Dependent Paired/Overlap Z-Tests (Percents)

For data that is either paired or has overlap one of the Dependent Paired/Overlap Z-Tests is the appropriate test.

A table is Multi if its banner is a set of attributes which qualify the respondent for the respective banner points and whose rows are some other data about the respondents. For example, you want to compare the percent for age groups of those who consumed Coke with the percent for age groups of those who consumed Pepsi and the percent for age groups of those who consumed Sprite. The rows of the table are the age breaks and the banner points are the following:

Have consumed in the last 30 days:

Coke Pepsi Sprite

----------------- ------------------ -----------------

In this example a respondent may have consumed none, all three or any combination. Since respondents can drink Coke, Pepsi and Sprite, the same respondent can appear in all three columns of the table which makes the percentage dependent. In this example, Dependent Paired/Overlap (Multi) is the appropriate test.

A table is LOC+/VAR+ if its banner is a set of attributes which qualify the respondent for the respective banner points and whose rows are separate measurements relating to the banner points. For example, you want to compare the percent for ratings of Coke for those who rated Coke with the percent for ratings of Pepsi for those who rated Pepsi. The rows of the table are the rating scale and the banner points are the following:

Rated Coke Rated Pepsi

----------------- -----------------

Since respondents can rate both Coke and Pepsi, the same respondent can appear in both columns of the table which makes the percentage dependent. In this example, Dependent Paired/Overlap (LOC+/VAR+) is the appropriate test.

IMPORTANT: The Dependent Paired/Overlap Z-Test (LOC+/VAR+) is only appropriate when the LOC instruction or VAR instruction is used in the banner. The LOC instruction lets you show data from the same respondent in multiple data locations (can be used with ASCII or Variable type data). The VAR instruction lets you show data from the same respondent in multiple variables (can only be used with Variable type data).

Given the possibility of finding "significant" differences due only to the degree of overlap of the two samples, WinCross has adopted the safeguard of declaring all such differences not significant if the (a+b)/[(1+a)(1+b)] as defined in A Note on Spurious Significance is less than 5%.

When a "Total" column is being used and denoted as such in a comparison group, given the possibility of finding "significant" differences due only to the degree of overlap of the part to the whole (Total), WinCross has adopted the safeguard of declaring all such differences not significant if the fraction of the part to the whole is less than 5% or greater than 95%.

Z-Test Options

Confidence levels

Select one or two confidence levels for the chosen test.
You can choose 1 or 2 confidence levels from the pre-listed percentages or specify custom confidence level percentage(s).
WinCross will not allow you to specify a pre-listed percentage and specify a custom confidence level percentage.

Show significance indicators in banner

Displays the significance annotation displayed for each banner column.

Exclude cells that are 0% from analysis

Select Exclude cells that are 0% from analysis if you don't want to include cells that are 0% in calculations for percentage tests.

Note: The Exclude cells that are 0% from analysis option is automatically selected and disabled if you are using the WinCross selects Z-Test or Dependent Paired/Overlap (Multi) statistical testing options. You cannot show significance against blank cells for these two statistical tests.