Everyone says it is. Counting it turned the question into a different one halfway through.
A fighter who has been knocked out is in more danger next time. Anyone who watches boxing has heard it many times. Something is left behind, and the next time out he goes down again. Is it true?
The population is bouts from 1970 on, where both men had at least 5 prior bouts and the result was decided, and where the previous bout ended either inside the distance or on the cards: 23,724 corner rows. Rows whose previous bout was a draw (4,638), a newspaper decision (2,342) or a disqualification (1,204) are out.
Start with the plain win rate, split by how the previous bout ended.
A corner row is one side of one bout, so a bout makes two of them. It is not a bout count.
After a stoppage loss it is 19.0%; after a decision loss, 24.3%. Exactly as the folklore says. This section could stop here.
This site's rating treats a loss inside the distance more heavily than a loss on the cards. A fighter who has been stopped has already had his rating cut, at that moment.
So the question is not whether losing by knockout is costly. That is already priced in. The question left is whether the cut was ENOUGH. And the four groups differ in more than how the bout ended.
The men coming off a loss are the smaller rating, carry more uncertainty, and have been out of the ring longer. The raw win rate mixes all of that together, so the previous section on its own settles nothing.
The site publishes a win probability before the bout. Actual minus predicted is what the rating missed. Over the whole population it is +0.1, which is well calibrated.
After a stoppage loss it is -8.3; after a decision loss, -7.9. A difference of -0.4. Splitting the same rows on something meaningless -- a hash of the bout -- moves the answer by 0.10, so the difference does clear its control.
Now move the year floor and the opponent's minimum, and measure the same difference again.
Of 9 settings only 5 clear their own control, the difference runs from -2.9 to +0.2, and the sign flips. The knob is the opponent's minimum: with no floor the difference survives, with one it nearly vanishes.
Nothing in this data decides which of those is right. So this article does not say that a stoppage loss costs more than a decision loss. The same question measured on the MMA site did produce a difference. Whether that is the sport or the way the population was cut is not something this can settle.
Read the same column on the MMA site
Fold the four groups into two -- after a loss, after a win -- and a different shape appears. Laid out by predicted band:
Corner rows in brackets.
After a loss every band is below zero; after a win every band is above it. The opening runs +5.3 to +25.0. Across the 9 settings the losing side runs -8.9 to -5.3, always negative, and the winning side +0.8 to +7.0, always positive. Not once the other way round. Remember the whole population sits at +0.1: this is not a miscalibration, it is a split.
A fighter who lost tends to wait longer for the next one, and this site has already measured, in another column, that the rating goes soft on a man who has been away. Is this the same thing under another name? Split it by the months since.
Corner rows in brackets.
The stopped side is flat across all four bands: short gap or long, it is soft by the same amount. The outpointed side deepens as the gap grows, and that one does look like the lay-off effect. So the figure on the stopped side is not the lay-off in disguise.
This is the third sighting. In the lay-off column the rating was too kind to men coming back. In the unbeaten column it was too harsh on men still climbing. Here it is too kind to a man who just lost and too harsh on a man who just won.
All three say the same thing. A rating can only be built out of the past, so it points not at where a fighter is but at the average of the road he came by. The error therefore points the way he is moving.
The next time a result of this shape turns up, suspect this first. It may not be an effect at all, only an estimate running behind.
Age is not measured here, nor weight class, nor the quality of the card. What this split would look like with those in it is not known. And there are two cells in the cross tables that are an order apart from everything around them.
A cell standing on fewer than 30 rows is treated as not measured. These two are above that. Above it and still thin is what this section is about.
Neither agrees with the overall -0.4. Both are above the row threshold, but nothing is claimed from the two of them alone: cut a corpus into enough bands and you get one extreme cell per cut.
When you look at a predicted win probability on this site for a fighter coming off a loss, read it as too high. For one coming off a win, too low. These are win-rate points, not rating points.
And it is not because he was knocked out. It is because he lost. How he lost is not something this record could tell us.
Measured 2026-09-27