The No-Hitter Hangover?
On the first of June, left-hander Johan Santana labored through 134 pitches and meticulously navigated around five walks to become the first player in New York Mets history to throw a no-hitter.
It was assuredly a special moment for the organization, as well as the entire fan base. The Mets’ manager, Terry Collins, understood the magnitude of the situation. Despite the fact that Santana missed all of 2011 with a shoulder injury and had largely been limited in his pitch counts throughout the year to that point, Collins stuck with his 33-year-old veteran in an attempt to rewrite the history books.
To counteract the extra strain put on the shoulder in that no-hitter, Johan Santana received extra rest before his next start. That decision caused the left-hander to develop rust, according to his manager, and Santana was not sharp against the New York Yankees his next start — he surrendered six runs over five innings, including four home runs.
Collins blamed the extra rest. After also watching White Sox right-hander Philip Humber implode in the start after his perfect game this year, however, I began to wonder if pitchers performed abnormally poorly in their first start following a no-hitter. After Matt Cain posted his worst start of 2012 last night in his follow-up to his perfect game, my suspicions only grew, and I decided to take a closer look at whether these pitchers are part of a broader trend
The hypothesis relied on the abnormal schedule between starts for pitchers who throw a no-hitter. The media demands become more intense. The sheer attention doted upon pitchers who throw no-hitters puts more pressure upon them to succeed, or even to replicate Johnny Vander meer’s back-to-back no-hitters in 1938. Perhaps those factors ultimately combine to adversely affect performance immediately after the emotionally taxing no-hitter.
Though carefully dismissing them as factors in his dreadful performance against the Red Sox in his first start following his perfect game, Philip Humber brought up the distractions – such as appearing on Letterman and speaking with the President of the United States — that ultimately arise.
“I did my best to try to eliminate the distractions and I felt like I was focused coming into the game. I don’t think anything that happened Saturday affected tonight. It was just a bad day. I’ll just chalk it up to one of those days.”
Since the live ball era, there have been 170 no-hitters thrown by a single pitcher, but of course many of those occurred before the invention of the 24 hour news cycle. In an attempt to ascertain if any discernable trend in performance exists under today’s conditions, I chose to isolate the regular season no-hitters over the past ten seasons and pull data from the subsequent starts, comparing their ERA and FIP from that specific start to their averages from that season to determine whether or not each individual pitcher performed better or worse than their own average from that season.
| Pitcher | Year | IP | ER | HR | BB | K | ERA | Season ERA | Difference |
|---|---|---|---|---|---|---|---|---|---|
| Carlos Zambrano | 2008 | 1.7 | 8 | 1 | 3 | 1 | 43.11 | 3.91 | 39.20 |
| Philip Humber | 2012 | 5.0 | 9 | 3 | 3 | 5 | 16.20 | 5.92 | 10.28 |
| Johan Santana | 2012 | 5.0 | 6 | 4 | 1 | 5 | 10.80 | 2.96 | 7.84 |
| Francisco Liriano | 2011 | 3.0 | 4 | 1 | 3 | 1 | 12.00 | 5.09 | 6.91 |
| Mark Buehrle | 2009 | 6.3 | 5 | 1 | 3 | 7.11 | 3.84 | 3.27 | |
| Matt Cain | 2012 | 5.0 | 3 | 1 | 4 | 4 | 5.40 | 2.34 | 3.06 |
| Edwin Jackson | 2010 | 5.0 | 4 | 3 | 4 | 7.20 | 4.47 | 2.73 | |
| Jon Lester | 2008 | 5.0 | 3 | 2 | 3 | 5.40 | 3.21 | 2.19 | |
| Matt Garza | 2010 | 7.0 | 4 | 2 | 1 | 9 | 5.14 | 3.91 | 1.23 |
