It's always surprising that sports reporters avoid computer-assisted reporting classes offered by NICAR in newsrooms around the country.
Of course, sports reporters keep strange hours, are often out of the office, and take month-long breaks to make up for the grueling hours they put in during the season. But no other beat depends so much on framing numbers in an entertaining and insightful way.
First, NICAR is looking to help. We're searching for a good time of year to hold two-day classes in computer-assisted reporting just for sports reporters.
Instead of government budgets and school reports, these would focus on covering high school football and stadium issues.
Let us know when you think you could fit it in.
That said, sports reporters have an instinctive understanding of statistics. Here are a few more tips for sports reporters that can make those numbers-laden stories just a little more insightful:
Understand Averages
The single biggest problem in sports reporting, particularly on salaries, but also on other player statistics, is a confusion between the words, "average" and "typical."
Get to know the concept of a median rather than a simple average, and use it in your stories. A median is often called the typical value, because, for example, it's the middle salary rather than the total paid divided by the number of players. That has a huge effect on what most readers view as astronomical player salaries. If one player makes $7.2 million, but the other eight make closer to $500,000 then the median is half-a-million. The average is something like $1.3 million. Which one more accurately portrays how much players get? The median. Which one says more about the finances of the team? The average. Pick yours wisely.
Pay attention to the database
Everyone understands in their guts that batting .333 is very different when a player bats 1-for-3 versus one who bats 33-for-100. The first could obviously be a fluke. The second shows consistent performance. That's easy. But what happens when it's 2-for-6? Or 3-for-10? In this space in September, John Sullivan discussed the use of significance testing in the newsroom. Here's an obvious case for a chi-square statistic. It tells you whether the batting average is likely to have occurred by chance. You can use a spreadsheet, a piece of statistical software, or a pocket calculator to test it. You'd set up your test like Table A.
The .449 says that a difference as big as the one shown could happen about 45 times out of 100 just by pure, dumb luck. In other words, if a player hits 3-for-10, he might or might not be better or worse than random chance (or 5-for-10). We don't need to set it up to expect hits to happen randomly. We can make our test more realistic by setting them up proportionately. This is usually a fairer way to test. Look at Table B. This time, I calculated "expected" hits by figuring the team's average of 15 percent hits (or 15 of 100). That percentage was applied to our star player's at-bats to compute the expected 2 of 10 hits, and to the rest of the team to figure the 14 of 90 hits. Now the 0.000 means that we'd rarely find a player to do this differently from the average purely by chance, or 0 of 1,000 times. Yes, it's a little work to get. But if your sports story revolves around a star's performance in a limited number of games, it's worth checking to see whether this could happen without any particular talent.
Evaluate players salaries
As many of my colleagues know, I'm not a big fan of gratuitous use of statistical techniques in the newsroom. But player salaries may be one area in which a regression can help. The problem is, it's a classic case of a difficult variable to test using simple linear regression.
In fact, it's one of the best examples of why reporters have to pay attention to the kind of data they're trying to predict or explain using regression.
In a linear regression, reporters try to guess a salary, a school's score or any other value based on a series of other figures that they think will explain its performance.
In this case, you'd try to guess a salary based on the idea that better players get paid more. The trick is to figure out whether it's consistent - if you get 40 percent more hits, do you almost always get a 10 percent higher salary?
How can you get a model?
Rather than reinventing the method needed to accurately predict or explain salaries, borrow from another field. Studies abound in explaining chief executive salaries in major corporations.
In fact, one consultant uses a method that helps tell shareholders whether the executive is worth the money he or she is paid.
Why not try adapting this model to sports? These techniques argue strongly for better use of the information that most sports reporters keep in books, in nightly reports and in their heads.
If they'd just type some data into a spreadsheet instead of writing it on paper, they'd be able to work through any of these problems on the fly. It's worth it.