The data was provided by The original data was in SPSS format, so we need to use the foreign package to import it into R. The original paper can be found here. Remember, this was a randomized controlled experiment to understand the impact of gender. i.e. applications were randomized to have either male or female names.
library(foreign)
hiring <- read.spss("Moss-Racusin_etal_replicate.sav", to.data.frame=TRUE)
The data looks like
## Subject cond gender likeable hireable competence competence2 salary mentoring
## 1 4 male male 2.666667 3.000000 0.277414539 4.000000 30000 4.333333
## 2 5 male male 3.500000 3.000000 0.843618853 4.000000 45000 3.666667
## 3 7 male male 3.333333 3.333333 0.057517929 3.666667 30000 4.666667
## 4 13 male male 2.000000 3.333333 -0.342407302 3.333333 25000 5.666667
## 5 14 male male 3.000000 3.333333 0.462666850 4.000000 35000 4.333333
## 6 15 male male 4.333333 4.000000 0.004375458 3.000000 40000 5.666667
where
cond is the (randomized) “gender” of the applicantgender is the gender of the evaluatorWe have the following split for the gender of the applicants
##
## male female
## 63 64
but also the following \(2\times 2\) split in who evaluates who:
## gender
## cond male female
## male 48 15
## female 45 19
i.e. 48 male candidates were evaluated by men while 15 male candidates where evaluated by women.
This is the overall comparison (i.e. not taking into account the gender of the evaluator)
## Response variable: numerical, Explanatory variable: categorical
## Difference between two means
## Summary statistics:
## n_male = 63, mean_male = 3.7804, sd_male = 1.2385
## n_female = 64, mean_female = 2.9245, sd_female = 1.0423
## Observed difference between means (male-female) = 0.8559
##
## Standard error = 0.2033
## 95 % Confidence interval = ( 0.4575 , 1.2544 )
Overall, it seems that on average men had higher “hireability” scores.
Let’s split the data set by the gender of the evalator using the filter command
hiring.male <- filter(hiring, gender=="male")
hiring.female <- filter(hiring, gender=="female")
We show 95% confidence intervals of \(\mu_{M}-\mu_{F}\) for the male evaluators:
## Response variable: numerical, Explanatory variable: categorical
## Difference between two means
## Summary statistics:
## n_male = 48, mean_male = 3.7361, sd_male = 1.2395
## n_female = 45, mean_female = 2.9593, sd_female = 1.1232
## Observed difference between means (male-female) = 0.7769
##
## Standard error = 0.245
## 95 % Confidence interval = ( 0.2966 , 1.2571 )
We show 95% confidence intervals of \(\mu_{M}-\mu_{F}\) for the female evaluators:
## Response variable: numerical, Explanatory variable: categorical
## Difference between two means
## Summary statistics:
## n_male = 15, mean_male = 3.9222, sd_male = 1.2675
## n_female = 19, mean_female = 2.8421, sd_female = 0.8416
## Observed difference between means (male-female) = 1.0801
##
## Standard error = 0.38
## 95 % Confidence interval = ( 0.2651 , 1.8951 )
For both female and male evaluators, seems that on average men had higher “hireability” scores.