One. That is the number of tracked cases in this dataset so far of a sitting member of Congress dying in office since Too Damn Old began recording completion probabilities. One case is not a trend. It is a single data point, logged in full, with no adjustment after the fact. This post explains what the site tracks, what that one case looks like, and why you should resist the urge to draw a pattern out of a sample size of one. The number will grow over time. Right now, it is one.
Here is everything currently on record. Lindsey Graham, Republican, Senate, South Carolina, died at age 71. He had served 31 years in Congress at the time of his last recorded data point. His last recorded completion probability, the site's estimate of how likely he was to finish his current term, stood at 82 percent. The special election triggered to fill his seat cost an actual, verified 12 million dollars. Across all tracked cases, the average last-recorded completion probability is also 82 percent, because there is only one case to average. It is not a statistical average in any meaningful sense yet, it is just Graham's number, repeated as the site's current aggregate. Total actual special-election cost triggered by tracked deaths in office so far: 12 million dollars, which is also just Graham's cost, not a sum across multiple cases. That is the entire dataset. No ranges, no distributions, no year-over-year comparison, because a distribution requires more than one point, and this site is not going to manufacture one.
Too Damn Old assigns every sitting member a completion probability, a running estimate of whether they will finish their current term, built from age, actuarial data, and time already served. The point of tracking actual outcomes against those predictions is accountability for the model itself, not just for Congress. If the site says a member has an 82 percent chance of finishing a term and that member dies in office, the case gets logged here, permanently, whether it makes the model look good or bad. That is the discipline behind this page. Readers should use it as a check on the site's own forecasting, not as a prediction engine for who is next. With one case logged, the honest statement is that the track record exists and is being built in public, in real time, as it happens. It is not yet large enough to tell you whether an 82 percent completion probability means what it says 82 percent of the time. It will take many more tracked cases, across many more members and many more years, before that claim can be made responsibly. Treat every number on this page as a running tally, not a verdict, and treat the word average with suspicion when the sample size is this small. The same caution applies to the total special-election cost. Twelve million dollars is real money attached to one real seat. It is not yet evidence of what deaths in office cost Congress on average, because averaging one number tells you nothing about the range of possible outcomes. A future case could cost more, could cost less, could involve a House seat instead of a Senate seat, could involve a state with different special-election rules entirely. The only responsible move right now is to report the one number honestly and wait for more.
The completion probability is a per-member estimate generated from age and actuarial data, recalculated as new terms begin and new information becomes available. When a tracked member dies in office, the site logs the last completion probability that was on record for them before death, their age at death, years served, and the actual dollar cost of any special election triggered to fill the vacancy, drawn from official state and federal election records. Nothing here is estimated after the fact or adjusted retroactively. The 12 million dollar figure for Graham's seat is the actual, reported cost of that specific special election, not a projection or an estimate. As more cases accumulate, this page will report an aggregate average completion probability across all tracked deaths and a running total of special-election costs, exactly as it does now with one case. The methodology does not change based on sample size. What changes is how much weight a reader should put on the resulting numbers, and right now, with one case, the answer is not much.
For the complete explanation of how completion probabilities are built, what actuarial inputs are used, and how the site verifies special-election costs, see toodamnold.com/data.