Shown below is the proportion of men and women who responded to our survey. Suppose that we know that the proportion of women in the whole organization population is actually 55% and not 48% as in our sample.
Since women are underrepresented in our sample, their responses need to be given a larger impact in order to avoid a biased result. Also, as shown below, the distribution between departments is not exactly the same in the sample and in the population:
This means that we need to weight each of the country response groups differently as well. How do we do this?
If weight variables are not already a part of the dataset, the process begins with defining a new weight variable.
Click on the `Weighting´ button in the ribbon and then click on ‘Insert’. As we want to give weighting for both country and gender, name the new weight variable `Gender and Department´.
Click on ‘Insert Variable’ and select the two variables, ‘Gender’ and ‘Department’, which defines the Weight variable ‘Gender and Department’.
The default weight to each of the categories is an equal distribution of countries and genders.
Since we know that there are 55 % women and 45% men in the organization, the values of each category are modified to match these percentages. In regards to department, the respective values are modified so the proportion between the departments corresponds to our total population. The effects that these weights have on our data are, for example, that the data from a woman from HR weights more heavily than the data from a Male from Marketing. To see the logic in this we compare the size of the response groups in our sample, and the size of them in relation to the population. In our sample, women and HR were underrepresented compared to their size in the population, while male and Marketing were overrepresented.
To see the effect of our weighting we will:
Create a table with country as the primary variable, and gender as the background variable, and
Apply the weights under the `Weight´ tab. We now select the weight we just recently created in the drop-down menu ‘Weighting’ at the bottom of the window:
Below are two tables showing the difference between a table with weighted data and one without:
Comparing these two tables shows that the size of each group is now different. Both the relative sizes of men and women, and the relative size of each country group, have changed. The effect is that a group that was previously quite big, for example, Male from Marketing is now valued less. Likewise, the responses from a group that previously was small, Female from R&D, is now valued more. The logic behind this is as explained before, Female from R&D are underrepresented in our sample, and hence we need to value their responses higher (and vice-versa for Male from Marketing).
The Base in weighted tables will per default be expressed as the unweighted base. In ‘Settings’ you can for the report change the setting to ‘Show the Weighed Base.