Tetrachoric Correlation Calculator

The tetrachoric correlation estimates the Pearson correlation between two latent (underlying) continuous variables that have each been dichotomized into binary (0/1) outcomes. It is commonly used in psychometrics, epidemiology, and genetics when you observe two dichotomous variables but assume they reflect underlying continuous traits.

Use tetrachoric correlation when both observed variables are dichotomous but you believe they represent continuous underlying constructs — for example, pass/fail scores on two tests, or presence/absence of two symptoms. The phi coefficient is appropriate when the dichotomy is truly categorical, while tetrachoric is better when the dichotomy is artificial (e.g., a threshold on a continuous scale).

Cell A is the count where both variables equal 1, Cell B is where Variable 1 = 1 and Variable 2 = 0, Cell C is where Variable 1 = 0 and Variable 2 = 1, and Cell D is where both variables equal 0. Together, A + B + C + D equals your total sample size N.

The calculator uses the cosine-pi approximation: r_tet ≈ cos(π / (1 + √(AD/BC))). For more precision, iterative numerical methods based on the bivariate normal distribution are also used. This tool applies the widely accepted approximation formula, which is accurate for most practical use cases.

The standard error (SE) quantifies the uncertainty in your r_tet estimate due to sample size. Larger samples yield smaller SEs. It is used to compute the confidence interval and assess how precisely the true tetrachoric correlation is estimated from your data.

The p-value comes from a chi-square test of independence applied to the 2×2 table. A p-value below your chosen alpha level (commonly 0.05) indicates that the two binary variables are not independent — supporting the conclusion that a genuine correlation exists between the underlying continuous traits.

A minimum of 50–100 observations is generally recommended, but larger samples (200+) produce more stable estimates. Very small cell counts (below 5 in any cell) can lead to unstable or inaccurate results, so verify that all four cells have sufficient counts before interpreting r_tet.

Yes — tetrachoric correlations are commonly used to build polychoric correlation matrices for factor analysis when items are dichotomous, such as in educational testing or psychological questionnaires. Software like R's 'psych' package or LISREL uses these matrices to perform factor analysis appropriate for binary item data.