Debunking the weather myth: Let it freeze, rain, let it snow (2 Viewers)

[Hipple Long Ware & Peete]

Apparently it's a monsoon in Jax right now. This will spawn an inevitable onslaught of 'i'm benching _____ b/c of weather threads tomorrow am and cause many owners to fret and panic. But I've often thought this was a bit of an overreaction, as the defenses have to play in the same ####ty weather, and anyone who has played in a monsoon knows that knowing where you are going is a huge advantage when it comes to keeping your footing. How many people benched Brady/Pats Wr's last year in a howling 30mph snowstorm only to watch Brady throw for 300 and 4 tds in the first half?However when I searched the annals of the SP I see pages upon pages of 'week __ weather' threads, but nary a thread devoted to whether (all puns intended) the weather matters?Doing a bit of internet scouring I found this (and I think i have posted this before, but prolly buried in some weeks weather thread) and I think it would make for some good discussion, and possible some of the uber 's here can run updated algorithms and #### to see if it still hold true.But according to this FF , rain, snow, bitter cold etc all have negligible effect. Even 'windy' conditions aren't so bad. As long as it's not incredibly gusty.

How Does Weather Affect Fantasy Football Performance: Part 1In 2007, Tom Brady was taking the fantasy football world by storm, averaging almost 30 points per game and making any team that owned him coast into the playoffs. But then something happened on the way to fantasy football immortality. In week 15, which is the playoffs in most fantasy football leagues, Brady put up his only bad game of the season, throwing for only 140 yards and no touchdowns. What happened? And, more importantly, could Brady owners have predicted this poor performance? The answer to the first question is simple, it was the weather. The answer to the second question is what we'll use Minitab Statistical Software and Stepwise Regression to figure out.So what were the conditions like in Foxboro the day that Brady had his terrible game? According to Weather Underground the temperature was 39 degrees Fahrenheit, it was raining, the wind speed was 12 miles per hour, and there were wind gusts of 23 miles per hour. So which condition was it that ruined Brady's day? Was it the temperature, the rain, the wind, or all of them?We collected weather conditions on every single NFL game in 2009. The variables include temperature, dew point, humidity, visibility, wind speed, gust speed, and the conditions (rain, snow, fog, or clear). We did not include games that were played in a dome or in stadiums that have a retractable roof. We then compiled the fantasy scores for every quarterback who played in those games. We adjusted each quarterback's score for the defense of the opposing team to make sure when a player's score was low it's because of the weather conditions, not because they played a good defense. Now we're ready to use Minitab's Stepwise Regression to see which weather variables affect player performance.We use Stepwise Regression because we have many variables, and we want a fast way to model the relationship between weather and player performance. Stepwise Regression adds or removes one variable at a time to the model based on statistical significance. The variables that remain in the final model are the ones that significantly affect the player's fantasy score.



The model that Stepwise Regression provided included only "Gust Speed" because there were no other statistically significant variables. This means that temperature, wind speed, and weather conditions don't have any significant effect on a quarterback's fantasy football score. To better understand this, look at the Fitted Line Plot of a quarterback's adjusted score and the temperature.



If temperature affected a quarterback's performance, lower temperatures would result in lower scores. But the plot shows that there is no relationship between temperature and scores, and that in 2009 the average fantasy score was slightly higher in lower temperatures.

If we made similar plots for the other variables, we would see similar results. So when Brady had his subpar game in 2007, we know it wasn't because of the temperature, the wind speed, or the rain. It was the gust speed.

So what is gust speed, and how is it different than wind speed? Wind speed is, of course, how fast the wind blows. A gust speed is a wind that blows 18 mph or higher and varies by more than 10 mph. This means that quarterbacks can play well in a steady wind, even if it's strong. But when the wind speed is constantly gusting from high speeds to low speeds, they don't perform as well. Stepwise Regression gives us a coefficient of-0.127, which means the quarterback's fantasy score drops 0.127 points for every mile per hour of gust speed. And when you include only quarterbacks that you would regularly start in fantasy football (players averaging 15 or more points per game), that number nearly doubles to -0.225. This means on a day with wind gusts of 25 miles per hour, the quarterback will perform almost 6 points worse than their average (25*-0.225 = -5.625). That's one less touchdown, and one more interception.

So wind gusts aren't the entire reason for Tom Brady's poor performance in 2007, but they definitely played a large role. Keep this in mind when you look at weather forecasts Sunday morning, especially if you own two equal quarterbacks. If one of them will play in the cold, rain, or snow, don't worry. But, if the wind is howling at the stadium one of them plays in, you should start the other.

food for thought before you go benching Aj Green tomorrow.

