20 is the New 18#

March 26, 2025 | Last Updated: April 24, 2025

Are Big Leads No Longer Safe?#

“When comparing the chances of pulling off a 20-or-more point comeback the shift is more about 2 points, not 8.”

There’s a perception that, in recent times, teams are climbing out of big deficits more often than ever before. Just as Kevin Pelton and Baxter Holmes noted in 2019, so did ESPN reporter Andrew Lopez in ‘20 is the old 12’: Why no lead is safe in the NBA anymore. There he points out that the 2023-24 season had already seen the most 20-point comebacks in a single season since 1996-97 and continues with analysis by Steve Kerr:

“Way more early possession 3s now. It just feels like you are up 12 and the other team gets two quick stops on you and they race down, they throw it ahead and they hit two 3s. It’s a six-point game now. So 20 is the old 12; 12 is the old seven. I mean, there’s definitely an awareness from everybody that leads are not safe.”

But is this true? In short… not really. While there’s certainly been a significant shift, when comparing the chances of pulling off a 20-or-more point comeback the shift is more about 2 points, not 8.

To show this, let’s look at a chart showing the percent chance of coming back when a team is down N or more points at some point in a game:

Looking at this data, the chance of coming back from 20 points-or-more down (~5.3%) is about the same as coming back from 18 points-or-more down (~4.9%) in the earlier era. And this shift remains fairly constant comparing other points on the graph.

To put this in perspective, this overall 2 point shift is on par with but not quite as pronounced as the advantage a home team has over a road team when attempting a comeback. Note, there are many ways to compare the comeback chances of the eras at different points in time during the game which affects the size of the shift (explained below), but roughly 2 points is a pretty good average. ()

To note, there has been a marked shift in the data: there are about 1.8 times as many 20-points-or-more comebacks comparing the eras. And when viewed through the lens of a point spread in sports betting, a 2 point shift is significant.

But as a fan watching the game, the data shows that leads are roughly as safe as they have been, perhaps needing an extra bucket to maintain the same level of security.

Breaking Down The Eras#

To frame this, I decided to break up the available play-by-play data into:

  • 1996-97 to 2016-17 (old-school) versus

  • 2017-18 to 2024-25 (modern)

The change in increased scoring and comeback chances appears to start changing most dramatically around 2016. And if you:

So—always wanting as many games as possible to reduce the statistical noise—I felt that was the fairest breakdown: 1996-2016 v 2017-2024.

Win Percentages When Max Deficit is N Or More Points#

Again, looking at a chart of comeback odds versus being down N or more points over a game:

The percent chance of coming back versus point deficit is fairly normally distributed under most conditions. So when plotted on a normal probability plot and normal trend lines are fit to the statistical data one can then better examine the lower probability events and also more easily compare different eras or situations.

And when comparing these two eras, one can see a mostly stable and constant shift of about 2-3 points to the left for the modern era versus the past.

You can interact with this chart and hover over and click the points to see which games compose a point and compare it with a list of biggest comebacks if so inclined (on mobile, go full screen before clicking). To focus on a few we get:

Win % Increases When Comparing Modern Versus Old School Eras#

Points Down Or More

1996-2016 Win %

2017-2024 Win %

Total Win % Increase

30

0.15 %

0.42 %

2.8x (180% increase)

20

2.89 %

5.31 %

1.83x (80% increase)

18

4.88 %

7.93 %

1.59x (60% increase)

15

9.22 %

12.82 %

1.39x (39% increase)

So while coming back from down 30 or more happens 2.8 times more often than in the past, it’s still very unlikely. In fact, you need to move over a little less than one three-pointer to get about the same chance: in the old school era, if you were down -27 or more there was about a ~0.50% chance of winning.

Note, this table above uses the raw game data points, which is a little more intuitive. You can also do this using the trend line in the chart, which cleans up the noise in the data and is statistically more accurate. Overall, either way draws the same conclusion.

As time dwindles, this shift is slightly smaller. Looking at biggest 4th quarter comebacks we get:

Now, the shift is about 1.5 points, an even smaller shift.

