# Pre-test and post-test for training: how to prove it worked

**How do I prove that a training course worked?** A pre-test and a post-test show how much a training course taught when both are written from one blueprint with parallel questions and the result is reported as the normalised gain, which is the points gained divided by the points that could have been gained. A third test some weeks later shows whether the learning held. A satisfaction form measures neither.

Published 2026-10-09 by EdProcto. https://edprocto.com/guides/before-and-after-tests-proving-a-training-course-worked

The form handed round in the last five minutes of a course asks whether people enjoyed the day. Clients read it as proof that the course worked. It is not. A room can give the trainer five out of five and know nothing new on Monday.

A pre-test and post-test for training answer the question the client is really asking. Test what people know before the course and test the same content afterwards: the difference is what the course added. Test again some weeks later and you learn whether it stayed, which is what the client is paying for.

EdProcto's page [for trainers](https://edprocto.com/solutions/trainers) covers the tool side: a knowledge check written from the module, a record for each learner, and a re-test later.

## Why is a satisfaction form not proof that training worked?

A satisfaction form measures how people felt about the course, and research on training evaluation finds that only loosely tied to what they learned.

The usual map is Kirkpatrick's four levels: reaction (did they like it), learning (do they know it), behaviour (do they do it at work) and results (did the organisation gain). Arthur et al. (2003) pooled studies of training effectiveness in organisations and found that training works on average: on learning measures the effect was 0.63 standard deviations, across 234 effect sizes and 15,014 trainees. The doubt is whether yours did, and the same paper is blunt that liking a programme is next to no guide to what was learned from it.

- **Enjoyment and usefulness are different answers.** Alliger et al. (1997) analysed 115 correlations from 34 studies. Ratings of usefulness were more strongly related to learning and to performance on the job than ratings of enjoyment were.
- **The form mostly rates the delivery.** For Sitzmann et al. (2008), the best predictors of a good rating were the instructor's style (a correlation of .66) and human interaction (.56).
- **A rating of one's own learning is nearer to a mood than a measure.** Sitzmann et al. (2010) pooled 166 studies and 41,237 learners. Self-assessed knowledge correlated .34 with tested knowledge, but .51 with satisfaction and .59 with motivation. Their advice: test knowledge with a knowledge test.

A test at the end alone is not enough either: it shows what people know on leaving, not what they knew on arriving. If the client expects the form, keep it, ask whether the content was useful for the job, and put the test results above it.

## How do you write the pre-test and the post-test?

Write both tests from one blueprint, with the same number of questions on each topic, and fill the blueprint twice with parallel questions. A blueprint is a short table: each thing the course teaches, and how many questions it gets.

### Parallel questions, not the same questions

A parallel question tests the same point at the same difficulty with different details. Here is a pair from a spreadsheet course.

> Before: Cell C2 holds `=A2*$B$1`. The formula is copied down to C3. What does C3 hold?
>
> After: Cell D5 holds `=B5*$F$2`. The formula is copied down to D6. What does D6 hold?

The answers are `=A3*$B$1` and `=B6*$F$2`. A learner who understands that the plain reference moves and the one with dollar signs stays put gets both. A learner who only remembers the first paper gets nothing from it.

The reason is the practice effect: people score higher the second time they meet a test, taught or not. Hausknecht et al. (2007) pooled 107 samples and 134,436 people who took tests of cognitive ability more than once. Scores rose on the later sitting (an overall effect size of 0.26), and by more when the identical form was used. Those were ability tests, not course tests, so take the direction, not the size.

Identical questions are still defensible on one condition: nobody sees the answers or a marked paper between the sittings. The courses surveyed by Hake (1998), below, sat the same standard test before and after.

