This experiment data comes from a study that sought to understand the influence of race and gender on job application callback rates. The study monitored job postings in Boston and Chicago for several months during 2001 and 2002 and used this to build up a set of test cases. Over this time period, the researchers randomly generating resumes to go out to a job posting, such as years of experience and education details, to create a realistic-looking resume. They then randomly assigned a name to the resume that would communicate the applicant's gender and race. The first names chosen for the study were selected so that the names would predominantly be recognized as belonging to black or white individuals. For example, Lakisha was a name that their survey indicated would be interpreted as a black woman, while Greg was a name that would generally be interpreted to be associated with a white male.
Format
A data frame with 4870 observations, representing 4870 resumes, over
30 different variables that describe the job details, the outcome
(received_callback), and attributes of the resume.
- job_ad_id
Unique ID associated with the advertisement.
- job_city
City where the job was located.
- job_industry
Industry of the job.
- job_type
Type of role.
- job_fed_contractor
Indicator for if the employer is a federal contractor.
- job_equal_opp_employer
Indicator for if the employer is an Equal Opportunity Employer.
- job_ownership
The type of company, e.g. a nonprofit or a private company.
- job_req_any
Indicator for if any job requirements are listed. If so, the other
job_req_*fields give more detail.- job_req_communication
Indicator for if communication skills are required.
- job_req_education
Indicator for if some level of education is required.
- job_req_min_experience
Amount of experience required.
- job_req_computer
Indicator for if computer skills are required.
- job_req_organization
Indicator for if organization skills are required.
- job_req_school
Level of education required.
- received_callback
Indicator for if there was a callback from the job posting for the person listed on this resume.
- firstname
The first name used on the resume.
- race
Inferred race associated with the first name on the resume.
- gender
Inferred gender associated with the first name on the resume.
- years_college
Years of college education listed on the resume.
- college_degree
Indicator for if the resume listed a college degree.
- honors
Indicator for if the resume listed that the candidate has been awarded some honors.
- worked_during_school
Indicator for if the resume listed working while in school.
- years_experience
Years of experience listed on the resume.
- computer_skills
Indicator for if computer skills were listed on the resume. These skills were adapted for listings, though the skills were assigned independently of other details on the resume.
- special_skills
Indicator for if any special skills were listed on the resume.
- volunteer
Indicator for if volunteering was listed on the resume.
- military
Indicator for if military experience was listed on the resume.
- employment_holes
Indicator for if there were holes in the person's employment history.
- has_email_address
Indicator for if the resume lists an email address.
- resume_quality
Each resume was generally classified as either lower or higher quality.
Source
Bertrand M, Mullainathan S. 2004. "Are Emily and Greg More Employable than Lakisha and Jamal? A Field Experiment on Labor Market Discrimination". The American Economic Review 94:4 (991-1013). doi:10.3386/w9873 .
Details
Because this is an experiment, where the race and gender attributes are being randomly assigned to the resumes, we can conclude that any statistically significant difference in callback rates is causally linked to these attributes.
Do you think it's reasonable to make a causal conclusion? You may have some health skepticism. However, do take care to appreciate that this was an experiment: the first name (and so the inferred race and gender) were randomly assigned to the resumes, and the quality and attributes of a resume were assigned independent of the race and gender. This means that any effects we observe are in fact causal, and the effects related to race are both statistically significant and very large: white applicants had about a 50\
Do you still have doubts lingering in the back of your mind about the validity of this study? Maybe a counterargument about why the standard conclusions from this study may not apply? The article summarizing the results was exceptionally well-written, and it addresses many potential concerns about the study's approach. So if you're feeling skeptical about the conclusions, please find the link below and explore!
Examples
head(resume, 5)
#> # A tibble: 5 × 30
#> job_ad_id job_city job_industry job_type job_fed_contractor
#> <dbl> <chr> <chr> <chr> <dbl>
#> 1 384 Chicago manufacturing supervisor NA
#> 2 384 Chicago manufacturing supervisor NA
#> 3 384 Chicago manufacturing supervisor NA
#> 4 384 Chicago manufacturing supervisor NA
#> 5 385 Chicago other_service secretary 0
#> # ℹ 25 more variables: job_equal_opp_employer <dbl>, job_ownership <chr>,
#> # job_req_any <dbl>, job_req_communication <dbl>, job_req_education <dbl>,
#> # job_req_min_experience <chr>, job_req_computer <dbl>,
#> # job_req_organization <dbl>, job_req_school <chr>, received_callback <dbl>,
#> # firstname <chr>, race <chr>, gender <chr>, years_college <int>,
#> # college_degree <dbl>, honors <int>, worked_during_school <int>,
#> # years_experience <int>, computer_skills <int>, special_skills <int>, …
# Some checks to confirm balance between race and
# other attributes of a resume. There should be
# some minor differences due to randomness, but
# each variable should be (and is) generally
# well-balanced.
table(resume$race, resume$years_college)
#>
#> 0 1 2 3 4
#> black 28 22 132 493 1760
#> white 18 18 142 513 1744
table(resume$race, resume$college_degree)
#>
#> 0 1
#> black 675 1760
#> white 691 1744
table(resume$race, resume$honors)
#>
#> 0 1
#> black 2310 125
#> white 2303 132
table(resume$race, resume$worked_during_school)
#>
#> 0 1
#> black 1069 1366
#> white 1076 1359
table(resume$race, resume$years_experience)
#>
#> 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
#> black 19 177 95 259 264 409 274 288 78 72 86 33 80 78 21 43 2 36
#> white 26 175 99 278 243 408 267 290 81 58 87 36 74 71 13 51 1 41
#>
#> 19 20 21 22 23 25 26 44
#> black 24 11 21 4 4 3 53 1
#> white 22 24 26 4 5 4 51 0
table(resume$race, resume$computer_skills)
#>
#> 0 1
#> black 408 2027
#> white 466 1969
table(resume$race, resume$special_skills)
#>
#> 0 1
#> black 1638 797
#> white 1631 804
table(resume$race, resume$volunteer)
#>
#> 0 1
#> black 1426 1009
#> white 1440 995
table(resume$race, resume$military)
#>
#> 0 1
#> black 2187 248
#> white 2210 225
table(resume$race, resume$employment_holes)
#>
#> 0 1
#> black 1349 1086
#> white 1339 1096
table(resume$race, resume$has_email_address)
#>
#> 0 1
#> black 1267 1168
#> white 1269 1166
table(resume$race, resume$resume_quality)
#>
#> high low
#> black 1223 1212
#> white 1223 1212
# Regarding the callback outcome for race,
# we observe a very large difference.
tapply(
resume$received_callback,
resume[c("race", "gender")],
mean
)
#> gender
#> race f m
#> black 0.06627784 0.05828780
#> white 0.09892473 0.08869565
# Natural question: is this statisticaly significant?
# A proper analysis would take into account the
# paired nature of the data. For each ad, let's
# compute the following statistic:
# <callback rate for white candidates>
# - <callback rate for black candidates>
# First contruct the callbacks for white and
# black candidates by ad ID:
table(resume$race)
#>
#> black white
#> 2435 2435
cb_white <- with(
subset(resume, race == "white"),
tapply(received_callback, job_ad_id, mean)
)
cb_black <- with(
subset(resume, race == "black"),
tapply(received_callback, job_ad_id, mean)
)
# Next, compute the differences, where the
# names(cb_white) part ensures we matched up the
# job ad IDs.
diff <- cb_white - cb_black[names(cb_white)]
# Finally, we can apply a t-test on the differences:
t.test(diff)
#>
#> One Sample t-test
#>
#> data: diff
#> t = 5.1896, df = 1322, p-value = 0.0000002437
#> alternative hypothesis: true mean is not equal to 0
#> 95 percent confidence interval:
#> 0.02021562 0.04478816
#> sample estimates:
#> mean of x
#> 0.03250189
#>
# There is very strong evidence of an effect.
# Here's a similar check with gender. There are
# more female-inferred candidates used on the resumes.
table(resume$gender)
#>
#> f m
#> 3746 1124
cb_male <- with(
subset(resume, gender == "m"),
tapply(received_callback, job_ad_id, mean)
)
cb_female <- with(
subset(resume, gender == "f"),
tapply(received_callback, job_ad_id, mean)
)
diff <- cb_female - cb_male[names(cb_female)]
# The `na.rm = TRUE` part ensures we limit to jobs
# where both a male and female resume were sent.
t.test(diff, na.rm = TRUE)
#>
#> One Sample t-test
#>
#> data: diff
#> t = 0.80699, df = 465, p-value = 0.4201
#> alternative hypothesis: true mean is not equal to 0
#> 95 percent confidence interval:
#> -0.01283147 0.03071416
#> sample estimates:
#> mean of x
#> 0.008941345
#>
# There is no statistically significant difference.
# Was that the best analysis? Absolutely not!
# However, the analysis was unbiased. To get more
# precision on the estimates, we could build a
# multivariate model that includes many characteristics
# of the resumes sent, e.g. years of experience.
# Since those other characteristics were assigned
# independently of the race characteristics, this
# means the race finding will almost certainy will
# hold. However, it is possible that we'll find
# more interesting results with the gender investigation.