Archive for the ‘Background’ Category

Expected Wins – 2016 season

January 27, 2017

In 2015 we introduced the idea of Expected Wins (“Expt Wins”) which reviews a team’s season, on a game by game basis, versus what they were expected to do using bookmaker’s odds. There are undoubtedly drawbacks to this which are expanded upon in the original article (here) however it is an interesting prism to view the season through rather than just looking at those that had a “good Championship”

2016 Top Performers

Team Games played Wins Win % Expt Wins above/below Expt Pts Win % rank Expt Wins rank
Clare 14 9 64% 5.465 +3.535 5 1
Dublin 16 15 94% 12.218 +2.782 1 2
Longford 10 5 50% 3.430 +1.570 =11 3
Louth 11 7 64% 5.483 +1.517 6 4
Carlow 10 4 40% 2.750 +1.250 =18 5

As can be expected those teams that have a good winning percentage are also high up on the Expt Wins ranking. Keep winning and you’ll consistently outperform your odds and thus your Expected Wins. Dublin, despite their very short odds at times and thus very high Expt Win total, being the poster child for this.

The two teams that had the most unexpected run in the Championship – Clare and Tipperary – are both comfortably in the top10 (Tipperary are joint 6th with an Expt Win total of +0.889) as they made a mockery of their odds at times. Indeed Clare also had an excellent league campaign, winning five of their eight games and picking up the Division 3 title, which helped them climb to first in the table.

Longford had a lesser league campaign winning three of their seven games but completed one of the shocks of the Championship beating Monaghan away in the qualifiers at odds of 15/2 which accounted for nearly all their positive Expt Win total. Removing the bookmaker’s margin they were expected to win that game ~8% of the time.

Perhaps the biggest surprise is Carlow. They won three games in the league for the first time since 2012 (when there was an extra game as Kilkenny played boosting everyone’s win volume in Division4!) whilst also recording only their second Championship win since the end of the 2011 campaign. You won’t read about Carlow having a good season anywhere … but the fact they outperformed expectations to such an extent should ensure it is considered as such.

Versus the Handicap

Another way to review a team is through the prism of the handicap. Bookmakers will tell you that they set the handicap line according to what they think their customers will back, thus giving them an even book, rather than the respective merits of the teams. Public perception, along with the strengths of the team, thus feeds into the handicap line. Even still it is quite a good barometer of how teams are expected to perform. So how did teams fare against the handicap and also the two rankings above?

Team Games played Games covered Cover % Cover Rank Win % rank Expt Wins rank
Kerry 13 10 76.9% 1 =2 =6
Louth 11 8 72.7% =2 6 4
Fermanagh 11 8 72.7% =2 =20 11
Tyrone 13 8 61.5% =4 =2 =8
Cavan 13 8 61.5% =4 =10 9

Only Louth appear in the Top5 of both the Expt Win and Handicap rankings though in truth all 5 that covered the handicaps most often were also high up in the Expt Wins Ranking.

Fermanagh may have only won four of their 11 games in 2016 (two at home in the league, the home win versus Antrim in the Ulster Championship and then away to Wexford in the qualifiers) however they were a tough nut to crack with only 3 teams managing to beat them on the handicap (Cavan winning by 6 in Brewster Park, Derry in the opening round of the league and Donegal in the Ulster championship). Indeed that Cavan game was the only one of the five that Fermanagh played at home where they failed to cover the handicap. We saw something similar in 2015 when they covered in five of the six games played at home. Fortress Brewster!

But this all buries the lead. Kerry were excellent against the handicap (see Note1) where they only failed to cover in the opening two rounds of the league, against Dublin and Roscommon, and again in the league final. We have seen a phenomenon of double digit favourites tending not to cover (see Note2) but when a big favourite in 2016 (11 + 10 versus Clare in their two games and 9 versus Tipperary in the Munster final) Kerry covered each time. They don’t get dragged down by “lesser” opposition – they play to their own level.

2016 Worst Performers

Team Games played Wins Win % Expt Wins above/below Expt Pts Win % rank Expt Wins rank
Cork 11 5 45% 6.617 -1.617 15 28
Limerick 10 1 10% 2.799 -1.799 31 29
Monaghan 14 4 36% 5.829 -1.829 =20 30
Down 9 0 0% 1.960 -1.960 32 31
Armagh 10 2 20% 3.963 -1.963 29 32

It is no surprise, given that league games can make up 60% to 70% of a poor team’s season that relegated teams feature heavily here. Between them Cork, Down, Armagh & Limerick mustered five wins from their 28 league games.

Down didn’t register a win in 2016 however Armagh, despite winning two games, came out slightly worse on Expt Wins. This is due to the fact that they were expected to be competitive in Division2. Outside of the Round6 game away to Tyrone, when the reality of their form was beginning to catch up with their odds, the largest price they were in any league game was 7/4. They also played three (see Note3) Championship games where they were given a good chance of winning each (2/1, 11/10 & 9/4) but came away with three losses.

Down on the other hand started slowly and continued to plummet. They may have lost all nine games but were only ever expected to win ~1.96 of those. Armagh were expected to win ~3.96 of theirs.

Monaghan had an average league campaign winning three of their seven games when they were expected to win ~3.1 games. Their issues were in the Championship where they had two 50:50 games with Donegal but won neither and then lost as 1/12 favourites at home to Longford. That loss alone equates to -0.893 Expt Wins.

Versus the handicap

Team Games played Games covered Cover % Cover Rank Win % rank Expt Wins rank
Laois 12 4 33.3% =28 =24 =23
Derry 12 4 33.3% =28 17 25
Waterford 9 3 33.3% =28 28 27
Down 9 3 33.3% =28 32 31
Wicklow 9 2 22.2% 32 =24 21

This is the “hope you weren’t relying on these guys throughout the year” listing.

Similar to the positive performers table there is only one team – Down – that appears in the worst list for both Expt Wins and handicap rankings. Again like the top performers those at the bottom of the handicap ranking are also in the bottom third of the Expt Wins ranking.

Waterford & Wicklow appear in the bottom 5 for the second year running. Combining 2015 & 2016 they are running at a 28% cover rate (9 games from 32).

We have already seen one of the teams relegated from Division2 – Armagh – appear in the “worst performing” category so it’s no surprise to see the other two teams that filled out the bottom three of that division make an appearance here. Division2 was seen as ultra-competitive at the start of the year and this was followed through with a high volume of games with a close handicap (see Note4). Once teams started to underperform they were going to struggle against the handicap.

Notes

Note1; what is more remarkable, and a cautionary tale re taking trends at face value, is that Kerry were in the bottom two in this metric in 2015. Then they covered in 31% (4 out of 13) of their games. We must always remember that we are dealing with very small sample sizes (and margins) here where a point or two can have a huge effect.

Note2; twenty teams since the start of the 2015 league campaign have started as double digit favourites. Only 7 (35%) have managed to cover this.

Note3; Although the first Laois game was declared null and void it was played as if it was a Championship game and we have odds & a result for it so it is included overall.

