This article describes how to replicate the statistical assumptions settings of SPSS Custom Tables Significance Tests.
- Select the tables you wish to have the testing applied to.
- In the object inspector on the right, go to Properties > APPEARANCE > Significance and select Compare columns.
- Go to the Properties > RULES and click on the button to add a new rule, then select Significance Testing in Tables > Independent Samples Column Means and Proportions Tests.
- If you get a warning message, press OK. The rule options dialogue box will appear.
- If you're using the SPSS Custom Tables setting of Estimate variance only from the categories compared, set Numeric data variance assumptions (t-test) to Equal variance (categories compared). Otherwise, set it to Equal variance (all non-ignored categories).
- Set the Multiple Comparison Correction to whichever of Bonferroni or None is selected in SPSS.
- If you wish to specify particular comparisons (e.g., comparing to the NET), check Set column comparisons (eg: A/B, C/D) and modify the comparison as appropriate.
- Set the other options as shown below and press OK to apply the rule.
If you wish to set up the testing to apply as a default for all items in the table, go to Properties > RULES, right-click on the rule, and select Apply > Add to every item in the document. Note that you should not do this if you are modifying the comparisons as these will need to be specified differently depending on the number of columns in each table (to do this, remove and reapply the rule based on the steps above).
- SPSS Custom Tables is not meant to be used with sampling weights. To quote from its documentation (Version 23): "The case weights (or frequency weights) are supposed to be integers representing number of replications of each case. In column proportions tests, we will only check if the column marginal 's are integers. If not, they will be rounded to the nearest integer." Displayr does not replicate this rounding. This may cause line-ball results to change.
- SPSS Custom Tables does not perform tests where the percentages in a group are either 0% or 100%. Displayr does perform tests in this situation.
- Displayr does not attempt to replicate SPSS's treatment of multiple response data in statistical tests.
- SPSS Custom Tables does not provide detailed diagnostics (e.g., p-values, t-statistics, z-statistics), which are necessary in order for us to verify that our computations replicate SPSS computations. Consequently, while we have performed tests, there may be other scenarios where Displayr's results differ from SPSS's results.