Skip to content
Omnikalc

Tetrachoric correlation calculator

Enter the four counts of a 2 by 2 table made by cutting two normal variables to get the approximate tetrachoric correlation, the phi coefficient, the odds ratio and the cut points of each variable.

Updated on
Your inputs

Value out of range

Value out of range

Value out of range

Value out of range

Results

Approximate tetrachoric correlation

0.707

Phi coefficient
0.492
Odds ratio
9.00
Cut point of the row variable (in standard deviations)
0.253
Cut point of the column variable (in standard deviations)
0.126
Share in row 1
40.0 %
Share in column 1
45.0 %

The tetrachoric correlation calculator gives the approximate correlation of two latent normal variables from the four counts of a 2 by 2 table, with the phi coefficient, the odds ratio and the cut point of each variable in standard deviations.

How the calculation works

  • Enter the counts a, b, c and d of the 2 by 2 table.
  • The odds ratio is computed from the cross products ad and bc.
  • Read the tetrachoric correlation, phi and the cut points.

Example

A table with 30, 10, 15 and 45 has an odds ratio of 9.00 and a phi coefficient of 0.492. The approximate tetrachoric correlation is 0.707. Row 1 holds 40.0 % of the cases, a cut point of 0.253, and column 1 holds 45.0 %, a cut point of 0.126.

Why it matters

The calculator helps psychometricians, students and researchers who analyze pairs of binary items, diagnoses or survey answers and want the correlation of the underlying traits, for example before a factor analysis of items scored right or wrong.

Good to know

Check that a normal latent trait behind each answer is a reasonable assumption. The cosine formula is an approximation, and the exact value requires software that solves the bivariate normal integral. The estimate has a larger standard error than a Pearson correlation of the same sample size. With small samples or a zero cell, add 0.5 to each cell, and avoid reading the result as precise. Report the phi coefficient as well, state that the tetrachoric value rests on the normality assumption, and compare it with the correlation you would get if you could measure the underlying variables directly, as the two should be close when the model holds.

The Yule Q calculator gives another association measure from the same table, the Cramér V calculator handles larger tables, the point-biserial calculator treats a binary and a continuous variable, and the Matthews calculator scores a binary classifier.

Frequently asked questions

What is the tetrachoric correlation?

The tetrachoric correlation estimates the Pearson correlation between two continuous normal variables when each has been observed only as a yes or no, above or below a cut point. It is used with items scored right or wrong, in diagnosis with a threshold, or in surveys, when a latent trait is assumed behind the answers.

What is the formula used here?

The exact value needs an iterative solution of the bivariate normal. A good and simple approximation is the cosine formula, r = cos(π / (1 + √OR)), where OR = ad / bc is the odds ratio. It is exact when the table has equal margins and is close in other cases, within a few hundredths for moderate margins.

How does it compare with phi?

The phi coefficient is smaller in absolute value than the tetrachoric correlation, because dichotomizing a continuous variable loses information and attenuates the correlation. The loss is greater when the cut points are far from the median. The tetrachoric value tries to undo this attenuation.

When should I not use it?

The method assumes that both latent variables are bivariate normal, which is a strong hypothesis. If one of the variables is truly binary, such as sex, the phi coefficient or the point-biserial correlation is more suitable. Small counts or empty cells make the estimate unstable, and the result can go beyond the range when the approximation fails.

Was this calculator helpful?

Thanks for your feedback!

Results are estimates provided for information purposes. They do not replace professional advice.