Predictions
Statistical
models provide predictions that represent the expected value
of the outcome. For continuous
variables, the expected value is the mean; for dichotomous
or categorical variables the expected values are probabilities
of each outcome value. For example, predictions for males and
females of a pain score are simply the mean according to each
group; predictions of mortality for these groups are probabilities.
When continuous
or multiple predictors are considered, the predictions can no
longer be directly compared to empirical estimates. The number
of observations will be too low in cells formed by patients
with similar values of a continuous variable or by the combination
of multiple predictors. Residuals
may then be studied, i.e. the difference between observed and
predicted values.
For survival
data, predictions can be made in two ways.