two different factors in groups being compared

#1
My aim was to assess the impact of "x" factor on disease progression. I divided patients into two groups: 1) possessing the x factor 2) not possessing. I compared the tempo of disease progression in both groups with Whitney Mann test (there was no normal distribution in both groups) and it showed significant difference p<0.001 between groups. However, it turned out that 50% of 1. group and only 20% of 2. group possessed another factor, let's call it "y", that is a well-established bad prognosis factor what actually might have impact on my result. How can I proof that the result of my study is not due to "y" factor.

I compared: x(+)y(+) and x(-)y(+) groups, and, x(+)y(-) and x(-)y(-) with p=0.04 for first, and p=0.0001 for second
 
#4
How large are your groups, and how exactely was the dependent variable measured?

With kind regards

Karabiner
200 for group one (with x factor) and 1100 for group two (without x factor)

I compared the decline rate of a scale that is used to assess patients with this disease = (how many points they lost / disease duration)
as, for example, if sb lost 1 point in 5 months time, his decline rate is 0,2, or if sb lost 5 points in 6 months, his decline rate is 0,83,
 

Karabiner

TS Contributor
#5
If the dependent variable can be considered as interval scaled, then an
analysis of variance with the factors "x" and "y" can be performed,
(without interaction). Regarding the normal distribution stuff, you
could search this forum or other sources why this is not important,
given your sample size.

With kind regards

Karabiner
 
Last edited:
#6
If the dependent variable can be considered as interval scaled, then an
analysis of variance with the factors "x" and "y" can be performed,
(without interaction). Regarding the normal distribution stuff, you
could search this forum or other sources why this is not important,
given your sample size.

With kind regards

Karabiner
Kolmogorov-Smirnov test and Shapiro-Wilk test stands for 0.00001 value for both groups

so I assume I can't use two-way ANOVA test
 

Karabiner

TS Contributor
#7
You do not need to shout. Your p values are irrelevant. Your assumption is wrong. If at all, then normality assumptions are important if total sample size is small (n <30).
 
#8
Sorry, I didn't want to shout. I just thought it looked better that way.
With two-way ANOVA I've got:
factor 1: 0.005
factor 2: 0.000
factor 1 * factor 2: 0.668
so I hope that's enough for a proof that factor 1 irrespective of factor 2 has an impact on disease prognosis.

edit
However, if I perform t-student test between groups x(+)y(+) vs x(-)y(+) p value is 0.057. What stays in opposition to two-way ANOVA result.
 
Last edited: