clock menu more-arrow no yes mobile

Filed under:

2019 Cal Football Season in Review: Did the Bears Meet Expectations?

In which I make the argument that the Bears underperformed, overperformed, AND met expectations. Buckle up, this is going to get interesting.

NCAA Football: California at Stanford
Judging by this moment? Absolutely. Judging by the October slide? Not so much...
Neville E. Guard-USA TODAY Sports

The 2019 season brought an up-and-down conclusion to an up-and-down decade for Cal football. Had you known in August that we would sweep the OOC slate and beat Washington, UCLA, and Stanford, you probably would have expected it to be a banner year for the Golden Bears. And if you had known that we would start the season off with a 0-10 deficit to UC Davis, score a whopping three touchdowns in October, and manage to score fewer points than 2018’s offensive disaster, you probably would have expected an abysmal year. Without any of that advance knowledge, our preseason predictions were somewhere between those two scenarios. Our predictions produced a projected win total of 7.21, which was right on target for our 7 regular-season wins. But the journey to those 7 wins was probably not what many of us had envisioned.

Revisiting our 2019 Predictions: How the Bears Met Our Expectations

As is our custom, in August we asked you all to give us the Bears’ chances of winning each game in the 2019 schedule. Our predictions had us as heavy favorites in three games, UC Davis, North Texas, and Oregon State, slight favorites against ASU, moderate underdogs in three games, @UW, @Oregon, and @Utah, and slight favorites in five toss-ups: @ Ole Miss, @LSJU, @UCLA, and at home against WSU and USC. Let’s see how the season unfolded compared to our preseason predictions.

Opponent Win Chance Outcome
UC Davis (5-7) 94.16% W (27-13)
@ Washington (8-5) 39.18% W (20-19)
North Texas (4-8) 85.36% W (23-17)
@ Ole Miss (4-8) 59.50% W (28-20)
ASU (8-5) 63.51% L (17-24)
@ Oregon (12-2) 33.30% L (7-17)
Oregon St. (5-7) 83.51% L (17-21)
@ Utah (11-3) 37.97% L (0-35)
Wazzu (6-7) 58.29% W (33-20)
USC (8-5) 54.32% L (17-41)
@ LSJU (4-8) 56.56% W (24-20)
@ UCLA (4-8) 55.62% W (28-18)

Overall, the Bears performed almost as exactly as you would expect based on Cal’s 8-5 record and the record of our opponents. The Bears beat everyone with a worse record, except Oregon State, went 1-2 against teams with the same 8-5 record, and lost both games against teams with better records. Similarly, the Bears fared pretty well in games in which we favored them (6-3) and less well in games where they were underdogs (1-2).

While it may be tempting to expect Cal to have a better record in games where they were favored, recall that we had five toss-up games and we favored Cal in all of them. Flip a coin five times in row. You’ll only have a 3% chance of getting five heads. To go 4-1 in those toss-ups is pretty good.

To illustrate how our season unfolded compared to how it should have unfolded (assuming our predictions were accurate), take a look a the plot below. I show the probability that Cal ends up with each possible record (1-0 or 0-1 after one game, 0-2, 1-1, or 2-0 after two games, and so on). The number in the box indicates the probability of a given record (multiply it by 100 to get the chance) and I’ve put gold boxes around where we actually were after each game.

The actual number of wins (gold boxes) superimposed on the spread of our projected number of wins over the course of the season.

After a 4-0 start, a 4-game losing streak, and a 3-1 finish to the regular season, we landed at 7-5, the most likely outcome based on our preseason predictions. Comparing the final record to our preseason predictions, we can make the argument that the Bears met expectations in 2019.

Revisiting our 2019 Predictions: How the Bears Underperformed After a 4-0 Start

Now let’s take another look at that above plot, particularly during the first four games. We had a 94% chance of being 1-0 after UC Davis, a 37% chance of being 2-0 after UW, a 32% chance of being 3-0 after North Texas, and a 19% chance of being 4-0 after Ole Miss. According to our probabilities, being 4-0 was only the third-likeliest outcome after those first four games. We should have been 3-1 or maybe 2-2.

