Wednesday, July 06, 2011

Understanding Statistics - IP, K/9, ERA

One of the most misunderstood statistics is IP, "innings pitched", for pitchers. This misunderstanding then extends to a variety of other statistics which are based on it including K/9 (strikeouts per nine innings) and ERA (earned run average)

Formulas: IP=Outs/3 , K/9=K/(Outs/27), ERA=ER/(Outs/27)

Explanation: IP, "innings pitched", is a measure of how many outs were achieved while a pitcher is on the mound. Because there are three outs in an inning, the number of outs is divided by three.

Of course, an actual "inning" starts with no outs and ends with the third out. This is not the case with "innings pitched". A pitcher may have pitched in multiple innings and still have one or fewer IP.

K/9 measures the proportion of outs made by strikeout.

ERA measures the number of earned runs allowed proportional to the outs achieved while the pitcher is on the mound.

Statistical Artifacts:

Innings Pitched (IP) is a pretty straight forward. Its only important to remember that a double play, pick off, outfield assist etc. can all produce outs with little or no real pitching involved. Its theoretically possible to have an "inning pitched" without throwing a single pitch.

With K/9, the fewer outs a team makes, the higher a pitcher's K/9 will be for the same number of strikeouts. In other words, there are two ways to increase a K/9, one is more strikeouts, the other is making fewer outs.

ERA uses "earned runs". I am not going to spend a lot of time explaining the details of earned runs. But it is important to understand that once an error would have made the third out in an inning, the rest of the runs are "unearned" regardless of how they score. The second oddity is that, while outs are attributed to the IP of the pitcher who is on the mound when the out was made, runs are attributed to the pitcher who put the runner on base.

Why we should care:

IP is actually a pretty good measure of the quality of a starter. Obviously the goal of a pitcher is to get outs and IP measures that. Moreover, a team will need to get a certain number of outs in each game and season. That number is relatively fixed, varied from team to team only by extra inning games and home/road records. So a relief pitcher will need to get every out a starter doesn't get.

K/9 provides a general idea of how heavily a pitcher relies on strikeouts. To the extent there are productive outs for hitters, those outs are unproductive for the pitcher. So there are situations where a strikeout is better than an out. If two pitchers have the same number of outs, the one who gets more of them by strikeout is likely going to be more successful.

ERA is generally useful for evaluating starting pitchers. The better pitchers tend to have better ERA's. For relief pitchers it is a lot less reliable, both because of the statistical artifacts and sample size issues.

Why we shouldn't care:

Comparisons of innings pitched stop being very meaningful with less than about 160 IP. It is hard to pitch more than that without being somewhat successful, although obviously that success. But below 160, what happened while the pitcher got those outs becomes more and more important.

The central goal of a pitcher is to get batters out. How they do that doesn't matter as much as how often. K/9 measures the how, not the how often.

ERA is one of those statistics that means something in context. But single season ERA's are easily distorted by outlier components. Consider a season where a pitcher pitcher has 180 IP. That is 20 games of 9 IP. A single outing where a pitcher gives up 10 earned runs will raise their season's ERA by .50.

How its used or misused:

What is important to realize is that IP is no more a measure of how much a pitcher pitched, than the number of hits or walks. It is a way of measuring a specific result, the number of outs, while they are on the mound. The number of outs, of course, depends on the quality of the pitcher's fielders and the ballpark, as well as the pitcher's own contributions.

To understand how k/9 can be misused, take this blog comment complaining about the Twins closer:

"It's (Twins closer Matt Capp's) 5.3 K/9IP rate on the season. Consistent success is almost unachievable for a reliever allowing that much contact."

In fact, K/9 does not measure how much contact a reliever allows. It ignores hits entirely. This is probably partially the BABIP myth I discussed in an earlier post. But it is also misunderstanding of k/9. A simple example.

Two pitchers both face six batters:

Pitcher1 strikes out 2, gets a ground out and gives up three hits.

Pitcher2 two strikes out 3 and gets three ground outs.

Pitcher1's k/9 is 18

Pitcher2's k/9 is 13.5

Pitcher2 had fewer balls in play and struck out a higher percentage of batters faced but has a lower k/9.

No comments:

MLB Twins Updates