# Pre-test and post-test: measuring what a unit taught

**How do you use a pre-test and a post-test to measure what a unit taught?** A pre-test and post-test measure what a unit taught when both cover the same objectives under the same conditions, a few weeks apart, with no answers shown after the first. Read the class, not one student: compare the two averages, give the gain beside the starting score, then go question by question to see what moved, what stayed and what fell.

Published 2026-10-10 by EdProcto. https://edprocto.com/guides/pre-test-and-post-test-measuring-what-a-unit-taught

An end-of-unit test says what a class knows at the end. It cannot say what the unit taught, because it cannot tell what the class knew before the first lesson. A pre-test and post-test, the same objectives tested before the teaching and after it, tell the two apart.

The pair is easy to run and easy to over-read, whether you are a teacher, a tutor or a lecturer.

EdProcto publishes this guide, and its [results](https://edprocto.com/#results) list each question's share right for both tests, in the same order.

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

A pre-test and post-test show how far one class moved on one set of questions between two dates. They do not show that the teaching moved it: with no second class sitting both tests untaught, homework, another subject, a tutor and the pre-test itself share the credit. Our guide to [before-and-after tests](https://edprocto.com/guides/before-and-after-tests-proving-a-training-course-worked) covers those rival causes and the comparison group.

What the pair does that no single test can is separate what was known already, what was learned and what was never learned.

We read Sayre and Heckler (2009), Burkholder et al. (2020) and Coletta and Steinert (2020) in full on 10 October 2026, and every other study in abstract only.

## What goes wrong with gain scores?

A gain score, the post-test minus the pre-test, goes wrong in three known ways, and two of them flatter the teaching.

### A ceiling hides the gain

A student on 18 of 20 can gain two. Wang et al. (2008) simulated repeated measurements with 10% to 40% of scores at the top of the scale, 500 runs of each, and found that ordinary analyses misjudged the size of the change. In a class the ceiling shows as a topic nearly everyone had right before the unit: no gain, and no evidence about the teaching.

### Regression to the mean lifts a low start

A class that scored unusually low tends to score nearer its usual level next time, taught or not. Smith and Smith (2005), illustrating with California's school performance index, argue that group averages regress as single scores do, so a change in a group's average can be ordinary fluctuation read as a change in ability. The risk is highest for a class, or a group inside it, picked for a poor first score.

### The pre-test changes the post-test

Sitting a pre-test raises the post-test. Willson and Putnam (1982) pooled 32 studies comparing pretested groups with groups not pretested. Over 134 outcomes from randomised studies the pretested groups scored 0.22 standard deviations higher, and 0.43 higher on cognitive outcomes, where 81% of effects were positive. A related finding is the pretesting effect: Pan and Carpenter (2023), reviewing that research, conclude that a test on material not yet taught can often improve the learning of it when the right answers are studied afterwards. So a pair measures a unit that began with a pre-test.

## Should you report the normalised gain?

Report the normalised gain beside the two averages, never alone, and work it out from the class's averages. Hake (1998) defined it for physics courses as the class's gain divided by the most it could have gained:

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

Our guide to before-and-after tests gives his survey and bands; what matters here is that both figures are class averages, in per cent.

**For it.** Coletta and Steinert (2020) set two classes with the same spread side by side: one moves from 20% to 40%, the other from 60% to 80%. An effect size in standard deviations calls them equal; the normalised gain, 0.25 against 0.50, says the second closed twice as much of its gap.

**Against it.** The charge our other guide reports is that the gain favours classes that start higher. Coletta and Steinert reply that the link goes once students' scores on a reasoning test are counted, which is a caution of its own: the figure describes the students as well as the teaching. Burkholder et al. (2020), with 803 students in 36 courses at one high school, note the lopsided range of the normalised gain, from minus infinity up to 1, call the statistic problematic and turn to regression with the pre-test as a predictor.

