Tuesday 6 January 2015

Championship Week 24: PDO vs Shots on Target - which has the biggest effect so far?

One of the things I'm happiest to see coming together in this Championship season is teams getting increasingly rewarded for their performances in driving play.
For most of the season, unlike much of the rest of the football league, the points accrued by teams has had a far stronger correlation to its PDO score than its efficiency at producing shots at the opponents goal.
However, we are starting to see that trend correct quite strongly now, although it has not reversed itself just yet.

Position Team Played Won Drawn Lost Goals For Goals Against Goal Difference Points
1 Bournemouth 24 14 6 4 54 25 29 48
2 Ipswich 24 13 8 3 41 22 19 47
3 Derby 24 13 6 5 46 23 23 45
4 Middlesbrough 24 12 7 5 38 17 21 43
5 Watford 24 12 5 7 43 26 17 41
6 Brentford 24 12 4 8 40 35 5 40
7 Norwich 24 10 7 7 44 29 15 37
8 Wolves 24 10 7 7 28 33 -5 37
9 Blackburn 24 9 8 7 35 33 2 35
10 Sheffield Weds 24 8 10 6 18 21 -3 34
11 Nott'm Forest 24 7 10 7 33 33 0 31
12 Cardiff 24 8 7 9 32 35 -3 31
13 Charlton 24 6 13 5 24 28 -4 31
14 Birmingham 24 8 7 9 27 37 -10 31
15 Bolton 24 8 5 11 26 31 -5 29
16 Reading 24 8 5 11 30 40 -10 29
17 Fulham 24 8 4 12 35 43 -8 28
18 Huddersfield 24 7 7 10 31 42 -11 28
19 Rotherham 24 5 11 8 21 30 -9 26
20 Leeds 24 6 6 12 25 36 -11 24
21 Brighton 24 4 11 9 26 32 -6 23
22 Millwall 24 5 8 11 24 40 -16 23
23 Wigan 24 4 8 12 24 31 -7 20
24 Blackpool 24 2 8 14 18 41 -23 14


As we can see from the two graphs further below, team shots on target share now has a correlation (r2) value of 0.5363 with points won, compared to an r2 value of 0.6699 for PDO (a rather catch-all statistic for "luck") and points.
Just a few weeks ago the shots of target correlation was much lower than this.

Position Team Shots for total Shots against total Corsi/TSR shots on target for total shots on target against total Shots on target share Shooting % For Save % PDO
14 Birmingham 328 397 0.452 102 130 0.44 26.46 71.54 98
9 Blackburn 384 332 0.536 127 117 0.521 27.56 71.79 99.35
24 Blackpool 271 385 0.413 86 122 0.413 20.94 66.39 87.33
15 Bolton 304 375 0.448 97 125 0.437 26.8 75.21 102
1 Bournemouth 397 282 0.585 148 86 0.632 36.49 70.93 107.42
6 Brentford 318 348 0.477 123 120 0.506 32.52 70.83 103.35
21 Brighton 365 309 0.541 100 93 0.518 25.99 65.61 91.6
12 Cardiff 305 363 0.457 95 107 0.47 33.68 67.3 100.98
13 Charlton 254 375 0.404 83 112 0.425 28.92 75.01 103.93
3 Derby 321 316 0.504 115 103 0.528 39.98 77.67 117.65
17 Fulham 317 353 0.473 104 114 0.477 33.65 62.28 95.93
18 Huddersfield 352 349 0.502 122 118 0.508 25.4 64.4 89.81
2 Ipswich 350 353 0.498 115 96 0.545 35.66 77.09 112.75
20 Leeds 284 364 0.438 77 114 0.403 32.48 68.43 100.91
4 Middlesbrough 360 260 0.581 116 70 0.624 32.76 75.7 108.46
22 Millwall 313 317 0.497 97 111 0.466 24.75 63.97 88.72
7 Norwich 411 243 0.628 131 85 0.606 33.58 65.9 99.48
11 Nott'm Forest 351 301 0.538 101 108 0.483 32.67 69.44 102.11
16 Reading 307 309 0.498 92 104 0.469 32.61 61.56 94.17
19 Rotherham 332 342 0.493 103 101 0.505 20.4 70.29 90.69
10 Sheffield Weds 319 294 0.52 85 86 0.497 21.18 75.58 96.76
5 Watford 339 347 0.494 122 102 0.545 35.23 74.51 109.74
23 Wigan 295 283 0.51 83 97 0.461 28.92 68.04 96.96
8 Wolves 347 327 0.515 102 105 0.493 27.45 68.58 96.04

These graphs serve two purposes: first to give an overall idea of the spread of the points gained by teams throughout the league, and second, by labeling the data points we can see which teams have over- or under-achieved on points obtained compared to their performances - a potentially important method of deducing how teams are likely to perform throughout the rest of the season.


Let's start with the shots on target (SoT) vs points graph.
We can see the line of best fit dividing the points fairly evenly - some teams actually sit directly on the line, but most are varying distances away. One way we can look at this graph is to gauge the difference from where clubs should be on the linear relationship, and where they actually are. These distances are important - the further away from the line of best fit, the greater the under- or over-achievement.

In this instance, teams above the line have collected a smaller points total than their share of the shots on target warrants (more shots on target, but fewer points). At the top end of the table, we can see Bournemouth, Middlesbrough and Norwich should actually be even better than they off already, while further down the table Brighton (who I've raised many times previously) stand out as deserving of more points.
In contrast, those below the line have out-performed their shots on target share and collected more points than they deserve. Ipswich, Derby, Brentford and Wolves stand out here, while lower down the table Charlton, Birmingham, Bolton and (most worryingly, perhaps) Leeds have taken more points than they could have been expected to claim following this trend.


If we look at the PDO vs points graph we can see the teams in general more closely grouped around the line of best fit - hence the stronger r2 correlation value.
If we want to interpret this graph the same way we have the previous one, we need to assume that PDO is a linear predictor of points. While this may appear to be true at the moment, it's widely held (though not by all) that PDO is give-or-take entirely random, hence the frequent references to it being a signifier for "luck" or variance within performance.

However, in this instance, if we want to suggest that PDO does have a linear relationship (at least temporarily) then for interpreting the graph it highlights those teams that have benefited most from PDO so far this season (above the line) or suffered from it (below the line).
And again the bigger the distance, the greater the effect.
So, for example, Bournemouth has actually collected 48 points, which they should have "needed" a PDO of around 110 to do so. Meanwhile, third placed Derby's PDO is so high that the Rams should have collected around another 10 points.

As I say, much of this discussion can be considered irrelevant (if you believe PDO to be essentially random) - but it acts as a nice indicator to show those teams perhaps more blessed by good PDO during the season so far.
Interestingly, again, Leeds are quite well above the line. If both these trends correct then it could be an even messier second half of the season for the Yorkshire club.


Much of this backs up the shots on target vs PDO graph higher up the page, but I feel it helps to show some of these trends in a different way sometimes.

Finally, Norwich.

I deliberately mentioned earlier on that Norwich were one of those most under-served in the shots on target vs points comparison. The Canaries sacked/pushed overboard/accepted the resignation of Neil Adams this week, something I believe was probably more over-reactionary to a rough one month spell of PDO and a cup upset at Preston.
I'll have a separate post soon going in to more detail on Norwich's performance this season, but suffice to say, sometimes shit happens and its how you react to it that counts.
I don't think the Norwich board have reacted well at all.

No comments:

Post a Comment