2012 Turnout Analysis: Methodological Notes

This is a semi-live document where I’ll answer various questions people have about my recent MonkeyCage blog post

  • I used Catalist’s turnout score and Obama support score the two microtargeting scores that I reference in the post. Neither was affected by 2012 campaign activity, thus reducing endogeneity problems. Special thanks to Catalist for the data and helping me understand its nuances.

  • Chris Kennedy rightly points out that other organizations besides the official Obama and Romney campaigns were engaged in targeted GOTV. The correct interpretation is to read “Obama campaign” and “Romney campaign” as stand-ins for “Democratic efforts” and “Republican efforts.” I apologize for the error.

  • Figure 1 was distorted by Monkey Cage. Here’s the real one:

  • Sporadic voters are defined as those having ex-ante turnout probabilities of below 85% (which is about the median probability, and much higher than the mean value of 71%). Campaign effect estimate are much lower among voters with the highest turnout probabilities (>95%); inclusion of borderline GOTV voters (with vote propensity scores 85%-95%) would increase the estimated net Obama effect, but it’s not clear that these voters were targeted by the campaigns, so they are left out. Republicans are defined as having a pre-campaign, Catalist-estimated likelihood of supporting Obama below 40%; Democrats, above 60%. One standard error is shown as error bars in the Figures. The nationwide turnout of 66% is calculated from Catalist’s voter file and includes inactive voters (using official designations) in the denominator.

  • The turnout and partisanship scores were developed by Catalist and used by them in Spring 2012. No 2012 campaign activities affect these microtargeting scores. This helps avoid endogeneity issues.

  • This analysis’ Democratic advantage is consistent with Enos and Fowler’s findings, though my GOTV effects are smaller for the straightforward reason that Enos and Fowler analyze the cumulative effect over time of living in a battleground state. Their turnout effect estimates include the efforts of past campaigns and are more susceptible to competitiveness effect (i.e., more people mobilizing because a race is closer). By controlling for an individual’s a priori expected 2012 vote, this analysis better isolates 2012 campaign interactions. Each type of analysis has its place, depending on what effect a researcher is interested in measuring.

  • The Florida calculation is here. Crucially, I caclulate votes gained not voters gained. The former is more valuable than the latter, because some modeled Democratic supporters will vote Romney and vice versa.

  • The battleground states are: OH, PA, FL, NH, NV, VA, NC, IA, WI, CO, MI

  • Using official party registration from the voter file (for states with party registration) rather than partisanship score reveals a similar pattern. (This protects against the possible bias created by using a private firm’s probabilities.)

  • Spillover effects (e.g., independents living with Democrats who would thus also receive Obama GOTV communication by virtue of a household member being targeted) may bias the estimates in this analysis downward.

  • Aggregated data, including n sizes and standard error calculations, is available upon request and subject to approval by Catalist.

  • Below is “Figure 0” — a chart I wanted to include in the blog post, but which was cut (fairly) for length. It shows the overall turnout underperformance. Note that this figure includes high- and low-propensity voters so that it isn’t exactly comparable with the Figure in the blog post.

This entry was posted in Uncategorized. Bookmark the permalink.