American political-party affiliation as a predictor of usage of an adultery website

American political conservatism emphasizes maintenance of the social order, individual interests (as opposed to collective interests), limited government, aggressive foreign policy, and traditional or religious values. The American political arena is dominated by two major parties, Republican and Democratic, with the Republican Party being more conservative. In general, more politically conservative Americans are also more sexually conservative, in both the attitudes they espouse and the behavior they self-report. Fried (2008), comparing Republicans and Democrats in General Social Survey (GSS) data, found that Democrats saw premarital sex as more acceptable, had had more sex partners, and were more likely to have had a same-sex partner or participated in prostitution (whether buying or selling). Gallup (2003) found similarly that 80% of liberals but only 42% of conservatives saw premarital sex as acceptable. A meta-analysis by Whitley and Lee (2000) found that conservative views, as well as the related personality constructs of authoritarianism and social-dominance orientation, were associated with more negative attitudes towards homosexuality. Since more religious people (especially conservative Protestants) are more likely to be Republicans than Democrats (Pew Research Center, 2015), and religions (especially conservative Protestantism) tend to promote sexual conservatism, it is no surprise that more religious and more religiously conservative Americans tend to also have more negative views of premarital sex (Petersen & Donnenwerth, 1997), homosexuality (Hill, Moulton, & Burdette, 2004), and pornography (Carroll et al., 2008).

One might wonder how much conservatives' actual sexual behavior is in line with their attitudes, and how honest they are in reporting their behavior. Sex can be tempting, and people with stricter sexual standards should be all the less willing to admit their indiscretions to prying researchers. Some writers, inspired by sex scandals involving conservative public figures such as Republican senator Larry Craig (McArdle, 2007) and commentator Bill O'Reilly (Steel & Schmidt, 2017), have speculated that more sexually conservative people are paradoxically more sexually deviant in practice. Some support for this idea comes from two state-level analyses. Edelman (2009) found that states with more conservative attitudes towards religion, sexuality, and marriage buy more subscriptions per capita to pornographic websites. MacInnis and Hodson (2015) found that more religious states had more Google searches for the word "sex," and more politically conservative states had more Google Images searches for "sex."

Why might stricter attitudes be associated with less restricted behavior? Freud (1910) popularized the idea that when people refuse to act on sexual urges, possibly unconscious ones, out of ethical concerns or other restraints of civilization, they may "flee to sickness, in order to find with its help a surrogate satisfaction" (p. 213). According to Freud, the "surrogate satisfaction" could be neurotic behavior with no apparent sexual significance, but it could also be sexual "perversion." Freud also argued for a hydraulic model of male sexuality in which going without ejaculation created "pressure on the walls of the seminal vesicles" (quoted in Apfelbaum, 1984). However, empirical evidence that unexpressed sexual desire builds up over time is weak at best (see Arfer, 2016, for a review). To return to conservatives in particular, some investigators have even taken sexual repression (Sidanius, 1985) or sexual preoccupation (Adorno, Frenkel-Brunswik, Levinson, & Sanford, 1950) to be a defining feature of the right-wing personality.

Even when simple self-report of sexual behavior is examined, we find evidence that people fall short of their own standards. In a sample of urban adolescents, Zabin, Hirsch, Smith, and Hardy (1984) found that 83% of those who had had sex endorsed a best age of sexual debut that was older than their own age of sexual debut. Similarly, 88% of the women who had had a child endorsed a best age for one's first pregnancy that was older than their own age of first pregnancy. In college students who had had sex or petted outside a concurrent committed relationship, Feldman and Cauffman (1999) found the mean acceptance of such activity to be 1.86 on a scale from 1 (totally unacceptable) to 4 (totally acceptable). At the same time, both these studies found positive relationships between people's acceptance of such behavior and their reports of committing it.

By discussing discrepancies between attitudes and self-reported behavior, we do not mean to imply that people are always honest about their behavior. On the contrary, studies have found that the bogus-pipeline method, in which people are falsely told they are subject to some kind of lie-detecting device, can change the rate at which people report deviant behavior. Alexander and Fisher (2003) found that a bogus pipeline increased women's odds of reporting masturbation and pornography consumption. Strang and Peterson (2016) found that a bogus pipeline increased men's odds of reporting sexual assault. Oddly, Fisher and Brunell (2014) found that a bogus pipeline decreased men's odds of reporting extradyadic sexual activity. Overall, while Zabin et al. (1984) and Feldman and Cauffman (1999) point to a gap between attitudes and behavior, and bogus-pipeline studies show that self-report may not be honest, neither kind of study offers a full alternative to self-report.

