Hitting the mark: targeting by turnout

Willie Sutton, a notorious lawbreaker, was asked why he robbed banks. His response: "Because that's where the money is." Targeting campaigns toward likely voters is doing the same thing; it's going where the votes are.

But in campaigns, unlike bank robbery, when you target some voters you are also eliminating others. For example, if you're running in a district with 100,000 registered voters and, based on past experience, you know that only about 20,000 usually vote in primaries, you may decide to select the 40,000 or 50,000 voters who are most likely to vote and concentrate your direct mail, phone banking and door-to-door canvassing to make sure you hit them -- assuming the other 50,000 or 60,000 don't matter because they won't vote.

Eliminating the wrong voters from campaign coverage can be a disaster -- which is why there is a practical need to distinguish between high propensity and consistent voters before you start eliminating large numbers of voters from the scope of your campaign efforts.

It is possibly true, however, that even though voting at rates higher than other voters, high propensity voters may not be consistent voters, and they may not even comprise the majority of persons who indeed vote. A small group voting at a higher rate may still not contribute as much to turnout as a large group voting at a marginally lower rate. These possibilities suggest the importance of evaluating the extent to which those who vote in low turnout elections are consistent voters. The evaluation of consistency in voting is approached through two questions:

Question #1: To what extent do those who vote in one low turnout election vote in the next similar election?

Question #2: To what extent are those who vote in a low turnout election the same people who voted in the prior similar election?

Data obtained from the Voter Contact Services "On-line National Political Database" for two elections in each of three different political jurisdictions can provide some answers. The elections discussed:

* A 1991 city election and run-off in Pomoma CA. The run off was decided by less than 100 votes, the candidates in the run off were both Democrats, on an anglo woman liberal and the other a Chicano male (more) liberal. Both candidates ran extensive voter ID and absentee voter programs and had extensive GOTV efforts. The initial election was March 5, 1991 with the run off six weeks later on April 15.

* The Presidential Primary in March of 1992 and the statewide Primary in August 1992 in Colorado for the voters in the City and County of Denver. Denver is predominantly a Democratic area and there were meaningful primaries on the Democratic side in the Presidential (Brown, Clinton, and Tsongas ran about even) and in the statewide primary for U.S. Senate (Campbell, Lamb and Heath).

* There same races for voters in El Paso County Colorado most of whom are in Colorado Springs. This County is predominately Republican, very conservative and host to many politically active, evangelical, born-again Christian organizations. There were no meaningful primaries on the Republican side at the statewide level, but there were a number of "stealth" candidates sponsored or supported by the religious right. The table below presents the rates of voting in these elections.

Risky Business

These data show about the same number of people voted in each of the two consecutive Ponoma city elections (20.6%) in the first and 21.7% in the second) and the two primary elections in Denver (24.7% in the first and 24.8% in the second) while El Paso County showed a slight decrease in turnout for the second primary (20.8% for the first, vs. 18.4% for the second.) Although the turnout is similar for the first election in each of these jurisdictions, it is clear that those voting in the two consecutive elections are far from being the same people.

For the city of Ponoma with consecutive elections held only six weeks apart, with two continuing candidates, only 71% of those who voted in the first election and only 67.1% of those who voted in the second election has also voted in the first. Of those voting who could be classified as high propensity voters by virtue of having voted in either local elections, only 52.6% who voted in both could be classified as consistent voters.

For Denver, only 58.8% of those voting in the presidential primary voted again in the state and local election held less than six months later, while only 58.7% of those voting in the primary for state and local offices had previously voted in the presidential primary. Of those voting who could be classified as high propensity voters by virtue of having voted in either of the primary elections, only the 41.6% who voted in both could be classified as consistent voters.

In El Paso County 51.4% of those voting in the presidential primary voted again in the state and local election and, 58.2% of those voted in the second primary had also voted in the prior one. Of those voting who could be classified as high propensity voters by virtue of having voted in either of the two primary elections, only 37.5% who voted could be classified as consistent voters.

Even in these "best case" studies, with two consecutive similar elections occurring in a short time span, the extent of inconsistent voting makes clear the risks of focusing a campaign exclusively on high propensity voters. A significant proportion of those identified as most likely to vote from having voted in an immediately prior election will in fact not do so (in these studies, from 29.0% in Ponoma to 48.6% in El Paso County.)

Strategic concern to candidates and campaign consultants, an equal or greater proportion of those who do vote will be those who were not identified as high propensity voters because they did not vote in the immediately prior similar election (from 32.9% in Ponoma to 41.3% in Denver and 41.8% in El Paso County) and could be incorrectly eliminated from campaign activities. Finally, it is noted that about half or less of those who voted in any two consecutive similar low turnout elections are consistent voters -- only 52.6% in Ponoma, 41.6% in Denver and 37.5% in El Paso County of those voting in either of the two similar elections actually voted in both.

Substantial Danger

The bottom line of this analysis is that there is a substantial danger of eliminating up to two thirds of those who may actually vote in a low turnout election by only including those who voted in a prior similar election as the target group of high propensity voters. On the other hand, targeting on the basis of some type of prior vote history is by far the best method to use to identify those who are, statistically, more likely to vote in such elections.