 

[Hipple Long Ware & Peete]

Very interesting

Two weeks ago, we used Minitab Statistical Software's Stepwise Regression to determine that gust speed significantly affects a quarterback's fantasy football performance. This week we'll use Stepwise Regression again to see if weather affects running backs or kickers. Our initial analysis leaves us with an unusual conclusion, so we'll also show how to use General Regression to uncover the lurking variable that was affecting our analysis.

Will a running back's score go up in cold weather because he is getting more carries? Does a kicker's performance suffer when he has to kick a cold ball? Let's go to Minitab to find out.

Running Backs

The detailed setup for this week's analysis is available in the week 11 article, How Weather Affects Fantasy Football Performance: Part 1.





The model that Stepwise Regression provided included only "Humidity" because there were no other statistically significant variables. The coefficient is 0.025, which means that, as humidity goes up, fantasy scores go up. But this sounds a little odd. You'd think that a running back would be less effective in very high humidity. Before we conclude that a running back's fantasy score increases as the humidity goes up, we should look for a lurking variable.A lurking variable is a variable that is not included in the analysis but may affect the interpretation of relationships among variables. Our analysis included many weather variables, but nothing else. When we looked at the data, we saw that Chris Johnson had 3 of the top 9 scores from last year. Chris Johnson also plays in Tennessee. So does higher humidity really result in high scores, or was humidity significant because the best fantasy running backs in 2009 played in humid cities? To find out, we need to add "Player Name" as a categorical variable to our model, and see if humidity is still significant. We can use Minitab's General Regression to do this.





We now see that once "Player Name" was added to the model, humidity has a p-value of 0.36 which is no longer significant. The reason that humidity appeared to be significant was because the better running backs just happened to be playing in humid cities.

So we learned that you need to pay attention to the possibility of a lurking variable, especially if your conclusion seems odd. And because we found a lurking variable, we know that weather doesn't significantly impact the fantasy football performance of running backs. You should start running backs who have performed well all season and sit the ones who haven't, regardless of the weather.

KickersKicker is usually a hard position to predict in fantasy football, but what if we knew how weather affects kickers? Right now David Akers is leading all kickers in fantasy points. But will his scores decrease as the weather turns cold in Philadelphia? We'll use Stepwise Regression again to find out.





You can see from the statement in the output "No variables entered or removed", that Minitab did not enter or remove any variables from the model. This means that none of our weather variables affected kicker fantasy scores. This may seem odd, as we know that it's harder to kick a football in cold and windy weather. But this analysis shows us that a kicker's score depends more on other factors, such as their team's offense.For example, look at the game earlier this season where the San Francisco 49ers were shut out against Tampa Bay. 49ers kicker Shane Andrus scored 0 points that day because he never even attempted a kick. His offense never put him in position to even make a field goal, so the nice weather in San Francisco didn't even matter. And then there was the Eagles-Bears game last Sunday where temperatures were in the high 30s with wind speeds and gust speeds over 20 mph. The kickers from both teams were 5/5 in field goals and 6/6 on extra points despite the weather. Even if either kicker would have missed a field goal or two because of the weather, they would have still had good fantasy scores because their offenses put them in position to score.

So our analysis shows us that, as the season goes along, David Akers' fantasy scores depend more on Michael Vick moving the football up and down the field than on the weather conditions. And if the Eagles quarterback continues to play the way he has all season, Akers will be just fine.

Speaking of Michael Vick moving the football up and down the field, let's get to our week 13 projections, where we think the Eagles quarterback is going to have a big day.

 

[Trizzler]

Great posting. It's funny how many times I see sharks acting like guppies in here; c'mon guys. A smart fantasy player relies on proven trends and statistical analysis, not wives tales.

Now, it is true that in the Panthers/Jags game, both QBs struggled mightily. But I think that had more to do with Gabbert playing in his first NFL game ever and Newton playing overconfident early and then utilizing a conservative gameplan to win the game late.

There is no definitive reason for anyone to bench Gresham, AJ Green, or Mike Thomas, assuming you do not have better options. Would I trust Dalton or Gabbert as fantasy starters this week? Probably not. But if I'm desperate for a bye-week QB replacement and I have to choose between Dalton in the rain or McNabb in a dome...that's a much harder call.

