The anatomy of an upset

The first round at the Australian Open served up a number of upsets that have set the tennis world aflutter. After just two days of competition at Melbourne Park, five of the men’s seeds and 12 of the women’s seeds had been been sent home. And not just any seeds…

By the end of Day 2, two of the top contenders for the title were out: No.5 seed Rafael Nadal and No.2 seed Simona Halep. Nadal lost to countryman Fernando Verdasco in what was their fourth meeting at a Grand Slam and Halep succumbed to qualifier Shuai Zhang of China, who registered her first main draw Grand Slam win in 15 attempts.

When huge upsets like these occur, our natural reaction is to search for explanations. Was the underdog simply playing out of their mind? Did the top seed not have enough tournament play after the off-season, or was this just an off day?

To better understand how upsets occur, the Australian Open data analytics team looked for patterns of performance in all upsets that occurred in the 2015 Grand Slams. A match was considered an upset when a seeded player lost to an unseeded player, an event that happened 46 times in men’s matches and 61 times in women’s matches at the 2015 majors.

Men’s upsets
When an underdog defeats a top player, does he or she have to raise their game considerably to do so? To answer this, we compared the performance of the upset winner – or the underdog – to their average performance on the same surface.

A number of serve and return performance stats were significantly better than expected when the underdog achieved a surprise win over a seeded player (see table below). The biggest improvement was found for break-point conversion, where underdogs raised the level of their conversion rate by 26 per cent. Performance on second serve, whether on serve or return, also had a notable boost in upsets for the unseeded player, each increasing by more than five per cent above the player’s personal average.

When we look on the other side of the net at the loser of the upset, we found the opposite. Where underdogs played better than their personal average in upsets, seeded players played much worse than their normal level. The greatest drop in performance was found for break point conversion, where seeded players converted 38 per cent fewer points than usual.

Interestingly, while aces, double faults, and first-serve win percentage are some of the most popular stats discussed in tennis, none of these skills were found to be related to an underdog‘s win over a seed.

Women’s upsets
Although the characteristics of performance differ between men and women (with the serve typically being more influential in the men’s game, etc), the patterns defining an upset were nearly identical (see table below), suggesting that there is something universal about what makes an upset an upset.

Like upsets in the men’s game, the women who pull off an upset raised their game most strongly in break points converted, and performance on second serve points.

Seeds were particularly off their normal game on the return, converting 42 per cent fewer break points and winning 13 per cent fewer return points than their personal average.

Summary
After beating Nadal at the 2016 Australian Open, Verdasco said he played “unbelievable” that afternoon. If the Nadal-Verdasco match was anything like upsets in the recent past, Verdasco got half of the equation right. A huge upset win will usually take more than an incredible day for the unseeded player; it also needs a less-than-unbelievable day for the seed.

Presented by the Australian Open analytics team in conjunction with Victoria University