Sex, Drugs and Economics

Saturday, 13 August 2022

Is there a premium for very unattractive workers?

The beauty premium has been widely established in the literature on labour economics. There is a whole book by Daniel Hamermesh on the topic, entitled Beauty Pays (which I reviewed here). However, this 2018 article by Satoshi Kanazawa, Shihao Hu, and Adrien Larere (all London School of Economics and Political Science), published in the journal Economics and Human Biology (ungated version here), presents a more nuanced view. Kanazawa highlight this earlier article (ungated version here) by Kanazawa and Mary Still (University of Massachusetts, Boston), and summarise the findings of that earlier article as:

...while unattractive individuals earned less than others, very unattractive workers always earned more than unattractive workers, sometimes more than average-looking or even attractive workers, seemingly contrary to previous findings in the economics of beauty.

The issue may be that earlier studies group together unattractive and very unattractive workers into a single category, and therefore may fail to identify a difference between the two groups. In the 2018 article, Kanazawa et al. look at this very-unattractiveness premium using data from the National Longitudinal Study of Adolescent Health (Add Health), including in their sample around 10,000 people who participated in all of the first four waves of the study. Rather than looking at labour market outcomes, Kanazawa et al. looked at 'mate value', measured as:

...whether the respondent was currently married (1 if currently legally married, 0 if currently single, excluding cohabitation) or cohabiting with a partner (1 if currently cohabiting, excluding marriage, 0 if currently single) at 29.

Attractiveness was measured on a scale of 1 to 5, measured closest to the time when their current relationship began (although they are not clear on how they treat the attractiveness of people who are not currently in a relationship). Comparing 'mate value' by attractiveness in their basic analysis, Kanazawa et al. find that:

...very unattractive respondents - both men and women - were always more likely to be married or cohabiting than unattractive respondents, sometimes more than average-looking or attractive respondents (except for cohabitation for men, when physical attractiveness was measured at 16 or at match), but this pattern was much stronger for marriage than for cohabitation and among women than among men.

Having re-confirmed the results from the earlier study, Kanazawa et al. then try to tease out the mechanism underlying this finding. They reason that intelligent men are attracted to very unattractive women:

According to the Savanna-IQ Interaction Hypothesis (Kanazawa, 2010) or the intelligence paradox (Kanazawa, 2012), more intelligent individuals are more likely to acquire and espouse “unnatural” preferences and values that go against their evolutionary design. Since men are evolutionarily designed to value physical attractiveness in their mates... more intelligent men may be more likely to prefer to mate with very unattractive women.

I'm not sure that I buy into this theoretical argument [insert duty-bound statement about my marriage being a counter-example to Kanazawa et al.'s theory]. Nevertheless, they do find some modest support for their theory:

...very unattractive women’s spouse or partner always earned significantly more than those of unattractive women except when physical attractiveness was measured at 17... Further, very unattractive women’s spouse or partner earned significantly more than those of average-looking women when physical attractiveness was measured at 29... in sharp contrast, none of the regression coefficients were statistically significant in the same direction among men.

However, there are a couple of problems here. First, Kanazawa et al. don't have a measure for the spouse's intelligence, so they are using earnings as a proxy. So, it may be that other mechanisms underlie the relationship they find wherein higher income men tend to marry very unattractive women. However, a more subtle problem is that the analysis is based on less than 300 people who were rated as very unattractive (at most - the number rated as very unattractive varied based on the age at which attractiveness was measured), and less than 500 people who were rated as unattractive, out of a total of around 10,000 people (this is obvious only from the supplementary materials to the article, as these numbers are not reported in the article itself). It wouldn't take much measurement error here to generate almost any result at all. In fact, Kanazawa et al. note that, in the earlier Kanazawa and Still article:

The mean [inter-rater agreement] for unattractive, about average, attractive, and very attractive individuals ranged from 0.6352 to 0.8280 for women, and from 0.6341 to 0.8527 for men. For very unattractive individuals, however, it ranged from 0.0180 to 0.1398 for women, and from 0.2184 to 0.3890 for men.

So, the proportion of evaluators who agreed on who was very unattractive was very low, especially for women, and it is only for women that Kanazawa et al. find the relationship with spouse' earnings. Low agreement suggests that measurement error in who is evaluated as very unattractive is very high. It would be helpful if Kanazawa et al. had provided an evaluation of how stable the attractiveness ratings were over time. That is, were very unattractive people consistently rated as very unattractive, or was it essentially random? Without knowing that, it is hard to take their results too seriously. With that in mind, I think it's going to take a lot more to make a convincing argument that there is a very-unattractiveness premium.

