Week in week out

It seems that, according to most accounts, the new All-Ireland format is very popular. A massive increase in the number of competitive matches, meaningful draws, and the addition of home matches for (nearly) every team have all been positive outcomes of this round robin structure. However, while it seems to be an improvement over the old system, the new format is not without its flaws.

There is imbalance between Leinster and Munster, as relegation to the Joe McDonagh tier only looms for Munster teams if Kerry suddenly become a dominant hurling power. While Kerry are safe from relegation this year, and they managed to win a pre-season game against Cork last winter (Kerry’s first win over them since their All-Ireland win in 1891) they are also at zero risk of reaching the 2018 Joe McDonagh final, so this seems unlikely for the time being.

Another issue is the imbalance with the home venues; Walsh Park in Waterford was deemed unfit for championship matches, and so Waterford have been forced to play their ‘home’ games at neutral grounds. Additionally, due to unusual rules from the GAA, these neutral grounds must be in Munster; instead of having a quick drive up to Nowlan Park for their match against Tipperary, they instead had to play in the Gaelic Grounds; this stadium is only about 20 minutes from the Tipperary border, whereas you have to tack on at least an extra hour to reach county Waterford.

The main issue, however, has been the imbalance in the frequency of matches. All teams in Leinster and Munster must play 4 matches in 5 weeks. However, there are four teams that must play these in consecutive weekends, with no gap in between, and another four who will have 3 matches in a row. This can be a big disadvantage, in an already tough schedule, as teams don’t have the same amount of time to recover and prepare. The question is, however: how large is this disadvantage, and what can be done about it?

Wins over expectation

It’s already been pointed out that, at time of writing, no team has won a match in their third or fourth week in a row; the current record is four losses and two draws. However, it’s important to consider that the losing team was not the favourite coming into this fixture; It seems likely that Offaly would still have lost to Wexford or Dublin in week one, for example.  As such, it’s important to look at the expected result of the match.

In order to get an expected result, the BAINISTEOIR ratings for each team at the time of the match were used. These ratings are not perfect, but overall tends to perform well; its favourite has gone on to win over 70% of this year’s matches in divisions 1A and 1B of the league, and the Leinster, Munster and Joe McDonagh groups in the championship, at time of writing. Though the BAINISTEOIR rating system does account for time between matches, it doesn’t calculate this as an advantage for either team; it simply reduces its certainty when the teams haven’t played in a while.

All matches in Leinster and Munster where there were a difference between how many consecutive weeks each team had played were looked at. So far, there have been 8 such matches. It is worth noting that this is a small sample size, so any results below could still change.

For each match, their actual result minus their expected result was plotted against the difference in consecutive weeks played. For example, when Wexford beat Dublin, they were given a 76% chance to win, so their ‘actual minus expected’ value equals 0.24. Dublin were given a 19% chance to win (the remaining percentage is the chance of a draw), so their ‘actual minus expected’ was -0.19. This was Dublin’s second week in a row, and Wexford’s first, so Dublin had a ‘consecutive week difference’ of 1, while Wexford’s was -1. The graph below displays the results for all 8 games:

Performance.png

The trendline for the chart has a slope of -0.0845. This means that, for every extra week in a row, a team’s odds to win go down by about 8%. This could easily be enough to make the difference between winning and losing in many situations, even with only one week’s difference. The correlation coefficient between the actual minus expected performance was -0.516, indicating a strong negative correlation. For reference, a correlation coefficient of 1 or -1 indicates a perfect linear relationship, with no outliers, and a coefficient of 0 indicates no relationship. An absolute value of 0.4 to 0.7 is typically considered to be a strong correlation. As mentioned above, the sample size is very small, so this result might be overly dramatic, but it still points towards a definite advantage for the more rested team.

Scoring margin

In many of the matches with the imbalance in consecutive weeks played, the team that won was already the favourite to win. Given the small sample size, it’s still very possible this was a false positive, and that the same correlation could have happened by coincidence. Though this is still an issue, the scoring margin was also looked at. This helps account for situations where the favourite still won, but maybe won by more or less than expected.

Again, the BAINISTEOIR system was used, as it produces an expected winning margin for each match (and its predictions for Munster, Leinster and the Joe McDonagh are published before each round). Though it’s rare that the ‘most likely’ result is the actual final result, the system performs consistently overall; it will overestimate the scoring margin as often as it underestimates it. Therefore, with a sufficiently large sample size, it’s possible to use this measure to try to determine a specific factors effect on the outcome of a match.

