Research Methods and Statistics in Psychology-Law
This domain tests whether you can design, critique and cross-examine applied research, and whether you can explain statistical evidence to a court without committing a fallacious inference.
Design and validity
Forensic questions are often quasi-experimental because defendants cannot be randomly assigned. Address selection, history and instrumentation threats explicitly. Separate statistical conclusion validity from construct validity (does a student 'jury' index real jury deliberation?) and external validity (does a lab volunteer generalise to a representative venire?).
Probability logic
Sensitivity P(test positive given condition) is not the positive predictive value P(condition given test positive). PPV depends on prevalence: at low base rates even a specific instrument yields many false positives. The prosecutor's fallacy transposes P(evidence given innocence) with P(innocence given evidence); the defence fallacy transposes P(evidence given guilt) with P(guilt given evidence). Report effect sizes and confidence intervals (d, odds ratio, AUC); a p value is not the probability that a hypothesis is true.
Worked example
A risk tool has 90% sensitivity and 90% specificity and the offence base rate is 5% in 1,000 people. True positives = 0.90 x 50 = 45. False positives = 0.10 x 950 = 95. PPV = 45/140, about 32%. Over two-thirds of positive results are wrong, so testimony that a positive score means a 90% chance of offending is misleading.
Traps
Ignoring base rates; reading an AUC of .75 as 75% accurate; confusing inter-rater with test-retest reliability; and failing to correct for multiple comparisons when many predictor-outcome tests are run on one dataset.