0
   

Alternative to Two-Way ANOVA for non-parametric data

Forums:
Email this Topic • Print this Page
 
 
Nat K
 
Reply Fri 21 Jan, 2022 09:06 am
Hello, I am currently using SPSS for my analysis and got stuck with looking for an alternative test for the Two-Way ANOVA for non-parametric data.

I would like to study the effects of two independant variables (both nominal) on a dependent (numeric) variable. For example, how does diagnosis (3 levels) and gender (2 levels) affect the biomarker A.
The collected data of biomarker A is skewed and therefore doesn't follow a normal distribution. Because of this I am not allowed to use the Two-Way ANOVA, is that correct?
Does this assumption of normal distribution still apply with a data set of 130-170 data points or is it neglectible?
I already tried doing Friedman's two-way ANOVA but got an error saying that the ''Friedman's Test can only be performed on two or more ordinal and continuous fields''.

I would be really grateful for suggestions and words of advice! Thank you!
  • Topic Stats
  • Top Replies
  • Link to this Topic
Type: Question • Score: 0 • Views: 306 • Replies: 5
No top replies

 
engineer
 
  1  
Reply Fri 21 Jan, 2022 09:51 am
@Nat K,
I think you should be fine with the ANOVA. You should not expect the biomarker data to be normal. Suppose you had a nominal variable with two conditions (A and B), each producing a normal distribution (around 1 and 2 respectively). When you combine those two distributions you do not get a normal distribution.
Nat K
 
  -1  
Reply Mon 24 Jan, 2022 06:03 am
@engineer,
Hi, thanks a lot for your reply!
Do you think that taking, for example, a Log10 of the data so it becomes (more) normally distributed would help?
I tried this with my biomarker and the distribution became quite normal.
Does this have an impact on the validity of the data set?
Many thanks!
engineer
 
  1  
Reply Mon 24 Jan, 2022 06:42 am
@Nat K,
You shouldn't expect the data to be normal, please don't try to force it. It's actually good that it is non-normal as it implies there is something in the dataset besides noise. After you run the model, it would be great if the residuals are normal, but the starting dataset should not be.

No, if you take the log of the data, you change the fundamental model in a way that doesn't make sense from what you describe.
Nat K
 
  1  
Reply Mon 24 Jan, 2022 08:34 am
@engineer,
I see, thanks again for your reply..!

I ran the Two-Way ANOVA on my skewed data set and tested the residuals for normality...they are also skewed (skewness =1.54, kurtosis =3.48).
What could I do/is there anything to be done about this in your opinion?

Thank you so much for helping out!
engineer
 
  1  
Reply Mon 24 Jan, 2022 10:27 am
@Nat K,
That means that you have not identified all the independent variables. Your model is missing something. This shouldn't be surprising if looking at medical data. It could be something like age, or income or gender. It doesn't invalidate your analysis. If it showed a significant difference between some of the parameters I would believe it.
0 Replies
 
 

 
  1. Forums
  2. » Alternative to Two-Way ANOVA for non-parametric data
Copyright © 2024 MadLab, LLC :: Terms of Service :: Privacy Policy :: Page generated in 0.03 seconds on 10/07/2024 at 04:22:34
▲ ▼