Research designs
Research designs
Questions here ask which design supports a causal claim, how to judge a study, and which validity threat explains a flawed result. You need to distinguish design, sampling, and measurement.
Core designs
- True experiment: random assignment to conditions, manipulated independent variable, control group. It supports causal inference and high internal validity; lab settings may reduce external validity.
- Quasi-experiment: treatment occurs without random assignment, often using pretest–posttest nonequivalent groups, interrupted time series, or natural experiments. Causal claims are weaker because selection and history can confound.
- Correlational survey: measures variables as they naturally occur; r ranges from −1 to +1 and shows direction and strength, not causation.
- Case study: deep contextual description, useful for theory building and rare phenomena, low generalizability.
- Meta-analysis: statistically combines effect sizes across studies; addresses sampling error but can inherit publication bias and poor primary studies.
- Longitudinal design: repeated measures establish temporal order but suffer attrition and testing effects.
Validity
Internal validity threats include history, maturation, testing, instrumentation, regression to the mean, selection, and mortality. External validity concerns generalizing to people, settings, and times. Construct validity asks whether measures capture the intended concept; statistical conclusion validity concerns power and error.
Worked example
A firm reports r = .45 between engagement and performance and concludes engagement drives performance. That is unjustified: high performers may become engaged, or a third factor such as supportive leadership may drive both. A stronger test is a field experiment with staggered rollout and random assignment of teams, checking performance before and after; if not feasible, a longitudinal design with controls is better than cross-sectional correlation.
Traps
Correlation does not imply causation. Random selection (who is in the sample) is not random assignment (who gets the treatment). A lab study is not automatically invalid outside the lab; judge the specific claim. Beware pretest sensitization and convenience samples presented as representative.