Kevin Roland Viel wrote:
> Continuing with my line of inquiry, please consider:
>
> data test ;
> class = "x" ;
> do _n_ = 1 to 10 ;
> DV = 5 + rannor( 1 ) ;
> output ;
> end ;
> class = "y" ;
> do _n_ = 1 to 5 ;
> DV = 10 + rannor( 1 ) ;
> output ;
> end ;
> class = "z" ;
> do _n_ = 1 to 15 ;
> DV = 2 + rannor( 1 ) ;
> output ;
> end ;
> run ;
>
> proc glm data = test ;
> class class ;
> model DV = class / solution ;
> run ;
> quit ;
>
>
> Dependent Variable: DV
>
> Standard
> Parameter Estimate Error t Value Pr > |t|
>
> Intercept 2.182730812 B 0.24853563 8.78 <.0001
> class x 3.068968176 B 0.39296934 7.81 <.0001
> class y 7.446994749 B 0.49707127 14.98 <.0001
> class z 0.000000000 B .
>
> By virtue of the coding, I consider CLASS = "Z" to be the referent
> group. Thus, may I consider each line above as a test of whether CLASS =
> "X" and CLASS = "Y" are each statistically different from CLASS = "Z".
> This seems painfully obvious, so I thought I'd take my chance to be a
> pain
>
> If so, then by carefully naming the CLASS variables, I can adjust the
> referent group? For instance, by changing "x" to "zX":
>

This is certainly true. Where are you going with this? Have you looked
into contrasts, which allow you to simply test differences between
groups, and are not dependent on the coding scheme?

> Standard
> Parameter Estimate Error t Value Pr > |t|
>
> Intercept 5.251698988 B 0.30439274 17.25 <.0001
> class y 4.378026573 B 0.52722370 8.30 <.0001
> class z -3.068968176 B 0.39296934 -7.81 <.0001
> class zx 0.000000000 B . . .
>
>
> Thanks,
>
> Kevin
>
> Kevin Viel
> Department of Epidemiology
> Rollins School of Public Health
> Emory University
> Atlanta, GA 30322