Part II · Chapter 5 · Lectures 4 & 5

Neuropsychology & dissociations

How studying damaged brains links a mental process to a brain region — why you need controls, why ceiling and floor effects hide deficits, and exactly what a double dissociation buys you that a single one can't.

1. The logic of a neuropsychological experiment

Cognitive neuroscientists bridge Marr's algorithmic and implementation levels: they link representations and processes to the tissue that runs them. You can't ethically switch off part of a healthy person's brain, but nature sometimes does it for you — stroke, tumour, injury.

Definition Neuropsychological experiment — a behavioral experiment with patients who have neurological disorders, used to relate brain and behavior: how is behavior disrupted when a particular part of the brain is damaged? How is a cognitive process linked to a particular area?

2. Round 1: no control group

Hypothesis: the right hemisphere is essential for music perception. Participants: 10 patients with right-hemisphere (RH) damage. Task: recognize songs.

Music	RH patients
Folk songs	25%
TV jingles	98%
Pop hits	3%

Is the hypothesis supported? You can't tell. 25% on folk songs might be how everyone does. We don't know the baseline performance of people without RH damage. So: add 10 matched controls.

Music	RH patients	Matched controls	Verdict
Folk songs	25%	90%	Impaired — big gap
TV jingles	98%	99%	Ceiling — everyone near max
Pop hits	3%	5%	Floor — everyone near min

Lecture 4's numbers. Lecture 5 reshows the table with slightly different values (TV 97/98, pop 5/7); the conclusion is identical.

Ceiling effectPerformance so high (task too easy) that differences can't be detected — everyone is near the maximum.

Floor effectPerformance so low (task too hard, near chance) that differences can't be detected — everyone is near the minimum.

Drag the difficulty slider. The patient group really is impaired in this simulation, by the same amount at every setting — watch when you can and can't see it:

A real deficit hidden by ceiling and floor

Task difficulty

What to notice: patient ability is lower than control ability by a fixed amount, but the gap in percent correct collapses at both ends of the slider. A null result on an easy (TV jingles) or impossible (pop hits) task says nothing about whether a deficit exists — the test lacked sensitivity. Toy logistic model.

The lecture's two conclusions from round 1:

Unclear support: impaired on folk songs, but the jingles and pop results are inconclusive (no sensitivity), so we can't claim a general music deficit.
Limited explanatory depth: even if true, it links "music" to a region without saying which mental process is involved.

3. Round 2: add a cognitive model → single dissociation

Fix the second problem with a model. Music perception = two processes:

Rhythm recognition — extracting the timing and beat.
Melody recognition — extracting the sequence and relationship of pitches.

New hypothesis: the right hemisphere is essential for melody recognition. Rhythm becomes the control task. Which result supports it?

RH patients vs controls. Rhythm (control task) / Melody (experimental task):

Why: Melody impaired while rhythm is spared points specifically at melody. (4) looks like support but can't be told apart from general brain damage — the patients are just worse at everything. (3) supports the opposite claim. (1) means no deficit.

Definition Single dissociation — a brain lesion impairs one cognitive process while sparing another, suggesting the two are at least partly distinct. Uses one patient group.

Three weaknesses of a single dissociation

Weakness	What it means here
1. Task sensitivity	Maybe the melody task is just harder — more sensitive to any damage — than the rhythm task.
2. General impairment	Maybe poor melody performance is a general consequence of brain damage, not damage to a specific process.
3. Model ambiguity	Can't tell whether the two processes are serial (one after the other) or parallel (independent).

Serial vs parallel — why one patient group can't decide

Two models of music perception

What to notice: in the serial model, melody sits downstream of rhythm. Break melody → only melody fails; that's what both models predict. Break rhythm → in the serial model melody loses its input and fails too. So the telling case is the one the RH group didn't show: rhythm impaired, melody spared.

The step the slides skip Why does a double dissociation "support the parallel model"? Because the second group (rhythm broken, melody fine) is impossible under the serial model — if melody needed rhythm's output, it couldn't survive rhythm's loss. The RH result alone (melody broken, rhythm fine) is compatible with both models, which is the "model ambiguity" weakness.

4. Round 3: double dissociation

Add a group impaired in the other process: 10 RH patients, 10 left-hemisphere (LH) patients, 10 matched controls.

	Rhythm	Melody
Controls	Baseline	Baseline
RH group	Normal	Impaired
LH group	Impaired	Normal

Definition Double dissociation — two patient groups show opposite patterns: damage to area A impairs process 1 but spares 2; damage to area B impairs process 2 but spares 1.

Single-dissociation weakness	How the double dissociation fixes it
Task sensitivity	Shows both tasks can detect impairment — rhythm isn't just "too easy".
General impairment	Rules out a generic effect of brain damage — each group is fine on one task.
Model ambiguity	Helps distinguish serial vs parallel — supports separate, parallel processes.

Dissociation lab — set the results, get the verdict

Group	Rhythm task	Melody task
RH patients
LH patients

Click a cell to cycle Normal → Impaired → not tested (—). What to notice: the verdict only reaches "double dissociation" when the two groups are mirror images.

One-liners to memorize Single: lesion impairs X but spares Y.
Double: lesion A impairs X not Y; lesion B impairs Y not X.

Double dissociations you'll meet in this course

Process 1	Process 2	Patients	Chapter
Language comprehension	Language production	Wernicke's aphasia (comprehension ✗) vs Broca's aphasia (production ✗)	6
Face recognition	Object recognition	Prosopagnosia (faces ✗) vs object agnosia, patient CK (objects ✗)	8
Melody	Rhythm	RH vs LH damage	5

5. Check yourself

Patients with frontal damage score 99% on a memory test; controls score 100%. The researcher concludes frontal cortex isn't involved in memory. The biggest problem is…

Why: Everyone is near maximum, so a real deficit couldn't show up. There is a control group. Floor would be everyone near zero.

Group A is impaired on both task X and task Y. What can you conclude about X and Y?

Why: No dissociation at all — no task was spared. Could be a shared process, a serial dependency, or just general brain damage.

Under a serial model Rhythm → Melody → Music, which patient pattern should be impossible?

Why: Melody depends on rhythm's output in the serial model; lose rhythm and melody should fail too. Finding this pattern (the LH group) is what supports the parallel model.

Try itDesign a double dissociation for faces vs objects

A patient with a lesion near the fusiform gyrus can't recognize his wife's face but names the shrine behind a sumo wrestler instantly. Someone argues: "faces are just harder objects; any damage hurts them first." What evidence would defeat that argument?

Step 1 — Which weakness is the critic using?

Task sensitivity: the face task might just be harder.

Step 2 — What would defeat it?

A patient with the opposite pattern: object recognition impaired, face recognition intact. If faces were just harder objects, anyone bad at objects should be even worse at faces.

Answer & the takeaway

Patient CK (object agnosia) is exactly that — Chapter 8. Prosopagnosia + object agnosia = double dissociation → faces and objects are distinct processes. The pattern you're practicing: when someone says "that task is just harder", find the mirror-image patient.