Predicting Turnout in a Presidential Election

by HAL MALCHOW
Campaigns & Elections, September 2004, volume 25, issue 8, pages 38-40

With Election Day only two months away, many campaigns are grappling with two questions. Which voters will show up on Nov, 2? And which voters should be the targets of get-out-the-vote (GOTV) activities?

This is the presidential election. Turnout is higher than in any other political contest. That higher turnout means that a high percentage of voters are almost certain to vote and need little encouragement to show up at the polls. And among the rest of the electorate, everyone is a good prospect to cast a ballot.

Who Will Vote?

Of all of the targeting problems a campaign might face, predicting voter turnout is the easiest, because unlike predicting candidate preference you know exactly what each voter did in the past. Most voter lists contain an extensive voting history for each voter. You know who votes in every election, who is a sometime voter, and who hardly ever shows up at all.

But voter history is just one factor in predicting turnout. A number of other variables can help refine your predictions and tell you, with surprising accuracy, the likelihood that any voter will show up at the polls. In Ohio in 2000, several factors predicted turnout.

The first factor is age. The pattern in Ohio is one of the most universal patterns in politics. Voters in their 20s vote at low levels. Participation increases for each age group until the 60s, then starts to fall off in the 80s and 90s. A little less than half of voters under 30 will show up to vote, but almost all other age groups have a turnout level in excess of 70 percent.

Many political scientists will tell us that income and education are important predictors of voting participation. That may be true in predicting registration. But among voters already registered, the level of income or education in the neighborhood predicts turnout only slightly. In Ohio, the lowest income neighborhoods turn out slightly less than middle or upper income neighborhoods, but once neighborhood income reaches the $40,000 to $50,000 there is little difference at all. Neighborhood education levels showed an even smaller impact.

Likewise, there was little difference in turnout by ethnicity. In Ohio, black neighborhoods turned out slightly less than white neighborhoods, but the difference was not great. In other states, this same analysis would show almost no difference at all. The one exception to this pattern is Hispanic voters who consistently vote at lower levels than either black or whites.

Another important predictor of voter turnout is party registration. There is seldom much difference between the turnout levels of registered Democrats and Republicans. But Independents regularly participate at levels substantially less than the registrants of the two major parties. Ohio does not have party registration. However, in 2000, the registered Democrats and the Republicans in Florida turned out at roughly equal levels. Those not affiliated with either party voted at a level 15 points lower.

Finally, length of registration is an important predictor of voting participation. In low turnout municipal elections, length of registration can be as important as voter history in predicting turnout. But even in a presidential election it matters. In the case of Ohio in 2000, voters who had been registered for less than 10 years voted at rates 20 points less than voters registered 10 years or more. New registrants (four years or less) can make up as much as 30 percent of the electorate. For these voters, there is little voting history, so knowing their turnout levels is important.

Finally, previous voting history is the most important predictor. Voting in the four previous major elections in Ohio--the general and primary elections of 1998 and 1996--impacted turnout in 2000. Voting in even one of those elections meant a voter had at least a 70 percent probability of voting. Somebody having voted in two of the four previous elections translates to a 90 percent probability of voting.

Creating Voter Predictions

So how do you take this information and convert it into a scheme that predicts turnout for individual voters? One way is to use turnout tables. Turnout tables utilize age and previous voting history to calculate a probability that any voter will turnout in an upcoming election

To use a turnout table, you decide what turnout range you are looking for and select the voters with the appropriate scores. For example, you are planning a persuasion mailing and want to send it to Independent voters who have at least a 60 percent chance of voting. Using the turnout table, you would select all Independent voters who voted in at least two of the last four elections as well as voters between the ages of 30 and 79 who voted in at least one out of the last four elections.

