Dafuge & Data: Getting PPDA into your FM21 save
I mentioned in a previous post that one of the things I’m interested in is getting PPDA into the game. I’m not the only one and recently it has been mentioned by Football Manager Therapy in their podcast, and by FMStag in their recent thread on stats in Football Manager. I’m sure there are plenty of other people who want it in too.
If you’re not aware PPDA just means Passes Per Defensive Action. At it’s most basic it is just a ratio of the number of passing actions made by one team, in comparison to the number of defensive actions made by the other. The more pressure being put on the team in the possession, results in more defensive actions per pass, bringing the ratio down. High pressing teams get low PPDA scores, and teams that do not press as much tend to get higher scores. The value just tells you how many passes are allowed on average before a defensive action like a challenge happens.
So teams can use this to assess how much pressure they are putting the opposition under. With my Scarborough save I want to put teams under a lot of pressure, and I want to disrupt the game and break up play. I don’t want the opposition to have lots of time and space to pass the ball well. So I’m wanting a low PPDA value.
The eagle-eyed, and anyone who watched the video above, will have noticed that I’ve mentioned actions. This sounds quite vague. Passing actions can be successful, unsuccessful, intercepted passes, and passes that end up just out of play. Likewise defensive actions aren’t just tackles. These can be challenges both successful and unsuccessful, fouls, blocks and interceptions.
It’s an increasingly useful real world stat. But it’s not one in FM21. Luckily all the data you need to work it out yourself is. I don’t know if anyone else has done it yet but I’m going to go through my process for working it out in FM21.
We have to set out a few general criteria for our PPDA.
First of all how are we going to calculate it? Well basically we are just going to use the simple ratio suggested in the Coach Rees video above. The number of passing actions (PA) divided by the number of defensive actions (DA). We’re not going to do anything fancy like weighting different actions more or less depending on importance. We don’t have a model to work from so any weightings we gave would just be guess work. The only way around this would be to run some modelling/statistical analysis using hundreds of games of data to work out likely weightings of individual actions for the different actions. I can do it, but it would take so long I’d still be doing it when FM23 arrives.
So for us it’s simply PA/DA.
The next thing to consider is how much of the pitch do we want to use? As the video above suggests there are some differences out there. We are going to align ourselves with the restrictions we have in the game. In FM21 you can define the pitch in the following ways:
Everything purple is the selected pitch area. But you can see that it is broken down into sections. Defensive half, middle third, and attacking half. There’s also the option to split the pitch along the wings as well.
We are just going to focus on the defensive half and middle third for the defensive actions, and the attacking half and middle third for attacking actions. They both essentially overlap for the attacking and defending team.
We could use the whole pitch but as PPDA is meant to be measuring pressing unless you press the opposition everywhere across the pitch it is not needed. Even with a high press we tend not to do that. Using the full pitch would skew the results as plenty of passes in the attacking teams own defensive third are not going to be challenged by the opposition. If you feel like using the whole pitch though then all you have to do is just keep the whole pitch selected.
For our passing actions we are going to be recording:
- All successful passes
- Unsuccessful passes
- Intercepted passes
- Passes made that went out of play
We are also going to consider crosses in this as well and just treat them as a type of pass. So again, both successful and unsuccessful will be counted for this.
Extracting Passing actions from FM21
Extracting makes it sound like some sort of surgical procedure is going on. Or that you’re going to become some sort of elite hacker. In reality all the data is in the game we just need to manually collect it. It’s just a bit dull and time consuming to do so.
That said, once you get into the habit of doing it it probably only takes a minute after each game to collect the information.
If you go to a result in your schedule you can select the scoreline to bring up the stats from the fixture. Over on the right we have the option to select match analytics.
From here we have a couple of options, Team or Player level stats. For this we only need Team. We only need player if we think some of the stats aren’t being recorded correctly at the team level. This can happen. One of the things often raised (and often ignored on the official forums) is that sometimes the stats and match events just don’t add up. If you see a team level stat that doesn’t seem quite right you can go to player level stats to double check.
Lets assume the team level stats are okay though. They have been when I’ve been doing it so far (bar one or two elements – see later in the post). The next thing we need to do is set the pitch areas so we are only collecting data for the correct areas, for the correct teams. As we are looking at passes first we need to select the middle of the pitch, and the attacking half. This will align with the attacking possession of our opposition. In this example it is Brighton. If you’re not sure about whether you have the right areas selected click on events like shots or passes and you’ll be able to see the direction of play/attack.
With the right pitch areas selected and Brighton selected (or whoever the attacking team are) you can then open the stats menu’s on the left for passing and crossing. You can see the break down of successful, unsuccessful and out of play events. Just add these up and make a note of them.