Note4; 86% (25 games out of 29) had a handicap of 2 or less in Division2. This was 52% for Division1, 83% for Division3 and 48% for Division4

Thinking out loud; In GAA betting the handicap is intrinsically linked to the match price. If those with high Expt Wins (essentially odds on shots) generally cover at a better rate, and the corollary is true for those with low Expt Wins are the linkages between match price and handicap line “out”?

Early Conversion Rates are poor – why?

November 10, 2016

Early Conversion Rates

Whilst uploading the 2016 data into the database I was noodling around in the numbers and produced a simple chart for production on Twitter.

graph-1-overview

Something was quite obviously happening in the first 10 minutes that saw the cumulative Conversion Rates much lower than the average. There were two initial thoughts

1. a number of “lower level” teams were dragging the average down early in games (either through just poor shooting or an inability to get “quality” shots off against better teams early on when the scoreboard was close)
2. shooting types, and where shots were being taken from, were so different in the frantic opening periods of games that the early Conversion Rates were being skewed

Upon doing some more superficial digging it appears that neither were the case

1. Conversion Rates by teams

graph-2-by-team

The phenomenon (of Conversion Rates being lower early on) was observed in three of the four semi-finalists (NOTE1) whilst all other teams followed the overall trend to a tee. The only outliers – unsurprisingly – were Dublin.

2. Expected Points over time

graph-3-expt-pts

The above is a replica of the Conversion Rate chart but replacing Conversion Rates with Expected Points (Expt Pts). Although the shape of the chart is different than the original the occurrence of poor early returns is still evident. And by using Expt Pts we remove the shooting types as an issue as Expt Pts bakes in the difficulty of a shot (NOTE2). All shots are being converted at a lower rate than expected until around the 30th minute but teams are noticeably struggling in the first 15 minutes.

Conversion Rates by shot type

So the phenomenon is real but cannot be attributed to a specific team type nor to shot selection/execution. It is across the board except for Dublin. Three shots types – free kicks, point & goal attempts from play – make up ~97% of all shots. Is there anything we can determine from investigating these shot types independently to explain this poor shooting in those early exchanges? And is there anything therein that explains how Dublin are managing to avoid this poor shooting early on?

Free kicks

graph-4-by-free

This is probably the most surprising, and hardest to attribute, of all the results. When the very first chart was produced on Twitter I mischievously suggested that whatever all the back-up teams were doing to get teams warmed up they needed to change it. There were some good responses re the intensity of teams, especially in the pressure applied to shots, being higher early on. Or that teams were defensively more conservative early on leaving less space for clear shots. All plausible and probably have a grain of truth. However none applicable to free kicks – and the phenomenon of poor conversion rates early on is noticeable here too.

Now by slicing the volumes into the first 10 minutes of one season’s games we are running in to sample size issues. Specifically for this segment the volume is 47 so this comes with a rather large health warning.

Assuming games are now 80 minutes the first 10 minutes make up 12.5% of the game; the 47 frees in the first 10 minutes make up 13% of all frees. On top of that the two main free takers – D Rock & C O’Connor – make up 21% of all frees in the first 10 minutes whereas they make up 25% of all frees in the database for 2016. So the first 10 minutes, low sample size and all, are representative of the whole year. So what happens in those opening 10 minutes?

Shots Scores Expt Pts Conversion % v Expt Pts
All frees 47 29 32.8 62% -3.8
Rock & O’Connor 10 9 7.8 90% +1.2
Others 37 20 25.0 54% -5.0

What the above table shows is that Rock & O’Connor were on point from the get go. Overall for the year they combined for an 86% Conversion Rate and in the first 10 minutes they were 90%.

If the two main protagonists were on point the rest of the free takers must be dragging the averages down from 71% overall to 62% in the first 10 minutes. And as the table shows this is the case. Indeed they were very poor returning a paltry 54% (the 80 minute average for all free takers outside Rock & O’Connor was 66%).

And this somewhat negates the argument for lower Conversion Rates early on being affected by what the opposition’s defence is doing. The opposition can’t really affect free taking. Outside of Rock & O’Connor it looks like free takers were just not ready early on (NOTE3).

Points from play

graph-5-from-play

The Conversion Rate for 2016 was 44.2% and for the five years from 2012 was 45.8%. For the first 10 minutes of 2016 games the conversion rate was 36% and only rose to a cumulative 38% by 20 minutes. Again the Expt Pts was lower in the first 10 minutes (-15.70) as against the remainder of the game (+5.84).

I do track whether a shot was taken under pressure however have only used it anecdotally to date as it is a simple “Y/N” flag and is probably not nuanced enough for any concrete use. Having said that however there is only one person applying the flag so we would expect a certain degree of consistency of application across the ~1,000 shots tracked here.

In the first 10 minutes I charted 53.6% of all point attempts occurring whilst under pressure. The remainder of the time it was 54.2%. Near enough as makes no difference.

So the poor shooting for points from play is real, is not linked to poorer shot types (as evidenced by the Expt Pts return) and from the empirical data we have is not linked to greater pressure applied earlier on in the game. I am completely open to the intensity of the pressure being different early on (NOTE4) but if this was the case you would expect some uptick early on in the percentage of shots marked as taken under some/any pressure in this timeframe. There is none.

There may be other non measurable factors such as nerves (these are amateurs after all) but as of now I can’t come up with anything other than the aforementioned “mischievous” reason that players are just not at peak performance early on. Maybe this is to be expected?

So what of Dublin? We saw that their early conversion Rates outperformed everyone else. This is in part due to the fact that Dean Rock went 5 from 5 on his frees but how was their shooting from play?

Shots Scores Expt Pts Conversion % v Expt Pts
Dublin 23 9 10.2 39% -1.2
Mayo 28 8 12.0 29% -4.0
Tyrone 19 8 8.7 42% -0.7
Donegal 17 4 7.4 24% -3.4
Tipperary 13 5 6.8 38% -1.8
All first 10 179 65 80.7 36% -15.7

Again volumes are low (NOTE5) but Dublin were no great shakes early on. Yes they were above the average for the first 10 minutes but they still underperformed when compared to the whole game average and their Expt Pts – like all the teams above – was below 0.00.

Perhaps the most striking return here is Mayo. From the 10th minute onwards they were exactly in line with Dublin (Mayo 49% on 126 shots with an Expt Pts of +5.59; Dublin 49% on 132 shots with an Expt Pts of +5.54) but for those first 10 minutes they were much poorer.

Another theory for the poor start was not where teams were shooting from but who was shooting – less pressure on returns early on so midfielders/defenders were more inclined to “have a pop”. So I had a look at Mayo’s shot distribution. In the first 10 minutes 64% of their shots came from what I would state are obvious offensive players (A Moran, A O’Shea, J Doherty, A Dillon and the two O’Connor’s). From the 10th minute onwards, and adding E Regan, C O’Shea and A Freeman to this mix who didn’t have a shot in the first 10, these forwards accounted for 60% of point attempts (NOTE6).