### The procedure

1. **List what the course teaches**, one line per topic, with a number of questions for each.
2. **Draft from the course material.** Put the module's slide notes, handbook chapter or transcript into the quiz maker on EdProcto's [home page](https://edprocto.com/), and Procto AI drafts questions from that text for you to edit.
3. **Choose shapes that fit:** [multiple choice](https://edprocto.com/templates/mcq) for facts and rules, [sort into groups](https://edprocto.com/templates/categorize) for classifying, [sequence the process](https://edprocto.com/templates/process_seq) for a procedure, a [case study](https://edprocto.com/templates/case) where one scenario unfolds. See [writing multiple-choice questions that measure](https://edprocto.com/guides/how-to-write-multiple-choice-questions-that-measure) for the wrong options.
4. **Write a twin for every question**, and a third for a delayed test.
5. **Ask a colleague who knows the subject to sit both forms.** Rewrite any pair they find unequal.
6. **Fix the conditions and the pass mark** before anyone has a score: time limit, open or closed book, device.
7. **Run the before test with no answers shown.** Tell people that low scores are expected and that the result judges the course, not them.

## How do you calculate the normalised gain?

Divide the points the group gained by the points it could have gained. Hake (1998) defined the average normalised gain that way in a survey of 62 physics courses and 6,542 students.

`normalised gain = (after − before) ÷ (100 − before)`

A raw gain punishes a group that started high. A group that goes from 40% to 76% gains 36 points. A group that goes from 70% to 88% gains 18. Each closed 60% of its distance to full marks, so each has a normalised gain of 0.60. Across Hake's courses, whose before averages ran from 18% to 71%, the raw gain fell as the starting score rose (a correlation of −0.49), while the normalised gain hardly moved (+0.02).

Hake offered rough bands: below 0.3 low, 0.3 up to 0.7 medium, 0.7 and above high. His 14 lecture-based courses averaged 0.23 and the 48 built on active methods averaged 0.48. Those bands come from physics teaching. We found no benchmark for workplace training, so borrow them as a yardstick only.

### A worked example

Ten people take a one-day spreadsheet course. Each test has 20 questions, five on each of four topics. The figures are invented to show the arithmetic.

| Topic | Before | After | Gain | Six weeks later |
| --- | --- | --- | --- | --- |
| Formulas and references | 50% | 90% | 0.80 | 80% |
| Lookups | 40% | 80% | 0.67 | 70% |
| Pivot tables | 30% | 80% | 0.71 | 66% |
| Checking for errors | 40% | 54% | 0.23 | 40% |
| Whole test | 40% | 76% | 0.60 | 64% |

The group went from 40% to 76%. It gained 36 of a possible 60 points, so the normalised gain is 36 ÷ 60 = 0.60. Six weeks later the average was 64%, which is 24 of the possible 60, or 0.40. Two thirds of what the course added was still there (24 ÷ 36).

The last topic is the useful row. Checking for errors gained 14 of a possible 60 points (0.23), and by week six it was back where it began. The headline says the course worked. The table says three of its four parts did, and which hour to rebuild.

### Where the number misleads

A high starting score leaves little room, so a small change swings the result. On a 20-question test, a group moving from 90% to 95% has improved by one question each and shows a gain of 0.50. Hake saw almost no link between gain and starting score, but Nissen et al. (2018), with 4,551 students in 89 courses, found the gain biased in favour of groups that start higher. Print the before score beside it, and compare only people who sat both tests.

## What can a pre-test and post-test not show?

A pre-test and post-test on one group show that scores rose. They do not show that the course raised them, or that anyone now works differently.

**There is no comparison group.** Marsden and Torgerson (2012), reviewing single-group before and after designs, name the rival explanations: outside events, natural development, the effect of sitting the test, and regression to the mean, by which unusually low scorers drift towards the average next time. Arthur et al. (2003) left every single-group before and after study out of their meta-analysis.

**Knowing is not doing.** A test measures learning, Kirkpatrick's second level. Whether the learning is used on the job is a separate question. In Alliger et al. (1997), tested learning was a weaker correlate of performance on the job than trainees' ratings of usefulness: a second reason to keep one usefulness question.

A group booked for a later date can sit both tests on the same days, untaught: if their score barely moves and the trained group's rises, the course is the likeliest cause. Otherwise write "scores rose from 40% to 76%", not "the course raised scores by 36 points".