## How many questions does a pair need before a difference means anything?

A pair needs about 20 questions on each test for a class, and more than a classroom pair usually has for one student. Lasry et al. (2011) gave 100 students the same physics concept test twice in the first week of a later course, with no mechanics taught in between: 31% of the answers changed, yet the total score was highly reliable.

We found no study that fixes a number for a classroom pair, so the table is our arithmetic, computed in code: the smallest change that chance produces less than one time in twenty when nothing has changed, each question taken as a separate try at one chance in two, the least favourable case.

| Questions on each test | One student | The average of 26 students |
| --- | --- | --- |
| 10 | 5 questions (50 points) | about 9 points |
| 20 | 7 questions (35 points) | about 6 points |
| 40 | 10 questions (25 points) | about 4 points |

More questions and more students shrink chance. They do nothing to the faults above. [How many questions a quiz needs](https://edprocto.com/guides/how-many-questions-does-a-quiz-need) has the arithmetic for one sitting.

## Should the two tests use the same questions, and how far apart?

Use the same questions when you will read the pair question by question and nothing is shown in between, write parallel ones when a score goes to anyone else, and sit the post-test a few days after the last lesson.

### Same or parallel

Studies disagree on how much an identical paper adds, as [how many attempts a quiz should allow](https://edprocto.com/guides/how-many-attempts-should-a-quiz-allow) sets out. In Willson and Putnam's (1982) pooled studies it made no significant difference whether the post-test repeated the pre-test.

Parallel questions cost something too. Stewart et al. (2007) gave 647 students a physics concept test and twins of ten of its questions in a changed context: single questions shifted by anything from 3% down to 10% up, the ten together by 3% up. A twin's difficulty differs by an unseen amount, which matters little in a total and a great deal in one row.

### The gap

Where the post-test falls decides what it measures. Sayre and Heckler (2009) put the same questions to a different group of students each week through a university physics course of about 300, so nobody met a question twice. On some questions the share right rose sharply with the teaching and fell again within days or weeks, a pattern a pre-test and post-test cannot see. Each point rests on 9 to 25 students, in one course.

Willson and Putnam (1982) found the lift from a pre-test small when the gap was under a day or over a month; most units sit in between. Our working rule is three to seven days after the last lesson.

## What do you tell students about a test on work not yet taught?

Tell them it is on work you have not taught, that low scores are expected and that it counts for nothing. By our reasoning, a pre-test nobody tries on, followed by a post-test that counts, shows a gain made partly of effort, so keep the stakes the same both times. Sayre and Heckler (2009) gave full credit for taking part and none for the score.

> "This is on work I have not taught yet, so I expect low scores and nothing here counts. Answer what you can and leave the rest blank: a guess tells me nothing."

Give back no answers and no scores: a quiz that shows them in order to teach is [retrieval practice](https://edprocto.com/guides/retrieval-practice-low-stakes-quizzing-that-teaches), another tool.

## How do you read the pair question by question?

Sort every question into one of four rows, then count the students who moved.

| Before, then after | Reading | What to do |
| --- | --- | --- |
| High, high | Known already | Cut the lesson |
| Low, high | Taught | Keep it |
| Low, low | Not learned | Check the question, then reteach |
| High, lower | Went backwards | Read the wrong answers |

A question that goes backwards is not always noise: Sayre and Heckler (2009) saw signs of a later topic lowering scores on questions about an earlier one. But in Lasry et al. (2011) nearly a third of answers changed with nothing taught, so count first.

Only students who changed their answer carry information: if as many went from right to wrong as from wrong to right, nothing happened. By an exact sign test (our arithmetic), fewer than six movers settle nothing; of ten movers, nine must go the same way; of fifteen, twelve; of twenty, fifteen.