It is difficult to attribute offensive/defensive tags to all players in today’s game but if there was a decisive split in who was shooting for teams you would expect it to show up in the team with perhaps the worst split. But it doesn’t.

Goal Attempts

graph-6-goal-attempts

To be honest I am just including the above for consistency and to help explain Dublin’s apparent ability to start faster than others. Whilst I have consistently cautioned against low sample sizes it is an overarching feature of this shot type and can explain a lot of the variance within the five minute groupings above. In total there were 137 goal attempts with just 15 in the first 10 minutes and 36 within the first 20.

Having said all that …. the Conversion Rate for goal attempts was 53% in 2016 and only crawled up to 40% after 15 minutes. With the evidence we have teams again were not converting on goal attempts early on in games.

Dublin? They had six goal attempts in the first 10 minutes scoring 3-00. 50%. And there is their apparent early start in a nutshell. They were 50% on goal attempts, 100% on deadballs (as well as Rock’s aforementioned frees he was 2 from 2 on 45s as well) and slightly below average at 39% on point attempts – giving them the aggregate of ~52% early doors.

Overview

This is based on one year’s data (NOTE7) but poor early conversion rates were definitley a “thing” that year

There is no evidence that shot selection (through Expt Pts), opposition pressure (through the simple “Y/N” flag) nor type of shooter (using Mayo as an example) is any different in the first 10 minutes to the rest of the game

It is also evident in early free taking, except for the very best in Rock & O’Connor who were on point from the very start, which somewhat nullifies the theory that it is something the opposition is doing to affect the shooting.

There are undoubtedly other factors at play. Some can be measured; first shot in the game, effect of new surroundings, debutants vs more experienced players, intensity of pressure. Some we may never be able to measure – nerves, mentality of players early on versus later in the game, etc.

But as of now, and taking all of the above into account, I cannot escape the initial gut reaction that players are just not ready – for whatever reason – early on

NOTE1 – we need to be careful with any segmentation. There are only 1,640 shots in total being reviewed here with 249 in the first 10 minutes. Segment that further by team and you get some ridiculous numbers; Kerry just have the 7 shots across two 2016 games; similarly Tipperary only have 16 shots in the same timeframe. You can’t make any judgements on those numbers. In truth I would not normally use a chart with such low volumes but I include it here as it was the chart that sparked me into looking deeper into the issue.

NOTE2 – for more on why this is so please see here

NOTE3 – I had a further look at the non Rock & O’Connor frees to see if any one player was having an effect. There was none really. 34 of the 37 were a player’s first attempt in the game which makes sense as it is uncommon for a team to have two shots at goal from a free in the opening 10 minutes.

This leads to a further corroboration that could be investigated – across the year’s how does a player’s very first free kick equate to the rest of their results?

NOTE4 – I started to grade pressure on a sliding 0 – 3 scale for the two All Ireland finals. It feels a lot more robust as having to apply a grade makes you stop and think. It will be very instructive from here on in but as of now I’m not inclined to go back over the entire season to retrospectively apply the grade(s)!

NOTE5 – This table lists all the teams with >10 shots from play in the first ten minutes. Again we are running into sample size issues.

NOTE6 – the non offensive players with a shot in the first 10 minutes were B Moran, D Vaughan, K McLoughlin, L Keegan, P Durcan & T Parsons. Other defensive players with shots post the 10th minute were C Boyle, K Higgins, S O’Shea, S Coen, B Harrison & K Keane

Note that whilst some of these could be moved into the “offensive” pot their individual shot volumes are such that it wouldn’t make a material difference to the overall point.

NOTE7 – why one year? Because for some unknown reason I didn’t track the time (outside of 1st half/2nd half) for previous years. Hell of an oversight in retrospect! The only reason I started in 2016 was I was so bored of looking at kickouts so decided to look at the rest of the game. I have one or two other pieces I newly gathered in 2016 so hoping to get another long form piece out on those

Moving from Weighting to Expected Points

April 20, 2016

Expected Points

Since the blog’s inception weighting has been the main comparative tool. Towards the back end of the 2015 season Expected Points (ExpPts) began to be used in the commentary. This is a very similar concept to Weighting, in that it is created using historical averages, but (hopefully) it is much more instinctively understood and one that the blog will be switching to in 2016.

On top of this there will be a major tweak to how the old weighting, and now ExpPt, will be calculated. Up until now the weightings have been generated purely on a “percentage of shorts scored” basis with the conditions being

• Where did the shot originate
• Was it a deadball (& what type) attempt or from play
• Was it an attempt on goal

One omission from the above, amongst many! is the impact the defense was having on shooting returns. This changed with the Big Fish Little Fish piece where we observed big shifts in Conversion Rates depending on the opposition. This concept will now be introduced to ExpPts.

Converting Weighting to ExpPts

This switch is relatively straightforward for everything bar goal attempts. In the 2015 Review we saw that 37% of point attempts were converted from Sector 4. In old parlance if you converted a shot from here your weighting was +0.63 and if you missed it was -0.37. The weighting was based on the Conversion Rate.

The same principle applies for ExpPts. As 37% of all shots are converted a player is expected to score 0.37 points for every shot taken from here. If he converts that’s +0.63 ExpPts (1 – 0.37) or -0.37 if he misses. Do that across all the shot types in all the sectors and we have an ExpPts model.

Goal attempts are slightly trickier. The weighting only ever had a value between 1 and -1 however on goal attempts we know there are three different returns; 3 points, 1 point or 0 points. There can be no direct correlation between goal shots and weighting. Instead we’ll create an ExpPts for goal shots.

Over the past four years 531 goal attempts have been charted with a total of 192-50 scored. This haul equates to 1.18 points per attempt ((192*3)+50)/531). So the ExpPts for a goal attempt from play, without overlaying any factors, is 1.18.

As ever this is a linear application. Not all goal shots are created equally (a pass across the square for an easy fisted goal is very different to a desperation shot with three defenders and a goalie in the way) but whilst shot location has been charted goalie & defender positions have not. Further sub division of goal attempts will be the main 2016 project.

Deadballs are easier. All bar 38 (29 penalties & 9 attempts at goal from a free) of the 1,411 charted were point attempts so we have a direct correlation between weighting and ExpPts using the above methodology. For the record the ExpPts for a penalty is 2.48 (24 of 29 converted) and for an attempt at goal from a free is 0.33 (1 goal scored from 9 attempts).

A nice example of how this works is the drawn Mayo-Dublin semi-final. Below is the ExpPts of both teams throughout the game. You can see how Mayo tracked their ExpPts throughout (i.e. were basically getting what you expect from the shots attempted) but Dublin were nearly always ahead with the gap widening around the 55th minute. Here they scored 2 goals getting 6 points versus an expected return of 2.36.

