Eli's QB Rating by Wind and Temperature

you would need to do two separate graphs

Winds vs. Rating

and

Temp vs. Rating

Then evaluate the correlation based on those results

If you want to post the data, I can run the true analysis for you

That's the kind of constructive input that I need! I'll fuss with it a bit and see if I can't get it to work.

Do you have an email address that you'd be willing to post? I'll send you the Excel file if you'd like.

.

.

after defibrillation from her heart failure after reading all the Shockey threads today.

haha.

Pretty interesting, thanks for posting.

In brief, R squared is the relative predictive power of a model. R squared is a descriptive measure between 0 and 1. The closer it is to one, the better your model is. By "better" we mean a greater ability to predict. A value of R squared equal to one, would imply that your quadratic regression provides perfect predictions.



When looking at the relationship between QB Rating vs. Temp, there is very little correlation. An R-squared value of .0871 means that based on the data we have, if we were to make a prediction of Eli's QB Rating based on the temperature, we would only be accurate about 8.8 times out of 100. This translates into little or no statistical relevance.




When looking at the relationship between QB Rating vs. Wind Speed, there is also very little (thought slightly more) correlation. An R-squared value of .1481 means that based on the data we have, if we were to make a prediction of Eli's QB Rating based on the Wind Speed, we would only be accurate about 14.8 times out of 100. This translates into little or no statistical relevance, though it does carry more statistical relevance than QB Rating vs. Temperature.

is gonna be pissed. Stealing his thunder.

for the data

would be logicial no?

other than saying "Line go down slightly.. Eli = Bad"

;)

I couldn't quite figure out what Weatherman was asking me to do, but here is the data sorted by QB Rating rather than chronologically. I think it's a bit more clear:



And as in the first post, here is the link to the larger image:

Link to Larger Image - ( New Window )

based on the lack of correlation, I don't think a cost/benefit analysis of putting a roof on the stadium will show that a dome would be worth it.

lukes.

that jlukes, having some sort of clue about how statistics works, beat me to the punch!

Nice work!

Your comment is just plain mean. You meanie.

a scatter plot too, commented too fast - jlukes nailed it, those two charts show the direct comparison (QBR indexed to the temp/wind variables) to show correlation

why let them put money in their pockets. its not like our tickets are any less because we dont have a roof. on top of that were prob going to get smacked with psl's

I was an Econ major, I had to take all sorts of statistics and forecasting classes -- glad I can finally put my degree to use!

I teach music theory. Not too much use for statistics there, though I can write a mean fugue...

...have trouble deciphering it, lol. Seriously, I think I forgot everything. A low R squared means that the variation is mostly due to randomness and NOT due to the the independent variable. That's how I remember it. R squared= SSR/SST.

Am I getting the jist? Do you work for a job that incorporates stats?

Right? 0.14 isn't much.

"An R-squared value of .1481 means that based on the data we have, if we were to make a prediction of Eli's QB Rating based on the Wind Speed, we would only be accurate about 14.8 times out of 100."

I wonder if it's just a different way of looking at it. But as I said the way I learned it, it means that 14.8% of the variation from the mean is due to the regression model and 85.2% is in fact RANDOM.

Maybe over the same length of time that you looked at Eli. That would be helpful.

LOL, not you got me going I'm a big fan of stats!

We have game by game data for Eli over his whole career. I don't think we have that data available for Favre, unless we can find a Green Bay fan as obsessive as I am who has kept the accurate records...

the R^2 value of 0.1481 would show that the variable (in this case wind) can be said to account for 14.81% of the variation in QBR... not a significant proportion. Other factors could account for other percentages, its not directly implied that the remainder is random.

correct and incorrect the closer the R-Squared is to zero, the less the dependent variable is "influenced" by the independent variable. You can't necessarily say that the closer to zero the more random it is, because there may be a 2nd, 3rd, or 4th variable that comes into play.

I am a strategy analyst in the logistics field. Graduated college in 06. I do a lot of data analysis, but don't do too much advanced analysis like this (yet).

