.-
help for ^cic^                                
.-

Backward model selection using change in coefficient criterion
--------------------------------------------------------------

   ^cic^ [^if^ exp] [^in^ range] [^weight^=exp] [, model(string) modelopts(string)
      outcome(varlist) primary(varlist) alist(varlist) blist(varlist) 
      criterion(real 5) ^complete^ ^nomodels^ ^nodetail^ ]

Description
-----------

^cic^ automates backwards model selection using the change-in-coefficient criterion
for a specified primary predictor. The primary predictor and an A-list of covariates
are included in the model by default. Then ^cic^ cycles through a B-list of covariates 
eligible for removal. On each cycle, a model is fit omitting each of the remaining 
B-list covariates one at a time, and the absolute percent change in the coefficient
for the primary predictor is calculated. Then the covariate that induced the smallest
change is removed from the B-list, provided the change is less than specified maximum
change criterion.  Cycling continues until no remaining B-list variables induce changes 
smaller than the allowable maximum.

In contrast to more sophisticated built-in Stata stepwise commands, ^cic^ requires the
user to specify the model type, required model options, outcome, primary predictor, and
A- and B-list covariates as separate options.

Options
-------

^if^, ^in^, and ^weights^ are specified as usual. The specified weight must be allowed 
    by the regression command.

^model^(string) specifies the regression command to use (e.g., regress, logistic,
    stcox, xtlogit, xtmixed). ^This option must be specified.^  The data must already have 
    been stset or xtset as required by the specified command.  

^modelopts^(string) specifies any required model options (e.g. || id: time in a mixed
    model with random intercept and slope). This option need not be specified.

^outcome^(varlist) specifies the outcome variable. ^This option must be specified,^ 
    except for Cox and other survival models using stset data.

^primary^(varlist) specifies the primary predictor. Only continuous or binary predictors
    can be used. Binary predictors may be specified using the i. prefix; if so, they are 
    assumed to be coded 0 = no, 1 = yes. ^This option must be specified.^

^alist^(varlist) specifies the A-list covariates to be included in any model by default.
    Factor variables (e.g. i.age_category) are allowed. This option need not be specified 
    if all covariates are eligible for removal (i.e., on the B-list).

^blist^(varlist) specifies the B-list covariates eligible for removal from the model.
    Factor variables (e.g. i.age_category) are allowed. ^This option must be specified.^

^criterion^(real) specifies the maximum allowable change in coefficient, ^in percent.^
    The default value is 5%.

^complete^ specifies that only observations with complete data on all variables including
    the primary predictor, A-list, and B-list covariates be used in all models. This is
    the default in Stata stepwise commands. If this option is not specified, each model
    is run using observations with complete data on the outcome, primary predictor, A-list,
    and ^remaining^ B-list covariates.

^nomodels^ turns off printing of standard output for the initial and final models.

^nodetail^ turns off default printing of standard output for the initial and final models,  
    as well as the variable removed and percent changes in the coefficient for the primary 
    predictor within cycles. 

Examples
--------

 . cic [pweight=pwt], model(regress) outcome(bmd) primary(eeu) ///
	alist(age lweight i.estrogen calsupp diuretic etid momhip) ///
	blist(usearms i.tandstnd has10 gs10 gaitspd poorhlth caffeine drnkspwk smoker) ///
	complete nomodels

 . recode bmd min/.7=1 .7/max=0, gen(lowbmd)
 . cic, model(logistic) modelopts(vce(robust)) outcome(lowbmd) primary(eeu) ///
	alist(age lweight i.estrogen calsupp diuretic etid momhip) ///
	blist(usearms i.tandstnd has10 gs10 gaitspd poorhlth caffeine drnkspwk smoker) ///
	criterion(10) 

 . stset years, f(death) 
 . cic if age<30, model(stcox) primary(eeu) alist(age i.edu_cat) ///
	blist(usearms i.tandstnd has10 gs10 gaitspd poorhlth drnkspwk smoker) ///
	nodetail
	
 
Author
------
   Eric Vittinghoff, University of California San Francisco, San Francisco, CA, USA
   (eric@biostat.ucsf.edu)    [This is Version 1.0.3 (17Feb2013).]