Training evaluation methods: 7 models and how to choose one

Training evaluation models compared

Seven training evaluation models, the question each answers, and the effort each takes

Model	Origin	Main question	Best for	Effort
Kirkpatrick's four levels	Donald Kirkpatrick, 1959 to 1960; updated by Kirkpatrick Partners in 2026	Did learners react well, learn, apply it, and get results?	A shared frame for almost any program	Low at Levels 1 and 2; higher at 3 and 4
Phillips ROI Methodology	Jack and Patti Phillips, ROI Institute (founded 1992)	Was the program worth what it cost?	Expensive, visible programs	High
CIPP	Daniel Stufflebeam, 1960s	What is needed, how should we do it, is it running as planned, and did it succeed?	Long programs you can still change	High
Kaufman's five levels	Roger Kaufman and John Keller, 1994	Did the program help the organization and the people it serves?	Public sector, health, and nonprofit programs	High
Success Case Method	Robert Brinkerhoff, 2003	When it works, how well, and why does it fail for others?	Explaining uneven results quickly	Low to medium
Kraiger, Ford, and Salas	Journal of Applied Psychology, 1993	What kind of learning happened: cognitive, skill, or affective?	Designing learning measures	Medium
LTEM	Will Thalheimer; version 13 released 2024	Can learners make decisions and do the task, and do they use it at work?	Moving past attendance and satisfaction data	Medium

A model tells you what to measure. A technique, such as a survey, a test, or an observation checklist, is how you collect the evidence. You need both: the model decides what evidence counts, and the techniques collect it.

1. Kirkpatrick's four levels

Donald Kirkpatrick introduced the four levels in a series of articles in 1959 and 1960, and the model became the most widely used way to evaluate training (Pearlstein, 2009). Kirkpatrick Partners defines the levels this way today:

• Level 1, reaction: whether learners find the training favorable, engaging, and relevant to their work.
• Level 2, learning: whether they gain the intended knowledge, skills, attitude, confidence, and commitment.
• Level 3, behavior: whether they perform the critical behaviors on the job, with support and accountability.
• Level 4, results: whether the targeted organizational outcomes occur.

Plan in reverse: Kirkpatrick Partners tells teams to start with Level 4 and the reason the program exists, then work back to the behaviors and learning that would produce it. They also advise checking behavior well before 90 days, so you can fix a program that is drifting, and note that most Level 4 indicators are already tracked somewhere in the organization.

In 2026, an update led by Vanessa Milara Alzate added the performance environment to the model: the conditions at work that help or block transfer. The update also names three return measures. Return on expectations is success as stakeholders define it in advance. Return on performance checks whether the behaviors happened and contributed to the outcomes. Contributive ROI states the share of improvement reasonably linked to the program instead of claiming training as the only cause (checked October 2026).

Two research findings help you use the levels well. A 1997 meta-analysis of 34 studies found that "utility" reactions, such as whether learners found the training useful, were more strongly related to learning and on-the-job performance than whether they liked it (Alliger and colleagues, 1997). Ask about usefulness on your Level 1 survey. And a 2021 review of the model in higher education found that evaluators lean toward the lower levels and that evidence for the causal chain between levels is thin (Cahapay, 2021). You still have to measure Level 3 directly, even when Level 1 and 2 scores look good.

Level 2 evidence should match the skill. For a language course, a placement test taken before and after, such as an online French level test, gives you a clean comparison. For procedures and judgment calls, short scenario questions inside the training work better. In Mindstamp, a Likert question can collect usefulness and confidence ratings, scored questions measure learning at the moment it is taught, and the report shows each learner's answers. Try the Kirkpatrick evidence desk example to see the levels as choices in a video.

2. Phillips ROI Methodology

Jack and Patti Phillips extended Kirkpatrick's chain to five levels: reaction, learning, application, impact, and return on investment. Their ROI Institute, founded in 1992, reports work with more than 9,000 organizations in over 70 countries. The method's distinctive steps sit between Level 4 and Level 5:

1. Isolate the effects of the program from other causes, for example by comparing a trained group with a similar group that has not trained yet.
2. Convert the impact measures to money.
3. Record intangible benefits you will not convert, such as teamwork.
4. Count fully loaded costs, including learners' time away from work.
5. Calculate the ROI.

The ROI Institute's formulas are ROI (%) = (program benefits − program costs) ÷ program costs × 100, and benefit-cost ratio = program benefits ÷ program costs (ROI Institute calculator). If a program costs $80,000 and produces $200,000 in benefits you can defend, the ROI is 150% and the benefit-cost ratio is 2.5.

Use Phillips when leaders will decide a large budget on financial evidence. The isolation step is the hard part, and it is where Kirkpatrick Partners' 2026 contributive ROI takes a different position: it reports the training's share of an improvement instead of trying to separate training as the sole cause. Pick one approach before you collect data and state it in your report. Our guide to measuring the ROI on training walks through the calculation, and the training ROI calculator runs the numbers.

