Written and maintained by CASRAI Editorial Board
Published
On this page
A mental rotation task measures how people compare objects shown at different orientations, usually by deciding whether two figures represent the same object or different objects. Researchers analyse accuracy and response time, often across rotation angles. The experimental task and the paper-and-pencil Mental Rotations Test are related, but their items, scores and supporting evidence differ.
Mental rotation task: definition and research purpose
In a mental rotation experiment, a participant judges a spatial relationship while the researcher varies orientation. The classic Shepard–Metzler task presents perspective drawings of three-dimensional objects: some pairs can be aligned by rotation, whereas reflected pairs cannot. Its central finding was an approximately linear increase in correct-response time with angular difference. This makes the task useful for investigating the time course of spatial transformation, rather than simply counting solved puzzles.
The response still includes seeing the figures, choosing a strategy, deciding and pressing a response control. A total response time therefore is not a direct stopwatch measurement of an isolated mental operation. Shepard and Metzler explicitly recognised this distinction. Specify the comparison, manipulation and recorded outcome when describing a study; the label “mental rotation” alone does not identify its measurement procedure.
Task versions, items and response formats
The original experiment and later implementations answer different methodological needs. A trial-based design retains the response and timing for each object pair. A paper test instead yields an item score under its administration rules. Choose the format around the research outcome: angle-dependent response times require trials with known orientations, while an item total describes performance on a particular test form.
Vandenberg and Kuse’s original 1978 Mental Rotations Test contains 20 items. Each presents one reference figure and four alternatives, including two correct rotated matches. Their recommended item scoring awards credit only when both choices are correct; otherwise the item receives none. Do not silently substitute a one-correct-answer rule or a partial-credit rule and describe the resulting total as the same score.
Computer implementations also differ. PsyToolkit’s documented demonstration uses a two-dimensional reference above two alternatives. Millisecond’s Inquisit implementation uses three-dimensional cube-object pairs from Ganis and Kievit, with separate same and different trials. Its default practice block differs from the published validation procedure. The data table below compares three documented protocols, distinguishing experimental studies from software defaults; its trial counts are examples, not a prescribed sample size.
Scoring formulas: accuracy, response time and rotation slope
There is no single score shared by every mental rotation task. For a binary trial task, define accuracy as A = C / N, where A is the proportion correct, C is the number of correct test responses and N is the number of test trials included in the accuracy denominator. Multiply A by 100 for percentage accuracy. State whether omitted or timed-out trials remain in N; changing that denominator changes the interpretation.
For a descriptive angle analysis, fit RTᵢ = a + bθᵢ + εᵢ. Here RTᵢ is the response time in milliseconds for observation i, θᵢ is its angular disparity in degrees, a is the fitted intercept in milliseconds, b is the slope in milliseconds per degree, and εᵢ is the residual difference between observed and fitted time. An observation might be a condition mean in a demonstration, or a trial in an appropriately specified study model.
The ordinary least-squares calculations are b = Σ[(θᵢ − θ̄)(RTᵢ − RT̄)] / Σ(θᵢ − θ̄)² and a = RT̄ − bθ̄. The bars indicate arithmetic means and Σ means summation across included observations. When b is positive, its reciprocal can express an implied speed: v = 1000 / b degrees per second. The factor 1000 converts milliseconds to seconds. The slope measures time per degree; the reciprocal measures degrees per time.
Worked example using published response times
Ganis and Kievit’s Table 1 reports mean response times of 1801, 2415, 3026 and 3191 milliseconds at 0°, 50°, 100° and 150°, respectively. These published values combine same and different trials. The following calculation is a descriptive reanalysis of those four rounded means, not a new participant-level result.
- The mean angle is (0 + 50 + 100 + 150) / 4 = 75°. The mean response time is (1801 + 2415 + 3026 + 3191) / 4 = 2608.25 ms.
- The angle deviations are −75, −25, 25 and 75. Their squared sum is 5625 + 625 + 625 + 5625 = 12500.
- The time deviations are −807.25, −193.25, 417.75 and 582.75. Multiplying corresponding angle and time deviations and adding gives 60543.75 + 4831.25 + 10443.75 + 43706.25 = 119525.
- The slope is b = 119525 / 12500 = 9.562 ms/degree. The intercept is a = 2608.25 − 9.562 × 75 = 1891.1 ms.
- The implied speed is v = 1000 / 9.562 = 104.58 degrees/second, approximately 105 when rounded to a whole number.
