Wednesday, December 27, 2023

L’affaire Gay

Three days ago I wrote about a Swedish example of “correlation doesn’t prove causation.” Now comes a Daily Mail article in which statisticians make similar claims about the published research of Harvard President Claudine Gay.

Jonatan Pallesen, a Copenhagen-based data scientist working for the Confederation of Danish Industry, tweeted that he had examined her use of data in her PhD thesis, and a 2001 American Political Science Review (APSR) paper.

The 2001 paper - 'The Effect of Black Congressional Representation on Political Participation' - was one of four peer-reviewed political science articles which secured her 2005 tenure at Stanford.

Pallesen posts on X:

The claim is that having a Black representative in a district causes lower White voter turnout. But what is being measured in the study are correlations between the two.

Might it not be equally likely that lower White voter turnout causes (increases the likelihood of) the election of black representatives? How about if some extraneous variable causes both? Something like the pressure by courts and federal agencies to form districts with Black majorities? 

Might not Whites who live in such districts have different characteristics than those who live in districts where they are the majority? Be for example on average poorer, have less education, a greater propensity for substance abuse? I daresay all of those would reduce voting propensity.

Say it together with me: Correlation does not demonstrate causation. It suggests a possible relationship between the two worth exploring further, although even that isn’t a sure thing. Coincidences do happen.

BTW, unwillingness to share one’s raw data, if requested, is another academic research no-no. Established scholars cited in the DM allege they have requested and Gay has refused. At Manhattan Contrarian, Francis Menton writes:

As far as I am concerned, failure by a researcher to share the underlying data as to published work is prima facie evidence that the data have been falsified.

Ouch, that stings. Hat tip to RealClearPolicy for the link.