At time of writing, we’ve just come through a tense Leinster final replay. Pushed to a draw in the first match, and tested in the second, Galway finally came away as comfortable 7 point winners in the end. They had generally been considered strong favourites at the outset, but the initial draw caused many to question this idea: was the drawn game just a blip; a bad day that every team has once in a while? Or, was it a sign that Kilkenny had risen to the challenge, and had taken on Galway at their own physical game?
These types of questions tend to emerge every time a significant match goes to replay. One school of thought feels that the favourite will storm out the gate the second time around, as performances revert to the mean, and the more consistent team pulls ahead. They feel that the underdog had their chance to win and blew it, and that another close game will be unlikely.
Others may feel the opposite, and that the momentum now lies with the underdog; they will come out with the confidence that they can take on the favourite and win, that maybe there isn’t as big a gap between them as people thought, and that they have the tools to cut out their opponents advantages.
Determining the favourite, and their favourable odds
A database of replayed matches, from either the All-Ireland or the provincial championships, was gathered. It consisted of 72 games, from the recent 2018 Leinster final, back as far as the 1938 Leinster final, 80 years prior. The BAINISTEOIR rating system was used to determine the expected odds for each team in each of these matches. Though it doesn’t always predict the victor, the confidence with which it predicts tends to be accurate; in other words, if it predicts that one team will win 60% of the time, over enough matches they will finish with a roughly 60% win percentage. The expected winning percentage was modified to assume that the match couldn’t be drawn; if there was more than one replay, only the final replay’s result was taken into account.
Additionally, the expected winning margin was determined for each match. This value is based off the expected winning percentage; the greater the perceived advantage, the more likely they are to win by more points. As with the win percentage, this value might be off considerably for a single game. However, over enough time, it should give the average result. If it doesn’t average out over a large number of matches, it would imply a correlation between a common factor (in this case, who is the favourite in a replay) and the result (whether the favourite scores more or less than expected).
These expected values were compared with the actual results. If the actual performance was considerably less than expected, it could be assumed that the underdog usually overperforms in replays. If the actual performance was significantly greater than expected, it would imply that the favourites had the greater momentum.
The chart below compares the average expected winning odds for a favourite in a replay, and the overall winning percentage across each examined match:
The average favourite had been given a 73% chance to win their replayed game; instead they won 79% of them. Typically in replays, there are no extra factors at play; they are typically played at the same venue, or at another neutral venue, they are played within the same championship, they are played between the same two teams, and they are often played within a week or two of the drawn game, preventing the emergence of too many changes. Though this 6% difference isn’t huge, the lack of other factors which could cause it means that it’s potentially quite significant.
The scoring margin difference initially provided a more confusing picture:
Despite winning more games, the favourites tended to win by slightly less than expected. They still had a positive average margin, typically winning the replay by 3 points, but it was very slightly less than expected. Perhaps the underdog wasn’t good enough to beat them, but the scoreline in both the drawn game and the replay was proof that they had been underrated?
However, looking at the median values instead of the average give a bit of a different picture:
The average expectation had been skewed slightly upwards by the occasional very heavy favourite, but the main issue was that the actual margins had been skewed downwards by the occasional heavy loss: a team may win by a point more than expected 20 times, but a single loss by 25 instantly negates this when calculating the average. The difference isn’t very large, even using the median, but thus far the favourite has typically won by a larger margin than expected. Across the list of replays, the favourite had a better scoring margin than expected in about 53% of all matches. They also won by a wider margin than expected in 65% of the matches they won, which does appear more significant.
Still, over a large enough number of matches, the average should approach the median, but this hasn’t quite happened. This brings us to one slightly tangential, but very important point:
The frequency of replays
The evidence above shows that the favourite tends to win more replays than expected; it supports the theory that the drawn game happens because one or both of the teams were playing at a different level to their usual standard, rather than the underdog being underrated. However, it’s very important to remember that this is a very small sample size.
In the last 50 years, the database this website uses lists 58 replays in the All-Ireland, or one of the provincial championships leading into it; this is only about 5% of the total All-Ireland matches played in this same span of time. The following table shows how many replays you could expect in a given year’s championship, based on a Poisson distribution:
Based on the above, you’re more likely to have zero replays in a championship than to have more than one! The above odds obviously change year to year, as the format changes, but the point remains that draws are rare, and draws in knockout matches that can’t go to extra time are rarer still.
All this to say; it’s hard to make a definite conclusion. So far we have seen the favourites overperform. However, we have also had many wonderful underdog performances succeed. The data could still swing the other way, with only a handful of outlier results, but replays will likely remain rare enough that the results won’t settle very strongly on side or the other. If your team is in a replay, and they’re the favourites, don’t get complacent. Similarly, if they’re the underdog, don’t lose faith that they’ve missed their chance. Based on everything that we’ve seen so far, it’s tougher for the underdog to win a replay than a regular match. However, the correlation is still small enough that it’s far from a sure thing, and there will surely be many teams in the future who’ll be eager to go against the existing trend.