Expected Pointsv2

Introducing Defensive Adjustments

Using the above model gets us closer to reality on how a player or team performed shooting wise however as noted it is still very “simple”. There are many factors that are not taken into consideration one of which is the quality of opposition. Scoring a point against Mayo surely cannot equate to scoring a point against xxxxxxx (insert name of a particularly poor team)

During the Big Fish Little Fish article it was highlighted that there were large discrepancies in returns depending on the opposition. During that article teams were split into semi-finalists and non semi-finalists and the outcome of shots taken in each type of game compared. This gives us four distinct “game types”

• semi-finalist shooting against other semi-finalists
• non semi-finalists shooting against non semi-finalists
• semi-finalists shooting against non semi-finalists
• non semi-finalists shooting against semi-finalists

In the Big Fish Little Fish piece the argument was made that the non semi-finalists probably need to be further sub divided (into Div1 & Div2 v Div3 & Div4 perhaps) but we do run into sample size issues. We could also arbitrarily pick teams to be “Big Fish” (Dublin, Kerry, Mayo, Donegal definitely. Tyrone?). Again one to be reviewed with more data.

So to wrap up we now have two Expected Points models; a bog standard one (ExptPts) based on pure averages – this is what was introduced within the two league semi-final reviews. The only difference with this metric from historic weightings is goal attempts with every goal shot having an ExpPts value of 1.18.
We have also created a defensively adjusted Expected Points (will have to come up with some shorthand. I’m not writing that out every time!). Where this will really come to light is in the early round of the Championship when the player’s returns will be adjusted based on the quality of opposition.

Big Fish Little Fish

November 30, 2015

2015 saw a number of very one sided games between the big 3 (Dublin, Kerry & Mayo) and the rest. Kerry’s shellacking of Kildare in the quarter finals. Dublin ripping through Leinster. Mayo destroying Sligo.

The blog, whether using weighting or expected points, has always used averages as a base. The more positive the weighting the better the team played when compared to the average. But these one sided beat downs must be affecting the averages. And if they are to what extent?

On top of just pure skill differentials there must also be something else happening in these games. How is it that a team like Kildare can beat Cork one week but lose heavily to Kerry the next? Do weaker counties play differently when facing the bigger teams? How do the big teams’ numbers compare when facing each other as opposed to the cannon fodder earlier in the Championship?

What is a Big Fish?

First we must define a big team. It would be very easy to just pick Dublin, Kerry & Mayo but we run into sample size issues when isolating games between similar teams. The semi-final replays of the last two years have helped but restricting “big teams” to the aforementioned three would leave us with only eight games over four years. To expand the big team pool all semi-finalists in any particular year are included (See Note 1 in the Appendix). This increases the volume of games between big teams to 14.
 
Deadballs

Shots Scores Success % Shots per game
non semi-finalist v non semi-finalist 465 323 69.5% 6.84
non semi-finalist v semi-finalist
non semi-finalist 358 241 67.3& 6.75
semi-finalist 365 257 70.4% 6.89
semi-finalist v semi-finalist 184 124 67.4% 6.57

 

The most remarkable aspect of this table is its blandness. No matter the game type the difference in volumes of deadballs per game, and the accuracy of those deadballs, is negligible.

In a way this makes sense as there is no real outside influence when you take a free (frees make up ~85% of all deadballs so in many ways the term deadball & free are interchangeable). It is just the player and the ball; the opposition cannot influence the outcome (See Note2 in the Appendix)

What is surprising is that in games between semi-finalists & non semi-finalists the volume of shots is the same. Yet we know that the volume of attacking play is heavily weighted in favour of the semi-finalists. In 12 games of this nature (semi-finalist v non semi-finalists) this year the semi-finalists averaged 41 attacks whilst the non semi-finalists averaged 32. The volume of deadballs has stayed steady over the years meaning that the semi-finalists gave up a deadball shot every ~4.7 attacks; the non semi-finalists were at ~6.1.

Now maybe this is due to better attacking play from the semi-finalists, enabling them to avoid getting caught in possession, but equally it could be the better teams fouling early to get back into position and/or stop a goal threat. Or maybe it’s just sympathetic referees in blow outs! But whatever the reason the better teams do foul at a higher rate when faced with inferior opposition.

For point from play
 

Shots Scores Success % Shots per game
non semi-finalist v non semi-finalist 1,214 544 44.8% 17.85
non semi-finalist v semi-finalist
non semi-finalist 847 338 39.9% 15.98
semi-finalist 1,156 598 51.7% 21.81
semi-finalist v semi-finalist 533 257 48.2% 19.04

 
The deadball overview may have been unremarkable – this is another beast entirely.

    Non Semi-Finalists

When playing each other non semi-finalists converted point attempts at 45%. This dropped to 40% when playing semi-finalists. The volume of shots also dropped by ~10%.

The deadball returns showed that all things being equal there are no real differences when comparing the various game types. Things are not equal here however as the opposition has a huge bearing on your returns.

But how does that manifest itself in terms of the reduced accuracy noted above? There are two main elements that have been tracked to date (there are others!) where the opposition can have an impact on your shooting – where the shot is taken from and whether the shot is taken under pressure.
 
Non semi-finalists’ % of all shots taken

Outside 45 4 5 6 7 8 9
v non semi-finalists 2% 24% 25% 16% 13% 11% 9%
v semi-finalist 3% 26% 23% 16% 14% 11% 8%

 
Nothing to see here. Irrespective of opponent non semi-finalists’ shooting tendencies remain constant. Remarkably so.
 
All teams – frequency & impact of pressure

Pressure applied Success % Pressure applied Success %
non semi-finalist v non semi-finalist 53% 42% 47% 48%
non semi-finalist v semi-finalist
non semi-finalist 57% 38% 43% 42%
semi-finalist 47% 48% 53% 55%
semi-finalist v semi-finalist 48% 43% 52% 53%

 
If the lower conversion rates are not due to shot location are they impacted by pressure? As expected semi-finalists do manage to pressure more shots than non semi-finalist in games between the two – 57% v 47%.

That however does not tell the whole story. “Pressure” is recorded as a binary “yes/no”. The *type* of pressure – be that multiple players, or tighter marking from better defenders – is not accounted for. That, on top of the mere presence of extra pressure, could easily be responsible for the drop in the overall Conversion Rate from 45% to 40%.

What’s really interesting though is what happens when there is no pressure applied. You would think, both from a logical standpoint as well as from the results observed in deadball situations, that the returns here would be similar irrespective of the opponent. There is no outside influence. But in contests against other non semi-finalists, where no pressure was applied, players had a return of 48%. In the same scenario against semi-finalists the return dropped to 42%. That’s a bigger drop off than when pressure is applied (see Note 3 in the Appendix)!

    Semi-finalists

The fact that something “strange” is happening with non semi-finalists is further confirmed when we look at the semi-finalists. As expected there is little difference in the returns when no pressure is applied; against non semi-finalists they convert 55% and against other semi-finalists they convert 53%. The difference is well within any margin of error.

The earlier point re the type of pressure being different is also underlined. Semi-finalists, when facing non semi-finalists convert 48% of shots taken under pressure. Against other semi-finalists that becomes 42%. The assumption again being that the drop off is due to the *type* of pressure being applied by the better teams.
 
