James P. Scanlan, Attorney at Law

FHA/VA Steering Study

(April 20, 2013; rev. June 19, 2013)

In July 2012 a collaborative group comprised of the California Reinvestment Coalition, Empire Justice Center, Massachusetts Affordable Housing Alliance, Neighborhood Economic Development Advocacy Project, Ohio Fair Lending Coalition, Reinvestment Partners, and Woodstock Institute issued a report styled “Paying More for the American Dream IV – Racial Disparities in FHA/VA Lending” that purported to “examine[] systemic inequities in the mortgage market, as reflected in neighborhood lending patterns based on race and ethnicity.”  The report compared differences between the proportions of home-purchase and refinance loans made to black, Latino and white borrowers were FHA and VA loans.  The proportion of FHA/VA loans comprised of loans to blacks and Latinos were several times the proportion FHA/VA loans comprised of loans to whites. 

Typical of the way the report’s findings were characterized in the media  is the headline of the Wall Street Journal blog stating  “Report Raises Questions Over Racial Lending Disparities.”   The item described the report’s authors as stating that “ their analysis of mortgage data raises questions about whether lenders are steering minority borrowers into government-backed loans that are slightly more expensive than conventional mortgages.” 

A report like this that found that disparities were larger than would be expected to occur for nondiscriminatory reasons would be subject to the problems discussed on the Underadjustment Issues and Partial Picture Issues subpages of the Lending Disparities page of jpscanlan.com.

But the remarkable thing about the study is that it did not attempt to estimate what the size of the disparity would be absent discrimination.  FHA and VA loans provide an avenue for securing mortgages for borrowers who will have trouble securing conventional   Minorities have greater difficulty qualifying for conventional loans than whites.  Therefore, a higher proportion of minority than white loans will be FHA/VA loans regardless of any discrimination.  Thus, a study that merely describe differences without providing any benchmark to against which the differences might be compared provides no evidence whatever that there exists discrimination.[i] 

The study does, however, provide some data that are illustrative of the statistical patterns I have described on the Scanlan’s Rule page and various other pages this site[ii] whereby the rarer an outcome the greater tends to be the relative difference in experiencing it and the smaller tends to be the relative difference in avoiding it.  And some of the discussion in the report illustrates that problem with efforts to appraise the size of differences between outcome rates without consideration of such patterns. 

While describing the disparities as “stark” several times, the report noted that the disparities were “especially stark” in the refinancing market.   But anyone familiar with the statistical patterns described on this site would understand that relative differences in adverse outcomes (in the context of the study, receipt FHA/VA rather than conventional loans) would tend to be larger for refinance loans than purchase loans simply because receipt of FHA/VA rather than conventional loans was much less common for refinance loans than for purchase loans.  As shown in Charts I and II of the report, for all cities combined the average proportion of purchase loans that were FHA/VA was 31.0% in predominantly white neighborhoods and 66.6% in predominantly black neighborhoods, while these figures were 8.2% and 27.2% for refinance loans.  On the other hand, however, relative differences in proportions of loans that were conventional (the favorable outcome) would tend to be larger for purchase loans than refinance loans.  That is, while the black rate of receiving FHA/VA loans was 3.3. times the white rate for refinance loans and 2.1 times the white rate for purchase and 3.1 time the white rate for purchase loans, the white rate of receiving a conventional loan was 2.1 times the black rate (69.0%/33.4%) for purchase loans but only 1.3 times the black rate (91.8%/72.8%) for refinance loans.

Table 1 below shows that, whether the data are analyzed according to applicant race or predominant race of neighborhood, for each city and for the all-city average, the relative difference in the proportion of loans that were FHA/VA was greater for refinance loans than for purchase loans, while the relative difference in the proportion of loans that were conventional was greater for purchase loans.  The final three columns provide the same sort of information using the EES (for “estimated effect size,” the measure described on the Solutions subpage of the Measuring Health Disparities page that is theoretically unaffected by the prevalence of an outcome).  The columns show that the difference tended to be larger for purchase loans than refinance loans, though that varied somewhat depending on whether the analysis was according to race of applicant or predominant race of neighborhood.    

