Poisson regression
Predict a count, and read slopes as incidence rate ratios.
When to use it
Use Poisson regression when the outcome is a count: absences, errors, visits. The slope is on the log-count scale. The incidence rate ratio (IRR) says how many times larger the expected count is for a one-unit increase in the predictor.
Assumptions
Counts are non-negative. The mean and the variance of the count are similar. If the variance is much larger than the mean, a negative binomial regression is the usual next model. Tensr does not choose that for you.
Running it in Tensr
Options
Prop
Type
Reading the output
Absence counts generated as Poisson, with anxiety raising the mean a little. These counts are not the over-dispersed visit counts.
| Term | Coef | 95% CI | IRR | p-value |
|---|---|---|---|---|
| const | 0.025 | [-0.524, 0.575] | 1.026 | .928 |
| anxiety | 0.136 | [0.034, 0.237] | 1.146 | .009 |
Poisson regression of absences with 1 predictor(s). Poisson regression of absences with 1 predictor(s). Metrics: AIC = 333.11; Deviance = 110.082; Pearson χ² = 93.399; Deviance/df = 1.171; Observations = 96.
Reporting (APA 7)
Poisson regression of absences with 1 predictor(s). Report the estimate in the table. This procedure is not summarised by one p-value.
Coming from SPSS
Analyze → Regression → Poisson.
SPSS Generalized Linear Models is the usual menu for a Poisson log link. Tensr’s menu path is Analyze → Regression → Poisson.
Related
If the count variance is larger than the mean, see negative binomial regression.