Anova

This application provides an interface for performing Anova analyzes on the database variables on the R software. Type III sum of squares are used.

Warning

This software is meant for exploratory analysis. When you use it you are likely doing multiple comparisons however, the reported P values are not corrected. You must be very careful when interpreting the results.

The interface is divided in a control panel at the left, a main plot at the top right, and a results table at the bottom right.

Hint

When you hover over any point in the main plot you will see the id of the associated subject, if you right click you will see a context menu that will let you switch to it in other applications.

The following sections describe the steps necessary for performing an Analysis. Notice that they can be accomplished in any order.

Selecting an Outcome

In order to select an outcome click on the box labeled outcome. It will give you choices based on your favorite variables (see Configuring Braviz) and previously used outcomes. It will also have an option labeled select outcome. Whe you click on it an enhanced variable select dialog will open.

Notice that at the moment only numerical outcomes are available.

If you have already added some regressors and interactions, the dialog will be able to display several plots that integrate the regressors with the possible outcomes. In this way you can get a better a feeling of how the outcome relates to the regressors before doing the actual calculation. The following pictures show examples of these plots

When you are happy with the selection click on Save and select

Selecting Regressors and Interactions

Regressors and interactions are displayed in a table together with their associated degrees of freedom (DF).

In order to add regressors to the table click on the button labeled Add regressor. This will open an enhanced variable select value.

Notice that at the variables list at the top-left there is an additional column labeled Importance. This column displays the importance of each variable for predicting the value of the outcome. This value is calculated using the randomForest R package. When the dialog is first opened this values are not available, in order to perform the calculation click on the column label. To add variables to the table of current regressors click on the button labeled save and add. You may also remove regressors by right clicking on them in the table at the bottom left and clicking on remove. When you are happy with the selection click on Done.

To add interactions click on the button labeled Add Interaction. The following dialog will open

In order to add a single interaction term select all the desired factors by clicking on them while holding the Ctrl key and afterwards click on Add Single Term. The Add all combinations button will add all possible products from the current regressors. When you are done close the dialog.

In the main window you can see the list of regressors and interactions at the left side. In order to remove any of them right click on it and select remove. Notices that if you remove a regressor, all of the interactions that contain it will also be removed.

If you double click on any of the rows in the regressors table the main plot will display the relation between this regressor and the outcome.

Choosing a Sample

At the lower right you will find the following tree which shows the current sample composition.

In order to select a different set of subjects to work click on the button labeled Modify sample. The sample select dialog will open (see Creating Samples)

Hint

If you double click on any node of the sample tree the subjects it contains will be highlighted in the main plot.

Fitting the model

When you have an outcome, and at least one regressor the Calculate Anova button will become active. When you click on it the R calculation will take place and you will see the table at the bottom right fill with the results. Also the main plot will show diagnostics from the residuals.

If the anova hypotheses are met

• The residuals variance should be constant along the outcome values
• The residuals histogram should look like a normal distribution

The results table contains columns for regressor name, sum of squares, degrees of freedom, F statistic value, and the associated P value.

Hint

If you double click on any of the rows in the results table you will see a corresponding plot.

Exporting data

The file dialog contains an entry labeled export which inside contains items that let you:

• Export the data used in the analysis
• The main plot as a vectorial graphic

Scenarios

The file menu contains options for saving and loading scenarios. This scenarios will include the current variables, current sample, and current plot.