So… none of my kicker models have worked out very well. Is it possible that kickers are just random? Should I just quit? No I say! I will not give in!

I’ve tried linear models and generalized linear models combining various terms (predicted scores, which teacms are playing, the day of the week), but so far nothing has been consistently better than guessing. I have a few mode ideas to try, and I wanted to start playing around with one of them today. Specifically, the weather.

I downloaded the weather from NFLWeather.com which includes the temperature, wind speed and direction, and weather conditions (cloudy, rainy, clear, etc.) for every game in 2015 and 2016. For dome games, I said that the temperature was 67 degrees F (found that number on a blog somewhere) and the wind speed was 0. Let’s explore the data a little:

# Effect of weather on total score

It’s pretty clear that most games don’t have a lot of interference from the weather with respect to rain and snow. The conditions I would consider bad (rain, snow, fog) happen infrequently when compared to clear/cloudy conditions (which I doubt have much impact). I ordered them (roughly) with the most disruptive conditions toward the top.

That doesn’t look like much is there, but what does ANOVA say:

## Analysis of Variance Table ## ## Response: totalscore ##Df Sum Sq Mean Sq F valuePr(>F) ## Conditions 13 8772674.793.87 3.833e-06 *** ## Residuals1010 176108174.36 ## --- ## Signif. codes:0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

That nice and low p-value is saying that there is at least one group that differes significantly from the mean. Looking at it, it almost looks like you get slightly higher scores when it’s raining and lower scores when drizzling. Not sure what this means, though. We’ll see later if this has any effect on fantasy points.

Maybe some very weak dependence on temperature

Hey, that’s something! The total score decreases by -0.5112359 points per mph of wind. It’s not a huge effect, but maybe it has some effect on kickers and fantasy points. Let’s start to dig into that, shall we?

# Effect of weather on kicker fantasy points

From this trio of figures it looks like kickers score fewer points in adverse conditions (snow, rain, and drizzle), while wind and temperature have weak effects.

# Adding weather to kicker models

In a previous article (Which I would like to, but I haven’t actually posted yet), I showed that my model from last year, Yahoo’s predictions, a GLM with 4 terms, and randomly choosing a kicker every week gave equivalent results. The only model that stood out at all was a slightly reduced GLM with just 3 terms: predicted score, predicted opponent score, and which team (kicker) you were modeling. Behind the scenes I tried adding Temperature, Wind Speed, and Weather Conditions to that model, and the only one with significant results was Weather Condition. Let’s see what would have happened last year had I used that model all year.

# Conclusions

Weather condition is a statistically significant term to add to the model (p = 5.009137310^{-4}), but it wouldn’t have produced results significantly different from the model with only Score, Opponent score, and Team playing in 2016. It’s probably worth trying in 2017 since the weather data from 2016 was better than the weather data from 2015 (more nuance in the forecast). Still, disheartening that none of these are (statistically significantly) better than just guessing a random kicker each week.

Since I can’t beat random, that means that my models are telling me that I’ve already devoted too much time to this. I’ll get back to QBs and Defences with my next work.