In this study, we will focus on sexual infidelity, particularly adultery, which is reasonably common (17% of GSS respondents said they had had extramarital sex; Burdette, Ellison, Sherkat, & Gore, 2007) but widely frowned upon regardless of political orientation (a survey of 1,005 people found that 6% of conservatives and 9% of liberals deemed it acceptable; Gallup, 2003). More religious and more religiously conservative people are less likely to state that they or their spouse have had extramarital sex (Burdette et al., 2007; Tuttle & Davis, 2015; but contrast Atkins, Baucom, & Jacobson, 2001), as do Republicans compared to Democrats (Fried, 2008), and more politically conservative people estimate adultery in their marriages as less likely (Buss & Shackelford, 1997).

Our study obtained an objective, individual-level measure of adultery (more precisely, intention to cheat in one's own or another's committed relationship) by drawing on data from the 2015 leak (Victor, 2015) of Ashley Madison (AM; http://www.ashleymadison.com), a dating website that specializes in cheating (slogan until 2016: "Life is short. Have an affair."). Previous uses of this data in social science include Grieser, Kapadia, Li, and Simonov (2016) and Griffin, Kruger, and Maturana (2016). We linked AM users with voter registration records to determine how usage of AM could be predicted on the basis of political-party affiliation, gender, age, and state.

We present two basic analytic approaches: a simple approach that groups the sample into bins and counts frequencies, and a more complex approach based on logistic regression and cross-validation. Our analyses included subjects belonging to no party as well as to any of four parties, namely, the two major American parties, Democratic and Republican, plus two of the largest minor parties, Green and Libertarian. In our informal judgment, Greens are less conservative than Democrats, whereas Libertarians are harder to rate in terms of overall conservatism, but on average are more conservative than Republicans. Note that as shown in Table 1, not every state listed any voters registered as Greens or Libertarians.

In the simple approach, we counted how many voters used AM within each state and party. The results are shown in Figure 1. We see that across all five states, Democrats had the least AM usage of all party groups, and in the four states where Libertarians are present, they used AM the most, with a rate of at least 1 in 300 in each state. Republicans and unaffiliated voters had intermediate rates that were similar to each other. Greens used AM more than Republicans in California and Florida, but less in New York. As for effects of state alone, ignoring parties (and hence including voters in parties other than these four), California had the most AM usage (1 in 560 voters), Oklahoma had the least (1 in 943), and the remainder were intermediate (1 in 659 for Kansas, 1 in 665 for New York, and 1 in 698 for Florida).

g/amr

The analysis represented by Figure 1 is easy to understand, but crude, because it does not account for, for example, gender differences in party affiliation. In our data, 63% of Libertarians but only 42% of Democrats were male, and 1 in 301 men but only 1 in 8,686 women used AM, so it is worth distinguishing the effect of party from the effect of gender.

To estimate the predictive value of party above and beyond confounding variables such as age and gender, we investigated a family of logistic regression models, each regularized with an L₂ (ridge) penalty. Logistic regression is an appropriate technique for modeling the probabilities of binary events. Regularization, also known as shrinkage methods, improves predictive accuracy by biasing regression coefficients towards 0 and hence countering overfitting, the tendency of statistical models to mistake noise for signal. Our models treated each voter as a Bernoulli trial (that is, a 0-or-1 binary event) with the outcome being whether the voter used AM. The models used state, gender, age, and party affiliation as predictors. Subjects with unimputed genders, missing ages, or parties other than the four analytic parties were removed, leaving a sample of 44,172,769 voters, 69,023 of which (1 in 640) used AM. State and party affiliation were dummy-coded, with New York and unaffiliated voters as the respective reference groups. Age was coded as the difference between the voter's year of birth and 2012, then standardized to have mean 0 and SD 0.5 (Gelman, 2008), and supplemented with a similarly standardized age-squared term.

