Odds and Ends: Week 8

A good sports betting column should be backed by a profitable gambler with a proven track record. It should offer picks generated by a sophisticated and conceptually sound model. Most importantly, it should treat the subject with the seriousness it warrants.

This is not that column.

Instead, this will be an offbeat look at the sports betting industry—why Vegas keeps winning, why gambling advice is almost certainly not worth the money, and the structural reasons why even if a bettor were profitable, anything they wrote would be unlikely to make their readers net profitable, too.

While we're at it, we'll discuss ways to minimize Vegas' edge and make recreational betting more fun, explain how to gain an advantage in your office pick pools, preview games through an offbeat lens (with picks guaranteed to be no worse than chance), and touch on various other Odds and Ends along the way.

Small Samples. Big Results.

Our model picks three games every week. Those picks have returned a profit in all seven weeks, going 15-5-1 against the spread. We have one Lock of the Week. After opening 0-2, that pick has reeled off five straight wins. If you had tailed every single one of our picks this season and gotten (-110) odds on every bet, you'd be making a 40% profit right now. (Well, 39.9%, but what's a tenth of a percent between friends?)

I've been hanging a lampshade on it for weeks, but right now I want to address that record head-on: it is entirely a product of random chance. As I make explicit every week, the "Lock of the Week" is literally selected by a (pseudo)random number generator. There is genuinely a mathematical model driving the three primary picks, but it is every bit as stupid as I have alluded to from Day 1. I suspect several readers have managed to reverse-engineer that model already (despite repeated admonitions not to), and they could all attest that there is no "there" there.

Our record against the spread, in other words, is a fluke. But it is not an accident.

One of the secrets of an internet tout is that everything about the process is designed to make a fluke like this as likely as possible. It would be trivial for me to use my "model" to pick every game every week, and yet I don't. Partly, this is to create the illusion of discernment, to plant the seed of the idea that I am picking and choosing my spots to maximize my advantage.

But mostly it's because small samples make extreme outcomes more likely.

I always say that for the vast majority of bettors, picking against the spread is essentially a coin flip. Over the long run, you should be right about 50% of the time and wrong about 50% of the time, with Vegas skimming the vig off the top to keep the lights on.

Note that this must be the case: if the typical bettor was right significantly more than 50% of the time, Vegas would go broke. On the other hand, if the typical bettor was right significantly less than 50% of the time, they could borrow a page from George Costanza and do the opposite, at which point they would be right significantly more than 50% of the time, and Vegas is once again struggling to pay the electric bill. So we can easily model an average bettor's outcomes with a simple coin flip.

If you flip a coin 20 times, the odds of getting heads on 15 or more of those flips are about 2%, or roughly 1 in 50. That's low, but not especially so. If you flip a coin 40 times, the odds of getting 30 or more heads fall to 0.1%, or about 1 in 1000. (We're rounding a bit—the odds of the former are closer to 1 in 48, while the odds of the latter are more like 1 in 900—but again, what's a tenth of a percent between friends?)

By deliberately keeping our sample sizes smaller, I greatly increase the chances of getting an extreme outcome. Now, that outcome is as likely to be extremely bad as it is to be extremely good; the math for 15 tails is the same as the math for 15 heads. But extremely bad records aren't as much of a risk when I'm capable of wiping the slate clean at any time; I can switch to a new model, or rename my "lock of the week", or change my name and pretend to be an entirely different person with entirely different picks.

I've been writing this column since the beginning of the 2022 season. I've been waiting to write this specific article since... the beginning of the 2022 season. I knew if I kept my samples small and kept wiping the slate clean, I'd eventually have a record that justified it. I'm a fairly patient person, but I also knew that I wouldn't have to be that patient. Hot streaks are a fluke, but given enough tries, they're also an inevitability.

Lines I'm Seeing

The Eagles were blown out by the Cowboys last week, but the Colts barely covered in a low-scoring slugfest, and the Giants beat the Texans in overtime despite being massive underdogs. (Once again, if you let me retroactively pretend that my selections were meant to be moneyline picks, this feature would be much more profitable. Are you sure you won't let me do that? Alas.) We already covered the overall record above, so I won't get into it again.

Today's variable in the model is a bit circular, but it does real work: the "hot hand" adjustment. Every time we pick a team to cover, and they successfully do so, we increment their "hot hand" score by 1. This score continues to increase until we pick the team and they fail to cover the spread, at which point it resets to zero. So if a team is winning for us, we keep picking it over and over again until it finally "cools off".

Arizona (+7) at Dallas

Arizona is the hottest team in our sample, with wins against the spread in Weeks 1, 3, and 6. In fact, Week 1 and Week 6 weren't just covers; they were outright wins against a heavily favored opponent. As a result, the "hot hand" score absolutely loves the Cardinals and will push us to keep taking them until they finally fall short.

Indianapolis (+10) at Jacksonville

The Colts were twice selected as the Lock of the Week (as a 2.5-point underdog against Houston in Week 3 and as a 3.5-point favorite against Washington in Week 4), and have won both games. The model doesn't care about that, though! The model only cares about its own picks, so Indianapolis only gets a +1 point "hot hand" bonus for barely covering the spread last week.

Tennessee (+11.5) at Cincinnati

The Titans are the only "cold" team we're picking this week. They were a winning pick in Week 4, but a losing pick in Week 5, and the "hot hand" adjustment only cares about what they've done for us lately. But let me remind you that the "hot hand" isn't everything; the model includes 14 other variables, and the sum of all of those other variables works in Tennessee's favor this week.

The Lock 'Em Sock 'Em Robot Lock of the Week

A win by the Steelers means the (pseudo)random number generator has successfully picked five straight games. Is this the first five-game winning streak in (p)rng history? Yes and no; it has never managed it in a single season before, but it did nail its final four picks in 2024 and its first selection in 2025.

Can it run its streak to an unprecedented six straight? Definitely maybe!

NY Jets (-1.5) vs. Las Vegas

I could try to unpack the logic of picking the Jets the week after they handed the Dolphins their first victory of the season, but ours is neither to understand nor to meddle, only to observe. And besides, if there's underlying logic, it's no longer (pseudo)random.

Follow @AdamHarstad Follow @Footballguys