The same procedure was implemented as above, except with the actual minus expected scoring margin substituted for the actual minus expected performance. The results are below:

Again, the small sample size is an issue, but there appears to be an even stronger negative correlation here, with a correlation coefficient of -0.625. The slope of the chart indicates that, for every additional consecutive match played, a team can expect to reduce their scoring margin by 3 points; in other words, the average team playing their second match in a row could expect to concede an extra goal to a team who’ve had a week off, based on these early results. This is, of course, hugely impactful; for example, Dublin lost to Wexford by only 2 points, playing their second match in a row versus Wexford’s first. A three point swing in the other direction, and they’d have won, resulting in them qualifying for the knockout stages, instead of being stuck in 4th place. The entire landscape of this year’s championship would be different!

Adjusting the predictions

If you assume that these results are accurate (which, again, is not a safe assumption, given the small sample size), then how will this affect the upcoming matches? Looking at the BAINISTEOIR predictions for the scoring margin for this weekend’s matches in Munster and Leinster, and modifying them according to the results above, both matches in Munster now have a different favourite!

The rating system originally expected a massive 8 point win for Tipperary. However, as they go into their 4th straight week against a fresh Clare team, this flips to a narrow win for Clare, by a single point (and, as a Clare supporter, I would be perfectly fine with this prediction holding up). Though a less drastic change, it also changes its more tentative prediction for the Limerick-Waterford match; originally a one point win for Waterford, it now expects Limerick to win by two.

Looking at Leinster, the favourites for each match remain the same (Galway over Dublin, Kilkenny over Wexford), but the margins do swing quite dramatically. Though it initially predicted a 9 point win for Galway, this becomes a less certain 6 point win, giving Dublin some room for an upset. On the other side of things, a fresh Kilkenny are now expected to trounce Wexford, entering their fourth consecutive week; originally a reasonably close 5 point Kilkenny win was expected, while now this goes up to 14 points!

How can this be fixed?

Time will tell whether these trends hold up; 8 matches is still plenty of time for flukes and coincidences to cloud the numbers. However, given the strength of the correlation, it does seem that there is some kind of negative effect when playing more weeks in a row than your opponent. If this effect proves to be anywhere near as strong as the initial figures indicate, something will have to be done, as this is simply too much of a competitive advantage for the teams who get a break after their second match, compared to those who play four in a row. Much was made of Waterford’s lack of home matches, but the fact that they had to play four in a row is likely a far greater issue for them, especially given their recent injury problems.

There have been a couple of solutions thrown around already, though its likely that no perfect solution exists. One idea, especially favoured by Offaly fans in order to avoid relegation, is to increase the number of teams in the groups to six. An even number of teams could prevent any one team from playing more often than any other, as everyone has someone else to play; there would be no more ‘spare’ teams over a given weekend. Though the plan is to reduce to five next year, the Joe McDonagh operated on this format this year, with each team playing weekly, with a break between the third and fourth matches. This has some obvious drawbacks, however; though it would be more equal, it’s an even busier schedule. Teams would be at increased risk of injury, and the cumulative fatigue could lead to some flat performances by the quarter final stage.It would also reduce the competitiveness of the matches; Kerry hurling is probably as strong as its ever been, competing with the better division 2A and Joe McDonagh teams, but it would be a big surprise if they could even draw against any of the rest of Munster. Similarly, Offaly lost by an average of over 15 points per match; it would be strange to add another, potentially weaker team on top of that, essentially giving the other four teams another free win each. Additionally, two round robins of six teams followed by knockout playoffs seems far too similar to the current league format to keep both competitions interesting.

Another idea would be to stretch out the All-Ireland. Though there would be some teams playing more frequently than others, the idea is that having more two week gaps would mitigate the disadvantages caused by the schedule. This idea does seem more plausible, but it would require a dramatic 180 by the GAA, who’ve already gone to lengths to try to set aside more time exclusively for clubs. Though it’s questionable how much this time is being used, it would still be a dramatic move to drop the idea entirely so quickly, and would kill the small progress that’s been made in trying to appease the demands of clubs.

Many more ideas may be thrown around in the coming weeks, especially if the impact of this issue is genuinely as bad as the early indicators suggest. Hopefully a fairer solution will be found, without compromising the advantages and excitement gained through this new format. Otherwise, some teams may just need to try to time their peaks for the years where they get the coveted gap between their second and third match, and simply write off the years where they have to play four in a row.