Some knowledge shows only against its nearest neighbour. A student who can recite what metaphase is may still not tell it from anaphase on a slide. A compare two images question tests that: two pictures side by side, and a question that only the difference between them answers.
This guide is published by EdProcto, whose Compare two images shape is two pictures, one of them right, marked by rule.
What can a question with two images ask, and what does each kind test?
It can ask four things: which picture is which, which one is correct, whether the two show the same thing, and what differs between them.
| Kind | A stem | What it tests | Shape in EdProcto |
|---|---|---|---|
| Which is which | "Which slide shows metaphase?" | Telling a thing from its nearest neighbour | Compare two images |
| Which is correct | "Which weld is acceptable?" | Visual judgement against a named standard | Compare two images |
| Same or different | "Do both graphs show the same relationship?" | Seeing past a surface difference | Compare two images, with the captions "Same" and "Different" |
| What differs | "What changed between map K and map M, and why?" | Finding the feature unaided, then explaining it | Caption it, marked against your rubric |
The first three are picks between two, so half of all blind guesses are right, and with image questions the number of options matters. Rutgers et al. (2018) reviewed 369 image-based questions from seven progress tests for Dutch radiology residents: those with more answer options were generally harder and separated strong candidates from weak ones better.
So never rest a judgement on one pair. A student guessing through ten pairs gets eight or more right about one time in 18 (56 chances in 1,024). A third picture helps too, and so does asking for the reason.
For one right picture among several, use picture choice; for where a thing is, a hotspot.
What does research say about learning from comparison?
It says that studying two cases together usually teaches more than studying them one at a time, and that what learners compare decides what they learn.
| Study | Who and what | Found |
|---|---|---|
| Alfieri et al. (2013) | A meta-analysis of 57 experiments | Comparing cases taught more than single cases or traditional instruction (d = 0.50), and more with perceptual content and the principle given afterwards |
| Rittle-Johnson and Star (2007) | 70 seventh-graders; two solution methods, compared or studied in turn | Comparing: greater gains in procedure and flexibility, equal gains in concepts |
| Kok et al. (2013) | 61 medical students; chest X-rays of 12 diseases, with or without a normal film beside | Helped with diseases at one location; no significant difference for those in both lungs |
| Kok et al. (2015) | 84 medical students; four kinds of pair | Most efficient: two films of one disease, or of two diseases |
Does telling students what to look for help?
For a feature of a picture, yes, and fast. Biederman and Shiffrar (1987) asked untrained people to judge the sex of day-old chicks in 18 pictures. Told where the critical structure was and which contrast in shape to look for, they rose from slightly above chance to a level comparable with a sample of experts.
For the idea behind the pictures, telling first can cost something. Schwartz et al. (2011) taught adolescents physics from contrasting cases. Students told the formula first used it on word problems as well as students who invented it from the cases and were told afterwards, but the inventors learned the underlying structure better and transferred it more often.
Left alone, students may compare the wrong things. Kok et al. (2015) tracked eye movements: comparisons of normal anatomy were especially common (45.8%), and those between abnormalities few. So name the criterion in the stem, and put the telling in the explanation shown afterwards.
What do dermatology and histology add?
They add perceptual learning: practice at classifying many varied pictures, with feedback. Krasne et al. (2013) gave medical students a module of skin histopathology images, one at a time with a choice among four. Accuracy and speed improved, and scores on unseen cases six to seven weeks later stayed well above the pre-test. Rimoin et al. (2015) report large gains (effect sizes of 1.1 to 2.2) from such a module on skin lesions, in self-selected groups from one class.
That is evidence for discrimination training, not for pairs as such. Its lesson for a quiz: Krasne et al. kept 48 of their 261 images out of the module for the tests. Test with pictures the class has not seen, or you test memory of a picture.
What we could not confirm
These studies are about learning. We found none that set a two-image question against a one-image question in a quiz, so the step to assessment is our reasoning. We read Krasne et al. (2013) in full and the others in abstract alone.
How do you choose a pair that differs in one thing that matters?
Start from a confusion you have seen in class, and find two pictures alike in everything except the feature that settles it.
- One difference that matters. Metaphase beside anaphase, a painting beside its copy, two designs for one poster. If three things differ, a right answer does not say which one the student used.
- Everything else held still. One stain and magnification, the same axes and scales on two graphs, one scale and orientation on two maps, one viewpoint for a before and after.
- Both kinds of pair in a same-or-different set. "Same" pairs that look different, and "different" pairs that look alike.
