Why splitting sports data creates smaller samples

A hitter has a profile with 120 recorded batted balls. You want to know how they hit sliders from right-handed pitchers, so you narrow the data. The resulting number can be useful, but it isn't based on all 120 balls anymore.

A split is a filtered part of a larger record. Handedness keeps observations against a selected type of opponent. A pitch-class split keeps observations involving selected pitches. Combining filters makes the question more specific and often leaves fewer examples.

That specificity is the benefit. It is also the tradeoff: the final value has less evidence behind it than the overall profile. Read the filtered count alongside the filtered statistic, not the count from the page's main heading.

In our invented window, 80 of the 120 batted balls came against right-handed pitchers. Twelve of those 80 were against sliders. So the final slider-versus-right-handers split uses twelve batted balls.

The 40 batted balls against left-handers don't belong in that final split. Nor do the other 68 balls against right-handers that came on different pitch types. They remain in the wider profile, but they answer different questions.

This filtered sample doesn't tell you how many total sliders the hitter saw. Some sliders may have been taken, fouled or missed. A contact count cannot silently become a pitch count or a plate-appearance count.

Suppose the twelve slider batted balls have a strong contact-quality value. That tells you something about those balls. It doesn't establish the hitter's full offensive performance against sliders, because outcomes without contact aren't represented in a contact-only measure.

It also doesn't prove that the hitter will keep producing the same quality. With only twelve balls, a few standout contacts can have a large influence. A narrow split may look unusually strong for exactly that reason.

Compare like with like. A hitter's full xwOBA and their contact-only slider value can both be meaningful, but they count different events. The pitch label, window and metric definition have to travel with the value.

A clear description is: “This value covers twelve batted balls against sliders from right-handed pitchers in the selected window.” It takes one sentence to prevent several common misunderstandings.

Then look at how relevant the split is to the matchup. If the pitcher rarely throws that pitch, the slice may be less important than the bigger picture. If they use it often, it deserves attention, with its small count still visible.

Use filters to explore a question, not to keep narrowing until an impressive number appears. Check the overall record too. A useful split adds context; it doesn't replace the rest of the evidence.