How a Statistical Anomaly Misled a Generation of Research on Representation

• In Event: New Directions in the Study of Inequality and Democratic Representation

Abstract

While scholars agree that important representational inequities exist in the U.S. political system, most research on political responsiveness to different income groups shows highly nuanced and contingent patterns. Factors such as the state, party in power, and the policy matter for who gets represented, and sometimes policy corresponds more closely to the preferences of those in the middle or even the bottom of the income distribution (e.g., Bhatti and Erikson 2011; Brunner, Ross, and Washington 2013; Ellis 2013; Elkjaer and Iverson 2020; Lax, Phillips, and Zelizer 2019; Wlezien and Soroka 2011).

Research by Gilens (2005, 2011, 2012) and Gilens and Page (2014, 2017) remains an exception to the broader literature, consistently finding “when their views differ from those of more affluent Americans, government policy appears to be fairly response to the well-off and virtually unrelated to the desires of the low- and middle-income citizens” (Gilens 2005, 789). It turns out that these findings of unequal representation depend on a previously unrecognized and problematic statistical anomaly. When we correct the anomaly, those findings change.

This paper explains the statistical anomaly that has misled research on political representation. Specifically, I show how the seemingly innocuous decision to analyze preference gaps biases estimates of responsiveness to the affluent upward and biases estimates of responsiveness to the middle and lower income groups downward. I then offer two potential solutions to eliminate this bias. The analysis reconsiders Gilens’ (2012) findings as well as a parallel study by Kopkin and Roberts (2023) that finds when women’s and men’s preferences diverge, policy only reflects the preferences of men. In both cases, when analyzed correctly, evidence of unequal responsiveness changes drastically. I then reconsider Gilens’ multivariate analysis, which attempts to correct for measurement error in group preferences. The results show that the conclusions from this multivariate analysis are extremely sensitive to the specific measurement error correction used, making it impossible to use this approach to determine which income group(s) receive representation.

In sum, while examples of unequal representation in the United States exist, the specific conclusions of Gilens, Gilens and Page, and others following their methods need to be revisited. The paper concludes by discussing the many implications of these results, offering recommendations for future research, and providing a template for how scholars should describe the literature on policy responsiveness moving forward.

Author

• Peter Enns, Cornell University