 

[Nick Nasty]

Interesting read. It's hard to not think a mOnsoon would affect performances though

 

[Hipple Long Ware & Peete]

Interesting read. It's hard to not think a mOnsoon would affect performances though

Not as much as you would think, as if it were truly a monsoon I imagine they would cancel the game. I would also wonder why they were playing a football game in Southeast Asia in the first place...

 

[davidwb]

I've never seen the term "lurking variable" before -- statisticians usually call it "omitted variables" instead.

 

[griff321]

Thanks Hipple. I'm so sick of the "well they'll have to run it once the weather gets bad" BS.

 

[Short Corner]

I just skimmed over it but it looks like you are using studies involving temperature and wind?

What does that have to do with the monsoon in Jacksonville, where the primary concern will be precipitation?

 

[Sabertooth]

 

[Hipple Long Ware & Peete]

Bump for the crazy weather this weekend and because this thread reminded me of this issue...

 

[Hipple Long Ware & Peete]

speaking of does anyone know what the wind is like at the meadowlands?

 

[Diamond]

Why is James Harrison tweeting about the fog in Pittsburgh?

No link, sorry, saw it scroll by on the Fox pregame.

Game is at 4pm, fog supposed to lift by 2pm according to weather.com.

If there is any weather besides wind that matters, it's fog. Kinda hard to catch what you can't see.

 

[Hipple Long Ware & Peete]

Bump for the impending inclement weather that will occur some places this weekend.

Basic conclusion for the lazy



So what is gust speed, and how is it different than wind speed? Wind speed is, of course, how fast the wind blows. A gust speed is a wind that blows 18 mph or higher and varies by more than 10 mph. This means that quarterbacks can play well in a

steadywind, even if it's strong. But when the wind speed is constantly gusting from high speeds to low speeds, they don't perform as well. Stepwise Regression gives us a coefficient of

<br style="color: rgb(55, 58, 62); font-family: Verdana, Arial, Helvetica, sans-serif; font-size: 13px; line-height: 19px; text-align: left; background-color: rgb(243, 249, 246); ">-0.127, which means the quarterback's fantasy score drops 0.127 points for every mile per hour of gust speed. And when you include only quarterbacks that you would regularly start in fantasy football (players averaging 15 or more points per game), that number nearly doubles to -0.225. This means on a day with wind gusts of 25 miles per hour, the quarterback will perform almost 6 points worse than their average (25*-0.225 = -5.625). That's one less touchdown, and one more interception.

 

[Shirtless]

Hate to make this more nerdy, but the R-squared values in these models are pretty low. R-squared is supposed captures how much variation can be explained by a regression model. In the above cases, it's 1.1% for the wind gust model, and 0.1% for the temperature model.

In other words, almost 99% of a quarterback's fantasy point production is explained by something other than the variables in the wind gust model, i.e. omitted variables. In the temperature model, it's almost 100%.

I should add that R-squared is not an end the conversation statistical value, more like a conversation piece. It can be be manipulated to suit a researcher's purposes, but it usually acts as a good place to begin a discussion.

 

[otb_lifer]

Hate to make this more nerdy, but the R-squared values in these models are pretty low. R-squared is supposed captures how much variation can be explained by a regression model. In the above cases, it's 1.1% for the wind gust model, and 0.1% for the temperature model. In other words, almost 99% of a quarterback's fantasy point production is explained by something other than the variables in the wind gust model, i.e. omitted variables. In the temperature model, it's almost 100%. I should add that R-squared is not an end the conversation statistical value, more like a conversation piece. It can be be manipulated to suit a researcher's purposes, but it usually acts as a good place to begin a discussion.

Pretty damn nerdy for a shirtless dude ... whodathunkit

 

[AB in DC]

Bump for the impending inclement weather that will occur some places this weekend.

http://www.accuweath...y-nfl-gam/42491

Umm....that article was from December 5, 2010.

 

[Hipple Long Ware & Peete]

Hate to make this more nerdy, but the R-squared values in these models are pretty low. R-squared is supposed captures how much variation can be explained by a regression model. In the above cases, it's 1.1% for the wind gust model, and 0.1% for the temperature model.

In other words, almost 99% of a quarterback's fantasy point production is explained by something other than the variables in the wind gust model, i.e. omitted variables. In the temperature model, it's almost 100%.

I should add that R-squared is not an end the conversation statistical value, more like a conversation piece. It can be be manipulated to suit a researcher's purposes, but it usually acts as a good place to begin a discussion.