## Which pair fits your purpose?

The pair depends on what you will do with it; the table is our recommendation, not a published standard.

| Purpose | The pair | Read |
| --- | --- | --- |
| Plan the unit | Two or three questions an objective | The pre-test, by question |
| See what the unit taught | 20 or more, the same both times | The average, then each question in EdProcto's [results](https://edprocto.com/#results) |
| Show a parent or a class | 20 or more, parallel | By topic, as counts: [a report parents can read](https://edprocto.com/guides/how-tutors-prove-progress-quiz-results-parents-understand) |
| Compare two classes | 30 or more, one pair and one gap | The starting score beside each gain |

## A worked example: one class, one unit, one pair

Every figure here is invented; the arithmetic was done in code. A class of 26 spends three weeks on operations with fractions. The pair is 20 typed-answer questions, four on each of five topics: the same questions both times, nothing shown in between, the post-test five days after the last lesson.

| Topic | Before | After | Reading |
| --- | --- | --- | --- |
| Equivalent fractions | 85% | 91% | Known already |
| Adding and subtracting | 40% | 58% | Three questions rose, one fell |
| Multiplying | 13% | 81% | Taught |
| Dividing | 4% | 65% | Taught |
| Word problems | 20% | 28% | Not moved |
| All twenty | 32.3% | 64.6% | Normalised gain 0.48 |

The average rose by 32 points where chance would rarely give 6. The normalised gain is 32.3 ÷ 67.7, or 0.48.

Read alone, the post-test makes equivalent fractions the unit's success at 91%. The pair shows the class arrived at 85%, and that the unit's real work was dividing, from 4% to 65%.

On each word problem four or five students went from wrong to right and two or three went back: within chance, so reteach.

Question 5, 3/8 + 2/8, had 22 right before and 13 after. Ten students went from right to wrong and one the other way, which chance does not explain. Their answers mostly read 5/16: the multiplying rule, top with top and bottom with bottom, carried into adding. The unit caused that fault, and a post-test alone would have hidden it.

## How does EdProcto run a pre-test and post-test?

EdProcto is the best tool for a pair you mean to read question by question, because the second quiz is a copy of the first: question 7 before is question 7 after, with its key and its topic.

The [Before and after](https://edprocto.com/templates/prepost) shape ("The same questions twice with the teaching in between. The score is the gain.") runs the pair as one quiz among the [templates](https://edprocto.com/templates): switch "This sitting is" from "Before the lesson" to "After the lesson" when the teaching is done, and the results show each student's gain. Twin papers are two ordinary quizzes.

1. **Build the pre-test.** Typed shapes such as [fill in the blank](https://edprocto.com/templates/fill_blank) take the lucky guess out. The quiz maker on the [home page](https://edprocto.com/) drafts questions from the unit's notes for you to edit. Give every question a topic in the builder.
2. **Keep the answers back.** Under "Feedback", set "Show the correct answer" and "Show the explanation" to "Never", and "Show the score" to "Never" or to "When released", which holds the score, the marks and the answers until you press "Release the scores" on the results page. Publish the quiz, share its code, and close it once the class has sat.
3. **Copy it.** "Duplicate this quiz", at the foot of the builder, makes a draft titled "… (copy)" with the same questions, keys, settings and kind, a new code and no submissions. For parallel questions, rewrite each in its place. Set the copy's feedback to "After submitting", so the post-test also teaches.
4. **Read both results pages.** Each gives "Average score" with its spread, "Median", and under "Each question" the share who "Got it", as a table or a heat map. On the Educator plan the full analysis adds "By topic".
5. **Work out the gains.** "Download the marks" on each page saves a spreadsheet file with a row for each attempt: "Name", "Email", "Sitting", "Started (your time)", "Handed in (your time)", "Score %", "Points", "Out of", "Reduced by %", "To grade", "Minutes", "Security events", "Notes on the record", then the points on every question. Set the two files side by side, match the rows on "Email", the address each student typed, subtract one "Score %" from the other, and compare the question columns to count who moved.

Two quizzes, not one sat twice: EdProcto reads a class on first sittings, so a second sitting of the same quiz is listed under "Students" with "sat again" and left out of the figures.