Although we were riding high on that 4-0 start, the following slide was particularly disappointing for two reasons. First, once we’re 4-0, those four wins are locked in. Let’s go back to that example of flipping a coin 5 times in a row. If, by chance, you get 4 heads in a row then you will be in a very improbable place. But after those 4 heads have been flipped, you still have a 50% of getting that 5th heads. Once those results are in the past, they don’t have any bearing on the next result. The fact that it’s only a 3% chance of getting 5 in a row doesn’t matter after you’ve already flipped 4 heads. It’s not like the coin suddenly drops from a 50% chance to a 3% chance for that final coin flip. Similarly, after we started with 4 wins, our predictions in the remaining 8 games suggested that we should win another 4.4 games. After winning 4 wins, we weren’t suddenly due to start losing simply because we overachieved in the first four games. Those wins were locked in and there was no reason to expect a 3-5 finish to bring us to our 7-win prediction. Instead of finishing 7-5, we were more likely to be 8-4 or 9-3 at that point. So ultimately finishing with a 7-5 record was an underperformance after a 4-0 start.

The second reason that slide was frustrating was due to correlated errors. Correlated errors occur when the same mistake is made over a series of predictions. This is why it wasn’t all that improbable that the 2016 election turned out the way that it did. The outcomes in Pennsylvania, Michigan, Florida, and Wisconsin weren’t all separate, independent events. Instead, they were linked. Polls in those four states weren’t all wrong by chance; they were wrong because they underestimated turnout of non-college-educated whites who voted for Trump. But this wasn’t an error isolated to a single state. Underestimating their turnout in PA means it’s more likely to underestimate their turnout in MI too. And if both those are underestimates, then it’s pretty likely that whatever mistakes were made in underestimating those turnouts in PA and MI will lead to underestimates in FL and WI too. Suddenly, a minor underestimate turns into a series of missed predictions. Now let’s apply that same logic to the Bears.

After that 4-0 start, we may have started to think we had underestimated the Bears in our preseason predictions. If they overperformed to the tune of a 4-0 start, maybe we had underestimated their chances in remaining games... Instead of expecting another 4.4 wins after the 4-0 start, maybe it was closer to 4.8 or 5.2 or maybe even 5.5. So now we’re thinking that we’re definitely not going 7-5, we’re going 9-3 or 10-2. As a result, falling back to 7-5 would be a major disappointment. By this line of reasoning, the Bears failed to meet expectations after that 4-0 start.

Revisiting our 2019 Predictions: How the Bears Overachieved and Were Lucky to Be 7-5

We obviously didn’t finish with that lofty 9- or 10-win record. In fact, it’s not unreasonable to say that rather than being a 10-2 team, we performed more like a 5- or 6-win team.

College football games produce binary outcomes: either you win or you lose. A team that wins every game 20-19 finishes with the same 12-0 record as a team that wins every game 105-0. But the team that won 12 toss-up games clearly was much luckier than the team that won every game in a spectacular blowout. This is because the final score provides some insight into how likely the winning team was to win that game. That’s the underlying logic behind the Pythagorean Projection, a measure used by Pro Football Outsiders to turn point differentials into estimated numbers of wins. They developed the following formula to predict wins based on point differentials.

WARNING: We have received reports that math is in the immediate vicinity. Please take precautionary measures to protect people and property from harm.

                               (POINTS SCORED)^2.37
Win likelihood =~  ---------------------------------------------
                    (POINTS SCORED)^2.37 + (POINTS ALLOWED)^2.37

NOTICE: The threat of math has been extinguished and there will be no further math. Please resume normal activities.

I used the final scores to calculate the Pythagorean Projection for each game of the 2019 football season. Below I’ve plotted those win likelihoods in gray. Anything above the 50% line is a win while anything below the 50% line is a loss.

Our chances of winning each game based on the final score.

Based on the results on the field, we only had one decisive win (vs. UC Davis) but we suffered three decisive losses (at Oregon, at Utah, vs. USC), all of which gave us only about 10% chance of winning. By contrast, six of our seven wins were around the 50-75% range, probable but far from guaranteed. With a handful of lopsided losses and wins that ranged from nail-biting to not-quite-comfortable, the Bears tended to perform decently in wins and rather poorly in losses. If we combine all those win probabilities across the 12 games into a projected number of wins, we only get 5.76 wins for the Bears. This suggests that Cal’s 7 wins were an overachievement and, in part, thanks to a fortunate 4-2 record in one-score games. The failures of UW, North Texas, and LSJU to score on their final drives made the difference between a 7-5 season and a 4-8 season.

Comparing Predictions to Reality

Throughout this section we will compare the outcome on the field to various predictions: our preseason predictions, our volatile pregame predictions, and the arguably more objective predictions from Vegas. First, let’s see how well the community predicted how we would fare this season.

Preseason Predictions vs. Reality

Outcomes compared to our preseason predictions.

A useful way of evaluating the team’s performance compared to our predictions is to compare the gray points and the blue points. If the blue point is above the gray one (North Texas, for example), the team underachieved compared to our preseason predictions. If they’re pretty similar (Big Game), then we met expectations. And if the gray point is notably higher than the blue one (UW, Wazzu) then we exceeded expectations.

