4th Down Percentage Decision Matrix Feed Back Needed

[AlexGreen#20]

Alright everybody, let's get some thick rimmed glasses on and our calculators out, because I've got some heavy nerd **** for you all here today.

I went into this with the premise that I wanted to see what the numbers say about going for it on 4th down vs. Punting vs. Kicking Field Goals.

I'll start out by saying that this is far from a finished product. There's a lot of room for error left in this work and there's a lot of room for the incorporation of new data as well. In the meantime, here's what I've put together. 

+++

I started out by tracking down the conversion rates for 4th down by down and distances for the first 5 weeks of the 2017 season.

My conclusion was that I didn't have near enough data to be making any kind of conclusions. Also note that the row for 11 yards should be 11+.

So I moved on to 3rd down conversion rates. My thought process was that basically a failure on either down typically results in the other team ending up with the ball. Therefore rates should be about the same.

This was a little better data in the sense that there were far more sample sizes, but I was concerned with how much higher the 3rd down conversion rates were than the 4th down conversion rates.

I had three possible rationalizations for the 4th down data being so much lower:

1. Defenses play much more aggressively on 4th down in comparison to 3rd down where the difference between a successful stop is a turnover rather than a punt.

2. A lot of 4th down scenarios occur in end of game or end of half situations, where there's no real incentive to converting the 4th down without getting into the endzone or at least getting chunk yardage, so offenses play with far less emphasis on converting the down and distance.

I choose to assume #2 was the most predominant reasoning because thatway my data wouldn't be entirely invalidated before I even started. Suck it Scientific Method.

I then took that data, and put it into a graph, in order to get a line of best fit.

The small blue dots making up a curved line are the plotted equation that I pulled in order to graph the system. The orange dots are the actual numbers pulled from the 3rd down conversion table above. It's not a perfect match, as it has a tendency to overestimate the conversion percentages in the 5 to 15 yard marker, but it smoothes out the data enough to use it as an adequate guess.

Once I had my conversion data figured out, I turned to an attempt to measure the expected drives based on the drive starting position. I gathered drive data from the first 5 weeks of the season, and I determined the following. 

I then put these points into a graph and a I ran another linear regression to find a model worth replicating. I got the following graph.

Basically what this tells me is that of all the drives that saw the ball get to the 50 yard line (excluding long plays like an 80 yard run, the offense will average 3 points on that drive.

With the conversion rate numbers calculated, and the expected points per drive determined, I needed to come up with a way to use this data to tell me something relevant. what I settled on was a decision matrix for comparing the decision to Punt/Go For It/Field Goal Attempt.

In order to do that, I needed to determine the approximate value of Field Goals and Punts.

Field Goals were easier. So I started with that one:

Basically Field Goal kickers are getting disturbingly close to automatic. When guys are hitting 70+% of their 50 yard field goals, that's crazy. If you want 

Punts were a little bit more challenging. I had to use my expected points per drive formula on the opposing team. 

I ran a regression for all punts kicked closer to the goal line than 59 yards, and then ran another regression for kicks kicked further from the goal line than that in an attempt to isolate drop punts from distance punts. 

Once I had that data, I could use it to determine the value of the opposing team starting on a certain yard line.

That became the basis for the rest of my analysis. I looked at all the outcomes through the lens of one of our drives, and then one of the oppositons' counter drives.

When evaluating Go-For-It situations I ran with the equation:

Conversion Percentage=%Chance of Converting a 3rd down based on the given distance from the first down marker

Expected Offensive Points=Result of the Points/Starting Position Regression based on setting the starting marker 1 yard beyond the first down marker.

 

Expected Defensive Points(Failed Go-For-It)= Results of the Points/Starting Position Regression based on setting the starting marker at the original line of scrimmage.

