HATFIELD PLAYS WITH HIS MODEL

PRESEASON OVERVIEW

At the prompting of our Commish, I’ve concocted a little preview of our 2012 campaign by doing a little analysis that, I hope, provides an objective way of comparing BARB rosters and their potential for winning games in the Diamond Mind simulation. I don’t have the current stats for the simulation, nor have I run any simulation based on last year’s stats. Nevertheless, I do have predictions based upon an informal model that experiences teaches me is likely to match up pretty well with performance in DMB. I have some misgivings about doing this, as I have some confidence that sharing the model provides some insights that could benefit other owners who are my rivals. But then again, there are many variables to account for and I doubt that everything I predict will transpire, and who will have the OCD necessary to do this sort of thing for themselves?

The first insight is that while many so-called “fantasy league” rankings are at best poor guides to the performance of hitters in OUR league, they are a good approximation of ability where starting pitchers are concerned. That’s because such rankings take into account durability and performance over time, and also the likelihood of their MLB team scoring runs to support them, a significant issue. Is C.C. Sabathia really a better pitcher than Madison Bumgardner? Hard to say, but it’s extremely likely he will have more wins because of the circumstances in which he pitches, and that actually fact makes such starting pitcher more valuable in DMB, because they tend to pitch more innings and get more wins, and the simulation tracks that.

The second insight is that the old saw that pitching is 60-70 percent of winning is really borne out by the simulation. Both the rotation and the bullpen contribute, but the rotations contribute more. BARB champions typically have staffs that are good enough to put them in the upper fifty percent of teams, and they typically have at least two top starters that they can employ in a short series to dominate the opposition. The only exceptions to this rule that I can recall are the founding draft-loaded 2003 Squirrels in the league’s inaugural season and Matt Caskey’s remarkable 2010 club that had a .300 hitter at every position. To summarize: pitching is important, and having top starters is even more important. I have therefore adopted the rule that starting pitching is 35-40 percent of a team’s success, while a bullpen is 25-30 percent, and the total contribution of a staff to winning is 65 percent.

Now why are those numbers important? Because, due to the way talent is concentrated in our league, it is very important to rank talent relative to other teams, rather than MLB teams that have an overall lower level of talent. To do this, I take an averaged ranking of starting pitchers and relievers, and then assign these numbers to pitchers on BARB rosters for both starters and relievers. Likely starters’ average rankings are summed to generate an average SP value for each BARB roster, the lower the better. BARB rosters are then ranked on the basis of this number and assigned a number 1-13, with 13 being the highest. That number is then multiplied by 0.6 to reflect the value of that roster’s group of starters relative to other pitchers. When this is done, it is clear that most of the top-ranked starting pitchers are in the East. Brooklyn (with the overall #1 ranked pitcher, Justin Verlander) has the best overall SP performance, but Sin City (now coming of age as a real BARB power) is right there with them, and the other teams in the division are certainly competitive.

Similar calculations are done to convert the rankings of relievers, but the spread of values being less, there are two important corrections applied that are highly-significant for bullpens: first, the average number of strikeouts per IP provides data about a vital edge essential in the most difficult of relief scenarios; secondly, bullpens that lack proven left-handed pitching are at a distinct disadvantage in getting favorable matchups in key situations: the importance of both of these is well-illustrated by the high level of performance achieved by Yuma’s bullpens, which are consistently underrated by many BARB owners.

Once the corrections are applied, the BARB bullpens are again ranked 1-13, but multiplied by a factor of 0.4 to indicate their relative contribution. The analysis indicates that Worcester is likely to have the league’s most dominant pen in 2012 and that (once again), Yuma’s unheralded relief corps will be one of the loop’s top performers.

Now, to finally reach the number that represents the relative strength of each of the 13 BARB squads in terms of the contribution made by pitching (estimated, remember, at 65 percent), the two numbers for SP and bullpen are added together. The result, interestingly enough, suggests that the East club with the best overall potential for pitching performance is Sin City, and the model suggests that this gives them a definite pathway to the post-season, and beyond.