Win Percentages When Teams Are Down N Points With So Much Time Left#

Another way to look at it and the more natural way to think about it while you are watching a game live (as opposed to describing a game after the fact) is to look at the win percentages when teams are down exactly N points with so much time left. Here’s a chart for the start of the 2nd half:

The data is a bit noisier here, because we are not accumulating the games as we move from left to right like we did when looking at points down or more. Here, for the old school era, we have the case that there was one game (11/27/1996 DEN @ UTA: 103-107) where UTA was down -34 at the half and won. But no team in that era won when down exactly -33, -32, -31, -30 or -29 at the half.

Now, the divide here is no longer as much of a shift as a change in slope of about 20% more for the modern era. So being down 20 in the modern era is about the same as being down 16.75 points in the old school era (a 3.25 point shift). And being down 10 points now is the same as being down 8.5 points in the past (a 1.5 point shift).

But as time decreases it tightens up – looking at starting the 4th quarter we see that the change in slope is about 13% now – so being 20 points down (~0.17%) is about the same as 17.75 points down when comparing new to old:

Win % Versus Time#

Collecting this data for every minute, we can focus on a single percentage and plot the time v point deficit for that percentage. Doing that for 10% while comparing the eras we get:

And doing it for 1% we get:

The point spread changes over time, but between 2 - 3 points is a fairly good characterization.

Real-Time Probabilities#

One interesting way to look at the data is to show what the running probabilities would be using data from the modern era versus the past. Note, this probability calculator doesn’t account for team or player strength or home court advantage – it’s based solely on win/loss data from the specified time ranges.

So at halftime, the Wolves had a 12.9% chance of winning using modern data versus a 9.1% chance if you use data from the past.

Here’s another game:

The same basic trend emerges: from a statistical standpoint, there are notable large shifts in percentages comparing the eras (3-4% in some cases). But the shift is not large enough to greatly affect how it “feels” as a fan to be down, say, 15 points at the half.

What An 8-Point Shift Looks Like#

I thought it would be useful to get a sense of what an 8-point shift looks like and the kind of trend I was expecting to see. So if we look at top 10 teams (out of 30) vs. bottom 10 teams in the modern era we get:

Here, the chance of a top ten team playing a bottom 10 team coming back from a 20-points-or-more deficit (~20%) is about the same for any team coming back from 12 or more down against any other team. And, over the last 8 years, the chance of coming back from 15 points or more is about 13% – but for a top ten team playing a bottom ten team, it’s about 34%. That feels very different and is what an 8-point shift in the data would foretell.

What Else Could Be Going On?#

Blowing a 20-point lead now has about the same chances as blowing a 17 to 18-point lead in the past. This two to three point shift is significant but I don’t think it matches what I see people saying about leads these days.

So I just want to consider a few points that I think are useful when thinking about this issue:

  • Sometimes people refer to the increase in percent chance, which is large, but overall the percent shift is not dramatic enough to support the narrative. For example, the percent of times a team came back from 15 or more down is about 12.8% recently compared to 9.2% in the past. While that’s an increase of about 40%, I don’t think having a 12.8% chance versus a 9.2% chance gives you a markedly different feeling about the game. If you compare that to the 8-point shift graph, here the percent chance is about 34% for a top 10 team to come back 15 or more against a bottom ten team over the last 8 years. That is dramatic and would warrant the kind of talk of “a 15-point lead doesn’t matter anymore”.

  • Scoring is up dramatically. The median score is 111 points in the modern era as compared to 97 in the past: a 14-point shift. Also, teams getting down 30 or more points happens 60% more often and teams getting down 20 or more points happens 30% more now than in the past.

    However, the fact that teams can score more quickly does not automatically lead to the conclusion that big comebacks should be more probable: if you are down and can score quickly, so can your opponent and keep you down. So while points are a little easier to come by, this effect is offset by the counterpoint that they are also a little easier to give up.

  • I do think part of the problem is availability bias: that is, things that happen recently, especially ones you’ve witnessed personally, lead to the untrue conclusion that they are happening more often than in the past. An example of this phenomenon is that after people hear about a plane crash in the news, they think that plane crashes are getting more common. After personally watching a series of garbage time games recently, I got to thinking “for sure this is an increasing phenomenon”. But when I look at the data it’s almost constant (there are slightly more large deficit games but this is offset by a 2-point shift in a team’s chances of coming back).

  • I do want to compare the average size and min/max size of runs between the two eras. It may be that large swings are much more common, leading to a perception of increased chances of coming back. But here again, if it’s an option for you to go on a run, it’s also an option for your opponent. For another day.