What a Small DFS Sample Can and Can't Tell You

By Splash Play Staff · A guide to reading the tools

A small sample in DFS can raise a question. It can almost never settle one. Three good weeks, one bad slate or five lineups that all cashed tell you less than they feel like they do, because with a few results a single outcome swings the whole number.

Count before you take a percentage

Something happens 1 time in 5. That is 20%. Something else happens 4 times in 20. Also 20%.

They are not the same evidence. Change one result in the first list and the rate moves 20 points, to 0% or 40%. Change one in the second and it moves 5. A percentage with a small count behind it is one result away from a different story, so write the count down with it: "1 of 5," never "20%" alone.

A thousand simulations are not a thousand slates

A simulation plays one slate many times under one set of assumptions. Running it more often makes the answer to "what does this model expect?" more precise. It does not test whether the model's assumptions are right. A wrong projection simulated many times produces a very stable wrong answer.

Real slates are the scarce thing. There is one NFL main slate a week.

Twenty lineups are not twenty tries

Your 20 lineups on one slate share players and share games. If your main stack fails, most of them fail together. Those entries are correlated results from one slate, not 20 independent tests of the strategy. Count slates when you count evidence, not entries.

What a short run can tell you

That a process broke. You entered a lineup with a player who was ruled out. One slate is plenty to learn that.
That an input was wrong. A projected starter didn't start. That is a fact about one input, and worth knowing.
What to look at next. A pattern in a few results is a reason to check, not a result.

What it can't

It can't tell you a player is "due," that a strategy works, or that one doesn't. The usual mistake is changing a sound process after two bad weeks and keeping an unsound one after two good ones. Decide in advance what you will count and for how long, then don't stop counting at the moment the number says what you hoped.

Where to see it

Simulation results are part of Squirt Squad VIP and Apex. Open the NFL tool for the current slate, read the field guide for the step-by-step workflow, or compare memberships. Simulation results are model outputs, and no number of them guarantees a result.