proc genmod data= ;
model y = x / dist= link= type3 ;
repeated subject= / corr= ;
lsmeans x / diff=control ('ref') cl adjust=Dunnett(or tukey);
Parameters estimated with maximum likelihood methods. This is an iterative process, i.e., the computer adjusts the parameter until the log likelihood function is maximised. Repeated statement: specifies the covariance structure of multivariate responses and iterative fitting algorithm for GEE model fitting
|normal||identity||continuous dependent variable, mean difference|
|binomial||log||risk ratio from exponentiation of the parameter estimate|
|poisson||log||risk ratio from exponentiation of the parameter estimate; robust error estimates can be requested with the REPEAT statement|
|multinomial||categorical response variable with more than two levels|
|Sums of squares||Effect type||Note|
|Type I||Sequential||The SS for each factor is the incremental improvement in the error SS as each factor|
effect is added to the model. In other words it is the effect as the factor were
considered one at a time into the model, in the order they are entered in the model
|Type II||Hierarchical or partially sequential||The SS for each factor is the reduction in residual error obtained by adding that term to a model consisting of all other terms that do not contain the term in question.|
|Type III||Marginal||The effect of each variable is evaluated after all other factors have been accounted for.|
Adjustment for pairwise comparison
- Tukey: all pairs
- Dunnett: paired with control
Generalized Estimating Equations
GEEs are used for analysis of correlated data, e.g., subjects are measured at different points in time, or subjects are clustered, i.e., share a common characteristic. GEE analysis can be performed in GENMOD by specifying a REPEATED statement which provides clustering information and a working correlation matrix. The REPEATED statement requests a GEE analysis.
- Subject: Responses from different subjects are assumed to be statistically independent, and responses within subjects are assumed to be correlated. Variables used in defining the subject-effect must be listed in the CLASS statement. The input data set does not need to be sorted by subject.
- Corr= specifies the correlation structure: Un unstructured; IND independent.