Detroit Sports Reporter

• Tigers
• Lions
• Pistons
• Red Wings

2026-08-09 · Detroit Tigers · Analysis

Are the Tigers bad at close games, or unlucky? Five seasons say both, and mostly the second one

Last week I wrote that the Tigers were the best team in the AL Central and in fourth place, and that the gap between their record and their run differential was a promise the rest of the season would keep.

The comments on that piece were better than the piece. The sharpest one said the reasoning was not just optimistic but conceptually wrong: run differential and Pythagorean expectation get read as a guarantee of regression, and they are not. Blowing out the Athletics does not put a run in the bank. A team that keeps losing one-run games might simply be a team that is bad at one-run games, and calling that luck is just refusing to look at it.

That is a real argument and I did not test it before publishing. So here it is, tested, with the answer up front: the reader is right about the mechanism and wrong about the size. Close-game performance is partly a real, repeatable property of a team. It is roughly 61 percent as repeatable as an equal number of ordinary games. That leaves Detroit better than its close-game record but worse than the pure-regression story implies, and it moves the projection by about two wins rather than the seven the Pythagorean gap advertises.

What Detroit's season actually looks like split by margin

Every game Detroit has played through the final whistle on August 8, from the MLB Stats API, regular season only.

A reader also corrected the framing I used, and the correction was an improvement. I had leaned on the 12-20 one-run split. The better number is 26-44 in games decided by three or fewer: same story, 70 games instead of 32. Their figure was 26-45 and the true one is 26-44, off by a single loss, which is closer than most published numbers get.

That 31-16 in blowouts is the part nobody talks about. Detroit beats good teams badly and loses to bad teams narrowly, and the distance between those two things is not normal.

Detroit is not merely at the top of that list. At +.288 the gap is more than 1.6 times the second-place Cubs at +.173, and the gap between Detroit and the Cubs is wider than the gap between the Cubs and ninth place. Whatever is happening here, it is happening to the Tigers more than to anyone else in baseball.

The number I found first, and why I am not using it

There is an obvious explanation sitting right there. Detroit has converted 22 of 47 save opportunities, 47 percent, with 25 blown saves, second most in baseball, behind a pitching staff whose 3.52 team ERA is fourth best in all of MLB. Good pitching, handed back after the seventh.

Across the 30 teams, save conversion rate correlates with close-game win rate at r = +.783, and with blowout win rate at +.069. That looks like a smoking gun: bullpen quality drives close games and nothing else.

It is not evidence, and I want to be the one to say so. A save opportunity is by definition a lead of three runs or fewer, and a blown save in a close game very often is the close loss. The two statistics are built out of overlapping events. Correlating them measures the dictionary, not the world. If I published that +.783 as proof, someone would eventually notice, and they would be right.

So the test has to be one where the answer is not baked in.

The actual test

If losing close games is a property of a team, then a team's close games should predict its own other close games. Deal each team's close games alternately into two piles, odd and even, and correlate the win rate in one pile against the other across every team. Same roster, same manager, same bullpen, no aging, no trades in between. That is the friendliest possible test of the skill hypothesis. Then run the identical test on all games as a control, because team quality is unquestionably real and has to survive it.

On the 2026 season alone, it does not settle anything:

Thirty teams is not enough. The coin-flip simulation shows that a correlation anywhere inside roughly plus or minus .30 is what pure randomness produces at these sample sizes, and both the close-game figure and its control sit inside that band. Anyone who stopped here and reported "+.097, therefore luck" would be reporting noise with a decimal point on it.

So I ran the same thing over the five completed seasons from 2021 through 2025, which gives 150 team-seasons instead of 30, on full schedules instead of two-thirds of one:

Now it separates cleanly, and the comparison that matters is the last two rows, because they use identical sample sizes. A random slice of a team's schedule predicts the rest of that team's schedule at +.583. The same number of close games predicts the team's other close games at +.290.

Stepping those up to full-season reliability gives .449 for close games against .737 for ordinary ones. A close-game record carries about 61 percent of the repeatable signal that the same number of ordinary games carries.

That is the answer to the objection, and it is a split decision. Close-game performance is not noise. Something real and repeatable is in there, and a reader who says "this team is bad at close games" is describing something that exists. But it is a good deal less real than the raw record makes it look, and treating 26-44 as a fixed property of the Tigers overstates it by a wide margin.

What that is worth, in wins

Regressing Detroit's .371 by that reliability puts their true close-game talent at about .442. Sixty percent of their games this year have been decided by three or fewer, so call it 27 close games in the 45 that remain.

Two wins. Not the seven that the Pythagorean gap dangles, and not zero either. Detroit is 2.0 games back of the third wild card. Two wins is the whole difference between playing in October and watching, which is a genuinely strange place for an argument about statistical reliability to land.

For what it is worth on the luck side of the ledger: if close games were coin flips, a team playing 70 of them finishes 26-44 or worse about 2.1 percent of the time. That is unusual. It is also the kind of unusual that shows up somewhere every year when 30 teams each get a turn.

The objection that survives all of this

A different reader made a point that none of the above touches, and it is the best argument against everything I wrote last week.

Detroit's worst split is against its own division: 11-18, a .379 clip. Last week's piece argued that the 20-odd remaining head-to-head games against the teams ahead of them were the path back into the race. Against a team playing .379 ball in exactly those games, the same schedule is the fastest available route to elimination.

Here is what I found looking into it, and it cuts both ways:

The division record is almost entirely a close-game record. Twenty-three of those 29 games were decided by three runs or fewer, which is a far higher share than the 60 percent Detroit runs overall. So the division problem and the close-game problem are not two independent pieces of evidence stacking up against the Tigers. They are largely the same fact counted twice, and the regression above applies to most of it.

But 0-6 against Cleveland, with five of those six decided by three or fewer, is its own small horror. Seven of Detroit's remaining 45 games are against Cleveland, and 23 of the 45 are inside the division. A team that has not beaten the Guardians once all year now needs that specific matchup more than any other on the schedule.

That is the honest state of it. The close-game record is about 61 percent as meaningful as it looks, which is worth roughly two wins and probably a wild card spot. The division record is mostly the same finding wearing a different hat. And a winless record against the team they will see seven more times is not something any correlation coefficient is going to explain away.

One correction, on a number nobody asked about

While pulling this data I found that my own earlier recomputation of Detroit's record gave 55-60 where the standings said 56-60. The missing game is April 4 against St. Louis, an 11-6 Tigers win called for rain. The MLB API returns it with a status of "Completed Early" rather than "Final", and my filter matched only the exact string "Final", so a real win vanished from every game-by-game calculation I ran. It is fixed here and in the backtest script. With Saturday night's 8-0 win in San Francisco added, the Tigers are 57-60.

It is a small thing. It is also exactly the kind of thing that quietly corrupts a number in a piece like this, and the only reason it got caught is that a recomputed figure disagreed with a published one by exactly one game and I wrote it down instead of shrugging.

Sourcing: every game result, margin and standing here comes from the MLB Stats API, pulled 2026-08-09. Season data covers 1,757 decided regular-season games in 2026 through August 8, plus the complete 2021 through 2025 regular seasons for the reliability test. Bullpen figures are the team pitching endpoint for the same date. The chart and every figure in it are generated by scripts/close_gap_chart.py; the reliability tests are scripts/close_games.py, which takes --margin and --seasons so anyone can re-run them at a different definition of "close". Both are in the repository along with this entry.

Not betting advice. Just analysis, made in public and kept in public.