Why couldn't they have just played in Buffalo?

Week 11 preview + talking about the weather again

Nov 20, 2022

As we get closer to Thanksgiving and into the second half of the NFL season, we are finally getting our first snow bowl of the year. Or, we were supposed to, at least:

A lot of clowns in the replies to that tweet missed the point - the game isn’t being moved because of the blizzard’s effect on the players, it is being moved to Detroit to protect the fan experience. With six feet of snow projected to hit upstate New York this weekend, roads will be hazardous, and with fans driving in from as far away as Rochester, the Bills (rightly) decided to not put their fans at risk. Luckily, the Lions allowed the game to take place at Ford Field:

Detroit plays in a domed stadium, so it should be around 72 degrees with no wind and a few feet less of snow than the game would’ve had originally. Aside from public safety concerns, how much of an impact does snow have on the on-field product?

Snowiest games (2010-present)

As with the ElasticNet model used previously, our dataset is limited to only games played from the 2010 season to now. Additionally, the hourly weather data from the meteostat Python package has shown to be somewhat unreliable at times, so we use daily weather data instead. When we sort games by those which were played in under 32 degree conditions by total daily precipitation, here are our top candidates for snow bowls:

Unsurprisingly, the most frequent home teams in this list are Green Bay and Buffalo.

For whatever reason, Pittsburgh seems to have to play in a lot of these miserable games as a visiting team even though they don’t share a division with Buffalo or Green Bay. I don’t think they mind, however, as they are 3-0 in those games (and 7-2 in snow games in general over this span).

The earliest snowy game during that time period was played on October 25th, 2020- a Week 7 matchup where the Broncos hosted the Chiefs.

Precipitation vs. scoring

I want to attempt a new graph here which might not be very readable:

Here, we can see the impact that precipitation has on scoring at a range of temperatures. Below freezing, the precipitation will tend to be snow, above freezing it will tend to be rain. At or below 30 degrees, snowing or not, total scores tend to not eclipse 40 points (the average is 45 in the entire dataset). Games played during some sort of precipitation also tend to be lower scoring than average, regardless of temperature.

The outlier, with 78.2 inches of rain, is Colts @ Panthers in 2015, which ended up being a muddy 55-point shootout, OT thriller, and Panthers win.

Adding precipitation to NFL models

Now that we’ve established how rain and snow impact scoring totals, we can add precipitation as a variable to our various models.

Total points

Once we re-fit the total point model, here are our weather coefficients:

The directions all still make sense - higher temperatures, lower wind, and lower precipitation make for higher scoring games. Thus, it follows that a blizzard game, like we would’ve had for the Bills-Browns showdown, would’ve had negative temperature, wind, and precipitation impacts, all else equal. Additionally, adding this new variable drops model error from 10.840 points to 10.827 points. This still isn’t quite as low as Vegas’ 10.61, but every little bit counts.

Spread

The additional weather point doesn’t make a difference in this model, it ends up still being 10.9 points off on average. It is interesting to see how the home team benefits from colder weather and more precipitation, but is hurt by more windy conditions. This definitely requires more investigation. One hypothesis I have is that some cold weather teams have been very good over the past 12 years (Patriots, Packers, Steelers all come to mind), and would score a lot of points despite this weather. Similar to in the head coaching change analysis, we’d have to look at how each team specifically has performed in various weather conditions to remove this bias.

In any case, the impact of weather variables in the spread model is small, even when scaling the coefficients based on their order of magnitude. Temperature, wind, and precipitation are the 19th, 32nd, and 52nd most important variables in the spread model, respectively.

For total points however, wind is the most impactful variable, while precipitation is 8th and temperature is 22nd. So, while weather hardly has any impact on the winner of NFL games, it has a strong impact on total points scored.

Week 11 ElasticNet picks

1:00 PM games

Chicago at Atlanta:

UPSET ALERT!! Chicago ML, Under 48.5 total points

Justin Fields is finding his groove, and Atlanta will have to get off to a hot start if they want to hold him off

Carolina at Baltimore:

Carolina +13.5

If PJ Walker was starting, Carolina would be favored.

Buffalo vs. Cleveland (@ Detroit):

Cleveland +6.5, Under 50.0 total points

The Bills lived out the movie Planes, Trains and Automobiles on their way out of snowy Buffalo to the game in Detroit. They’ll win, but it’ll be closer than people think.

Washington at Houston:

Washington ML, Over 41.5 total points

Enough said.

Philadelphia at Indianapolis:

No strong pick, but maybe Indianapolis +6.5

Will the Colts go 2-0 in the Jeff Saturday era? Can the Eagles bounce back from their embarrassing loss to the surging Commanders? I’m not at liberty to say, but I’d throw a small unit on the Colts to cover here.

New York Jets at New England:

NY Jets +3.5

Matthew Judon looks to get over the 12 sack mark on the season and solidify his DPOY candidacy. Some New England struggles on offense will keep this close, though.

LA Rams at New Orleans:

New Orleans is on the no-bet list but LA Rams +2.5 if you really want to

I will not be donating any more money to the sportsbooks on the Saints’ behalf until Jabo is the starting quarterback again.

Detroit at New York Giants:

Detroit + 2.5, Under 45 total points

Probably a strong candidate for mid-off of the week if not for the next game..

4:00 pm

Las Vegas at Denver:

Denver is on the no-bet list until Subway brings back the DangerWich (Denver -2.5)

Inverse AB strategy in full effect

Dallas at Minnesota:

UPSET ALERT - Minnesota ML

I don’t really have a statistical method for proving it, but this catch was miles above the OBJ catch from a few years ago. I have no idea why Dallas is favored in this game, it is free money.