Rather than recap this last week’s game, which went well generally for the Seattle Seahawks kicking unit, I decided to cover an age old question: Does icing the kicker actually work?
To answer this question I will be using the NFL player database, available through R here, for the years 2009 to 2016. Why these years? The R scraping tool can only reach back to 2009, and rather than pull from a still changing 2017 season I am using only complete season sets. Each of these sets contains over 40,000 individual NFL plays per season. From 2009 to 2016 there were 3,434 field goal attempts. In order to determine if a kick was iced I looked at the previous play and indicated “Iced” if that previous play was a timeout. “But wait!” I hear you saying, “you can’t ice your own kicker!” To which I say:
While it would surely be possible to go back and review all 956 possible iced kicks to make sure they were truly iced and not simply timeouts taken before lining up, this is a blog post for Field Gulls, not a dissertation. Thus, I beg the reader for some leeway and to consider such incidences as random noise of the sample.
The next step was going through each distance of a field goal and determining if we have sufficient sample size. As an illustrative example, here is a histogram for all field goal attempts from 2009 to 2016 broken down by made, missed, and blocked.
We can see in this example that the overall number of field goal attempts is drastically reduced beyond 53 yards, with the sample becoming unusable after about 55 yards. Additionally, at about 55 yards is where we start seeing our cross over of a miss or block being about as equally likely as a good attempt. For this reason, I limited our search of field goals from 18 to 55 yards. Outside of this data window, there simply isn’t the robust sample size needed to make good comparisons.
Overall, with allowance for some variance, the trend from 2009 to 2016 seems to be that icing the kicker produces vary similar percentages of field goals made. According to the data, an iced kick was good 82.1% of the time versus 84.9% of the time normally. While there is a slight difference, it certainly isn’t much. Next, I applied a Welch two sample t-test to formally test whether this was evidence to suggest rejecting the null hypothesis; the null hypothesis is that our two sample averages are equal. Not too surprisingly, this test did not find significant evidence to reject the null hypothesis, in fact our p-value here was .41. However, There does seem to be a larger difference if we look specifically at the long range region beyond 45 yards.
Testing this region, and this region alone, we still don’t get statistically significant evidence to suggest rejecting the hypothesis that icing the kicker has no effect. However, at a p-value of .11 it is much closer. Even with this result, the methodology becomes a little dubious, we are narrowing in on a specific section we think we see a difference and then testing if a difference exists. With that said, there’s nothing explicitly wrong with doing so. We are simply defining the population we are interested in differently. I.E. we narrow from a population of all field goals, to those attempted from 45 to 55 yards.
Out of curiosity I ran this test one more time, this time only looking at field goals from 50 to 55 yards. Now, this is a pretty narrow sample of a very particular range. But there does appear to be a statistically significant difference (p-value below .05) between the average accuracy of an iced and non-iced kicker between 50 and 55 yards.
The sample size beyond 50 yards for iced kickers is fairly small, so I am hesitant to declare icing works at this distance. It is far more likely that the reduced sample size has lead to increased variance as compared with the overall population. The conclusion here reaffirms the conclusion found in earlier studies, there is not statistically significant evidence that icing the kicker changes their accuracy.
Even when you may or may not have tried to ice the kicker at the end of a divisional round game and it may or may not have backfired: