Have the Dolphins Really Improved this Year?

[Shouright]

Looks like yet another posting of the disclaimer below might be in order. :)

 

[roy_miami]

Now we know why WV likes Shourights threads, they can be very entertaining.

Spoiler

:snack:

P.S. That Calvin cartoon pretty much sums it up, good work.

 

[Shouright]

Now we know why WV likes Shourights threads, they can be very entertaining.

...as well as a hell of a study in social psychology. :)

 

[tazthenomad]

You are picking the 3rd week to justify this model?

How would you know the increase in wins wasn't due to the variables that tend to fluctuate randomly, rather than more stable attributes of teams?

Of course you could call any increase in wins an "improvement" by that measure alone (i.e., the team's record), but if that improvement is being accomplished via variables that tend to fluctuate randomly, rather than via those that tend to reflect more stable attributes, then it can be said the "improvement" is being done with "smoke and mirrors" if you will, and isn't likely to be sustained.

This is I think what we saw during the first three games of the season, when the team was 3-0, yet its average number of estimated wins for a 16-game season during that period -- based on the more stable variables that strongly correlate with winning -- was merely 9.4.

And what a surprise -- we're set to win about "9.4" games this year (in quotes because it'll likely be either 9 or 10).

OK... but looking at the plot pretty much shows had bad this stat is a tool for predicting anything other than how bipolar the posters will be here. I modified my first plot to add the moving average and you can see that it doesn't move much because the averaging process obscures the variation. This is why time based plotting and not averages are best for detecting performance trends. You have asked for input on how to make your use of stats more technically correct. One thing that you do often is use averages rather than time-based series to measure performance. This is almost always bad practice because averages obscure variation - which is what we are interested in.

The weekly metric goes up and down like a yo-yo, just like many of the bipolar reactionary posts we get on this board. (I think you could predict when certain posters will come out with this metric.) :) The moving average and the stat for the third game which you quote is almost the same number. But as you can see the moving average for the stat has barely moved during the past 5 weeks. This stat is awful on a week-to-week basis and the moving average approach doesn't match what we are seeing at all. I don't see how you can pick the 3rd week only just to validate this as a prediction model.

 

[Shouright]

We are a smart well coached team, other than the frequent avoidance of rushing attempts. I hope that is a thing of the past.

Lately our offense is incredibly dangerous, creating spacing everywhere. That aspect has been all but ignored while praise is heaped on Tannehill.

I'm not going to get carried away if the passing numbers aren't ideal. That's true of the Dolphins or any other team.

This year we have cornerbacks who have made plays when they are available, and the key penalties have been next to none, outside the first Patriot game and a stretch in the Ravens game. Overall we certainly can't complain about the officiating, given our low penalty numbers.

Nothing is 100% true to the numbers, as shouright has pointed out countless times. But if we want to make the jump forward that so many will no doubt be comfortable applying toward 2014, our passing attack has to improve considerably without sacrificing anything on the defensive end. I'm never happy relying on interceptions, given how that category can change on a dime.

An 8.1 quarterback provides margin of error that a 7.1 quarterback can't begin to threaten.

Thanks for your post. What I bolded above is what I meant by the distinction between variables that tend to fluctuate randomly, and those that reflect more stable attributes, although interception rates tend to be more stable than numbers of interceptions alone.

 

[Shouright]

OK... but looking at the plot pretty much shows had bad this stat is a tool for predicting anything other than how bipolar the posters will be here. I modified my first plot to add the moving average and you can see that it doesn't move much because the averaging process obscures the variation. This is why time based plotting and not averages are best for detecting performance trends. You have asked for input on how to make your use of stats more technically correct. One thing that you do often is use averages rather than time-based series to measure performance. This is almost always bad practice because averages obscure variation - which is what we are interested in.

