Abstract

Researchers can design a study to characterize a single instance of a phenomenon or to make an inference about a phenomenon in a population via a sample. Single-subject (or case) studies are justifiable when sampling is difficult or inappropriate. Psychosocial cases aimed at solving a specific problem usually require qualitative methods. Clinical cases are reports of diagnosis or treatment of injury or illness and are usually based on quantitative assessments and qualitative analysis. Non-clinical quantitative cases involve repeated sampling on a single subject and a quantitative inference about the subject generally. Sample-based designs are either observational or interventional, and most are aimed at quantifying a causal effect, in which changes in a predictor variable on average cause changes in a dependent variable. Establishing such causality in observational designs is problematic, owing to difficulties in adjusting for bias in the effect arising from confounders (variables that cause changes in the predictor and dependent). This problem is eliminated in interventions, but the necessary inclusion of a control treatment introduces bias mediated by differences between the groups in administration of treatments, compliance with study requirements, or imbalance in subject characteristics. Use of blinding and randomization at the design stage and inclusion of covariates in the analysis generally lead to trustworthy outcomes by reducing bias in interventions, but observational studies are sometimes the only ethically or logistically possible choice. In both types of study the role of a potential mechanism (or mediator) variable can be investigated by including it in the analysis as a covariate. The observational studies in approximate ascending order of the quality of evidence they provide for causality are case series, cross-sectional studies, case-control studies, and cohort studies. The corresponding approximate order for interventions is pre-post single group, post-only crossover, pre-post crossover, pre-post parallel groups, and post-only parallel groups. Methodological designs are also of interest to researchers; these are special kinds of cross-sectional study aimed at characterizing the validity, diagnostic accuracy, reliability or factor structure of a measure. Finally, reviews are another kind of cross-sectional study in which the “subjects” are study-estimates of an effect and in which the analyst estimates the effect of different settings on the outcome. Each design has particular strengths that offset its weaknesses and make it the most appropriate for a research question. Verf.-Referat