Goal Attempts

Shots Scores Success % Shots per team pts per attempt
non semi-finalist v non semi-finalist 141 55-16 39.0% 2.14 1.28
non semi-finalist v semi-finalist
non semi-finalist 118 31-12 26.3% 2.23 0.89
semi-finalist 219 96-14 43.8% 4.13 1.38
semi-finalist v semi-finalist 82 34-08 41.5% 2.93 1.34

 
The same patterns observed when attempting a point are observed when going for a goal.

For non semi-finalists the accuracy plummets as the opposition quality increases. For semi-finalists the accuracy is more or less maintained but they manage ~50% more goal attempts per game when facing non semi-finalists.
 
 
Based purely on point taking ability and volumes (we’ll take deadballs as a wash) the big fish of this piece have about a five point head start. That grows out to ~8.5 points once we bake in goal attempts. That is a huge gap to plug but the little fish of this piece are not helping themselves. Too many of their own shots are taken under pressure whilst they are also allowing their goal to come under siege. That’s not to mention malfunctioning when under no pressure. Nor the unedifying thought that they are not cynical enough and should be following the big fish’s lead and fouling more.
 
 
Appendix
 
 
Note 1: The big fish are thus

2012 games; Cork, Donegal, Dublin & Mayo
2013 games; Dublin, Kerry, Mayo & Tyrone
2014 games; Donegal, Dublin, Kerry & Mayo
2015 games; Dublin, Kerry, Mayo & Tyrone

There is an argument to be made that the non semi-finalists – the “little fish” of the piece – needed to be further subdivided whether that be by league position, losing quarter finalists or some subjective manner. Lumping the Cork & Monaghans of this world in with the Leitrim & Longfords is not fully representative. This is accepted but again we run into sample size issues.

Still to be honest I nearly did it anyway so that I could use the “big fish”, “little fish” and “cardboard box” nomenclatures. Next time.

Note 2: Not strictly true I guess as the opposition can – in many cases – choose where to foul and thus impact the outcome of those frees. Where deadballs attempts originated from in games between semi-finalists & non semi-finalists are listed below.
 

Outside 45 4 5 6 7 8 9
v non semi-finalists 13% 18% 20% 18% 5% 19% 7%
v semi-finalist 10% 19% 25% 17% 6% 18% 5%

 

The semi-finalists do get more shots off from Sector5 whilst giving up more outside the 45. This could be used to support the “clever” fouling argument in that they are fouling earlier to protect the defence. It could also be the case that the weaker teams take whatever opportunity they get so have more long range attempts.

This is the only real difference however. Generally speaking teams foul in the same areas.

Note 3: The sample size here is robust (~950 shots) so the issue is real. Pressure is a nebulous thing – one man’s pressure is another’s lazy arm – so consistency in the definition could definitely be an issue. There is however only one source tracking the games so we have to assume there is consistency.

Other issues like game state (more pressure on shots in closer games), game type (semi-finalists by nature are meeting in bigger, more pressurised games) or pitch (shooting in Croke Park into the hill against the Dubs vs Clones perhaps) will all have an undetermined effect.

2015 Season Review – Part II

November 10, 2015

In Part I it was observed how the volume of shots dropped from the 2014 high of 30.9 a game back to 27.8 in 2015 (in line with previous averages from 2012 & 2013). With the quantity down was the quality affected? Yes – but in a positive manner.

The overall accuracy on all shots increased. Between 2012 and 2014 (3 years, 74 games and 4,246 shots) 51.2% of shots were converted with little year-on-year variance; 51.7% in 2012, 50.5% in 2013 and 51.3% in 2014.

2015 saw a 5.2% increase on this three year average to 53.8% (26 games & 1,446 shots). Like the deadball increase observed in 2014 (more on that below) I would be loath to read too much into one year’s worth of data however it is a noteworthy movement given (a) the size of the jump and (b) the fact that there was a jump at all after the steadiness of the previous three years.

So how was this increase achieved? Shots are broken down into three main constituent parts; deadballs account for 26% of all shots, goal attempts account for 9% with the remaining 65% coming from attempts for a point from play. The 2015 returns for all three are reviewed below.
 
Deadballs
 

Shots Scores Success Rate
2012 347 232 66.9%
2013 389 259 66.6%
2014 328 239 72.9%
2015 347 240 69.2%

One of the main findings from the 2014 review – expanded upon here – was the fact that deadball accuracy jumped after three years of remarkable consistency (although not shown in the above table the 2010 season had a Success Rate of 66.3%).

Whilst that increase was not sustained in 2015 the overall returns were still very good in a historical context. To be of an average intercounty standard your team needs to convert 70% of deadballs assuming a normal spread of distances & type.

So how is this 70% achieved?

Shots Scores Success Rate
Frees 304 215 70.7%
45s 34 19 55.9%
Penalties 7 6 85.7%
Sidelines 2 0 0%

 

Only 47 penalty & sideline attempts have been charted since 2012; much too low a number to make any concrete conclusions on. [As an aside 83% of the penalties were converted and 28% of sideline attempts]

The number of 45s converted continues on its upward curve (40% Success Rate in 2012, 50% in 2013, 52% in 2014 and now 56%) to give an overall average of 49.4% over the four years. This increase has little effect on the overall deadball Success Rates however as 45s only account for ~12% of all deadballs.

So that leaves free kicks. As ever with deadballs it is free kicks where the real movement happens. They account for ~85% of all deadball attempts (and 21% of all shots in total).

In 2014 the Success Rate for free kicks jumped to 76% from 70% & 71% the two previous years. There was no real trend as to why this was except to say that accuracy improved across the park. This year? That accuracy dropped back to 70% – bang in line with previous norms. Hello regression to the mean.
 
From play – for a point
 
Two thirds of all shots are attempts at a point for play. Though the Success Rates in the other shot types are important a team’s bread and butter can be found here.

Shots Scores Success Rate per game
2012 887 419 47.2% 17.74
2013 888 397 44.7% 17.76
2014 1012 453 44.8% 21.08
2015 963 468 48.6% 18.52

 

2015 saw a drop of ~2.5 shots per game which, though dramatic, is still ~0.75 shots higher than observed in 2012 & 2013. This lower volume did produce a higher quality however with a Success Rate of 48.6%. That is a ~8% increase on the previous two years.

Although there was a similar return in 2012 I had a look at where the shots originated to see if there was any discernible change in pattern (more shots from easier sectors). There wasn’t – if anything there were less shots from the easiest sector – Sector8 – just in front of goal.

Sector Outside 45 4 5 6 7 8 9
’12 – ’14 2% 23% 24% 17% 12& 13% 9%
2015 1% 24% 24% 18% 13% 11% 9%

 
Seeing as the ease of shot hasn’t changed the conclusion is that the quality haS increased. Ignoring shots taken from outside the 45 – which only account for ~2% of all shots – the Success Rate increased for all sectors bar Sector5 which remained stable.

Sector Outside 45 4 5 6 7 8 9
’12 – ’14 37% 37% 50% 35% 42& 71% 46%
2015 27% 43% 49% 41% 47% 75% 52%

 
 
From play – for a goal
 
The first thing to note is that the prevalence of goal attempts has not changed in any real sense. In 2015 goal attempts made up 9.4% of all shots; it was 9.6% the two previous years.

