Help for GLM power links                                       [STB-12: sg16.1]
------------------------

The syntax for ^glmp^ is

^glmp^ depvar varlist [^fw^=weight] [^if^ exp] [^in^ range], ^f(gau^|^poi^|^gam^|^invg)

     ^p(^#^) sc(d^|^c^|^n) ex(^varname^) o(^varname^) le(^#^) it(^#^) lt(^#^) ef^orm

where ^p()^ is the power value and ^sc()^ is a scaling option. ^glmp^ allows
the user to model the full range of acceptable power links to the following
distributions: Gaussian, Poisson, gamma, and inverse Gaussian.  

^p()^ is required for any power other than the default value of 1. Typically 
used powers are the square, p(2), square root, p(.5), and inverse, p(-1). 
^glmp^ can also model all identity links using p(1) and canonical links by: 
Gaussian, p(1); Poisson, p(0); gamma, p(-1); and inverse Gaussian, p(-2). 
Log links use the p(0) option. Hence, to model a lognormal regression type: 

                     ^glmp depv varlist, f(gau) p(0)^ 



The ^sc()^ option allows the user to select the scale upon which the standard 
errors are to be adjusted. The default is to scale by the deviance ^(d)^ 
rather than by chi2 ^(c)^. ^(n)^ is to be selected if one desires no scaling. 
In general, you should use the (n) option with poisson and the default 
deviance with the gaussian, gamma, and inverse gaussian distributions. 
However, if there is evidence of substantial dispersion then you should 
scale the poisson standard errors by either the deviance or chi2.  Both 
dispersion values are presented on the screen. A better fitting model within 
a distribution is one that has low deviance and a pattern of significant 
explanatory variables.




Example
-------

^.  glmp mpg weight length, f(invg) p(2) sc(d)

Iter 1 : Dev =    0.1117
Iter 2 : Dev =    0.0765
Iter 3 : Dev =    0.0764
Iter 4 : Dev =    0.0764
                                                     No obs.    =        74

Chi2       =  .0819838                               Deviance   =  .0764327
Prob>chi2  =         1                               Prob>chi2  =         1
Dispersion =  .0011547                               Dispersion =  .0010765

Inverse Gaussian: Power = 2 
---------------------------------------------------------------------------
mpg      |      Coef.   Std.Err.        t     P>|t|    [95% Conf. Interval]
---------------------------------------------------------------------------
weight   |  -.0933343     .04528     -2.061   0.043   -.1820815   -.0045872
length   |  -3.785392   1.769495     -2.139   0.036   -7.253539   -.3172446
_cons    |   1452.548   213.9344      6.790   0.000    1033.245    1871.852
---------------------------------------------------------------------------
Standard errors adjusted by sqrt(dispersion)


For additional help: Joseph Hilbe, Department of Sociology
                     Arizona State University, Tempe, AZ 85287

or Voice: 602-860-4331
   Fax  : 602-860-1446 
   Email: atjmh@@asuvm.inre.asu.edu