,
8 tweets,
2 min read
Read on Twitter
One of my statistics heroes Cosma Shalizi (@cshalizi) argues that you can talk about the causal effect of race and gender. I disagree somewhat. My argument is you can't talk about a variable's effect without specifying the intervention you would perform to change it. Here's why: 
Consider men getting paid more and possible mechanisms. 1. Behaviorally. Men negotiate more and this leads to higher pay. 2. Biologically. Maleness causes muscles and muscles cause higher pay. 3. Socially. Society gives more pay to anybody who presents as male.
Consider the variable manipulation of making participant X male. I set SEX=1 to SEX=0. The real world intervention could be:
1. Give that person more testosterone
2. Change their name in a book to male
1. Give that person more testosterone
2. Change their name in a book to male
3. Give that person assertiveness training
4. Give that person a fake beard.
The lack of specificity matters because depending on type of intervention some causal mechanisms will be engaged and others not.
4. Give that person a fake beard.
The lack of specificity matters because depending on type of intervention some causal mechanisms will be engaged and others not.
My second point is race and gender are not fixed. They are social categories that are defined by their members and by society. So one also needs to think about how to intervene on these variables in a ways that don't alter the conception of category itself.
To summarize. In order to sensibly talk about causal effects, one must either:
A. define an intervention (and thus avoid thinking deeply about causal mechanisms) or;
B. Define the causal effect with such specificity that the associated intervention would be obvious.
A. define an intervention (and thus avoid thinking deeply about causal mechanisms) or;
B. Define the causal effect with such specificity that the associated intervention would be obvious.