You want to target supporters who are between a 25 percent and a 75 percent chance of voting for GOTV. You will notice that in a high turnout election such as this one, there are no voters whose voting probability is less than 25 percent. So we would eliminate voters with a probability greater than 75 percent. That would include all voters who voted in at least three out of the last four elections. It would also include voters between the ages of 30 and 79 who had voted in two of the last four elections. Finally, it would include voters between 50 and 59 who voted in one of the last three elections.

And what about new registrants where little or no voter history is available? Where no voter history is available, you can fall back on age. If you wanted to send mail to 60 percent probability voters, you would select new registrants between the ages of 30 and 89.

These turnout tables are based upon elections in several states where voting patterns may be different from your own election. Overall, turnout patterns are remarkably similar from one state to another. But these turnout tables will not be as accurate as turnout estimates based upon the 2000 election in your own state. To compute turnout tables for your own campaign, categorize voters by age groups and voting history. Then look at the turnout in your own state in 2000 and compute voter turnout in that election for each of your age/voter history categories. The exercise can be done by anyone with a little skill in using a database program.

Finally, there is a still better way to predict turnout. That method is regression analysis. Regression analysis considers voter history, age, party registration and any other predictor that works in your own state. Regression analysis will examine all the factors that predict turnout, weigh them and produce an equation that is used to calculate the turnout probability of each voter. Regression analysis is more accurate than turnout tables and provides a highly reliable prediction. Comparing predictions to actual post-election turnout has repeatedly validated these predictions.

If your party organization does not provide turnout probabilities, you can sometimes obtain them through your voter file vendor or you can license them for about $1,500 per state.

In approaching the issue of turnout targeting, it is important to emphasize that probabilities can give your campaign a much better targeting scheme. In the past, many campaigns viewed the turnout issue as a yes/no proposition. Will the voter vote or will he stay home. We can never know that answer in advance. But we can know probability.

More importantly, probability tells us something about each voter that a yes/no decision can never do. Probability measures value. For persuasion communications, probability tells you that a 90 percent probability voter is more valuable than a 70 percent probability voter-even though both are likely to vote. If you mail a hundred 90 percent probability voters your mail will likely reach 90 voters. If that mail cost 50 cents to send, then the cost of reaching a voter is about $.56. Using the same math, the cost to reach a voter when mailing 70 percent probability voters is $.71. So there is a big difference in value. In allocating scarce campaign resources, knowing a voter's turnout probability is a valuable tool that can stretch a campaign's dollar.

Turnout Strategies in a Presidential Election

Once you have created or purchased turnout probabilities, you need to think about how to use them. Obviously, higher turnout voters are your most efficient targets for mailings and other voter contact. The tougher question is voter turnout.

Many smaller campaigns would be well advised to avoid spending money on GOTV in the presidential election year. For eight to 10 months, voters will have been bombarded with information about the national campaign. In many states, party organizations and other groups are mounting large, and expensive voter turnout efforts. The ability of a non-presidential campaign to influence turnout will be small. Studies done by Don Green and Alan Gerber, two political science professors at Yale, have actually measured the impact of phone calls, mailings and door-to-door visits in lower turnout elections [See Campaigns & Elections, May 2004, p. 51]. While these activities do increase turnout, the impacts tend to be small even in lower turnout elections. In a presidential election, these activities are sure to have a lesser impact.

If you do undertake turnout activities, which voters should you target? There is little value in spending money to turnout a voter who is already 90 percent likely to vote. Logic would suggest that the perfect target for turnout activities would be the 50 percent probability voter. Under that strategy, your target might be to target turnout activities to supporters whose voting probability is between 30 percent and 70 percent or even 20 percent to 80 percent.

But Green and Gerber's experiments with more than 25 campaigns suggest that the voters most likely to be influenced by mail and phones are partisans whose vote likelihood is in the higher range. Their data suggests that you have a better chance of reaching and motivating voters who are 70 percent likely to vote than the voter who is only a 30 percent likelihood to cast a ballot.

Whatever strategy you undertake, turnout targeting can help you pinpoint your target and do so with highly accurate results.

Turnout Table for 70% Turnout Elections