Streamlining Player Evaluation with Dropback Models 🏈

Hi, I'm Parker Fleming, the VP of football strategy at Dropback.

Today I want to show off a little bit of the work we've done to make professional grade analytics and data systems available and accessible to college programs across the country.

We're talking a lot today about our model builder.

A model in Dropback is just a collection of variables with some weights that helps you formalize your player evaluation process.

No longer moving just to filtering and sort, banging your head against spreadsheets, but instead a unified system that helps you decide what's important, evaluate players and get results quickly.

Let's dive into an example.

I've been building a couple of starter kit models.

Side note, these are all in the model library.

A collection of resources where you can fork, duplicate and recreate models to help you get started, across sports in your player evaluation journey.

What I have right here is an offensive line starter kit.

So I've selected some variables and I've created different importances for these variables.

We have some physical metrics like height and weight.

We have some performance metrics like pass block grade or pass block efficiency.

We also have participation like pass blocking snaps.

How often did this player play?

All of those go into a model to help me get a baseline for how good a player is and whether they're a fit for our team.

We have some filters here as well.

We can build subsets of these models and drill in to create different standards for different position groups like offensive line, interior and tackles.

Because we want to compare players based on the relevant comparison group.

For this model, I can set the weights, my Overall weight is 100%.

I can go through each of these variables and say how important they are, what is going to factor into this model and what's going to make sense for how I want to evaluate players.

I can also go through and verify my model based on different players.

So for instance, if we pull up Tulane transfer to Houston Shadre Hurst, we can see that the interior, offensive line model applies to him and we can see his measurables and how he grades out on each of these different pieces of the model that I think are important.

We also can change the output.

Right now I've got it on a 0 to 100 scale.

We could change it to percent of expectations so that we're getting an idea of how much better than our target he is.

We also can do evaluation, an index.

We could do a tier of low, medium, high.

We could do pass fail or even a custom output.

There again, to meet your needs for how you Want to evaluate players.

Now, how do I create this model?

Very, very simply, all I do is just add variables to my canvas.

So I'll select a metric here and let's use impact run block percent.

We type in the model variable that we want.

We select the variable, then we change a couple settings to make sure we're getting the latest season.

And then we can go to the data explorer on our variable to figure out what are targets for whether a player is good.

We have filters for active in College Division 1 and we can even add another position filter for interior offensive line.

We click analyze and we'll get some results.

The data summary gives us the overall distribution.

It gives us the min, mean, median and max and it tells us the R squared, how stable this metric is year over year.

From this we can grab the median is 8.82%.

So we'll set that as our target for this variable.

And for best, the highest is 100%.

We'll set it there.

Now we see that this impacts our score.

Hurst is getting 63% from being above our target on this model.

We want to do two things.

One, because the distribution is skewed to the right, we want to use a logarithmic curve that's going to smooth out some of those outliers and help us really get signal for are you above our target or not?

Secondly, we want to check the weights.

So we go back and see that we're at 148% overall.

That's because impact run block is defaulted to 50% weight.

If we bring that down back to 100.

We see that now, Hurst is getting a much more respectable 1.53% boost from his run block and his Overall score is 94.3.

From this we can go explore and get a list of players who meet our criteria across a host of variables and put them into our scouting workflow.

On the Explore page you can quickly reference players who grayed out highly in your model.

You can click for more information.

You can add tags and even add the player to a board.

This is just the start of your scouting workflow, taking models, variables and evaluations into a pipeline that helps you evaluate better, get information faster and make better decisions with drawback.