Research Method Advisor

What problem it solves

In psychology or social-science analysis, the blocking point is often not “can I run the code?” but “which test should I use first: a t-test, ANOVA, chi-square, correlation, or regression?” Choosing the wrong method can violate assumptions, overstate the claim, or leave the follow-up analysis without support. The tool turns statistical-method selection into a set of inspectable decisions, using research goal, dependent-variable type, group count, design, number of factors, and assumption status to propose a suitable method plus assumptions and alternatives.

How it works

Typical inputs include goal (compare/relationship/predict), dv_type, n_groups, design, n_factors, iv_type, and assumptions_met. It maps cases to methods across three task families:

• Comparison: t-test, Wilcoxon, Mann-Whitney U, Kruskal-Wallis, Friedman, ANOVA, chi-square, or Fisher exact test
• Relationship: Pearson / Spearman correlation and chi-square independence
• Prediction: linear regression, logistic regression, and ordinal logistic regression

Each recommendation includes applicability, assumptions, parametric/nonparametric alternatives, follow-up analysis, and entry points for Python packages such as scipy, statsmodels, and pingouin. For scale research, it also points to Cronbach's α, CR, EFA/CFA, AVE, and HTMT checks; for mediation and moderation, it distinguishes Bootstrap, PROCESS, and hierarchical-regression routes.

Boundaries and cautions

It is best used for pre-experiment method planning, pre-analysis test checks, or finding nonparametric alternatives when assumptions fail. It does not produce final conclusions and cannot replace judgment in the concrete research context: parametric tests still require checks for normality, homoscedasticity, independence, and residuals; multiple comparisons need Bonferroni, Tukey, or similar corrections; reported results should include effect sizes, confidence intervals, and model diagnostics, not only p values.