Patrick Ruffini’s Interesting Beliefs About a Regression
“Patrick Ruffini is a strategist, thinker, and organizer focused on data and technology’s disruptive impact on politics and business. “
Here is a tweet he just posted (retweets are not necessarily endorsements so don’t blame Barro)
In a model to predict the swing to Trump, Black/Latino % in a county is not predictive at all. Whites+education, % Mormon, % Italians are. pic.twitter.com/BX3pmfXKxk
— Patrick Ruffini (@PatrickRuffini) November 11, 2016
Since the percent Asian + … + percent other always must add up to 100% the computer can’t estimate coefficients on all the variables. It excluded percent Asian. This means that, to the extent the country level (I assume) data tell us anything about ethnicity and voting, they say the shift in votes from 2012 to 2016 was similar for Asians, Blacks and Hispanics.
If the excluded grouu had been non college whites, then Black and hispanic would have been extremely predictive.
Nothing can be inferred from such coefficients without thinking about the excluded group.
Treating coefficients on an indicator as a measurement related only to the group indicated by the variable is absurd.
Now we know from exit polls that the percentages of votes for Romney and Trump of blacks were similar (both roughly zero) and the same for Hispanics (neither roughly zero). The only new information that the regression could give us regards turnout of blacks and hispanics.
Ruffinis conclusions are probably correct (say exit polls) but they are not supported by the regression results.
This is a top Republican data guy and his party just won it all.