Converting chances: MLS versus the World Cup / by Matthias Kullowatz

By Matthias Kullowatz (@MattyAnselmo)

Throughout the World Cup, we kept shot data here at ASA for all 64 games. When we converted that data into Expected Goals output, we had to use our MLS data from 2013 and 2014 to estimate shot values. At first, I assumed that MLS finishing rates would be a crude estimate for those of the World Cup. However, despite a post-expansion record 171 total goals this past Cup, the finishing rates lined up surprisingly well with those from the United States' first division.

The widest paintbrush shows that overall finishing rates in MLS (10.1% +/- 0.5%) are actually slightly higher than those in the 2014 World Cup (9.8% +/- 1.4%), though not by a statistically significant margin (p-value = 0.66). Our Expected Goals model was a shade on the other side, with World Cup players scoring 166 non-own goals* versus the 159 goals that our MLS model estimates should have been scored. This discrepancy is because World Cup players were forced to take more shots from at least 24 yards away (zone 5), lowering their expected goals output while also lowering their finishing rate.

Location Frequencies
Location MLSlocations WClocations
1 0.054 0.060
2 0.304 0.263
3 0.180 0.174
4 0.209 0.160
5 0.237 0.316
6 0.017 0.027

If we delve deeper, we see that there are no statistically significant differences in any of the six zones, or by headed versus kicked shots. The chart below shows finishing rates (MLSpct and WCpct) broken down by location and body part. P-values are based on two-sample proportions tests.

Finishing Rates Foot
Location Part MLSshots WCshots MLSPct WCpct Pvalue
1 Foot 363 54 0.391 0.407 0.820
1 Head 300 47 0.193 0.234 0.516
2 Foot 2155 241 0.246 0.274 0.342
2 Head 1556 204 0.086 0.093 0.738
3 Foot 2059 274 0.070 0.088 0.303
3 Head 135 20 0.052 0.050 0.972
4 Foot 2520 270 0.052 0.030 0.104
5 2894 534 0.023 0.024 0.867
6 Foot 206 45 0.039 0.000 0.179

MLS players are able to finish at the same rates as World Cup players, but why? I presume that World Cup players are better than those of MLS, but as best we can measure, it doesn't have anything to do with shot placement. An MLS model that accounts for goal mouth placement estimates that World Cup players should have scored 166 goals, which is exactly what they scored and not much more than the original estimate of 159. 

Obviously, there are two sides to a shot: the guy trying to score it and the guy(s) trying to keep it from being scored. What our model doesn't take into account is what's probably the culprit for such similar finishing numbers. We can't control for proximity of the defender---defensive pressure on the shot---or the pace on the shot. MLS games may include a lot more chances like those Germany got against Brazil, mitigating the differences in offensive talent. Additionally, World Cup shooters could drive the ball harder at keepers that are more prepared to stop it, an effect that would again cancel out before we ever saw it in our data.

What I do know for sure is that World Cup teams scored 2.59 offensive goals per game, and MLS teams have scored 2.59 offensive goals per game. If you like goals, MLS is likely to produce just as many.

*There were five own goals during the World Cup that are not included in the shot analysis.