| Anibal Sanchez | 2006 | 7.0 | 3 | 2 | 1 | 8 | 3.86 | 2.83 | 1.03 |
| Dallas Braden | 2010 | 8.0 | 4 | 1 | 1 | 5 | 4.50 | 3.50 | 1.00 |
| Justin Verlander | 2007 | 6.0 | 3 | 2 | 6 | 4.50 | 3.66 | 0.84 | |
| Kevin Millwood | 2003 | 6.0 | 3 | 2 | 2 | 7 | 4.50 | 4.01 | 0.49 |
| Clay Buchholz | 2007 | 4.7 | 1 | 2 | 5 | 1.93 | 1.59 | 0.34 | |
| Jonathan Sanchez | 2009 | 6.0 | 3 | 2 | 1 | 8 | 4.50 | 4.24 | 0.26 |
| Mark Buehrle | 2007 | 7.0 | 3 | 1 | 1 | 4 | 3.86 | 3.63 | 0.23 |
| Roy Halladay | 2010 | 7.0 | 2 | 1 | 7 | 2.57 | 2.44 | 0.13 | |
| Randy Johnson | 2004 | 7.0 | 2 | 1 | 1 | 5 | 2.57 | 2.60 | (0.03) |
| Jered Weaver | 2012 | 6.0 | 1 | 2 | 2 | 1.50 | 2.61 | (1.11) | |
| Justin Verlander | 2011 | 8.0 | 1 | 3 | 7 | 1.13 | 2.69 | (1.57) | |
| Ervin Santana | 2011 | 9.0 | 1 | 2 | 7 | 1.00 | 3.38 | (2.38) | |
| Ubaldo Jimenez | 2010 | 7.3 | 2 | 5 | - | 2.88 | (2.88) | ||
| Total | 132.0 | 73 | 21 | 42 | 111 | 4.98 | 3.44 | 1.54 |
Seventeen of the 22 pitchers on this list posted a higher ERA in their start after their no-hitter than their season average, and 12 of those 17 were at least one full run higher than their own average for that year. As a group, these pitchers posted a 4.98 ERA, more than 1.5 runs per nine innings above their overall average of 3.44 for the seasons in which we’re evaluating them. There’s no question that the observed phenomenon of late is that Humber, Santana, and Cain are the norm and not the exception. Of course, we’re dealing with small samples and ERA isn’t a great way to evaluate whether they actually pitched worse, but using FIP brings you to a similar conclusion, as they posted a 4.44 FIP in their 132 innings of work, still quite a bit higher than their 3.70 season average. And remember, this is a sample of mostly good pitchers, so while regression to the mean is expected, this is regression well past their own mean. This is regression to Tommy Hunter‘s mean.
Twenty-two pitchers and 132 innings is not enough to say with certainty that there’s a real effect here, but going back further in time is problematic because of the changes in landscape of the sports world in recent years. If we’re hypothesizing that the distractions of the day are a factor in the pitcher’s struggles, we can’t go back to a time before those distractions existed or were as prevalent.
So, instead of expanding our sample, we’re going to dig a bit deeper into the quality of pitches thrown by guys in their after-no-hitter starts, and see if we can’t find a explanation in their stuff for their decline in performance. Jack Moore will be publishing a follow-up on using PITCH F/x data this afternoon, and see whether pitcher’s are throwing their usual stuff in their after-no-hitter start, or whether the combined workload of going nine innings and then dealing with an adjusted routine might be manifesting itself in what they have to throw in their next start. This data isn’t conclusive, but it is interesting enough to warrant further study. And at least Cain, Santana, and Humber can take solace in the fact that they’re certainly not alone in experiencing a big letdown.
Interesting stuff, JP, thanks.
Do you not need to strip out the no-hitter start in calculating the season averages for these pitchers?
Yes. If this were me doing the analysis, I’d additionally leave out the stats from the start after the no-hitter, so that the “season ERA” is independent from the stat we are comparing it to.
By yes I meant no…No, you do not not need to do this. Yes, you should do this.
Yes, you should probably do that too. A rough-and-ready calculation would be that the no-hitter decreases season ERA by about 0.2, and FIP by a bit less [Matt Cain excepted!]