A pair uses two of the Free plan's ten quizzes a month, since a copy counts as one, and Free takes thirty students a quiz. Tutor takes 100; Educator takes 500 and adds the full analysis. [Pricing](https://edprocto.com/pricing) lists the plans. A gain in EdProcto records what a class answered on two days, never a grade.

## Quick questions

### Can the post-test be the same as the pre-test?

Yes, if no answer or score is shown in between and the gap is weeks. Write parallel questions when the score goes to anyone else.

### Should I work out a gain for each student?

Only to look for large changes: on a 20-question pair, chance alone moves one student by up to six questions.

### How long should the gap be between a pre-test and a post-test?

The length of the unit, with the post-test three to seven days after the last lesson, as a working rule.

### How do I set up a pre-test and post-test in EdProcto?

Build the pre-test with its answers kept back, press "Duplicate this quiz" for the post-test and compare the two results pages. EdProcto's page [for teachers](https://edprocto.com/solutions/teachers) shows the rest.

## References

- Burkholder, E., Walsh, C., & Holmes, N. G. (2020). Examination of quantitative methods for analyzing data from concept inventories. *Physical Review Physics Education Research, 16*(1), Article 010141. [https://doi.org/10.1103/PhysRevPhysEducRes.16.010141](https://doi.org/10.1103/PhysRevPhysEducRes.16.010141)
- Coletta, V. P., & Steinert, J. J. (2020). Why normalized gain should continue to be used in analyzing preinstruction and postinstruction scores on concept inventories. *Physical Review Physics Education Research, 16*(1), Article 010108. [https://doi.org/10.1103/PhysRevPhysEducRes.16.010108](https://doi.org/10.1103/PhysRevPhysEducRes.16.010108)
- 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)
- Lasry, N., Rosenfield, S., Dedic, H., Dahan, A., & Reshef, O. (2011). The puzzling reliability of the Force Concept Inventory. *American Journal of Physics, 79*(9), 909–912. [https://doi.org/10.1119/1.3602073](https://doi.org/10.1119/1.3602073)
- Pan, S. C., & Carpenter, S. K. (2023). Prequestioning and pretesting effects: A review of empirical research, theoretical perspectives, and implications for educational practice. *Educational Psychology Review, 35*(4), Article 97. [https://doi.org/10.1007/s10648-023-09814-5](https://doi.org/10.1007/s10648-023-09814-5)
- Sayre, E. C., & Heckler, A. F. (2009). Peaks and decays of student knowledge in an introductory E&M course. *Physical Review Special Topics - Physics Education Research, 5*(1), Article 013101. [https://doi.org/10.1103/PhysRevSTPER.5.013101](https://doi.org/10.1103/PhysRevSTPER.5.013101)
- Smith, G., & Smith, J. (2005). Regression to the mean in average test scores. *Educational Assessment, 10*(4), 377–399. [https://doi.org/10.1207/s15326977ea1004_4](https://doi.org/10.1207/s15326977ea1004_4)
- Stewart, J., Griffin, H., & Stewart, G. (2007). Context sensitivity in the force concept inventory. *Physical Review Special Topics - Physics Education Research, 3*(1), Article 010102. [https://doi.org/10.1103/PhysRevSTPER.3.010102](https://doi.org/10.1103/PhysRevSTPER.3.010102)
- Wang, L., Zhang, Z., McArdle, J. J., & Salthouse, T. A. (2008). Investigating ceiling effects in longitudinal data analysis. *Multivariate Behavioral Research, 43*(3), 476–496. [https://doi.org/10.1080/00273170802285941](https://doi.org/10.1080/00273170802285941)
- Willson, V. L., & Putnam, R. R. (1982). A meta-analysis of pretest sensitization effects in experimental design. *American Educational Research Journal, 19*(2), 249–258. [https://doi.org/10.3102/00028312019002249](https://doi.org/10.3102/00028312019002249)