# How to Replicate Survey Reporter Significance Tests

Jump to navigation
Jump to search

- Make the following changes in Statistical Assumptions:
- Change
**Show significance**to**Compare columns**. - Change both
**Multiple comparison correction**settings to**None**. - Set the
**Minimal sample size for testing**to its level in Survey Reporter (which defaults to 30). - Change the
**Proportions**setting to**Survey Reporter Proportions**. - Change the
**Means**setting to**Survey Reporter Means**. - Change
**Weights and significance**to**Kish's approximation**. - Uncheck
**Recycle column letters**. - Replace the
**No test symbol**with`**`. - In the
**Significance levels and appearance**grid, copy the one uppercase**Column letters**fields (i.e.,`A,B,C,...,`) and paste it over the top of each of the lowercase fields (i.e.,`a,b,c,...,`). Note that this will remove Q's default of using uppercase and lowercase to denote different levels of significance. However, it will not precisely replicate Survey Reporter's use of column letters as Q assigns a letter to NET fields but Survey Reporter does not. Two ways of resolving this are to:- Remove the NET categories from within Spans.
- Manually change the contents of
**Column letters**to exactly replace Survey Reporter if so required, but this will need to be redone for each question in the Brown Drop-down Menu. - Where a Pick Any question is in the Brown Drop-down Menu or analyzing a Number - Grid, Pick Any - Grid or Pick One - Multi SUMMARY set
**Overlaps**to**Dependent**.

- Change
- Move the NET categories to be at the beginning top/left of the tables (and sections within a table where Spans are used).
- Ensure that the
*base*used in statistical comparisons is equivalent (e.g., on a Pick Any - Grid Survey Reporter's base is the total number of responses, whereas Q's is the total number of respondents. - Ensure that the rounding is being computed in a way consistent with Q (Q rounds up; by default Survey Reporter rounds to even).
- Check that the percentages and means being compared in the statistical tests are the same. Where they differ this likely relates to the base. Sometimes Filters can be used to get Q to replicate Survey Reporter's percentages (however, keep in mind that this is something that is only done to aid in comparison rather than being a desirable approach to compute percentages).
- Ensure that any filters are equivalent (including Survey Reporter's
*Interview Filter*settings).