How to Interpret "Inconsistent" Statistical Testing Results

Displayr contains multiple systems for computing and displaying statistical testing results. Each of these systems does different things and, consequently, their results can seem inconsistent. However, unless the user has modified default settings, the seeming inconsistencies are generally due to misinterpretations of outputs rather than genuine inconsistencies.

Methods

Cell comparisons (arrows and font color) versus column comparisons

In the table below, both Column comparisons and Cell comparisons (also called Exception Tests) are shown (this has been done by selecting Font colors in the Object Inspector > Appearance > Significance menu, and then selecting Statistics > Cells > Column Comparisons from the Data tab in the object inspector .

In the first row, no letters are shown, and instead, the "-" symbol is shown, which indicates that no columns have been compared. This is because, by default, columns are only compared if they are in the same span. That is, in this table, the Non-workers columns are compared with each other, and the Work columns are compared with each other (see also How to Specify Columns to be Compared in a Table). However, the font color, which represents the Cell comparisons (also called Exception Tests), shows that in June, the percentage for If I wanted to eat fried chicken, I would go to KFC is higher among the Work segment than the Non-workers segment, as the cell comparisons do not take the spans into account.

The second row of the table more starkly illustrates the difference between the column comparisons and the cell comparisons. The font colors indicate that Non-Workers, in general, find the side orders more appealing than Workers do. The column comparisons show that among the Non-Workers, the June score was significantly higher than the April score.

Planned comparisons (alpha) versus Cell and Column comparisons

In the second row of the table below, the only significant cell comparisons are for May (i.e., these are the only comparisons in color). But, if you select the April cell and press from the Object Inspector > Appearance > Significance in the object inspector, the result is shown as being Significant, as shown below. On the face of it, this appears to be inconsistent.

The reason for the difference is that by default in Displayr, Cell and Column comparisons have multiple comparison corrections automatically applied to them, whereas planned tests, like the one conducted with the button, do not.

If you look at the output shown below, you will see that this seeming inconsistency is also described in the sentence immediately below, where it is marked as significant. This difference in behavior is by design.

Note also that when selecting a single cell and pressing from the Object Inspector > Appearance > Significance (or else right-clicking > Statistical Test), it will show the results of a Cell comparison (also called an Exception Test), whereas if you select a pair of cells, Displayr will show a Column comparison.