Can Empathy Explain Gender Differences in Economic Policy Views in the United States?

ABSTRACT

This paper shows that different levels of empathy of men and women explain the well-documented gender differences in interventionist government economic policy views in the United States. Using the Davis Interpersonal Reactivity Index (IRI) to measure empathy, the study finds that more empathic people support more interventionist policies. While greater empathy leads both men and women to support more government action, there is no gender difference in the effects of empathy on policy views. When policy views are separated by area, gender differences on policies concerning poverty, inequality, and social welfare disappear once empathy is accounted for, though they persist in views on free markets.

KEYWORDS:

• Public policy
• gender
• empathy
• distribution
• experiment
• United States

JEL Codes:

• A13
• H00
• I38

ACKNOWLEDGMENTS

We gratefully acknowledge financial support for this project from Haverford College and both the Economics Department and the Leavey School of Business at Santa Clara University.

SUPPLEMENTAL DATA

Supplemental data for this article can be accessed at https://doi.org/10.1080/13545701.2018.1493215.

Notes

1 See, for example, Center for American Women and Politics (CAWP; 2012a, 2012b, 2014a, 2014b) and Pew Research Center (2012, 2015). In the US, the term “liberal” refers to progressive or left-leaning policy views. The political system of the US is primarily a two-party system, with Democrats and Republicans dominating federal and state politics. The Democratic Party supports an activist government, a broad safety net, and liberal social policy. The Republicans are more conservative in their views on social policy, and they support a small federal government with low taxes and few regulations.

2 The issue of how empathy is acquired is of considerable interest and has been extensively investigated in the literature (see, for example, Eisenberg and Strayer [1987]; Hoffman [2000], De Waal [2009]; Decety and Ickes [2011]; Decety [2012]). A reasonable conclusion from this literature is that empathy is both an innate, heritable characteristic (the effect of the evolution of social animals for whom the good social relationships created by empathy promote survival of the group) and a trait that is learned as humans develop.

3 There is some evidence that the growth in the gender gap in party preferences has been due to men shifting toward the Republican Party while women’s preferences have been more stable (Carroll 2014).

4 Similarly, the psychology and neuroscience literature distinguishes between theory of mind (ToM) or mentalizing, which is the ability to know another person’s beliefs, emotions, or intentions; empathy, which entails sharing the feelings of the other; and compassion or prosocial concern, which is feeling concern for the other and being motivated to improve his or her well-being (Zaki and Ochsner 2012; Singer and Tusche 2014).

5 For recent reviews of the neuroscience literature on empathy, see Jamil Zaki and Kevin Ochsner (2012) and Tania Singer and Anita Tusche (2014).

6 See Leonardo Christov-Moore et al. (2014) for a recent comprehensive summary of the literature on sex differences in empathy in humans and nonhuman animals. Bhismadev Chakrabarti and Simon Baron-Cohen (2006) provide a detailed discussion of reasons why women are more empathic than men.

7 In adding two correlated explanatory variables to the regression, we are subjecting the regression to multicolinearity and the resulting problem of reduced statistical significance. Both variables have substantial and behaviorally important effects on policy views when entered alone. Once we enter both, the variation used to estimate the separate coefficients is their independent variation, not the variation shared with the correlated variable. With less variation used to estimate the coefficients, the probability that these coefficients become statistically insignificant rises.

8 One might also be concerned that empathy is developed within individuals growing up receiving government services, and these individuals are likely to support large government. In our analysis we control for income and find no evidence that income controls reduce the effect of empathy on policy views.

9 Each statement is designed to articulate a conservative or liberal point of view, and the respondent can rate how closely the statement reflects his or her preferences. The survey was developed and implemented in the spring of 2012, when domestic political conversations were focused on large US fiscal deficits. At that time, liberals were advocating both raising taxes and cutting spending to deal with budget shortfalls. Therefore, statement 11 was aimed at articulating the then current liberal point of view. We are aware that left-leaning liberals would not advocate cutting spending, and while most of our analysis includes statement 11, we do make sure that the results persist when statement 11 is dropped.

