This year at the Midwest Political Science Association’s Annual Meeting I will be presenting a paper that is part of my dissertation and a greater research agenda committed to understating the social and psychological motivations for overreporting turnout in political surveys. In my paper I propose that the assumption that overreporting is the result of socially desirable responding (SDR) can be confirmed through the use of response latency measures. Social psychologists has found that providing deceptive responses reliably increases response latencies (Mayerl 2013; Vendemia, Buzan & Green, 2005; Verschuere et al, 2011; Walczyk et al 2003) along with a link between response latencies and SDR (Holtgraves, 2004). Since overreports of voter turnout are in essence deceptive responses because they provide false information to survey researchers, overreports should result in significantly lengthier response times for the vote self-report question in surveys.
See the slides for my presentation on Saturday April 9th at the panel titled “Why People Vote (or Don’t Vote But Say They Did)” from 4:45-6:15pm.
Vote validation have focused on identifying correlates of overreporting, and not the social psychological mechanism that cause this phenomenon. Only on study has found a link between SDR and overreporting.
There are two types of SDR and they differ in intentionality, one is intention other-deception and the other is unconscious self-deception.
As I mentioned before lying takes time.
To test these expectations I use 2014 CCES data, which has over 56 thousand respondents and includes vote validation data. Since this is an online self-administered survey page timing data was made available for this study. The page timing data in the CCES is measuring the period during which the survey question becomes visible to the respondent, the respondent reads the question, formulates an answer, makes a report and then moves on to the rest of the survey. These page timings will serve as response latency measures.
The overreport variable is a dummy where overreporting is the indicator. I trim the dependent variable, the vote self-report page timing, by eliminating all values above the 95th percentile and eliminating any participants with a value of zero seconds. Eliminating extreme values from the vote self-report page timing allows for the use of mean response latencies in descriptive inference and OLS regression to test the effect of overreporting on page timings. I use a baseline timing control measure in my OLS model. The baseline timing is the calculated average individual timing from 4 questions. Two from the pre-election and two from the post election survey o the 2014 CCES. I convert the values above the 95th percentile to the value at the 95th percentile to not loose observations. The questions in the baseline are similar in construction to the vote self-report question, but differ in that they are not susceptible to socially desirable responding. This baseline controls for any and all individual differences in response latencies, there is no need for other control variables.
I restrict my analysis to all respondents who reported they “definitely went out to vote” in the 2014 general election. the best comparison for this study is that of persons who gave the same response to the same question but were found to be either honest or deceptive using vote validation data.
Descriptive statistics suggest overreporting does increase response latencies for the vote self-report question in the 2014 CCES.
These plots show the distribution of the vote self-report timing and the Baseline timing on either side. The red line indicates the mean page timing for each plot. Though I have included the baseline timing as a control in the OLS model for the effect of overreporting on the vote report timing I also include a placebo model as a robustness check on the baseline. I use the party id timing from the post election survey as my placebo.
Overreporting has a positive and significant effect on increasing response latencies for the vote report question. Overreporting does not affect response lattices for the placebo question.
The following plot shows that the probability of overreporting increases as a function of residual timing controlling for baseline timing.
This paper provides supporting evidence for the widely held, but almost untested assumption that turnout overreports are the result of SDR and concludes that these are intentional other-deception because of the significant cognitive effort involved in falsely reporting turnout.