Define/Review Model

Click the OK button in the Variance Estimation and Precision Startup Panel to display the Define/Review Model dialog box, which contains one tab: Quick. Use the options on this tab to review the default design and make any necessary changes.

Estimating method. Use the Estimating method drop-down list to choose the method to be used for estimating the variance components for the random effects. When ANOVA is selected, a general linear model is fit to the data and when REML is selected, a mixed linear model will be used. For more information, see Computational Details.

ANOVA. Select ANOVA to estimate the variance components of the random effects in the model by equating their variances to their expected mean squares.

REML. Select REML to estimate the variance components of the random effects in the model by Restricted Maximum Likelihood (REML) estimation. The basic idea behind REML estimation is to find the set of weights for the random effects in the model that minimize the negative of the natural logarithm times the likelihood of the data (the likelihood of the data can vary from zero to one, so minimizing the negative of the natural logarithm times the likelihood of the data amounts to maximizing the probability, or the likelihood, of the data).

Sum of squares. The options under Sum of squares are used to choose the type of sums of squares decomposition when estimating the variance components for the random effects in the model.

Type I. Select the Type I option button to attribute shared variation between effects to the effect which enters the model first, as determined by the order in which they were selected to be included in the model. Unlike the Type II or Type III decompositions, this decomposition produces sums of squares for the effects that add up to the total sums of squares. Type I sums of squares, however, typically depend on the order in which effects are included in the model.  

Type II. Select the Type II option button to test an effect in the presence of all other effects except any higher order interactions in which it is involved. Type II sums of squares typically do not add up to the total sums of squares.  

Type III. Select the Type III option button to specify that shared variance between effects is not attributed to any effect. Type III sums of squares do not depend on the order in which effects are entered in the model, but typically do not add up to the total sums of squares.  

Type V. Select the Type V option button to specify a combination of the methods employed in computing Type I and Type III sums of squares. Specifically, whether or not an effect is eligible to be dropped from the model is determined using Type I procedures, and then hypotheses are tested for effects not dropped from the model using Type III procedures. For more details, see Type V Sums of Squares.

Note that there are several modules in STATISTICA that will perform Analysis of Variance for factorial or specialized designs. For a discussion of these modules and the types of designs for which they are best suited refer to Methods for Analysis of Variance. Note also that the General Linear Model (GLM) module can analyze designs with any number and type of between effects and compute ANOVA-based variance component estimates for any effect in a mixed-model analysis using any of the six types of sums of squares.

Save design. Click this button to save the design for the currently selected response variable(s) to the spreadsheet. If you have specified separate designs for each response variable, STATISTICA will store the metadata associated with each response variable. If the design has been saved, then in future Variance Estimation and Precision analyses, the stored metadata will be used to populate the dialogs, making it easier to progress to the Variance Estimation and Precision Results dialog box. If, during the same session, you make changes to the design after you have saved it (e.g., you go back and specify a covariate), the new default design may not apply. In that case, STATISTICA will reapply the initial generic design as suggested by the data.

Note that only one design per response variable can be stored in the spreadsheet at a time.  Also note that while this option saves the metadata to the spreadsheet, it does not save the actual spreadsheet. You must save the spreadsheet manually by clicking the toolbar button or selecting Save from the File menu. Once you have saved the actual spreadsheet and closed the analysis, you will not be able to view Variance Estimation and Precision Startup Panel. That is, in subsequent sessions, you will not be able to make changes to the factors or covariates selected for the design. You can, of course, select new responses by clicking the Responses button (see below).

Saving designs in macros. As with all STATISTICA analysis, you can create a macro that recreates the steps in the current analysis (see Create macro, below). When, in the course of your analysis, you save the design to the spreadsheet, you are changing the spreadsheet. If you should then create a macro (that recreates the saving of the design), you would not be able to apply it to the changed spreadsheet. However, you could apply the macro to a new data file and save the design to the new data file via the macro. If you want to create a macro to analyze a data set with a saved design, you should record the macro after you have saved the design to the spreadsheet.

A note about metadata. The metadata for Variance Estimation and Precision designs (e.g., whether a factor is fixed or random and the design effects for response variables) are stored as variable properties. To view the Variable Properties dialog box, double-click the variable header and click the Properties button on the Variable specification dialog box. To view the value of a specific keyword, select the keyword in the Properties list box. The corresponding value will be displayed in the Value list box. Note that you should not change the values shown in the Variable Properties dialog box. All changes to a Variance Estimation and Precision analysis should be made via the options on the Define/Review Model dialog box.

Use the design for the first selected dependent variable as the default design to save. When a Variance Estimation and Precision analysis is first launched, STATISTICA generates a default design based on factor coding in the data set. For example, if the indexes of a factor (A) continue to increment across the levels of another factor (B), then in the default design factor A is nested within factor B [A(B)]. This default design will be displayed in the Design representation box and used to generate results in the ANOVA table even if no response variables have been selected for the analysis. If you want to replace this initial spreadsheet default design with the design specified for the first response variable selected in the Responses list box, select this check box and click the Save design button. The spreadsheet default design will be changed to the design specified for the first response variable. In future analyses, this spreadsheet default design will be used when a new response variable is added to the analysis (i.e., a response variable for which you have not specified a design) or if you launch an analysis without specifying any response variables.

Customize design. Click this button to display the Define Custom Design dialog box, which can be used to further refine the model. The options in this dialog box enable you to make crossed or nested effects using the selected factors. Additionally, options are provided to designate factors and effects as either fixed or random.

Design representation. The box contains a representation of the current design and a list of all random factors. The default design is based on the coding schemes for levels of factors in the data file. Specifically, if the indexes of a factor repeat across levels of other factors, then that factor is crossed with those other factors. If the indexes of a factor continue to increment across levels of other factors, then that factor is nested within those other factors. In the initial (default) design interactions between random and fixed effects are random, nested effects are random and crossed effects are fixed. The design is represented using industry-standard symbols (e.g., A crossed with B is represented as A*B, and A nested within B is represented as A(B). This box will be updated when customizations are made to the design. Note that if a design has been previously saved with the current data set, that design will be shown here. Factors and effects can be designated as fixed or random on the Define Custom Design dialog box (accessible by clicking the Customize design button). For more details on the rules used for generating the default design, see Determining the Default Design.

Dependents. Click the Dependents button to display a variable selection dialog box, which is used to select dependent (or response) variables to use for the analysis. The variables selected in that dialog box (or via the Variables option in the Startup Panel) will then be shown in the Dependents list box. To designate a dependent variable for the current design, select the variable name in the Dependents list box. A different design can be specified for each dependent variable. By highlighting more than one dependent variable in the list box, all current design specifications will propagate to each highlighted dependent variable.

OK. Click the OK button to display the Variance Estimation and Precision Results dialog box.

Cancel. Click the Cancel button to return to the Variance Estimation and Precision Startup Panel without accepting any unsaved changes.

Options. See Options Menu for descriptions of the commands on this menu.

Modify. Click the Modify button to display the Variance Estimation and Precision Startup Panel. You will then be able to modify the current analysis.