005 3.739*** 4.205*** 2.179***

(0.401) (0.415) (0.652)

Year = 2006 6.362*** 6.368*** 4.517***

(0.428) (0.496) (0.824)

Year = 2007 7.736*** 8.476*** 4.594***

(0.708) (0.677) (1.492)

Year = 2008 9.228*** 10.268*** 5.540***

(0.937) (1.233) (1.689)

Year = 2009 9.227*** 10.535*** 5.076**

(1.011) (1.122) (1.970)

Year = 2010 13.962*** 15.950*** 9.587***

(1.325) (1.763) (2.031)

Year = 2011 14.455*** 16.937*** 9.498***

(1.480) (1.950) (2.209)

Year = 2012 15.623*** 18.803*** 10.140***

(1.733) (2.313) (2.821)

Year = 2013 16.282*** 19.930*** 10.303***

(1.965) (2.803) (3.101)

Year = 2014 20.774*** 24.864*** 14.132***

(2.051) (2.724) (3.333)

Year = 2015 21.953*** 26.431*** 14.632***

(2.413) (3.180) (3.918)

State FE Yes Yes Yes

R‐squared 0.882 0.905 0.862

N = 999 475 524

Notes: *, **, and *** indicate significance at the ten, five, and one percent level, respectively. The sample covers 2003‐2015.

Table 5: Fixed effects model of the share of 23 and 25 year‐olds in the state who are living with parents, urban v. rural

(1) (2) (3)

All Lower half grad rate Upper half grad rate

Percent of the state population living in ‐1.283*** ‐1.398*** ‐0.082

urban areas (0.092) (0.120) (0.101)

State public college tuition mean, $1000s, IPEDS ‐1.659*** ‐2.630*** ‐0.302

(0.552) (0.816) (0.568)

% living in urban areas * state mean tuition 0.020*** 0.028*** 0.011*

(0.005) (0.007) (0.006)

Employment to Population, QCEW ‐0.019 0.160*** ‐0.067

(0.053) (0.045) (0.067)

Youth unemployment, state, CPS 0.042 0.032 0.049

(0.053) (0.073) (0.077)

Average Weekly Wage, $100s, QCEW ‐2.453*** ‐3.822** ‐1.535*

(0.806) (1.587) (0.835)

House Price Index, 100 in 2000, CoreLogic ‐0.003 0.003 ‐0.011

(0.008) (0.010) (0.026)

Age = 23 7.026*** 6.498*** 7.259***

(0.395) (0.429) (0.552)

Year = 2005 3.720*** 4.287*** 2.181***

(0.392) (0.405) (0.537)

Year = 2006 6.193*** 6.245*** 4.602***

(0.382) (0.488) (0.783)

Year = 2007 7.699*** 8.550*** 4.797***

(0.626) (0.668) (1.040)

Year = 2008 9.345*** 10.363*** 5.879***

(0.849) (1.205) (1.173)

Year = 2009 9.463*** 10.751*** 5.420***

(0.901) (1.079) (1.211)

Year = 2010 14.391*** 16.375*** 9.852***

(1.198) (1.686) (1.348)

Year = 2011 14.912*** 17.430*** 9.708***

(1.317) (1.779) (1.425)

Year = 2012 16.234*** 19.351*** 10.288***

(1.578) (2.163) (1.927)

Year = 2013 17.109*** 20.548*** 10.450***

(1.833) (2.722) (2.016)

Year = 2014 21.486*** 25.326*** 14.250***

(1.851) (2.503) (2.041)

Year = 2015 22.658*** 26.744*** 14.726***

(2.165) (2.914) (2.519)

State and Age FE Yes Yes

R‐squared 0.884 0.906 0.864

N = 1019 475 544

Notes: *, **, and *** indicate significance at the ten, five, and one percent level, respectively. The sample covers 2003‐2015.

Table 6: County‐level flow regression of parental co‐residence, Moving in

(1) (2) (3) (4) (5) (6)

Employment to Population, county, QCEW ‐0.203*** ‐0.226*** ‐0.284*** ‐0.286*** ‐0.306*** ‐0.308***

(0.050) (0.052) (0.039) (0.040) (0.041) (0.041)

Average Weekly Wage, $1000s, QCEW ‐1.249*** ‐1.148*** ‐1.463*** ‐1.502*** ‐1.456*** ‐1.501***

(0.360) (0.364) (0.372) (0.382) (0.357) (0.365)

House Price Index, county, 100 in 2000, CoreLogi 0.043*** 0.053*** 0.068*** 0.067*** ‐0.007 ‐0.008

(0.009) (0.009) (0.009) (0.009) (0.016) (0.016)

State public college tuition mean, $1000s, IPEDS ‐ 1.255*** ‐ ‐0.166 ‐ ‐0.194

(0.207) (0.214) (0.214)

State & Age FEs Yes Yes Yes Yes Yes Yes

Linear trend No No Yes Yes No No

Year FE No No No No Yes Yes

R‐squared 0.028 0.034 0.036 0.036 0.042 0.042

N = 21,921 21,827 21,921 21,827 21,921 21,827

Mean of dependent variable 17 17 17 17 17 17

Notes: *, **, and *** indicate significance at the ten, five, and one percent level, respectively.

The sample covers 2003‐2015. Observations other than tuition are measured at the county level.

Table 7: County‐level flow regression of parental co‐residence, Moving out

(1) (2) (3) (4) (5) (6)

Employment to Population, county, QCEW 0.051 0.075 0.082 0.076 0.052 0.046

(0.101) (0.097) (0.097) (0.098) (0.102) (0.102)

Average Weekly Wage, county, $1000s, QCEW ‐3.993*** ‐4.195*** ‐4.046*** ‐3.913*** ‐4.006*** ‐3.870***

(0.662) (0.651) (0.652) (0.634) (0.671) (0.652)

House Price Index, county, 100 in 2000, CoreLogic 0.026 0.022 0.028* 0.028* ‐0.010 ‐0.009

(0.020) (0.016) (0.016) (0.016) (0.025) (0.025)

State public college tuition mean, $1000s, IPEDS ‐ ‐1.523*** ‐ 0.537 ‐ 0.535

(0.197) (0.607) (0.606)

State & Age FEs Yes Yes Yes Yes Yes Yes

Linear trend No No Yes Yes No No

Year FE No No No No Yes Yes

R‐squared 0.020 0.030 0.035 0.036 0.037 0.038

N = 18,873 18,806 18,873 18,806 18,873 18,806

Mean of dependent variable 38 38 38 38 38 38

Notes: *, **, and *** indicate significance at the ten, five, and one percent level, respectively.

The sample covers 2003‐2015. Observations other than tuition are measured at the county level.

Joint Center for Housing Studies
Harvard University

Effect of Changing Demographics on Young Adult Homeownership Rates

Rachel Bogardus Drew

February 2015

W15-2

© 2015 by Rachel Bogardus Drew. All rights reserved. Short sections of text, not to exceed two
paragraphs, may be quoted without explicit permission provided that full credit, including © notice, is
given to the source.

Any opinions expressed are those of the authors and not those of the Joint Center for Housing Studies of
Harvard University or of any of the persons or organizations providing support to the Joint Center for
Housing Studies.

Abstract: Changing socio-demographic characteristics of young adult households – those with

householders ages 25 to 34 – are having an impact on their propensities for homeownership.

Increases in the share of minority and unmarried householders are placing downward pressure

on homeownership rates for this group, while at the same time higher levels of income and

educational attainment are providing a boost. But events in housing markets over the last

twenty years have masked these effects, first by making homeownership more attractive and

attainable in the years leading up to the Great Recession, thus pushing homeownership rates

up, then by lowering them after 2005 as constraints on credit and increasingly poor economic

conditions inhibited home purchases by young adults. Untangling the combined effect of these

trends requires analyses that can decompose demographic trends from macro and micro

market conditions, to isolate the effects that specific changes in characteristics have had on

young adult homeownership rates over time. This paper describes such an analysis based on

econometric methods that estimate the expected change in homeownership over time due to

socio-demographic factors, and finds that absent the boom and bust in housing markets over

the last two decades young adults would likely have lowered their homeownership rates by

over 5 percentage points, with much of that decline caused by changes in marital and family

status. It concludes with some commentary on the implications of these findings for the

homeownership tendencies of young adults going forward.

Introduction

The housing boom and bust of the last twenty years has produced some dramatic swings in

homeownership rates. Arguably, one of the groups most affected are young adult households

(with householders ages 25 to 34), who during the boom experienced the largest increase of

any age group in homeownership rates, followed by the greatest decline during the housing

market downturn. Their homeownership rate rose from 45 percent in 1995 to 50 percent by

2005. After reaching that peak, however, the rate declined to 40 percent by 2014 (Figure 1).

Figure 1: Homeownership Rates by Age, 1995-2014

Source: Tabulations of the 1995-2014 Current Population Survey.

The primary cause of this rise and fall in young adult homeownership rates over this period was

the extraordinary conditions in the market for homes and home mortgages. The late 1990s and

early 2000s saw an unprecedented boom in the national economy, which elevated incomes and

house prices, making homeownership both more attractive and more attainable. Coupled with

innovations in mortgage markets that made financing home purchases easier for more

households, these changes encouraged more households to buy than might have otherwise

done so. As a result, homeownership rates rose to record levels by the mid-2000s. Soon after,

45%

50%

55%

60%

65%

70%

75%

80%

85%

1995 1997 1999 2001 2003 2005 2007 2009 2011 2013

65-74

55-64

45-54

35-44

25-34

however, the favorable conditions for homeownership abruptly changed as house prices

stagnated and then declined, and the worst recession since the Great Depression set off a wave

of mortgage defaults and foreclosures that caused lenders to restrict credit to potential

homebuyers. The lingering economic malaise that followed further inhibited home purchases,

especially for young adults with increasingly bleak income and employment prospects, rising

student debt levels, and wary perspectives on the wealth potential of homeownership (Fisher

and Gervais 2011; Fry 2013).

During these wild swings in housing markets, however, young adults have continued to undergo

substantial shifts in their personal characteristics that were mostly underway before the

homeownership boom, with important consequences for their propensities towards

homeownership. The share of minority young adults, for example, has increased from under 20

percent in the 1970s to 40 percent currently, mostly due to the growth of native and foreign-

born Hispanic and Asian populations. As minorities and immigrants are less likely than

comparable whites and natives to own, this shift likely contributed to a decline in

homeownership for young adults. Lower rates of marriage and family formation are also placing

downward pressure on homeownership rates, as are higher shares of young adults living in

central cities. Increasing attention to educational attainment, meanwhile, has also raised the

share of young adults who are college graduates, thereby improving their income prospects and

thus their demand for homeownership.

Previous research has suggested that the combined result of these socio-demographic shifts

has been to lower the overall homeownership rate of young adults, with economic conditions

offsetting that decline during the housing boom and exacerbating it during the bust (e.g.,

Gabriel and Rosenthal, forthcoming). Yet little attention has been paid to untangling the

individual effects of these distinct trends in young adult characteristics, or to quantifying how

much they contributed to the fall in homeownership rates over the last decade. This paper digs

deeper into this issue, by detailing how the socio-demographic characteristics of young adults

have changed over time, and then modeling the expected impact on homeownership rates of

these changes during both the boom and bust.

Young adults play a particular and important role in housing markets that warrants this

investigation of their homeownership tendencies. Despite being a minority of households, they

represent the majority of first time homebuyers (Fisher and Gervais 2011). The current and

rising generation of young adults is also poised to be the largest to come of age since the Baby

Boomers in the 1970s and ’80s (U.S. Census Bureau 2012). Not only will their numbers influence

housing demand in the near future, but they will also, with their different tastes and

preferences relative to prior generations, drive trends in the types and locations of housing that

will be built. Surveys show that young adults today increasingly prefer smaller homes close to

urban amenities, rather than the large houses in suburban and exurban locations that their

Baby Boomer parents purchased in the 1980s and 1990s (Demand Institute 2014). Given that

young adults will be called on to replace Baby Boomers as the latter start to leave the housing

market in the next few decades, the implications of young adult preferences for housing will

have significant impacts on the makeup of the housing stock for years to come (Nelson 2013).

Understanding what drives trends in their home purchasing tendencies, therefore, is important

to predicting and preparing for these changes going forward.

The paper begins by describing foundational and recent research on factors that influence

housing tenure (i.e., whether a household owns or rents), with emphasis on studies of socio-

demographic characteristics and the homeownership tendencies of young adults from the last

two decades. This is followed by a description of the data and trends demonstrating the extent

of changes in young adult characteristics over this time period. Regression models then isolate

the individual contributions of these characteristics to young adult homeownership

propensities, with a shift-share analysis to decompose their actual from expected effects on

changes in homeownership shares over time. The result of these analyses suggests that, based

on changes in socio-demographic characteristics alone, young adult homeownership rates

should have declined by over 5 percentage points from 1995 to 2014. Most of the expected

decline, moreover, is due to changes in the marital and family status of young adults over time,

as they are increasingly delaying marriage in favor of cohabitation and additional educational

attainment and career development. The final section discusses the implications for the near

future of these findings about socio-demographic forces

on young adult homeownership rates.

Literature Review

Most studies of the determinants of housing tenure outcomes in the United States consider

homeownership to be the product of combined demand for shelter as a consumption good and

for housing wealth as an investment (Henderson and Ioannides 1983; Ioannides and Rosenthal

1994; Ortalo-Magné and Rady 2002). The investment side of the equation is determined by

macro-economic and housing market conditions (e.g., user costs, price appreciation, and value

of alternative investments) and the financial condition of households (e.g., wealth and income,

risk tolerance, liquidity needs), which influence both demand for and constraints on tenure

choices (Ioannides and Rosenthal 1994; Ortalo-Magné and Rady 2002; Sinai and Souleles 2005;

Di and Liu 2007). Consumption demand, on the other hand, is assumed to be driven by personal

preferences for housing attributes that are associated with owning and renting (e.g., stability

versus mobility of residence, control versus freedom from responsibility for property). Since

such preferences are difficult to directly observe and measure, many tenure studies use socio-

demographic characteristics as proxies (Megbolugbe, Marks, and Shwartz 1991; Timmermans,

Molin, and van Noortwijk 1994; Jansen, Coolen, and Goetgeluk 2011). Traits such as

race/ethnicity, age, marital/family status, and educational attainment are assumed to represent

the lifestyle and life stage of households, which shape their needs and tastes for housing. Some

studies consider how changes in these characteristics over time predict transitions from renting

to owning and vice versa (e.g., Clark and Dieleman 1996; Clark, Deurloo, and Dieleman 2003). In

such analyses, personal financial conditions again play a role, but more as a constraint on than

as a driver of preferences.