Life expectancy in the age of COVID-19

It seems self-evident that the coronavirus pandemic has reduced life expectancy, both in individual countries and globally in aggregate. However, just how extraordinary declining global life expectancy is, is not so obvious. In this recent article published in the journal Population and Development Review (open access, with non-technical summary here), Patrick Heuveline (UCLA) summarises the changes. Using a mix of life tables from the United Nations, supplemented by excess deaths data from the World Mortality Database, Heuveline constructed new estimates of life expectancy for each country and each quarter in 2020 and 2021 (using a machine learning algorithm to fill in some of the data gaps). The estimated effect of the pandemic on global life expectancy is summarised in Figure 2 from the article:

As you can see, the long-run increase in global life expectancy was interrupted and reversed course substantially in 2020 and 2021. However, even that figure understates just how remarkable this reversal is. As Heuveline notes (emphasis added):

The increase in the number of deaths during the pandemic had a substantial impact on the global life expectancy. After 69 years of uninterrupted increase from 1950 to 2019, the global life expectancy is estimated here to have declined by −0.92 years between 2019 and 2020 and by another 0.72 years between 2020 and 2021... In 2021, the global life expectancy is estimated to have dropped below its 2013 level.

The coronavirus pandemic ended at least seven decades of increasing global life expectancy. Of course, not all countries have been equally affected. Heuveline reports that:

...many countries experienced substantial changes in life expectancy... Between 2019 and 2021, life expectancy is estimated to have declined by more than two years annually (four years overall) in eight countries... five in America (Peru, 5.6; Guatemala, 4.8; Paraguay, 4.7; Bolivia, 4.1; and Mexico, 4.0 years) and three in Europe (the Russian Federation, 4.3; Bulgaria, 4.1; and North Macedonia, 4.1 years)... Among those with sufficient data, the only countries that did not reach the two-year mark at any point between 2020 and 2021 are countries in Eastern Asia, Australia, New Zealand, and European countries, west of a line running from the Baltic to the Balkans.

For comparison:

Instances of life expectancy declines, from one calendar year to the next, remain rare in the UN time series and relatively modest though. The main exceptions to this generalization are found for Cambodia (up to −4.63 years per year) and Rwanda (up to −5.02 years per year) - two countries that experienced massive increases in violent mortality, in the late 1970s and early 1990s, respectively - and a few sub-Saharan countries during by the AIDS pandemic. According to the UN estimates, the impact of AIDS mortality on life expectancy was most severe in Eswatini in the late 1990s (up to −2.10 years per year).

So, in the worst affected countries the impact of the coronavirus pandemic was roughly equivalent in its impact on life expectancy to the worst years of the HIV/AIDS pandemic in the worst affected country. And, the impact of the coronavirus pandemic in the worst affected countries was nearly half of the annual impact of the Khmer Rouge or the Rwandan genocide. That certainly puts things in perspective.

However, there is one ray of hope in the article:

Comparing life expectancy estimates for each of the eight 12-month periods, however, the decline in global life expectancy appears to have stopped in the last quarter of 2021...

We can only hope that is a signal of a return to increases in global life expectancy (albeit at an apparently decreasing rate over time), and possibly even some catch-up growth. Now, we just need to know what that means for the discrepancies in life expectancy between rich and poor.

[HT: N-IUSSP]

Wednesday, 10 August 2022

The beauty premium and student grades for in-person and online education

The beauty premium is the idea that people who are more attractive are paid more in the labour market. I've posted on this topic many times and, although we usually focus on wages, the premium is not limited to the labour market. There is a small beauty premium in education as well (such that more attractive people receive better grades - see here). So, what happens to the beauty premium when education is moved from in-person (where there are many face-to-face interactions, and attractiveness might pay a relatively larger role in grades) to online (where students can hide behind Zoom with their cameras switched off, limiting 'face-to-face' interactions)?

That is the question that this new article by Adrian Mehic (Lund University), forthcoming in the journal Economics Letters (sorry, I don't see an ungated version online), addresses. Mehic uses data from the Industrial Engineering Programme at Lund University, and restricts his attention to the first two years of the programme, which is made up of 15 mandatory courses in mathematics, physics, computer science, business, and economics. He categorises mathematics and physics as quantitative subjects, and the others as non-quantitative (which would probably come as a surprise to most economists and computer scientists!). The sample includes 307 students, who commenced their studies sometime between 2015 and 2019, and experienced online study from the second semester of the 2019/20 academic year. That means that:

...students who started the program in 2018 had two online courses in their second year, whereas students starting in 2019 had two online courses in their first year, and eight online courses in their second year.

The attractiveness of each student was measured by asking 74 volunteers to rate the attractiveness of each student's ID photo on a scale of 1 (extremely unattractive) to 10 (extremely attractive), with each volunteer rating half of the sample. Then, looking at the relationship between attractiveness and grades overall, Mehic finds that:

When all courses in the program are considered, there is a positive, albeit statistically insignificant relationship between attractiveness and grades.

However, there is a difference between the 'quantitative' and 'non-quantitative courses', and:

...the coefficient for attractiveness is highly significant for the non-quantitative courses. The results suggest that one standard deviation higher beauty is associated with around 0.08σ higher grades. The magnitude of the estimated coefficient is slightly lower when the full set of controls is included. Concomitantly, there is no significant relationship between attractiveness and grades for the non-quantitative courses.

