Details. Here is a good example of the kind of detail work being done to nurture this election.
by Tom Schaller
Fri Oct 8th, 2004 at 12:49:55 GMT
Michigan's voter registration deadline closed earlier this week, on October 4. The state's Board of Elections has not certified and published the final registration totals (due out next week, according to Steven Luck of the MBoE). Michigan also does not record the party affiliation of registered voters. So, is it possible to get a sense of whether the registration efforts of the two parties and their affiliated groups this year have made a difference, and if so, how?
Actually, I think the answer is a qualified "yes."
The Michigan elections board has published the preliminary, county-level registration numbers (as of July 6, 2004) on its website. Coupling those totals with both the final, county-level registration figures from October 2000, as well as the final, county-level voting results from the 2000 presidential election, we can get a basic sense of the relationship between the current registrant base in each county and how that county voted four years ago.
But the data, at first blush, appear to send what the president might call "mixed messages." If you compare MBoE's final, official October 2000 county registration totals with the July 2004 (pre-final but official) county registration totals, the Republican-leaning counties have made greater registration gains during the past four years. However, if you compare the actual turnout figures from the November 2000 election with the July 2004 registration totals, the Democratic-leaning counties have made more registration gains.
Allow me to explain my analyses in order to unpack these (seemingly?) contrary findings:
- First, and for both sets of analyses, I calculated the Democratic performance percentage of the 2000 county-level presidential vote in 2000 (i.e., Gore's percent). For example, in Alcona County (MI's first, alphabetically, of 83 counties), Gore received 2,696 votes, or 45.0%;
- Next, I calculated two county-level figures from 2000: (a) the final registration total for each county as a share of all statewide registrants; and (b) the voter turnout figure for each county as a share of all statewide voters. Alcona's 9,491 registrants constituted .1384% of the state's 6,859,322 registrants in 2000; the 5,997 Alcona voters who voted in 2000 constituted .1417 percent of the 4,232,501 voters statewide that November. (N.B.: These percentages are roughly 14/100ths of a percent, not 14 percent.);
- I then computed the same proportions for the one county-level measure we have thus far - the MBoE registration totals published as July 6, 2004. For example, Alcona's datum is .1362 percent - 9,417 of 6,916,340 registrants;
- I then subtracted the county-level 2000 registration shares from the preliminary 2004 preliminary registrant shares for every county, and did the same with the 2000 turnout shares. I purposely used absolute values, rather that ratios or rates of change, so that county weights obtain, as they should for doing the correlations. (That is, tiny Alcona is weighted less than, say, the larger counties, like Kent or Wayne.)
- Finally, I ran two (Pearson's) correlation statistics for the county-level data: The first correlated the 2000registration/2004registration county-level share changes with the Gore2000 voting percentage for each county; the second correlated the 2000turnout/2004registration share changes with the Gore2000 percentage. A negative statistic would indicate that the more Democratic-voting counties shrunk as a share of the registered voting population, and vice versa for positive statistic.
The results are surprising and, at first blush, perplexing. Using 2000-v-2004-registrant measure, the correlation is -.460, which indicates that the more Republican-voting counties from 2000 have added more registrants. I suspect that suburban/exurban growth in Michigan has changed the landscape as it has in other states, and so it's no surprise, for example, that Macomb County has gone from 7.91% of registrants in 2004 to 8.04% now, while Wayne County is shrinking.
But the correlation statistic when using the 2000turnout-v-2004registrant measure is +.375, which indicates the more Democratic-voting counties have added registrants faster. (Stat-geeks sidebar: These correlates are statistically significant to the .001 level, and I triple-checked the inputs, my Excel formulas, and my SPSS correlates.)
How to square this apparent circle? The most obvious explanation is that the lower turnout rate in Democratic-leaning counties from 2000 creates a comparative baseline that leads to what appear to be, but are not, contradictory results. That is, if more Michigan residents have been registered during the past four years in low-turnout, Democratic-leaning counties, then the registrant-v-turnout growth in these counties can yield a strong, positive correlate at the same time that the changes in the registrant-v-registrant growth yields a strong, negative correlate. (Of course, this analysis is performed at the county level, not at the precinct or ward level, which might produce different results.)
My sense of the findings is that the combined Democratic party and 527 efforts have succeeded in registering people in precisely the places where there are Democratic voters...but who turn out at lower rates - a common phenomenon not only in Michigan, but around the country.
Registrations count in October, sure; but what matters are votes in November. The good news is that increases in registrations in Democratic-leaning counties should be sufficient to counteract the overall registration rate gains in Republican-leaning counties in a state Gore carried by five percent - but only if those newly-registered voters vote.
Now there's a straightforward concept that requires no fancy math whatsoever.
