GEE
Overall effect for a yes-or-no outcome that clusters in groups.
When to use it
Use GEE when the outcome is yes or no and the rows cluster, and you want the average effect across clusters. Pass or fail inside a clinic is the usual case. A random intercept you want to read as a variance is a GLMM.
Assumptions
The outcome is binary. The predictors are numeric. The grouping column has at least two clusters. The working correlation is exchangeable: two people in the same cluster are treated as equally correlated. There is no switch for the family or the correlation.
Running it in Tensr
The dialog reuses the mixed-model form, so it also shows random slopes and a REML switch, and it labels the dependent as numeric. This request sends the outcome, the predictors, and the group. The outcome has to be binary. Random slopes and REML are not part of this fit.
Options
Prop
Type
Reading the output
The clustered pass/fail outcome, population-average, exchangeable correlation within clinic.
| Term | Coefficient | SE | z | p | 95% CI low | 95% CI high |
|---|---|---|---|---|---|---|
| const | -1.48 | 0.768 | -1.926 | .054 | -2.986 | 0.026 |
| hours | 0.326 | 0.15 | 2.175 | .030 | 0.032 | 0.62 |
GEE binomial (exchangeable) for passed clustered by clinic. GEE binomial (exchangeable) for passed clustered by clinic. Metrics: Family = binomial; Correlation = exchangeable; Groups = 12; Observations = 120.
Reporting (APA 7)
GEE binomial (exchangeable) for passed clustered by clinic. Report the estimate in the table. This procedure is not summarised by one p-value.
Coming from SPSS
The product menu is Multivariate → Mixed Models → GEE (clustered binary). SPSS fits this with GENLIN, a binomial distribution, and a repeated subject with an exchangeable correlation. The path string stored for this item says Analyze → Generalized Linear Models → GEE. The item itself is on the Multivariate menu.
Related
A random intercept for the same kind of outcome is the GLMM. Independent rows: logistic regression.