The weekly metric goes up and down like a yo-yo, just like many of the bipolar reactionary posts we get on this board. (I think you could predict when certain posters will come out with this metric.) :) The moving average and the stat for the third game which you quote is almost the same number. But as you can see the moving average for the stat has barely moved during the past 5 weeks. This stat is awful on a week-to-week basis and the moving average approach doesn't match what we are seeing at all. I don't see how you can pick the 3rd week only just to validate this as a prediction model. View attachment 11927

I appreciate your response and the deeper look at the data, including the plot, but the model predicts overall win percentage, not necessarily week-to-week performance. I was using the first three games not to validate the model, but to show that, despite the 3-0 record at the time, the team wasn't performing at a high level overall during that period with regard to the stable variables that predict overall win percentage. "We're on-pace to be 16-0," or even "we're on-pace to be 12-4," certainly weren't things that should have been said around that time, given the data.

I believe CK appeared here during that period and adjusted his season win prediction upward by one game (IIRC from 8 to 9) and was lambsted by some for it. I think this gives some insight into why. The team wasn't playing all that well with regard to stable variables that predict winning, despite its record.

Also, I'm very interested in how the team's performance has shifted up and down during the season, which is why I decided to gather game-by-game data and look at it.

 

[nyashfan]

Now I know how Torquemada's victims felt.

 

[FinfanInBuffalo]

What I find confusing is that on the one hand, people tend to call Tannehill a "developmental" quarterback, or they caution us that patience is required while he develops, yet on the other hand, Tannehill's performance -- during the same "developmental" period, no less -- with regard to the statistical measures of quarterback play that distinguish the all-time greats from the rest of the pack, are attributed by those same people to other parts of the team.

And it would, if you were comfortable deeming correlation to equal causation, without supporting it with anything objective.

Once again, we have an inconsistency: Tannehill is supposedly developing and needs time to perform well, but at about the time the experts believe the light comes on for developmental quarterbacks (about a thousand pass attempts), the improvement in his performance is attributed to the offensive line, despite the fact that it can't be supported objectively.

Sort of like attributing YPA solely to the QB despite the fact that it s computed using measures of someone else's performance......

 

[FinfanInBuffalo]

Once again, how do you know the wins weren't accomplished via variables that tend to fluctuate randomly, versus those that are strongly correlated with winning and tend to reflect more stable attributes of teams?

Who says that stats that are strongly correlated to winning reflect more stable attributes of teams?

 

[Shouright]

Who says that stats that are strongly correlated to winning reflect more stable attributes of teams?

The people who have found that those variables are correlated with themselves at different points in time.

There are statistics that are strongly correlated with wins that are correlated with their own future occurrence, and there are those that are strongly correlated with wins that are not correlated with their own future occurrence. The former tend to be more stable, and the latter tend to be more random.

 

[Shouright]

Now I know how Torquemada's victims felt.

If only they could've been armed with an effective disclaimer....

 

[FinfanInBuffalo]

The people who have found that those variables are correlated with themselves at different points in time.

There are statistics that are strongly correlated with wins that are correlated with their own future occurrence, and there are those that are strongly correlated with wins that are not correlated with their own future occurrence. The former tend to be more stable, and the latter tend to be more random.

what is the correlation between YPA and wins?

 

[2413fanphins]

This thread blows goats.

 

[tazthenomad]

I appreciate your response and the deeper look at the data, including the plot, but the model predicts overall win percentage, not necessarily week-to-week performance. I was using the first three games not to validate the model, but to show that, despite the 3-0 record at the time, the team wasn't performing at a high level overall during that period with regard to the stable variables that predict overall win percentage. "We're on-pace to be 16-0," or even "we're on-pace to be 12-4," certainly weren't things that should have been said around that time, given the data.

I believe CK appeared here during that period and adjusted his season win prediction upward by one game (IIRC from 8 to 9) and was lambsted by some for it. I think this gives some insight into why. The team wasn't playing all that well with regard to stable variables that predict winning, despite its record.

Also, I'm very interested in how the team's performance has shifted up and down during the season, which is why I decided to gather game-by-game data and look at it.

My point for the post is that the model does not predict it. Not on a week-to-week basis and not on a moving average basis. You can't pick one point and say that it worked.

 

[Shouright]

My point for the post is that the model does not predict it. Not on a week-to-week basis and not on a moving average basis. You can't pick one point and say that it worked.

You mean for the sample consisting of the 2013 Miami Dolphins?