There are 31 venues used as the home stadium for NFL teams. Eight of these stadiums are either fixed domes or have retractable roofs. The other 23 are open year-round, exposed to the natural elements. As a result, we’ve seen NFL games played in a variety of weather conditions. We’ve had the Ice Bowl, the Snow Bowl, the Freezer Bowl, the Fog Bowl, the Snowplow Game, and the Monsoon Bowl. The players fight through it all. The NFL could go ahead and adopt the “neither snow nor rain nor heatâ€¦” motto of the US Postal Service.

However, these extreme conditions must have some impact on the performances of the players. I used this dataset of weather data and this dataset of player game logs in order to analyze the impact of weather on player performances (if there even is one).

I’m primarily focusing on wind in this article because the sample size of rainy / snowy games in the NFL frankly isn’t all too big, and many of the ones that *did *occur are in the postseason where statistical trends vary. There will be a few mentions of precipitation, but most of the attention will be on wind.

First, here’s a histogram of all the average wind speeds recorded at games since 1985:

The mean of all average wind speeds recorded at NFL games since 1985 is 6.95 mph, while the mode is 0 mph. The windiest game in the dataset was this low-scoring 2011 outing between the St. Louis Rams and the Cleveland Browns. The wind was recorded at a whopping speed of 40 mph in Cleveland.

Now, let’s look at how wind speed impacts quarterback performance in the NFL. Here are the average passing yards, passing yards per attempt, passer rating, and completion percentage for various wind speed ranges.

There is a noticeable downward trend in all four passing stats as the average wind speed increases. When the wind speed is recorded at less than 10 mph, NFL quarterbacks complete 60.31% of their passes on average. This figure drops to 54.65% when wind speeds are at least 20 mph. The average touchdown percentage drops from 4.29% to 3.58%, while interception percentage increases from 2.99% to 3.11%.

I also calculated adjusted net yards per attempt (ANY/A) for different wind speeds because it’s the most popular all-in-one individual passing metric.

Once again, there’s a clear downward trend. Quarterbacks enjoy an average ANY/A of 5.79 when wind speeds are less than 10 mph. When wind speeds are over 20 mph, the average ANY/A drops to 4.62.

I used an ANOVA test to test for a statistically significant difference in ANY/A between games with wind speeds of 0-10, 10-15, 15-20, and 20+ mph. The resulting p-value was less than 0.01, which allows us to reject the null hypothesis. As one would expect, wind has a negative impact on the efficiency of the passing game in the NFL.

What about placekicking? Strong wind gusts should make it harder to accurately kick a football, right? Let’s see if the stats back it up.

Right off the bat, I can just look at the average field goal percentage as wind speeds increase. The NFL’s average field goal percentage is 83.8% when wind speeds are slower than 10 mph. When wind gusts exceed 20 mph, the average field goal percentage drops to 76.9%. It’s an open-and-shut case, right?

Well, it would be a bit silly to disregard the fact that field goal percentage is directly correlated with average field goal distance. The violin plot shows that there is a clear shift in the distribution of field goals when wind speeds are over 20 mph. If you’re playing in extremely windy conditions, you’re probably gonna be a little more reluctant to send your kicker out there to boot a 60-yarder. This study would be incomplete if we didn’t account for this (an idea which I played around with about a year ago).

The blue points represent the average league-wide field goal percentage at a certain distance. As you can see, there’s a clear trend. Short field goals are easier than long field goals. Who knew?

I fit a logarithmic function according to this trend, which can be used to calculate *expected *field goal percentage (xFG%). The idea is simple: if you attempt four 20-yard field goals and hit two of them, that’s a 50% rate. If another person attempted four 60-yard field goals and hit two of them, they also had a 50% success rate. However, according to the model, the probability of nailing a 20-yard field goal is 98% versus just 35.8% for a 60-yarder. Without a way to quantify this through xFG%, though, we’d be stuck just looking at the FG% without any contextualization.

Now, I can apply this idea of xFG% to our wind question. The NFL’s average expected field goal percentage is 88.8% when wind speeds are over 20 mph. When the speeds are less than 10 mph, the xFG% drops to 83.5%. Let’s compare these expected percentages to the actual success rates:

If wind had no impact on kicking, all of those points would be close to the line (y=x), because the expected field goal percentage would be almost equal to the actual field goal percentage. There is obviously some margin for error because the model used to calculate xFG% is not very complex. However, we can clearly see that strong wind gusts do have a significant impact. on placekicking in the NFL. When wind speeds are below 10 mph, the actual FG% is actually a hair above the expected FG%. When wind speeds are over 20 mph, though, there is a gap of 11.86% between xFG% and FG%.

The same graph can be made with different forms of precipitation instead of wind speeds:

The results are quite similar. Rain does not seem to have a huge effect on placekicking but there *is *an effect. Meanwhile, snow has a massive effect.

On a side note, it’s worth mentioning that the average wind speed in NFL games is higher when there is precipitation in the form of rain or snow, which could be impacting this data as well. The difference is only about 2 mph, though, so I don’t think the impact would be too severe.

Finally, there’s one last thing I want to take a look at: kickoffs and punts. We’ve already covered two of the four ways the ball travels through the air, which is when Mother Nature would have the best opportunity to impact a football game. Might as well tackle the other two ways.

The goal on a kickoff is to boot it far for a touchback (at least before the recent rule change, but our weather data set covers 2016 at the earliest anyway). So, does wind speed and precipitation affect the touchback rate on kickoffs?

Yes, it certainly does. That’s a drastic difference. Over 40% of kickoffs are touchbacks when there is no precipitation. This rate drops to less than 25% if it is snowing.

What about punts? On punts, you *don’t *want a touchback. The goal is to pin your opponent back as close to the end zone as possible without it being a touchback. Punting is more about finesse and accuracy than sheer distance like a kickoff, so it should be interesting to see what impact weather has on it.

Unfortunately, all of these graphs don’t perfectly follow the expected trends like everything else on this article so far. But the results definitely aren’t *entirely* unexpected. In the presence of precipitation, less punts are downed in the 20 and there are more touchbacks — both trends that are obviously negative on punts. The average distance of punts (that are not touchbacks, to be clear) decreases when it is raining or snowing, as you would expect. The same negative trend in average distance exists when wind speeds increase.

It gets a little wonky when you look at the ‘touchback rate’ and ‘punt inside 20 rate’ graphs for wind speed. I’m not sure why there’s a spike in the percentage of punts downed within the opponent’s 20-yard-line when wind speeds are over 20 mph. I’m also not sure why there’s a spike in the touchback rate of punts when wind speeds are between 10 and 15 mph.

Overall, the ANOVA test outputted a p-value below 0.05 when the the variables from three of the six graphs were inputted. The trend in both ‘touchback rate’ graphs were not statistically significant, as with the ‘punt inside 20 rate’ graph for wind speed. It certainly seems like precipitation and strong wind gusts impact the distance punts travel. Basically everything else related to punts is up in the air.

Anyway, I think this is a good stopping point. There’s plenty of more analysis that could be done, but this was enough for a single article. Some of the ideas from here could be applied to future projects, like evaluating NFL kickers in a way that takes into account weather. Maybe that’ll be the topic of the next article.