3. CIPP model

Daniel Stufflebeam built the CIPP model to help school districts, state education departments, and the federal government assess projects in the US War on Poverty of the 1960s and 70s. Its four parts each ask a question you can use before, during, and after a program (CIPP Evaluation Model Checklist):

• Context: what needs to be done? This is your needs assessment.
• Input: how should it be done? Compare designs, budgets, and delivery options.
• Process: is it being done as planned? Check delivery while the program runs.
• Product: is it succeeding? The checklist splits this into impact, effectiveness, sustainability, and transportability to other settings.

CIPP fits a multi-month program, such as a new-manager academy, where you can still change the design. Where a results study looks back after delivery, CIPP gives decision makers information while the program is still running. Western Michigan University's Evaluation Center publishes the CIPP checklist with its other evaluation checklists. For sample questions at each CIPP stage, see our list of evaluation questions.

4. Kaufman's five levels

Roger Kaufman and John Keller published "Levels of Evaluation: Beyond Kirkpatrick" in Human Resource Development Quarterly in 1994. Their model has five levels: enabling and reaction, acquisition, application, organizational outputs, and societal outcomes (ERIC record). The first level asks whether the resources and delivery were good enough, alongside learners' reactions. The last asks what the organization's results mean for its clients and the wider community.

Use Kaufman when an organization's results are judged outside its walls. A hospital's infection-control training, for example, can be traced from nurses' practice (application) to infection rates on the ward (organizational output) to patient outcomes after discharge (societal).

5. Brinkerhoff's Success Case Method

Robert Brinkerhoff's Success Case Method skips the average learner. It finds the most and least successful participants and studies them in detail to answer two questions: when the program works, how well does it work, and what gets in the way when it does not (Brinkerhoff, 2005). His 2003 book sets out five steps, summarized by BetterEvaluation:

1. Focus and plan the study.
2. Build an impact model that describes what success looks like.
3. Survey all participants to find the best and worst cases.
4. Interview and document those cases.
5. Report findings, conclusions, and recommendations.

A study can finish in weeks and gives stakeholders concrete stories with the conditions behind them, such as a manager who coached after the course or a team whose tools did not allow the new process. It pairs well with Kirkpatrick Level 3. It does not give you an average effect, so do not use it alone to claim how much a program improved performance.

6. Kraiger, Ford, and Salas learning outcomes

Kurt Kraiger, J. Kevin Ford, and Eduardo Salas argued in the Journal of Applied Psychology in 1993 that "learning" is three kinds of outcome, and each needs its own measure:

• Cognitive: what learners know and how they organize it. Measure with knowledge tests and questions that ask learners to explain or connect ideas.
• Skill-based: what learners can do, and how smoothly. Measure with demonstrations, role plays, and work samples.
• Affective: attitudes and motivation, including confidence that they can do the task. Measure with attitude and self-efficacy scales.

Use this framework when you write Level 2 measures. A safety course that tests only recall misses whether workers can perform the lockout steps and whether they believe the steps matter.

7. Learning-Transfer Evaluation Model (LTEM)

Will Thalheimer's LTEM has eight tiers and was last revised as version 13 on October 31, 2024. Where Kirkpatrick puts all learning in one level, LTEM separates knowledge, decision-making, and task performance. Thalheimer says measures in the lower tiers cannot validate learning success, while measures in the top half can show that learning transferred to the job.

LTEM is most useful as a check on your current metrics. If your dashboard reports completions and satisfaction scores, it tells you that people showed up and liked the course. Adding realistic decision questions, like the one in the video at the top of this page, moves your evidence up a tier without a long field study.

You may also meet the CIRO model (context, input, reaction, outcome) from Peter Warr and colleagues' 1970 book on evaluating management training. It overlaps with CIPP and Kirkpatrick, so the advice above applies.

Training evaluation techniques: how to collect the evidence

Common training evaluation techniques and the level of evidence each provides

Technique	What it shows	Level	Tip
End-of-course survey	Relevance, usefulness, barriers learners expect	1	Ask how useful the course was for their job
Pulse checks during the course	Problems while you can still fix them	1	One or two questions at the end of each module
Pre-test and post-test	Knowledge gained	2	Use equivalent questions, not identical ones
Knowledge checks and scenario questions	Recall and decision-making at the moment of teaching	2	Write wrong answers that reflect real mistakes
Demonstrations, role plays, work samples	Skill performance	2 to 3	Score with a short rubric shared in advance
Confidence and commitment ratings	Early signals of whether learners will apply it	2	Ask again after 30 days and compare
Manager observation checklists	Behavior on the job	3	List three to five critical behaviors, observable in a normal week
Self, peer, and 360-degree feedback	Behavior from several views	3	Pair with records or observation so one view does not decide the result
Interviews and focus groups	Why results differ between people	3 to 4	Interview the best and worst cases, as in the Success Case Method
Business records and KPIs	Results such as errors, sales, incidents, turnover	4	Record a baseline before training starts
Cost and benefit data	Financial return	5 (Phillips)	Include learner time in costs

LMS and video data, such as completion, time watched, and attempts, show participation. They become learning evidence only when you attach scored questions to them. For worded examples at every level, see our sample evaluation questions, and use the evaluation question selector to match a question to the evidence you need.

