GB% and G/F Rate

If you remember back to the stone age, before FanGraphs existed and we had a veritable cornucopia of batted ball stats at our disposal, you may recall a time when pitchers were judged by their ratio of ground balls to fly balls. G/F rate was often used as a tool to describe a pitcher’s type of batted balls allowed. In general, it works fairly well. At some extremes, however, it breaks down.

Take Lenny DiNardo, for instance. He returned to the majors last week, being called up by Kansas City to fill a hole in their rotation down the stretch. DiNardo’s primary skill has always been an ability to rack up groundballs, and his first start of 2009 was no exception. In fact, if we looked at his 8.0 G/F rate, we would think that hitters were pounding the ball into the ground all day.

They weren’t, however. There’s an additional batted ball type – line drives, and DiNardo gave up six of them. When you include those in the balls in play denominator, his GB% is 53.3%, which highlights the fact that he got a majority of groundballs but doesn’t suggest the same crazy performance than an 8.0 G/F rate implies.

Now, this is obviously a tiny sample, and the correlation between G/F rate and GB% is very high. 95% of the time, either one will give you the same answer. But it’s that 5% that G/F rate may lead you astray. Let’s use a slightly more realistic scenario, also from this year.

Who has been more of a groundball pitcher this year – Jason Hammel or Brett Anderson? Hammel has the higher G/F rate, 1.46 to 1.36, though the difference is small enough that you might just conclude that they’ve been basically the same. However, when looking at their overall batted ball profile, Anderson has a 48.8% GB% and Hammel has a 45.5% GB%, which is a big enough gap to say that Anderson has clearly been better at getting ground balls.

Their line drive rates (23.4% for Hammel and 15.4% for Anderson), not included in the G/F measurement, hide the fact that Anderson has been more of a groundball guy than Hammel. In fact, by ignoring the line drives that Hammel is giving up, it actually gives an incorrect answer to the question.

In practical terms, this is more of a current issue with minor leaguers, where we don’t have the same quality of batted ball data, and a pitcher’s G/F rate is still quoted with some regularity. Just think of this as a word of caution – while G/F rates work when line drive rates are equal, line drives are not always equal.

Print This Post

Dave is the Managing Editor of FanGraphs.

11 Responses to “GB% and G/F Rate”

You can follow any responses to this entry through the RSS 2.0 feed.
  1. Good point to bring up, I’ve been wondering about this for a long time now, why LD is left out.

    Vote -1 Vote +1

  2. Andrew says:

    Thought you would enjoy this, Dave.

    Keep up the great work. You’re one of the best in the business.

    Vote -1 Vote +1

  3. Kincaid says:

    What if you don’t just want to know who actually gave up more ground balls, though, but who has a better talent for inducing ground balls? Over yearly samples, at least, most of the variation in LD rates from year to year comes from something other than the observed value in the previous season. Just because Anderson has only allowed 15.4% line drives and Hammel has allowed 23.4% line drives doesn’t mean that Anderson is that much better at preventing line drives, nor that we should expect Anderson to allow that many fewer LDs going forward than Hammel.

    We would still assume both will probably have close to average LD rates going forward, so when we have widely differening LD rates, should we really prefer the stat that is influenced by extreme LD rates (GB%) to the one that assumes an average LD rate (GB/FB ratio)? I think it depends on what we want to know. For measuring who actually induced more ground balls in a sample, GB% works better when the LD% is extreme, but for estimating who is better at inducing ground balls or trying to project future performance, that’s not necessarily true.

    Isn’t the quality of batted ball data in the minors a separate issue? If you have batted ball types at all, you can calculate GB% just as easily as GB/FB (they list GB% at MinorLeagueSplits, so it’s not like the data isn’t there or is hard to find), and if the data is just not as reliable, it will affect both figures. GB/FB should be about as good for minor leaguers anyway since we care more about how they project in the future than how they did in the sample.

    Vote -1 Vote +1

  4. Xeifrank says:

    So what’s your solution? GB / (GB+FB+LD)?
    vr, Xei

    Vote -1 Vote +1

    • Bill says:

      Why not GB / (FB+LD)?

      Should get the noise out of that.

      Vote -1 Vote +1

    • Kincaid says:

      What is whose solution for what? GB/(GB+FB+LD) is just GB%.

      GB/(FB+LD) is the same thing as GB%, just turned into a ratio instead of a rate. If you let GB/(FB+LD)=x, then GB/(GB+FB+LD) will always equal x/(1+x). So you don’t get any information out of one that you don’t get out of the other.

      Vote -1 Vote +1

      • Xeifrank says:

        If you are going to criticize a stat with offering a solution… then what’s the point? That was where my original question came from. If GB/FB is a poor stat, then what should be used in its place?
        vr, Xei

        Vote -1 Vote +1

      • Kincaid says:

        I would guess the solution Dave is offering as an alternative to GB/FB is the one that he spends the whole article talking about using instead of GB/FB.

        Vote -1 Vote +1

  5. Al Dimond says:

    The article’s claim is that GB% (which is the same as GB/(GB+FB+LD)) is a better measurement than GB/FB. So that’s exactly what the article is suggesting, use GB% when it’s available. When it’s not… you can’t factor in line drives unless there’s a source for that data, so you can’t find GB%.

    I don’t know exactly what’s available from the minor leagues, but if you look at historical major-league data you don’t even get GB/FB. You get GO/AO. And counts LD as FB in their GB/FB (Fangraphs doesn’t), so b-r is actually listing GB/(FB+LD) in Fangraphs’ terms. GB/(FB+LD) seems to be usually lower than GO/AO (I guess FB+LD have higher BABIP than GB?) but GB/FB is lower (presumably because few LDs are caught for outs and GB have higher BABIP than FB).

    My guess is that if a data source gives no batted-ball data but GB/FB they’re going to count LD in their FB, but that’s just a guess based on the, “well, we have to put these balls in some bucket,” concept.

    Vote -1 Vote +1

  6. Paul Thomas says:

    Isn’t LD rate for pitchers essentially random? I was under the impression that it was one of the major causes of year-to-year BABIP fluctuations.

    Vote -1 Vote +1

  7. John Thomas says:

    Kincaid and Paul have excellent points.

    That Hammel and Anderson have different LD%’s doesn’t mean that one is better than the other at preventing preventing line drives. Nor does it mean that one or the other is better at getting ground balls.
    The in LD%, which amounts to 17 LDs for Hammel, could easilly be explained by random chance.

    Previous studies have concluded that the variation in LD%’s for pitchers could easilly be explained by random chance, wheras the difference in GB/FB rates could not be. The example above does nothing to prove otherwise.

    If you really were interested in proving that GB% is a better indicator of the skill of inducing ground balls, then you would have to show that LD% is influenced by something other than random chance. You’ve yet to do so, which means that this entire article is completely falwed.

    Vote -1 Vote +1