Good stuff guys, I apologize if my questions are so elementary, but this is groundbreaking stuff (for me, at least). A lot is being said, so to distill/organize this for the more simple minded (me), SSOG and wdcrob, would you say then that the answer to my questions above (now numbered) are:1. Yes, the subject has been adequately tested to conclude with high level of confidence that there is no relationship between number of carries and decline in effectiveness. 2. Yes, even if the relationship in Doug’s (by the way, who is Doug) study holds, that indicated 800 carries = 1 year, it is still true that we should assume that a 26 year old with 1800 carries is equally close to decline (in effectiveness, which is what is most important imo) as a 26 year old with 900 carries (i.e., I am not interpreting what SSOG or Doug is saying correctly about a 1 year difference).3. Yes, if talent is no question, it makes no difference (in a positive way) if a player has 800 fewer carries in any evaluation (i.e., if you own the player, you wouldn’t prefer that your player has very few carries vs. a ton of carries).4. Yes, the fact that CJ Spiller, a wildly talented but previously underused back, has so few carries (and intuitively to some, less wear on his tires) due to non-use is irrelevant (in a positive way) to his overall outlook. Last question
o you believe (A) that the number of carries over time (not including ridiculous loads like Larry Johnson) does not increase the wear and tear on the player (i.e., number of hits is irrelevant). Or, do you believe (B) that the number of hits does take a toll on players over time but that the lack of predictive value makes looking at number of carries a useless exercise?
1. No. Again, that's not how statistical analysis works. You never prove that there's no relationship. Hell, you never prove that their IS a relationship, either- when statistical analysis returns a positive result, it's not saying "there is definitely a relationship between these two variables", it's instead saying "it's highly unlikely (but not altogether impossible) that you'd see the results you observed unless there was some sort of relationship between the two variables". The standard threshold for significance is 95%, which is essentially saying "there's only a 1-in-20 chance, based on your observations, that these two variables are unrelated". But even that isn't proof- if you find 20 statistically significant results, on average one of them will have been a false positive. If you flip 100 coins ten times each, at least one of them will come up heads enough times that statistical analysis will say, to a statistically significant degree, that the coin is likely weighted. But the coin isn't weighted, of course- it's just the nature of random events. You'll see results that look meaningful to our pattern-seeking brains, but truly random results will frequently look meaningful when they're not. And even if statistical analysis identifies a meaningful relationship, it offers no insight into the relationship. For instance, one of the biggest predictors of a child's intelligence is how many books could be found in their home growing up (this is a true relationship, by the way). Identifying the relationship is nice, but what does it mean? Do children who are exposed to a lot of books wind up becoming more intelligent (the nurture story)? Or are parents who are intelligent more likely to own books, and parents who are intelligent also more likely to have smart babies (the nature story)? A combination of both? Something else? People who are very rich are actually less likely to be robbed than people who are very poor, but in this case there are multiple variables at work- rich people are more desirable targets (increases their chances), but they have less interaction with potential criminals, live in safer neighborhoods, and have better security systems (decreasing their chances). Sometimes when you have two competing explanations, one factor is stronger and outweighs the other. Other times, both factors cancel each other out returning a "no relationship" result when there is, in fact, a relationship- it's just complicated. So you can never be sure whether a relationship exists (even when statistical analysis suggests it does), or doesn't exist (even when statistical analysis suggests it doesn't). And even if you feel confident that a relationship exists, you can't know the nature of the relationship without running carefully controlled experiments, which is impossible for the casual fantasy owner. In short- statistics is a hard, messy, and largely unsatisfying endeavor. Can I say with certainty that there's no relationship between workload and decline? No, but I've seen enough of the numbers for me to conclude that there's no substantive reason for me to assume that there IS a relationship.2. Statistical analysis tells us two things- what is the nature of the relationship, and how likely is it that the relationship actually exists instead of just being random noise in the numbers. In Doug's study, the answer to the first question was "if the relationship is real, one year is similar to 800 carries". The answer to the second question is "we should not be at all confident that the relationship actually exists instead of just being random noise in the numbers that appears meaningful on cursory examination". 3. Worst case scenario is that there is no relationship between workload and decline. In that case, assuming all else truly is equal, preferring the low-workload player costs you nothing. Best case scenario, there is a relationship, and preferring the low workload player benefits you something. So in this hypothetical- all else truly being equal- favoring the low-workload player is rational. The problem is that all else is never equal, so the question becomes how much you weight workload compared to the other factors. Would you prefer a low-workload back over a comparably talented high-workload back a year younger? Would you prefer a low-workload back over a slightly more talented high-workload back the same age? In both cases, I would not. We aren't comparing CJ Spiller against Bizarro Spiller, who is identical in all ways except he played under Cam Cameron and had a huge workload. We're comparing your Spiller's to your Ray Rices, your Jamaal Charleses to your Marshawn Lynches, and in those comparisons, the relatively light workload is a factor that carries virtually no weight to me. 4. Yes. I love Spiller because he's an unstoppable beast with top 10 pedigree. His workload (or lack thereof) does not play a role- either positive or negative- in my evaluation. Finally, gun to my head, ignoring the stats entirely, I do think that carries cause repetitive wear... BUT, I also think that coaches are generally smart and give carries to backs capable of handling the workload, so that RBs with lots of carries are not more "worn down" than RBs with few carries. If you could somehow create an alternate universe and create a bizarro Stephen Jackson who entered the league at age 26, then sure, I'd think he'd have more left in the tank. But you can't, so for practical purposes, past carries hold little predictive value when estimating future carries.
