by: Grebner
Sun May 11, 2008 at 20:37:40 PM EDT
A few days before the 2006 election, State Republican Chair Saul Anuzis sent out a taunting email, claiming the Republicans had launched a boldly re-engineered “Get-Out-The-Vote” drive, whose unprecedented effectiveness would push their candidates several percentage points above their standing in the late polls. Saul believed he had caught the MDP napping, and he wanted to stake his claim before the ballots were counted, in order to get credit for the surprise win he expected.
As we all know, the only surprise on Election Day was how badly Republican candidates did. Granholm and Stabenow were re-elected by very large margins (14 and 16 points, respectively), the Dems swept the statewide education posts, re-captured the State House, came within a whisker of taking the State Senate, and were within hailing distance of upset wins in a couple of Congressional seats – which nobody thought were even in play.
The conventional wisdom is that the landslide was caused by a record turnout, especially among Democrats. This post looks at patterns of turnout at the level of individual voters, seeking confirmation of that hypothesis, but not finding it. To my surprise – shock really – Saul was right. No matter how I look at the data, the numbers come out the same – the Republicans turned out their vote more efficiently than the Democrats did. The Democratic landslide was not based on turnout patterns, but shifts of voter allegiance away from the Republicans.
(This posting is adapted from an article published last year in Bill Ballenger’s Inside Michigan Politics.)
The conventional wisdom is simply that “the Democrats turned out in record numbers”, overwhelming a weak Republican turnout. But looking at the precinct tallies two days after the election, I didn’t find any evidence of a large Democratic turnout advantage – the actual situation seemed to be much more complicated. Heavily Republican precincts turned out at least as well as heavily Democratic precincts. It appeared Democratic candidates did well not because of who turned out to vote, but how they voted. In other words, more or less the usual collection of voters showed up, but there voting patterns shifted toward the Democratic Party. It looked as most Dems voted straight-ticket, most independents leaned heavily toward the Democrats, and even Republicans split their tickets more than usual. That’s not the conventional wisdom – but it’s what the evidence seemed to show.
When data became available from Michigan’s Secretary of State showing exactly which individuals showed up at the polls in November and which didn’t, I linked it with my firm’s coding of political preference, which is the most extensive and accurate data available. Given the Democratic blowout, I expected that when I looked at Democrats and Republicans who have similar past voting records, I’d find a slightly higher turnout among the Dems, simply because recent national events would have tended to motivate the Democrats, while demoralizing the Republicans. That theory was completely wrong!
The actual pattern I found is very simple – but shocking. In every comparison, the Republican turnout was higher than the Democratic, often by five or eight percentage points. In other words, it looks as if the Saul and the Republicans really did out-hustle the Dems. The effect showed up clearly in my logistic regression analysis of voting; the coefficient for partisan orientation was highly significant even when I controlled for age, previous voter history, use of absentee ballots, number of voters in the household, and everything else I could think of.
To create a comparison easier for non-statisticians to understand, I grouped all the potential voters based on their previous voting patterns. In other words, first I looked at people who had never previously voted in either a primary or a general election. Among that group of very weak voters, 16% of the Republicans voted, compared with only 7% of the Dems. At the opposite end of the spectrum, among people who hadn’t missed a primary or general election in 8 years, the Republican turnout was 98%, about equal to the Dem turnout among that group. Stepping through each permutation of voting or not voting in each of sixteen elections, I calculated the 11/2006 turnout for Democrats and Republicans with identical previous histories. I found the Republican advantage averaged across all voting history groups was about six percentage points. That advantage was larger among weak voters, and narrower among people who always vote. Similarly, the average difference was 7% among people who have never voted in an August primary, but only 2% among people who have voted in at least one primary. Whether the difference was large or small, it was consistently in favor of the Republicans.
This pattern of a consistent, substantial Republican turnout advantage – in the midst of a Democratic landslide – leaves me puzzled. I’ve never seen anything this large before. I can’t tell why I didn’t find evidence of Republicans were demoralization or Democratic enthusiasm. All I can say is that I’ve looked at the records of the 3.9 million people who voted, and the data shows what it shows.
