Learn how to create and interpret crosstabs.
Register for this 20min webinar to learn:
If you're new to market research or just want to brush up on your basic skills, register for this live webinar and become a crosstab aficionado!
This webinar is targeted at people that are new to doing their own analysis. It focuses on when and how to use crosstabs in market research.
Just in case you aren't aware, the crosstab is the workhorse for all survey analysis. It's the most used and most useful tool in the entire survey analysis toolbox.
In this session we will look at:
We start with some data we want to understand.
Here i have some data about attitudes towards Coca-Cola.
We can see that 33% of people love Coke. But, who are these people? Is this among everybody? Or just some segments?
What's the score among young people?
Well, that's interesting, the 18 to 24s have more love than the total sample.
What about the 25 to 29 year old age group?
Now, fun is this is, you'll rightly think there must be a better way. There is.
This is a crosstab, it shows the data from one key question, summarized by different sub-groups. In this case the sub-groups are age categories.
Looking at this table, we can see that the highest level for Love of Coke is with the 18 to 24s, and this declines with age.
The last column, the NET, just shows the total result without any filter. It's the same result as shown in the very first table.
Why we use crosstabs
Sometimes we have a theory about a particular market. For example, we may believe that people who like Coke are very traditional, and not into new experiences. We can test that by crossing attitude to Coke by openness new to new experiences.
As we can see here, there are only pretty trivial differences between whether people say they are open to new experiences and whether they love Coke. For example...
So, we conclude that attitude to Coke isn't relate to openness to new experiences.
Why we use crosstabs ... Hunting for patterns
In practice, we don't just test theories. Commercial researchers run lots and lots of crosstabs, looking to see what insight they can extract from the data. I'll return to this later.
What makes a good crosstab?
What makes a good crosstab? I'll start by showing you a bad crosstab.
This is a bad crosstab
This is a bad crosstab. Why?
Let's make it a good crosstab
I'm glad you asked. On this page I have listed eight basic rules for making a good crosstab.
We will return to our bad crosstab, and make it better.
Making a bad crosstab good
Each cell on the table has three numbers. A Column %, a Row %, and a Count. Back when I was learning reserach, this was routine. It was routine because it took a week to get a table, so we wanted tables which had all the data we could possibly want.
Today, computing is quick and easy, so we should make simpler tables. We should as a general rule, only show the column %.
It has a third as much data, so can be understood in one third the time, or less.
The second thing is that as a general rule, we should show demographic data and other profiling data in the columns.
This way, we can intrepret each column as a stand alone summary table, and more easily compare results.
The next thing is to report showing whole numbers. That means remove the decimals.
Get rid of the percentage sign.
Why do we get rid of the percentage sign? The more text on a table, the harder it is to see parterns. With this table, about a third of the characters are % signs, so all they do is make it hard to read.
The next thing is to merge together data or create nets in the rows.
As a general rule, with attitude stuff we just about always want to look at the combined love and like data, whihc is commonly known as top 2 boxes.
If we have weights, we need to remember to apply them.
It can be useful to look at sample sizes in column as well.
Many researchers won't even look at data with less than 30 respondents, and would merge the 65 or more category with the 60 to 64 category.
But, I prefer to be guided by stat tests.
Step 8 is to put stat tests on the table.
The last step is to merge together any columns that have no interesting results and small sample sizes.
I'm going to merge the under 40s
There are some common additional customizations that are popular as well.
Adding statistics (e.g., averages)
It can be useful to add the average to a table
Statistics > Below > Average
We may choose to use conditional formatting to highlight cells.
We can sort the data, either manually or automatically. In this case, I've sorted from highest to lowest on the NET column. As the data has a natural order sorting makes little sense in this case, but I just wanted to make the point.
Additional questions in the banners
Old timers in market research refer to the columns in a table as banners or banner points and the row headings as stubs.
This table just has age as the banner, but it can be useful to add additional questions. For example, gender.
We can add additional headings onto tables, which are sometimes called spans.
Now for something a bit more complicaed. This table looks like a crosstab, but it isn't
A crosstab shows the relationship between two questions.
But, this is all from one big grid question. We've got attitudinal data for each of six brands.
We can crosstab this, but we end up with a mess.
Tables like this are really hard to read.
So, we need to instead flatten our grid.
It's showing the same data as before, but in a single column. That is, the data has been flattened.
Now we can crosstab with age. This tends to be much easier to understand.
Using stat tests to save time
Earlier I talked about the need for crosstabs to show stat tests. Why is this so important? It's all about saving time.
Apologies if you have seen this analogy before. This is Wally.
Where's Wally. Can you find him?
The main reason we use stat testing in market research is to help us know where to look.
The stat testing tells us where we don't need to look. It's quite hard to find Wally in this image.
Where's Wally with stat testing
But, he's a lot easier to find here.
Attitude ... Q12
When we have a theory, we need to carefully read the data.
But, if we are just skimming through lots of crosstabs to find anything, stat tests save so much time.
We have some blue and red here, which tells us that we have some statistically significant differences. So, we need to figure out if they are interesting. I will return to this in an upcoming webinar.
Nothing's significant here. So, we don't need to even attempt to read it.
Again, nothing's significant, so we don't need to read it.
To save time, I can completely automate this and just delete all the tables that aren't statistically different.
As you can see, lots of the crosstabs have just been deleted.
I've now got many fewer tables to read.
Resources for learning more
In our resources you will find a webinar that provides A beginners guide for survey analysis.
You'll also find more depth on statistical significance.
And, lots more advanced topics as well.
And, our next few webinars will be all about how to turn the crosstabs into insight-filled reporting.