Base Rates: The Anchor Most Prop Research Skips

Watch how a typical prop conversation opens. Somebody mentions the matchup, the recent form, an injury on the other side, a stadium, a grudge. All of it is specific, all of it is about today, and all of it arrives before anyone has established the one thing that would give it meaning: how often the event in question happens under ordinary circumstances.

That ordinary frequency is the base rate, and it is the most reliably skipped step in amateur research. Not because it is difficult. Because it is boring, and because it feels like it says nothing about this particular game. It says more than any of the specifics do, and everything the specifics contribute is measured relative to it.

A base rate is what happens before you know anything

A base rate is the frequency of an outcome across a large reference class, computed before you condition on today's details. How often does a starting pitcher in this role clear six strikeouts. How often does a rotation regular at this workload reach twenty points. How often does a fixture in this competition produce three goals. The reference class is a group of comparable situations, and the base rate is what happened across that group.

In statistical language it is a prior: the position you hold before evidence, which evidence then updates. In practical language it is the anchor. Every specific fact you learn about today should move you away from the base rate by an amount proportional to how much that fact actually predicts. No fact entitles you to abandon the anchor and start from scratch, which is exactly what most narrative research does.

Notice how this connects to shape. A base rate is a probability at a threshold, not an average, which means it is already the right kind of object for prop research. The average is a summary of a whole range; the base rate answers the yes or no question the offer asks. That distinction is the subject of reading a distribution, and base rates are the empirical cousin of the same idea.

Why vivid specifics beat boring frequencies in your head

The reason base rates get skipped is not ignorance. It is that specific information feels far more informative than it is. Tell someone that a player has cleared this line in four straight games, or that the opposing defense allowed a huge number to a similar player last week, and the base rate stops feeling relevant. The specific detail crowds it out. Psychologists call the resulting error the base rate fallacy, and it is remarkably resistant to knowing about it.

The classic demonstration involves a rare condition and an accurate test. If a condition affects one person in a thousand, and a test correctly flags it 99 times in 100 while producing a false alarm 1 time in 100, a positive result is still much more likely to be a false alarm than a real case. The reason is arithmetic: there are so many more people without the condition that even a small false alarm rate produces more false positives than there are true positives to find. People shown this problem overwhelmingly guess the opposite, because the accuracy of the test is vivid and the rarity of the condition is abstract.

Prop research runs the same machinery constantly. A four game streak is a vivid test result. If the underlying base rate for clearing that line is low, the streak is far weaker evidence than it feels, because streaks of that length occur regularly by chance across the hundreds of players on any slate. Somebody is always on one. The question is never whether the streak exists. The question is how many players would have produced a streak like that even if nothing had changed, and sample size is what answers it.

Name the reference class out loud

The discipline that makes this usable is embarrassingly simple: say the reference class in a full sentence before you look at anything about today. Not in your head. In words, written down if you are keeping a research log.

• Starters in this rotation role, at roughly this pitch count, against lineups of roughly this quality.
• Rotation regulars at this minutes level, in this offensive role, against defenses in this range.
• Wide players who start in this competition, at this share of team attempts.
• Fighters in this weight class, in bouts scheduled for this number of rounds.

Saying it out loud exposes the two failure modes immediately. Too narrow and the class collapses: this exact player against this exact opponent on this exact surface leaves you with four observations and a base rate that is mostly noise. Too broad and it stops describing the situation: all players in the league tells you very little about a specialist in a limited role. The workable class is the widest group that still resembles the situation in the ways that matter for this stat.

There is a useful test for whether you have gone too narrow. Ask how many observations the class contains. If the honest answer is under a couple of dozen, you do not have a base rate, you have a small sample dressed as one, and the correct move is to widen the class and accept a less tailored anchor. A broad anchor with real support beats a bespoke one built from six games.

Adjust from the anchor, do not replace it

With the class named and a frequency in hand, today's specifics get their turn. This is where the process differs from narrative research, which treats each fact as a fresh argument. Here each fact is an adjustment with a size, and the size has to be defensible.

Some adjustments are large and structural. A change in role is large: a player moving into a starting spot, a pitcher on a restricted count, a footballer shifting from a wide role to a central one. Confirmed absence of the primary option ahead of a player is large. These change the opportunity that generates the stat, and opportunity dominates. Other adjustments are small even when they feel enormous: an opponent's rank in a defensive category, a recent hot stretch, a venue effect, a rest advantage. They are real, they are measurable, and they are worth low single digit percentage points, not a rewrite of the anchor.

The general rule that falls out of this: adjustments that change how much opportunity a player receives are worth far more than adjustments that change how well he converts it. Opportunity is chunky and observable. Conversion is noisy and regresses hard. Most confident prop takes are built entirely from the second category while ignoring the first, which is also why a role change is the piece of news worth catching earliest.

Notice when you have overshot

The most valuable habit in this whole process is the audit at the end. Compare your final estimate to the base rate you started from and ask a blunt question: how far did I move, and does the evidence I gathered actually support a move that size?

Overshooting has a signature. It happens when several pieces of evidence all point the same direction, and you count each one at full value as though they were independent. They usually are not. A hot streak, a friendly matchup rating, and a bullish preview article are frequently three descriptions of one underlying fact, or three consequences of the same recent performance. Counting correlated evidence as if it were separate is how a defensible move of four points becomes an indefensible move of twenty.

The other signature is arriving at a probability that is too far from the market without being able to name the mechanism. If your estimate says an offer clears 78 percent and the posted line implies something closer to a coin flip, one of two things is true. Either you know something specific and structural that the price does not reflect, and you can state it in a sentence, or you have talked yourself into a number. Being unable to finish the sentence is the tell. Where a genuine market reference exists, that comparison is informative; where it does not, you are doing model versus line analysis and the discipline matters even more, which is the subject of how to read probability without fooling yourself.

What this looks like at the board

In practice the base rate step takes about a minute per prop and it changes the character of a research session. You stop asking whether you like a player and start asking whether today is different enough from ordinary to justify a specific move away from ordinary.

1. Read the offer and state the exact threshold before anything else.
2. Name the reference class in a sentence, and check it has enough observations to mean something.
3. Write down the base rate for clearing that threshold in that class, even as a rough figure.
4. List today's specifics, and mark each as opportunity changing or conversion changing.
5. Move from the anchor by an amount you could defend to somebody skeptical, then compare the result to the posted line.

Projection systems do a version of this internally. Shrinking a player's observed rates toward a population baseline, weighted by how much of his own data exists, is the mechanized form of the same idea: the fewer observations you have, the more the anchor should dominate. Slateline's engines apply that shrinkage before any simulation runs, and the graded record that shows whether the resulting probabilities hold up is published in the Model Room, including the misses.

The habit is unglamorous and it is most of the edge available to a careful researcher. Start from how often the thing happens. Move only as far as the evidence pays for. When the honest move is nowhere, that is a finding too, and passing is the correct output. You can see how our projections sit against posted lines across every live board on the signal board.