Why do you analyse the 5 injuries seperately? You could collapse them into one variable, as it seems.
But if you want to analyse injury 5 separately, you could better use Chi square test.
Hi Everyone! I am conducting a retrospective research study to determine if a new specific sign of an injury is correlated with more related injuries to the region than two other signs.
Patients have been sorted into one of three groups (one independent variable):
- Group 1: Cluster of signs that DOES NOT include the sign of interest (N=22)
- Group 2: Cluster of signs that DOES include the sign of interest (N=19)
- Group 3: Another cluster of signs that DOES NOT include the sign of interest (N=50)
There are five types of injury that we have measured in each patient (five dependent variables):So, we hope to show that Group 2 has more of Injuries 1-5 than Groups 1 or 3.
- Injury 1-Injury 4 are counts of the number of that injury in a patient. e.g. Patient 39 has 5 ligament tears, 2 fractures, etc.
- Injury 5 is whether or not there was any injury to a particular anatomical structure.
In SPSS, I ran a one-way ANOVA with a Bonferroni post-hoc. This seems to give me what I want: 10 P-values for the null hypotheses that Group 2 has the same number of [Injury X] as [Group 1/3].
I have two questions:
1. I used a dummy variable for Injury 5, assigned 0 if the injury is absent, 1 if the injury is present. I recall a professor teaching me that this was acceptable, but I'm having doubts because I can't find examples on the Internet. Is this acceptable? What are the implications if it is?
2. My statistics are a little rusty, so I've been doing a lot of reading and was reminded that, before I chose my post-hoc test, I should have checked that my error variances were equal. A Levene's test gave the following significances:Injury 1: 0.810So, it appears that my error variances are not equal. From what I've read, I would have been better off with a Dunnett T3 post-hoc. Do you agree with that and would it be responsible to switch at this point?
Injury 2: 0.000
Injury 3: 0.386
Injury 4: 0.174
Injury 5: 0.002
Last edited by Crunch; 02-21-2014 at 12:10 PM.
Why do you analyse the 5 injuries seperately? You could collapse them into one variable, as it seems.
But if you want to analyse injury 5 separately, you could better use Chi square test.
Crunch (02-25-2014)
Collapsing injuries into one variable is a great idea, thank you. Unfortunately, I can't do that here.
If I analyze Injury 5 with a chi-square test, what should I do about type I error? Would I use a Bonferroni correction and just call significant results anything from the ANOVA and chi-square tests with P<0.05/2?
Tweet |