Research Methods and Statistics
Design
Internal validity concerns causal inference; external validity concerns generalizability. Threats include history, maturation, testing, instrumentation, selection, attrition, and regression to the mean. Random assignment to conditions controls selection; random selection from a population supports generalizability-do not confuse them. Pre–post designs without a control group invite regression-to-the-mean and maturation artifacts. Single-subject ABAB reversal designs demonstrate experimental control. Cross-sectional comparisons are confounded by cohort effects.
Measurement and inference
Scales: nominal, ordinal, interval, ratio. Type I error is a false positive, governed by alpha; Type II error is a false negative, reduced by more power (larger N, larger effect, better reliability). Meta-analysis pools effect sizes. Statistical significance is not clinical significance; report effect size (Cohen's d benchmarks: 0.2 small, 0.5 medium, 0.8 large).
Relationships among variables
Correlation does not establish causation. A mediator explains how X affects Y; a moderator changes when or for whom the effect holds. In a 2×2 ANOVA, a main effect is the overall effect of one factor; an interaction means the effect of one factor depends on the level of the other.
Worked example
A clinic enrolls its most distressed clients, who improve, and claims the program works. Without a comparison group, the improvement is confounded with regression to the mean-some improvement is expected from extreme scores alone.
Ethics
IRB approval, informed consent, minimal risk, and deception only when justified and followed by debriefing.