Concept Primer No. 28 · The Number and the Thing
The Sports Page
Making the numbers mean something since the first pitch
Concept No. 28Constructs and Indicators Back to The Sports Page →

Nobody Has Ever Counted Power. We Count Home Runs and Hope.

The things fans argue about — power, clutch, hustle, a good eye — cannot be measured directly. Statisticians call them constructs. The numbers in the box score are indicators: things we can count because we think they move with the thing we care about. Every stat argument is really an argument about that link.
Tier 1 · The Two-Minute Version

The stat is a window. The construct is the room.

Ask a fan who the best power hitter is and you will get an answer, fast. Now ask what power is. You can not put it on a scale. You can only watch what it does: home runs, doubles, balls hit hard and far. Those you can count.

That gap has a name. The thing you care about is a construct. The thing you count is an indicator, sometimes called a proxy or a manifest variable. Home runs are an indicator of power. Hits are an indicator of contact. Runs batted in are an indicator of … well, that is the trouble.

A good indicator moves when the construct moves, and stays put when it does not. A bad one also moves for other reasons. RBI move when a hitter gets better. They also move when the two men batting ahead of him get on base more, which has nothing to do with him at all.

So every time someone wins an argument with a stat, ask two questions. What is this number supposed to stand for? And what else makes it go up?

Tier 2 · If You Want to Go Deeper

Two kinds of arrows, and why one window is never enough.

Draw the construct as an oval and the stat as a box, with an arrow between them. Which way does the arrow point? For home runs it points from power to the stat: power causes home runs. That is a reflective indicator. Several of them — home runs, doubles, exit velocity — should rise and fall together, because they share a cause. If they do not, at least one of them is measuring something else.

RBI work the other way. Power, contact, teammates on base, a spot in the lineup and staying healthy all point into the RBI count. It is a composite — a sum of many causes, only some of them the hitter’s. That is why two hitters with the same bat can finish 50 RBI apart.

Because every indicator carries some of the wrong signal, careful measurement uses several indicators per construct and looks for what they share. What the windows agree on is the room. What only one window shows is probably a smudge on that window.

The test of whether a stat measures what it claims has a name too: validity. A stat can be perfectly reliable — the same every time you count it — and still not be valid for the thing you are using it for. RBI are counted with great care. Whether they measure the hitter is a different question.

Tier 3 · Take It to the Textbook

Latent variables, measurement models and construct validity.

The formal treatment: Cronbach and Meehl (1955) on construct validity and the nomological network; Campbell and Fiske (1959) on convergent and discriminant validity across multiple methods; the common-factor model, in which observed indicators are a latent variable plus unique error; reflective versus formative measurement (Bollen and Lennox, 1991) and why the choice changes what a model can say; and structural equation models, which put the measurement model (ovals to boxes) and the structural model (ovals to ovals) in one picture. That last idea is where this series is heading: first the boxes, one stat at a time, then the ovals, then the arrows between them.

Where You Have Seen This