Watch it fade.
Bend a forgetting curve with spaced retrieval, compare Asch's and Milgram's published variations side by side, then drag a correlation and watch causation refuse to follow.
The curve, and how testing bends it
FIG. 01 · 4.2 · MODEL, NOT DATAPick a revision schedule on the right. Each retrieval pulls the curve back to the top — and makes the next fall slower.
Honest label: this is a model, not a dataset. It uses the exponential shape Ebbinghaus (1885) reported for his own nonsense-syllable recall, with a stability parameter you control. No published study produced these exact numbers.
Same number of sessions, different spacing. Watch what happens to day 30.
The stability of the first trace. Deep processing and good cues start you higher up this dial.
Retrieving a memory is not neutral — pulling it out changes it. That is the testing effect.
What the exam wants. AQA 4.2 does not examine Ebbinghaus — it examines why the fall happens: interference (proactive and retroactive) and retrieval failure through absent cues. Godden and Baddeley's 1975 divers are the standard evidence for the second: recall was around 40% worse when the learning and recall environments did not match. This widget is the mechanism made visible, and it is also a 4.7.3 graph-reading exercise — read the axes before you read the shape.
The variations, as published
FIG. 02 · 4.1 · ASCH · MILGRAMChoose a variation. Nothing here is modelled or interpolated — each bar is the figure that study actually reported.
Asch, S. E. (1951; 1956). Milgram, S. (1963); Milgram, S. (1974) Obedience to Authority.
What the exam wants. 4.1 asks you to explain conformity and obedience through situational variables — group size, unanimity, task difficulty for Asch; proximity, location and uniform for Milgram — and evaluation marks go to candidates who quote the actual figures rather than saying “a lot of people conformed”. The single most useful pair here: one dissenter cuts Asch's 36.8% to 5.5%, and moving the experimenter to a telephone cuts Milgram's 65% to 20.5%. Both are evidence that the situation, not the person, is doing most of the work.
Drag r. Causation stays put.
FIG. 03 · 4.7.1 · 4.7.4 · SIMULATED DATASixty simulated participants. Slide r and watch a convincing-looking relationship appear out of nothing at all.
−1 is a perfect negative correlation, 0 is none, +1 is perfect positive. Real psychological findings rarely pass 0.6.
What the exam wants. 4.7.1 asks for the difference between a correlation and an experiment, and the mark is for saying that no variable was manipulated — so the third-variable problem and the direction-of-causality problem both stay live. 4.7.3 wants r² read as shared variance: r = 0.5 means 25% of the variation is accounted for, so three-quarters is still unexplained. And when the data are ordinal and you are testing a correlation, the test is Spearman's rho.