Red Voter, Blue Voter

[marknau]

The American National Election Studies is a collaboration between the U of Michigan and Stanford. They do polling of individuals during election years, asking them various questions about their political views, views of candidates, and demographics data. They make the full dataset available for download, so one can do data analysis

Comments

[marknau]

So, the two biggest factors to individual partisanship are ethnicity and biblical origin. Each is associated with an almost 80-point swing in the temperature ratings. This is the difference between one person who rates Bush and Republican party at 60, and Kerry and the Democrats at 40, while another person switches those around.

Income comes next, with the difference between top and bottom being half as large as either of the two major factors.

And then being married, and having a post-graduate degree each have an effect about half as big as income.

[zamiel]

Which is interesting, because it would seem to imply that either:

• Non-African Americans don’t believe that the Bible is the inerrant Word of God, or
• Blacks find their own race is twice as compelling an influence than their white counterparts with the same beliefs.

Since I think the first one is testable from this dataset, this suggests an intriguing experiment.

If the latter is pointed out, it suggests that the African-American population legitimizes stereotyping. Since, at that point, it’s not - it’s an accurate description of the mass. Which, as a properly enlightened cynic, I find amusing.

[candid]

It seems like the results would be easier to interpret if you included a B2 dummy variable (I am assuming that B2 covers all other cases) instead of the constant term, then you'd have (if I am doing stats correctly)

B1: -5.2
B2: -26.9
B3: -78.5

Also, I am unhappy with how constrained your dependent variable is (always between -2 and +2), and would probably rescale it to [0,1] and run a logit model.