# comparing 3 stratified groups

#### robbyd

##### New Member
Hello all,

I would like to make a comparison of some data collected from a burn center. The study subjects are stratified into groups based on the size of their burn (0-10%, 10.1%-30%, 30.1% or more). The cause of burn is reported as a percentage. See data below.

What is the appropriate test? How do I report the results?

Last edited:

#### Archidamus

##### Member
I am reading this as you want to see if there is a significant difference in coded burn source between groups? You've already got the data in a contingency table, so my initial idea would be to do a chi-squared test for independence. That test would let you know to what degree you can make a claim that the burn mode and group are dependent. Assuming you get a p-value below you pre-determined alpha, you could make an inference that the burn source has influence on which group they end in.

Assuming the chi-squared returns a significantly low p-value, which my eyeballs are telling me so, you could do a 2-proportion test to determine if the probability of ending in the 3 groups differ when burned by the same type source. Think of this as a coin flip, or in this case a 3 sided object. The null hypothesis is that when you get burned by a flame you have an equal chance of ending in group 1, 2, or 3. The alternative is that the probability of one of these groups is not equal to another. This could be done to show that the probability of being a group 3 when burned by a flame is higher than ending in group 1.

#### ask.biostat

##### Member
Hello, I would like to add that better way is to use measured burn sizes instead groups. In this case you would use more information for analysis and results would be more accurate.

#### Archidamus

##### Member
Hello, I would like to add that better way is to use measured burn sizes instead groups. In this case you would use more information for analysis and results would be more accurate.
Excellent point, get the specific body burn% of each individual, that way you can do more powerful analysis like regression.