How lucky are the Cowboys?

Blogging The Boys’ Tom Ryle wrote something recently that got me thinking:

Luck is such a huge part of the game.

There were two times when a pass was affected directly by a defensive player as it left the quarterback’s hand, one for each team. When Prescott got one slapped by a pass rusher, it went directly into the hands of Oakland cornerback Sean Smith, leading to an interception. But when almost exactly the same thing happened to Derek Carr, the ball landed in the hands of one of his own receivers.

Now, Tom was speaking about specific plays in a specific game but it got me wondering whether the Cowboys have been a lucky or unlucky team over the last 25 years or so? Specifically, have the Cowboys been lucky in 1-score games?

I wondered about 1-score games because over the years the analytics crowd has advanced the idea that 1-score games are basically a coin-flip: over time teams will win or lose 1-score games at the same rate. BTB Hall of Famer rabblerousr repeatedly made this point:

The problem is that close games are highly subject to luck; in large sample sizes, all teams' winning percentages in such games, regardless of overall record, are roughly .500.

Good teams aren't good because they win close games, but because they managed to avoid close games.

Bill Barnwell has written extensively on this topic. Here’s his basic outline of the reasoning:

Record in Close Games

In a Sentence: Teams are incredibly inconsistent from year to year when it comes to winning games that are decided by one touchdown or less.

How It Works: No obtuse formula here. Just count up each team’s number of games that were decided by one touchdown or less, check their winning percentage, and then see if they were similarly good or bad during the following season. When you do, we bet that you’ll find it’s essentially random.

Why It Works: Because, as we mentioned in the intro, a few close games per year isn’t enough to draw any conclusions.

Prove That It Works: Let’s start with a group of teams that were dominant in close games during given NFL seasons. Our arbitrary group of teams played six or more games that were decided by a touchdown or less in those seasons and each of them won 75 percent or more of those games. In all, those teams went a combined 449-102 (81.5 percent) in close contests. If there were really something consistent about how a team performs in the tight ones, these teams would at least emulate their record during the following season. Instead, they went a combined 256-249 (50.7 percent) in those same close games the following year.

If we accept the premise that all teams should win around 50% of 1-score games, then any team that significantly deviates from that 50% number would be considered particularly lucky or unlucky. So how have the Cowboys of the Jones era performed, have any been particularly lucky or unlucky? To find out I looked at every game of every season since 1989 and segmented them based upon score differential:

• Games decided by 7 or fewer points are defined as 1-score games
• Games decided by more than 7 points are defined as “blowouts”.

I then looked at only the 1-score games and determined how much the Cowboys deviated from the expected number of wins. For example, the 2000 team played 7 1-score games; the expected win total for that team would be 3.5 (7 * 0.5). However, the team actually went 2 - 5 in such games. The difference between the expected win total (3.5) and the actual win total (2.0) would be attributed to “bad luck”.

Here’s the Cowboys year-by-year deviation from expected win totals in 1-score games:

This is a verrry interesting chart to me, so some observations:

• The “luckiest” Cowboys teams seem to be the 2003 and 2016 squads (each winning two more games than expected in 1-score games). Make those teams .500 in 1-score games and the Quincy Carter-led 2003 squad misses the playoffs with an 8-8 record while the 2016 team would have gone 11-5, but still have won the NFC East and enjoyed a first-round playoff bye.
• The “unluckiest” Cowboys teams seem to be the 1999 and 2010 teams that lost 2.5 more games than expected in 1-score games. That 1999 team finished 8-8 and made the playoffs as a wild card so two or three more wins might have won the division (Washington went 10-6 to claim the East that year).
• The overall-up-and-down, relatively randomness of the results supports the idea that 1-score games are 50-50 propositions.

But then I noticed four things that really caught my interest:

1. The 1-15 1989 team was also unlucky, losing two more 1-score games than expected.
2. The early 90’s triplets dynasty was generally quite lucky, winning a total of five more games than expected over the 91-96 seasons.
3. The bad Dave Campo era was unlucky every single season, losing 4.5 games more than expected over 3 seasons.
4. The 2014-2016 teams saw the team’s “luck” exactly mirror the team’s overall performance:

• “Good luck” in both 2014 and 2016 when the team enjoyed big winning seasons (75% or higher win percentages)
• “Bad luck” in 2015 when the team suffered a bad losing season (25% winning percentage).

Hmmm......

If the theory that overall win percentage has no effect on close games there should be no correlation between overall win percentage and 1-score win percentage. So let’s check those numbers; the following shows the same chart as above and adds the team’s overall win percentage (the red line):

Well, that’s an interesting chart. While it doesn’t perfectly correlate, you can definitely see that when the team is better overall, it’s usually also better in 1-score games.

I’ll admit, at this point I’m beginning to question my own belief in the idea that 1-score games are largely a matter or luck. There’s simply too much correlation here between good Cowboys’ seasons and “lucky” results in 1-score games. So, I decided to dig a little deeper.

