Last week I published an article which showed that in season 2018/19 the Elo FDR was better at predicting fixture results than the Official FDR. I now carried out an additional test to see if there was a correlation between FDR rating and fixture goal difference.
Again, I would have preferred to test if there was a relationship between fixture difficulty and total FPL points a team’s players accumulated in the fixture, however, for now, I did not have this data.
Test: Goal Difference
In a fixture where a team had a good FDR rating, did they have a high positive goal difference? Likewise, in fixtures where teams had poor FDR ratings did they have high negative goal differences?
The FDRs changed throughout the season and looking back, the only time I had comparable ones was at the start of the season. Therefore, I used the Official FDR dated 9thJun 18 and the pre-game week 1 Elo FDR. (A copy of the Official FDR was provided by @FantasianPL).
Also, an attempt was made to compare the start of the season FDR performances with the end of the season. As the FDRs were from the start of the season, it was expected that the correlation would be higher at the start and would fall off towards the end of the season.
For both FDR’s and from the home teams perspective the following graphs were plotted and R2 values calculated.
- GW 1 – 38 Goal Difference vs Official FDR rating
- GW 1 – 38 Goal Difference vs Elo FDR rating
- GW 1 – 6 Goal Difference vs Official FDR rating
- GW 1 – 6 Goal Difference vs Elo FDR rating
- GW 33 – 38 Goal Difference vs Official FDR rating
- GW 33 –38 Goal Difference vs Elo FDR rating
A few notes on the graphs, some of these will be obvious but are included for clarity:
- The input, FDR rating was plotted on the x-axis.
- The output, fixture goal difference was plotted on the y-axis.
- Positive points above the x-axis represented home wins.
- Negative points below the x-axis represented home defeats.
- Official FDR, points to the right of the y-axis represented difficult fixtures.
- Official FDR, points to the left of the y-axis represented easy fixtures.
- Elo FDR, points to the left of the y-axis represented difficult fixtures.
- Elo FDR, points to the right of the y-axis represented easy fixtures.
GW 1 – 38 Goal Difference vs Official FDR rating
- Official FDR trend lines had negative slopes as good fixtures had low FDR values.
- Elo FDR trend lines had positive slopes as good fixtures had high FDR values.
- This graph highlights the granular nature of the Official FDR and the fact no fixtures were rated as 1.
GW 1 – 38 Goal Difference vs Elo FDR rating
- The less granular nature of the Elo FDR is shown in this graph.
GW 1 – 6 Goal Difference vs Official FDR rating
GW 1 – 6 Goal Difference vs Elo FDR rating
GW 33 – 38 Goal Difference vs Official FDR rating
GW 33 –38 Goal Difference vs Elo FDR rating
R2 Summary Table
R2 is a measure of the correlation between two variables and can range from 0 to 1. In this case, the variables were fixture FDR rating and actual goal difference. The highest R2 in the results was 0.3715 or 37.15% for Elo FDR over the first 6 game weeks. This meant that 37.15% of the variability in goal difference was due to the variability in Elo FDR rating.
- For all game week ranges R2 was higher for the Elo FDR than it was for the Official FDR. This showed that the Elo FDR was a better model than the Official FDR. Although, the correlation was not great, it was still better than the Official FDR.
- As expected, R2 was higher for GW01-06 than for the full season and for GW33-38. This is was true for both FDRs. This confirmed the result shown in the first article, in that the further they look out the less accurate they become.
- The correlation was not as strong as I had hoped and maybe this gave credence to the theory, that some manages have, about form over fixtures.