Neuropsychology

Double Dissociation Detective

Six fictional case files, two tasks each, and one question: what does this pattern of scores actually establish?

Simulated — invented people, invented tasks, invented control data

Learning objective

By the end you should be able to tell a double dissociation from a single one, and say what each rules out.

About 30 minutes. Nothing you do here is saved or sent anywhere.

These are not patients

Every case file, score, task and control group here was written for teaching. No real person is described and no clinical instrument is used. Nothing on this page diagnoses, assesses or gives a prognosis for anybody.

Commit to an answer to the opening question.
Work through the six case files in order. For each one, read the scores against their own control group, open the task demands, and return a verdict.
Case 5 has a test-length slider. Move it before you commit.

This part needs JavaScript

The case files are drawn and scored in your browser, so they cannot run with JavaScript switched off. The rest of this page is readable without it.

Key terms

Single dissociation: One person doing badly on one task and normally on another.
Double dissociation: Two people showing opposite patterns to each other: each performs normally on the task the other is impaired on.
Control standard deviation: The unit the scores here are given in — how far below its own control group's average a score sits, measured in that group's own spread. Each task has its own control group.
Precision: How tightly a score pins down the thing it is estimating. Case 5's slider changes it by changing how many items the tasks had.

Case file

Which case

Test length

The percentage correct stays where it is; only the precision of the estimate changes. Watch the intervals in the chart.

Your verdict Not enough evidence to say A general severity difference A single dissociation A double dissociation A crossover, but the tasks differ in more than the process of interest

How far below the control average, in control standard deviations

The numbers

Raw score, fictional control-group average and standard deviation, control-referenced score and its 95% interval
Person	Task	Score	Controls	Distance from controls	95% interval

Case notes

What each task actually demands

What each task requires the person to see, hold in mind and do
Task	Input	Response	Also required

Two tasks never differ in only one thing. Whether the extra differences matter is a judgement about the case, not a number the chart can give you.

A fictional report describes two people. One is far below the control range on face recognition and within it on object naming; the other shows the opposite. Both tasks use the same input, the same response and the same number of items, and both differences are well outside measurement error. The report concludes: "face recognition is performed by one brain region and object naming by another."

What does the crossover actually support? The conclusion as written - two tasks, two dedicated regions That the two tasks draw on at least partly separable processing, with the anatomy still open Only that the two tasks are equally difficult Nothing - two people are never enough

A single dissociation has a boring rival

If one person is poor at task B and fine at task A, the obvious explanation is that A and B need different things. The boring rival is that B is simply more demanding, so any general reduction in capacity shows up there first. One person cannot tell those apart, however dramatic the gap.

A double dissociation rules that rival out

If a second person shows the opposite pattern, task difficulty can no longer be the explanation: a task cannot be the harder one for one person and the easier one for another purely by being harder. That is the whole logical work a double dissociation does, and it is real work. It is also the limit of what it does.

Compare each score with its own control group

Raw percentages are not comparable across tasks. A task on which healthy controls average 52% and a task on which they average 95% are different rulers. Case 4 on this page has a 58-point raw gap and no gap at all once both scores are referenced to their own controls. Whenever you see two percentages compared directly, ask what the control groups did.

Precision decides what you are allowed to claim

Case 5 changes verdict as the test gets longer without a single score changing. Short tests produce wide intervals, and a difference inside the interval is a difference you cannot claim. This is not a technicality: single-case dissociations are often built on tasks short enough that the difference would not survive being treated seriously as an estimate.

Two tasks always differ in more than one thing

Task impurity is unavoidable. Two tasks with different names differ in input, response, speed pressure, familiarity, strategy and how much they load on everything else at once. A crossover between two tasks isolates the process you had in mind only if nothing else that differs between them could produce the same crossover. Case 6 is a real crossover produced by two peripheral sensory differences, and nothing about it is a trick - it is what most real cases look like before the follow-up testing.

What this does not demonstrate

Not a localisation. Nothing here says where any damage is. Two separable processes can be distributed across the same tissue, and one region can serve both.
A double dissociation constrains, it does not prove. It rules out a single graded resource shared by the two tasks in the same proportions. It is still compatible with overlapping networks, with different strategies, with a shared component plus two separate ones, and with connectionist models that produce crossovers from damage to a single undivided system.
The numbers were written to make the verdicts come out. Control means, standard deviations and scores are not norms, prevalences or published values, so no percentage here should be quoted.
The statistics are simplified. The intervals use a binomial standard error and treat the control mean and standard deviation as known exactly. Proper single-case comparison uses a modified t test that allows for the control sample being finite, and gives wider intervals than these.
Two people are not a syndrome. Real dissociation evidence accumulates over many cases, converging methods and usually imaging. Individual variation in how any task is performed is large enough that a single pair proves nothing on its own.

Double Dissociation Detective

Learning objective

These are not patients

First, a judgement

The case files

Case 1