Q1.02.2Confirmatory hypothesis testingresearch

Confirmatory research tests prespecified hypotheses

Aliases: confirmatory research · preregistered study · hypothesis testing

What it is

Confirmatory hypothesis testing evaluates whether new evidence is compatible with a claim whose hypothesis, measures, sample, and analysis rules were specified in advance. The claim may concern causal effects, associations, group differences, or qualitative theoretical predictions. Confirmatory describes how evidence confronts a bounded claim; it does not guarantee confirmation and is not synonymous with null-hypothesis significance testing.

Why it happens

When outcomes, subgroups, or models are selected after results are visible, random variation participates in that selection while ordinary uncertainty estimates ignore it. Restricting the analysis space in advance, or testing in independent data, restores interpretable error and disconfirmation. The key property is that evidence inconsistent with the prediction can survive the workflow, not that the method makes the desired answer appear.

Studying it

Specify direction, population, treatment or comparison, primary outcome, exclusions, stopping rule, and analysis model. Quantitative studies plan precision or power around a smallest effect of interest. Qualitative confirmation can state the situations, processes, and counterexamples predicted by a theory and use purposive sampling to test whether the account covers new cases. Deviations do not automatically make the entire analysis exploratory: prespecified adaptive rules remain part of the plan, and timestamped revisions made before outcome inspection for result-independent reasons usually preserve more confirmatory force than outcome-informed selection of measures, cases, or models. Disclose every deviation and assess its consequence from what changed, when it changed, what result information was available, and how much researcher choice it introduced.

Where it stops holding

Preregistration cannot repair a weak construct, biased measure, or unrepresentative sample, and it cannot turn association into causation. Exact replication tests an operation without necessarily establishing a theory across tasks and populations. A nonsignificant result from an imprecise study is not evidence of absence; effect size, uncertainty, data quality, and external validity remain necessary.

Related

  • Same group: Q1.02.1 Exploratory research generates hypotheses · Q1.02.3 Reusing exploratory data creates circular confirmation
  • Adjacent: Q3.18 Experimental design and control · Q3.21 Statistical significance and effect size
  • Search terms: confirmatory research · preregistration · hypothesis testing

Cards in the same group

Quick Actions

Share

Share this page

ios_share

https://hci.top/en/handbook/Q1.02.2