So, there is only a beauty premium for the non-quantitative courses, which Mehic explains may arise due to the following:

In our setting, the tasks faced by students in non-quantitative subjects, for instance in marketing and supply chain management, are likely to be seen as more ”creative”, and significantly contrast the more traditional book-reading and problem-solving in mathematics and physics courses, the latter presumably perceived as more monotonous. Together with the large use of group assignments in non-quantitative courses, these theoretical results imply that socially skilled individuals are likely to have a comparative advantage in non-quantitative subjects.

I'm not sure how much I buy that explanation, as Mehic includes computer science and economics in the non-quantitative courses, and (much as we may wish it was not true) most people would not characterise those subjects as 'creative', even relative to mathematics or physics.

Mehic then turns to a 'triple-differences' analysis, comparing the difference in the effect of beauty on grades between male and female students, and between online and in-person classes. He finds that:

...the triple interaction between Online, attractiveness, and the indicator for non-quantitative course is highly significant for female students. This finding suggests that the grades of female students deteriorated in non-quantitative subjects, with grades declining with attractiveness. There is no equivalent relationship for males.

So, the results show that there is a beauty premium, but the beauty premium exists only in the non-quantitative courses overall. Also, the shift to online teaching reduced the beauty premium for female students, but the beauty premium in non-quantitative courses was unaffected for male students. I could understand the beauty premium reducing for both male and female students, but it is difficult to understand why it may be that only the beauty premium for female students reduces in the online teaching mode. Mehic offers only that:

...relative to other students, attractive men are more successful in peer influence, and are more persistent, a personality trait positively linked to academic outcomes...

That seems really unconvincing to me. I think we're going to need a lot more research on this, before we can draw a firm conclusion on how the shift to online teaching has affected the beauty premium in education.

[HT: Marginal Revolution]

Online teaching and gender bias in teaching evaluations

The gender bias in student evaluations of teaching is well established (see my most recent post on the topic here, or the links at the end of this post). However, during the pandemic teaching underwent a sudden and unexpected change to an online mode. It is reasonable to ask, has online teaching reduced gender bias in evaluations? On the one hand, we know that women faced additional difficulties in managing the transition to online work, and especially juggling work with unequal home and family responsibilities. This may have impacted on female teachers' ability to deliver teaching in the online mode effectively. On the other hand, at the risk of gross generalisation, women tend to have a different teaching style that favours connections over content, which may have better helped students with the transition to online learning. It is therefore unclear whether female teachers would be helped, or hurt, in terms of student evaluations of teaching, by the move to online teaching.

This new article by Sara Ayllón, published in the journal Economics of Education Review (ungated earlier version here), provides us with an initial answer. Ayllón uses data from the University of Girona in Spain. Using a difference-in-differences approach, she compares the difference in teaching evaluations between the first and second semester of the 2018/19 academic year, with the difference in evaluations between the first and second semester of the 2019/20 academic year. Since online teaching was enforced in Spain for much of the second semester of 2019/20, this comparison is intended to pick up the impact of online teaching on evaluations. In her baseline analysis, she finds that:

...when I use controls (student’s age, its square, gender, whether the student is repeating the course, and field of study) and robust standard errors clustered at the student level... teaching evaluations during the online semester were, on average, no different from those in previous semesters. Interestingly, though, separate regressions by gender of the lecturer indicate a different story... the online semester had, on average, no impact on the evaluation of male lecturers; but for female lecturers, the average evaluation score decreased by 0.063 points in the online semester compared to previous semesters (about 5.4% of a standard deviation). Thus, while the new teaching environment had, on average, no effect on men’s scores, it did negatively impact the scores received by women...

Ayllón then attempts to tease out the reasons underlying the negative impact of online teaching on female lecturers' student evaluations. She finds no difference in how students felt about how well the course materials were adapted to online learning between male and female lecturers. She also finds no robust difference in grades between students with female lecturers and students with male lecturers. And there is no difference in students' opinions on various aspects of lecturer performance. Ayllón notes that:

...the gendered difference in the teaching evaluation result of the online environment does not appear to be driven by (potentially more objective) aspects of the teacher’s performance. The bias creeps in when students evaluate overall performance...

Students couldn't have easily sorted themselves to have different lecturers, because they chose their programme of study at the start of the academic year. Nevertheless, Ayllón shows that the results hold when the sample is limited to compulsory courses. Finally, looking more deeply at the characteristics of lecturers and students, she finds that:

...the results are particularly negative for young female instructors without a permanent contract, and are strongly driven by male students and low achievers who - even before they know their final grade - retaliate against female instructors, but not against male teachers. The findings are most apparent in Social Sciences. Online teaching did not lead to any positive bias on the part of female students towards female instructors. Yet a considerable degree of discrimination in favour of male instructors is found among high-achieving students.

These results are not dissimilar to other results in the literature. However, rather than showing the underlying gender bias in teaching evaluations, this study shows that the gender bias is larger during online teaching. Unfortunately, as with most studies like this, the solution to the problem is unclear. With online teaching, female lecturers can't give their student evaluations a bump by giving students chocolate. However, these results, especially if confirmed in other studies, should make us even more cautious about using data from student evaluations of teaching in promotion or tenure or appointment decisions, unless we want to perpetuate gender bias.