New Stat Viz

The NFL’s website is getting on the data visualization bandwagon. They recently unveiled a project called ‘StatLab’ which involves two types of plots.

The first one isn’t terribly interesting. It just illustrates each division in terms of W-L record. The more wins, the bigger each team’s circle. This is a textbook lesson is when a table is better than a chart, but it’s a start. How about spicing it up with a dynamic chart that recreates the horserace of the season? Teams grow and shrink as the weeks click by. Or year-to-year improvement/decline in season records?

The other set of charts is more interesting and interactive. There are ‘radar’ charts for player statistics. Each stat is depicted as a ‘dimension’ or distance from the center of a plot. The better the stat, the further out the point. The points for each stat are connected and filled, so the bigger the shape, the better the performance. You can select a number of players to compare. The average stats are depicted for the selected position too. (Don’t show this to Joe Flacco’s agent. It’s amazing how close Flacco’s numbers hew to average in nearly every category. The one exception is interception total.)

Under the main plot, there's also a week-by-week plot of selected stats, plus a year-by-year illustration. I can't seem to make the week-by-week plot work and the career year-by-year plot is very odd. It's a string of bubbles--the bigger the bubble, the better the stat that season.

It’s kind of cool, but in the end it’s only as useful as the stats used to create the chart. Yardage totals or raw tackle numbers aren’t any better metrics in graphical form than they are in a table. Still I’m a nut for cool visualizations, so I appreciate the effort.

I’d do it a little differently. I’d want the stats normalized so that the average is always midway out on the plot. Otherwise, the chart tends to produce the same odd shape over and over again with small deviations for each player. It might be a little tough because outlier numbers would literally be ‘off the chart’. I’d also suggest making different average plots for each defensive position instead of one general one. It’s not useful to see that a defensive lineman has fewer than average interceptions and a linebacker has more than average tackles.

Here's the link: StatLab

  • Spread The Love
  • Digg This Post
  • Tweet This Post
  • Stumble This Post
  • Submit This Post To Delicious
  • Submit This Post To Reddit
  • Submit This Post To Mixx

4 Responses to “New Stat Viz”

  1. Ian says:

    My opinion is that the interactive visualization obsession in tech/business/[the stat community] is a pretty big waste of time. People are just visualizing for the sake of it. I think it's taking good stat guys away from what they do best (torturing the data for truth), and it's enriching tech companies that put together cool looking but ultimately pointless software.

    I think there's a place for visualization, but 9 times out of 10 an excel chart will be simpler and more understandable.

    On the radar chart: I'm not a fan because there's no sense of which stats are most important, and the area inside the lines is meaningless. Imagine one quarterback who is good at a bunch of generic stats on a radar chart but is bad at any/a. He might have a great area but he's actually a bad QB.

  2. Ian Simcox says:


    I agree with that. We spend far too long at my office trying to make interactive visuals because mangers see a good bit of analysis summed up with a suitable vis and think that the vis is the way forward.

    They don't seem to realise the analysis behind the numbers is not something that can be generalised. Each problem requires specific techniques, and similarly the best vis is different every time.

  3. Unknown says:

    I'm not named Ian, but I agree with the other comments.

  4. David says:

    There are two cases where I think visualization really helps:

    In the first, you have recurring metrics and you can see how they change over time and compare them. For football, I'd like to see metrics year by year or even game by game for players of the same position. Take a concrete example where you show each quarterback's passer rating (a horrible stat, but bear with me) for each regular season game over the past three years. You have 3 * 16 * 32 datapoints all on a single chart! Compare that chart with the visualization for discovering who's been a better player lately.

    The other case is where you are trying to find something interesting in a mountain of data and a visualization can help pinpoint some interesting tidbits. Sometimes interesting stuff is found in unlikely places. In the viz you have to select the players you're going to compare.

    I'd say this viz is cool but functionally not very useful. However, I wouldn't write off all visualization as a total loss.

Leave a Reply

Note: Only a member of this blog may post a comment.