Todays stats are redundant

[zimpy27]

Usually you use statistics for a controlled experiment. Sometimes you use stats to observe things that aren't in control but you do it by adding standards or finding an internal standard. And usually you try to observe or retrieve as much different types of data from a system that you are trying to understand.

Usually when planning an experiment you think of the data analysis that is required to answer your question. Then you think about what data you need for that analysis, then you design an experiment to retrieve that data.

The way people try to build stats on data that is retrieved without purpose seems broken to me. It seems like all these analyses are just ad hoc and trying to cling to a set of arbitrary parameter because that's all we have.

Does anyone know of anyone who does this sort of work themselves by taking down their own stats from games?

[D Nice]

Usually you use statistics for a controlled experiment. Sometimes you use stats to observe things that aren't in control but you do it by adding standards or finding an internal standard. And usually you try to observe or retrieve as much different types of data from a system that you are trying to understand.

Usually when planning an experiment you think of the data analysis that is required to answer your question. Then you think about what data you need for that analysis, then you design an experiment to retrieve that data.

The way people try to build stats on data that is retrieved without purpose seems broken to me. It seems like all these analyses are just ad hoc and trying to cling to a set of arbitrary parameter because that's all we have.

Well, you're right about this, but I think some fleshing of out of how this exactly pertains to basketball (and specifically certain metrics) would serve you well. In theory the "test" would be "players with these similar {x Metric} scores in the future will help their teams out to a similar degree of players with {X} scores in the past," the problem with that is it is an incredibly difficult result to actually evaluate (because the "player" variable is never truly isolated) and we kind of already have seen an exceedingly number of bogus results where scrubs end up above stars (if you're talking about RAPM or xRAPM anyway).

Things like W/S and PER are insanely useless in assessing, well, anything specific really, they're only good for catch all looks at what tier a player's production falls in, but I've completely given up fighting that battle. People actually thing there is some magical information gleaned from PER that isn't seen in a player's composite box score (when really it's the opposite). You're post doesn't make it clear if that's really what you're referring to though.

Personally I do find it a bit amusing that people use predictive methodologies as a means of formulating "past impact estimation values" and are probably not even aware that's what they're doing but it at least gives an alternate look at otherwise rote data.

Does anyone know of anyone who does this sort of work themselves by taking down their own stats from games?

I mean Synergy used to provide possession stats (P&R efficiency, post-up efficiency, etc), but those are no longer available to the public. The thing is, these days, anyone who does do this probably wouldn't give their data away for free. Not until virtually all NBA teams have "in-house'd" whatever it is they are doing and have eliminated that "market inefficiency" so to speak. As long as there are a number of teams that would rather pay for the data than develop a collection and evaluation infrastructure themselves, you probably won't see the type of data you're asking for available for free.

In "Free Land," personally I'd be interested in "follow-rate" (the % of a guys misses end up as put-backs partially adjusted for offensive rebounding strength of the team's PFs and Cs). I think it would go a long way to showing some of the "hidden" value of different playstyles. There was an article written I believe in 2012 on this effect with Kobe (led the league), and I think it would definitely extend to some other guys as well (healthy Rose, healthy Wade, Westbrook, etc).

[Kohanz]

Does anyone know of anyone who does this sort of work themselves by taking down their own stats from games?

From my point of view, this is what working with SportsVu data is. You take raw observations and go from there.