The fitted equation is therefore RT = 1891.1 + 9.562θ. At 100°, it predicts 2847.3 ms, compared with the observed mean of 3026 ms. That difference is a reminder to inspect the observations, rather than assume an exact straight line. The authors also describe flattening at the largest angle. Four group means cannot establish the distribution of individual slopes or the uncertainty of a population estimate.
Interpretation: what each result means
Interpret the outcome in its units and alongside the task conditions. For the example, an additional degree corresponds to about 9.56 ms of additional fitted response time across the observed angles. This is an average relationship within that summary, not a universal conversion factor for spatial ability.
| Result | Reading guidance |
|---|---|
| Larger positive b, in ms/degree | More additional response time per degree in the fitted model; describe it as a steeper time cost. |
| Larger positive v, in degrees/second | A faster implied transformation rate, because v is the reciprocal of the time slope. |
| Intercept a | The fitted time at zero angular disparity; do not label all of it “rotation time.” |
| Higher accuracy | More correct responses under the stated scoring and denominator; interpret with timing and omitted responses. |
Millisecond’s manual separately records accuracy and mean correct-response time, including values by angle and trial type. Keeping those outputs separate makes an apparent speed improvement accompanied by more errors visible. Before comparing conditions, check that the retained trials, response deadlines and scoring rules are comparable.
Psychometric properties and validity evidence
Vandenberg and Kuse reported KR-20 internal consistency of .88 and a test–retest correlation of .83 in their original research. The full paper also reports .70 in an age-corrected sample. It describes stronger associations with spatial tests than with verbal measures. These are findings from particular samples and procedures, not coefficients automatically inherited by any task displaying rotated shapes.
Separate three questions in a study plan: whether items produce a coherent total, whether scores remain stable across occasions, and whether the outcome relates to the intended construct. CASRAI’s guide to convergent and discriminant validity explains the last distinction. For a new implementation, report evidence for its own stimuli, timing, scoring and population, rather than attach a historical reliability number to a different measurement.
Peters and colleagues found substantial practice effects, including transfer between alternate MRT versions. Consequently, switching forms does not by itself make improvement evidence of an intervention effect. A repeated-measures design needs an explicit account of prior exposure, practice and comparison conditions; see the discussion of systematic change in test–retest reliability studies.
Assumptions and common mistakes
Specify the rotation axis and the meaning of “different.” A reflection and a rotation are different transformations. Ganis and Kievit note that two of their “different” objects can align around a nonvertical axis. Their operational comparison concerns vertical-axis rotation. Treat stimulus classification as part of the experimental definition, rather than assume every different pair is impossible to align under any transformation.
Keep accuracy and timing exclusions distinct. The documented PsyToolkit output distinguishes correct responses, errors and responses that are too slow. Preserve those categories in a trial log. As an analysis recommendation, define timing exclusions before inspecting condition differences, retain an exclusion count, and identify which responses enter each mean. Deleting every timeout before calculating accuracy can make a difficult condition look deceptively accurate.
Check whether the linear summary fits. The worked example describes a trend across angle means. In a study, inspect the angle-specific pattern and trial types before interpreting a reciprocal slope. A zero or negative fitted slope does not support a meaningful positive rotation-speed estimate. A pooled slope also does not show that every participant has the same slope.
Check score range and exposure. Inspect whether participants cluster at the task’s minimum or maximum score; such floor and ceiling effects can hide differences. Record previous task exposure and practice separately from experimental trials. A transparent protocol should let another researcher determine what participants saw, what counted as correct and which observations contributed to each reported outcome.
How to report a mental rotation task: APA-style example
Describe the stimulus source, dimensionality, rotation axis, angle levels, trial types, practice and test counts, response controls, deadline and scoring rule in the methods. In the results, identify the analysis unit and provide accuracy alongside correct-response timing. If participants supply repeated observations, describe how the analysis handles those observations rather than treating all trials as unrelated people.
APA-style descriptive example using the published aggregate data: “A descriptive regression of the four angle-specific means reported by Ganis and Kievit (2015, Table 1) yielded a response-time slope of 9.56 ms/degree and an intercept of 1891.10 ms. The reciprocal slope corresponded to 104.58 degrees/s. The analysis used rounded group means collapsed across same and different trials.”
This example deliberately reports no participant-level significance test or confidence interval: those cannot be recovered from the four means alone. For an original study, report the actual participant sample, retained observations and uncertainty from the model used. Keep a group-level descriptive reanalysis separate from claims about individual differences or intervention effects.
