Statistics in Crosstabs
Overview
MarketSight runs a variety of statistical tests based on the column variables, row variables, and options you have specified in your crosstab design. For more information on how to interpret your results, see Interpreting Significance.
Standard Rules
All statistical tests in MarketSight:
- Ignore rows and columns of data that sum to zero
- Assume that samples are taken randomly from large populations
- Assume that data contained in cells are independent
- For tests on means, MarketSight assumes the data in each column are independent and normally distributed
- For tests on frequencies, MarketSight assumes the data in each cell are independent
Statistical tests are NOT run in the following circumstances:
- The data you want to compare is in two or more separate tables
- A continuous variable without value ranges is placed in columns
- Shared values for variables are displayed in columns
- Cell count is under 5
- There is unequal variance
Tests Comparing Means
| Test Name | Description |
|---|---|
| ANOVA | Analysis of Variance (ANOVA) is tested using an F-test. The F-test determines whether or not the row variable's means are equal to one another for all columns. A prerequisite for ANOVA is the Levene test for equality of variances.ANOVA is not run for multiple response variables. |
| T-test | T-tests, in the form of Fisher's Least Significant Difference (LSD) test, are used to find whether or not two independent means are equal to one another. For both contrast and pairwise tests, this form of T-test is run when the option to "Correct for Type I errors" is disabled.A modified version of a T-test is run for multiple response variables, for both contrast and pairwise tests. |
| Corrected contrast T-test | Corrected T-tests, in the form of Scheffé tests, are used for contrast tests when the option to "Correct for Type I errors" is enabled.No correction is applied for multiple response variables. |
| Corrected pairwise T-test | Corrected T-tests, in the form of Tukey-Kramer tests, are used for pairwise tests when the option to "Correct for Type I errors" is enabled. No correction is applied for multiple response variables. |
Tests Comparing Column Proportions
| Test Name | Description |
|---|---|
| Chi-squared | Tests whether or not a relationship exists between the column variable(s) and the row variable in relatively large crosstabs.A modified version of a Chi-squared test is run for multiple response variables. |
| Fisher's Exact | Tests whether or not a significant relationship exists between the column variable(s) and the row variable in relatively small crosstabs with 2 cells per row and 2 cells per column (2x2).Used for both contrast and pairwise tests to determine whether two column proportions are equal to one another for cells with low cell counts.Fisher's Exact test is not run for multiple response variables. |
| Z-test | Determines whether or not two column proportions are equal to one another for relatively large crosstabs.Used for both contrast and pairwise tests. A modified version of a Z-test is run for multiple response variables for both contrast and pairwise tests. |