We considered six models, which were distinguished by the choice of predictors, drawing from among state, gender, age, and party. All nontrivial models were fit with Python 3.6 and Hy 4fce884, using the LogisticRegression class from scikit-learn 0.18.1. We used tenfold cross-validation to estimate the predictive accuracy of each model, randomly splitting the voter data into ten folds and then, for each fold in turn, fitting model parameters using the other nine folds and having the model predict the probability of using AM for each voter in the held-out fold. The L₂ penalty size C was chosen among {10⁻¹², 10⁻¹¹, …, 10¹¹, 10¹²} using inner rounds of fivefold cross-validation. We compared predicted probabilities to actual probabilities with mean squared error (MSE), also known in the context of probabilistic classification as the Brier score, which is a proper scoring rule (Brier, 1950; Bröcker, 2009).

Descriptions of each model and cross-validation results are shown in Table 2. By MSE, the models from least to most accurate were Trivial, Parties, Demo, IntDemo, DemoParties, IntDemoParties. Hence, not only does political party have predictive value on its own; it adds to the predictive ability of a model already accounting for age, gender, and state. Examining p₀ and p₁ helps to get a more concrete idea of this predictive performance. p₀ is the mean predicted probability of AM usage among voters who did not actually use AM, so that a p₀ of 1 in 500 would mean that the model predicted on average that nonusers had a 1 in 500 chance of using AM. p₁ is the same value among voters who did use AM. Thus, more accurate models have smaller p₀s but larger p₁s. We see that p₀ does not improve much in any case: the models were generally unable to identify nonusers as less likely to use AM than the base rate. However, p₁ improved substantially as predictors were added to the model, and interaction terms also improve p₁, although only slightly. We will emphasize DemoParties in interpretation, since it is almost as accurate as IntDemoParties while being substantially simpler, and our interest in this study was more in the overall effects of party affiliation than interactions of party affiliation with state, age, or gender.

Table 2. Results of the cross-validation analyses with the six models. "Terms" counts the number of terms in the model, including the intercept. p₀ is the mean predicted probability of AM usage among voters who did not actually use Ashley Madison, and p₁ is the same value among voters who did use Ashley Madison.

Model	Description	Terms	MSE	p0	p1
Trivial	No predictors	1	0.001560127	1 in 640	1 in 640
Parties	Predictors: party only	5	0.001559982	1 in 640	1 in 604
Demo	Predictors: state, gender, age	8	0.001555791	1 in 642	1 in 232
DemoParties	Predictors: state, gender, age, party	12	0.001555600	1 in 642	1 in 226
IntDemo	Demo, plus all first-order interactions	22	0.001555783	1 in 642	1 in 231
IntDemoParties	DemoParties, plus all first-order interactions	51	0.001555565	1 in 642	1 in 224

Having validated DemoParties and IntDemoParties as predictively accurate, we fit each model to all of the data (again using fivefold cross-validation to choose C) and examined the coefficients, which are shown in Table 3. We see that women were especially unlikely to use AM (since the coefficients are in logit units, the value −3.31 indicates that a male voter with a 50% chance of using AM would have his probability reduced to 3.5% if he were female); recall, however, that this effect is likely to have been inflated by women's underrepresentation in the AM credit-card data. In DemoParties, the terms for age describe a positive effect on AM usage from ages 25 to 63, peaking at age 44 with an effect of +1.04. Compared to New York, all states except California were associated with less AM use (except that IntDemoParties gives Florida a very small positive effect). In order from least to most AM-using, the parties were Democratic, unaffiliated, Green, Republican, and Libertarian. This agrees with the simple analysis of Figure 1, with the possible exception of the position of the Greens, which was also inconsistent between DemoParties and IntDemoParties.

Table 3. Coefficients of the DemoParties and IntDemoParties models when fit to all of the data. The reference category for the party terms is unaffiliated, and the reference category for the state terms is New York.