10 Students were recruited by e-mail and by flyers posted around campus. The message explained that students were invited to participate in an on-campus research study asking them to make some decisions about money and to fill out a survey. They were told that everyone who participated would be paid a US$10 show-up fee and could earn additional money, with average payoffs expected to be about US$18–US$20, and that the study would take about forty minutes. Students were given a link to a form where they could indicate their availability for sessions at various times. They were then e-mailed the session time and date to which they were assigned. Nine sessions were held with the number of students ranging from fourteen to twenty-eight, and average session size of twenty. All personal information that would allow the identification of any person(s) described in the article has been removed.

11 Experiment instructions are provided in the Supplemental Online Appendix. This paper reports the findings of the IRI and public policy exercises; the other exercises will be analyzed in a separate paper.

12 Factor analysis was conducted for each subscale and for the total IRI. The factor analysis for the subscales all calculated large eigenvalues between 2.3 and 3.2 for the first factor, with the first factor explaining roughly ten times as much variance in the variables as the second factor. For robustness purposes, we calculate the factor score for each category of empathy, an index of the latent variable the questions are asking about. We then conduct a factor analysis on these indices, and again the eigenvalue of the first factor dwarfs those for subsequent factors, and the factor loadings are similar across all indices except the index for empathic distress, which is lower. We calculate the factor score for the full set of empathy categories. This factor score is highly correlated with our average IRI score (r = 0.95). In all our subsequent analysis we repeat the analysis using this factor score rather than the straight average. None of the results change.

13 We performed a similar factor analysis on the policy variables within category. Again, the eigenvalues for the first factor were many times larger than subsequent eigenvalues, and factor loadings on the variables for the first factor were quite similar, thus validating our survey tool. There were a couple of exceptions; the factor loading on immigration reform in the social welfare policy factor was lower than other factor loadings, and the factor loadings on the two questions identifying the government’s major focus were lower than the factor loadings on other questions in the government budget category. As a result, we calculated factor scores for each category and then conducted a factor analysis of these scores to create a factor score for total policy views. This factor score is highly correlated with our average policy score (r = 0.98). We repeat all analyses with the factor score measuring policy views, and all results stay the same.

14 See Linda Kamas and Anne Preston (2012) for a detailed description of the methodology utilized.

15 In the regression without the empathy score, the income controls (four dummies with the highest income category as the omitted group) have positive coefficients and are jointly statistically significant. The major controls (dummies for an economics major, an engineering major, and a business major) are all negative and jointly statistically significant. With the inclusion of the empathy score, the income coefficients fall in statistical significance and only the two lowest income categories have statistically significant coefficients. The major variables continue to be jointly statistically significant. The dummy for Catholic, which has a negative coefficient, becomes statistically significant at the 0.05 level. There is no evidence that the controls have different impacts on policy views for men and women, as woman interactions terms are all statistically insignificant.

16 We also have measures of altruism. They include the amount the participant gives in a dictator game (a subject is given US$10 and then asked how much of the US$10 he or she would like to give to an anonymous person in the experiment); whether the subject donated to charity in the last year; and the amount the participant donated to charity in the last year. Only the dummy variable identifying those who donated to a charity in the last year is correlated with empathy at the 5 percent level (r = 0.15). We include these variables as controls (separately) instead of the social preference controls in the models of columns 3 and 4 of Table 5. None of them has a statistically significant impact on public policy preferences, and inclusion of these altruism variables does not alter the size or statistical significance of the woman and empathy coefficients.

17 We reestimate the column 3–6 regressions adding interactions between woman and the controls, and none of these interactions are statistically different from 0. There is no evidence that the controls affect policy views differently for men and women.

18 In the case of regulation, poverty and inequality, and social welfare, because there are subjects at one or both of the ends of the distribution, we repeat the analyses of Table 6 using Tobit regressions, and the results do not change.

Additional information

Notes on contributors

Linda Kamas

Linda Kamas is Associate Professor of Economics at Santa Clara University. Her research is in the area of behavioral and experimental economics and focuses on gender, altruism and other-regarding preferences, confidence, competitiveness, and empathy. Professor Kamas’s recent work links economic behavior in laboratory experiments to real-life labor market outcomes.

Anne Preston

Anne Preston is Professor of Economics at Haverford College. Her research interests include nonprofit labor markets, careers of scientists and engineers, the effects of sexual orientation on labor market outcomes, and the link between experimental work on altruism and competition and real-world behaviors and outcomes. Her research studies often focus on different economic outcomes of men and women.