Here's the number: 1. That's how many tracked cases we have so far where a member of Congress died in office and we can check our prediction against what actually happened. One.
We're not going to dress that up as more than it is. This post exists because transparency means showing our work even when the sample is small. So let's look at the one case we've got, and let's be honest about what it does and doesn't prove.
Lindsey Graham, Republican senator from South Carolina, served 31 years before he died at age 71. At the time our model last recorded a completion probability for him, that number stood at 82 percent. In plain terms: our system estimated an 82 percent chance he'd finish out his term. He didn't.
When a senator dies in office, states don't just leave the seat empty. They trigger a special election, and special elections cost real money: ballots, staffing, polling places, security, the whole apparatus of democracy spinning up on short notice. In Graham's case, that triggered cost came to $12 million.
Across our one tracked case, the average last-recorded completion probability is, unsurprisingly, also 82 percent, because averaging one number with itself gives you that same number back. And the total actual cost of triggered special elections across all tracked cases so far is that same $12 million. There's no averaging to be done yet. There's just one data point.
Here's why we're publishing a page about a sample size of one instead of waiting until we have more data: because the whole point of Too Damn Old is showing our work in public, including when the work is early and thin.
Our completion probability isn't a prophecy. It's a data-driven estimate of how likely a sitting member is to finish their current term, based on factors like age and tenure. An 82 percent probability doesn't mean someone will almost certainly serve out their term. It means the model still saw meaningful room for something else to happen, and in this one tracked case, it did.
A 71-year-old with 31 years of service isn't an outlier in Congress. That's practically the norm in a body where seniority is currency and there's no mandatory retirement age. Members of Congress are people, and people serving well into their seventies and eighties face the same mortality realities as anyone else that age, whatever their office.
What this means for you as a reader: use our completion probabilities as a lens on risk, not a guarantee. With only one tracked case, we can't tell you how often 82 percent probabilities pan out versus fall short. We can only tell you what happened this one time. Don't read a single outcome as proof the model runs too hot or too cold. Read it as the first entry in a ledger we intend to keep filling in, in public, as it happens.
Every member tracked on Too Damn Old gets a completion probability: our estimate of how likely they are to finish their current term, calculated from data like age and years already served. We record that number as a snapshot, and when a term ends early, whether through death, resignation, or another exit, we go back and check the last recorded probability against what actually happened.
When death in office triggers a special election, we also track the actual public cost of that election where the data is available. That $12 million figure for Graham's seat isn't an estimate. It's the recorded cost of the process his state had to run because a Senate seat came open outside the normal election cycle.
Over time, as more cases accumulate, this page will let us calculate real accuracy metrics: how often our probabilities line up with outcomes, whether certain age brackets are systematically over or under-predicted, and how much triggered special elections cost taxpayers in aggregate. Right now, with one case, we can't calculate any of that meaningfully. We can only show you the one data point we have and promise to keep updating this as the list, unfortunately, grows.
This page will get more useful as our dataset grows, and we'll keep updating it honestly, including when our predictions miss. If you want the full detail on how completion probabilities are built, what data feeds them, and how we track outcomes over time, head to toodamnold.com/data.