In today’s tennis landscape, Davis Cup is a weird anachronism, in which no-name players contest five-set epics for national glory. The usual guidelines about the invincibles and the journeymen are set aside, and we can watch a few days of raw, emotional tennis.
That’s the story, anyway. And when John Isner beats Roger Federer on clay (or loses to Thomaz Bellucci on hard), it makes for good copy. As Sam Querrey battled Thiago Alves in Jacksonville last weekend, the USTA’s Tim Curry tweeted, “On paper the @USDavisCupTeam is in good shape. No. 20 Sam Querrey (USA) v No. 141 Thiago Alves (BRA) but @DavisCup is never about the chalk.”
In the end, of course, it was about the chalk. Most of the time, it is. Legendary Davis Cup upsets stand out because of their rarity, not because they define the event.
Quantifying Davis Cup favorites
To determine whether there are a disproportionate number of upsets in Davis Cup, we first need to know what a proportionate number would be. We can get there via two (similar) routes: using a projection system to determine how many upsets there should have been, or comparing Davis Cup results to another group of similar matches.
Since the beginning of 2009, there have been exactly as many Davis Cup upsets as we would have predicted, and almost the same upset rate in Davis Cup matches as in Grand Slam matches from the same time frame.
(I’m including matches contested between top-200 players, and live rubbers from all levels of Davis Cup–though the ranking requirement means we’re mostly looking at World Group and WG Playoffs. Projections are surface-specific and are derived from jrank; I’ve also discarded matches where one player has very few [<10 clay or <30 hard matches] recent results on the surface.)
In these last four-plus years, 352 Davis Cup matches and 1853 Grand Slam matches have fit these parameters. 93, or 26.4%, of the DC matches were upsets, against 474, or 25.6%, of the Slam matches. I’m using surface-specific jrank to define “upset” here, in an attempt to remove surface (and the home team’s ability to choose it) as a confounding variable.
The similarity of those percentages starts to cast some doubt on the “different game” theory of Davis Cup. But it isn’t the whole story.
Raw tallies of upsets don’t tell us how big the upsets were, or how lopsided the average match was. For that, we need more detailed projections.
Projecting the outcome of each one of those 352 Davis Cup matches (using only data that would have been available pre-match) gives us an estimate of 92 upsets. That’s almost identical to the observed total of 93 upsets.
Importantly, the prediction algorithm I’m using here is derived from ATP results. Thus, when we say that the number of Davis Cup upsets is the same as expected, what we’re really saying is that the number of upsets is the same as would be expected if they were five-setters on the ATP tour.
Davis Cup is unusual, it is fun, and it can be thrilling. But it is “about the chalk” no more and no less than your average ATP tour event.