Gambling Statistics and the Snapshot Problem
Last year's problem gamblers aren't this year's
Editor’s Note: I’ll soon be contributing to the new Boys and Men Online Substack. If you’re interested, please give it a follow. Also, happy new year! Thanks for reading.
In a 2018 blog post titled Do the Rich Get All the Gains from Economic Growth?, the economist Russell Roberts makes a simple observation: when we compare the bottom 20% of earners in one period to the bottom 20% in another, we’re not looking at the same people.
Some stay poor, but many move up the income distribution and are replaced by new entrants at the bottom. To understand how people are actually faring, you need longitudinal data that tracks the same individuals over time.
Roberts’s conclusion is optimistic. When researchers follow the same people, they find that 84% of children—93% of those in the poorest quintile—out-earn their parents. Headline statistics about stagnant household income, meanwhile, can mislead: as marriage rates have fallen and more Americans live alone, household income can decline even as individuals do better. “When you follow the same people over time,” Roberts writes, “the largest gains often go to the poorest workers; the richest workers often make no progress.”1
“People who have gambling issues, they’re going to have a gambling issue,” DraftKings CEO Jason Robins told Fortune in 2024. “There is a small number of bettors whose activity can become disordered,” wrote Bill Miller, president of the American Gaming Association, the same year in Newsweek.
How small? The National Council on Problem Gambling estimates that 2.5 million adults meet the criteria for a severe gambling problem in a given year, with another 5-8 million experiencing “mild or moderate” problems.2 As I’ve argued elsewhere, these figures likely understate overall harm, since most affected gamblers won’t reach these thresholds. But even taken at face value, they are snapshots. It’s different people in the frame each year.
Longitudinal studies are time-consuming, expensive, and rare. As a group of experts wrote in a 1999 report commissioned by Congress:
There is almost no research that examines the incidence of pathological or problem gambling cases over a representative, recent time period. Nor are there longitudinal studies that provide trend data for population cohorts or that track the progression of individuals into or out of the states of pathological or problem gambling.
The few studies that do exist show high turnover. One longitudinal analysis from Missouri found that only one in eleven problem gamblers at the first measurement remained problem gamblers seven years later. A 2017 WHO report reached a similar conclusion, noting, “In any given year, around half move out of the problem gambling category and are replaced by ‘new’ problem gamblers.”
This might sound, at least in part, like good news—people recover. But many don’t drop out until after the damage has been done: their savings drained, job lost, relationships ruined.3 A thousand people in crisis for a month each is far more dangerous than a hundred people in crisis for a year—even though the total amount of time in crisis is less—because it’s ten times as many people at risk of hitting bottom.4
The episodic nature of problem gambling makes static prevalence estimates misleading. If 5% of gamblers have serious problems at any moment, and half cycle out each year to be replaced, then over a decade the share of people affected isn’t 5%. It’s closer to 15 or 20%.
Churn exists in other addictions too. But gambling harm can accumulate unusually fast.5 A drinker who loses control usually wakes up with a headache; a gambler who loses control might not have a home to wake up in.6
In Mario Puzo’s 1978 novel Fools Die, the casino boss Alfred Gronevelt explains:
Percentages never lie. We built all these hotels on percentages. We stay rich on the percentage. You can lose faith in everything, religion and God, women and love, good and evil, war and peace. You name it. But the percentage will always stand fast.
Gronevelt is describing the law of large numbers. Over enough bets, variance disappears. But while casino revenues may be smooth, individual gamblers’ results are not. Their years are punctured by spikes: the parlay win they’ll never forget, the night of excessive loss chasing they’d rather erase.
Roberts followed the same people over time and found reason for optimism. Applying the same methodology to gambling cuts the other way. When you freeze the frame, gambling harm can look contained. But real life doesn’t stand still, and over time a lot more people wander into the shot.
I don’t fully agree with Roberts’s conclusions—mainly that inequality/lack of social mobility in America hasn’t grown and may not be as large of a problem as many believe—but that doesn’t negate the importance of his observations or this methodological approach. Also, fewer marriages and smaller households is probably bad. The most interesting finding to me from similar analyses is that 73% of Americans were in the top income quintile for at least one year.
To note: I believe these numbers basically haven’t been updated in 5+ years, which strikes me as obviously flawed given far more people are gambling today and thus far more people likely fall into these categories. There have also been lots of debates surrounding these and similar numbers.
This is not to suggest all, or even a majority, of problem gamblers (defined however one chooses) will experience consequences this severe—just that they’re at elevated risk of doing so.
I am not 100% sure of this, and do not have the data to back it up, but I am quite confident in it. If anyone has access to or has seen data that would support or oppose this please reach out. My logic is as follows (focusing purely on the financial losses associated with gambling): in addition to harm being “spiky,” in that bettors often lose a lot in a single session, it is front-loaded. A gambler’s first month in crisis will likely generate the most losses, because that’s when they have the most to lose. Once one has lost everything, it is (at least in theory) impossible to lose it all again. So a hundred people each in crisis for twelve months aren’t experiencing twelve times the harm of their first month. In the most extreme example, imagine if everyone at serious risk lost all their money in the first month. Total losses from the thousand people at risk for a month each would then be (roughly) ten times the total losses from the hundred people at risk for a year (minus whatever the hundred earn/borrow and lose in the subsequent eleven months, which is harder to model).
Though I haven’t found great public data on this, the idea is supported by: common sense, my lived experience, internal casino/sportsbook data I’ve seen, and conversations with hundreds of bettors and people who work in and around the gambling industry. Also of note is that gambling products have changed over time and are increasingly optimized to accelerate addiction/the losing process. For more on that and the lifecycle of gamblers’ losses, I recommend Natasha Dow Schüll’s book Addiction by Design. (One interesting observation Schüll, and others, have made is that casinos often don’t want players to lose too much too quickly, because it’s a bad experience and makes them less likely to want to play again, decreasing their customer lifetime value. Of course, that doesn’t always stop gamblers from doing so.)
This is obviously an oversimplification. I’m not trying to minimize the dangers of binge drinking (even once), but highlight that harms from gambling, especially the financial ones, are often “spikier” than those from other vices.




This felt honest. That’s rare with stories from this side of the fence.