Expected Defensive Points(Succeed Go-For-It)=(% of scoring Drives at the new LOS)*(Results of the Points/Starting Regression, setting the starting marker at the 25 yard line)+(% of Punt Drives)*(Points/Starting Regression based on setting the marker forward 15 yards from the converted line of scrimmage and then adding the expected punt yardage)+(% of Turnovers)*(Points/Starting Regression based on setting the marker at the original LOS)

NET POINTS GO FOR IT=(Expected Offensive Points)*Conversion Percentage+Expected Defensive Points(Succeed Go-For-It)*Conversion Percentage+Expected Defensive Points (Failed Go-For-It)*100%-Conversion Percentage

Kick Percentage=Percentage Likelihood that the FG will be good.

Field Goal Expected Offensive Points=3*Kick Percentage

Expected Defensive Points(Field Goal)=(Result of the Points/Starting Position Regression based on setting the starting marker at the 25 yard line)*(FG Conversion Rate for a kick at the LOS+9 yards)+(Result of the Points/Starting Regression based on setting the starting marker at the kick location)*(The FG Failed Conversion Rate for a kick at the LOS+9 Yards)

NET FG EXPECTED POINTS=Field Goal Expected Offensive Points-Expected Defensive Points(Field Goal)

NET PUNT EXPECTED POINTS=The result of the Points/Starting Position Regression based on setting the starting marker based on the most likely starting position of the drive based on the Punt/Starting Marker Regression.

You then compare the NET POINTS GO FOR IT with the NET FG EXPECTED POINTS with the NET PUNT EXPECTED POINTS:

Here's the 4th and 1-10 Graphs For Reference

Conclusions: NFL Teams should be going for it way more than they are based on what this data is telling me. 4th and 1 should be an automatic go for it just about anywhere. Punting on 4th and 1 between the 40s is basically unforgivable. 

And here's a thing that I made so that I can tell in the game how much Josh McDaniels is pissing me off based on his 4th down decisions.

I want to compare Crosby and Scott with the Punt and FG metrics as well. But that's for tomorrow. 

I'm tired. This sucked.

Give me your thoughts

 

[incognito_man]

nerd

[Packer_ESP]

That's some great analysis Alex. Couple of quick observations:

- In your graph, punts are always negative, and that makes sense because you're giving away possession so whatever expected points the opponent will have after fielding it are a positive for you. However, I don't know if the effects of pinning your oponent inside the 10 and the possible points from turnovers are accounted for. Granted, this would be a minimal effect and probably doesn't impact your data, but it could sway the decision one way or the other in these gray areas between the 40s.

- I maybe missing something with the expected net points formula but it's kinda weird to me that field goals max at one expected point, even from the 99. Shouldn't it be a lot closer to 3? Same for going for it, 4th and 1 from the 99 can't be less than 2 expected points, right?

- (EDIT) BTW, current score should probably be taken into account when making the decision. It does NOT impact this chart since the probabilities and points are the same, but in certain situations you can settle for a FG even if the expected points are lower because it has a higher probability of converting.

[Leader]

Damn dude. My brain hurts.

This comment drew a laugh.....

"I choose to assume #2 was the most predominant reasoning because thatway my data wouldn't be entirely invalidated before I even started. Suck it Scientific Method"

 

[Uffdaswede]

Stand back! He’s working with some very unstable herbs!

[AlexGreen#20]

3 hours ago, Packer_ESP said:

That's some great analysis Alex. Couple of quick observations:

- In your graph, punts are always negative, and that makes sense because you're giving away possession so whatever expected points the opponent will have after fielding it are a positive for you. However, I don't know if the effects of pinning your oponent inside the 10 and the possible points from turnovers are accounted for. Granted, this would be a minimal effect and probably doesn't impact your data, but it could sway the decision one way or the other in these gray areas between the 40s.

- I maybe missing something with the expected net points formula but it's kinda weird to me that field goals max at one expected point, even from the 99. Shouldn't it be a lot closer to 3? Same for going for it, 4th and 1 from the 99 can't be less than 2 expected points, right?

- (EDIT) BTW, current score should probably be taken into account when making the decision. It does NOT impact this chart since the probabilities and points are the same, but in certain situations you can settle for a FG even if the expected points are lower because it has a higher probability of converting.

As far as your first two comments, the analysis is done in a paired drive sequence.

