Often it is useful to compare results between different sub-groups. For example, how do the preferences of males differ from females? The main way of doing this is to create crosstabs. This article describes:
How to read a simple crosstab
The crosstab below shows how likely people said they would buy a product by gender. Each column represents a sub-group, where NET represents the entire sample. It shows us, for example, that:
- 28% of men said they would definitely buy the product.
- 20% of women said they would definitely buy the product.
- 24% of all people (i.e., the NET) would definitely not buy the product.
In this study, the total sample size is 300 people. Of these 134 are male. Of these males, 38 said they would definitely buy. The 28% figure is computed as 38 / 134. Similarly, the 24% figure for the total sample is 71 / 300 = 23.47% which, after rounding, becomes 24%.
In essence, this table is created by combining a summary table from the total sample, with two filtered summary tables, one for men and one for women. Refer to Introduction to Statistics for more about how to read the column percentages in the table above.
The table above is somewhat vanilla. A more customized table is shown below. The customizations involve:
- Creating banners, with multiple questions across the top.
- Using nets - TOP 2 BOX and BOTTOM 2 BOX - to summarize the data. "Top 2 box" is a term of art in survey research. It refers to the sum of the two most positive statements, which in this case are I would definitely buy it and I would probably buy it.
- Using spans to emphasize the positive (+VE) and negative (-VE) responses.e
- Removing the nets for each question in the banner (top of the table) and removing the question names.
- Changing the coloring of the table.
- Using letters to show statistical significance.
- Adding an additional statistic, the Average, to the table.
Creating lots of crosstabs
Most of the analysis of a survey involves creating and interpreting lots of crosstabs, often with filters. There are three common workflows:
- Creating a "deck" of crosstabs. The approach here is to create a large number of crosstabs. For example, perhaps you may create crosstabs showing every question in the survey in rows, with all the key demographic variables in the columns. Then, you read through all the tables looking for something interesting.
- Creating an analysis plan. This approach involves thinking carefully about the goal of the study and working out, prior to looking at the data, what the most insightful analyses will be. Then, these summary tables and crosstabs are created and interpreted. In the case of the iLock study, where the key question is priced purchase intent (this is discussed in more detail in Overview of How to Finding Insights in Data), suggesting we should crosstab everything by priced purchase intent.
- Interactively exploring the data. This approach involves creating summary tables of all the questions in a survey and then using gut feel and judgment to create additional tables to explore interesting results.
While most researchers have a preference for one of these three workflows, in practice most people use a mix of the three.