He Got Skills

Last time I left us some questions. The first, and most fundamental question is; just what is a skill? We think we can recognize skills in ballplayers (and people) and sometimes we are right. In baseball though, with its panoply of statistics, we can be easily confused. A statistic is not a skill. The two can sometimes be related, but there are stats that should not be confused with a skill, especially some particular stats. Many of the so-called sabremetric or new baseball stats are more closely related to actual baseball skills, because incorporated in them is the notion of skill we are looking for.

Quite simply, a skill is an ability to complete a task, and complete it the same way in a highly predictable fashion, due to the fact that the person who possesses it has the right set of characteristics, mental and/or physical that allow for this. It should be also noted that a skill is an ability to complete a task repeatedly in circumstances that are fair to all of the many different possible outcomes of the task, and do it at at level beyond that which you would expect beyond random chance. So, for instance, it is not a skill to be able to flip a coin for a head and then a tail; this you would expect from random chance. It is not even a skill to flip a coin six, seven or eight times in a row for a head. This too could be just random chance, or it could represent a loaded coin (not fair to the two outcomes that are possible). Beyond this, at ten or twelve times in a row, you would have to expect that either the coin was loaded, or the person flipping it had some sort of skill to complete the task repeatedly, because beyond 8 or 10 or 12 flips in a row all heads (assuming no cheating), it is highly statistically unlikely that it is just chance determining the outcome.

I won't go in to the math here, but this would qualify as a skill, if the coin flipper could do this with regularity (always able to flip 8, 10 or more heads in a row, most of the times he was asked to do this task). In a fair environment, being able repeatedly to complete a task at a predictable level of ability beyond the level of probability of success that can be explained by random chance, qualifies as a skill.

OK then, with this definition of skill in hand, let's venture out into the world of ballplayers. There are a shocking array of statistics about baseball players all vying for your attention. Which ones should we pay attention to? Which ones should we pay the most attention to? Which ones are pretty much useless and should be ignored. This is where our new idea of a "skill" comes in. One thing I neglected to mention above about skills is that in order to be useful to us in evaluating baseball players, these skills must be measurable in some way. The simplest form of measurement is counting ("how many times did he do "X"?); the next simplest is rate measurement ("how many times did he do "X" for every opportunity, or for every time "Y" happened)?

Counting is easy, but unfair, as it credits those that have more opportunities than those who have less opportunities. Rate measurement is more fair, and levels the playing field, but in a game whose goal is not to perform at a certain rate, but to simply outperform your opponent's runs scored, we need to look for those players who actually perform at a high rate over many many opportunities (after all, a player who is injured or sick or suspended for bad behaviour for 5 days of the week, but performs at a high rate in some skill, is not, at the end of the week, worth as much to his team as the player who performs at a somewhat lower rate, but never misses an inning).

These are the two basic forms that most baseball statistics take: counting statistics, and rate statistics. Depending on context, each kind is important. But how do we know which statistics are most closely related to skills? In general, rate statistics are easier to evaluate as valid measurements of skills, because they remove the unfair advantageous influence that increased opportunity can introduce to counting statistics. Without going into the math though, it is very important when using rate stats as a means of determining skill to ensure that the number of opportunities used to measure the rate (the "Y" above) is sufficiently high to guarantee that you aren't just observing random chance or luck at work. Remember the coin flip. Two or three heads in a row is not anything unusual. However, 11 or 12 heads out of 12 flips is statistically significant, and is very highly unlikely to be due to just random chance or luck.

With this in mind, what we are looking for are statistics that reflect skills, that is, those statistics that, for a given player, do not vary dramatically from year to year (the player can repeatedly complete a task in a consistent fashion). Of course players get injured, and often play injured, and over time a player's skills can improve as he matures; and, as he ages, decline, but these two factors can be accounted for accurately when one is using a good statistic, one that actually measures a skill. I can't stress this enough: the best statistics to use to evaluate a player's skills are those that do not vary considerably from year to year, or those that show a predictable increase as the player grows and gets more experience, and then decline as he ages. These statistics are the ones that are the most reflective of a player's skills, because, we are safe to assume, the skill of a player is fairly constant for several years, so well thought-out and defined statistics should reflect this, and remain fairly constant, allowing us to compare players in a fair manner.

The flip-side of this of course is that we should be suspicious of statistics that swing widely in value for a given player from season to season, as these fluctuations are likely to reflect the influence of factors other than a player's skills. These highly variable stats could be showing the effect luck has on the outcome of his performance, or the influence of the skills of the other players on the team he plays on. These stats are of limited use to us in evaluating players and their skills and contribution to winning (and therefore each's objective value as a player). Of course it's always important to remember that we are looking for skills and stats that are related to the scoring (or prevention of scoring) of runs, as runs are what contributes to winning. It's great if a pitcher can consistently throw the ball at 102 mph (and therefore has this as a skill), but only if this can be shown somehow to contribute to winning ball games.

NEXT TIME: Which stats are good ones and which common stats are poor ones if our aim is to evaluate a player's skills and contribution to winning ball games? Remember, we are going to stick to the player's we can vote for on the All-Star ballot, so we won't look at pitchers stats ... well ... OK maybe this one time. Your homework is to answer this: which stat do you think better represents a pitcher's skills, ERA (Earned Run Average) or K/9 (Strikeouts per nine innings pitched)? For batters which is a better measure of a skill over two or three seasons; a player's RBIs (Runs Batted In), his AVG (batting average) or his K/PA (strikeouts per plate appearance). I'll give you a hint ... go look up the numbers for some of your favourite players and see which of them vary the least from season to season.


Until next time ... James