MerQuR+Statistics Glossary

Statistics Glossary → Expected frequency

Expected frequencyconcept

The number of observations a table cell would be expected to contain if the null hypothesis were true; chi-square tests compare the observed counts with these values.

In a test of independence, if the row and column variables are unrelated, the expected count in a cell is E = (row total × column total) / n. In a goodness-of-fit test it is computed from the theoretical proportions as E = n × p₀. The Pearson chi-square Σ (O − E)²/E summarises the differences between observed (O) and expected counts. Cell-level residuals such as (O − E)/√E, or adjusted standardised residuals, show which cells drive the difference.

The chi-square approximation is reliable when the expected frequencies are large enough. The common rule (Cochran) is that no more than 20% of cells have E < 5 and no cell has E < 1. If this fails, Fisher's exact test, a Monte Carlo p value or a sensible merging of categories is considered. Note that the rule applies to expected, not observed, counts: a cell can have an observed count of 0 and still have a large expected value.

An everyday example

In a school, 120 of 200 pupils are girls and 80 are boys, and 50 pupils in total have joined a voluntary club. If sex and participation are unrelated, the expected number of girls in the club is 120 × 50 / 200 = 30. If 40 girls actually joined, this cell is 10 above expectation, and the chi-square test asks whether that gap can be explained by chance.

In MerQur

In the chi-square test of independence, the 'Expected counts' checkbox gives the table of E = row total × column total / n. Chi-square goodness of fit, the crosstab and the log-linear model also use expected counts, and multinomial logistic and probit regression compare observed and expected counts when assessing goodness of fit.

Analyses where this concept appears

Chi-Square IndependenceChi-Square Goodness of FitLog-Linear ModelCrosstabMultinomial Logistic RegressionProbit Regression

Related concepts

Standardized and adjusted residuals (contingency tables)Continuity correction (Yates)