I sent him this thread link as an fyi on twitter.

There is a lot of detail in your new comment and the older ones. I'll listen for now.

## Nylon Calc notice

### Re: Nylon Calc notice

A lot here, but briefly, comparing players in year to year will never add up to season totals as rookies enter every year and guys fall out of the league every year. Players with fewer than 800 minutes were dropped both because their ratings are more noisy and because it's a floor for a subjective limit on a rotation player in both years.

This was done simply as a presentation of the public stat's correlation year to year, not as a projection. The public single year stat gets cited often so I wanted to explore how stable it is and how that has been impacted by changing teams.

IT's change was definitely one of the outliers in the dataset, especially if you look at ORPM year to year.

For players changing teams with 800 minutes in both years, regression formula was -.468 +.524 prior RPM, coefficient std error of .043.

This was done simply as a presentation of the public stat's correlation year to year, not as a projection. The public single year stat gets cited often so I wanted to explore how stable it is and how that has been impacted by changing teams.

IT's change was definitely one of the outliers in the dataset, especially if you look at ORPM year to year.

For players changing teams with 800 minutes in both years, regression formula was -.468 +.524 prior RPM, coefficient std error of .043.

### Re: Nylon Calc notice

Before responding point by point to your reply, let me just say that your initial results are interesting, less because of the sign of the effect (i.e. it is reasonable to believe and unsurprising to find, for a bunch of reasons, that a traded player would become less productive relative to remaining on his home team) but because of the magnitude. If I were a +3 NBA player and was at risk of being traded, knowing that I would face being "devalued" by +1.05, all else equal, this would matter to me a whole lot (as it should to any team looking to hire me).

So, improving the quality of this estimated effect seems to me a very praiseworthy goal, and in this direction, one should at least evaluate the statistic on its own terms, by regressing base year RPM with aging effect added on next year RPM. If my intuition is correct, it wouldn't change the slope of a univariate regression much, but it would improved the precision of the estimate.

This having been said...

But particular year(s) average RPM wasn't the primary point. Why I was scratching my head when looking at the data was that the scatter plot simply looked strange given such a large fraction of the observations below the unit slope line - what implies (I think) that the observations above the line correspond to much higher number of minutes played on average, and that doesn't seem quite right...but maybe it is.

So, improving the quality of this estimated effect seems to me a very praiseworthy goal, and in this direction, one should at least evaluate the statistic on its own terms, by regressing base year RPM with aging effect added on next year RPM. If my intuition is correct, it wouldn't change the slope of a univariate regression much, but it would improved the precision of the estimate.

This having been said...

Yes, but this effect is relatively small, and, as rookies and oldies have quite negative RPM on average and don't play a large share of total minutes. (Checking numbers for the seasons ending in 2015 and 2016, the non-rookie and retiree average RPM was about 0.16 in both years.) Slightly more significant is eliminating the 799 minute and under subset, where the rotation player and starter average is 0.31 for the four seasons ending 2015-18.

But particular year(s) average RPM wasn't the primary point. Why I was scratching my head when looking at the data was that the scatter plot simply looked strange given such a large fraction of the observations below the unit slope line - what implies (I think) that the observations above the line correspond to much higher number of minutes played on average, and that doesn't seem quite right...but maybe it is.

A case can be made for exclusion (as well as inclusion). I am curious if you have the results to share for this subset and/or the regression results for the entire sample.

I am not sure I understand the distinction between correlation and projection. To my interpretation, all else equal, the correlation is a simple projection. Pending refinements, I should think that a team not intending to make the effort to understand the origin of a player's success and facilitate his transition accordingly should expect, on average, to be rather disappointed with the results of player acquisition.

The observed deterioration in both IT and Jae Crowder's performance is noted; I just cannot find either of them in your scatter plot.

0.043? The plotted data is much noisier than that figure implies, no?