Term	DemoParties	IntDemoParties
Intercept	−6.20	−6.22
Age	−1.07	−1.10
Age²	−2.23	−2.14
Female	−3.31	−3.13
Female × Age		0.52
Female × Age²		0.43
Female × Party: Democratic		0.18
Female × Party: Green		−0.08
Female × Party: Libertarian		0.03
Female × Party: Republican		−0.01
Female × State: California		−0.14
Female × State: Florida		−0.14
Female × State: Kansas		−0.23
Female × State: Oklahoma		−0.23
Party: Democratic	−0.18	−0.17
Party: Democratic × Age		−0.13
Party: Democratic × Age²		0.08
Party: Green	0.02	−0.19
Party: Green × Age		0.22
Party: Green × Age²		−0.29
Party: Libertarian	0.46	0.47
Party: Libertarian × Age		−0.34
Party: Libertarian × Age²		−0.34
Party: Republican	0.19	0.28
Party: Republican × Age		−0.21
Party: Republican × Age²		0.05
State: California	0.14	0.16
State: California × Age		0.22
State: California × Age²		−0.19
State: California × Party: Democratic		0.06
State: California × Party: Green		0.23
State: California × Party: Libertarian		−0.15
State: California × Party: Republican		−0.16
State: Florida	−0.04	0.01
State: Florida × Age		0.19
State: Florida × Age²		−0.22
State: Florida × Party: Democratic		−0.13
State: Florida × Party: Green		0.42
State: Florida × Party: Libertarian		−0.25
State: Florida × Party: Republican		−0.10
State: Kansas	−0.10	−0.16
State: Kansas × Age		−0.33
State: Kansas × Age²		−0.59
State: Kansas × Party: Democratic		−0.07
State: Kansas × Party: Libertarian		−0.08
State: Kansas × Party: Republican		−0.25
State: Oklahoma	−0.31	−0.15
State: Oklahoma × Age		0.23
State: Oklahoma × Age²		−0.14
State: Oklahoma × Party: Democratic		−0.26
State: Oklahoma × Party: Republican		−0.20

Table 4 provides an example to make the models' predictions concrete. The order of parties was the same between the two models except for, again, the position of Greens. We see that while all men of this age in New York were expected to have a higher rate of AM usage than the base rate (1 in 640), party affiliation can still make a big difference.

Table 4. Predicted probability of Ashley Madison usage for a 40-year-old man registered to vote in New York, broken down by party affiliation and model.

Party	DemoParties	IntDemoParties
Libertarian	1 in 117	1 in 98
Republican	1 in 152	1 in 138
Green	1 in 180	1 in 219
Unaffiliated	1 in 184	1 in 189
Democratic	1 in 219	1 in 223

Using two analytic strategies, one simple and one complex, we found that a registered voter's probability of spending money on AM for something other than deleting their account—hence, apparently intending to cheat in a romantic relationship—varied substantially based on their political party. Libertarians were most likely to use AM, Democrats were least likely, and Republicans, Greens, and unaffiliated voters were in between. This pattern mostly coincides with political conservatism, with members of more conservative or more right-wing parties using AM more often.

Adorno, T. W., Frenkel-Brunswik, E., Levinson, D. J., & Sanford, R. N. (1950). The authoritarian personality. New York, NY: Harper.

Alexander, M. G., & Fisher, T. D. (2003). Truth and consequences: Using the bogus pipeline to examine sex differences in self-reported sexuality. Journal of Sex Research, 40(1), 27–35. doi:10.1080/00224490309552164

Apfelbaum, B. (1984, November). Sexual reality and how we dismiss it. Paper presented at the meeting of the American Association of the Advancement of Science, San Francisco State University, San Francisco, CA. Retrieved from https://web.archive.org/web/20010627190317/http://www.bapfelbaumphd.com/Sexual_Reality.html

Arfer, K. B. (2016). Diversity of sexual practices is not bad. In Empirical sexual attitudes. Los Angeles, CA: Author. ISBN 978-1-5238-1453-4. Retrieved from http://arfer.net/w/esa/practices

Atkins, D. C., Baucom, D. H., & Jacobson, N. S. (2001). Understanding infidelity: Correlates in a national random sample. Journal of Family Psychology, 15(4), 735–749. doi:10.1037/0893-3200.15.4.735

Biddle, S. (2015, August 19). What's actually inside the nightmare Ashley Madison leak. Gawker. Retrieved from http://gawker.com/1725077740

Blevins, C., & Mullen, L. (2015). Jane, John… Leslie? A historical method for algorithmic gender prediction. Digital Humanities Quarterly, 9(3). Retrieved from http://www.digitalhumanities.org/dhq/vol/9/3/000223/000223.html