Of primary interest for this paper are specific types of socio-demographic characteristics that

are known to correlate with tenure choice (owning versus renting), and which may help in

estimating how young adult homeownership tendencies have changed over time. The rest of

this literature review thus focuses on studies that emphasize these factors. Race and ethnicity,

for instance, are common topics of interest among tenure studies, which have consistently

identified lower rates of homeownership among minorities relative to white households. Some

of this gap is explained by differences between whites and minorities in other characteristics

such as age, income, location, and education (Wachter and Megbolugbe 1992; Coulson 1999;

Painter, Gabriel, and Myers 2001; Haurin, Herbert, and Rosenthal 2007). The lower

homeownership rates of minorities that remain, even after controlling for these differences,

are often attributed to reduced access to homeownership for minorities, rather than to lower

preferences for owning relative to similar white households (Herbert et al. 2005).

An important source of racial differences in homeownership rates is nativity status, with

immigrants generally less likely to own than native-born householders. The gaps in

homeownership rates by nativity are mitigated, however, by longer durations of residence in

the U.S., stronger command of English, financial literacy, and education—so much so that some

long-term and high achieving immigrants have been found to surpass comparable native-born

households in homeownership attainment (Myers, Megbolugbe, and Lee 1998; Coulson 1999;

Drew 2002; Haurin and Rosenthal 2009). The children of immigrants, meanwhile, all else equal,

also have high propensities for homeownership relative to children with native-born parents

(Rosenbaum and Friedman 2004).

Household life stage, indicated by the marital status and family composition of residents, also

directly influences preferences for different housing types and tenures. Married couples, for

example, are presumed to favor more (financially and residentially) stable living situations that

reflect their long-term relationship commitment, and thus to prefer to own rather than to rent

(Clark, Deurloo, and Dieleman 1994; Clark and Huang, 2003; Grinstein-Weiss et al. 2011). Dual

incomes also help increase the affordability and accessibility of homeownership for partnered

versus single-person households (Hendershott et al. 2009). The presence of children in the

household further promotes ownership, which is often associated with both larger dwellings

and locations in neighborhoods close to family-friendly amenities and better schools (Clark and

Davies Withers 2007). The positive effect on homeownership of being married and having

children also endures over the life course; divorcees and empty-nesters are more likely to own

than similar adults who have never been married or had children (Carliner 1974; Drew 2014).

One aspect of lifestyle that influences homeownership, because the relative availability of

owned and rented dwellings varies according to place, is the locational preference of

households. Living in a dense center city generally restricts ownership options, as the majority

of the housing stock located there is offered for rent, and most homes available for purchase

are condominiums in multifamily structures (Schwartz 2013). Suburbs and rural areas,

meanwhile, have fewer rental options and generally more homes for sale relative to cities.

Location also determines how affordable the housing stock is: people who choose to live in

high-cost markets may find their purchase options more constrained by their budgets than they

would in more moderately-priced parts of the country (Schwartz 2013).

Other socio-economic conditions associated with homeownership reflect the financial

resources of households, including current income and expected future income as determined

by educational attainment. High income households are more likely to own than those with

lower incomes, not only because they are better able to afford the down payment and high

transaction costs of owning, but also because they likely have higher investment demand for

real estate, as well as the means to pursue it as part of a diversified portfolio (Ioannides and

Rosenthal 1994; Ortalo-Magné and Rady 2002; Sinai and Souleles 2005). Having a high school or

college degree, meanwhile, substantially increases the expected lifetime earnings of individuals

relative to those without as much education, thus ensuring continued ability to afford

homeownership for the long term (Gyourko and Linneman 1997).

One of the most important personal characteristics related to homeownership, however, is age.

Homeownership rates tend to correlate positively with age for those under 50 years old, and

then level off before declining slightly among older seniors (Gyourko and Linneman 1997;

Gabriel and Rosenthal, forthcoming).1 The most rapid rise in homeownership rates occurs

between the early twenties and mid thirties, when most people are forming their own

households and settling into careers and lifestyles that will define them through their adult lives

(Haurin, Herbert, and Rosenthal 2007).

Indeed, several studies have focused on determinants of homeownership for young adults,

given the importance of this group to housing markets, and have confirmed that similar

demographic and financial forces determine tenure choice at this stage of life. Haurin,

Hendershott, and Kim (1994) studied the homeownership choices of individuals ages 20 to 33,

and found income to be a primary factor, along with the relative costs of owning and renting,

availability of resources for down payments, and demographic characteristics. Subsequent

analyses of the same data further found that marital status was an important indicator of

homeownership for young adults, and that married couples with two working spouses were

more likely to buy (Haurin, Hendershott, and Wachter 1996), while the presence of borrowing

constraints had a strong negative association with homeownership (Haurin, Hendershott, and

Wachter 1997). Gyourko and Linneman (1997), meanwhile, found that among young adults

with similar financial circumstances, gaps in homeownership rates by educational attainment

and race expanded between 1960 and 1990. At the same time, among households under 36,

the effect of marital and family status on homeownership was found to have decreased, as

more young adults delayed marriage and childbearing in favor of seeking further educational

and career opportunities (Gyourko and Linneman 1997).

The recent decline in homeownership rates among young adults has renewed interest in

studying this subset of households to understand their tenure decisions. Fisher and Gervais

(2011) identified two primary reasons for trends in homeownership observed between 1980

and 2000 among households with heads 25 to 44: declines in the share of such households that

are comprised of married couples, and increases in their long-term earnings risk. Gabriel and

1 Cross-sectional data on homeownership rates by age mask trends in cohort attainment of homeownership, which
suggest increasing shares of households owning homes up through age 70, and declines only among older seniors
(Masnick and Di 2001).

Rosenthal (forthcoming) look at more recent data to decompose the effects of socioeconomic

factors and market conditions on homeownership rates among households (segmented by

age), and find that personal and financial characteristics (including marital status, race, income,

educational attainment, disability status, nativity, labor market status, metro status, and three

housing market indicators) collectively contributed little to changes in homeownership rates

during the housing boom (2000-2005) and bust (2005-2009). Only income, metro area house

prices, and metro price volatility show much effect, and mostly among households in their mid

thirties and younger; specifically, having lower income, higher prices, and more volatility

decreases the likelihood of homeownership for younger households, though moreso in 2000

and 2009 than during the height of the housing boom in 2005. Gabriel and Rosenthal’s

conclusion that shifting demographic characteristics were relatively unimportant may well

simply reflect the short time frame of their analysis, over which shifts in such factors as marital

status and racial/ethnic composition would be less pronounced. In contrast, this period of

boom and bust was marked by sharp fluctuations in housing market and economic factors.

The analysis below is similar in some respects to that of Rosenthal and Gabriel (forthcoming), in

that it employs a shift-share analysis, using regression models, to separate the effects of

personal characteristics from those of market conditions on changes in homeownership rates

over the housing boom and bust period. It differs, however, in its exclusive focus on young

adults (ages 25-34), its emphasis on socio-demographic characteristics, its longer time frame for

the analysis, and its identification of specific characteristics that are driving demographically

expected shifts in tenure status over time. The following section describes these characteristics

and their trends among young adults through the housing boom and bust; a subsequent section

delves into the econometric analysis of their collective effects on homeownership rates.

Data and Descriptive Analysis

The data used in this analysis comes from the Current Population Survey’s Annual Social and

Economic (March) Supplement for the years 1995-2014 (CPS). The time frame covers the

presumed entirety of the homeownership boom and bust, as the national homeownership rate

in 2012-14 returned to its average 1960-2000 level. In addition to the wild swings in

homeownership rates observed during this period, the CPS shows some dramatic changes

occurring in the socio-demographic status of young adult households over the last twenty

years.2 This section describes some of these trends and discusses the expected effects of each

on young adult homeownership rates.

Race/Ethnicity and Nativity

One of the most striking changes in the composition of the young adult households during the

past two decades has been a substantial increase in the share of minorities, from 28 percent in

1995 to 41 percent in 2014 (Figure 2). Young Hispanic householders account for biggest

component of this growth: their share rose from 11 to 18 percent of households, with most of

these gains occurring during the boom years. Indeed, only 2.3 of the 13 percentage point

minority share gain occurred after 2005. The growth in minority households among young

adults has been caused in part by rising immigration rates, since many immigrants arrive in the

U.S. during their twenties. The share of foreign-born among 25-34 year old household heads

grew from 12 percent in the mid-1990s to 19 percent a decade later, before falling back slightly

during the recession to 17 percent.

2 All data presented in this section is based on the author’s calculations of weighted counts and shares of households
from the CPS, and may not match published tables available from the U. S. Census Bureau.

Figure 2: Minority and Foreign-born Shares of Young Adult Households, 1995-2014

Note: White young adult households are not shown in the chart and comprise the remainder of the distribution in
each year. Whites, Blacks, and Others are non-Hispanic, while Hispanics may be of any race. Other includes Asians
and multi-racial categories.
Source: Tabulations of the 1995-2014 Current Population Survey.

Given that non-white and foreign-born households generally have lower homeownership rates

than whites and native-born householders, shifts in the distribution of young adult households

towards the former likely served to depress overall homeownership rates for the age group as a

whole during this period. Yet minority and immigrant homeownership rates were also rising

during the housing boom (Herbert et al. 2005), potentially offsetting some of the dampening

effect of their share increases on young adult homeownership. Since the collapse of the

housing market, however, minority homeownership rates have declined more than those of

whites3 (U.S. Census Bureau 2014), and may have exacerbated the downward pressure placed

on homeownership rates by the increase in the minority share of young adult households since

2005.

3 Non-Hispanic white homeownership rates declined 3.5 percentage points from 2005-2014, versus 5.6 percentage
points for non-Hispanic Blacks, 3.3 percentage points for Hispanics, and 4.2 percentage points for non-Hispanic
Asian/Others, according to the Current Population Survey.

Income

The household income of young adults over the period analyzed here, when adjusted for

inflation and measured in constant categories, also underwent some noticeable changes. In

particular, the share of young adult households in the highest category, i.e., with real incomes

over $75,000 (expressed in 2014 dollars), increased from 28 percent in 1995 to 37 percent by

2002 (Figure 3). This gain was offset by declines in the share in the lowest income category,

under $25,000, from 21 to 15 percent. Since 2001, however, the low-income category has been

gaining share, and as of 2014 included 22 percent of young adult households.

Figure 3: Income Distribution of Young Adult Households (2014$), 1995-2014

Note: Incomes are expressed in 2014 dollars and adjusted for inflation by the CPI-UX for all consumers from the
Bureau of Labor Statistics.
Source: Tabulations of the 1995-2014 Current Population Survey.

Higher incomes are strongly associated with higher propensities for homeownership, so the

gains of young adults during the late 1990s likely increased their probability of owning homes

during the boom, even though the share in the highest income group declined slightly in the

early 2000s. Likewise, the rising share of low-income households subsequent to the housing

boom probably had a depressing effect on the homeownership rate of young adults in the more

10%

15%

20%

25%

30%

35%

40%

<$25K

$25-50K

$50-75K

$75K+

recent period.

Marital/Family Status

Another notable trend among young adults has been declines in the share that are married

and/or living with minor children. Like the increasing minority share, this shift is actually a

continuation of trends that date back to the 1970s. But while the rate of growth among non-

white households has slowed, the increase in unmarried and childless households has

accelerated in recent years, possibly in response to the recession (Cherlin et al. 2013). The

share of young adult householders living with a spouse, which declined from 60 percent in the

early 1980s, actually held steady at around 50 percent through most of the housing boom,

before falling to 42 percent by 2014 (Figure 4a). Most of this decline was offset by rising shares

of unmarried partner households, from 5 to 13 percent between 1995 and 2014, who despite

their coupled status are more like single young adults in their home buying behavior. The share

of young adults with children has also decreased over the last two decades, from around 54

percent to 48 percent (Figure 4b).

Figure 4: Marital and Parental Status of Young Adults, 1995-2014
a)

10%

20%

30%

40%

50%

60%
19

19
96

19
97

19
98

19
99

20
00

20
01

20
02

20
03

20
04

20
05

20
06

20
07

20
08

20
09

20
10

20
11

20
12

20
13

20
14

Married

Partner

Single

Multi

Note: Married couples include married with spouse absent, but not separated. Single-adult households include
single parents with no other adults living in the residence. Multi-adult households are households with more than
one adult who is not the spouse or unmarried partner of the householder (e.g., roommates or other family
members). Households with children include only those with the householders’ own natural, adopted, foster, or
step-child(ren).
Source: Tabulations of the 1995-2014 Current Population Survey.

The likely effect of these shifts in marital and family status has been to lower homeownership

rates among young adults, as married couples and parents tend to have much higher

propensities towards owning than unmarried and childless households. Indeed, recent research

suggests as much as half the decrease in young adult homeownership since 1980 is the result of

lower rates of marriage and family formation (Fisher and Gervais 2011). Lower marriage rates

among young adults also reduce the likelihood of some unmarried householders owning homes

as a result of their prior marital status, i.e., remaining homeowners following a divorce,

separation, or death of a spouse with whom they lived in an owned dwelling. Indeed, the

reduction in the share of young adults that have ever been married, from 70 percent in 1995 to

54 percent in 2014 (Figure 4b), is even more dramatic than the decline in the share of young

adults who are parents. Both of these shifts further reduce the likelihood of young adults

owning homes today relative to the past.

Educational Attainment

A potentially positive force on young adult homeownership rates over the last two decades has

been the growth in post-secondary educational attainment among this age group. Almost half

of all young adult-headed households now include someone with a college degree, up from 32

percent in the mid-1990s (Figure 5).4 Nearly all this growth is due to an increase in the

percentage of high school graduates who attain college degrees; the overall percentage of high

school graduates (i.e. the inverse of the share without a high-school diploma), meanwhile, has

increased only slightly..

Figure 5: Maximum Educational Attainment among Residents in Young Adult Households

Source: Tabulations of the 1995-2014 Current Population Survey.

Though the time it takes to pursue a college education may delay entry into the workforce and

the achievement of financial and residential stability associated with home purchases, young

college graduates still tend to have higher homeownership rates than their less educated peers.

4 Note that these data reflect the highest educational attainment achieved by all adults in the household, rather than
just the householder, since more education of household members tends to elevate household income, regardless of
whether it is the head that holds the degree. As a result, the share of young households with a college graduate is
larger than the share of all such graduates among the young adult population (35 percent in 2014).

10%

20%

30%

40%

50%

60%

70%

19
95

19
96

19
97

19
98

19
99

20
00

20
01

20
02

20
03

20
04

20
05

20
06

20
07

20
08

20
09

20
10

20
11

20
12

20
13

20
14

College Graduate

High School
Diploma only

No High School
Diploma

Indeed, Gyourko and Linneman (1997) point to the increasing importance of educational

attainment in predicting homeownership outcomes for young adults, which suggests that an

increase in the share of households with college graduates should have a positive effect on the

homeownership rates of young adult households overall.

Over the last two decades, other noteworthy changes in the socio-demographic characteristics

of young adult households include small increases in the following shares: those living in central

cities, those with female household heads, and those in the bottom half of the age range (25-29

years old) (see Appendix A). All of these changes likely have a slight depressing effect on the

homeownership rates of young adults.

Econometric Analysis

Regression Models

To decompose the simultaneous effects of these shifts in socio-demographic characteristics on

young adult homeownership rates, regression analyses were run on the tenure status of

householders ages 25 to 34 controlling for their race, nativity, education, marital and family

status, central city location, income distribution, and age. An additional variable is included in

the analysis for the estimated median monthly mortgage principle and interest cost, which is

calculated from local median house prices in the year prior to the analysis and median

prevailing interest rates for a 30-year fixed rate mortgage, assuming a 10 percent down

payment.5 All of the socio-demographic characteristic variables are expressed as binary

indicators, with respondents who have a specified characteristic assigned a value of one for

that variable, and a value of zero otherwise. All categorical variables (race, marital status,

income, and education) have one characteristic excluded from the models as a reference

category (see Appendix A). While regressions with binary dependent variables are traditionally

estimated with a non-linear model, this analysis uses ordinary least squares (OLS) to facilitate

5 Local prices are the median price for the metro area (or state, if metro area is not available or identified) of the
household as reported by the National Association of Realtors® (NAR). Approximately 70-80 percent of
observations in the three years modeled for the analysis were in an identified metro area matched to NAR data.
Estimates of monthly owner costs do not include taxes or insurance payments that are often added to mortgage costs.

interpretation and post-estimation calculations of model coefficients.6

The regressions are performed for three separate years of weighted CPS data: 1995,

representing a period before the housing boom; 2005, at the height of the boom; and 2014,

after the worst of the housing market downturn and the most recent year for which data are

available. The coefficients represent the estimated difference in the homeownership rate for

young adult households who have each socio-demographic characteristic relative to those in

the reference category, all else equal (see Appendix B). For the local monthly mortgage cost,

the coefficient is the estimated difference in homeownership rates from a $1 increase in the

monthly cost of owning a home. As expected, in all three years modeled, the homeownership

rate is lower (i.e., coefficients are negative) for unmarried, minority, female, foreign-born, and

central city households, and higher (i.e., coefficients are positive) for higher income, higher

educated, older, parents, and previously married householders, when all other characteristics

are controlled for. The coefficients also suggest a negative relationship between the monthly

cost of owning a home and homeownership rates, as households are less likely to own when it

is more expensive to do so. In the three models, all variables were statistically significant at the

0.1 percent

level.

Looking across the three models, there are some trends in the coefficients that warrant

mention. The coefficients for marital status, for example, increased in absolute value during the

housing boom, reflecting an increasing importance of marital status differences for predicting

homeownership among young adults. The subsequent decrease in the aftermath of the boom

suggests that, in the current environment, marital status is becoming less relevant to tenure.

The opposite, however, is true of race/ethnicity, which had coefficients closer to zero (i.e.,

smaller differences in homeownership rates) during the boom; recently, these coefficients have

been rising in absolute value. The effect of income distributions on homeownership appears to

have declined consistently throughout the study period, with the exception of the coefficient

for those earning between $25,000 and $50,000 a year, which increased between 1995 and

6 The results derived from the OLS models are similar in magnitude and significance to those produced by
estimating marginal effects after a more traditional binary probit regression.

2005. Education, meanwhile, was associated with larger differences in homeownership rates

during the boom than before, and smaller differences currently. Metro status and nativity are

both associated with differences in home ownership rates during the boom that remained in

effect after the market turned, while being female and in the older half of the age group

became less relevant over time. Finally, the coefficients for local monthly mortgage costs

declined slightly between 1995 and 2005, suggesting a decreasing influence of higher costs as a

deterrent to homeownership at the peak of the housing boom, followed by a sharp increase in

2014. All differences in coefficients across models are statistically significant at the 5 percent

level.

A final comment on the results of the regression models concerns their explanatory power,

which is measured by the adjusted R-square as a proxy for the amount of variation in tenure

status that is explained by the variables. The combined effect of the variables included in the

models declined slightly from 0.27 in 1995 to 0.24 in 2014. This suggests that only around a

quarter of young adults’ homeownership rate is predicted by their personal characteristics and

local monthly owner costs, and that the role of other forces (e.g., attitudes towards

homeownership and macro-economic factors) has increased somewhat over time. Thus

regardless of the influence of individual socio-demographic factors, collectively these conditions

remain less relevant to homeownership than market factors and unobserved drivers of tenure

choices.

Shift-Share Analysis

An advantage of the OLS models estimated to assess socio-demographic effects on

homeownership is that the product of model coefficients and mean variable values when

summed over all independent variables in the model, plus the constant is equal to the mean

value of the dependent variable. The mean values of all independent variables in the three

models described above are shown in Appendix A. When multiplied by the coefficients in

Appendix B, the result is the mean value of the tenure choice indicator, i.e., the sample

homeownership rate. For the three years modeled in the regressions above, these

homeownership rates were 44.7, 49.9, and 40.1 percent, respectively.

A shift-share analysis of the three regression models offers the opportunity to examine the

effect of changes in means versus coefficients. That is, the product of the mean values in one

year and the coefficients in another allows us to separate changes in the homeownership rate

resulting from shifts in the characteristics of young adults from changes in the propensities of

each characteristic for predicting tenure status. This form of shift-share analysis is

demonstrated in Appendix C, which first applies coefficients from the 1995 model to

distributions observed in 2005, to see what the expected homeownership rate for young adults

would have been if only the mean values of the socio-demographic variables, i.e., the

proportions of the sample that exhibited each of these characteristics, changed during the

decade. The result is a homeownership rate of 40.6 percent, or 4.1 percentage points less than

the actual 1995 rate. The combined effect of shifts in socio-demographic characteristics and

changes in local monthly owner costs thus was expected to lower young adult homeownership

rates by this amount, while in reality the rate rose by 5.2 percentage points, due mostly to

favorable market and economic conditions for home buying and positive views about

homeownership.

The second shift-share applies 2005 coefficients to 2014 distributions, to estimate the effect of

changes in socio-demographic characteristics over this period on the homeownership rate of

young adults. The result of this calculation is an expected homeownership rate of 51 percent,

which is 1.1 percentage points above the observed rate of 49.9 percent as of 2005. This

expected rate is heavily influenced by the relative increase in affordability of homeownership

that occurred during this period, as house prices declined while interest rates reached historic

lows. Absent this decline in monthly mortgage costs, the expected homeownership rate based

on socio-demographic characteristics alone should have been lower than the observed rate, by

around 1.7 percentage points. This suggests less impact from socio-demographic changes in the

second decade of the study period relative to the first, a decrease that is consistent with the

slowing growth of the minority share and the slight decline in the foreign-born share among

young adults between 2005 and 2014. This period also saw some increase in homeownership-

positive factors, such as share of young adult households with college graduates. This analysis

also reaffirms the primary role of macro-economic and market factors on homeownership,

which at the time were considerably less conducive to home purchases among young adults,

thus lowering the actual young adult homeownership rate to 40.4 percent as of 2014.

The third shift-share calculation spans the nearly two decades covered by this analysis,

combining 1995 coefficients with 2014 distributions. The result is an expected homeownership

rate for young adults of 41.9 percent; this suggests that the combined effect of all socio-

demographic changes that occurred within the young adult population over the last nineteen

years, along with changes in local monthly owner cost estimates, should have lowered the

young adult homeownership rate by 2.8 percentage points from its 1995 level (Figure 6). This

expected rate of 41.9 percent is also slightly higher than the observed 2014 homeownership

rate of 40.1 percent, suggesting that more young adults would be owners now than actually are

if homeownership tendencies from 1995 still prevailed. This result, however, is skewed by the

relative affordability of owning today versus in the mid-1990s. Subtracting out the positive 2.2

percentage point influence of local monthly owner costs, the socio-demographic effect alone

should have lowered young adult homeownership rates by 5 percentage points, bringing it

nearly in line with its current level.

Figure 6: Actual vs. Expected Homeownership Rates of Young Adults, 1995-2014

Note: Expected homeownership rates (i.e., the first and third shift-share calculations in Appendix C) are the
product of 1995 regression model coefficients and distributions of young adult characteristics (i.e., mean values of
binary variables) in indicated years, summed over all characteristics in the model, plus the constant term.

The shift-share analysis further allows for the decomposition of differences between expected

and actual homeownership rates according to the variables included in the models. To isolate

the effect of changes in a particular socio-demographic characteristic, the product of the

distribution and coefficient for a given variable in one year is subtracted from the product of

the same coefficient and the distribution for that variable in a later year. In the first shift-share

analysis, for example, holding coefficients constant at 1995 estimates shows that over a third of

the expected 4.2 percentage point decrease in the homeownership rate between 1995 and

2005 was due to changes in local monthly mortgage costs, while much of the remaining decline

was attributable to shifts between these years in the racial/ethnic and nativity status of young

adults, which each contributed about six-tenths of a percentage point to that decrease.

Changes in marital and living statuses of young adults over this period, including the increase in

never-married householders, added an additional percentage point to the expected decline,

while the increase in higher income young adult households actually added 1.2 percentage

35%

40%

45%

50%

55%

1995 2005 2014

Observed
homeownership rate of
young adults

Expected HO rate based
on 1995 coefficients

points to the expected homeownership rate (Figure 7a).

The second shift-share calculation shows that changes in estimated monthly mortgage costs

were wholly responsible for expected increases in homeownership rates between 2005 and

2014; absent these dramatic swings in house prices and interest rates, shifts in socio-

demographic characteristics would have lowered homeownership rates. The most important

factors in the expected demographically-induced decline were shifts in marital and living

statuses, as the decrease in the share of married households accelerated during the housing

downturn and recession (Figure 7b). Growth in the share of lower income households also

placed downward pressure on homeownership during this period. Small changes in the

racial/ethnic and nativity distributions of young adults, meanwhile, had only a negligible effect,

while changes in educational attainment among young adults were actually expected to

increase homeownership by over one-half a percentage point. For the whole nineteen-year

period (i.e., the third shift-share calculation), marital status shifts were still the largest socio-

demographic driver of expected homeownership changes between 1995 and 2014 (Figure 7c).

Race and nativity combined added another 1.1 percentage points to the expected decline,

while higher shares of young adults living in center cities contributed 0.7 percentage points.

Figure 7: Variable Contributions to Expected Decline in the Homeownership Rate of Young
Adults, 1995-2014
a)

These results make clear several important facts about the role of socio-demographic factors in

shaping recent homeownership rates among young adults. First, absent the dramatic swings in

housing markets and macro-economic conditions, we would have expected young adult

homeownership rates to be similar to what they actually are. Most of that decline, moreover,

would have occurred during the 1990s and early 2000s, when the racial/ethnic, nativity,

gender, and age distribution of householders age 25 to 34 were shifting more rapidly. Second,

the additional decline that was expected to take place after 2005 was mostly driven by changes

in the marital status and living arrangements of young adults, which were themselves

potentially influenced by the declining economy of that period. Third, however, given the low

R-square values of the regression analysis, and the actual rise and fall in homeownership rates

for young adults over the last twenty years, it is clear that personal characteristics play a

somewhat limited role in determining the tenure status of young households. In contrast,

market factors account for around three-quarters of the variation in observed tenure status,

with that share growing slightly over the period examined in this analysis.

Discussion and Implications

The analysis above reveals informative details about the drivers of tenure choices among young

adults over the past two decades, and in particular about the effects of demographic changes in

this population. It does not, however, provide insight into the future tenure status of young

households. Such a prediction would require estimates of the number and composition of

young adult households going forward. The Joint Center for Housing Studies (JCHS) recently

revised their household projections using current Census population projections (McCue 2014).

According to their calculations, the number of households with heads ages 25 to 34 is expected

to increase by 1 million between 2015 and 2035. The JCHS projections further segment

expected change in households by race and marital/family composition, using estimates of the

existing population by age and race, expected immigration flows, and current rates of marriage

and childbearing. These estimates thus make some assumptions about future trends in socio-

demographic characteristics of young adults that do not consider the effects of potential

economic conditions or other external shocks on the composition of households. Still, these

projections offer some basis for discussing how further changes in socio-demographic

characteristics of young adults may impact their homeownership rates.

All of the projected growth in young adult households, according to the JCHS calculations, is

expected to be among minority heads, who will increase their share of all young households to

fully 50 percent by 2035 (McCue 2014). This ten percentage point shift in the racial/ethnic

distribution of young adult households over a twenty year period reflects a continuation of the

trends observed over the past two decades, which, as the analysis above shows, contributed

over a full percentage point to declines in the young adult homeownership rate. That decline,

however, was calculated after taking into account changes in the marital, educational, income,

and locational distributions of young adults, which are also affected by the minority share of

households. Specifically, increases in the minority share of young adult households are likely to

further decrease the share of these households comprised of married couples, those having

college educations, and those living outside central cities; all of these decreases in turn place

further downward pressure on the homeownership rate of this population. The full impact of

higher minority shares on young adult homeownership rates is therefore difficult to foresee,

and may actually be even greater than one percentage point.

It is important to note that these projections are based on recent estimates of headship rates

for young adults, i.e., the share of individuals heading their own household, which may not

remain at their current level going forward. A rise in the headship rate could occur if the

economy, and particularly the job prospects for young adults, improves greatly in the near

future. The JCHS’s projections also do not account for other socio-demographic trends known

to influence homeownership rates, such as the gender distribution of householders, the

increasing appeal of urban living, and higher educational attainment among young adults. Any

distributional changes in these socio-demographic characteristics of young adults will have

further implications for their homeownership rates.

With continued socio-demographic changes likely to have further depressing effects on

homeownership rates among young adult households, it will be up to the economy and housing

markets to offer countervailing forces to encourage young adults to buy homes. As the analysis

in this paper shows, the effect of favorable mortgage terms, affordable housing costs, and

increases in income can be stronger drivers of tenure outcomes than socio-demographic

characteristics, as evidenced during the housing boom. When both characteristics and

economic conditions are less favorable to home purchases, however, young adult

homeownership rates can fall precipitously, as happened after the collapse of the housing

market in 2005.

Conclusion

The dramatic rise and fall in young adult homeownership rates observed over the past two

decades has largely been a function of economic and market conditions, with a smaller role

played by changing socio-demographic characteristics. Indeed, increases in the shares of young

adult households that are minority, unmarried, and living in center cities should have lowered

their homeownership rate during the 1990s and early 2000s; instead, favorable lending

conditions and enthusiasm for homeownership increased the homeownership rate among

householders ages 25 to 34. In the subsequent decade, after the peak of the housing boom, the

homeownership rate for this group fell precipitously, even as changes in their demographic

characteristics moderated relative to the prior decade.

Among the characteristics shown by this study to have the greatest effect on young adult

homeownership rates, the decline in the share of married couples (both currently and formerly

married) had the largest impact, accounting for 40 percent of the expected change in

homeownership due to socio-demographic shifts. Most of this effect occurred during the

downturn, when economic conditions may have themselves been inhibiting marriages and

family formation. Changes in the race and nativity of young adults, meanwhile, have slowed in

recent years, reducing the effect of these factors on homeownership. The future of these

trends is uncertain, but if recent experience is any guide, they will continue to place downward

pressure on young adult homeownership rates in the near term.

