Finally, the Secretary of Defense is home. Though the game ended in the dreaded 0-0 tie, Tim Howard’s debut with the Rapids was surely the biggest storyline of the week. When his signing was originally announced, there was no doubt he would take over immediately as the starter, but it did create a tremor of controversy. The Rapids had just bought Zac MacMath, a young but experienced goalkeeper, after a year-long loan from Philadelphia, and people wondered what might happen if he made a statement performance in the first half of the season. Would Colorado deal him to another team? If he was playing well enough, would it even be wise to make the switch to Howard? Now that the inevitable has arrived, it seems important to consider this point. Are the Rapids really best served by changing their goalkeeper after letting in the fewest goals of any team after 16 games? Or have the defenders been doing the heavy lifting here?
For a brief glimpse into this, I decided to look at how teams’ goals and expected goals (xG) numbers broke down to see if MLS teams facing the Rapids/MacMath found it harder to get good chances. These are rough interpretations, but I’d expect a team’s xG allowed to primarily be a result of their defenders, rather than a goalkeeper, specifically as a result of their ability to stop offensive developments. From there, any systematic deviations in goals allowed from the expectation (xG-G) would be a result of goalkeeper skill. Good goalkeepers stop shots that might otherwise go in, causing teams to underperform their xG.
To keep things comparable, I limited the results I analyzed to only those teams the Rapids played before July 4. Then, I used those teams' games against Colorado as the “treatment” group, compared to their average against teams other than Colorado. My method is often the first test statistic introduced in a statistics course: Student’s t-test. This test provides confidence intervals to compare the means of a variable across two groups.
The results are quite interesting. They fit the narrative that Colorado has had great defensive production in all phases. On average, Colorado's opponents amassed 1.00 xG when playing the Rapids, compared to 1.41 xG against all other teams (p = 0.013). I feel confident saying that the Rapids defensive unit is above average, in that they limit the quantity and quality of opponents’ shots.
MacMath and his defense have allowed 0.69 goals per game, even better than the stellar xG-allowed figure already quoted. That 0.31 xG difference could be attributed to MacMath, which looks good compare to the 0.06 difference* we’ve seen in all games other than MacMath’s. However, that difference is not statistically significant (p = 0.23), and our keeper ratings suggest MacMath is closer to 0.00 than to -0.31. Those ratings take shot placement into account, indicating that opponents have been unable to consistently put the ball out of MacMath’s reach to this point. That is likely a hat tip to Colorado’s stingy defense and maybe some good fortunes, rather than anything the keeper is doing.
Howard put in a couple huge stops against Portland, but it’s not a stretch to say that MacMath could have made them too, nor is it a stretch to say that Howard also benefited from some lucky misses. The point is that no one should be concerned about a decline in the Rapids’ defensive quality with the switch to Howard, largely because the Rapids defense has made sure that few dangerous opportunities ever actually make it to the keeper. One is left to wonder if such a big investment would have been more useful on Colorado’s offensive side of play.
*Because the xG model is built across multiple seasons, the league average in any given year is not likely to be zero since finishing rates fluctuate.