(Page 16 of 28) - Walking the Talk? What Employers Say Versus What They Do authored by Pager, Devah. and Quillian, Lincoln.

Table 1. A Comparison of Employers’ Self-Reports
and Behavioral Outcomes for Overlapping Sample

Audit Results

Likely to Hire
Drug Offender No

Call-Back Call-Back

No 56 4

(93.3 %)

(6.7 %)

Yes 89

(92.7 %)

(7.3 %)

Difference of percentages

0.6%

95% confidence interval

(-8.6%, 8.8%)

Among those who reported a favorable likelihood of hiring an applicant with a prior
felony drug conviction on the survey, 7.3 percent made calls to the tester with the
criminal record in the audit study, relative to 6.7 percent of those expressing an
unfavorable likelihood.

This difference is in the expected direction, but is only slightly

greater than zero (0.6%), and is far too small to reach statistical significance.

There remain two limitations of this analysis that must temper its conclusions. First, the
respondent in the survey may be different from the individual who reviewed the testers’
applications. To the extent that hiring practices vary within firms depending on the
individual manager or human resource officer, the correspondence between survey results
and audit results will be attenuated. This will attenuate correlational consistency,
although it should not systematically alter the level of aggregate consistency. In cases
where they differ, there is little reason to believe that the hiring agent would be
systematically more or less likely to consider hiring ex-offenders than would the survey
respondent, leading to no overall effect on the level of aggregate consistency observed.

A second limitation concerns the size of the sample of employers who completed both the
survey and the audit; the relatively small sample size limits the reliability of the estimate
of the small difference in percentages (0.6%). Fortunately, statistical theory allows us to
assess this second problem. We can make an exact assessment of the conclusions we can
draw from a sample of this size by calculating a range of values that bound the extent of
correlational consistency that would be likely to hold if we did have a much larger
sample. In other words, we can calculate confidence intervals for the difference between
the hiring percentages. A 95% confidence interval for the difference in percentages is a
range of –8.6 percent to 8.8 percent, indicating that we are 95% confident that employers
who indicate “yes” are no more than 8.8 percentage points more likely to make a callback
than employers who indicate “no”.

A difference of 8.8 percentage points in making a

Given the small sample sizes in this final comparison, a further breakdown by race of the tester would be

impossible. Analyses, therefore, include all call-backs to testers in the criminal record condition, regardless
of race.

This confidence interval is calculated using the “plus 4” method of Agresti and Caffo (Agresti and Caffo

2000). The Agresti and Caffo method has the important advantage that it provides reasonably accurate
(and slightly conservative) intervals even when the count of successes or failures is very small, as is the