Saturday, August 7, 2010

Estimating Primary Turnout

Even though I’ve already submitted an answer to the Primary Prediction Contest I’m going to take a crack at actually running some numbers to come up with a more accurate turnout prediction.

As you can see from the graph below, turnout in Minnesota primaries as a percentage of eligible voters has fallen over the years.

But turnout in absolute numbers has remained basically the same. This is due to the year by year steady growth in eligible voters.

The peaks in the above graph roughly correspond to when there were competitive primaries, which there hasn’t been the last two cycles. For that reason it’s not really fair to use the turnout percentages of the two previous cycles to estimate this one, with a three person DFL race and an IP primary that will drive turnout higher.

This is from the Winona Daily News supports that conclution:

Nearly 22,000 voters have cast absentee ballots ahead of Tuesday's primary, a level of absentee turnout that already tops the last two gubernatorial elections.

If we were to look to Minnesota’s recent past to find a similar primary 1998 is the first one that sticks out, that year there was an open seat election with multiple DFL challengers. Overall turnout that year was just north of 20%, with the DFL turnout at around 12%. I’m going to use that 12% figure, actually 11.8%, as a baseline for estimating turnout going forward.

Now the question becomes how to factor in the change from a September date to an August one. What I did was to figure out the Democratic turnout, as a percentage of eligible voters, for states with non-summer primaries and for states with summer primaries. I limited this to states that have similar PVI’s to Minnesota’s (within half a standard deviation away), since I’m attempting to measure turnout in a partisan election I figured I should try and control for a states partisan tilt, and I used information from the last two gubernatorial election cycles in those states.

The three non-summer states I used are Ohio (5/4), Pennsylvania (5/18) and Oregon (5/18). The summer election states are New Mexico (6/1), Nevada (6/8) and Michigan (8/3). Why are New Mexico and Nevada considered summer primaries? Because I needed data points, that’s why. If I really wanted to be meticulous about this I could have gathered information from every state and attempted to figure out a turnout curve based on time of year the primary is held, but I didn’t want to devote nearly that much time to this problem. Instead I’m just going to use a small sample of states with similar partisan makeups.

What this analysis shows is that states with summer primaries experience an approximately %15 percent decrease in primary turnout compared with states that have non summer primaries. The three non-summer states average about 11% turnout and the three summer states average a little over 9%.

If we use 15% as our “summer penalty” the turnout estimate comes in at 381,850. There are a few things that need to be considered though, first off Oregon votes by mail which leads to higher turnout in that state than there would otherwise be. Additionally, in Pennsylvania this year the gubernatorial primary was overshadowed by the Sestak-Spector senate primary, which likely drove turnout higher.

Because of those factors and the uncertainty inherent in such a small sample (twelve elections) I’m going to use 10% as a summer penalty, instead of 15%. With a 10% penalty the turnout estimate is 404,312. I would say than that turnout is likely to be between 375,000 and 425,000. That is my actual estimate of turnout.

What if we apply no penalty though, what if the change in the election date is offset by the GOTV efforts of the three DFL campaigns, if this is the case than the turnout estimate is 449,235. We’ll see after the primary which, if any, method arrives at the right conclusion, but here are the guesstimates all together.

No comments:

Post a Comment