In Table 1 the abbreviations that are not self-explanatory are: PWA = Purchase White Adverse Outcome Rate i.e., proportion of loans that were FHA/VA); RWA = Refinance White Adverse Outcome Rate;  PAR – Purchase Adverse Ratio (in terms of the rate of the black FHA/VA rate to the white FHA/VA rate); RAR – Refinance Adverse Ratio; PFR = Purchase Favorable Ratio (in terms of the ratio of the white rate conventional rate to black conventional rate); RFR = Refinance Favorable Ratio; PEES = Purchase EES; REES = Refinance EES.  The PWA and RWA are shown to provide an indicator of the general prevalence of the adverse outcome for each type of loan.  See note iii infra.

Table 1:  Measures of White and Black Differences in Adverse and Favorable Outcome Respecting Mortgage Type and EES [ref b3625b6]

One might also use these data to attempt to show patterns of relative differences in favorable and adverse outcomes according to prevalence by city.  The pattern by city would be less consistent with the prevalence-related factors than the pattern by loan type because the differences in the sizes of the disparity (EES) would tend to be greater relative to differences in prevalence by city than by loan type. See Section A.9 of the  Scanlan’s Rule page and the Life Table Illustrations subpage of the Scanlan’s Rule page.[iii]

Notably, in March 2013, the Woodstock Institute, one of the co-authors of the FHA/VA study, issued a Fact Sheet reflecting preliminary results of a study of gender differences in mortgage approval rates titled “Unequal Opportunity: Disparate Mortgage Origination Patterns for Women in the Chicago Area,” which measured disparities according to relative differences in favorable outcomes.  The full study is to be released later this year.  Had that approach to measuring disparities been employed in the FHA/VA study, it would have reported larger disparities for purchase loans than refinance loans. 

A summary released with the Fact Sheet stated that disparities were most pronounced among African Americans, citing the fact that the black women were 34% less likely to have their loans originated than black men, while white women were 22% less likely to have their loans originated than white men.  But had the authors relied on relative differences in adverse outcomes, as was done in the FHA/VA study, they would have found the disparities greater among whites than African-Americans.  For discussion of the common situations where observers attribute significance to larger relative differences in adverse outcomes among whites than blacks or among any advantaged subpopulation without consideration of the extent to which the pattern is to be expected because adverse outcomes tend to be less common among advantaged subpopulations and without recognizing that one would commonly observe the smaller relative differences in the favorable outcome in the advantaged subpopulation, see page at 16-17 of the Harvard University Measurement Letter and in my “Race and Mortality” (Society, Jan/Feb 2000, reprinted in Current, Feb. 2000) and “The Perils of Provocative Statistics” (Public Interest, Winter 1991).

The Woodstock gender study discusses larger relative differences in favorable outcomes for refinance loans than for purchase loans.  In this instance the favorable outcome tends to be less common for refinance than purchase loans, the opposite of the pattern observed the FHA/VA study where the adverse outcome was less common for the refinance loans.  But that pattern is not consistent and the differences in prevalence are usually not substantial (except in the case of blacks, as shown in Figures 2 and 3 of the Fact Sheet). 

It would only be as to blacks that one would find the situation where the relative difference in the favorable outcome is larger for the refinance (the rarer outcome) while the relative differences in the adverse outcome is larger for purchase loans (where that is the rarer outcome).  The pattern of opposite reaching opposite conclusions reached about whether gender disparities are larger for purchase or refinance depending on whether one relied on relative differences in favorable or relative differences in adverse outcomes would not hold for whites and Hispanics.  But as to those two groups, the differences in prevalence are not as great as for blacks and hence the prevalence-related forces are less likely to predominate in the face of other factors.