****************
****************
** lincom     **  estimates linear combinations of most recently fit model
****************
****************

Overview:  The logistic or logit command estimates a logistic regression for 
           a binary outcome variable. Lincom can then be used to estimate
           the expected value and variance of any linear combination of the 
           parameters.


Usage:     "logistic Dvar X1 X2 X3"
             Where - Dvar is the disease variable (1=disease, 0=control)
                     Xj are predictor variables 

           "lincom exp"
            where exp is a linear combination of the coefficients. Use the names 
            associated with the independent variable for each coefficient in the 
            expression e.g. 100*ALC + 40*TOB

Summaries:  lincom will provide the estimated value of the linear combination on
            either the log or original scales (see options). It also provides a
            95% CI. The standard error may not be too useful if the result is
            given on the original scale instead of the log scale. All CI's are
            computed on the log scale and transformed, if necessary.

Options:    Useful options are level(#) to specify the confidence level and or
            (equivalent to rrr) to specify that the estimate be given on the
            odds ratio scale instead of the linear predictor (log) scale. or is
            the default if the prior command was logistic.


****************
****************
** predict    **  predictions / fits & logistic regression
****************
****************

  Overview:  The "logistic" command estimates a logistic regression for 
             a binary outcome variable.  To obtain summaries of the
             fitted probabilities, or the fitted log odds ("X beta")
             we can use the predict command.


  Usage:     "logistic Dvar X1 X2 X3"
             Where - Dvar is the disease variable (1=disease, 0=control)
                     Xj are predictor variables 

             "predict pfit" -- gives fitted P( D | X1, X2, X3 )

             "predict lgfit, xb" -- gives fitted log odds

  Summaries: The argument "pfit" is simply a name you choose for the 
             variable that you would like to be the fitted values.
             Similarly "lgfit" is a name use choose for the fitted 
             log odds. 


  Options:   (1) "predict pfit"  -- this creates "pfit" as the fitted
                 probability for each observation.  Recall that this
                 is just exp(xb)/[1+exp(xb)]  where "xb" is the 
                 fitted "X beta" for each observation.

             (2) "predict lgfit, xb"  -- this is used to create
                 the variable lgfit which equals the estimated
                 regression log odds, "X beta".





****************
****************
** fracpoly   **  Using fractional polynomials (this command not necessary for
****************  homework 5, but could be used to help identify function form.
****************

  Overview:  The "fracpoly" command allows use of general polynomial
             regression models (even using fractional powers such 
             as sqrt(X) = X^{1/2}, and negative powers such as
             1/X = X^{-1}).
             This allows a very flexible curve to be used for a 
             continuous covariate.

  Usage:     "fracpoly  logistic chd bmi"

             "fracplot bmi"

             Note that the useage of "fracpoly" comes before a general
             regression command.

  Summaries: The "fracpoly" function actually uses a variable selection 
             method to chose the powers of X that will be used.  This is
             why the fitting procedure can be somewhat slow.  When a model
             is determined the routine will tell you the powers of X that
             are used.

             I don't find the coefficients of these variables interesting,
             but the shape/form can be shown using the "fracplot bmi"
             command.  This includes the regression model with standard error
             bars.


  Options:   (1) "fracplot bmi" -- shows the fitted curve with 95% CIs.