guide
The Kirkpatrick model — what each level really measures
Four levels, sixty years old, still the shared language of training evaluation — and still quietly abused. Here is what each level measures, where it breaks, and what a form can honestly deliver.
The four levels in one breath
Donald Kirkpatrick published the model in 1959 (revised 1998). Level 1 — Reaction: did the room find the training relevant and well run? Level 2 — Learning: can participants now do the core things? Level 3 — Behavior: is it showing up in the work weeks later? Level 4 — Results: did business numbers move?
Where each level breaks
Level 1 breaks because satisfaction is easy to collect and pleasant to report — a room that enjoyed itself tells you almost nothing about change. Level 2 breaks when it is measured with vague questions ("I understand the topic"); it survives only with narrow, checkable ones ("I can apply this without help"). Level 3 breaks because nobody sends the follow-up: the thirty-day email is the whole level, and it is the first thing a busy calendar drops. Level 4 does not break — it simply is not a survey: isolating training's effect on business numbers needs comparison, baselines and a tolerance for confounding that no feedback form provides.
The honest boundary is the feature. A form that claims Level 4 produces numbers nobody believes — and then nobody believes the rest of the form either.
What a form can actually deliver
Grouped reaction items (content versus delivery, so you know which half to fix), narrow confidence self-reports as a Level 2 proxy, and a day-30 follow-up that counts uses — that is Levels 1 through 3, cleanly separated, on one page a manager can forward. The training evaluation form folder ships exactly that: 8 reaction items, 3 confidence items, keep/change questions and the day-30 email template — print or online, CC BY 4.0. The lighter post-training survey covers Levels 1-2 in five minutes.
How to report it without lying
Three rules that keep the whole model credible: report reaction and behavior separately, never averaged together; when the confidence items have no baseline, write the words "self-report, no baseline" on the page; and report the lowest item alongside the mean — a 4.1 average hiding a 2.2 on pace is a different session than a flat 4.1.
Questions people ask
Is the Kirkpatrick model still worth using?
Yes — as a map of what you are and are not claiming. Its power is the discipline of the four levels, not the scores. Skip it and you either over-claim (smile sheet as proof of impact) or under-collect (nobody ever checks behavior).
How fast can I add Level 3 to a training I already ran?
Thirty days after any future session — that is the whole lead time. Send three questions: used it how often, what helped or blocked, what support would make it stick. A third of the in-room response rate is normal and still counts as evidence.