Do elected officials serve the poor on health care? Evidence from a field experiment on members of Congress and state legislators

ABSTRACT

To what extent do legislators respond to the poor? While extensive research demonstrates the poor are largely ignored in legislators’ policy calculations, little research examines the degree to which they discriminate against the poor with respect to providing constituency service. We examine this question using a series of correspondence experiments on both the offices of members of Congress and state legislators on the topic of health care. Consistent with previous studies we find no evidence that members of Congress discriminate by economic class and only mix edevidence that state legislators discriminate along these lines. We also find limited, but potentially important, evidence of partisan bias in service responsiveness for state legislators.

KEYWORDS:

• Responsiveness
• Congress
• State legislatures
• Field experiment

Highlights

• We find no evidence that members of Congress discriminate in service responsiveness on health care by economic class.

• Only mixed evidence that state legislators discriminate along class lines.

• We find limited evidence of biases in responsiveness by legislator party for state legislators.

Acknowledgements

The authors would like to thank Christian Grose, Paul Herrnson, Carl Klarner, Elin Naurin, Jeremy Pressman, Lyle Scruggs, Matt Singer, Antoine Yoshinaka, participants at the UConn Political Economy Workshop and SWPSA Field Experiments and Political Elites Conference for comments on earlier drafts. The authors would also like to thank Sam Dorman, Adam Ghalmi, Steven Colon, Franklyn Barrueco Jr., Dan Dennis, and Sahar Iqbal for their research assistance in collecting data for this article. The authors would like to thank the Alan R. Bennett Foundation and the Political Science Honors Bennett RA program at the University of Connecticut for partial support for data collection of this project.

Notes

¹ Since the implementation of the ACA, the number of people without health care has fallen considerably, as a report in 2014 showed a decrease of 8% of the uninsured population from the previous year (Tavernise, 2014). Recent estimates show the non-elderly uninsured population has dropped to around 10% in 2017 from 18% in 2010 (Kaiser Family Foundation [KFF], 2017).

² See “Obama’s Health Law: Who Was Helped Most” http://www.nytimes.com/interactive/2014/10/29/upshot/obamacare-who-was-helped-most.html

³ While our study builds on the work of Carnes et al. (2019) our experimental design differs in important ways. First, Carnes et al. (2019) use confederates living in local areas to send requests to their elected officials. Our study instead relied upon the use of fictitious addresses. Second, while Carnes & Holbein contact local officials such as school principals and mayors, as well as state legislators, our studies focus on the state (state legislators) and federal level (Congress). Third, Carnes & Holbein’s use occupational background as a signal for class status. Our experimental design explicitly mentions constituent income.

⁴ We chose the very common name Amy Johnson so as not to prime any racial characteristics. Use of a common name might also make it more difficult for legislators to search the voter rolls to obtain additional information about our fictitious constituent.

⁵ In our follow-up experiment on members of Congress (November 2015), the emails were sent a year after the first audit study (October 2014). This long period of time between studies makes it unlikely legislators or their staffs were aware of the study. We did not receive any indication in the responses that anyone became aware of our audit study.

⁶ An analysis examining the ability to detect a difference in response rates of 10 points shows that the statistical power increases from .55 to.73 for the low-income condition between studies 1 and 2. The statistical power increases dramatically when moving from the first two studies to the examination of state legislators (.99). The full results for this analysis can be seen in Appendix A. Additionally, we pool all studies and employ a fixed effects regression as another way to increase sample size. The results of this analysis, found in Appendix C, show no statistically significant differences in responsiveness between the low-income and control conditions.

⁷ The NJ 1st, NC 12th, and VA 7th seats were vacant and therefore we did not email these districts.

⁸ At the time of the experiment, Tennessee and Pennsylvania did not publicly list email addresses for state legislators and thus were left out of the experiment. A state legislator in Wyoming responded that they were aware this was part of an experiment and would take active measures to distort our number of responses. Thus, this state is also excluded, as it is possible this would bias the results. The results, however, are unchanged when Wyoming is included in the analysis.

⁹ When pooling the results for studies #1 and #2 we observe no statistically significant difference between conditions as well.

¹⁰ One concern is that our treatment conditions are heterogeneous as cost of living varies greatly by state. In poorer states such as Alabama, for instance, $60,000 may be well above middle income. The possibility exists that our middle-income treatment may not prime attitudes toward middle-income constituents in such cases. This is perhaps less of a concern for the low-income condition as $20,000 in most states is considered quite low. To address the concern that our results are biased against low-income areas, however, we conducted difference in means tests for states in the middle half of the income distribution. While the results are substantively similar for Congress (study 1), there is one important difference in results for state legislators (study 3). Here, state legislators are found to be significantly less responsive the middle-income condition (compared to the control) and less responsive to the middle-income compared to the low-income condition. We also examined this issue at the state legislative level by splitting the sample by median income (household and family) for state senate districts using data from the 2014 Census. Here we find no statistically significant differences in responses for state legislators by income.

¹¹ We include a figure with difference in means tests as an alternative way to illustrate these findings in Appendix B.