Ok, this is how I learned it. SST= The difference between the actual "Y" and the mean of "Ys" (mean without taking into account the independant variable(s)). This includes SSR, which is difference that's due to the regression model and SSE which is the difference from the regression line "Y" and the actual "Y", and thus is only due to randomness. So R squared looks at the SSR, which is how much of the difference is due to the regression model, as the part of SST. Meaning, we're checking how much of the difference is due to the regression. A low R squared says that most is due to SSE, meaning most of the difference is RANDOM. So really we see that the reason Eli isn't great in cold weather and wind is random.

it would be cool if we could get that data for other QBs because then we could compare that data.

For instance, compare Eli to Ben.
Would Ben's R-squared value be lower or higher? If higher, it would mean that Ben is "more affected" by temperature changes than Eli.

You just have to be careful when making statements like that because when you are dealing with such small r-squared values, there isn't that much of a difference between .088 and .21 -- they would still be considered statistically insignificant in predicting QB rating from temperature.

...the R squared, but that's why you do an adjusted R squared to take that into account.

for a 100% true statement, leave the word 'random' out of it.

The best conclusion to draw is that Eli's QB Rating CANNOT be accurately predicted based on Temperature or Wind Speed.

...was before your post explaining it. I guess I see what you mean. You can say that the dependant variable is less influenced by the independant variable, but you can't make that leap and say that it's more random?

...make when they say that one variable causes another rather than correlates. Or in a hypothesis test that you "accept" a hypothesis rather than "do not reject"? I guess statistics is a very strict field and making leaps of judgement is inaccurate.

based on our model

QBR = (0.6007x(temp))+39.778
QBR = (0.6007x(3))+39.778
QBR = 41.58

So, based on the historical data and trends, there is an 8.8 % chance that Eli will have a 41.58 QB Rating on Sunday

Again, because of the low r-squared, that is not deemed to be accurate or statistically significant

exactly

to try and figure out what his QB rating will be this week or to see if there is a coorelation between the weather and his QB rating?

and by showing that if you did try to predict using such a statistically insignificant variable, your forecast would be wrong/not make much sense.

i meant the latter

my goal was to see if there is a correlation between weather and rating, but to be honest, I don't see a significant trend over 60 games to say for sure. There just aren't enough very cold games for us to be really sure.

However it would be much more useful if one could control for injuries on offense and the caliber of the opposing defenses. Then, if there is any correlation, that should be compared with the correlation for other NFL QBs to see if Eli is more or less effected by weather conditions than other QBs.

given jlukes' work, a significant correlation when controlling for other factors is unlikely.

One thing that I'm interested in. I learned also that you can't extrapolate outside of the range of temperatures that are given. In other words if Eli Manning never played in temeperatures as low as 3 degrees then you can't put that into your equation. Am I misunderstanding it?

stuff, but I do find his charts interesting. QB rating looks to be evenly scattered throughout all temp ranges. Ergo temp is not a major factor.
Wind is another story. In eleven games played where wind was 12mph or above, Eli has only one game with a rating over 80 vs seven games below 60. Many well below.
Now analysts and QB's have been saying for years that wind is the biggest factor affecting the passing game, so I suspect that every QB in the league would show the same trend. However, I also suspect that Eli is more affected than most because it seems to me he has alot of wobble in his passes. That wobble means more surface area exposed to the wind.

Eli has a little 3 or 4 game streak where he has a consistent streak with good to acceptable QB rating. In between he is a whipsaw. We're 3 games into this consistent streak with his highest QB rating group ever. Hopefully he has a 4th in this streak and then breaks out of his trend with a 5th. :-)

no, it doesnt matter what the temperature is. Plug any temperature into that linear equation and you will only have an 8.8% chance of predicting what his QB rating will be at that given temperature

Could you do a regression to take into factor both wind and temp and forecast the rating? I vaguely remember being able to do somthing like that, but don't have my SPSS software with me to do it.
Perhaps a low temp + high winds might show more deterioration for Eli... though I suppose that would hold true for any QB unless he was the abominable snowman

just kidding, i did enjoy them last week but thankfully they can be put away now.