Pretty damn nerdy for a shirtless dude ... whodathunkit

he showed his work. care to recrunch the #'s with what you feel is a more appropriate R2?

 

[Shirtless]

Hate to make this more nerdy, but the R-squared values in these models are pretty low. R-squared is supposed captures how much variation can be explained by a regression model. In the above cases, it's 1.1% for the wind gust model, and 0.1% for the temperature model.

In other words, almost 99% of a quarterback's fantasy point production is explained by something other than the variables in the wind gust model, i.e. omitted variables. In the temperature model, it's almost 100%.

I should add that R-squared is not an end the conversation statistical value, more like a conversation piece. It can be be manipulated to suit a researcher's purposes, but it usually acts as a good place to begin a discussion.

Pretty damn nerdy for a shirtless dude ... whodathunkit

he showed his work. care to recrunch the #'s with what you feel is a more appropriate R2?

Can't. I would need his dataset. Plus you can't produce a different R-squared just by inputting the same variables into a model. You would need other variables, meaning I would have to append the dataset with other stuff. And judging by the size of the R-squared, it would have to be a lot of other stuff. I'm not saying what the guy produced is wrong. In fact, it's probably right, just incomplete, as all things like this have to be. If I had to interpret his findings, I'd say it shows that wind gust matters more than the other weather variables he included in his model with respect to affecting QB ff points, but that the weather variables in his model are much less important than other omitted variables.

 

[Shirtless]

OK, spoke too soon, methodology might not be too great either. Without knowing what he did exactly, it looks like the wind gust model he's offering only includes the wind gust variable, depending upon how his software did the stepwise regression. A one variable regression has a greater chance of achieving statistical significance than multi-variate regressions, but at the expense of valid interpretations.

Here's an example. I work in educational research. I could run a regression with the dependent variable being student test scores, and the single independent variable being, let's say, race of student. Almost guaranteed I would get statistical significance. However, that's not very sound research. My boss would come back at me and say to add more variables, like socioeconomic status. With two independent variables (race and socioeconomic status), two things would happen: 1) the significance of the race variable would decrease and 2) the R-squared value of the entire model would increase. This is because part of the significance of race is explained by socioeconomic status, and having the two variables together in the model explains more variation in student test scores than just race by itself (bigger R-squared). The goal, so to speak, is to get a model with as many variables as possible, and still end up with at least one that is statistically significant. That narrows the field more by 1) eliminating insignificant variables and 2) increasing R-squared to decrease the chance of omitted variables being a factor.

My guess is the statistically significant model only has the wind gust variable because if you add all the other weather variables then nothing is statistically significant. This would lead to the interpretation that weather is even less important than we thought, with wind gust being the least insignificant out of a handful of insignificant variables.

To be as short as possible, screw the weather, start your studs.

 

[ratbast]

OK, spoke too soon, methodology might not be too great either. Without knowing what he did exactly, it looks like the wind gust model he's offering only includes the wind gust variable, depending upon how his software did the stepwise regression. A one variable regression has a greater chance of achieving statistical significance than multi-variate regressions, but at the expense of valid interpretations.Here's an example. I work in educational research. I could run a regression with the dependent variable being student test scores, and the single independent variable being, let's say, race of student. Almost guaranteed I would get statistical significance. However, that's not very sound research. My boss would come back at me and say to add more variables, like socioeconomic status. With two independent variables (race and socioeconomic status), two things would happen: 1) the significance of the race variable would decrease and 2) the R-squared value of the entire model would increase. This is because part of the significance of race is explained by socioeconomic status, and having the two variables together in the model explains more variation in student test scores than just race by itself (bigger R-squared). The goal, so to speak, is to get a model with as many variables as possible, and still end up with at least one that is statistically significant. That narrows the field more by 1) eliminating insignificant variables and 2) increasing R-squared to decrease the chance of omitted variables being a factor. My guess is the statistically significant model only has the wind gust variable because if you add all the other weather variables then nothing is statistically significant. This would lead to the interpretation that weather is even less important than we thought, with wind gust being the least insignificant out of a handful of insignificant variables. To be as short as possible, screw the weather, start your studs.

That was quite interesting.

 

[Leroy Hoard]

To be as short as possible, screw the weather, start your studs.

 

[Wreckincrew]

To be as short as possible, screw the weather, start your studs.

Ahhh Leroy. "Coach, if you need 1 yard, I'll get ya 3, if you need 5 yards, I'll get ya 3."Gotta be a top 10 quote!

