Evaluation research has treated program theory as its own craft for decades. A logic model sketches the path from activity to outcome. A program theory states why each join on that path should work. Teams usually draw the sketch. Few put the reasons in writing.
An earlier report on this site already walked a five-stage funnel with one real metric at each join, built around a savings-goal feature. That piece is a logic model. This one sits one layer down: the causal claims evaluators have formalised for half a century, and the two failure modes that look identical until someone names the theory.
One of the cleanest cases on record ran almost exactly to plan and still pushed the harm it meant to stop higher. Delivery held. The causal idea did not. Program theory exists so that split can be made on purpose.
The story in four parts
Every initiative treats doing X as a path to Y, even when nobody ever writes that bet in those words.
The idea itself can be false, or a sound idea can be mishandled in the field. Those two breakdowns call for different repairs.
Listing the stages is the lighter job. Naming why each transition should fire, and in which conditions, is the part that can be tested.
An unspoken assumption cannot be checked before the budget is spent. A written claim can.
The evidence at a glance
Across nine randomised and quasi-randomised trials, a Campbell Collaboration review found “Scared Straight”-style programmes lifted reoffending by 1 to 28% relative to untreated controls.
Mexico's Progresa put its theory of change on trial in 506 of 6,396 eligible villages before national scale-up, among the earliest large randomised evaluations of a national anti-poverty programme.
The W.K. Kellogg Foundation's logic-model guide, already cited on this site, divides outcomes into short-term, medium-term, and long-term tiers, the layered map a single goal line leaves out.
Evaluators have kept the two pieces apart for a long time. Huey-Tsyh Chen's foundational split divides a program's theory into a normative half (what the programme does, and what it aims to achieve) and a causative half (why those actions should produce the result). Sound evaluation needs the causative half alongside the normative one. Funnell and Rogers draw the same line for practitioners. The logic model is the box-and-arrow sketch. Full program theory adds the causal assumptions and the conditions that have to be true for each arrow to hold.
That is the gap this site's outcome-logic report walks into without naming it. Its five-stage funnel, with a real metric at every join, is a logic model. Program theory sits underneath. It writes that a customer should move from “sent” to “opened” because attention has to clear a salience threshold, as an explicit claim someone can check.
Putting the causal chain on paper does more than describe the programme. It opens a diagnosis that stays unavailable when the chain stays implied. Carol Weiss's framing of theory-based evaluation sorts failure into two causes. A theory failure means the underlying causal idea was false. An implementation failure means a sound idea was mishandled in delivery. With no explicit program theory to compare against, an evaluator watching a programme fail cannot tell which cause applies, and the wrong call points the repair at the wrong place.
The abstract split is easy to recite. Here is the same split inside a programme tested again and again in the field. “Scared Straight”-style programmes brought juvenile offenders and at-risk youth into adult prisons to hear inmates describe incarceration first-hand, on the causal bet that vivid exposure to consequences would deter future crime.
Why did a program executed almost exactly as designed increase the crime it was built to prevent?
A Campbell Collaboration systematic review pooled nine randomised and quasi-randomised trials of Scared Straight and similar juvenile awareness programmes across eight U.S. states. Each study tracked subsequent offending as the outcome, leaving attendance and satisfaction aside. Pooled across those trials, reoffending rose 1 to 28% among treated youth relative to untreated controls.
What it teaches: delivery mostly matched the plan. Prisoners spoke, tours ran, sessions took place. Tighter management would not have rescued the result, because the causal bet underneath it, that fear of a vivid future consequence cuts present offending, failed for this population. Call that a theory failure. An implementation failure looks different, and only a written theory lets anyone tell the two apart after the fact.
Scared Straight is what follows when a false theory scales without a prior check. Mexico's Progresa programme ran the opposite sequence. Designed in 1997 by deputy finance minister Santiago Levy, Progresa (later renamed Oportunidades) paid cash transfers to mothers in poor households, conditional on children attending school and family members attending health checkups.
What made Progresa's theory testable instead of just hoped for?
Designers wrote the theory as a concrete causal chain rather than a soft goal. Cash tied to school attendance and health checkups would raise human-capital investment, and that investment would cut poverty in the next generation beyond short-term relief in the current one. That level of detail is what let evaluators randomise 506 of the 6,396 eligible villages into treatment and control groups before national rollout. It remains one of the earliest large-scale randomised evaluations of a national anti-poverty programme anywhere.
What it teaches: a sharp theory does not promise success. It promises that success or failure, and the reason for either, can be measured, which is why writing the theory down is worth the time.
Scared Straight and Progresa sit side by side as two tests of the same evaluation habit. Both cases support the same claim about how to evaluate. A named, specific program theory is what let Scared Straight's collapse read as a theory failure instead of an unreadable shrug. The same discipline is what let Progresa's designers put their causal claim on trial before staking a national budget on it. Nobody could see either outcome in advance. Writing the theory down, in enough detail to check, is what made each result readable after the money moved.
Here is the same discipline in a form a team can run in a meeting. Four checks, short of a full evaluation plan, because a theory that stays unwritten when the process feels heavy never gets the chance to do its job.