Brier, G. W. (1950). Verification of forecasts expressed in terms of probability. Monthly Weather Review, 78(1), 1–3. doi:10.1175/1520-0493(1950)078<0001:VOFEIT>2.0.CO;2

Bröcker, J. (2009). Reliability, sufficiency, and the decomposition of proper scores. Quarterly Journal of the Royal Meteorological Society, 135(643), 1512–1519. doi:10.1002/qj.456

Burdette, A. M., Ellison, C. G., Sherkat, D. E., & Gore, K. A. (2007). Are there religious variations in marital infidelity? Journal of Family Issues, 28(12), 1553–1581. doi:10.1177/0192513X07304269

Buss, D. M., & Shackelford, T. K. (1997). Susceptibility to infidelity in the first year of marriage. Journal of Research in Personality, 31(2), 193–221. doi:10.1006/jrpe.1997.2175

Carroll, J. S., Padilla-Walker, L. M., Nelson, L. J., Olson, C. D., Barry, C. M., & Madsen, S. D. (2008). Generation XXX: Pornography acceptance and use among emerging adults. Journal of Adolescent Research, 23(1), 6–30. doi:10.1177/0743558407306348. Retrieved from https://www.researchgate.net/profile/Laura_Padilla-Walker/publication/247721479/links/00b7d53455d09d6fcd000000.pdf

Coleman, L. M., & Testa, A. (2008). Sexual health knowledge, attitudes and behaviours: Variations among a religiously diverse sample of young people in London, UK. Ethnicity and Health, 13(1), 55–72. doi:10.1080/13557850701803163

Edelman, B. (2009). Red light states: Who buys online adult entertainment? Journal of Economic Perspectives, 23(1), 209–220. doi:10.1257/jep.23.1.209

Feldman, S. S., & Cauffman, E. (1999). Your cheatin' heart: Attitudes, behaviors, and correlates of sexual betrayal in late adolescents. Journal of Research on Adolescence, 9(3), 227–252. doi:10.1207/s15327795jra0903_1

Fisher, T. D., & Brunell, A. B. (2014). A bogus pipeline approach to studying gender differences in cheating behavior. Personality and Individual Differences, 61-62, 91–96. doi:10.1016/j.paid.2014.01.019

Freud, S. (1910). The origin and development of psychoanalysis. American Journal of Psychology, 21(2), 181–218. doi:10.2307/1413001

Fried, J. (2008). Democrats and Republicans—rhetoric and reality: Comparing the voters in statistics and anecdotes. New York, NY: Algora. ISBN 978-0-87586-603-1.

Gallup, G. H., Jr. (2003). Current views on premarital, extramarital sex. Retrieved from http://www.gallup.com/poll/8704/x.aspx

Gelman, A. (2008). Scaling regression inputs by dividing by two standard deviations. Statistics in Medicine, 27(15), 2865–2873. doi:10.1002/sim.3107

Geuss, M. (2016, December 14). Hacked cheating site Ashley Madison will pay $1.6 million to FTC for breach. Ars Technica. Retrieved from https://web.archive.org/web/20161214/https://arstechnica.com/tech-policy/2016/12/hacked

Grieser, W. D., Kapadia, N., Li, R., & Simonov, A. (2016). Fifty shades of corporate culture. Retrieved from http://ssrn.com/abstract=2741049

Griffin, J. M., Kruger, S. A., & Maturana, G. (2016). Do personal ethics influence corporate ethics? Retrieved from http://ssrn.com/abstract=2745062

Hill, T. D., Moulton, B. E., & Burdette, A. M. (2004). Conservative Protestantism and attitudes toward homosexuality: Does political orientation mediate this relationship? Sociological Focus, 37(1), 59–70. doi:10.1080/00380237.2004.10571234

Independence Party of New York. (2014). About/ Legislative Agenda | Independence Party of New York State. Retrieved from http://web.archive.org/web/20141106202740/http://www.nyindependenceparty.com/agenda

Kirby, D. (2007). Abstinence, sex, and STD/HIV education programs for teens: Their impact on sexual behavior, pregnancy, and sexually transmitted disease. Annual Review of Sex Research, 18(1), 143–177. doi:10.1080/10532528.2007.10559850

