clock menu more-arrow no yes mobile

Filed under:

Why Do Computers Hate the Huskies?

Or how I learned to stop worrying and love the algorithm

NCAA Football: Alamo Bowl-Texas at Washington Kirby Lee-USA TODAY Sports

It has been a long time since the Washington Huskies entered a season with this level of expectations. The national pundits have largely been leading the charge. A slew of writers have had Washington in the top-10 with oft-accused Husky hater Jon Wilner putting the Dawgs at #4 overall. That has led many fans to start dreaming about a trip to the College Football Playoff.

But those are humans. And an analytically-minded guy such as myself knows that humans are subject to bias. So what if we check out some of those good ol’ computer rankings. Surely those must be even better since they aren’t subject to #EastCoastBias.

Bill Connelly’s SP+: 17th

Kelley Ford: 19th

ESPN’s FPI: 21st

What the hell?!? We’re the home of Microsoft and Amazon. If anything, computers should love the team based out of Seattle. But apparently they don’t. So let’s try to figure out why that’s the case.

It becomes a much easier question when you just look at where things finished. The Huskies ended the season 8th in both the AP and Coaches Polls after an 11-2 season. Every team that lost 2 or fewer games except for Troy ended up in the top-9 of those polls while no team that lost at least 3 games penetrated that group. Win-Loss record is still very important to that grouping.

That wasn’t the case in these same computer-based metrics. SP+ had Washington 15th in the final ranking set of the 2022 season. FPI had the Dawgs 20th which was up 3 spots from before the bowl game. The simple answer for why Washington ranks so much lower now with the computers is that they started out so much lower than the general human consensus at the end of last year.

So that really divides our initial question into separate questions. “Why didn’t the computers like Washington in 2022?” and “Why don’t the computers expect improvement coming off of 2022?”

The first thing you always want to check when trying to figure out discrepancies between resume-based metrics and predictive metrics is the performance in close games. That varies depending on your definition. The Huskies were 5-1 in games decided by 7 points or fewer. If you choose to go count 8 points as still being within one possession then UW was 5-2. And if you go down to games decided by a field goal or less then the Dawgs were 2-0.

No matter how you slice it, the Huskies won more close games than probably would be expected for the average team. Bill Connelly’s Postgame Win Expectancy charts are instructive to explain the concept. A team could completely dominate the line of scrimmage but give up a pair of 99-yard scoop and scores and lose a game. That counts as a loss on the schedule but it’s probably not realistic to expect them to fumble it on the goal line in the future and probably not realistic to expect their opponent to come up with such huge swings in the future either. Looking at some of the per-play metrics helps to figure out when a team might be getting the dreaded “L” word: Lucky.

Washington had 6 games last year with a postgame win expectancy of 98.9% or higher (Kent St, Portland St, Michigan St, Stanford, Colorado, and Wash St). Those were all unsurprisingly all double-digit victories.

The Huskies played 3 more games with win expectancies somewhere between 72% and 97% which included a pair of their close wins (California, Oregon, and Arizona). Perhaps surprisingly, the Arizona game with its 49-39 final had a lower postgame win expectancy than the Oregon game at 37-34. Arizona was much closer in yards per play than Oregon and the score differential was closer in Eugene because of UW’s red zone struggles whereas it was the Wildcats who kicked too many field goals against UW in Seattle.

Next, we find 3 games where the postgame win expectancy was between 25-50% with Texas (45.9%), Oregon State (27.0%), and Arizona State (26.6%). Those are contests that UW probably should’ve lost but it’s reasonable to expect the Huskies to go 1-2. Instead, Washington went 2-1 thanks in part to Jonathan Smith deciding not to kick field goals in a giant windstorm (which a computer doesn’t know about) and failing to convert a pair of 4th downs in the red zone. The only game where Washington truly got its ass kicked was at UCLA (3.9%) and it took a frantic Husky comeback in psuedo-garbage time to get the game within 8 points.

When we put it all together, Washington’s expected win total was 9.6 compared to an actual total of 11. The +1.4 differential was the 9th highest in FBS last season. You would rather your team have a higher differential because it probably means they won a lot of games but it does make you a little more susceptible to a regression to the mean.

A lot of Husky fans would shrug that off when looking ahead to 2023 because of all of the returning talent that Washington is bringing back. The Husky offense has the national leader in passing yards returning along with the team leader in rushing TDs, a pair of all-conference offensive tackles, his leading 3 receivers, and his entire tight end corps. Plus to top it all off the offensive coordinator turned down Alabama to stay and coach this group for at least one more season.

That’s obviously great but it does gloss over the fact that Washington is losing its leading rusher in Wayne Taulapapa plus 3 starting offensive linemen and replacing them with some question marks. We’ve seen Matteo Mele, Nate Kalepo, and Julius Buelow at times and none have been able to secure starting spots in the past. It’s possible to project improvement at those spots but with the extra uncertainty it’s also reasonable for a computer to knock down Washington’s ceiling a bit due to only a combined 38 OL starts (Fautanu- 16, Rosengarten- 13, Buelow- 5, Kalepo- 3, Mele- 1). It’s hard to fault guys for failing to beat out Jaxson Kirkland and Henry Bainivalu in the past but that also makes it hard to think they’ll be better than that pair as well.

On the defensive end we also see 4 key contributors moving on. It’s reasonable to think that most Husky fans would only list Edge Jeremiah Martin as a huge loss given that he was 1st team all-conference last year and had a fantastic season as a pass rusher. S Alex Cook made 2nd team all-conference and LB Cam Bright plus CB Jordan Perryman were starters but UW has reason to think they’ve upgraded at all 3 spots this offseason. However, Cook was 1st on last year’s team in tackles and Bright was 3rd so the computers view the lack of returning tackles as a bigger issue than the fans who know that Edefuan Ulofoshio should be fully back from injury to replace Bright for example.

Some teams have just imported completely known quantities from the transfer portal to fill holes made by graduation or early draft entries. Washington made one such move with CB Jabbar Muhammad coming over after starting at Oklahoma State. Bringing in another P5 starter provides a theoretically high baseline for future performance even if it didn’t work out quite as well for UW last year when Cam Bright transferred to Washington.

Some of UW’s adds aren’t quite that simple. Running backs Dillon Johnson and Daniyel Ngata played very well when in the game but didn’t get starters’ reps. Other adds like Edge Zach Durfee and TE Josh Cuevas are moving up a level so it’s unknown how well their skills translate.

And on top of it all is the influence of Washington’s decreased recruiting. Not every system uses recruiting rankings but the Huskies have clearly dropped off over the last 5 years as we get further away from the peak of the Chris Petersen era which means the blue-chip ratio for the Huskies has gone down. The 2016 Huskies were at a similar level so it’s possible to be a top-5 team with this level of talent but it’s yet another red flag for some projections.

If you wanted to build the case against UW from a completely analytical standpoint you could sum it up like this. Washington overachieved last year and relied on some good fortune to end up with an 11-2 record. They played more like a good but not exceptional 9-10 win team with a great offense and an at best meh defense. That team lost 8 starters to graduation and is relying on internal improvement to replace at least 6 of those spots from a roster that finished 30th, 96th, and 25th in recruiting rankings over the past 3 seasons. Not exactly a top-five group.

Are there counterarguments for most if not all of those points that optimistic Husky fans would love to employ? Sure. But while you’re better able to argue with a computer now than at any other point in recorded human history, good luck doing so with a CFB ranking algorithm.