Importantly don’t include key passes or passes received in the final number. Passes completed is the same as passes received. It just highlights the passer or the receiver on the pitch depending on which you select. Using both will just double up and artificially inflate the final passing actions number. Likewise key passes are just passes that are been highlighted due to the chances created/xG value resulting from them. So including those doubles counts them.
Now we can count up the defensive actions. For our purposes these are going to be:
- Successful tackles
- Unsuccessful tackles
- Headers won
- Headers lost
A missed tackle or header is still pressure. Likewise with a foul. There are some areas of debate depending on your views (or model of PPDA) but simply you can tailor this list to fit your idea of what counts as a defensive action (relating to pressure). If it’s in the game and you can count it, you can add it to the list.
Extracting Defensive actions from FM21
This time we need to make sure we have the defending team selected (so the mighty Scarborough in this case) and the right areas of the pitch, the defensive half and the middle third. Again if you’re not sure select a few events, and even play the highlights, to check you’ve got the right team and areas showing.
Again, it’s important to note that we don’t need to include key tackles or headers. This will double count. Another issue is that missed interceptions and missed headers don’t ever seem to populate with a value. My team are good in the air but I doubt they’ve not missed a header or an interception across a full Premier Division season.
The PPDA Spreadsheet
Once you have the total passing actions, and the total defensive actions all you have left to do is divide. Or you can be slightly fancier and use a spreadsheet.
Behold. The rough and ready PPDA spreadsheet. Enter the individual data for the different types of passing and defensive actions and the spreadsheet will chug along and give you the PPDA value. It also keeps all the different actions recorded for you incase you want to break down your analysis even further.
As I mentioned above interceptions missed and headers missed doesn’t seem to populate with data. You could almost certainly get rid of them.
I’ve also use a quick formula so that the number of interceptions from defensive actions populates the interceptions in the passing actions. The two should really be the same value.
You can get the basic spreadsheet here. It’s fairly straightforward. Add and change what you want.
PPDA in my Scarborough Save
In its’ beautiful raw form it’s difficult to take too much from this. I’ve got data for 27 games. As I mentioned in the last Dafuge & Data post we made some tactical changes at the Chelsea game, which including playing with our lines of engagement. So I’ve only looked at data from this point onward.
We’re averaging over 100 defensive actions a game, and the opposition around 380 passing actions, giving us a PPDA on average of 3.37. The opposition make 3 passing actions before we put in some sort of defensive action. That sounds reasonable to me.
On its’ own the PPDA stat is more descriptive of a style. It doesn’t tell us how effective our pressing game is in keeping the score down or helping us win for example, it just tells us how aggressive our pressing game is. For example you can see on there that two of our lowest PPDA’s, our most aggressive or effective pressing games, resulted in a loss and a draw (Brighton and Chelsea). We pressed a lot but they were able to beat us. Whether that was luck or because the cut us open with their 2nd and 3rd passes easily we don’t know yet.
Where PPDA can really come into its’ own for us though is if we pair it with other stats. Does it correlate with xGA or goals conceded? What about xG?
PPDA and xGA
Beautiful. If we plot xGA versus PPDA we find a fairly strong and statistically significant positive correlation (r=.31 p<.001 for any nerds in the know).
Simply put as the PPDA value increases so does the xG against. Essentially the less pressing (the higher the PPDA) the more xG against we concede. There’s only one freak win when the PPDA is above 4 for us.
For us an aggressive press seems to work quite well at keeping down the xGA and as a consequence of that improve our chances of winning. Maybe not a huge surprise given FM21 alleged bias towards gegenpressing but something we know for sure with our team and our tactic. Even though we are minnows and don’t have players anywhere near as good as the other sides in the league if we press aggressively we can increase our odds of survival.
When you break down the defensive actions you can see that fouls account for around 20% of our defensive actions every match. We are being very disruptive and to be honest we often get that news item about how we stopped other teams from playing with constant fouling. Now we know those fouls are part of some decent PPDA numbers keeping the xGA down.
PPDA and xG
Not quite as strong a relationship. Again if you’re feeling nerdy it’s not quite significant, or as strong, but it indicates a potential positive correlation again.
As PPDA increase so does our own xG value per game. By pressing so aggressively we are not only disrupting the opposition but gaining possession. By gaining possession we are then giving ourselves the chance to attack.
So we are intercepting about 10% (being kind) of the oppositions passing actions, so directly gaining possession. What would be interesting to see here is how many of these tackles, blocks and aerial challenges are resulting in us recovering possession.
I didn’t even think about including in the original spreadsheet. You can add it if you want. I’m assuming this stat records all possession won in a tackle, header, interceptions, collected blocks, turnovers from passes out of play, being fouled and winning the free kick etc. If you do use it in your own analysis remember to turn the value into a percentage otherwise it’ll give you odd results.