How to calculate the slope in Excel, R and Python
All three examples use the same four published means and ordinary least squares, with angle as the predictor. They should return slope 9.562 and intercept 1891.1. Enter response time in milliseconds consistently; entering seconds instead changes the slope’s units and requires a different reciprocal conversion.
Excel
Put angles 0, 50, 100 and 150 in A2:A5, and times 1801, 2415, 3026 and 3191 in B2:B5. Use =SLOPE(B2:B5,A2:A5) for milliseconds per degree and =INTERCEPT(B2:B5,A2:A5) for the intercept. If the slope is in D2, =1000/D2 gives degrees per second. Response times are the first argument; reversing the ranges fits another regression.
R
angle = c(0, 50, 100, 150)
rt = c(1801, 2415, 3026, 3191)
fit = lm(rt ~ angle)
coef(fit)
1000 / unname(coef(fit)[2])
Python
from statistics import linear_regression
angle = [0, 50, 100, 150]
rt = [1801, 2415, 3026, 3191]
slope, intercept = linear_regression(angle, rt)
print(slope, intercept, 1000 / slope)
These calculations illustrate the descriptive fit, not a complete repeated-measures analysis. Retain participant and stimulus identifiers in the research dataset so the study analysis can use the intended level of observation.
Access, stimuli and licensing
The Ganis–Kievit Figshare repository supplies the stimulus set and validation data under CC BY 4.0, and asks users to cite the associated paper. Record the repository version and selected stimulus identifiers with a study. An open stimulus license and the right to run a particular software implementation are separate matters; Millisecond’s script carries its own copyright notice.
PsyToolkit provides demonstration code and a downloadable experiment package. Its copyright policy permits noncommercial research and educational use with acknowledgement, while third-party materials can have separate conditions. Commercial use requires permission under that policy. Identify the material actually reused, preserve its attribution and follow the applicable distribution terms. A historical test citation alone does not establish permission to reproduce its items.
Sources
- Shepard, R. N., and Metzler, J. (1971). Mental rotation of three-dimensional objects. Science, 171, 701–703. Checked 2026-11-14.
- Vandenberg, S. G., and Kuse, A. R. (1978). Mental rotations, a group test of three-dimensional spatial visualization. Perceptual and Motor Skills, 47, 599–604; full original paper. Checked 2026-11-14.
- Peters, M., et al. (1995). A redrawn Vandenberg and Kuse mental rotations test: Different versions and factors that affect performance. Brain and Cognition, 28, 39–58. Checked 2026-11-14.
- Ganis, G., and Kievit, R. A. (2015). A new set of three-dimensional shapes for investigating mental rotation processes: Validation data and stimulus set. Journal of Open Psychology Data, 3, e3. Checked 2026-11-14.
- Ganis and Kievit. Figshare stimuli and validation data; repository metadata and license. Checked 2026-11-14.
- Millisecond. Inquisit Mental Rotation Task technical manual. Checked 2026-11-14.
- PsyToolkit. Mental rotation task implementation and output documentation. Checked 2026-11-14.
- PsyToolkit. Copyright and use policy. Checked 2026-11-14.
Data: Mental rotation protocols: studies and software defaults
Original crosswalk of the Shepard–Metzler experiment, Ganis–Kievit validation study and Millisecond Inquisit defaults, compiled from their original paper and technical manual. Counts describe each named protocol, not interchangeable norms or recommended study sizes. Test trials exclude separately identified practice trials.
| Protocol | Angle levels (count) | Minimum angle (degrees) | Maximum angle (degrees) | Test trials per participant (trials) | Practice trials (trials) | Source |
|---|---|---|---|---|---|---|
| Shepard–Metzler original experiment (1971) | 10 | 0 | 180 | 1600 | Not enumerated in the paper | https://facultypsy.hope.edu/psychlabs/exp/rotate/readings/ShepardMetzler_1971.pdf |
| Ganis–Kievit validation study (2015) | 4 | 0 | 150 | 96 | 12 | https://openpsychologydata.metajnl.com/articles/10.5334/jopd.ai |
| Millisecond Inquisit v7 documented defaults | 4 | 0 | 150 | 96 | 16 | https://www.millisecond.com/library/v7/mentalrotation/mentalrotation/mentalrotation/mentalrotation.manual |
Sources
- Shepard and Metzler (1971), original experiment (accessed 2026-11-14)
- Ganis and Kievit (2015), validation study (accessed 2026-11-14)
- Millisecond, Inquisit v7 Mental Rotation Task technical manual (accessed 2026-11-14)
Compiled by CASRAI from the sources above. Free to reuse under CC BY 4.0 with attribution.