The obvious objection to the above chart is “well of course the overall results match the lucky results because the lucky results are driving the overall results”. This basically says those “lucky” results play a large factor in determining the overall results and as the lucky results go up (or down) so too do the overall results.

And that’s a perfectly reasonable objection. So, let’s go further and look at performance by 1-score games compared to “blowouts”. If 1-score results are really a 50-50 proposition there should be absolute no correlation between 1-score results and blowout results. The following compares the Cowboys performance in these two metrics:

While the lines don’t perfectly mirror each other, there’s definitely more correlation than you would expect if 1-score games should have no correlation to blowouts. So, is it really true that a team’s overall performance has no impact on a team’s performance in 1-score games? I’m not so sure any more.

Each of these are individual seasons where random variance can play a significant role. After all, with only 16 - 19 data points each individual season is significantly impacted by random results. Let’s try to “smooth” these results by creating three-year rolling totals.

Again, while not perfect, this seems to show that as overall performance goes up or down so too does the team’s performance in 1-score games.

These are individual season results, what do the aggregate results over these 28 seasons show?

I segmented each Cowboys’ season based upon their overall win percentage:

• Teams with >67% win percentage are defined as “good”
• Teams with <33% win percentage are defined as “bad”
• Teams between 33% and 67% are defined as “mediocre”

As expected, rabblerousr is right, over the 217 1-score games during this time-frame the Cowboys are 107 - 110 for a 49.3 win percentage. This is exactly what the analytics crowd would predict. However, that’s just a gross overall result. Here are the results when we look at good, bad and mediocre Cowboys teams:

Hmmm....this sure seems to indicate, for the Cowboys at least, that good teams in fact do win close games at a higher rate than poor teams. Some might claim the sample sizes here are too small but even the smallest group (the “bad” teams”) has a 40-game sample, which seems pretty large. Here’s the table:

If teams split 1-score games at a 50/50 rate regardless of overall record we would not expect the above results. We would instead expect the results to be somewhat similar in each team quality segment.

Now, it is noteworthy the bad team won a higher percentage of 1-score games than blowouts (30% vs 20%) and good teams won a higher percentage of blowouts than 1-score games (82% vs 62%). But the idea that “good teams win close games” seems to stand.

What about the number of “luck” wins? Do they correspond with the above? The following shows the aggregate “luck” record segmented by quality of season:

This tells us that “bad” Cowboys teams won eight fewer times than expected in 1-score games while “good” Cowboys teams won 7.5 time more than expected in 1-score games. Again, this seems to support the notion that “good teams win close games”.

We could also turn these metrics around. Rather than segmenting seasons into good or bad and then seeing how lucky they were in each season, we could segment seasons into lucky or unlucky and then see how those teams performed in blowout games.

• “Lucky” teams are defined as those who enjoyed 1.5 or more wins above the expected number in 1-score games.
• “Unlucky” teams are defined as those who suffered 1.5 or more losses below the expected number in 1-score games.

Again, if the idea is that 1-score results are based upon luck, we would not expect to see a correlation between 1-score results and blowout results. But that’s not what we find:

Now, some people could say this is all just random variance and the result of small sample sizes. But it seems very compelling to me that these numbers all seem to support the idea that the better Cowboys teams performed better in close games while the lesser Cowboys teams performed poorly in close games.

This would be a strong argument against the idea that 1-score games are coin-flip propositions where luck largely rules.

Most of the writing on this topic focus on “lucky” or “unlucky” team’s failure to sustain their luck status over time. Every season Bill Barnwell writes an article where he predicts which teams will perform better or worse than expected; teams that were outliers in terms of luck the previous season usually are involved. And he’s right; pretty much every time a team has extreme luck (such as the 2016 Cowboys with their 7 - 3 record in 1-score games) are unable to sustain that luck the next season (Cowboys are 1-2 in such games in 2017).

(Barnwell’s predictions are often quite accurate as well. This season he forecast the Jaguars, Eagles, Chargers, Browns and Cardinals to improve and the Cowboys, Giants, Raiders, Texans and Dolphins to decline; that’s 8 of 10 correct.)

But that doesn’t tell the whole story. Winning seven of ten games is hard in the NFL; only a really good team can do that. The 2016 Dallas Cowboys were a really good team even if you subtract their luck (6-1 in blowouts). Maybe the 2017 team didn’t win a high rate of 1-score games because the 2017 team simply isn’t as good (7-4 in blowouts).

I’m not saying the idea that 1-score games are coin-flips is wrong; I’m saying for the Cowboys the results over time don’t support the finding.

Here’s the entire table I used for this analysis for those interested:

Next time we’ll look at two franchises that should test these findings. The New England Patriots have been consistent winners for the last 17 years while the Cleveland Browns have been consistent losers for that same time period. IT should be interesting to see if the results we’ve observed the Cowboys hold up for those two franchises as well.