Methods · Guide
Which statistical test should I use?
The right statistical test depends on three things: what your outcome is (continuous, categorical or time to an event), how many groups or variables you compare, and whether the same cases are measured more than once. Match your situation to a row below; the second column is the usual choice and the third is the alternative when its assumptions do not hold.
어떤 검정을 쓸지는 결과변수의 종류(연속형·범주형·사건까지의 시간), 비교하는 집단·변수의 수, 같은 대상을 여러 번 측정했는지로 정해집니다.
Find your situation
| Your situation | Usual choice | If assumptions fail | Note |
|---|---|---|---|
| Compare one continuous outcome between two independent groups | Independent Samples t test | Mann-Whitney U Test | Normal-ish outcome and similar variances; otherwise use Welch's version or the rank-based test |
| Compare the same people measured twice (before/after) | Paired Samples t test | Wilcoxon Signed Rank Test | Differences roughly normal for the t-test |
| Compare three or more independent groups | One way ANOVA | Kruskal-Wallis Test | Follow a significant ANOVA with a post hoc test such as Tukey HSD |
| Compare groups across repeated measurements | Repeated Measures ANOVA | Friedman Test | Check sphericity; mixed models handle missing waves better |
| Test whether two categorical variables are related | Chi Square Test | Fisher's Exact Test | Use Fisher's exact test when expected counts are small |
| Measure the linear association between two continuous variables | Pearson Correlation Analysis | Spearman Rank Correlation | Spearman for ranks, outliers or monotonic but non-linear relationships |
| Predict a continuous outcome from several predictors | Multiple Linear Regression | Regularised Regression (Ridge, Lasso, ElasticNet) | Check residuals; regularize when predictors are many or collinear |
| Predict a yes/no outcome | Binary Logistic Regression | Random Forest | Logistic regression for interpretable odds ratios; tree ensembles for prediction |
| Model time until an event when some cases have not had it yet | Kaplan-Meier Survival Analysis | Cox Proportional Hazards Model | Kaplan–Meier to describe and compare curves; Cox to adjust for covariates |
| Check that questionnaire items measure the constructs you claim | Confirmatory Factor Analysis (CFA) | Cronbach's Alpha Reliability Analysis | CFA tests the structure; alpha only summarizes internal consistency |
| Estimate the effect of a policy or intervention with before/after data for treated and control groups | Difference in Differences (DiD) | Synthetic Control Method (SCM) | Needs parallel pre-treatment trends |
| Forecast a time series | ARIMA Model | Exponential Smoothing (Holt-Winters) | Check seasonality and stationarity first |
Still not sure?
Upload the data for a free pre-analysis diagnosis: Quorant reads the variable types, checks missing values and assumptions, and suggests the methods that fit, with an exact quote. Or ask the methods advisor in plain words.