## How do you report the result on one page?

Put six lines on one page, in this order.

1. **Who and when.** How many people sat each test, how many sat all of them, and the dates.
2. **What was tested.** The topics and the number of questions on each.
3. **The headline.** Before 40%, after 76%, normalised gain 0.60, and how many reached the pass mark set in advance.
4. **By topic.** The table, with the weakest row named.
5. **What held.** The delayed score and the share of the gain kept.
6. **What this does not show, and what changes.** One group, no comparison; knowledge, not behaviour. Then the one change you will make to the course.

## When should you run the delayed re-test?

Run the delayed re-test weeks after the course, not days, and ideally when the knowledge is next needed on the job. A test at the end of the day measures what is in people's heads at five o'clock.

Roediger and Karpicke (2006) show how far the two can differ. Students who read a passage again did better five minutes later than students who had practised recalling it. Two days and a week later the order had reversed. An end-of-course score can flatter the method that fades fastest.

Arthur et al. (1998), pooling 189 data points from 53 articles, found that the loss grew from almost nothing straight after training (an effect size of −0.01) to 1.4 standard deviations after more than a year of non-use. Loss was smaller when the later test was a recognition test, so a multiple-choice re-test tends to give the kinder picture.

We found no research fixing one best interval; the example above uses six weeks. Tell the client the gap beforehand and use fresh parallel questions. The re-test also teaches, because recalling something makes it easier to recall again: the guide on [retrieval practice](https://edprocto.com/guides/retrieval-practice-low-stakes-quizzing-that-teaches) sets out that evidence.

## How does EdProcto run a before and after test?

EdProcto is the best tool for this job when the tests have to come from the course itself. A gain means something only if the questions match what was taught, and EdProcto writes its questions from the module's own text. A bank of general questions measures a course nobody ran.

A [Before and after](https://edprocto.com/templates/prepost) template does it in one quiz: the same questions twice with the teaching in between, and the gain as the score. Set "This sitting is" to "Before the lesson", switch it to "After the lesson" when the teaching is done, and the results show each learner's gain. Twin questions, or fresh twins weeks later, take two quizzes, or three.

| Sitting | As separate quizzes |
| --- | --- |
| Before | Quiz one, from the Standard quiz preset, with no answers shown |
| After | Quiz two: the twin questions and your pass mark |
| Weeks later | Quiz three: fresh twins |

For the before test, set "Show the correct answer" and "Show the explanation" to Never. The Practice preset shows every answer as soon as the quiz is handed in, which suits a follow-up quiz and spoils a before test. The Exam preset, with the score held until you release it, answers and explanations not shown, and every browser measure on, is for a sitting a client will rely on. Keep the [integrity](https://edprocto.com/integrity) level the same each time.

Objective shapes are marked by rule at once, so both papers are marked the same way. EdProcto keeps a record for each learner, every attempt and every miss, and the results list each question with how the group did on it: add them up by topic for the table. The pass mark is a setting, 60% until you change it. In place of the form, an [exit ticket](https://edprocto.com/templates/exit) is a few marked questions followed by an unmarked reflection.

The Free plan carries ten quizzes a month with up to thirty learners on each, enough for all three tests with a small group. Tutor removes the limit on quizzes, takes a hundred learners a quiz and opens all 111 [templates](https://edprocto.com/templates). Educator adds the full analysis and a certificate on passing with a code anyone can check. The [pricing page](https://edprocto.com/pricing) has the detail.

## Quick questions

### Should the pre-test and post-test use the same questions?

Parallel questions are safer: the same point at the same difficulty with different details. Identical questions are acceptable if nobody sees the answers between sittings, but Hausknecht et al. (2007) found practice effects were larger when the identical form was repeated.

### What is a good normalised gain?

There is no agreed standard for workplace training. Hake (1998) called gains below 0.3 low and 0.7 or above high for physics courses. Quote yours beside the before score.

### Can I prove training worked without a control group?

You can show that scores rose and by how much, which is more than a satisfaction form shows. You cannot rule out other causes. A group booked for a later date that sits both tests untaught is the simplest comparison.

