Good news everyone. I have hit upon a genius idea to make my spreadsheets more accurate. Simply add 1 to every team’s predicted score, to take into account the effect of the new handball rule!

I have previously cited some grounds for treating my spreadsheet’s predictions with caution at the start of the season (the impact of new transfers; the suspect form of several teams post-lockdown; and, the unknown quantity that are newly promoted teams), but it became clear last weekend that I’d missed the most distorting effect of all. The new handball rule is hands down (or up, away from the body, and outside the body line) the single biggest factor making results and scorelines so unpredictable this season.

After three gameweeks, my spreadsheets are yet a register a single correct score forecast! Last season, they managed an average of 2 per week, which may not sound like many, but is actually decent. That said, they had the BHA 1 MUN 2 scoreline right, up until the 4th minute of additional time, and the CRY 1 EVE 1 prediction would have been two out of two if you discount the ludicrous penalty awarded to the visitors because Joel Ward had not superglued his arms to his sides!

Who saw MCI 2 LEI 5 and WHU 4 WOL 0 coming, though? Not my spreadsheet, that’s for sure, but then again, who did? FUL 0 AVL 3 and BUR 0 SOU 1 were the other results to diverge from my model’s predictions markedly.All I have to do to cheer myself up whenever I am feeling disappointed by my spreadsheet’s limitations, however, is compare its predictions with those deemed most probable by bookmakers. Half of the predictions tallied, but the bookies were out by more in 4 of the other 5, and this was enough to make the mean absolute error (**MAE**) and mean square error (MSE) of their predictions considerably worse than my spreadsheet (see below).

By way of comparison, my spreadsheets averaged a MAE of 0.8 last season, which is actually very respectable. Note, the lower the MAE the better. For ease of understanding, a MAE of 1.0 would equate to being one goal out (on average) for each team’s score prediction.

In other good news, the expectation is that my spreadsheet will become progressively more accurate with each passing gameweek, so let us turn now to the predictions for **GW4**. Here are the scorelines suggested by putting together each team’s most likely number of goals (see below). The probability of each scoreline is shown alongside.

17.8% for **WOL 2 FUL 0** is the highest probability my spreadsheet has given to any result this season so far. That said, WOL 1 FUL 0 has an even higher probability of 20.2%! Remember, the method used to arrive at these correct score forecasts doesn’t necessarily equate to the scorelines with the highest probability. In fact, only 4 of these are deemed to do so. The highest probability for the other scorelines are as follows: EVE 1 BHA 1; LEE 0 MCI 2; SOU 1 WBA 1; ARS 1 SHU 1; and, AVL 0 LIV 1.

So, you are probably wondering why I don’t just list the scorelines with the highest probability? The short answer is because the method I prefer yields better results. Last week, for example, going with the highest probability scorelines would have fared worse than the bookies with a MAE of 1.42.

The long answer would invite you to take the example of **ARS vs SHU** this weekend to understand why I prefer the method I use. Whilst it is true that 1 – 1 has a higher probability than 2 – 1, it is also true (paradoxically) that ARS have a higher probability of scoring twice than they do just the once, and I deem that to be the deciding factor.Further good news for **GW4 **is that there are 8 teams deemed more likely to score twice than once and, in descending order of expected goals, they are shown in the table below:

As highlighted by my Season Review blog earlier this month, Palace are nowhere near as defensively resolute away from home, and **Werner **sellers may come to regret their impatience this weekend. Last week’s **Aubameyang **sellers may likewise be nervously eyeing his forthcoming fixture vs **SHU** who will be without O’Connell. Meanwhile, **Jimenez **buyers disappointed by his blank vs WHU last weekend can be hopeful of belated returns vs **FUL**, the league’s current whipping boys. There is plenty of encouragement here too for owners of the usual suspects from **SOU**, **EVE**, **MCI**, **MUN **and **LIV**.

In further welcome news for **Jimenez **owners, **WOL **also feature in the top 5 projected highest scoring teams in 2 of the next 3 gameweeks (see below).

Courtesy of my spreadsheet’s player points prediction model, **Rashford **is my sole **MUN **attacker, and I was pleasantly surprised just now to discover that he is the least owned of the Reds’ front four. Especially so, given that he is at the summit of my spreadsheet’s **GW4 **expected FPL points table (see below). Some of you may remember that **Fernandes **was ranked highest of the quartet last week, and justifiably so as it happened.

Scepticism was expressed last week about **Sterling **not featuring in the GW3 table, and I anticipate similar protests this week about the notable absences here of **Fernandes**, **DCL**, **Jimenez**, **Aubameyang **and **KDB**, so for context they are provided in the *best of the rest* table shown below.

Remember, these are averages whereas players points tend to polarise between high and low. So, assuming no clean sheet points, a midfielder scoring a goal and earning a bonus point one week (8), but blanking the next (2), would have averaged 5 points.

At the other end of the pitch, meanwhile, there are much stronger candidates for clean sheets this week than there were last, although the unusually high probabilities assigned to **WOL **and **CHE **(see below) does fly in the face of the fact that they conceded 7 goals between them in GW3! The hope is that the 4 goals shipped vs WHU last week was just a bad day at the office for the former, whilst the chalk and cheese performances of the latter’s defence playing at home rather than away was something else picked up on by my Season Review blog.

Historically, my spreadsheet’s clean sheet probability calculations have been its strongest suit, so it was disappointing that none of the 4 teams highlighted in green last week (MUN, LIV, MCI and CRY) managed to keep a clean sheet.

I kept a record last season, up until lockdown, of how my model’s clean sheet probabilities fared against those implied by bookies odds, and made popular by @FPL_Salah, and on only 2 out of 12 occasions did the bookies do better than my spreadsheet (1 tie). Last week, the bookies were on average 1% more accurate though, so that makes it 3 from 13 for them now. Time has not permitted me to look back at how they fared over the first 2 gameweeks, so I will report back on those after the International Break.

Just when confidence in the **WOL **backline has taken a knock, they actually top my clean sheet probability projections for the next 6 gameweeks, so it will be interesting to see how they fare this weekend before committing to a possible double up in GW6.

One of the promoted teams are currently riding high in this same table in 4th place, which has me sitting up and taking notice, and bodes well for me making savings in my team’s defence budget.

Another thing I tested for last season was the accuracy of my spreadsheet’s longer-term forecasts, and the results were very encouraging indeed. I compared the projections for the following 5 gameweeks with current gameweek only ones provided by “the world’s most powerful predictive *fantasy football* algorithm”. And, despite that model having the informational advantage of being up-to-date prior to each of the following 5 gameweeks, my spreadsheet outperformed it in each category I ran correlation tests for (number of goals teams scored, correct score forecasts, and mean absolute error). In other words, even from 5 gameweeks away, with 4 intervening ones taking place, my spreadsheet proved to be more accurate.

May the GW4 flop be with you!

May your arrows be green!

Coley a.k.a. FPL P0ker PlAyer (@barCOLEYna)