Quick Reference Sheet for PS_FRAIL Macro
- by Youyi Shu -
****************************************************************************
Note: The official documentation of the macro is the following paper:
Shu, Y. and Klein, J.P. (1999). A SAS Macro for the Positive
Stable Frailty Model. American Statistical Association
Proceedings of the Statistical Computing Section, p.47-52.
****************************************************************************
I. Data set in SAS Data Step
(A) Data set must have the following ordering of its variables:
Group, Time, Censoring Indicator, Covariates.
(B) The censoring indicator needs to be specified as follows:
0-censored; 1-event.
II. Macro Loading
PS_FRAIL macro is loaded in the program via the "%INCLUDE 'filename'"
statement where 'filename' is the name of the PS_FRAIL macro file.
III. Macro Invocation
PS_FRAIL macro is invoked in the form of
%PS_FRAIL(dsn <,option 1 <...,option n>>)
where dsn is the data set name and it must be given, the symbol of angle
brackets identifies optional arguments. There are at most 13 options
available and they can be given in any order after the data set name. Each
option is specified as a keyword followed by an equal sign and a value
(not case-sensitive). The options are:
(1) GREPORT=Y: Prints the summary report of group sizes and the number
of events per group.
(2) ITPRINT=Y: Prints the iteration history and the main process of the
analysis.
(3) ITSUMM=Y: Prints the summary table of the iteration history.
(4) ALPHA=value: Confidence level option. To construct 100(1-alpha)%
confidence limits for the relative risks, the value is set to alpha. This
option has no effect unless the RL=Y option (see below) is specified. The
default value is 0.05, which results in the calculation of a 95% confidence
limits.
(5) RL=Y: Prints, for each explanatory variable, the 100(1-alpha)% confi-
dence limits for the relative risk. Here, alpha is determined by the ALPHA=
value option.
(6) COV=Y: Prints the estimated variance-covariance matrix of THETAHAT
(the estimate of the dependence parameter) and BETAHAT (the estimates of
the regression coefficients).
(7) OUTEST=value: Output data set option. If the value is specified as a
legal SAS data set name, then that data set will contain THETAHAT and
BETAHAT, the estimated variance-covariance matrix of THETAHAT and BETAHAT,
and the last computed value of the log full likelihood.
(8) BASELINE=value: Output data set option. If the value is specified as
a legal SAS data set name, then that data set will contain ordered event
times (_TIME_), baseline hazard rates (_BHR_) , standard errors of the
baseline hazard rates (_SEBHR_), and the estimate of the dependence parameter
(_THETA_).
(9) OUTPUT=value: Output data set option. If the value is specified as a
legal SAS data set name, then that data set will contain the grouping
variable, survival time, censoring status, covariates, estimated linear
predictor (_XBETA_), cumulative baseline hazard rate (_CBHR_), and the
estimate of the dependence parameter (_THETA_).
(10) CONVBETA=value: Option to set the convergence criterion based on the
estimates of the regression coefficients. The iteration sequence stops when
the maximum change among the estimates of the regression coefficients is
less than the specified value for this option. Default value: 0.001.
(11) LAMBDA=value: Option to set the blending constant used with
Marquardt's method to find the roots of the score equations in a modified
Cox profile likelihood analysis. Note that small value corresponds to
Newton-Raphson method while large value corresponds to the steepest ascent
method. Default value: 0.25.
(12) GRIDSIZE=value: Option to set the step size in the grid search
method. The macro starts from one and steps down by value at each step to
bracket the maximum likelihood estimate of the dependence parameter. Default
value: 0.05.
(13) GOLDLENG=value: Option to set the stopping criterion for the golden
section search algorithm. After initially bracketing a maximum through a
grid search method (see GRIDSIZE=value option above), the macro then starts
the golden section search to search for the maximum likelihood estimate of
the dependence parameter. The search routine is continued until the length
of the bracketing interval is less than the specified value for this option.
Default value: 0.001.