Clearly some caution against over generalizing from these studies is in order, especially since they vary in both "level" (local vs. primary) and the time between each election (six weeks to six months). However serious this limitation in the sample might be, the extent of the similarity of findings strongly suggests the importance of recognizing the risks as well as the rewards of targeting only on voting "propensity" for low turnout elections. The following ideas may prove helpful in countering problems which may well be inherent in the use of past voting history to diminish the universe.

* It is important to think about a target group in two ways. First is the traditional approach of identifying a smaller group of voters who will vote at a high rate. The rate at which persons in the group actually vote can be thought of as a hit rate -- a measure of the efficiency of the selection process. Generally, the more restrictive the measure of past vote history used to create a target group, the higher the hit rate will be, i.e., those who voted in all four past primary elections will vote at a higher rate than those who voted in, say, either of the two past general elections.

* However the more restrictive the criteria used, the smaller is the target group selected. At some point the efficiency gained by focusing on a small group of very likely voters tends to become ineffective for it omits the majority of persons who will vote. So it is equally important to keep in mind the percentage of those who actually vote who have been included in the target group and have been contacted as part of the campaign's strategy. This rate is called the cover rate, or the percent of those voting who had been included in the targeted group and therefore could be "covered" by campaign activities and programs.

Hit and Cover are inversely related -- higher hit rate usually means lower cover rate. One obtains 100% cover by selecting all eligible voters, but the hit rate decreases to the limit of the turnout rate. One obtains close to 100% hit by selecting only those who have voted in all previous elections of the same type in recent years, but the cover rate tends to drop substantially below even 50%.

* Voting propensity reflects conditional probabilities of voting. That's why the type of election used to diminish the universe is not as important as the number of voters included in the targeted group, if it produces satisfactory levels of hit and cover. If concerned about cover one may use prior elections with higher turnout rates, or if concerned about hit rates one may use prior elections with lower rates of turnout. In this formula voting = voting.

* One may also vary the level of consistency in voting in prior elections used to diminish the universe of voters. Selecting those voting in both of two prior elections produces higher hit and lower cover than selecting those voting in either of them. Alternatively, if higher cover is important to win the election a selection of those who voted in either prior elections will be a wiser choice, even though the hit rate may drop.

* Ranking elections by turnout rate facilitates diminishing the universe by using elections with higher or lower turnout rates, as in #3, and also provides a way of evaluating altenative elections which might be used for doing so. Elections might be ranked by turnout percent, from lowest to highest, as: local issue, city, state primary, state general, presidential. The following paired races will produce about the same rates of hit and cover:

EITHER PRIOR LOCAL ISSUE = BOTH PRIOR CITY

EITHER PRIOR CITY        = BOTH PRIOR PRIMARY
EITHER PRIOR PRIMARY     = BOTH PRIOR GENERAL
EITHER PRIOR GENERAL     = PRIOR PRESIDENTIAL

* It may be prudent to select at the most general level a base voter group of those who are identifiably more likely to vote in a lower turnout election than regular voters in general. The Base Voter Group will be comprised of segments, each with differing rates of voting propensity, and meriting different rates of contact by the campaign. Two examples below are drawn for Colorado's August Primary and the 1993 Primary in California. Since turnout for each is expected to be lower than for prior elections, hit rates are somewhat more at issue than cover rates. It is clear that a substantial proportion of those voting in the 92 general are not going to vote in the 1994 non-presidential general.

* A final idea: use a fairly restrictive definition for targeting voters based only on prior vote history and adding other voters who might be assumed to be favorable to the candidate or campaign if they do vote. This is consistent with the general role of working one's strength and may even work to increase turnout among potentially favorable groups.

These ideas are only some of the approaches to offset the inherent problem when targeting campaigns based only on past vote history. High propensity voters are not always consistent voters. Not all, or at times even most, of those who vote in low turnout elections are high propensity voters The inverse relationship between hit and cover appears to be inevitable. Campaign strategists and managers should exercise caution and attention to local turnout and vote behavior in reaching a decision regarding targeting.

COMPARATIVE CONSISTENCY OF VOTING IN CONSECUTIVE ELECTIONS

TOTAL       PONOMA          DENVER             EL PASO CO.
            40,150          259,716             197,536

1ST         8,255 (20.6%)   64,155 (24.7%)   39,089 (20.8%)
ONLY 1ST    2,397 (29%)     26,430 (41.3%)   18,991 (48.6%)
BOTH        5,858 (61.7%)   37,725 (58.7%)   20,098 (51.4%)

2ND         8,728 (21.7%)   64,294 (24.8%)   34,544 (18.4%)
ONLY 2ND    2,870 (32.9%)   26,569 (41.3%)   14,446 (41.8%)
BOTH        5,858 (52.6%)   37,725 (58.7%)   20,098 (58.2%)

EITHER     11,125 (27.7%)   90,724 (34.9%)   55,535 (28.5%)
ONLY 1ST    2,397 (21.6%)   26,430 (29.1%)   18,991 (35.5%)
ONLY 2ND    2,870 (25,8%)   26,569 (29.3%)   14,446 (27.0%)
BOTH        5,858 (52.6%)   37,725 (41.6%)   20,098 (37.5%)

Robert Hughes is a professor of sociology, at the University of Colorado, Colorado Springs

COPYRIGHT 1994 Campaigns & Elections, Inc.
COPYRIGHT 2004 Gale Group