Statistical Tests (t-test, ANOVA, Chi-Square, Correlation)

USMLE Step 1 trap: Applies chi-square to continuous outcomes instead of t-test or ANOVA. Chi-square tests are for categorical (nominal) data; continuous data comparing two groups requires a t-test, and more than two groups requires ANOVA.

Statistical tests are the decision-making tools of biostatistics — they tell you whether the difference you see between groups is real or just noise. USMLE Step 1 doesn't ask you to calculate these tests. It asks you to pick the right one given a study setup, recognize when an assumption is violated, and interpret what the output actually means. The three angles that show up repeatedly are: matching the test to the data type and number of groups, knowing when to switch from a parametric to a non-parametric test, and distinguishing what linear versus logistic regression actually produce.

The core skill is a two-step lookup: first, is the outcome categorical or continuous? Second, how many groups are being compared? Continuous outcome, two groups → t-test. Continuous outcome, three or more groups → ANOVA. Categorical outcome → chi-square (or Fisher's exact when sample sizes are small). Students get into trouble because they pick chi-square out of habit whenever they see a comparison, even when the outcome is something like blood pressure or serum creatinine — both of which are continuous and need a t-test or ANOVA.

The parametric vs. non-parametric distinction trips up students who think 'more powerful = always better.' Parametric tests (t-test, ANOVA, Pearson correlation) assume the data are normally distributed. When that assumption breaks down — small samples, skewed distributions — you swap to the non-parametric equivalent (Mann-Whitney U instead of t-test, Kruskal-Wallis instead of ANOVA, Spearman instead of Pearson). USMLE Step 1 loves giving you a vignette with a small, skewed sample and asking which test is appropriate. Defaulting to the parametric version without reading for normality is a reliable way to lose that point.

From real student decks

51%have cards covering this topic

35%have mature cards

Common misconceptions

Common mistake

Wrong: Students apply chi-square tests to continuous outcome data comparing two groups.

Right: Chi-square tests are for categorical (nominal) data; continuous data comparing two groups requires a t-test, and more than two groups requires ANOVA.

Chi-square tests assess whether two categorical variables are associated — they work on counts and proportions, not means. If the outcome is a measured number (blood pressure, weight, enzyme level), you need to compare means, which requires a t-test for two groups or ANOVA for three or more. Applying chi-square to continuous data is a category error, not just a minor mistake — the math literally doesn't apply to that data structure.

Common mistake

Wrong: Parametric tests are always preferred because they are more powerful.

Right: Parametric tests assume normally distributed data; when this assumption is violated (small samples, skewed data), non-parametric equivalents should be used despite lower power.

Parametric tests are more powerful, but that power is only valid when the data actually fit a normal distribution. When you have a small sample size or visibly skewed data, the normality assumption breaks down, and using a parametric test produces unreliable p-values. The right move is to drop to the non-parametric equivalent — you sacrifice some power, but you get a result that's actually trustworthy for the data you have.

Common mistake

Wrong: Logistic regression produces a mean difference as its output, like linear regression.

Right: Linear regression outputs a continuous predicted value (slope/coefficient); logistic regression outputs an odds ratio for a binary outcome.

Linear regression models a continuous outcome and outputs a coefficient that tells you how much the outcome changes per unit change in the predictor — it's a slope on a number line. Logistic regression models a binary outcome (disease yes/no, death yes/no) and outputs an odds ratio, which is a ratio of probabilities, not a mean difference. Confusing these two means you'd misinterpret the entire result of a regression study on the exam.

Free Deck audit

See if your Anki deck covers this topic.

Upload your deck →

Guided session

Stuck on this? An AI tutor that probes your understanding.

Start a session →

What the exam tests

Given a study outcome type (categorical vs. continuous) and number of comparison groups, select the correct statistical test from: t-test, paired t-test, ANOVA, chi-square, or Fisher's exact test.
Recognize when the assumptions for a parametric test (normality) are violated and identify the appropriate non-parametric substitute (e.g., Mann-Whitney U, Kruskal-Wallis, Spearman correlation).
Distinguish between linear and logistic regression by their input requirements and outputs — linear regression predicts a continuous value and reports a coefficient/slope; logistic regression handles a binary outcome and reports an odds ratio.

Can you avoid these mistakes?

A researcher measures systolic blood pressure in three diet groups (low-carb, low-fat, Mediterranean). Which statistical test should be used to compare the means across all three groups?

A study compares rates of surgical site infection (infected vs. not infected) between two hospitals. The expected cell count in one group is only 3. Which test is most appropriate — chi-square or Fisher's exact — and why?

A dataset of 15 patients shows a heavily right-skewed distribution of hospital length of stay. A colleague wants to use a t-test to compare two treatment groups. What is wrong with this approach, and what test should be used instead?

A logistic regression study reports an output of 2.4 for the association between smoking and developing lung cancer. What does this number represent, and how does it differ from what a linear regression would have reported?

Statistical Tests (t-test, ANOVA, Chi-Square, Correlation)

Common misconceptions

What the exam tests

Can you avoid these mistakes?

Related topics