Treating an outcome variable as continuous data
If an outcome variable is numeric data that doesn't reflect categorical data, it is probably appropriate to treat it as continuous data. It can then be estimated with a linear model with optional weights, with an event-count model, or with a failure-time model.
To treat an outcome variable as continuous data:
- In the Model view, select the outcome variable you wish to treat as continuous data
- Choose Model > Outcome Variable and ensure that Treat as Categories is unchecked
By default, a linear model is used. There are other options available from the "Model" pop-up button, including:
- Log-linear, which models the natural logarithm of the outcome variable
- Poisson, which models the outcome as a count of independent events
- Negative binomial, which models the outcome as counts that exhibit overdispersion (defined as variance exceeding the mean)
- Geometric, which models the outcome as counts of trial successes before a failure
- Proportional hazards, which models the outcome as time elapsed until a failure event
If the outcome is a date column, then only two models will be available: linear, and proportional hazards. When a proportional hazards model is selected, the date is taken to be the failure time.
See also: