Replication and reanalysis for: van der Lee and Ellemers 10.1073/pnas.1510159112 . Published reanalysis: 10.1073/pnas.1518936112

VDL_simpson_paradox.R

######################
# Replication and reanalysis for:
#   van der Lee and Ellemers 10.1073/pnas.1510159112

# load data from supplemental Table S1
# found at http://www.pnas.org/content/early/2015/09/16/1510159112/suppl/DCSupplemental

d <- read.csv("http://xcelab.net/VDL_data.csv")
d$rejects <- d$applications - d$awards

# replicate paper results, ignoring discipline

summary( m0 <- glm( cbind(awards,rejects)~gender , 
    data=d , family=binomial ) )

# now add discipline to control for different baseline success rates

summary( m1 <- glm( cbind(awards,rejects)~gender + discipline , 
    data=d , family=binomial ) )

# consider model that omits gender, but includes discipline

summary( m2 <- glm( cbind(awards,rejects)~discipline , 
    data=d , family=binomial ) )

# compare AIC scores

cbind(m0=AIC(m0),m1=AIC(m1),m2=AIC(m2))