February 13, 2016 saw the unexpected death of US Supreme Court Justice Antonin Scalia. Justice Scalia was born March 11, 1936, making him just shy of 80 years old at the time of his death. The death of Justice Scalia marks the first death of a US Supreme Court Justice since September of 2005, when Justice William Rehnquist died.
In collabortion with two of our colleagues, we published a paper in the Journal of Insurance Medicine in 2015 about the mortality of US Supreme Court Justices. In that paper we presented results from studying the life and death of all 112 Justices that had served from 1789 through the end of 2013.
One of the key analyses we performed was a Poisson regression model that tested several factors for relationships with mortality rates of SCOTUS Justices: Justice age, secular trends in mortality rates, career length, the effect of actively serving on the Supreme Court, and the effect of being Chief Justice of the Supreme Court. We reproduce the adjusted results of Table 2 from the paper, here:
The results of the model suggested that being Chief Justice does not significantly influence mortality risk for SCOTUS Justices. All other factors were highly significant predictors, though the magnitude of their effects was in some cases very small. This models shows us that, on average, mortality rates for Justices rise by 6% for each additional year of age, rates decrease by 1% for each additional year that passes (secular trend), and rates increase by 4% for each additional year served on the Supreme Court. The most striking result is the protective effect of being an active Justice on the Court: still being active reduces mortality rates by a substantial 60%.
It is important to note that in the original paper we explored the interrelationships of the variables in the model and found that their effects are all independent of one another, in spite of the collinearity of age, time served, and calendar year. It is also important to reiterate from the discussion of our original paper that the protective effect of being an actively serving Justice may be causally connected to remaining active in mind and body through ongoing judicial work on the highest level, or it may simply reflect a survivor bias, wherein retirement from the Court is a marker of declining health. The available data offer us little by which to judge between these two possibilities (or indeed to rule out others), and we remain agnostic on the direction of causation in this relationship.
With the passing of Justice Scalia, another death accrues in a little over two calendar years and just over 25 person-years of exposure time. More importantly, this death accrues among active Justices. This led us to wonder what effect this might have on our original findings. We have therefore updated our study to extend follow-up through February 13, 2016, re-running the exact same model. Here are the results:
The death of Justice Scalia might be expected to dilute the protective effect of the “Active” parameter, as he died while still serving on the Court. The updated model confirms this, though the effect is quite small. The estimate of the MRR changed from 0.42 to 0.43, and its 95% confidence very little. The estimate for the effect of being Chief Justice also shifted slightly, from 1.09 to 1.08. All other parameter estimates remained unchanged.
Standardized Mortality Ratios
In the original paper we also presented a figure that plotted the Standardized Mortality Ratios (SMRs) and their 95% confidence intervals for each decade between 1790 and 2009. With the additional follow-up time in this update, we can add one more SMR to that figure:
The SMR and confidence band for 2010-2016 is shown at the far right, in red. Whereas before this partial-decade category only had 4 years of follow-up and no deaths, it now has 7 years of follow-up and 1 death, namely that of Justice Scalia. This SMR, though displaying a wide confidence band, holds to the pattern of most of the SMRs since 1900: a point estimate of less than 1.0, with a wide confidence band. Note that the dotted line in the figure represents 1.0, or the null hypthosis, wherein there are no differences between the observed counts of death and the expected.
Finally, we took this opportunity to update the overal SMR as listed in Table 1 of our paper. There we noted that the SMR for 1789-2013 was 0.87 (95% CI 0.70-1.05). Adding the additional 25 years of person-time and the extra death changes this SMR to 0.86 (95% CI 0.70-1.04). As mentioned in the original paper, statistical power is limited in these analyses owing to the small number of Justices alive at any given time. Though the data suggest that Supreme Court Justices may have lower mortality than the general population of the US, statistical significance can be reached for only very strong associations.
Note on Reproducibility
The first model presented here does not exactly match that shown in Table 2 of the original paper, though the differences are trivial and no conclusions are altered. In our experience such small discrepancies are far too common after additional analyses morph the original code during the peer review and publication process and the memory of the analysis is no longer fresh. Such scenarios are exactly why version control software was created, and they are the reason we are moving to make our research more open, transparent, and, ultimately, reproducible. Check out the new Github.com repositories for Mortality Research & Consulting, Inc. Here we will post datasets (when possible - see below*) and code from our research, making available all the source files you need to reproduce (and hopefully improve upon) our research, using nothing more than open-source tools. We hope to continually add content to our repos on Github, so check back frequently.
*In some cases, access to original data is allowed (if at all) only after agreeing to terms and conditions stipulated by owners of said data or required by laws governing access to data that include health information and/or personal identifying information. In such cases we will point users to the original sources and will provide code for using these data if and when they are accessed appropriately.