Not sure how I missed this thread. Probably because I assumed the 3 massive nasty existing threads were enough. Suppose not..
*Disclaimer* I don't think vote fraud explains this phenomenon. It's my realist bias. Thus far, however, nothing fully explains the precinct-size correlation.
da32130:
While it's good you are trying to incorporate demographics (+rep), you're making several mistakes.
1 -- this research was already done 15 days ago, with the data and analysis posted for everyone's use:
Revisited the data. This time, rolled precinct up and grouped by just the demographics and fields I wanted to test.
Fields Tested:
Median Income (County)
% White (County)
% Over 65 (County)
Precinct Size
Precinct % of County Vote
% Female (County)
People per Square Mile (County)
Precinct Turnout %
With all parameters, R^2 = .437 (explains roughly 44% of the noise)
With just the top 4, R^2 = .398 (explains 40% of the noise)
With just the top 3, R^2 = .256 (explains 26% of the noise) [--> this one excludes precinct size]
Median Income, % White, and Precinct Size have comparable significance (t ~ 15)
% Over 65 is slightly less significant (t = 9)
Romney's vote is positively correlated with Median Income, Precinct Size, and % Over 65. Interestingly, he's negatively correlated with % White.
For this type of rough analysis, an R^2 of .40 is significant. That .14 comes from Precinct Size is also significant.
From this analysis, precinct size cannot be ruled out as a significant independent variable. This suggests that something else must be driving the correlation (other demographic, campaign activity, fraud, etc).
2 -- You're assuming that if B and C are correlated with A, then B and C are correlated. That can't be assumed, it must be shown. Your analysis, as I mentioned in my PM and in the other thread, does not demonstrate that these factors are correlated with precinct size and therefore explain the phenomenon. You've shown that these factors are positively correlated with Romney vote %, but you haven't shown that they are positively correlated with precinct size! Nor, more specifically, that they fully explain the precinct-size correlation.
3 -- You're disregarding the conclusions of the analysis that was done correctly and did incorporate demographics. Precinct Size cannot be ruled out based on the demographics tested -- it's still significant in an Income/% White/% Over 65/Precinct Size multiple regression on Romney Vote %.
4 -- Because your analysis is incomplete and does not build upon prior analysis, your conclusion is misleading and hurts the efforts to understand the phenomenon. Truth is, we still can't fully explain what's driving the precinct size / Romney positive correlation
5 -- You should release your data and analysis for peer review
6 -- Your final weighted graph assumes each of your variables are independent and not correlated. However, female % and republicanism are anti-correlated. They essentially measure the same thing, and are therefore double counted. The pivot table graph in the file below lets you view this type of information easily. You'll find that there is still precinct-size correlation even when the other demographics are normalized, just as the analysis in (1) above shows.
Still, most shocking of all, is that no one else has pointed out the obvious errors with your approach (#2, #6). Then again, based on the "analysts" in the other thread, I suppose it shouldn't be too surprising.
For the 50th time, if someone wants to do some demographic analysis, do not start from scratch!! Use this file:
http://dl.dropbox.com/u/63868969/RPF..._Analysis.xlsx
Conclusion: Not debunked
Connect With Us