Evaluation designs: showing the training made the difference

A design is how you compare results so a change can be credited to the training. Common designs, from simplest to most demanding:

• Post-test only: scores after training, with nothing to compare. Fine for a compliance record, weak as evidence of change.
• Pre-test and post-test: the same group before and after. Shows change but not its cause.
• Staggered rollout: one group trains now and a similar group trains later, so the later group serves as a comparison. This often works in companies where you cannot leave anyone untrained.
• Comparison group with pre- and post-tests: trained and untrained groups measured at both points.
• Time series: a business measure tracked for months before and after training, which shows whether the change outlasts normal ups and downs.

A 2003 meta-analysis of training in organizations found similar average effects for learning, behavior, and results criteria, but the evidence was uneven: 234 effects came from learning measures and only 26 from business results (Arthur and colleagues, 2003). Results-level evidence is the hardest to collect, so choose the design before training starts.

What factors influence the type of training evaluation design?

These factors decide how strong a design you need and can afford:

• Purpose: improving a program needs fast, practical feedback; deciding whether to fund or cut it needs stronger comparisons.
• Importance and cost: a program tied to safety, revenue, or a large budget justifies a comparison group.
• Scale: comparison groups need enough learners in each group to show a difference.
• Whether you can delay training for some people: if not, use a time series instead of a comparison group.
• Time: a decision due next quarter rules out a six-month follow-up.
• Expertise and data access: someone has to design the measures and get the business data.
• Culture: managers must be willing to observe and report on behavior.

How to choose a training evaluation model

Start from the question your stakeholders will ask, then pick the model built to answer it.

Match the stakeholder's question to an evaluation model

If the question is	Start with
Did people find it useful, and did they learn it?	Kirkpatrick Levels 1 and 2, with Kraiger or LTEM to design the learning measures
Are people doing it on the job?	Kirkpatrick Level 3, plus a Success Case study to learn why or why not
Was it worth the money?	Phillips ROI, or Kirkpatrick's return on expectations and contributive ROI
Should we build it, and is it running as planned?	CIPP
Does it help our clients or the community?	Kaufman's five levels

For most programs a workable plan is Kirkpatrick as the frame, scenario questions for Level 2, a manager checklist at 30 and 60 days for Level 3, and one business measure with a baseline for Level 4. Save a full Phillips ROI study for your most expensive programs. Our guide to measuring training effectiveness turns this plan into seven steps.

Training evaluation tools and software

Most teams combine four kinds of tools, because each covers different levels:

• Survey tools for Level 1 and for Level 3 follow-ups with learners and managers.
• Your LMS for enrollment, completion, and test scores.
• Assessment tools for Level 2. Our comparison of online assessment tools covers quiz builders and testing platforms.
• HR, CRM, or BI systems for the business data behind Levels 4 and 5.

When training is delivered by video, Mindstamp collects Level 1 and Level 2 evidence inside the lesson. A rating or Likert question captures usefulness and confidence, scored and scenario questions measure learning when it is taught, and voice or video responses let learners practice a conversation you can review. The report shows each learner's answers and attempts, exports to CSV, and on supported plans a SCORM package sends completion and score to your LMS. See Mindstamp for training for setups by use case.

Common questions

What is the best training evaluation model?

No model is best for every program. Kirkpatrick's four levels are the most practical default because stakeholders recognize them and they scale from a one-hour course to a year-long academy. Add Phillips when you need a financial figure, CIPP when you need guidance while the program runs, and the Success Case Method when you need to know why results vary.

Which training evaluation model is the most widely used?

Kirkpatrick's four-level model. It dates from articles Donald Kirkpatrick published in 1959 and 1960 and became the most widely used training evaluation model, as Pearlstein's 2009 chapter in the Handbook of Improving Performance in the Workplace notes.

What is the difference between training assessment and training evaluation?

Assessment measures the learner: a test, a demonstration, or a rubric score. Evaluation judges the program, using assessment results along with reactions, on-the-job behavior, business data, and costs. A strong assessment is one input to an evaluation, usually at Level 2.

Where can I read Practical Evaluation of Learning Systems in the Workplace?

It is a 2024 textbook by Dave Grant, published by Cognella. Check your course materials, your library, or the publisher for access.

Pick one program you will run next quarter. Write down the business result it should change, the two or three behaviors that would change it, and one Level 2 question for each behavior. Record a baseline for the result before the first session, and you will have the start of an evaluation instead of a satisfaction survey.