What's the difference between a descriptive metric & a predictive metric? Why does it matter?
A thread 👇
1/ Generally speaking:
- Descriptive metrics are intended to describe what has happened in the past
- Predictive metrics are intended to provide insight into what might happen in the future
2/ Let's look at an example of each type.
A 🏀 player's Free Throw % is a descriptive metric because it quantifies the efficiency with which a player shot free throws in the past.
3/ Win Probability, on the other hand, is typically a predictive metric because it makes a prediction about the likelihood of game outcomes in the future.
4/ Why does the distinction between descriptive & predictive matter?
Because past performance/results may, for a number of reasons, not resemble future performance.
5/ Let's use an example of a ⚾ player. We'll call him Joe.
Joe has hit 45, 40, 35 & 40 home runs in the last four seasons, respectively.
Thus, over that period, his Avg. Home Runs per Season is 40.0
6/ Now let's say Joe is a free agent and I'm the GM of a team interested in signing Joe.
If I look only at Joe's Average Home Runs per Season, I might think that he's worth a 5-year contract fitting for someone hitting 40 home runs in a season.
7/ But Average Home Runs per Season is only a descriptive metric in this case, and Joe's past performance may not resemble his future performance.
Why?
A number of reasons, but one of the most obvious is age.
8/ Let's say that Joe is now 34 years old and moving out of his prime.
Should we expect Joe to maintain his current level of performance as he continues to age out of his prime?
Probably not.
9/ Because of this, we shouldn't only rely on a descriptive metric like Average Home Runs per Season when estimating Joe's value in future seasons.
We might end up "paying for past performance" if we do.
10/ What would we look at instead?
Fortunately, analytics folk spend a lot of time building advanced metrics that account for this exact type of situation.
11/ Without getting too technical, these advanced metrics take into account factors like player age when projecting future performance.
If it wasn't already clear, these types of metrics are predictive metrics.
12/ A final point: In most cases, a predictive metric will be built using descriptive metrics as inputs.
What happened in the past is still valuable in predicting the future, as long as we account for what might change (like age).
13/ For a slightly deeper dive on the difference between descriptive and predictive metrics, check out my blog post on the topic: brendankent.com/2020/12/08/spo…
• • •
Missing some Tweet in this thread? You can try to
force a refresh
There aren't many better ways to get exposure to teams (that are hiring) than to perform well in this. Also, sports analytics is fun.
More specifically to this year's topic (the secondary), many football analytics folks I've spoken to (including on @MeasurablesPod) agree this is perhaps the most difficult area of the game to quantify.
Excited to see how people tackle this problem (pun intended, obviously).