I like to use student-generated data to illustrate course concepts. The core of the exercise in this case is a survey the students wrote, to which I added standard items on party identification and self-reported ideology.
The students’ 30 items covered many topics, some of which were very salient and others less so (which is good). I recall there being items on a TikTok ban, mask mandates, existence of climate change, trust in politicians/parties, and even attitudes toward the UK.
I plan to go back to these data as we move through the course. For now, the goals are to illustrate party identification, ideological self-identification, and revealed ideology; see to what extent these hang together; and just see what other things come to mind as we discuss the results.
One way to represent the data is as a set of ideal points. This helps start (or continue) a conversation about ‘revealed ideology.’
The first-dimension coordinates are highly correlated (0.7) with ideological self-placement:
The first-dimension score is a bit less strongly correlated with party ID (0.64 versus 0.7 for self-reported ideology). Meanwhile, party ID and self-reported ideology are highly correlated at 0.8:
My ideal point is (0.33, 0.33)! Modeling a second dimension raises accuracy from 88% to 92%. Items for which the second dimension markedly improves accuracy include: poverty reduction by state (not federal) government, whether federal officials are working in “your best interest,” school vouchers, “unconditional” aid to allies, soda/pop regulation, and whether law enforcement professionals should recuse themselves from politics.
Here are the three measures above, broken out by gender. I captured gender with an open-ended question (text box):
The survey was anonymous, but I gave people the option to supply a nickname. That lets those who did so find themselves in the data if they want.