What has changed, and in truth has been a noticeable trend since 2012, is the accuracy of these goal attempts. In 2012 a score (a goal or a point) was returned from 39% of goal attempts. This has risen year on year to 52% in 2015. When we only include goals as a score (probably a more accurate measure of goal attempts!) there is still a noticeable upward trend.

Shots Scores Success Rate per game
2012 117 40 34.2% 2.34
2013 136 44 32.4% 2.72
2014 142 52 36.6% 2.96
2015 136 56 41.2% 2.62

 

Teams are getting more scores, and more goals, from their goal attempts.

So there you have it. An overall increase fuelled by better accuracy from play – both in point & goal attempts – though the increase was somewhat dampened by a drop in free kick accuracy.
 
Dublin
 
Do Dublin, given the volumes they achieved during the year, have an overbearing affect?

The answer is probably in the question – of course they do. Taking goals only Dublin scored 16 on 32 attempts in 2015 meaning that the remainder of teams converted at a 38.5% clip. In 2014 Dublin only converted 28% (9 from 32) with the remainder returning 39%. Whilst not wholly reliant on Dublin’s returns (Mayo put 6 past Sligo whilst Kerry put 7 past Kildare) the fact that they have been responsible for 24% of all goal attempts means that the year on year increase has followed their outcomes.

Similarly when going for a point Dublin converted 57.3% on 17% of all attempts recorded with all other teams converting 46.8%. It is not just Dublin here however as Mayo converted 56.7%.

We use averages as a starting point as, with a large enough sample size, these “outliers” will be subsumed by the whole. However when viewing the Grade A teams (Dublin, Mayo, Kerry) a premium needs to be added to the average when reviewing their play whilst Grade B teams – those trying to break through (Cork, Tyrone, Galway) – need to aim far higher than the average.

2015 Season Review – Part I

October 29, 2015

Reviewing the major statistics – Attacks, Attack Rates, Shots, Shot Rates & shooting accuracy – from the season just gone, and comparing them to previous seasons, has become a staple of the blog. We do it for two main reason; the first to see if there are any major trends, or indeed areas of complete randomness, jumping out whilst the second opens a window on any elements we should be taking a closer look at.

We tracked where attacks originated for the first time in 2014. This was extended in 2015 to cover all possessions. This enables us to have a complete picture of an average (see NOTE 1 below) Championship game. How often teams have the ball, where those possessions emanate from and what they do with them.

Given this expansion I am going to split the review into two parts; below concentrates on everything up to shooting whilst the second piece (due soon!) will focus on shooting trends and accuracy. So without further ado here are the overall returns up to the point of shooting

 

Year Possessions Attacks Attack Rate Shots Shot Rate
2012 35.3 27.0 76.6
2013 36.3 28.3 77.8%
2014 39.8 30.9 77.5%
2015 49.7 36.8 74.0% 27.8 75.7%
avg 37.0 28.5 76.9%

 

Possessions volumes & origination

How a possession is defined is expanded upon within the definitions page but essentially a possession starts with gaining “control” (see Note2 below) of the ball and ends with either a shot or the opposition gaining control of the ball (a turnover). For the sake of this piece all turnovers are equal – whether forced or otherwise – with the only defining characteristic being where on the pitch control of the ball was obtained.

On average a team has the ball ~50 times (49.69 to be exact!) a game. This has always seemed low to me – essentially 2 possessions every 3 minutes allowing for injury time. I guess it emphasises the need to squeeze the most out of each possession though as ever we must be wary of averages. Especially averages built on one year’s data.

Where this single piece of information may be of most use is when chasing (or defending) a lead with “x” minutes to go. You can extrapolate how many possessions you’ll have; how many shots; what you need to break your way; go for points or when to push the “go for goals” button. Game endings can take on a life of their own but knowing you’ll have 10 possession in the last 15 minutes gives you a good starting point.

Where the possessions emanated from are listed below

 

Possession Origination # % all possessions
Own kickout 16.0 32%
Opp Kickout 6.9 14%
T/over own 3rd 17.2 35%
T/over mid 3rd 6.1 12%
T/over opp 3rd 0.9 2%
Other 2.6 5%

Essentially one third of all possessions come from a team’s own kickout, one third comes from ball gained inside a team’s own 45 with the remaining third spread across the game. There is huge emphasis placed on kickouts – both your own and the opposition’s – but teams get as much ball up to the opposition’s 65 from broken play as they do from kickouts. Do teams place as much emphasis in moving the ball, or defending counter attacks, as they do on kickouts? Does the general commentary? I’m as guilty as anyone. Tracking kickouts is easy but how (and why?) teams are able to move the ball in broken play is every bit as important.
 

Attack & Shot Rates

Last year we saw a spike in attacks; rising 11% from 35.80 in the two previous years to 39.83 in 2014. At the time quite a bit of the increase was being attributed to the black card (well the numbers weren’t used in the wider commentary – it was more the black card being acclaimed for leading to open play) however here on the blog judgement was reserved. The black card was part of the reason but it could as easily have been a one year outlier as much as anything.

That reticence seems well founded now as the 2015 average attack numbers have dropped back to 2013 levels. Looking across the four years a team will create an average of 37.0 attacks and, using 2015 data only, 74% of all possessions end up as an attack.

From 2012 to 2014 the average number of shots per game rose from 27.0 to 30.9 – a 14% increase. Over that time there had been a rise in shot rates but nowhere near the 14%. Instead the increase in shots had risen in line with the volume of attacks (with a small bounce from increased efficiency as the shot rate went from 76.6% to 77.5%).

In 2015 the number of shots per game dropped ~10%. As we can see from the above this is to be expected when the number of attacks drops but on top of the drop off in attacks there was also a drop off in the shot rate. For the first time in four years the Shot Rate came in below 76%. Again this is one of those movements that is probably more of a blip, or general randomness, but it is still worth monitoring. The sample size – 1,446 shots in 2015 & 4,246 in the previous three years – is robust.

So there we have it. On average a team has 50 possessions a gam converting 74% of those possessions to an attack and 77% of those attacks to a shot leaving a total of 28.5 shots a game. What teams do with those shots will be expanded up in Pasrt2 of the review.

Note1: Averages hide a lot and the fact that Dublin have been the pre-eminent attacking team in the period that these returns cover (probably involved in ~20% – 25% of all games) will have an effect. There is definitely scope for rerunning these numbers with/without Dublin and also in games completely excluding Dublin (or even Croke Park for that matter).

Note2; Control, and where it starts and ends, can be a very subjective thing. I tend to veer away from subjectivity – though to start to expand and add value I am going to have to dip my toe in that water (more of that anon) – however here I would estimate that ~98% of the time who is in control of a ball is obvious. The other 2%? I guess we’ll just have to go with it ….