We do not observe any major departures from reality over the first four games (although I told y’all not to take North Texas so lightly). Once Garbers broke his collarbone, it was all downhill, however. While he was out, our predictions were between 20 and 40 percentage points too optimistic for every game except for our lone win against the defensively challenged Cougs. Once Garbers returned (and stayed healthy) for a pair of rivalry games, our days of wildly overestimating the Bears came to a pleasant end.

Over the course of the season we had six instances in which the team underachieved (North Texas and every loss while Chase was injured), two games where they overachieved (UW, UCLA), and four games where they met expectations (UC Davis, Ole Miss, LSJU). Of course, we didn’t know in preseason that Chase would get injured and change the trajectory of the season, so let’s see how our pregame predictions compare to our preseason ones.

Preseason Predictions vs. Pregame Predictions vs. Reality

Typically our pregame predictions (gathered each week in the previous game’s Rating the Bears forms) are more volatile than our preseason predictions. Big wins like the UW win portend sunshine and Rose Bowls while terrible losses (OSU) make an eight-game losing streak to close the season seem entirely too possible.

Comparing our preseason and pregame predictions to the product on the field.

A funny pattern emerges here. After falling short of predictions against UC Davis, we lowered the Bears’ chances of winning the next game. After they then exceeded our preseason predictions against UW, we increased our predictions of a win over the next game. Over the course of the season, every time we exceed our preseason expectations we then give the Bears a higher chance to win the next game. Likewise, every time we failed to meet expectations, our pregame predictions then fell lower than the preseason predictions. Clearly we are a reactionary bunch. Not that there’s anything wrong with that. In fact, we should be praised for being receptive to new information and using it to update our assessments of the world. Go us!

Preseason Predictions vs. Pregame Predictions vs. Las Vegas’ Predictions vs. Reality

Finally, we add Las Vegas’ predictions to each game. I calculated these by using the point spread and the under/over to calculate Vegas’ prediction of the final score for each game.

Adding Vegas’ predictions to the mix...

Amusingly, compared to our preseason predictions Vegas gives us a lower chance of winning in EVERY SINGLE GAME. Only four times did Vegas predict a Cal victory (UC Davis, North Texas, ASU, and Oregon State). For several years now the oddsmakers in Vegas have consistently underestimated the Bears. Will that start to change now that the Bears have finished with a second consecutive winning season?

The table below displays all the above data so you can see the exact numbers across the various predictions (and reality).

Reality Preseason Pregame Vegas
UC Davis (5-7) 84.97% 94.16% 94.16% 82.04%
@ Washington (8-5) 53.04% 39.18% 25.74% 14.60%
North Texas (4-8) 67.18% 85.36% 93.31% 79.40%
@ Ole Miss (4-8) 68.94% 59.50% 53.52% 41.70%
ASU (8-5) 30.63% 63.51% 70.49% 60.88%
@ Oregon (12-2) 10.88% 33.30% 21.25% 9.05%
Oregon St. (5-7) 37.74% 83.51% 75.32% 72.91%
@ Utah (11-3) 0.00% 37.97% 12.70% 4.28%
Wazzu (6-7) 76.62% 58.29% 16.50% 30.39%
USC (8-5) 11.04% 54.32% 60.30% 40.23%
@ LSJU (4-8) 60.64% 56.56% 53.71% 47.18%
@ UCLA (4-8) 74.02% 55.62% 73.91% 47.68%
Projected Wins 5.76 7.21 6.51 5.30

Over the course of the season it seems like we watched three different teams: the middling offense and strong defense of the first three games, the ghastly offense and inconsistent defense while Garbers was injured, and the surprisingly effective offense complementing a decent defense during the Ole Miss, LSJU and UCLA games. Likewise, we have three different ways of interpreting the events of the season: an underachievement after the strong start, an up-and-down season that averaged out to meet our expectations, or a season where a number of narrow wins helped the Bears luckily avoid an injury-driven disaster. I am not sure which interpretation I subscribe to, but I do know this: 2020 is looking good for the Bears. Sunshine pumps eternal.


Which narrative best characterizes the 2019 Cal football season?

This poll is closed

  • 35%
    The Bears had a chance to achieve an 8- or 9-win season after the 4-0 start and ended up underachieving.
    (56 votes)
  • 47%
    7 projected wins. 7 actual wins. The Bears met expectations.
    (76 votes)
  • 16%
    Based on the number of one-score wins and the non-competitive mid-season slump, the Bears overachieved by earning 7 wins.
    (27 votes)
159 votes total Vote Now