Running a correlation between gained possession (as a percentage of defensive actions), and xG and xGA reveals what we would expected. The more possession you gain from your actions the higher the xG and the lower the xGA.
Our PPDA versus the Real World
If you’ve looked at PPDA in the real world you might be surprised by how low the values are here.
As pointed out in the 1 minute video above, the average for La Liga for example, has been about 9. Top pressing sides in Premier Division like Liverpool and Southampton were normally getting about 8. Yet we seem to be coming out with an average of around 3. I think this is just an issue with the data. The game wasn’t set up to consider PPDA so the amount of events hasn’t been balanced to fit the real world, and the amount of actions counted in real world models of PPDA might differ drastically from ours. Also as the Tifo video below explains, there’s a lot of defensive output that might not be measured.
So I don’t think we are 2-3 times better than teams like Liverpool. I just think the scale in FM21 is different. I’ve not had the time to do it but I imagine the PPDA average in FM21 is probably around 4, and we come in a bit under on average as we are pressing urgently and aggressively because of our squad attributes.
Some Practical PPDA issues
It’s simple, and it’s straightforward but there are a few things to consider.
Can we trust the data?
It’s been noted often and by many that the stats in FM21 don’t always work or line up correctly. And the match analysis screen as had issues going back at least as far as FM19 when it comes to events not linking to correct highlights.
But we just have to trust it. We have no other choice really. We have to hope any inconsistencies are well…consistent. Like with missed interceptions and headers. They’re not being counted but at least they are always not being counted.
We can also use the player level stats as a check against the team level ones. And one tip I would give is that if the stats don’t look right use the clear events button or reload the match (exit the screen and try again – you should see the match report load again).
Something you may notice is that despite choosing just the Attacking and middle thirds of the pitch you will sometimes get passes included from much deeper.
You can see a few passes here originating from deep in the attacking team’s own half, outside of the pitch selection we’ve made. I think that’s simply because they were launched into the active areas of the pitch we were interested in. I can’t see anyway around this. It ups the amount of passes considered, but it is perhaps balanced by the fact that long passes like that are more likely to be intercepted or won by a defender in an aerial battle. I’m just going to close my eyes and assume it evens out.
What about Crosses, Interceptions, and Dribbles?
Some of the actions included, or not, are potentially controversial.
Everything is weighted the same but a lot of recent stats work in football has highlight how crosses are actually pretty ineffective. They are a low success strategy or action on the pitch, especially in terms of xA or generating xG. Yet in our simple calculation they carry the same weight as a pass, a usually much more effective pitch action. For that reason you might want to remove crosses. I’ve kept them in because although we don’t know what weighting to give them, they should have something. I’d rather have them in and over estimate the passing actions than remove them entirely and under estimate.
Another issue is interceptions. Remember here that defensive actions are meant to be acting as a proxy measure for pressing or pressure. Tackles, aerial challenges, blocks all fit the bill. Interceptions could do if they are a result of defenders filling space and channels, or closing down quickly and getting into a position to collect a ball. But, there’ll also be interceptions recorded in there that are actually from speculative long balls that are intercepted normally, without being the result of pressing behavior. There’s no way for us to separate the two without painstakingly going through each interception highlight. I can’t be arsed with that.
Now the final point is about dribbling which might seem odd. PPDA is about passes per, not dribbles per, but again remember this is meant to be a measure that indicates the effectiveness of a team’s pressing game. A player dribbling the ball is still going to be closed down and tackled, and you expect that to be done aggressively so under most high press tactics. I’ve not included it but you could if you wish include dribbles/carries in there. It is one of the categories recorded in the match analysis.
Will PPDA ever get into Football Manager?
Obviously I don’t know. And I don’t know how easy it is to code or not. But given that all the data needed exists, even down to pre-set divisions of pitch area, I find it hard to believe it’s something that can’t be done.
Maybe in practice it is difficult to do but I feel like it should be a simple job for the game to call the information from the categories, for the teams, for the pitch areas needed, and just work out two sums and a division. It’s a glorified series of IF THEN functions.
Until it does I’ll keep spending 30-60 seconds looking at it the game and adding it to the spreadsheet.
If you liked PPDA…
If this hit the stats spot for you check out some of the other posts on the Dafuge & Data series, such as solving problems like a rightback, choosing stats and violence, DNA, and my striker health check.
I’ve also written about getting xG manually before we had it in Football Manager. And for Dictate the Game there are some articles about working out your DNA, using xGA to evaluate keepers, and working out expected assists (xA). As well as experiments and investigations into the impact of professionalism, different striker traits, and the impact of the medical team and physio on success.