Impact of Game Situation

May 15, 2013

Whilst the weighting is a step further on from success rates, helping to explain how a team’s, or player’s, shooting compares to the average, it is by no means complete. There are many elements to a shot that are omitted from its calculation but one important one, game situation, is tackled below.

Where we stand

What we know at present is that 51% of all shots are converted

Table1

Even that simple statement is slightly misleading as the type of shot is hugely important. Shots from play are converted 45% of the time whilst attempts at goal from deadballs (frees, 45s in the main) are converted 67% of the time.

Delving a bit further we know that where the shot is taken from is of huge importance. Dividing the pitch into 9 segments (see here for details) we can see that as expected the closer in to goal, and the more central the shot, the better.

Table2

There have been tweaks along the way, notably separating frees from 45s and shots at goal from shots for points, but these two factors, shot type & shot position, have been the general thrust of the weighting since the blog began.

We know that there are other elements to a shot’s success; pressure on the shot from the opposition, weather, quality of the opposing defence to name but a few. Another one is the game situation.

Game situation

If you are behind, or ahead, does this affect the returns? My natural inclination was to believe that it was easier to shoot from the front – that pressure to keep up would see lower returns for those chasing the lead (there is also the fact that the shooter’s team is losing for a reason – maybe their shooting is just poorer!)

Table3

There is definitely evidence of improved performance whilst shooting with a lead but it’s a mixed bag. The deadball striking was baffling at first. I couldn’t understand why it was that players would perform better from deadballs whilst chasing a lead.

The answer came however when I looked at the type of shot being taken, Table 2 above shows us that the closer in to goal you are the more likely you are to score. Teams chasing the lead take 5% more of their shots from inside the 45m. When protecting a lead are teams more willing to foul an opponent (Mayo V Dublin in last year’s semi final) or is this just normal variation that will even out over time? The volumes are relatively robust (in excess of 550 frees) which would suggest there is an element to the former.

From Play

The difference in shooting from play is more pronounced and with a lot more data points. Also comparing the shots from play does not have, to the same extent, the issue of where the shots are coming from as in the deadball commentary above. Nearly 20% of all shots from deadballs come from outside the 45; only 2.5% of shots from play come from the same sectors.

The below graph shows the returns from play for each game situation along with a cumulative return. Looking at the cumulative return there appears to be an argument that the further ahead, or behind you are then the more magnified the effect.

Chart1

There is a lot of noise in the above graph so if we concentrate on the -8 to +8 range, which accounts for 89% of all shots taken, we can see the effect more clearly

Chart2

There’s quite a bit in this graph. Taking shots when teams are ahead (red line) first you can see that apart from when team’s are 4 points up the returns are grouped relatively tightly together around the 49% average (I cannot give a reason for the drop at 4 points – we’ll just have to accept that it is random for now).

Shooting whilst chasing the lead (blue line) is altogether more volatile. Success rates for teams that are 2 or 4 points in arrears are quite high. One reason for this could be how teams defend. Teams drop back ‘protecting’ the goal offering up shots, with less pressure, from further out than they would normally as they know even if the opposition scores they still have a one score cushion. There may also be a mental cushion for the shooter – his is not the shot that “must” be made.

A similar dynamic could explain why it is that shots for those teams 3 points behind are significantly worse – with a less than optimal chance players might shoot for goal to draw level from positions where they would normally tap the ball over for a point.

Affect on weighting

As described above the weighting is based primarily on a sectoral basis. As such we have to ensure that there is a good spread (enough data points) across all the relevant sectors before making any major adjustment – and amending the weighting for game situation would be a major adjustment.

We only have 867 shots from deadballs, which means that 4 of the 9 sectors have a volume of shots less than 100. I’m reluctant to make any changes to the weighting based on such a sparse spread.

The spread for shots from play is much better. Firstly we have more shots (2,523) and secondly there are primarily only 6 sectors used. Sectors 1, 2 & 3 only account for 2.5% of shots. I would like to make adjustments within point groupings ( 1-3, 4- 6 etc.) however we just don’t have the volumes. As such for the 2013 Championship we’ll add game situation (behind, level, ahead) to the weighting for shots from play for Sectors 4 – 9.

Review of Basics

April 29, 2013

For those new to the blog, or who haven’t visited in a while, below are some definitions as to the various terminologies used as well as some background as to how the numbers are created.

Definitions
Possession
Any time a team has control of the ball (can come from a kickout, a throw in, gathering a breaking ball from your own shot or a turnover). “Control” at times can be subjective but ~98% of the time it is clear. Your possession ends with either (a) a shot or (b) the opposition gaining control of the ball.

Up until the start of the 2015 season possession was defined as having control inside the opposition’s 45. From 2015 onwards that is now an attack.

Attack
When a team has control of the ball inside the opposition’s 45 or takes a shot from outside the 45 (without having crossed the 45).

Attack Rate (%) = (attacks / possessions)

Shots – unlike mainstream calculation of shots (wides + scores) this includes all shots including those that are blocked, saved, hit the woodwork or drop short. 99% of the time this is straight forward however occasionally a call has to be made as to whether a shot was indeed a shot or a lob into the square (prime example being the debate around Fionn Fitzgerald’s injury time equalizer in the 2015 Munster final).

Shot Rate (%) = (shots / attacks)

Scores = (points + goals)

Success Rate (%) = (scores/shots)

avgs – these are taken from 74 games charted in the 2012-2014 Championship seasons. I also charted 36 games in the 2010 season, which are no longer used but would have fed in to the earlier calculation of averages & weighting. The averages are updated yearly as we get more data – note that if reviewing older games I do not retrospectively update the averages. Those were the applicable averages at the time the piece was written.

Deadball – free, 45, sideline or penalty

Weighting (also called “Return Vs Expected” or ”Vs Expected”)
Ultimately whilst useful none of the above stats go any way towards informing us about the type of shots taken. If the average return for shots from play is 46.5% how good a day did the player scoring 3 from out on the wings have? Similarly is getting 4 from 8 (50%) from in front of goals an “average” day? The weighting attempts to rank these two very different shooting displays against the average (how the weighting is created can be viewed here).

In its first incarnation the weighting looked at two variables
• Where on the pitch the shot was taken from. The pitch is divided into 9 segments (see here) using the large parallelogram on the vertical and the 21m & 45m lines on the horizontal. These were used as they are constant throughout all the pitches and easily identified on the TV.
• Whether the shot was from play or a deadball

Doing the blog it was obvious that, whilst instructive, there wasn’t enough in just those two variables. During the years two relatively easy and obvious tweaks were made
• Shots from 45s were originally treated separately from other deadballs. I have since weighted sideline & penalty attempts separately as well. There are now five categories of shots; frees, sidelines, penalties, 45s and shots from play
• Shots from sector 8 were divided into shots at goal and shots for points.

Note once I make a change I have not gone back and reran the weightings – bear this in mind if reviewing old(er) entries.

Kickouts
A kickout is “won” by whomever emerges with control of the ball.

Length wise kickouts are divided into
“short” – landing inside the 45
“mid” – landing between the 45 & 65
“long” – landing past the 65
Again this can be subjective as a player catches a high ball around the 65 or the lines are not obvious on the broadcast.