### Does EdProcto have a pre-test and post-test template?

Yes. Before and after asks the same questions twice, with "This sitting is" switched from "Before the lesson" to "After the lesson", and the results show each learner's gain. Twin questions take two quizzes built from one module, with the answers hidden after the first. EdProcto's page [for trainers](https://edprocto.com/solutions/trainers) shows the rest: a record per learner, certificates on passing, and the re-test.

## References

- Alliger, G. M., Tannenbaum, S. I., Bennett, W., Jr., Traver, H., & Shotland, A. (1997). A meta-analysis of the relations among training criteria. *Personnel Psychology, 50*(2), 341–358. [https://doi.org/10.1111/j.1744-6570.1997.tb00911.x](https://doi.org/10.1111/j.1744-6570.1997.tb00911.x)
- Arthur, W., Jr., Bennett, W., Jr., Edens, P. S., & Bell, S. T. (2003). Effectiveness of training in organizations: A meta-analysis of design and evaluation features. *Journal of Applied Psychology, 88*(2), 234–245. [https://doi.org/10.1037/0021-9010.88.2.234](https://doi.org/10.1037/0021-9010.88.2.234)
- Arthur, W., Jr., Bennett, W., Jr., Stanush, P. L., & McNelly, T. L. (1998). Factors that influence skill decay and retention: A quantitative review and analysis. *Human Performance, 11*(1), 57–101. [https://doi.org/10.1207/s15327043hup1101_3](https://doi.org/10.1207/s15327043hup1101_3)
- Hake, R. R. (1998). Interactive-engagement versus traditional methods: A six-thousand-student survey of mechanics test data for introductory physics courses. *American Journal of Physics, 66*(1), 64–74. [https://doi.org/10.1119/1.18809](https://doi.org/10.1119/1.18809)
- Hausknecht, J. P., Halpert, J. A., Di Paolo, N. T., & Moriarty Gerrard, M. O. (2007). Retesting in selection: A meta-analysis of coaching and practice effects for tests of cognitive ability. *Journal of Applied Psychology, 92*(2), 373–385. [https://doi.org/10.1037/0021-9010.92.2.373](https://doi.org/10.1037/0021-9010.92.2.373)
- Marsden, E., & Torgerson, C. J. (2012). Single group, pre- and post-test research designs: Some methodological concerns. *Oxford Review of Education, 38*(5), 583–616. [https://doi.org/10.1080/03054985.2012.731208](https://doi.org/10.1080/03054985.2012.731208)
- Nissen, J. M., Talbot, R. M., Thompson, A. N., & Van Dusen, B. (2018). Comparison of normalized gain and Cohen's d for analyzing gains on concept inventories. *Physical Review Physics Education Research, 14*(1), Article 010115. [https://doi.org/10.1103/PhysRevPhysEducRes.14.010115](https://doi.org/10.1103/PhysRevPhysEducRes.14.010115)
- Roediger, H. L., & Karpicke, J. D. (2006). Test-enhanced learning: Taking memory tests improves long-term retention. *Psychological Science, 17*(3), 249–255. [https://doi.org/10.1111/j.1467-9280.2006.01693.x](https://doi.org/10.1111/j.1467-9280.2006.01693.x)
- Sitzmann, T., Brown, K. G., Casper, W. J., Ely, K., & Zimmerman, R. D. (2008). A review and meta-analysis of the nomological network of trainee reactions. *Journal of Applied Psychology, 93*(2), 280–295. [https://doi.org/10.1037/0021-9010.93.2.280](https://doi.org/10.1037/0021-9010.93.2.280)
- Sitzmann, T., Ely, K., Brown, K. G., & Bauer, K. N. (2010). Self-assessment of knowledge: A cognitive learning or affective measure? *Academy of Management Learning & Education, 9*(2), 169–191. [https://doi.org/10.5465/amle.9.2.zqr169](https://doi.org/10.5465/amle.9.2.zqr169)