 

[davidwb]

OK, spoke too soon, methodology might not be too great either. Without knowing what he did exactly, it looks like the wind gust model he's offering only includes the wind gust variable, depending upon how his software did the stepwise regression. A one variable regression has a greater chance of achieving statistical significance than multi-variate regressions, but at the expense of valid interpretations.Here's an example. I work in educational research. I could run a regression with the dependent variable being student test scores, and the single independent variable being, let's say, race of student. Almost guaranteed I would get statistical significance. However, that's not very sound research. My boss would come back at me and say to add more variables, like socioeconomic status. With two independent variables (race and socioeconomic status), two things would happen: 1) the significance of the race variable would decrease and 2) the R-squared value of the entire model would increase. This is because part of the significance of race is explained by socioeconomic status, and having the two variables together in the model explains more variation in student test scores than just race by itself (bigger R-squared). The goal, so to speak, is to get a model with as many variables as possible, and still end up with at least one that is statistically significant. That narrows the field more by 1) eliminating insignificant variables and 2) increasing R-squared to decrease the chance of omitted variables being a factor. My guess is the statistically significant model only has the wind gust variable because if you add all the other weather variables then nothing is statistically significant. This would lead to the interpretation that weather is even less important than we thought, with wind gust being the least insignificant out of a handful of insignificant variables. To be as short as possible, screw the weather, start your studs.

As a professional econometrician, I can say that this is a good description of the statistical process. Shirtless is almost certainly correct that the multivariable model would give a better forecasting result than the single variable model. It's hard to say without the actual data whether or not the individual explanatory variables would be statistically significant or not. Of course, it is also possible that a "kitchen sink" model fits the data well, but is absolute crap with respect to forecasting -- it is very easy to overfit a model.

 

[Hipple Long Ware & Peete]

Bump for the impending inclement weather that will occur some places this weekend.

http://www.accuweath...y-nfl-gam/42491

Umm....that article was from December 5, 2010.

looking at this no places should have wind concerns for this weekend,if that even matters

http://www.donbest.com/nfl/weather/

<br class="Apple-interchange-newline">NFL FOOTBALLSat, December 17Forecasted Game Time Conditions303 Dallas

304 Tampa Bay

68° , Wind 6 MPH

Click for details & forecastSun, December 18Forecasted Game Time Conditions305 Washington

306 NY Giants

36° , Wind 5 MPH

Click for details & forecast307 Green Bay

308 Kansas City

50° , Wind 8 MPH

Click for details & forecast309 New Orleans

310 Minnesota

Played in Dome311 Seattle

312 Chicago

40° , Wind 7 MPH

Click for details & forecast313 Miami

314 Buffalo

36° slight chance light snow showers , Wind 10 MPH

Click for details & forecast315 Carolina

316 Houston

62° , Wind 7 MPH

Click for details & forecast317 Tennessee

318 Indianapolis

319 Cincinnati

320 St. Louis

Played in Dome321 Detroit

322 Oakland

55° , Wind 5 MPH

Click for details & forecast323 New England

324 Denver

56° , Wind 5 MPH

Click for details & forecast325 NY Jets

326 Philadelphia

40° , Wind 4 MPH

Click for details & forecast327 Cleveland

328 Arizona

57° likely light rain showers , Wind 5 MPH

Click for details & forecast329 Baltimore

330 San Diego

58° , Wind 3 MPH

Click for details & forecastMon, December 19Forecasted Game Time Conditions331 Pittsburgh

332 San Francisco

55° , Wind 5 MPH

Click for details & forecast

 

[davidwb]

But it was still a useful discussion, even if the generating post was a year old.

 

[Hipple Long Ware & Peete]

But it was still a useful discussion, even if the generating post was a year old.

Yeah. a little to much 420 yesterday. Buddy of mine in NYC was telling me that they were expecting bad gusting wind. When I searched for it and found what I was expecting to find I didn't notice that the weather report was last years. I do think that the 'gusting' theory makes sense though vis a vis quarterbacks. Ask any golfer. When the win is consistently blowing it is still playable. You have to adjust your shots but these guys are pros and they can do that. They can 'feel' how much to allot for wind, even one blowing hard as long as it is consistent, same with rain. When it gets maddening/difficult is when the wind is swirling and gusting as you plan for the win and then the wind disappears right as you shoot/throw or vice versa.

 

[Hipple Long Ware & Peete]

Bump for 2012

 

[Hipple Long Ware & Peete]

Bump for 30 mpg gusts in CHI?