Using interruptions to predict Supreme Court cases

Reading Justice Gorsuch

If you wanted to know which way Justice Gorsuch was going to vote in the 2017 Term, you could have placed your bets with 86% accuracy by observing just one statistic from oral argument—how many times Gorsuch interrupted each side.

In the 2017 Term, Gorsuch heard 60 arguments and voted 40 times for the Petitioner and 19 times for the Respondent (Washington v. United States was affirmed by an equally divided Court, and by convention no individual justice votes are recorded). On average, he initiated 1.6 more interruptions during the Respondent’s time than during the Petitioner’s (a few of these interruptions were of other justices but most were of the advocate). But like Chicago’s annual average temperature of 49°, that number masks all the interesting variation. In those cases where Gorsuch voted for the Petitioner, he interrupted the Respondent an average 3.8 more times than the Petitioner. Conversely, when voting for the Respondent, he interrupted the Petitioner 3.1 more times on average.

One way to visualize this kind of data is with a histogram.

Gorsuch Interruption Gap, Pro-Petitioner and Pro-Respondent Histograms 2017 Term

In the figure above, we have overlaid two different histograms of the interruption gap (interruptions of Petitioner minus those Respondent) for Gorsuch in the 2017 Term. The orange bars indicate the interruption gap in cases where the Justice voted for the Petitioner: it is clear that in the vast majority of these cases, Gorsuch interrupted less during the Petitioner’s time than Respondent’s. The blue bars show the interruption gap for the pro-Respondent cases. Here, in contrast, most of the weight of the figure is above zero—i.e. Gorsuch interrupted the Petitioner more often, though the shape of the blue distribution is flatter. We should note that because Gorsuch voted for the Petitioner twice as often as he voted for the Respondent (as the Court in general does), an orange bar represents twice as many cases as would a blue bar of the same height.

In the next figure, we further explore the distribution of Gorsuch’s interruption voting patterns through a boxplot.

Gorsuch Interruption Gap Boxplots 2017 Term

Boxplots are cool (trust us!) but they require a little explaining. The “box” part of a boxplot indicates the range separating the top 25% of the data from the bottom 25% (the interquartile range); think of this as the space between the 25 and 75 yard lines on a football field. The vertical line segmenting this box is the median of the data. So, in the figure above, a quick comparison of the light blue pro-Respondent box with the orange pro-Petitioner box shows that there is no overlap between the middle 50% of each of these categories. What is more, the orange box is entirely to the left of zero on the interruption gap scale and the blue box is entirely to the right. The white space between the boxes tells you at a glance that 75% of cases in the orange category have a lower interruption gap than 75% of cases in the blue category. And thus, when Gorsuch interrupts one side more than the other, he is usually showing his hand and telegraphing which way he will vote.

The whiskers extending away from the boxes extend close to the extremes of the data: they indicate about how far the data extends before you get to true outliers (in a normal distribution, 0.7% of observations lie beyond the upper and lower limits). The fact that the whiskers cross the zero point confirms that although the interruption gap is a very strong predictor for Gorsuch, it is not infallible. In fact, in the 2017 Term, relying on the interruption gap to predict Gorsuch’s votes would have been misleading in just 7 cases.

Some broader implications

The Supreme Court is under close scrutiny like never before. In this era of intense political polarization, the Court merits attention for reasons that go beyond the outcomes of individual cases or issues of methodology and jurisprudence. The justices themselves have become icons—or demons—representing the moral future of the country. Ruth Bader Ginsburg’s visage appears on cups, T-shirts, and various other paraphernalia, and the late Antonin Scalia is fetishized equally by the right. The two most recently appointed associate justices were subject to extraordinary confirmation processes. Gorsuch holds a “stolen” seat in the eyes of some, and Kavanaugh is defective in character and temperament, according to others. In a divided country where justices are feted and demonized with equal fervor, the public is now paying close attention to every aspect of the Court’s decision-making process, down to the meanings attributed to every word uttered. Oral argument deserves attention because it is the one public part of the Court’s process, and because it is public it gives us an opportunity to more rigorously assess aspects of judicial character that would otherwise be matters of supposition. One way to explore those more amorphous judicial traits is to look at the language justices use in the relatively unguarded arena of oral argument, as compared to tightly scripted case opinions.

Websites such as SCOTUSBlog provide excellent coverage of current Supreme Court cases using traditional modes of legal analysis: close reading and expert opinion. In addition, reporters such as Marcia Coyle, Dahlia Lithwick, Adam Liptak, and Nina Totenberg provide qualitative analysis of Supreme Court oral arguments, and there also are podcasts such as Amicus, First Mondays, and even an army of teenagers who write about the Supreme Court at the website High School Scotus. There is also excellent analysis by other empirically minded scholars, such as Tim Johnson, who has shown that oral argument can be important in judicial decision-making, and some have used transcript data in their research.

But ScotusOA offers something new. Unlike other commentators, we make voting predictions for each justice as the cases are heard. Predicting outcomes is always treacherous, as even a largely accurate model will have variation and noise: even an attempt to model prior data, where the results are known, will lead to imperfect “predictions.” Consequently, we buttress our empirical analysis by listening to the oral arguments, in order to follow the nuances of each case and to understand what the justices think they are communicating.

We also base our predictions on justice-specific models of features of oral argument that our early research has told us are important in the aggregate (Jacobi & Schweers, “Justice, Interrupted”; Jacobi & Rozema, “Judicial Conflict and Voting Agreement”; Jacobi & Sag, “Justices as Advocates”). The challenge is to develop a model that is a good predictive tool in individual cases. To do that, we have developed multiple models for each justice to predict their votes based on oral argument. What is striking about Justice Gorsuch is that he is most easily predicted using just one variable, the interruption gap.