“whether the combined workload of going nine innings and then dealing with an adjusted routine might be manifesting itself in what they have to throw in their next start.”
This is what I would have guessed.
I’d be interested in finding this out using a metric that isn’t beyond stupid.
Good stuff, J.P.
A couple months back, after Humber’s game, I took sort of a similar look at the follow up to each of the perfect games in history and found results pretty similar to what you’ve found with recent no hitters. It may (or may not) be of any interest:
http://southsideshowdown.com/2012/04/26/after-perfection/
I don’t think this was addressed but it would be interesting to see how many pitches each of those pitchers threw and if that has any correlation with the performance of the next start.
While I agree that the media distractions are a factor, i would think the unusual workload has more of an impact than anything else.
22 pitchers is a pretty good sample size to make some claims, considering it’s a “within subjects” analysis rather than a “between subjects” design. That buys you a lot of statistical power, typically.
Because you are trying to analyze the residual “hangover” associated specifically with no-hitters, however, a good comparison would be to how pitchers perform the next start after a CG shutout. In this case, starters will be prone to the same regression (ERA wise) in the following start, and are coming off the same innings load (though probably fewer pitchers, since pitchers like Santana will be pushed to a higher pitch count to chase a no-no than a CG, which is typically a function of reasonable pitch count entering the ninth). But the additional stressors and fallout associated with the no-hitter are absent.
So what you’re saying is RA Dickey is smarter than everyone by just throwing one hitters all the time.
And Lenard with the win!
This may just be a personal preference, but would median analysis be appropriate here? You point out that 17 of the 22 are above the average. However, if the distribution of ERA across starts had a long leftward tail, that would be exactly what you might expect
Of course, since ERA is bounded from below, but not from above, I suspect the median is below the mean, so you may be understating how badly (from a distributional perspective) pitchers are performing following their gems.
This is what I was thinking. Looking at that list, you see a few people who really crashed and burned the next time around, but by and large most people were really close to where they were before (11 observations were within 2 of their previous effort). Another issue: I’ll bet a lot of no-hitters happen against sub-par teams, and just by regression the competition in the pitcher’s next start is likely to be more difficult. These games are outliers for a number of reasons, and regression is expected; the question is whether the regression is more than what would be expected or not.
The fact that a no-hitter may have occurred against a poor team doesn’t really matter. The comparison here is season performance to the game after the no-hitter performance. Regression to the mean would simply result in roughly equal season performance and game after performance, not a 1.5 increase in ERA for the game after performance compared to season performance.
Maybe you could isolate the media effect by also running this sort of analysis on pitchers throwing one-hitters. because obviously R.A. Dickey had a pretty good outing after his first one-hitter of the year. And one-hitters aren’t substantially better than no-hitters, usually just luckier.
I would think that looking at how pitchers preformed after a no-hitter from years past would give you a control, to determine whether the increased media presence is having an effect. If in years past there was a similar rise in ERA, than something else must be going on (extra work in the no-hitter, for example). However, if you do not see a similar drop, than your hypothesis should followed up on.
If your hypothesis is that the modern media contributes to the decrease in performance, wouldn’t expanding your sample be exactly what you *do* want to do? Or, more specifically, wouldn’t you want to compare, say, these perfect starters to a similarly sized batch of pre-modern media perfect starters?
That wouldn’t answer the question once and for all, of course, but it would fill out the picture at least.
Just shows how awesome it was for Johnny Vander Meer to throw two in a row.
It also goes to show how awesome Ervin Santana can be when he’s not absolutely atrocious.
On scout.com, I looked at the last 10 years [20 no-hitters] and compared the four starts before the no-hitter to the four starts afterward;
The average before: 24.5 innings 10.3 ER
The average after: 25.3 innings 11.3 ER
I toked solace once. Not in ‘Nam of course.
What’s the p-value of the wilcoxon signed rank test for the fip data?