Sometimes the width can be described as “right” or ” left”. This can be accurate to gauge as there is no markings out the field however generally the D is used as the marking so anything left og the D is ” left” etc.

Turnovers
Only when the opposition gain control of the ball is a turnover considered to have occurred. The type of turnover can be very subjective – when a long pass is given, and misses the target, is that due to the player passing in the ball or the player receiving? If it is a long ball into Donaghy on the square – and it lands on the square – is that the pass? Donaghy losing a contested ball? Or just good defending?

This is being collated at player level however I generally don’t publish given the above points. You can only see so much from TV coverage.

The case for Return V Expected

November 15, 2012

I started the blog in May this year by, for want of a better term, throwing up the data I had. I knew that if I didn’t start the blog then, prior to the Championship starting, it just wasn’t going to happen.

Part of the raison d’etre for the blog was to find a method for ranking attacking performances – the old nugget that not all scores are equal therefore two players scoring 0-4 each did not necessarily have the same impact. That method was “Return V Expected”

I didn’t push this return during the year as (a) I hadn’t really explained how it was constructed and (b) obvious deficiencies in the process presented themselves as the season progressed.

I hope my last post shows the basic math behind the principle whilst I have updated the database to account for some of the glaring weakness. Namely
–> Returns from 45s are now treated separately from frees
–> Shots in Sector 8 that go for goal are treated separately from those that go for a point.
–> There isn’t enough data on Penalties (12 charted in 2 years) and shots from sidelines (19 charted) to treat them separately. Penalties are treated like shots on goals whilst sideline balls are included in frees

Despite my initial reticence I still believe however that using “Return V Expected” is a better measure of a player (and teams) performance than the raw scoring return. Take the below examples from the 2012 Championship

Player Game Shots Scores Success Rate Return Vs Expected Ranking
Ben Brosnan Dublin V Wexford 8 4 50% 0.7501 62nd
Jamie Clarke Armagh V Tyrone 8 4 50% -0.0731 236th

Both players took 8 shots with both scoring 4 points. Conventional scoring/ranking would equate these as being comparable returns. However when we break down where their shots, and scores, came from we can see that Brosnan took his shots from much more difficult positions on the pitch. Clarke’s were much closer to goal – an average county player would be expected to get more of Clarke’s kicks than Brosnan’s

B Brosnan J Clarke
Inside 21m 1/1 4/7
21m – 45m 1/5 0/1
Outside 45m 2/2

Clarke didn’t have a bad day – in fact he basically had a bang on average day – Brosnan had a good day scoring 4 points from positions that average county players just would not score from. Traditional ranking/analytical methods wont show this.

Attached below are the Scatter charts, for frees & shots from play, for all returns charted in 2012. Given that we are comparing everything to the average they look as expected … the ‘line’ crosses the X axis (An average Return V Expected = 0) at approximately half the volume. These outcomes again strengthen my belief that using “Return V Expected” will better rank performances – the outliers especially tell us exceptionally good (or poor) performances.

The -3 on the left hand side of the ‘Frees’ graph was Ben Brosnan’s horror show from dead balls against Dublin (the same game outlined previously when his shooting from play was very strong!). He had 4 frees and a 45 but missed all 5. Three of the frees were from Sector 6 which is directly in front of the posts whilst the other free was inside the 21m line. Knowing the high success rate there is for frees Brosnan got hammered for missing relatively simple kicks.

Explaining “Return Vs Expected”

November 5, 2012

“One running back runs for three yards. Another running back runs for three yards. Which is the better run? This sounds like a stupid question, but it isn’t. In fact, this question is at the heart of nearly all of the analysis on Football Outsiders.”

 This quote comes from Football Outsiders’ first line describing their propriety DVOA stat that compares one NFL play to another. They go on further to state that

 “Several factors can differentiate one three-yard run from another. What is the down and distance? Is it third-and-2 or second-and-15? Where on the field is the ball …. Conventional NFL statistics value plays based solely on their net yardage.”

 (If you want to read more go to http://www.footballoutsiders.com/info/methods. If you are just interested in the NFL I would strongly recommend that you visit the site and use it as a counterweight to the traditional analysis on the game.)

 The site got me thinking on how we view GAA and especially the scoring in a game. Bernard Brogan gets 0-06 in a game against Louth with 3 from frees. Darren Clarke gets 0-08 from frees in the same game. Has Clarke had a better game just because he scored more frees? Were all the frees in front of the posts? What if Brogan’s 3 points from play were from out on the sideline? How do we compare the two performances? As such I started to chart games with a view to comparing team’s attacking play.

 

Every shot charted is broken down into two main component parts

         Type of shot: whether the shot was from play or a deadball

         Segment: where on the pitch the shot was taken from

 This led to 18 categories of shot (9 segments x 2 types ). Once all the shots had been compiled I was then able to give an average success rate for each of the 18 shot categories. This is, inelegantly, explained in the very first few posts on the blog.

 The next stage was to attempt to give a value to each shot. We know a shot from out on the wing is harder to convert than a shot from in front of the posts on the 21m line. But by how much? And conversely when a player misses a shot how do we compare misses across the segments and against success in the same segment? We must chart failure as well as success.

 Again I turned to Football Outsiders. The basis for some of their returns is comparing a play to the average and working from there. After compiling enough data (3,392 shots in 61 games over the 2010 & 2012 Championships) I was in a position to compile an average success rate for the 18 shot categories. This meant that I could create a return for each shot by comparing the outcome to what was expected (expected in this case means the average for each of the 18 shot categories). In the absence of any meaningful name I christened this return the “Return Vs Expected” (or “Vs Expected”). How this return is compiled is explained below

 Return Vs Expected

The average return for a shot type within a sector = x%. This is then represented as a decimal. Eg if the average success rate = 65% this then becomes 0.65

Weighting for a score = (1-x) or (1-0.65) or +0.35

Weighting for a miss = (-x) or -0.65

 

Using a concrete example ; the average return for frees from Sector 5 is 84.5%(131 scored from 155 attempts). This is converted to a decimal, so that 84.5% = 0.845 therefore

A shot that scores = (1 – x) or (1 – 0.845) = +0.115

Whilst a shot that misses =  (-x) or -0.845

 

The average player will score, out of 100 attempts, from the same free 84.5 times. So therefore the more times a player hits that shot over 84.5 the more positive a return he will have and the more times he misses the more negative a return he will have

Same shot is hit 95 times = (95 * 0.155) + (5 * -0.845) = +10.5

Same shot hit 90 times = (90 * 0.155) + (10 * -0.845) = +5.5

Same shot hit 85 times = (85 * 0.155) + (15 * -0.845) = +0.5

Same shot hit 80 times = (80 * 0.155) + (20 * -0.845) = –4.5

Same shot hit 75 times = (75 * 0.155) + (25 * -0.845) = –9.5

 

We now have a method for giving a weighting to every shot – thus we can compare (and rank!) very disparate performances.