# Sleeping Your Way to Riches: The Econometrics Edition

A recent paper by economist Dr. Nick Drydakis finds that more sexual activity is correlated with higher wages. The analysis is based on a survey of nearly 5,000 Greek adults, who self-report wages, frequency of sexual activity, and several additional characteristics. The paper uses a sophisticated econometric technique called a “two-stage regression” to make the leap from correlation to causation. This sophistication is necessary, and perhaps not even sufficient, because the relationship between sex and wages is complicated. First, although sex might cause higher wages as hypothesized, it’s equally plausible that the causal arrow goes in the opposite direction: higher wages might result in more sex. The stereotype of rich, sports-car-driving men who are able attract young women as a result of their wealth comes to mind. Second, it’s likely that a third factor (e.g. self-confidence, physical attractiveness) might increase sex and wages independently, thereby creating the positive correlation observed. The first complication is often called reverse causality while the second is called confounding.

In its simplest form, a regression identifies a mathematical relationship between two variables. The two-stage regression, as the name implies, basically involves performing two regressions in sequence such that the output of the first regression becomes the input to the second regression. In this particular application, the first stage predicts sexual activity using factors (let’s call these factors A and B) that are correlated with sex but not with wages. The second stage takes the predictions of sexual activity generated from factors A and B and then uses these predictions to explain wage levels. Because the predictions of sex used in the second regression were generated from factors that are uncorrelated with wages, the second regression, if performed correctly, is no longer subject to concerns of reverse causality and confounding.

To come up with factors A and B, a certain degree of cleverness and creativity is required. What is correlated with sexual activity, but is also independent of wages? Drydakis, the paper’s author, makes the case that religiosity (measured as belief in god and church attendance) fits the bill. In the survey relied upon, a significant and strongly negative relationship exists between religiosity and frequency of sex. Drydakis also asserts that religiosity is uncorrelated with wages, though his own calculations show a moderately positive (albeit statistically insignificant) correlation. Accepting that religiosity is indeed a proper factor to be used in stage 1, Drydakis finds that increasing the frequency of sexual activity by one standard deviation results in a 3.2% increase in wages, holding all else equal.

A couple potential problems come to mind while assessing this paper’s causal inference. First, it is not clear that religiosity is fully independent of wages. Although Drydakis did not find a statistically significant correlation between religiosity and wages, this may be due to a non-linear or more complex relationship (i.e. involving a third, confounding factor). Second, this analysis relies upon self-reporting for all of its key variables: religiosity, sex, and wages. Misrepresenting actual behavior within responses can occur in surveys when questions touch upon highly sensitive issues like money, sex, and religion. The relationship between sex and religion might be further distorted because respondents who are more religious might underreport sex out of guilt or a sense of propriety, while those who aren’t religious might overreport frequency of sex. Finally, the way sexual activity is coded (0 = no sex, 1 = sex once or twice a year, …., 6 = sex more than 4 times per week) creates the false notion that a one unit increase from no sex to annual/semi-annual sex is equal a one unit increase from sex 2-3 times per week to sex 4+ times a week. In all regressions except one, Drydakis assumes that a one unit increase in sexual activity is equivalent across all levels of sexual activity. In the one regression that does not make this assumption, Drydakis has some strange results that don’t dovetail with his conclusions. In that model, sex once a week is actually associated with lower wages than sex 2-3 times per month or even sex once per month. In spite of these objections, this paper does make a valiant effort to quantify the complex relationship between sexual activity and wages.