Research and Program Evaluation
Research items test whether you can read a study correctly: design, threats to validity, statistics reported, and ethical safeguards. Program evaluation items ask which evaluation type fits a stated question.
Designs
Experimental designs use random assignment and manipulation, supporting causal claims. Quasi-experimental designs lack random assignment. Correlational and ex post facto designs describe association only. Single-subject designs (ABAB reversal, multiple baseline) establish control within one case. Qualitative approaches - phenomenology, grounded theory, ethnography, case study - generate meaning rather than test hypotheses.
Validity
Internal validity threats: history, maturation, testing effects, instrumentation, regression to the mean, selection bias, and attrition. External validity concerns generalization across people, settings, and time.
Statistics
Descriptive statistics summarize (mean, median, mode, range, standard deviation); the normal curve places roughly 68%/95%/99.7% of scores within 1/2/3 SDs. Inferential tests (t-test, ANOVA, chi-square, regression) estimate whether findings exceed chance. Alpha sets Type I error (false positive); beta relates to Type II error (false negative). A p-value is the probability of data this extreme if the null hypothesis is true - not the probability the null is true. Effect size (Cohen's d: .2 small, .5 medium, .8 large) reports magnitude independent of sample size.
Program evaluation and ethics
Needs assessment, formative (process) evaluation, and summative (outcome) evaluation answer different questions. IRB approval, informed consent, confidentiality, and debriefing after deception are required.
Worked example
Clients who attended more sessions also improved more. The reasoning: this is a correlation; more motivated or less impaired clients may attend more and improve more, so causation is unsupported without random assignment and a control group.
Traps that cost marks
Reading causation into correlations, misstating p-values, confusing Type I with Type II error, ignoring effect size and attrition, and assuming a large sample fixes a biased design.