Statistics help - Comparing means non-normal and normal distributed data??

#1
Hi to everone,
I would like to ask you for little help with my statistics.
I have sample of 30 subjects - there is done some calculations with 5 different formulas on each subject (to calculate intraocular lens)(om each subject was done calculations with 5 formulas).
These calculalations are named(see attached file spss output file in PDF) RR_celkovo, RR_SRK2, RR_HillFRF, RR_Haigis, RR_Barret and their absoute values ABS_R_celkov, ABS_R_SRK2.......
I want to compare means of RR_celkovo with all other RR_. Also want to compare ABS_R_celkovo with all others ABS_R_...
I have done tests for normality and only RR_SRK2 seems to be normal (p>0,05, Shapiro wilk). Which test should I use to compare these groups? Mann-whitney?
RR_ Histograms dont look so much different from normal distrubution to me.
Do you think I could use parametric (t-test??) for RR_ groups?
I would be very grateful for your help.
Best regards
Ivajlo Popov
 

Attachments

Last edited:

obh

Active Member
#2
Hi Ivipopi,

Since the number of subjects is 30 and the histogram is mildly skewed. the normality is not a problem.
So you can run a parametric test.

What is the difference in the formulas?
Now, what is significantly different? when you compare sample data you may say it is significantly different.
So is you e question if the mean will be different on the entire population?
or maybe you can say that different formula is different?

When you use the same subjects it is a dependent test.

So you can run multiple comparisons paired tests with the Bonferroni correction.
Or repeated measured ANOVA with Tukey HSD.
 
#3
Hi Ivipopi,

Since the number of subjects is 30 and the histogram is mildly skewed. the normality is not a problem.
So you can run a parametric test.

What is the difference in the formulas?
Now, what is significantly different? when you compare sample data you may say it is significantly different.
So is you e question if the mean will be different on the entire population?
or maybe you can say that different formula is different?

When you use the same subjects it is a dependent test.

So you can run multiple comparisons paired tests with the Bonferroni correction.
Or repeated measured ANOVA with Tukey HSD.
Thank you very much,
formulas calculte the same (dioptric power of the intraocular lens which will be implanted in the eye after cataract surgery). Each formula use biometric data of the eye. So I want to find out which formula is more precise.
As you said, on my sample would be OK to use paired T-test, yes?

Thank you again
You helped me a lot
 

obh

Active Member
#4
So after the surgery, you know the real value of the lens and you compare each of the 5 to the real value?

Don't forget to correct the significance level: for example for the significance level of 0.05: corrected-α=1-Root(α=0.05,n=5)=0.010206
But I probably it is better to go to the Tukey HSD.
 
Last edited: