Renting to Owning: An Exploration of the Theory of Planned Behavior in the Homeownership Domain

Abstract

This study extends the theory of planned behavior (TPB; Ajzen, 1991) to the domain of homeownership. We used a 4-year longitudinal data set of 919 low-and-moderate income renters to explore factors associated with greater homeownership intentions and actual home purchases. Our findings provide strong support for the TPB. Favorable attitudes and subjective norms and greater perceptions of control were all associated with greater homeownership intentions. Homeownership intentions, in turn, predicted home purchases during the following year. The analysis included relevant demographic and economic variables, and the significance of income and geographic location suggests a distinction between respondents' perceived versus actual control. Our use of a longitudinal panel data set represents an important advance over much of the prior TPB literature, which tends to use cross-sectional designs and focus on short-term goals. We discuss implications for behavioral prediction using the TPB as well as implications for housing policy.

ACKNOWLEDGMENTS

This study was supported by the Ford Foundation. We thank Catherine Zimmer for analytic guidance; Bert Grider and for help with data management; and Janneke Ratcliffe, Michael Stegman, Angie Calcaterra, and Kim Manturuk for helpful comments on earlier versions of this article.

Notes

¹CAPS is run by the Center for Community Capital at the University of North Carolina at Chapel Hill to help assess a secondary mortgage market program developed out of a partnership between the Ford Foundation, Fannie Mae, and Self-Help, a leading community development financial institution. The goal of mortgage program is to provide evidence to lenders, policymakers, and the secondary mortgage market that low-wealth borrows are “bankable” and that Fannie Mae (and, by implication, Freddie Mac) can significantly expand their purchase of affordable housing loans without compromising their balance sheets or the safety and soundness of their practices. The renters subsample analyzed in this study serves as a comparison group to the CAPS homeowners.

²The goal of the renter sample selection was to complete around 1,500 interviews of low-income renters who lived in the same areas as the CAPS owners. We particularly wanted low-income renters who lived in geographic proximity to CAPS owners so as to neutralize the impacts of local market conditions on homeowner outcomes, especially with respect to the financial impacts of homeownership. We also wanted to assess differences between renters and homeowners living in the same areas. To select the low-income renter panel, we limited our search to the 30 metropolitan areas with the largest number of outstanding loans, starting with the subset of CAPS owners in those areas who had participated in the 2003 owner survey. We then looked for “matching” renters, that is, those living in the same neighborhood as a CAPS owner. The term “neighborhood” was defined as the same census block group as the homeowner. If too few qualified renters could be found in a particular census block group, the search was extended to census tract level. If insufficient potential renter respondents were found in the census tract, the neighborhood was extended to a four-mile radius around the CAPS owner. The potential renter survey respondents were identified from a database created and maintained by Genesys Sampling Systems. To be eligible for participation in the low-income renter sample, a respondent had to be the person who signed the rental lease and paid the rent and had to meet CAPS income limits. For renters, the CAPS income limit referenced the household's 2002 income and was equal to 80% of the AMI when the percent minority population was less than 30%, or 115% of AMI when the percent minority population was 30% or greater in census tracts. A total of 15,935 households were sampled to ultimately locate 1,531 qualified, matching low-income renter panel participants. A small financial incentive was provided to respondents at each year.

Note. Each respondent that remained a renter 1 year later was entered in the data set as a separate observation for the following year. The final data set consisted of a total of 2,168 observations (919 respondents), including 159 respondents who became homeowners. Respondents were “censored” (i.e., dropped from further analysis) if they did not respond to later surveys or if they had missing data on one more of the variables of interest. NA = not applicable.

Note. N = 919 respondents. Relative income = (Income – Area Median Income)/10,000. The other-race category included Asian participants, Native American participants, multi-racial participants, and all other participants who did not identify themselves as White, Black, or Hispanic.

³Comparing the 919 CAPS renters in our sample with renters from the 2003 American Housing Survey (AHS; U.S. Census Bureau, Housing and Household Economic Statistics Division, 2005) indicated that CAPS renters were more likely to be minority (53% vs. 49%), female (71% vs. 54%), older (>50; 22% vs. 18%), married or partnered (47% vs. 35%) unemployed (39% vs. 34%), and have more than a high school education (83% vs. 77%). Although these differences were statistically significant in chi-square tests (p < .001), with the exception of sex, the substantive demographic differences between the CAPS sample and the AHS sample are small. We controlled for all of these demographic characteristics in our analyses.

⁴Of those who purchased homes between 2004 and 2005, 93% reported purchasing their home with a 30-year fixed mortgage that carried an interest rate of 6.06%. Respondents reported paying a median purchase price of $101,750 for the home. No data are currently available for loans that originated after 2005.

Note. N = 919 respondents. Means (with standard deviations) are presented on the diagonal. Pearson correlations are presented in the lower triangle. Responses ranged from 1 (strongly disagree) to 5 (strongly agree). All correlations were significant at p < .001.

⁵Several other variables from the 2000 Census were also examined: unemployment rate, homeownership rate, and cost of housing. All Census variables were measured at the level of the Metropolitan Statistical Area. When the Census variables were entered simultaneously, only area poverty rate was significant. For parsimony in presentation, we trimmed the nonsignificant Census variables from the model.

Note. N = 2,168 observations (919 respondents). Akaike's Information Criterion = 3,143.4, Bayes' Information Criterion = 3,172.4. Tests for all variables contained 1 df except where noted. Significant effects shown in bold. T.V. = time-varying (all other variables were assessed once and remained constant).

Note. N = 2,168 observations (919 respondents). Akaike's Information Criterion = 3,530.7, Bayes' Information Criterion = 3,559.7. Tests for all variables contained 1 df except where noted. Significant effects shown in bold. T.V. = time-varying (all other variables were assessed once and remained constant).

Note. N = 2,168 observations (919 respondents). Akaike's Information Criterion = 3,740.3, Bayes' Information Criterion = 3,769.2. Tests for all variables contained 1 df except where noted. Significant effects shown in bold. T.V. = time-varying (all other variables were assessed once and remained constant).

Note. N = 2,168 observations (919 respondents). Akaike's Information Criterion = 5,058.1, Bayes' Information Criterion = 5,087.0. Tests for all variables contained 1 df except where noted. Significant effects shown in bold. T.V. = time-varying (all other variables were assessed once and remained constant).

Note. N = 2,168 observations (919 respondents). Model Fit: χ2(29) = 182.60, p < .001, −2 Log Likelihood = 954.27. Nagelkerke R ² = .20, Cox & Snell R ² = .08. Tests for all variables contained 1 df except where noted. Significant effects shown in bold. OR =Odds Ratio. T.V. = time-varying (all other variables were assessed once and remained constant).

Note. N = 2,168 observations (919 respondents). Model Fit: χ²(12) = 177.73, p < .001, −2 Log Likelihood = 959.14. Nagelkerke R ² = .19, Cox & Snell R ² = .08. Tests for all variables contained 1 df except where noted. Significant effects are shown in bold. OR = odds ratio; T.V. = time-varying (all other variables were assessed once and remained constant).