How Race and Class Stereotyping Shapes Attitudes Toward Affordable Housing

Abstract

The development of affordable housing often involves a contentious siting process. Proposed housing developments frequently trigger concern among neighbors and community groups about potential negative impacts on neighborhood quality of life and property values. Advocates, developers, and researchers have long suspected that these concerns stem in part from racial or class prejudice. Yet, to date, empirical evidence supporting these assumptions is lacking. This study seeks to examine roles that perceptions of race and class play in shaping opinions that underlie public opposition to affordable housing. This study applies a public opinion survey to determine the extent to which stereotypes and perceptions of the poor and minorities relate to attitudes toward affordable housing. The results demonstrate that such perceptions are particularly strong determinants of negative attitudes about affordable housing. These findings provide advocates, planners, developers, and researchers with a more accurate portrayal of affordable housing opposition, thereby allowing the response to be shaped in a more appropriate manner.

Keywords:

• Low-income housing
• discrimination
• minorities

Acknowledgements

This research was conducted with the support of a US Department of Housing and Urban Development Doctoral Dissertation Research Grant and support from the University of Texas Graduate School and the School of Architecture. Special thanks to Elizabeth Mueller for overseeing the research and to Andrew Stackhouse for reading many early drafts.

Notes

¹ Ordinary least squares linear regression was used since the dependent variable is a mean index, and thus a continuous variable (albeit bounded). Each of the independent variables is normally distributed, and the data do not indicate a nonlinear relationship between any of the independent variables and the dependent variable. Bivariate correlations between every explanatory variable were also applied to test for multicollinearity, and the results indicate no issues with multicolinearity between the explanatory variables.

² Survey researchers differ on the subject of providing a neutral response category. On the one hand, providing a neutral category allows respondents who do not have concrete opinions about the question to accurately describe those views. On the other hand, a neutral response category can lead to satisficing, or taking cognitive shortcuts to answer a survey question, leading to a biased response (Groves et al., 2004, p. 208). This can seriously undermine the validity of the survey instrument as a result of response error. Therefore, a neutral option was not provided but rather a volunteered ‘don't know’ and ‘refused’ option allowed the interviewers to appropriately code such responses.

³ The sample for this survey was purchased from Survey Sampling International (SSI) and utilizes SSI's ‘Random B’ sampling technique. SSI takes their sample from a database of all ‘directory listed’ households, but does not provide data on the geographic location of respondents. The University of Texas Office of Survey Research (OSR) implemented the survey using Computer-Aided Telephone Interviewing facilities. OSR used within-household sampling once a call is successful to obtain a more appropriate population sample rather than the household-level sample obtained through random digit dialing. Although such methods may result in lower response rates, using this technique increases the representativeness of the final sample. All interviews were conducted in English only, and no incentives were provided. No households who were contacted were unable to take the survey due to language barriers, so it is unlikely that there is any bias emanating from language exclusion.

⁴ Low response rates can result in nonresponse bias, although a number of articles suggest that changes in nonresponse rates do not necessarily alter survey estimates (Curtin et al., 2000; Keeter et al., 2000; Merkle and Edelman, 2002). As suggested by Groves (2006), post-data collection weighting was explored on potentially biased variables, but no significant changes in estimates were found, thus the nonweighted estimates were used in the analysis.

⁵ Income in the census is measured in slightly different categories than in the study sample (the census bureau categories are < $20K and $20–50K; mine are < $25K and $25–50K), so the low end of the income spectrum may match better than the table indicates.

⁶ Measured using a liberal–conservative scale which delineates ‘1’ as ‘very liberal’ and ‘7’ as ‘very conservative’. Twenty per cent of the respondents identified themselves at 1–3 on the scale; 30 per cent at 4; and 50 per cent at 5–7.

⁷ For missing data, the mean was imputed in order to maintain as many recorded responses as possible.

⁸ The index shows very high internal reliability with a Cronbach's α score of 0.846.

⁹ Correlation between the index and question A009 is 0.652, and correlation between the index and question A010 is 0.697.

¹⁰ Correlations ranging from 0.471 to 0.557.

¹¹ The five-item index demonstrated strong reliability, with a Cronbach's α score of 0.818.

¹² The eight-question index proves reliable with an α score of 0.801.

¹³ The eight-question index demonstrates an acceptable internal reliability with a Cronbach's α score of 0.703.

¹⁴ The sample size used in this study is comparable to other small-scale studies. As discussed by Wimmer & Dominick (2010), Groves et al., (2004), and Dillman et al., (2009), one would ideally have 800+ responses in a national survey in order to reduce the margin of error to under 3 per cent. With a sample of 300, the margin of error is around 6 per cent. Most survey researchers recognize that in all but large-research centers, getting 800+ respondents is often cost-prohibitive. Thus, they state that as long as the sample is randomly drawn (which this is) and representative (which this is), sample size should not adversely affect the results.

¹⁵ In this study, the RACE Index and POOR Index present only a moderate correlation (0.447).

¹⁶ The residual plot demonstrates normality; collinearity statistics (Variance inflation factor and Tolerance) show nothing of concern regarding collinearity (VIF < 2; Tolerance>0.5) for all variables.

¹⁷ Future research may wish to consider structural equation modeling to determine the extent to which ideology influences housing attitudes both directly and indirectly via how stereotypes are framed.

¹⁸ For a history of the Fair Housing Act in the USA, see Carr (1999), HUD (2006), and Kennedy (1999).