Appendixes | Appendixes | Estimate Statements and the Estimate Builder

Estimate Statements and the Estimate Builder
Every fixed effect that you specify in the model is assigned a certain number of algebraic parameters in the linear model. The number of parameters for each effect is determined by the number of levels of each of the class variables in the effect. By clicking the + and - buttons, you assign coefficients to these algebraic parameters. For example, by clicking + for one treatment level and - for a control level, you create a difference between these two particular parameters.
All of the coefficients within an effect for which + is clicked are averaged, so the assigned coefficients sum to one. In this way you can construct differences of averages. For estimates of differences, the intercept coefficient should be left at 0. You can also construct estimates of means, for which the intercept coefficient is typically 1. The scientific null hypothesis tested is that the sum of all nonzero coefficients times their corresponding algebraic parameters equals zero.
When the coefficients in the boxes are as you desire, click Create an Estimate Statement Below to construct a single SAS ESTIMATE statement in the text box. Click Clear Entries Above to reset all coefficients to zero in order to start building another Estimate statement. You can also delete or edit the statements directly in the text box. In this way you can construct an arbitrary number of simple or complex hypothesis tests.
When finished, click Save Estimate Statements to Output File in order to save all of the statements to the output file. This file can then be used as input in the ANOVA or Mixed Model Analysis analytical processes. Make sure the Class Variables and Fixed Effects are specified in exactly the same order in both the Estimate Builder initial dialog and in ANOVA or Mixed Model Analysis, otherwise the order of the coefficients can be incorrect.
Note there is the possibility that the linear combination that you specify is not estimable. This is more likely with higher order factorial designs, unbalanced designs, and missing data. In this case the output can and usually does contain missing values. Refer to the SAS PROC MIXED documentation for additional details about ESTIMATE statements and estimability.