This site attempts to inject some science into combine stats predicting NFL career stats:
This is, in effect, what we do when we do basic analysis on combine stats. My example might be extreme, but it illustrates the mechanism at work. We trick ourselves into believing that these attributes don't matter when they really do. It's just that we only get to observe the very best players who have survived an extreme selection process. After surviving selection the differences among players in measurable attributes is swamped by their differences in "error," which is just shorthand for "all the other stuff that affects career success."
The survivor effect works like this: The fact that relatively low-attribute players were invited to the Combine or made it onto the NFL field means that they had other qualities and skills that compensated for the low attribute. So the selection process introduces bias into that "error" term. The error is no longer randomly distributed, but very likely to be highly positive.
The lesson here is that scouts should be using the measurables as a "filter" rather than a "coefficient." In other words, they should be looking for red flags where a player might be below a certain minimum level of an attribute, rather than using the attribute as a relative indicator of potential. And as I understand it, that's how teams generally use the Combine in practice.
[URL]http://www.advancedfootballanalytics.com/index.php/home/research/draft/234-combine-selection-effect[/URL]
Bookmarks