Appendix A: Mean Values of Variables used in Regression Analysis

1995 2005 2014
Number of households (000s) 19,474 19,331 20,033

Marital/ Living Status

Married Couple* 0.530 0.494 0.423
Partnered Couple 0.054 0.088 0.129
Single Adult 0.308 0.312 0.326
Multi Adult 0.108 0.106 0.123

Race/ Ethnicity

Non-Hispanic White* 0.716 0.621 0.598
Non-Hispanic Black 0.136 0.137 0.131
Hispanic 0.115 0.171 0.181
Non-Hispanic Other 0.033 0.071 0.089

Income Categories
(2014$)

Under $25,000* 0.214 0.183 0.221
$25,000-$50,000 0.274 0.275 0.256
$50,000-$75,000 0.229 0.214 0.196
$75,000 or more 0.283 0.327 0.327

Max Educational
Attainment in
Household

No Degree* 0.082 0.083 0.059
High-School Degree 0.598 0.524 0.483
College Degree 0.321 0.393 0.458

Presence of Children
in Household

No* 0.446 0.459 0.478
Yes 0.554 0.541 0.523

Gender
Male* 0.635 0.517 0.509
Female 0.365 0.483 0.491

Age Category 25-29 years old* 0.432 0.475 0.468
30-34 years old 0.568 0.525 0.533

Metro Status Non-Central City* 0.723 0.677 0.651
Central City 0.277 0.323 0.349

Prior Marital Status
Never Married* 0.308 0.377 0.465
Ever Married 0.693 0.623 0.535

Nativity Native-born* 0.879 0.809 0.824
Foreign-born 0.121 0.191 0.176

Local Monthly Owner Cost (2014$) $1,235 $1,432 $953
Homeownership Rate (Dependent Variable) 44.7% 49.9% 40.1%

Notes: Counts and distributions are calculated using weighted CPS data. Variables indicated
with an asterisk (*) are designated reference categories and thus excluded from the regression
analyses.

Appendix B: Results of Regression Analyses on Tenure Status for Young Adult Households

1995 2005 2014
Number of Observations (unweighted) 11,122 13,479 8,761

Marital/ Living Status

Married Couple* – – –
Partnered Couple -0.180 -0.218 -0.176
Single Adult -0.132 -0.184 -0.131
Multi Adult -0.155 -0.165 -0.124

Race/ Ethnicity

Non-Hispanic White* – – –
Non-Hispanic Black -0.152 -0.091 -0.133
Hispanic -0.078 -0.021 -0.025
Non-Hispanic Other -0.033 -0.044 -0.008

Income Categories
(2014$)

Under $25,000* – – –
$25,000-$50,000 0.092 0.100 0.091
$50,000-$75,000 0.232 0.216 0.183
$75,000 or more 0.352 0.334 0.306

Max Educational
Attainment in
Household

No Degree* – – –
High-School Degree 0.048 0.069 0.009
College Degree 0.034 0.132 0.072

Presence of Children
in Household

No* – – –
Yes 0.068 0.037 0.089

Gender
Male* – – –
Female -0.040 -0.008 -0.017

Age Category 25-29 years old* – – –
30-34 years old 0.118 0.106 0.067

Metro Status Non-Central City* – – –
Central City -0.096 -0.125 -0.123

Prior Marital Status
Never Married* – – –
Ever Married 0.053 0.012 0.017

Nativity Native-born* – – –
Foreign-born -0.088 -0.109 -0.111

Local Monthly Owner Cost (2014$) -0.00008 -0.00006 -0.00012
Constant 0.334 0.408 0.405
Adjusted R-Square 0.2684 0.2613 0.2372

Note: The models were all run using weighted data. All regression coefficients are statistically
significant at the 0.1 percent level. Variables indicated with an asterisk (*) are designated
reference categories and thus excluded from the regression analyses.

Appendix C: Shift-Share Analysis of Regression Results on Tenure Status for Young Adult Households
Shift 1: 1995-2005 Shift 2: 2005-2014 Shift 3:1995-2014

Coef95 x
Mean05

Diff.
from
1995

Total
Percentage
Point Diff.

Coef05 x
Mean14

Diff. from
2005

Total
Percentage
Point Diff.

Coef95 x
Mean14

Diff. from
1995

Total
Percentage
Point Diff.

Marital/ Living Status

Married Couple*

0.0000 0.0000

-0.63%

0.0000 0.0000

-1.43%

0.0000 0.0000

-1.81%

Partnered Couple -0.0158 -0.0061 -0.0281 -0.0089 -0.0232 -0.0134
Single Adult -0.0410 -0.0005 -0.0601 -0.0026 -0.0429 -0.0023
Multi Adult -0.0165 0.0002 -0.0203 -0.0027 -0.0190 -0.0024

Race/ Ethnicity

Non-Hispanic White* 0.0000 0.0000

-0.57%

0.0000 0.0000

-0.05%

0.0000 0.0000

-0.62%

Non-Hispanic Black -0.0208 -0.0001 -0.0120 0.0005 -0.0200 0.0007
Hispanic -0.0133 -0.0044 -0.0038 -0.0002 -0.0141 -0.0051
Non-Hispanic Other -0.0023 -0.0012 -0.0039 -0.0008 -0.0029 -0.0018

Income Categories (2014$)

Under $25,000* 0.0000 0.0000

1.22%

0.0000 0.0000

-0.58%

0.0000 0.0000

0.62%

$25,000-$50,000 0.0253 0.0001 0.0255 -0.0019 0.0236 -0.0017
$50,000-$75,000 0.0497 -0.0034 0.0424 -0.0039 0.0455 -0.0076
$75,000 or more 0.1151 0.0155 0.1093 0.0000 0.1151 0.0155

Max Educational Attainment
in Household

No Degree* 0.0000 0.0000

-0.11%

0.0000 0.0000

0.57%

0.0000 0.0000

-0.09%

High School Degree 0.0251 -0.0035 0.0333 -0.0028 0.0231 -0.0055
College Degree 0.0132 0.0024 0.0605 0.0086 0.0154 0.0046

Presence of Children in
Household

No* 0.0000 0.0000
-0.09%

0.0000 0.0000
-0.07%

0.0000 0.0000
-0.22% Yes 0.0367 -0.0009 0.0192 -0.0007 0.0355 -0.0022

Gender
Male* 0.0000 0.0000

-0.47%
0.0000 0.0000

-0.01%
0.0000 0.0000

-0.50% Female -0.0192 -0.0047 -0.0041 -0.0001 -0.0195 -0.0050

Age Category
25-29 years old* 0.0000 0.0000

-0.51%
0.0000 0.0000

0.08%
0.0000 0.0000

-0.42% 30-34 years old 0.0619 -0.0051 0.0562 0.0008 0.0628 -0.0042

Metro Status
Non-Central City* 0.0000 0.0000

-0.44%
0.0000 0.0000

-0.33%
0.0000 0.0000

-0.69% Central City -0.0310 -0.0044 -0.0437 -0.0033 -0.0336 -0.0069

Prior Marital Status
Never Married* 0.0000 0.0000

-0.37%
0.0000 0.0000

-0.11%
0.0000 0.0000

-0.84% Ever Married 0.0330 -0.0037 0.0065 -0.0011 0.0283 -0.0084

Nativity
Native-born* 0.0000 0.0000

-0.62%
0.0000 0.0000

0.16%
0.0000 0.0000

-0.49% Foreign-born -0.0169 -0.0062 -0.0191 0.0016 -0.0156 -0.0049
Local Monthly Owner Cost (2014$) -0.1112 -0.0152 -1.52% -0.0565 0.0284 2.84% -0.0740 0.0220 2.20%
Constant 0.3345 0.0000 0.4085 0.0000 0.3345 0.0000
Expected Homeownership Rate 40.6% -4.1% 51.0% 1.1% 41.9% -2.9%

References

Carliner, G. 1974. Determinants of homeownership. Land Economics 50, no. 2: 110-19.

