Main Questions:
Do we see labor market discrimination based on race?
How does implied race affect callback rates for job interviews?
Methodology:
This experiment was conducted as a field experiment where modified real resumes were altered and randomly assigned a black or white sounding name. Resumes were evaluated on quality to equally match the quality of the resume with the race it would represent. The actual resumes were sent to jobs advertised in newspapers in both Chicago and Boston and the experiment took place between 2001-2002. This experiment simply accounted for whether or not the resume assigned to a black or white name received a call back.
Main Findings:
The experiment found statistically significant evidence that there was a gap in callback rates finding specifically that a black candidate on average needed 8 more years of experience to receive a callback and 50% of the difference was able to be attributed solely to the name manipulation.
In terms of the quality of the resume, on average the higher quality resume had a better chance of a callback across both groups, but the better quality resume more positively impacted the white groups callback rates.
The experiment also captured data about the neighborhood in which the candidate lived and found candidates living in higher income neighborhoods had more callbacks than those from low income neighborhoods. The effect across groups for the neighborhood in which their address was listed was about the same across both groups.
The following table reports the variables mean value broken down by race for which data was captured during this experiment. The ‘b’ designation represents black and the ‘w’ designation represents white.
| race | avg_ofjobs | avg_yearsexp | avg_computerskills | avg_specialskills | avg_col | avg_h | avg_female |
|---|---|---|---|---|---|---|---|
| b | 3.658 | 7.830 | 0.832 | 0.327 | 0.723 | 0.502 | 0.775 |
| w | 3.664 | 7.856 | 0.809 | 0.330 | 0.716 | 0.502 | 0.764 |
The coefficient for the black dummy variable (listed below) represents that if the candidate is black, it is .032033*100 percentage points less likely that the candidate will receive a callback. (Note: This variable is significant at the 5% level of significance)
##
## Call:
## lm(formula = call ~ black, data = Data)
##
## Residuals:
## Min 1Q Median 3Q Max
## -0.09651 -0.09651 -0.06448 -0.06448 0.93552
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.096509 0.005505 17.532 < 2e-16 ***
## black -0.032033 0.007785 -4.115 3.94e-05 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.2716 on 4868 degrees of freedom
## Multiple R-squared: 0.003466, Adjusted R-squared: 0.003261
## F-statistic: 16.93 on 1 and 4868 DF, p-value: 3.941e-05Interpretation For Each Independent Variable Follows:
black: The coefficient for the black dummy variable (listed below) predicts that if the candidate is has a black sounding name, it is .0321751*100 percentage points less likely that the candidate will receive a callback. (Note: This variable is significant at the 5% level of significance)
The coefficients on black that were estimated under this model and the previous model were (nearly) the same. This is expected because none of the variables that were added to this model seem to be obviously correlated with race. The variables chicago, female, yearsexp, ofjobs,and education seem like they would not be independent from race.
chicago: The coefficient for the chicago dummy variable (listed below) predicts that if the candidate has a chicago address, it is .03739*100 percentage points less likely that the candidate will receive a callback. (Note: This variable is significant at the 5% level of significance)
yearsexp: The coefficient for the yearsexp variable (listed below) predicts that for every year of experience the candidate has it is .0031787*100 percentage points more likely that the candidate will receive a callback. (Note: This variable is significant at the 5% level of significance)
female: The coefficient for the female dummy variable (listed below) predicts that if the candidate is female, it is .0177709*100 percentage points more likely that the candidate will receive a callback. (Note: This variable is not significant at the 5% level of significance)
ofjobs: The coefficient for the ofjobs variable (listed below) predicts that for every job the candidate has previously had, it is .0088452*100 percentage points more likely that the candidate will receive a callback. (Note: This variable is significant at the 5% level of significance)
education: (broken down per factor level (excluded group: 0)): factor(education)1:The coefficient for the factor(education)1 variable (listed below) predicts that a candidate that is a high school dropout is .0022208*100 percentage points more likely to receive a callback versus a candidate with no education. (Note: This variable is not significant at the 5% level of significance)
factor(education)2:The coefficient for the factor(education)2 variable (listed below) predicts that a candidate that is a high school graduate is .0059974*100 percentage points more likely to receive a callback versus a candidate with no education. (Note: This variable is not significant at the 5% level of significance)
factor(education)3:The coefficient for the factor(education)3 variable (listed below) predicts that a candidate with some college education is .0035667*100 percentage points less likely to receive a callback versus a candidate with no education. (Note: This variable is not significant at the 5% level of significance)
factor(education)4:The coefficient for the factor(education)4 variable (listed below) predicts that a candidate with a college degree is .0038701*100 percentage points less likely to receive a callback versus a candidate with no education. (Note: This variable is not significant at the 5% level of significance)
##
## Call:
## lm(formula = call ~ black + chicago + yearsexp + female + ofjobs +
## factor(education), data = Data)
##
## Residuals:
## Min 1Q Median 3Q Max
## -0.16692 -0.09467 -0.07519 -0.05061 0.99632
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.1142922 0.0425316 2.687 0.00723 **
## black -0.0321751 0.0077652 -4.143 3.48e-05 ***
## chicago -0.0373900 0.0089829 -4.162 3.20e-05 ***
## yearsexp 0.0031787 0.0008121 3.914 9.20e-05 ***
## female 0.0177709 0.0099179 1.792 0.07323 .
## ofjobs -0.0088452 0.0036720 -2.409 0.01604 *
## factor(education)1 0.0022208 0.0586684 0.038 0.96981
## factor(education)2 0.0059974 0.0433681 0.138 0.89002
## factor(education)3 -0.0035667 0.0413339 -0.086 0.93124
## factor(education)4 -0.0038701 0.0406761 -0.095 0.92420
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.2708 on 4860 degrees of freedom
## Multiple R-squared: 0.01115, Adjusted R-squared: 0.009318
## F-statistic: 6.088 on 9 and 4860 DF, p-value: 1.497e-08There are other variables captured in this data set that were not previously explored. It seems reasonable that having computer skills would be relevant to a large portion of jobs, even if it is not essential to the job to indicate that you have some technological ability. It also seems reasonable that a candidate should have some special skill. Having an extra skill would set a candidate apart from those without one and it seems reasonable that an extra skill may increase the odds of a callback.
computerskills :The coefficient for the computerskills dummy variable (listed below) represents that if the candidate has a adequate computer skills, it is .0209*100 percentage points less likely that the candidate will receive a callback. (Note: This variable is significant at the 5% level of significance)
It is unusual to consider the idea that a person with computer skills is actually less likely to get a callback in this sample.
specialskills :The coefficient for the specialskills dummy variable (listed below) represents that if the candidate has a special skill, it is .064601*100 percentage points more likely that the candidate will receive a callback. (Note: This variable is significant at the 5% level of significance)
##
## Call:
## lm(formula = call ~ black + computerskills + specialskills, data = Data)
##
## Residuals:
## Min 1Q Median 3Q Max
## -0.15668 -0.10443 -0.07118 -0.03983 0.96017
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.092079 0.010141 9.080 < 2e-16 ***
## black -0.031349 0.007739 -4.051 5.18e-05 ***
## computerskills -0.020900 0.010086 -2.072 0.0383 *
## specialskills 0.064601 0.008235 7.845 5.30e-15 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.2699 on 4866 degrees of freedom
## Multiple R-squared: 0.01663, Adjusted R-squared: 0.01603
## F-statistic: 27.43 on 3 and 4866 DF, p-value: < 2.2e-16