For offense (which I estimate provides about 35 percent of a team’s competitiveness), the averaging of fantasy league ratings is a poor match for league performance, precisely because the categories in most fantasy leagues overrate the importance of stolen bases and hits to a very high degree. Carl Crawford is an above-average outfielder in the big leagues, and in our league as well…..he is not now, however, nor has he ever been one of the top 25 offensive players in baseball. It seems more useful to me to rank power, hitting and speed in decreasing order of relationship to offensive potential.

So, as a rough approximation, I simply ask myself three questions: does this player have a better than 50 percent chance of….(hitting 30 HR? hitting .300? stealing 20 bases?) If 30 HR…three points. If hitting .300…two points. If stealing 20 bases…..one point. I then add up the point totals for each roster, and derive a number. This number is used to rank BARB teams 1-13, and their rank is multiplied by 0.35 to generate their likely offense. You might think this is a very crude thing, where I’m discounting a ton of players who hit 20+ HR or who hit .290, but in point of fact this doesn’t seem to matter that much.

This relative offensive rating is then added to the relative pitching rating to produce an overall power rating. It should be said that this particular method probably underrates the actual offensive potential of two ballclubs (Worcester and Arizona) due to the small number of eligible hitters on their roster as of April. These clubs are likely to overperform their collective seeds in BARB: for example, it would not be unsurprising to see 2B Aaron Hill hit .205, but with 30-plus HR even though he hasn’t performed at that level in the majors for a couple of years, simply because (due to his BARB club’s roster size), he would get a very high number of at-bats. So, in the final charts for won-loss percentage, I’ve tweaked these two clubs upward. As a former Commish, I’m troubled by small rosters and know how they can start skewing the entire simulation. If the Arizona GM was inexperienced and didn’t build up his roster (as we would expect RM to do), then they could end up in a situation where their won-loss percentage could plummet below .300, which would lead to some ugly wins for other clubs, and a skewed divisional record for the other clubs.

Speaking of divisional records, how to account for the differences in division, talent-wise? This is actually pretty straight-forward. A divisional average rank can be determined by adding all the power ratings and dividing by the number of teams in each division. The team’s relative standing within the division can be determined by dividing its power rating by the divisional average rank. Then, treating the average as simply “1”, the Pythagorean for won-loss percentage can be approximated by dividing the team’s relative standing number by (the same number, plus “1”):

Similar calculations can be made for the league-wide number for record outside of divisional play, and the overall won-loss percentage for each team can be estimated as the average of these two numbers. This number can be affected by the relative number of divisional and extra-divisional games in the final schedule, but those adjustments being the same for all clubs, they would not be expected to alter the final order of standings….though, in the case of the Central, extra divisional games could significantly drop the contending clubs in the draft order by hiking their overall won-loss percentage. (Hmm. Something to think about when designing the schedule? What a headache!)

So, in conclusion, here’s my best guess for how things will finish. No great detailed drooling over the actual talent, just a snapshot derived by applying a pretty tight set of constraints to independent rankings or talent evaluations, then setting the resulting data sets in relative terms to one another in terms of “BARB strength”. You might wonder how well this works? Well, good enough for me to do it every year, good enough to predict (accurately) that my own club would make the playoffs after having never making it before. That won’t happen for me this year, though (sigh):

Let me end by saying that, while the model predicts outcomes, trades, injuries and sudden decline to aging are harder to factor in. Trades are largely useful to bolster a hand that’s already winning: they are unlikely to turn a sub-.500 club into a contender. On the other hand, catastrophic injuries can turn a contender into a .500 team pretty quick. This is the aspect that any predictive model is going to be lousy at, but the best predictor is the average age of the roster. So, while the model predicts Brooklyn will surmount early-season injuries to Howard and Utley, my hunch is that a younger Sin City roster will peak after August, and become the club no one wants to play.