Divided government and the bias against presidential restraint

ABSTRACT

Comparative rankings of presidential performance can be clouded with partisan biases. Here, we argue that the presidential rankings literature contains an overlooked bias favoring presidential activity. It is easier to observe the use of power than the exercise of restraint. As such, expert rankings of presidents may exhibit a selection bias whereby we are best able to evaluate those who are more proactive rather than those who, willfully or not, exercise political restraint. In this paper, we consider how presidential rankings of greatness are affected by measures of presidential restraint. In other words, we are not interested in the quality of presidencies, but in the quality of the rankings. We create five measures of restraint: divided government, use of the veto power, military interventions, army size, and changes in the size of government. We argue that the first two provide measures of involuntary restraint while the remaining three represent measures of willful restraint. Using OLS (for presidential scores) and ordered probit (for presidential rankings) for presidents up to George W. Bush, only one measure of restraint has a consistently negative and significant effect: divided government. This suggests the presence of bias in historical evaluation whereby presidents who had more room to act rank higher in presidential rankings.

KEYWORDS:

• Presidential Studies
• American History
• American Politics
• divided government

Acknowledgments

We received no funding for this article and have no conflicts to declare. We gratefully acknowledge the helpful comments of David Henderson, Zachary Gochenour, Alexander Salter, Alexandra Foucher as well as the four anonymous referees who provided useful suggestions and comments.

Notes

¹ This survey is explained in the cited article, but a newer version is available online at: https://www.boisestate.edu/sps-politicalscience/(Consulted 1 November 2018).

² As opposed to earlier surveys such as Schlesinger’s where presidents had to be ranked as Great, Near Great, Average, Below Average, or Failure (Schlesinger, 1948).

³ There exist disagreements within certain dimensions of the surveys (Nichols, 2012), but they do not have a sizable effect on the overall rankings and who sits at the top and bottom of them. One exception may be found in R. Vedder and Gallaway (2001). The authors created a ranking of their own based on changes in the size of government and inflation rate. There is no correlation between their rankings and the other rankings, and some of the worst performers sit at the top of Vedder and Gallaway’s ranking system.

⁴ The last of these elements is meant to reflect a recurrent criticism regarding how to judge a president in context (Bailey, 1966; DiClerico, 1979; Nichols, 2012).

⁵ The bias is assumed to be pro-democratic in general. Stanford historian Thomas Bailey claimed that the original survey conducted by Schlesinger (1948) was a ‘Harvard-eastern elitist-Democratic plot’ (as cited in Uscinski & Simon, 2011, p. 2).

⁶ It is also worth pointing out that the case of the Eisenhower presidency reinforces the idea of a bias against restraint. Unlike Coolidge and Cleveland who restrained themselves from acting, Eisenhower was proactive, but kept a hidden hand. Eisenhower’s discreet approach translated into an under-evaluation of his presidency according to F. I. Greenstein (1979); F. Greenstein (1988)). We thank an anonymous referee for pointing us in the direction of the work of Greenstein on Eisenhower and how it encapsulates the argument we make about the bias against restraint (and the appearance of restraint).

⁷ In Moe and Howell (1999a, 1999b)) point out that it is because they are not constitutionally specified that these powers exist. Because the powers of presidents are ‘ambiguous,’ there is room for ‘presidential imperialism that Congress and the courts cannot be counted upon to stop’ (Moe & Howell, 1999a, p. 176).

⁸ This is in part why Howell and Moe (2016) argue that a more powerful presidency would be socially beneficial.

⁹ We thank two anonymous referees for these suggestions regarding this mechanism by which presidential reputations might be affected adversely by divided government.

¹⁰ We also exclude presidents whose tenure was too short to generate much data (e.g., James Garfield, William Henry Harrison).

¹¹ We downloaded the control variables from the data files made available by Henderson and Gochenour (2013). However, we modified one of the variables – real GDP growth. They used an older version of the data made available by Johnston and Williamson (2020) as part of the Measuring Worth Project. We simply replaced the values in the dataset by the more recent ones which have been revised.

¹² We also tried different specifications such as excluding either the war-related variables, or the changes in government spending (because one would expect them to be related). This does not alter the results: divided government is never affected while changes in government spending always remains statistically insignificant. Because these results would dramatically lengthen the paper, we have opted not to include these tables. However, they figure in the online appendix available at https://tinyurl.com/ycqayfjd.

¹³ We also tried different specifications that include the share of time in a president’s mandate where Congress was divided and whether or not one, or both chambers were opposed to the president. The share of time is not significant, but the penalty appears to be present regardless of whether it is the House, or the Senate, or both who are opposed to the president. These results are presented in the appendix.

¹⁴ In fact, the small differences appear to be due in large part to the year of the C-SPAN survey used. These scholars used the 2009 version of the C-SPAN survey that was available. The version we used was for 2017. We tried to redo our results using that earlier survey. We found the same pattern of significance as in Table 5 and the coefficients for the control variables were similar to those of Curry and Morris (2010) and Henderson and Gochenour (2013). The results for our restraint variables were unaffected.

¹⁵ In Table 6, we only show the results with all the restraint variables together. However, the results remain unchanged when we run specifications 13 through 17 from Table 5. The full results are included in the appendix to this paper (which can be accessed here: https://tinyurl.com/ycqayfjd).

¹⁶ Because there are 40 presidents, there are 40 categories. Thus, we cannot produce tables of predicted probabilities as for each category as the resulting table would be unwieldy.

¹⁷ It is interesting to note that even absent regression analysis, the penalty is evident. In the APSA survey, the mean score of presidents with divided government is 48.1 while it stands at 63.3 for presidents with unified government. In the C-SPAN Survey, those figures stand at 522.2 versus 653.2. In the Faber & Faber survey, they stand respectively at 22.4 and 48.9. We thank an anonymous reviewer for this suggestion on how to highlight the importance of our divided government findings.