How do you prepare the two images so that only knowledge separates them?
Prepare them as one object: the same size, style and tag, with nothing in either that the other lacks except the feature.
- Use pictures you may use. Your own, or ones whose terms allow it, with no identifiable person in them. Read the terms attached to each picture: this is not legal advice.
- Crop both to one shape, save both at one size, and match brightness, background and colour. Make the background solid: the quiz screen is dark, and a drawing saved with a transparent background loses its black lines.
- Remove what tells. Labels, arrows, watermarks, any name or date printed on the image, and the file name: the address of each picture is sent to the student's browser, where
metaphase.jpggives the answer andk.jpgdoes not. - Tag them K and M. A and B are printed on the answer buttons by position.
- Choose the build. Two separate pictures, each an answer the student taps, or one combined picture above two answers in words. For the combined picture, place both crops on one canvas in a slide or image editor, at one height with a narrow gap, print K and M in the same corner of each half, and export one image file. Vary the side the right picture is on.
- Check it small. In EdProcto's Compare two images, by our arithmetic from the page's layout, a picture that is an answer is shown about 325 pixels wide at most and nearer 270 on a phone, and a combined picture about 300 on a phone held upright: 150 for each half. Shrink your picture to that width, crop tighter if the difference has gone, and check the question in "Preview as student".
Where the files should live, and what to offer a student who cannot see the pictures, are covered in the picture choice guide.
How do you mark an explanation of the difference?
Mark it against a rubric that pays for three things in turn: the feature, the picture it belongs to, and what it means.
| Point | Earned by | Not earned by |
|---|---|---|
| The feature | Describing what differs: "lined up across the middle" | A statement true of both pictures: "the cell is dividing" |
| The picture | Tying it to a tag: "in K" | "One of them has a line" |
| The meaning | Saying what the feature shows: "so K is metaphase" | The name alone, with no feature |
Accept plain words for the feature unless the term is the lesson, and write the model answer so that it describes both pictures, because whoever marks works from the words. A right pick with an empty reason was a guess, or something in your pictures gave the answer away.
Two worked examples: a slide and a river bend
Metaphase or anaphase, for a biology class
The stem: "Which of the two slides shows metaphase?" Each picture shows one dividing cell from an onion root tip, in one stain, magnification and crop. In Slide K, the right one, the chromosomes lie in a line across the middle of the cell. In Slide M they have parted into two groups drawn towards opposite ends.
A second question asks what shows metaphase, for two points: the line across the middle, and the contrast with Slide M.
| Student's answer | Pick | Reason, of 2 |
|---|---|---|
| "K. The chromosomes are lined up across the middle. In M they have split into two groups moving to the ends, which is anaphase." | Right | 2 |
| "K, because the cell is dividing." | Right | 0: true of both slides |
| "M. You can see the chromosomes clearly." | Wrong | 0: true of both; this student's rule may be "chromosomes visible" |
A river bend fifty years apart, for a geography class
One combined picture holds two sketch maps of the same reach, at one scale and orientation: map K, and map M drawn fifty years later. In K the river swings round a loop with a narrow neck. In M it runs straight across the neck, and the loop is a separate curved lake. The stem: "What has changed between map K and map M, and what process caused it?"
Three points: the loop cut off as an oxbow lake; erosion on the outside of the bends narrowing the neck until the river broke through; deposition sealing the old loop.
| Student's answer | Points, of 3 |
|---|---|
| "The loop in K is cut off as a lake in M. The river wore away the outside of the bends until the neck was thin, broke through in a flood, then dropped sediment that sealed the old loop." | 3 |
| "The river is straighter in M and there is a new lake." | 1: the change, no process |
| "It rained more, so a lake formed beside the river." | 0: the lake is not seen as the river's old loop |
How does EdProcto build, show and mark a comparison?
EdProcto is the best tool for a comparison question when you want both halves of the evidence: the pick, marked by rule on the server with the key never sent to the student's device, and the reason, kept for a reader.
- Building. Add a question in "Compare two images". It opens with four rows under "Options · mark the right one": remove two with × for a pair, or keep a third picture. Every row carries a letter and two boxes, "A caption" and "Image address": the tag goes in one, the picture's address in the other, and you mark the right row. A combined picture goes in the "Image address" box above "Prompt", with the rows' addresses left empty, so the answers are captions alone, such as "Same" and "Different".
