Hello everyone
I have a data set for two group of patients one group undergoing special physiotherapy treatment (Treatment group) and the other group undergoing classical home treatment (Control Group) and I want to deter effect size for each group (Pre treatment VS post Treatment) and I don't know what effect size method I should use, my supervisor told me use r not cohen's d and there is two type of r I found I will put the links bellow as well as I will attach the result of pre and post treatment of one of the groups of patients.
please I need your help to tell me which r method I should use between this to methods.
1- Pearson Correlation Coefficient
(https://www.socscistatistics.com/tests/pearson/Default2.aspx)
2- https://www.polyu.edu.hk/mm/effectsizefaqs/calculator/calculator.html
It sounds like you would want to use Cohen's d for paired samples. This would match a situation when you would use a paired t-test.
There might be a few ways of calculating this, but using the lsr package in R, it appears the paired-sample Cohen's d is the same as the mean of the differences divided by the standard deviation of the differences. For the data you attached, I got a Cohen's d of 2.32 by this method.
For some variants, see: http://www.real-statistics.com/students-t-distribution/paired-sample-t-test/cohens-d-paired-samples/
I think r is still Pearson's r in your second link. (I can't be sure. I didn't read through their equations). One way to use r for two independent samples is to convert them to two variables, one variable is 1 for Group1 and 2 for Group2, and the other variable is the measurement variable. You can then correlate these two and get an r. I think this is what the website does essentially, except through a simple equation. For your attached data, I got r = 0.867.
But I don't think this method is applicable for your data because you have paired observations (Before/After for each individual), so the samples aren't independent.
TLDR: Your advisor should explain to you what they have in mind. But I suspect the use of r as an effect size measurement isn't ideal in this case.
I agree with Salvatore S. Mangiafico. For a paired comparison such as you are conducting, the effect size is the difference in the means, relative to the standard error, and this precisely what d is designed to capture.
The only way I can see to use r is the method suggested by Salvatore, but this is a measure of how well the T-1 data predict the T-2 data, not of the effect of the treatment. In particular, in a regression model, the treatment would would be included as a dummy (1 or 0) variable, which would affect the intercept, not the slope. Imagine the T1 scores were identical to the T2 score, that would give an r=1.0 but with no effect of the treatment.
My suggestion is that you find a statistician to resolve this disagreement, rather than trying to convince your supervisor yourself.
Hi,
1. Analyses with a calculator:
Here, a calculator for computing Cohen's d directly (with means and std deviations of your two distributions):
https://memory.psych.mun.ca/models/stats/effect_size.shtml
with your data:
d=2.322 (Effect sizes expressing the degree of difference between means)
Little trick: all effect sizes indicators can be computed each with others
To compute r from d (sorry for the proposition but impossible to use mathematical objects):
r=d/ square root (d*d + 1)= 2.322/2.528= 0.92
r2= 0.84 (r squared : Effect sizes expressing the “proportion of variance explained”)
2. Analyses with SPSS:
Using SPSS and your data, I computed T-test for paired sample :
T-TEST PAIRS=beforetreatmnet WITH aftertreatment (PAIRED)
/CRITERIA=CI(.9500)
/MISSING=ANALYSIS.
t=-14.32 p
- Badges
- Science topic
- Similar topics
- Computer Science
- Research Topics