Nine training evaluation models, compared without the methodology wars

Reaction, learning, behavior, results. The shared language of training evaluation for seventy years, which is both its strength and the reason people attack it. Most stakeholders already think in its terms whether they know the name or not.

Reach for it when: you need a framework everyone in the room already accepts. Which is most of the time.

The official revision. Plan backwards from Level 4, measure the "required drivers" that make behavior change survive, watch leading indicators instead of waiting a year, and judge success as Return on Expectations rather than ROI. The practical effect is a longer, better list of questions at every level.

Reach for it when: you want Kirkpatrick's familiarity with a sharper measurement plan, or stakeholders ask about ROE.

Kirkpatrick's four levels plus a fifth: return on investment. The distinctive machinery is isolation, separating the program's effect from everything else that changed, usually via participant estimates discounted for confidence. Produces a percentage a CFO will engage with.

Reach for it when: finance holds the budget and wants the business case in their own units. Reserve the full treatment for your most expensive programs.

Ignore the average, study the extremes. A short screening survey finds the most and least successful participants; interviews with both produce documented success stories and a diagnosis of what blocked everyone else. Unusually persuasive with executives, because stories with names attached beat bar charts.

Reach for it when: you need to convince leadership a program deserves continued investment, or to find out why transfer keeps failing.

Eight tiers of evidence strength, from attendance at the bottom to organizational effects at the top. Built as a critique of how the industry actually measures: completions and smile sheets. Less a process to run than a standard to hold your evidence against.

Reach for it when: your client's L&D team is methodologically sharp and the credibility of the evidence itself is under scrutiny.

Kirkpatrick with the first level split in two (the materials versus the delivery) and a "mega" level above organizational results for societal impact. The Level 1 split is quietly useful for question design; the mega level matters mainly where the mission is the societal outcome.

Reach for it when: the client is public-sector or mission-driven, or you want cleaner diagnostics out of your reaction questions.

Context, Input, Reaction, Outcome. Half the model evaluates decisions made before delivery: was the need real, was the design right. Its other lasting idea is treating participant reactions as improvement suggestions rather than verdicts.

Reach for it when: honestly, rarely as a full process. Its instincts about pre-delivery evaluation survive inside newer models.

Evaluates the learning function rather than individual programs: check alignment with strategy, then choose the kind of evidence (ROI, expectations, benchmarks, capacity) your decision-makers actually value, and evaluate in that currency.

Reach for it when: the client is a UK HR team, or the real question is what to measure rather than how a single program performed.

Context, Input, Process, Product. A decision-oriented model from the education world: evaluate the needs a program addresses, the soundness of its design, the quality of its delivery, and its outcomes, to decide what to do with it. Stufflebeam's slogan was that evaluation's purpose is "not to prove, but to improve."

Reach for it when: the work is education-sector, public-sector, or accreditation-shaped, where the question is whether to adopt, continue, or restructure a program. In corporate L&D you'll meet it rarely, but its process-versus-product distinction quietly shows up in good evaluation design everywhere.