http://www.studioglyphic.com/blog/
Please update your links, bookmarks, and RSS Feedreaders. All comments have been disabled.
« Choices | Main | Reading list »
November 07, 2004
More analysis of election results
I'm fairly ready to take back what I said about the Republicans using an anti-gay agenda to their advantage. Several sources have stated that the evangelical turnout wasn't necessarily as proportionally large as may have been initially stated. I decided to run a regression analysis on the state by state results:
Dependent:
% Bush margin (e.g., 9.83% in Arkansas, -10.29% in California)
Independent:
Density (natural log of population/area, from Census 2004 population estimates)
% college-educated (percentage of population over 25 with bachelors degree or higher, from Census 2000 SF3)
Median income in 1999 (natural log of Census 2000 SF3 data)
Anti-gay ballot initiative (1 for yes, 0 for no)
Results:
R Square 0.65Coeff. t Stat
Intercept -1.30 -0.75
Density (ln) -0.07 -5.51
College or Higher -2.07 -3.64
Median Income (ln) 0.21 1.20
Gay Ballot 0.04 0.89
Check out the t Stat for density!
I took out the ballot initiative and ran the analysis again:
R Square 0.64Coeff. t Stat
Intercept -0.99 -0.58
Density (ln) -0.08 -5.60
College or Higher -2.11 -3.72
Median Income (ln) 0.18 1.06
That R Square value is pretty low. Any thoughts on what would make the model work better? Let me give you an example of how off it can be:
State Actual Predicted
Nevada 2.62% 30.64%
Maine -8.79% 13.00%
DC -80.27% -60.43%
Virginia 8.64% -8.46%
Nebraska 34.48% 15.81%
Utah 44.72% 11.95%
Perhaps dummy factors such as Coastal and Region might do a better job. If I had some data on church attendance, I'd throw that in, too.
Posted by glyphic at November 7, 2004 03:53 PM