Lamont, T. (2016, February 27). Life after the Ashley Madison affair. The Guardian. Retrieved from https://www.theguardian.com/technology/2016/feb/28/what-happened-after-ashley-madison-was-hacked

MacInnis, C. C., & Hodson, G. (2015). Do American states with more religious or conservative populations search more for sexual content on Google? Archives of Sexual Behavior, 44(1), 137–147. doi:10.1007/s10508-014-0361-8

McArdle, J. (2007, August 27). Craig arrested, pleads guilty following incident in airport restroom. Roll Call. Retrieved from http://web.archive.org/web/20150908081135/http://www.rollcall.com/news/-19763-1.html

Moe, K. (1984). Should the Nazi research data be cited? Hastings Center Report, 14(6), 5–7. doi:10.2307/3561733

Paton, C. (2015, August 24). Ashley Madison hack list: How to download and search leaked adultery website database email addresses. International Business Times. Retrieved from http://www.ibtimes.co.uk/1516808

Petersen, L. R., & Donnenwerth, G. V. (1997). Secularization and the influence of religion on beliefs about premarital sex. Social Forces, 75(3), 1071–1088. doi:10.2307/2580531

Pew Research Center. (2009). Trends in political values and core attitudes: 1987-2009. Author. Retrieved from http://www.people-press.org/2009/05/21/independents-take-center-stage-in-obama-era

Pew Research Center. (2015). A deep dive into party affiliation: Sharp differences by race, gender, generation, education. Retrieved from http://www.gallup.com/poll/8704/x.aspx

Prince, S. J. (2015, August 19). Where can I find Ashley Madison data? Heavy. Retrieved from http://heavy.com/news/2015/08/how-to-download-ashley-madison-full-email-list-dump-torrent

Sidanius, J. (1985). Cognitive functioning and sociopolitical ideology revisited. Political Psychology, 6(4), 637–661. doi:10.2307/3791021

Stanger-Hall, K. F., & Hall, D. W. (2011). Abstinence-only education and teen pregnancy rates: Why we need comprehensive sex education in the U.S. PLOS ONE. doi:10.1371/journal.pone.0024658

Steel, E., & Schmidt, M. S. (2017, April 19). Bill O'Reilly is forced out at Fox News. The New York Times. Retrieved from https://www.nytimes.com/2017/04/19/business/media/bill-oreilly-fox-news-allegations.html

Strang, E., & Peterson, Z. D. (2016). Use of a bogus pipeline to detect men's underreporting of sexually aggressive behavior. Journal of Interpersonal Violence. Advance online publication. doi:10.1177/0886260516681157

Tuttle, J. D., & Davis, S. N. (2015). Religion, infidelity, and divorce: Reexamining the effect of religious behavior on divorce among long-married couples. Journal of Divorce and Remarriage, 56(6), 475–489. doi:10.1080/10502556.2015.1058660

United States Census Bureau. (2012). Table 1. Annual estimates of the population for the United States, regions, states, and Puerto Rico: April 1, 2010 to July 1, 2012 (NST-EST2012-01). Retrieved from https://www2.census.gov/programs-surveys/popest/tables/2010-2012/state/totals/nst-est2012-01.xls

Victor, D. (2015, August 19). The Ashley Madison data dump, explained. The New York Times. Retrieved from http://www.nytimes.com/2015/08/20/technology/the-ashley-madison-data-dump-explained.html

Whitley, B. E., Jr., & Lee, S. E. (2000). The relationship of authoritarianism and related constructs to attitudes toward homosexuality. Journal of Applied Social Psychology, 30(1), 144–170. doi:10.1111/j.1559-1816.2000.tb02309.x

Yoon, H.-J., Tourassi, G., & Xu, S. (2015). Residential mobility and lung cancer risk: Data-driven exploration using internet sources. Lecture Notes in Computer Science, 9021, 464–469. doi:10.1007/978-3-319-16268-3_60. Retrieved from http://www.ncbi.nlm.nih.gov/pmc/articles/PMC4662563/

Zabin, L. S., Hirsch, M. B., Smith, E. A., & Hardy, J. B. (1984). Adolescent sexual attitudes and behavior: Are they consistent? Family Planning Perspectives, 16(4), 181–185. doi:10.2307/2134899