09-30-2015, 03:42 PM
I think the author accounted for changes in personnel by looking at the year-to-year correlation in other statistics, finding that those are much higher than "forcing turnovers". If every season were truly independent due to roster turnover, all the stats would be uncorrelated, but some, such as yards allowed, are more correlated than turnovers.
I think there is probably a better statistical way of answering the question, which is to look at turnovers within a season. I assume that turnovers per game can be modeled with a poisson distribution (https://en.wikipedia.org/wiki/Poisson_distribution), and you can look at between team departures from that distribution that would exceed the within team departures.
By the way, we've had 1, 1, 0, and 1 turnovers in our four games. If you model turnovers with a Poisson distribution, the 95% confidence interval of our turnover rate is 0.19 to 1.94 turnovers per game (that is, the observed data would be consistent with a long term rate of between 1 turnover every 5 games, in which case we have been really lucky, and 2 turnovers per game, in which case we have been unlucky).
BC
I think there is probably a better statistical way of answering the question, which is to look at turnovers within a season. I assume that turnovers per game can be modeled with a poisson distribution (https://en.wikipedia.org/wiki/Poisson_distribution), and you can look at between team departures from that distribution that would exceed the within team departures.
By the way, we've had 1, 1, 0, and 1 turnovers in our four games. If you model turnovers with a Poisson distribution, the 95% confidence interval of our turnover rate is 0.19 to 1.94 turnovers per game (that is, the observed data would be consistent with a long term rate of between 1 turnover every 5 games, in which case we have been really lucky, and 2 turnovers per game, in which case we have been unlucky).
BC