Swedish Twin Pairs: A common argument is that the wealthy practice patrimonial voting, i.e. voting for right-wing parties to maximize returns on their assets; these authors say this is a non-causal correlation

Uncovering the Source of Patrimonial Voting: Evidence from Swedish Twin Pairs. Rafael Ahlskog & Anton Brännlund . Political Behavior, Jan 30 2021. https://rd.springer.com/article/10.1007/s11109-020-09669-4

Rolf Degen's take: The belief that wealth generates a political shift to the right, which serves to protect the wealth, fails the critical test. https://t.co/c8wicIvEOW https://t.co/r9j8TLePcN

Abstract: The boom in wealth inequality seen in recent decades has generated a steep rise in scholarly interest in both the drivers and the consequences of the wealth gap. In political science, a pertinent question regards the political behavior across the wealth spectrum. A common argument is that the wealthy practice patrimonial voting, i.e. voting for right-wing parties to maximize returns on their assets. While this pattern is descriptively well documented, it is less certain to what extent this reflects an actual causal relationship between wealth and political preferences. In this study, we provide new evidence by exploiting wealth variation within identical twin pairs. Our findings suggest that while more wealth is descriptively connected to more support for right-wing parties, the causal impact of wealth on policy preferences is likely highly overstated. For several relevant policy areas these effects may not exist at all. Furthermore, the bias in naive observational estimates seems to be mainly driven by environmental familial confounders shared within twin pairs, rather than genetic confounding.

Discussion

A popular argument in political science is that the wealthy practice patrimonial voting in that they vote for right-wing parties to increase their fortunes. While this pattern is well documented, the underlying mechanism (if causal or not) is not fully understood. Descriptively speaking, wealthier people in this sample of Swedish middle-aged twin pairs are more supportive of free market policies, less supportive of redistribution and real-estate taxes, and more likely to identify as right-wing and vote for right-wing parties. Causally speaking, however, the effect of wealth (all or financial) on domain specific policy preferences about taxes and market regulations on the one hand, or questions about economic redistribution on the other, appear to be much more modest than ordinarily assumed, and are not statistically significant.

When moving to within-twin pair analyses it is revealed that naive correlations, rather than reflecting a true causal signal, are likely severely biased by unobservable familial characteristics. This bias is largely left unchecked by conventional statistical controls such as income, education, family size or occupational category, which underscores the importance of taking familial confounders into account. Bivariate variance decomposition models further showed that the sources of this confounding are more likely shared family environment than genetic overlap.

While previous studies using lottery-winnings have shown some effects on political preferences, it should be noted that these studies concern a fairly dramatic relative increase in net wealth. Most voters are unlikely to experience a wealth shock comparable to the magnitude of a large lottery prize sum, not even under extraordinary circumstances such as financial crises or recessions. If our results are quantitatively transferable, we should expect few individuals to significantly alter their views on most economic and social policy issues in response to a realistic change in individual net wealth. However—we do find remaining effects for a few outcomes even after taking familial confounders into account, although these findings are mostly weak in both a substantive and statistical sense. Most notably, the highly specific issue of property taxation appears to be an exception to the pattern above.

This finding falls in line with studies that find a general link between property ownership and resistance against property taxation (Brunner et al. 2015). As argued in the theory section, we think that the case of property taxes and real wealth, in this particular setting, should be seen as a type of most likely case for finding an effect. People can fail to precisely compute the general costs and benefits of government policy (as would often be required for the free-market and redistribution dimensions), and yet find this specific form of taxation of their property obtrusive. The connection between the value of ones’ house rising, and paying larger amounts in property taxes, is hard to miss.

We do find some remaining effects, although statistically weak, on self-reported left–right placement and party choice. These effect sizes are similarly highly attenuated when controlling for familial confounders. It is possible that these two indicators are picking up the effect from other pocketbook type mechanisms than those captured by the survey items included in our issue dimensions. For instance, the right-wing coalition abolished the wealth tax at the time of the survey and Swedish property owners were saved from collapsing house prices as the conservative administration was able to mitigate much of the impact of the financial crisis. Unfortunately, no survey items regarding these policy preferences were available.

There are, as always, a few concerns about the interpretation of the lack of results when using a discordant twin design. First, if twins tend to exert a marked influence on each other (as adults) in a way that is relevant to the variables of interest, this will violate independence assumptions and introduce bias. The bias can go in either direction depending on whether this effect makes the twins more or less similar (i.e. if they come to diverge into different behavioral niches as a consequence). One particular way in which this could function in the present case is if the type of self-interest mechanisms we’re interested in testing extends within the family—specifically to the other twin. If so, this may dampen within-pair effects if one twin adapts, for example, to prefer less taxation as his or her cotwin gets richer. Robustness checks using self-reported contact rates between the twins in Appendix A give no indication that this is a problem for our results, but these tests do not completely rule out this potential design weakness.

Another remaining possible limitation is concerns about reverse causality or collider bias: values might affect wealth acquisition, and may also affect some of the controls. Collider effects would bias the estimates downward, but do not appear to be an issue since the introduction of controls has only minor consequences for the estimates. Reverse causality on the other hand would not lead to any of the estimates being biased up or down, but would obviously render moot any causal conclusions for the results that remain. In the absence of fine-grained longitudinal data for political attitudes, we will have to rely on plausibility: it appears dramatically less likely, for example, that having a particular opinion about real-estate taxes causes the value of one’s house to change, rather than the other way around.

Something should be said about the implications of the bivariate ACE models for future observational research. As mentioned in the methods section, these results indicate, at least in the current setting, where to look for confounding factors to include in the models. The absence of any substantial genetic confounding in these relationships indicate that we can conceivably get by without genetically informed data—provided that the sources of the environmental confounding can be identified and controlled for. The largest part of this confounding appears to stem from shared environmental factors—possibly family environment or shared networks. If these are accurately captured, observational models may be able to avoid a large amount of the endogeneity problem that we document here.

Finally, it is difficult to conceive of reasons why the external validity of these findings should be confined specifically to twins, since it would require that having a twin in itself nullifies most of the causal relationship between wealth and preferences over government policy. The results are therefore likely generalizable to a wider Swedish context, at the very least for the cohorts included in the study. Another question is how far these results can travel in an international context given that the universal welfare system in Sweden reduces the need for private savings in times of financial distress. However, recent research suggests that this context likely constitutes a tough test for the patrimonial voting hypothesis in that a larger effect of asset wealth on voting is found in more liberal welfare systems like the UK (Quinlan and Okolikj 2019). Thus, we think it is reasonable to assume that our results are generalizable to most other OECD-nations.

Our results should encourage scholars to investigate the relationship between economic resources and political preferences further. We find, as others have before us, that there is a descriptive link between wealth and political attitudes and behavior. However, it seems that these naive comparisons—even when using a rich set of statistical controls—overestimate the causal effect of wealth quite substantially, if there is one at all. We suggest that future research should aim at finding situations where the economic stakes are clear and unobtrusive to voters, and where there is an unambiguous connection between the policy issue at hand and the immediate self-interest of asset holders. Further research may provide important new knowledge about the effects of increasing wealth inequality, as well as insights into processes of political preference formation.