The ICPHREG Procedure

Residuals and Diagnostic Statistics

The ICPHREG procedure computes two types of residuals. Residuals are available for the piecewise constant hazard model and the cubic splines model.

For the ith subject whose observed interval is $(L_ i, R_ i)$, the Lagakos residual is defined in Farrington (2000) as

\[ r_ i^ L = \frac{\hat{S}_ i(L_ i)\log [\hat{S}_ i(L_ i)]-\hat{S}_ i(R_ i)\log [\hat{S}_ i(R_ i)]}{\hat{S}_ i(L_ i)-\hat{S}_ i(R_ i)} \]

You can request Lagakos residuals in an output data set by using the keyword RESLAG in the OUTPUT statement.

The deviance residual is defined as

\[ r_ i^ D = \mr{sign}(r_ i^ L) \Big\{ 2 \log \Big[ \frac{\hat{S}_0(L_ i)^{\eta _ i}-\hat{S}_0(R_ i)^{\eta _ i}}{\hat{S}_0(L_ i)^{\exp (\bZ _ i' \hat{\bbeta })}-\hat{S}_0(R_ i)^{\exp (\bZ _ i' \hat{\bbeta })}} \Big]^{1/2} \Big\} \]

where

\[ \eta _ i = \frac{\log [\hat{\Lambda }_0(R_ i)]-\log [\hat{\Lambda }_0(L_ i)]}{\hat{\Lambda }_0(R_ i)-\hat{\Lambda }_0(L_ i)} \]

and where $\eta _ i=0$ if $R_ i=\infty $ and $\eta _ i=\infty $ if $L_ i=0$.

You can request deviance residuals in an output data set by using the keyword RESDEV in the OUTPUT statement.

The ICPHREG procedure computes the length of input intervals. For the ith subject whose observed interval is $(L_ i, R_ i)$, the length is defined as

\begin{eqnarray*} \mr{Len}_ i = \left\{ \begin{array}{ll} R_ i-L_ i & \mr{if} R_ i \ge L_ i > 0 \\ R_ i & \mr{if} R_ i > 0 \ \mr{and} \ L_ i=0 \ \mr{or} \ L_ i \ \mr{is \ missing} \\ -1 & \mr{otherwise} \end{array} \right. \end{eqnarray*}

You can request interval lengths in an output data set by using the keyword INTERVAL in the OUTPUT statement.