# The Case for Pau Gasol, Setting Up an Experimental Design

It’s been some time, but let’s continue the quest to shift NBA “stats” to NBA “Stats.”

Pau Gasol will be in the HOF. The current hot topic is, what value does he add to the identity-less Lakers? Pau’s argument: By being featured less on offense, the current Laker system has marginalized out Pau’s versatility; in turn his production has taken a blow. D’antoni’s argument: even when Pau gets his touches, he’s not productive; further Pau lacks the hunger and shows symptoms of Radmanovic levels of space cadetness.

Causal Analysis, Potential Outcomes, and Experimental Design provide good candidate tools to structure a more rigorous study than some barbershop speculation when getting your man weave.

Main Causal Question
What is the effect size and direction of “usage” on pau’s “production.”

We make an implicit assumption that some measurable outcome (Pau’s ‘production’ as measured by points) is causally related to some measure of usage. That is, we are not asking the causal “Why?” we assume that; we are asking the causal “How?”

Potential Outcomes
Consider the following:
Pau’s points given we force him to “do” option A
compared to
Pau’s points given we force him to “do” option B

The key point in potential outcomes: For each ‘trial,’ we only “observe” the outcome when “do”-ing one of the two options.

However, for trial “i” we would love to say something about both options
$Y_{i}(do A) - Y_{i}(do B)$

Treat Practices as Experiments
Blind the players (keep them in the dark that a specific practice play is actually part of an experimental study). We make the simplifying assumption by ignoring real world differences between effects seen in “practice” and effects seen during game time. Loosely speaking, this is Sasha Vujacic vs Allen Iverson.

Controlling for other Factors
Although the initial question is simply stated, we can and should roll up our sleeves by considering other possible factors that can cloud this “simple” relationship (eg consider controls).
Acknowledging that time is extremely limited (which affects the number of trials you can ‘run’) yet we still want to study many desirable factors, we should appeal to economical Fractional Factorials. Fractional Factorials are great when you want to consider the scenario when Trials << Factors.

Stay Tuned
So far, it’s all been soap-box talk. Stay Tuned for the example.

Advertisements