Turnout was especially lackluster among Democratic voters who are black, regardless of whether they live in Detroit or elsewhere. This suggests that the underlying cause was a less effective G-O-T-V program, since the Michigan Democratic Party program inevitably concentrates most of its resources on black voters. The turnout gap was smaller, but clearly present, among white Democrats.
Voters identified on my files as independents/ticket-splitters had turnouts that were exactly halfway between the Dem and GOP turnouts.
This entire line of reasoning is so unexpected, I’m sure it will be resisted. But I’d be surprised if anybody who looks carefully at the data comes to a different conclusion.
This higher Republican turnout didn’t make as much difference as you might think – the impact of get-out-the-vote drives is always over-hyped. For one thing, high-frequency voters generally comprise the bulk of the electorate that actually shows up on election day, and they aren’t affected very much by get-out-the-vote drives. Secondly, each percent added by a party to the electorate only increases its percentage by 1/2%. (That is, imagine a group of 100 voters that’s evenly divided. If you turn out one additional voter, you raise the tally to 51-50, which equals 50.5%)
As I calculate it, the benefit to the Republicans of pushing 5% more of their voters to the polls was to raise their statewide percentages in the statewide education races by almost exactly 1 percent – from 44%, up to 45%. Roughly the same math applies to higher profile races – it probably reduced Granholm’s and Stabenow’s margins of victory by about 2%. In a close election that might have made a difference, but in 2006, many supposed “Republicans” were so disgusted by national events they actually split their tickets; habitual ticket-splitters voted predominantly for Democratic candidates; while Democrats showed no inclination to do anything but vote straight-ticket. In spite of their morale problems, the Republicans seem to have designed and executed their program very well. But in the end, it was about as useful as a New Orleans homeowner buying a new sump-pump in preparation for Katrina.
Here is a simplified version of the analysis, dividing all voters into bands based on how many times they voted in the four August primaries and three even-year general elections from 2000 until August 2006. The turnout reflects the fraction who voted 11/2006.
VOTES REPS RTURNOUT INDS ITURNOUT DEMS DTURNOUT DLAG
0 46170 34.1 380635 21.5 117201 20.2 16291
1 127446 47.8 560518 41.8 237898 39.6 19508
2 189651 66.0 447759 60.3 221998 60.5 12210
3 292733 83.7 414448 79.1 256517 79.0 12056
4 213163 89.6 238325 86.2 215778 86.6 6473
5 185887 93.1 153401 90.9 174202 91.2 3310
6 161060 94.9 106027 93.5 149762 94.0 1348
7 170346 98.5 71806 98.2 128708 98.4 129
TOTAL 1386456 81.6 2372919 60.2 1502064 71.1 71325
Data from the current QVF, compared with PPC’s partisan indexes for voters. I excluded duplicate registrations and voters who appear to have died or moved.
VOTES – the number of elections in which each person voted, from Aug ’00 through Aug ’06.
REPS – the number of voters on PPC’s file coded as 75%+ likely to vote Republican.
RTURNOUT – for Republicans in each voting group, the percentage who cast votes 11/2006.
INDS – the number of voters 26-74% likely to vote Democratic.
ITURNOUT – notice that the AVERAGE turnout percentage (when combining all the groups of good and poor voters) is very low compared to the DTURNOUT and RTURNOUT, but that it’s almost exactly equal to the DTURNOUT when compared for each group.
DEMS – voters 75%+ likely to vote for Democrats
DTURNOUT –
DLAG – the number of Democratic votes which would have been added, if Dems had the same turnout percentage as Republicans who had cast ballots in the same number of elections.
As you can see, by this analysis – which uses a completely different approach from the one I describe above – the Dems left 70,000 votes on the table. If they had turned those people out, Granholm’s percentage would have been almost 1.0% higher, and DeVos would have been nearly 1.0% lower.
Leave a Reply