# compare 4 groups on several categorical characteristics

#### Hilde_S

##### New Member
Hi!

For my student research assistant job I have collected many pictures of dogs, and these dogs can be divided into four groups (young vs. old and breed 1 vs. breed 2). Before we are going to show the pictures to participants to compare these groups, we first want to make sure that some picture characteristics are equal across the 4 groups (e.g., if the dog is making eye-contact; the dogs' body position; how good you see the eyes etc.). These characteristics are all categorical and consist of 2 or more levels (for example: body position can be 'sit', 'layingdown', 'stand'). My colleague suggested to use statistical analyses to test if these 4 groups significantly differ on these characteristics.

My question is: Is this a good idea and what analysis can I use to compare 4 groups on several factors?

Hilde

#### obh

##### Well-Known Member
Hi Hilda,

A statistics test run on random sample data and gives the result for the entire population based on the sample.

If I understand you correctly, in your case you build the sample data to get almost identical groups for each characteristic.
For example if there are 3 body positions: A, B, C and 4 groups (G1, G2, G3, G4) and 21 dogs in each group

You want to build something like:
Expected group
Body, G1, G2, G3, G4
A,5,5,5,5,
B,6,6,6,6
C,10,10,10

But you succeeded to build

Actual group
Body, G1, G2, G3, G4
A, 6, 5, 4, 6
B, 7, 6, 6, 6
C, 8,10,11,9

Now you want to decide if the actual groups are close enough to the expected group?
"Chi-square test for association" will do something like this (a bit better as it calculates the expected values based on the actual data)

You may use a statistical test, not for the original purpose as a way to measure if the ratio between the group is similar.
You definitely not tied to the regular significance level of 0.05, you may decide that you need at least a p-value of 0.9 for similar groups.
You may as well use other non-statistical methods, as not more than 5% deviation from the average ratio of all groups.