Cherlin, A., Cumberworth, E., Morgan, S. P., and Wimer, C. 2013. The effects of the great

recession on family structure and fertility. The ANNALS of the American Academy of
Political and Social Science 650, no. 1: 214-31.

Clark, W.A.V., and Davies Withers, S. 2007. Family migration and mobility sequences in the

United States: Spatial mobility in the context of the life course. Demographic Research
17, no. 20: 591-622.

Clark, W. A. V., Deurloo, M. C., and Dieleman, F. M. 1994. Tenure changes in the context of

micro-level family and macro-level economic shifts. Urban Studies 31, no. 1: 137-54.

Clark, W. A. V., Deurloo, M. C., and Dieleman, F. M. 2003. Housing careers in the United States:

Modeling the sequencing of housing states. Urban Studies 40, no. 1: 143-60.

Clark, W. A. V., and Dieleman, F. M. 1996. Households and housing: Choice and outcomes in the

housing market. New Brunswick, NJ: Center for Urban Policy Research, Rutgers
University.

Clark, W. A. V., and Huang, Y. 2003. The life course and residential mobility in British housing

markets. Environment and Planning A 35: 323–39.

Coulson, N. E. 1999. Why are Hispanic and Asian American homeownership rates so low?

Immigration and other factors. Journal of Urban Economics 45, no. 2: 209-27.

Demand Institute. 2014. Millennials and their homes: Still seeking the American Dream. New

York: Demand Institute.

Di, Z. X., and Liu, X. 2007. The importance of wealth and income in the transition to

homeownership. Cityscape 9, no. 2: 137-52.

Drew, R. B. 2002. New Americans, new homeowners: The role and relevance of foreign-born

first-time homebuyers in the U.S. housing market. Research Note N02-2. Cambridge,
MA: Joint Center for Housing Studies, Harvard University.

Drew, R. B. 2014. Three facts about marriage and homeownership. Housing perspectives (Joint

Center for Housing Studies blog), December 17
(http://housingperspectives.blogspot.com/2014/12/3-facts-about-marriage-and-
homeownership.html).

Fisher, J. D. M., and Gervais, M. 2011. Why has homeownership fallen among the young?
International Economic Review 52, no. 3: 883-912.

Fry, R. 2013. Young adults after the recession: Fewer homes, fewer cars, less debt. Washington:

Pew Research Center.

Gabriel, S. A., and Rosenthal, S. S. Forthcoming. The boom, the bust, and the future of

homeownership. Real Estate Economics (pre-publication version available at
http://ssrn.com/abstract=2323889).

.
Grinstein-Weiss, M., Charles, P., Guo, S., Manturuk, K., and Key, C. 2011. The effect of marital

status on home ownership among low-income households. Social Service Review 85, no.
3: 475-503.

Gyourko, J., and Linneman, P. 1997. Analysis of the changing influences on traditional

households’ ownership patterns. Journal of Urban Economics 39, no. 3: 318-41.

Haurin, D. R., Hendershott, P. H., and Kim, D. 1994. Housing choices of American youth. Journal

of Urban Economics 35, no. 1: 28–45.

Haurin, D. R., Hendershott, P. H., and Wachter, S. 1996. Wealth accumulation and housing

choices of young households: An exploratory investigation. Journal of Housing Research
7, no. 1: 33-57.

Haurin, D. R., Hendershott, P. H., and Wachter, S. 1997. Borrowing constraints and the tenure

choice of young households. Journal of Housing Research 8, no. 2: 137-54.

Haurin, D. R., Herbert, C. E., and Rosenthal, S. S. 2007. Homeownership gaps among low-income

and minority households. Cityscape 9, no. 2: 5–51.

Haurin, D. R., and Rosenthal, S. S. 2009. Language, agglomeration, and Hispanic

homeownership. Real Estate Economics 37, no. 2: 155-183.

Hendershott, P. H., Ong, R., Wood, G. A., and Flatau, P. 2009. Marital history and home

ownership: Evidence from Australia. Journal of Housing Economics 18, no. 1: 13-24.

Henderson, J. V., and Ioannides, Y. M. 1983. A model of housing tenure choice. American

Economic Review 73, no. 1: 98-113.

Herbert, C. E., Haurin, D. R., Rosenthal, S. S., and Duda, M. 2005. Homeownership gaps among

low income and minority borrowers and neighborhoods. Washington: U.S. Department
of Housing and Urban Development.

Ioannides, Y. M., and Rosenthal, S. S. 1994. Estimating the consumption and investment

demands for housing and their effect on housing tenure status. Review of Economics
and Statistics 76, no. 1: 127–41.

Jansen, S. J. T., Coolen, H., and Goetgeluk, R. W. 2011. Measurement and analysis of housing

preference and choices. New York: Springer.

Masnick, G. S., and Di, Z. X. 2001. Cohort insights into the influence of education, race and

family structure on homeownership trends by age: 1985 to 1995. Research Note N01-1.
Cambridge, MA: Joint Center for Housing Studies, Harvard University.

McCue, D. 2014. Baseline household projections for the next decade and beyond. Working

Paper 14-1. Cambridge, MA: Joint Center for Housing Studies, Harvard University.

Megbolugbe, I. F., Marks, A. P., and Schwartz, M. B. 1991. The economic theory of housing

demand: A critical review. Journal of Real Estate Research 6, no. 3: 381-93.

Myers, D., Megbolugbe, I., and Lee, S. 1998. Cohort estimation of homeownership attainment

among native-born and immigrant populations. Journal of Housing Research 9, no. 2:
237-69.

Nelson, A. C. 2013. Reshaping metropolitan America: Development trends and opportunities to

2030. Washington: Island Press.

Ortalo-Magné, F., and Rady, S. 2002. Tenure choice and the riskiness of non-housing

consumption. Journal of Housing Economics 11, no. 3: 266-79.

Painter, G., Gabriel, S. A., and Myers, D. 2001. Race, immigrant status, and housing tenure

choice. Journal of Urban Economics 49: 150-67.

Rosenbaum, E., and Friedman, S. 2004. Generational patterns in home ownership and housing

quality among racial/ethnic groups in New York City, 1999. International Migration
Review 38, no. 4: 1492-533.

Schwartz, A. F. 2013. Housing policy in the United States: An introduction. 3rd ed. New York:

Routledge.

Sinai, T., and Souleles, N. S. 2005. Owner-occupied housing as a hedge against rent risk. The

Quarterly Journal of Economics 120, no. 2: 763-89.

Timmermans, H., Molin, E., and van Noortwijk, L. 1994. Housing choice processes: Stated versus

revealed modeling approaches. Netherlands Journal of Housing and the Built
Environment 9, no. 3: 215-27.

U.S. Census Bureau 2012. The baby boom cohort in the United States: 2012 to 2060 –
population estimates and projections. Current Population Reports P25-1141.
Washington.

U.S. Census Bureau 2014. Housing vacancy survey, third quarter. Washington.

Wachter, S. M., and Megbolugbe, I. F. 1992. Racial and ethnic disparities in homeownership.