It looks at the expected outcome of your drive based on the decision, and then runs it against the expected drive of the opposition based on the percentage outcomes.

The reason punts are always negative is that on a punt, your offense scores 0 points and their offense will score at least something. Even if your Punter pine the opposition down at the 1, the opposition is still expected to score like 1.2 points on their drive. This creates the always negative results you're seeing.

Similarly with Field Goals, even if you kick a 19 yard field goal at what amounts to an almost guaranteed 3 points, the opposition then gets the ball on their 25 after you kick off to them. The expected points of their drive, subtracted from the expected points of your FG is the net of your decision. 

4th and goal from the 99 has a guaranteed 7 offensive points if converted, and 0 points if failed at a .63ish conversion rate so you're seeing a likely offensive outcome of 4.41 points minus whatever the expected defensive drive is, probably something like 2.3 points if they're most likely getting it on their 25 after kickoff.

 

Good questions, hard to explain that on my description.

[deathstar]

This is pretty interesting. I guess the only monkey wrench is if third down doesn't equal fourth down. It's beyond the scope here obviously, but what we would want to do in an ideal world is:

Only look at fourth downs.

Parse out irrelevant fourth downs.

Look at the results of the current drive and the opponent's next drive.

So the irrelevant fourth downs are those within 2 minutes of the end of the game. We also probably want to eliminate fourth downs by teams that are trailing by 14 or more points in the fourth quarter, as teams in that situation are also playing more of a prevent defense. We also want to look at at least 10 years of this to establish enough data points to make it statistically relevant. 

Does that sound right? I don't even have excel but I am on paternity leave right now so I might devote some time to see if I can figure out how to scrape this data from nfl.com.

[MacReady]

Too long, didn't read, but I've always wanted to go for it past the 50 with our defense since 2011. 

[AlexGreen#20]

1 hour ago, deathstar said:

This is pretty interesting. I guess the only monkey wrench is if third down doesn't equal fourth down. It's beyond the scope here obviously, but what we would want to do in an ideal world is:

Only look at fourth downs.

Parse out irrelevant fourth downs.

Look at the results of the current drive and the opponent's next drive.

So the irrelevant fourth downs are those within 2 minutes of the end of the game. We also probably want to eliminate fourth downs by teams that are trailing by 14 or more points in the fourth quarter, as teams in that situation are also playing more of a prevent defense. We also want to look at at least 10 years of this to establish enough data points to make it statistically relevant. 

Does that sound right? I don't even have excel but I am on paternity leave right now so I might devote some time to see if I can figure out how to scrape this data from nfl.com.

That's one of a few items but the biggest one.

The problem I run into is that even with like 10 years worth of data, there are only like 4 examples of a team going for it on 4th and longer than 6 within the parameters you described.

[AlexGreen#20]

34 minutes ago, Outpost31 said:

Too long, didn't read, but I've always wanted to go for it past the 50 with our defense since 2011. 

One of the things I'm going to work on is putting in an adjustment for liking at quality of offense and defense for both teams.

If your offense is great and their defense is lousy, it would push the "Go For It" option. 

[deathstar]

I guess then is there any situation that a third and 6+ wouldn’t equal fourth and 6+ beyond the parameters you’ve already used? 

And in all honesty the data you found already is pretty similar to a lot of the data already out there on fourth downs.

 

https://www.nytimes.com/2014/09/05/upshot/4th-down-when-to-go-for-it-and-why.html

[Cakeshoppe]

Are you assuming a TD is worth 7 points? How many points is a TD actually worth? Or is this based on the actual point outcomes of the drives you looked at? The difference is likely to be small, but a lot of this is dealing with pretty small differences. 

[rcon14]

Oh yes this is the stuff I want to see on this forum.

Seems to fall in line with most of the research that is out there. The graphs are helpful visuals to showcase the loss in points from making poor decisions. One thing I would like to see is what teams are giving away the most free points every week. It's kind of like the old saying with the Rockets: They were up 8-0 every game from the start because they took smarter shots. If you can just make smarter decisions here, you could be up 2 or 3-0 at kickoff.