- Showing. A combined picture stands above the prompt. Answer pictures stand two to a row, or one above the other on a screen 720 pixels wide or narrower, and a tap anywhere on one chooses it. With "Shuffle the options" on, as it is by default, the two can swap sides from student to student: never write "left" or "right" in a stem.
- Marking. The right picture earns the question's points; the other, or none, earns nothing. With "Negative marking" on, a wrong pick between two took away as many points as the question was worth when we ran EdProcto's marking function, and a blank took none. A study quiz checks each answer as it is given; an exam is marked at hand-in.
- A written comparison. "Another shape" adds a question of a different kind. Caption it takes one image address, a prompt, "Model answer (what a full answer contains)" and "Rubric (what earns the points)". Its answer waits as "to grade" until you enter the points and press "Save my grade", or press "Ask Procto AI" on it: that sends the prompt, your model answer, your rubric and the student's words, and saves a mark you can change.
- On paper. "Paper" prints a tick box for each caption: put the pictures on the board, and see how to print a quiz.
Marking the differences on the picture itself is Spot the difference. Two pictures on a screen can be photographed like any others, so match the integrity settings to what the quiz decides.
Free gives ten quizzes a month with the core templates, marked by rule. Tutor opens all 111 templates and Procto AI's grading, and Educator adds the full analysis: see pricing.
Quick questions
What is a compare two images question?
A question that shows two pictures side by side and asks which is which, which is correct, whether they match, or what differs. EdProcto's Compare two images marks the pick by rule.
Should the two pictures be labelled A and B?
No. Use neutral tags such as K and M: the quiz screen prints A and B on the answer buttons by position, and shuffling moves the pictures.
Can students compare two images on a phone?
Yes, if the difference survives the size. In EdProcto two answer pictures stand one above the other on a phone, and a combined picture keeps them side by side at about half the width each.
Can Procto AI mark a written comparison?
It marks the words against your model answer and rubric when you press "Ask Procto AI" on an answer, so write what each picture shows into the model answer. Its mark is yours to keep or change.
References
- Alfieri, L., Nokes-Malach, T. J., & Schunn, C. D. (2013). Learning through case comparisons: A meta-analytic review. Educational Psychologist, 48(2), 87–113. https://doi.org/10.1080/00461520.2013.775712
- Biederman, I., & Shiffrar, M. M. (1987). Sexing day-old chicks: A case study and expert systems analysis of a difficult perceptual-learning task. Journal of Experimental Psychology: Learning, Memory, and Cognition, 13(4), 640–645. https://doi.org/10.1037/0278-7393.13.4.640
- Kok, E. M., de Bruin, A. B. H., Leppink, J., van Merriënboer, J. J. G., & Robben, S. G. F. (2015). Case comparisons: An efficient way of learning radiology. Academic Radiology, 22(10), 1226–1235. https://doi.org/10.1016/j.acra.2015.04.012
- Kok, E. M., de Bruin, A. B. H., Robben, S. G. F., & van Merriënboer, J. J. G. (2013). Learning radiological appearances of diseases: Does comparison help? Learning and Instruction, 23, 90–97. https://doi.org/10.1016/j.learninstruc.2012.07.004
- Krasne, S., Hillman, J. D., Kellman, P. J., & Drake, T. A. (2013). Applying perceptual and adaptive learning techniques for teaching introductory histopathology. Journal of Pathology Informatics, 4, Article 34. https://doi.org/10.4103/2153-3539.123991
- Rimoin, L., Altieri, L., Craft, N., Krasne, S., & Kellman, P. J. (2015). Training pattern recognition of skin lesion morphology, configuration, and distribution. Journal of the American Academy of Dermatology, 72(3), 489–495. https://doi.org/10.1016/j.jaad.2014.11.016
- Rittle-Johnson, B., & Star, J. R. (2007). Does comparing solution methods facilitate conceptual and procedural knowledge? An experimental study on learning to solve equations. Journal of Educational Psychology, 99(3), 561–574. https://doi.org/10.1037/0022-0663.99.3.561
- Rutgers, D. R., van Raamt, F., van der Gijp, A., Mol, C., & ten Cate, O. (2018). Determinants of difficulty and discriminating power of image-based test items in postgraduate radiological examinations. Academic Radiology, 25(5), 665–672. https://doi.org/10.1016/j.acra.2017.10.014
- Schwartz, D. L., Chase, C. C., Oppezzo, M. A., & Chin, D. B. (2011). Practicing versus inventing with contrasting cases: The effects of telling first on learning and transfer. Journal of Educational Psychology, 103(4), 759–775. https://doi.org/10.1037/a0025140