Lineup Protection: A Myth Worth Testing

Some beliefs about baseball are wrong and die quickly. Some are right and become boring. A small third category survives for a century without ever being clearly shown to be true, because the story is so tidy that nobody feels the need to check. Protection lives in that third category, and it is worth an entire article not because prop researchers need a verdict on it, but because working through it teaches a habit that pays across every sport on a board.

The habit is simple to state and hard to practice. When you meet a claim about why a player will produce tonight, do not ask whether it sounds right. Ask what the claim predicts, ask whether that prediction would be visible in data, and if the answer is no or barely, put the claim aside and go back to the base rate.

The mechanism people propose

Protection is the idea that a hitter performs better when a dangerous hitter is scheduled to bat immediately after him. The reasoning is a chain, and every link in it is individually reasonable. A pitcher would rather not walk the man in front of a slugger, because the walk turns a solo threat into a rally. Preferring not to walk him, the pitcher works in the strike zone more often. Working in the zone more often, he serves up more hittable pitches. Seeing more hittable pitches, the protected hitter does more damage.

Notice the shape of that argument. It is not a superstition about momentum or heart. It is a causal story with a named actor, the pitcher, and a named decision, pitch selection, and it is exactly the kind of story a model could in principle absorb. That is why it deserves testing rather than dismissal. It also explains its durability: anyone who watches a pitcher nibble at a hitter with the eight slot on deck and then challenge that same hitter with a star waiting has just watched the mechanism appear to happen.

What the claim predicts, stated precisely

A vague belief cannot be tested, so the first real work is turning the story into predictions specific enough to be wrong. If protection operates as described, several things should be observable.

• The protected hitter should see a higher share of pitches in the strike zone, since the entire chain begins with pitchers avoiding the walk.
• His walk rate should fall, because fewer pitches outside the zone means fewer free passes.
• His production on contact should rise, because hittable pitches are the payoff of the whole story.
• The effect should scale with the quality of the hitter behind him, and it should appear and disappear when a manager reorders the lineup.

That last one matters most. A real causal effect should move when its cause moves. If a hitter's output does not change when the man behind him changes, the mechanism is not doing what the story says it does.

Why the evidence has stayed thin

Baseball analysts have argued about protection for a long time, and the honest summary of that argument is that the pitch selection piece is easier to find than the production piece. It is not hard to believe pitchers change how they attack a hitter based on who is coming next. It has been persistently difficult to demonstrate that the change produces a meaningful lift in the protected hitter's results, and the estimates that do appear tend to be small relative to the noise around them.

We are deliberately not going to hand you a number here, because the number is where these discussions usually go wrong. Anyone can produce a figure by picking a window, a definition of dangerous, and a stat. What matters more is understanding why the question is so hard to settle at all.

Start with the fact that the treatment is not randomly assigned. Managers place good hitters in front of other good hitters. If protected hitters outproduce unprotected ones, the obvious first explanation is that they were better hitters to begin with, and untangling that requires comparing a player against himself across lineup arrangements, which shrinks the sample dramatically. Then consider what the effect must survive. A modest change in strike zone rate spread over a few hundred plate appearances is competing against a batting average on balls in play that swings wildly by itself, exactly the volatility the streaks article describes. A real but small effect can be entirely buried by ordinary variance, which means the data can neither confirm nor rule it out.

There is also a countervailing force the tidy story leaves out. If a pitcher throws more strikes to the protected hitter, the protected hitter is trading walks for swings. Walks are valuable. A hitter who walks less and slugs slightly more has not obviously gained, and depending on the market you are researching, he may have gotten worse: a walk is not a hit, but it is also not a strikeout, and it changes the batter's own on base outcomes in ways that cut across different prop types.

What lineup position really changes

Here is the part that actually pays in prop research, and it is not subtle at all. While protection remains debatable, batting order position changes two things that are measurable, stable, and large enough to move a projection.

The first is expected plate appearances. The top of an order comes to the plate more often than the bottom, roughly half a trip per game between the extremes, and that is a direct multiplier on every counting stat a prop can settle on. A hitter who drops from second to sixth has lost real chances tonight, and no matchup narrative replaces them. This is the foundation the hitter prop guide builds everything else on top of, and it is the first thing to confirm on any posted lineup.

The second is run scoring context, which is where the honest version of the protection intuition lives. Runs and runs batted in are team outcomes wearing an individual name. A hitter batting in front of productive teammates scores more runs because those teammates drive him in. A hitter batting behind on base heavy teammates collects more runs batted in because there are more runners to drive in. Neither of those requires a pitcher to change anything. They are arithmetic about who is on base and who is coming up, and they show up cleanly in the data because they are not fighting through a chain of behavioral inference.

A template for every other folk belief

Protection is one instance of a genre. Every sport has them: the team that always shows up flat after a long trip, the shooter who gets hot when the crowd is loud, the fighter who fades because of a rough weight cut nobody documented. Run each through the same four steps.

1. State the mechanism as a causal chain with a named actor and a named decision. If you cannot, the belief is a vibe and stops here.
2. Write down what the mechanism predicts in terms specific enough to be false, and include the prediction that the effect should vanish when the cause is removed.
3. Ask whether that prediction is observable at the sample size you actually have. Effects smaller than the noise around them are unusable even when they are real.
4. If the answer is no, default to the base rate and move on. Not knowing is a finding, and it is a cheaper one than acting on a story.

Step four is the one people skip. Defaulting to the base rate feels like giving up, but as the base rates article argues, the base rate is usually the best available estimate precisely because it has already absorbed every effect you are unable to isolate. If protection is real and small, it is already inside the historical performance of hitters in those lineup slots. You do not need to find it. You need to avoid adding it twice.

How our engine treats the question

Our MLB engine simulates the actual batting order one plate appearance at a time against the actual starter, with the bullpen taking over when the simulated leash runs out. That structure gives lineup slot its measurable consequences for free: the order determines how many trips a hitter gets, and because the whole lineup lives inside the same simulated game, the traffic in front of him and behind him shapes runs and runs batted in without anyone hand tuning a context factor.

What the engine does not carry is a protection coefficient. There is no term that rewards a hitter for who is on deck, because we have no measured basis for one, and an authored coefficient with no evidence behind it is a confident guess that will quietly propagate into hundreds of projections. When we cannot measure something, our preference is to leave it out and let the record show whether its absence hurts. That record, per market, at posted lines, with losses and voids included, is public in the Model Room, and the projections it grades are on the Signal Board. Test the belief, keep the part that survives, and let the ledger judge the rest.