Which test to use

KateB

New Member
Hi!
I am making a small scientific research and would really appreciate some home with statistics.
Excel file is in the attachment.
I am studying the expression of two similar markers in carcinomas depending on grade (column „grade”). There are 6 different grades GG0-GG5. The markers are FM and SM, the expression values are in %. The correlation coefficient between SM and FM is 0,8, the expression is pretty similar, but FM theoretically should be slightly better. What statistic should I use to prove that one marker is better than another meaning correlation between marker expression and GG is better? I know that the word „better” is not very scientific The solution should be quite simple, but I failed to find it by myself. I have stata and R
Apologies for my English.
Thank You in advanced.

Attachments

• 7.8 KB Views: 2

obh

Active Member
Do you want to see the best combination of FM and SM as predictors to grade, like Grade = 5.855577 - 0.121474 FM - 0.0235641 SM?
FM pval=0.00000
SM pval=0.00252555
like linear regression or better ordinal logistic regression?

Do you want to prove that FM is better than SM as a predictor for the grade? you may run two models and see which one gets better result

Last edited:

KateB

New Member
Hi!
I want to prove, that "FM is better than SM as a predictor for the grade"

obh

Active Member
You may run two ordinal logistic regression models and see which one support a better result

obh

Active Member
I can see both markers are significance with linear regression but the FM had lower p-value as you expected.
I assume you will have a similar result with ordinal regression, but you should try.
linear regression is okay but ordinal is better for your case.

obh

Active Member
Yes, the link is correct, shouldn't be too complex The values were for linear regression with 2 predictors, but you get similar results (say better for FM) with one linear regression model per predictor.
So probably similar results also for ordinal, but you should try

KateB

New Member
I will try it!
Final questions:
What software would You advise?
I do not understand what are these values?
Grade = 5.855577 - 0.121474 FM - 0.0235641 SM?
Thank You very much!  Karabiner

TS Contributor
It seems like you want to statisticaly compare "correlated" correlation coefficients.
You could use correlated correlations coefficients as a search string in a web search.
An online calculator (with references) for such analyses you can find here.

With kind regards

Karabiner

KateB

New Member
It seems like you want to statisticaly compare "correlated" correlation coefficients.
You could use correlated correlations coefficients as a search string in a web search.
An online calculator (with references) for such analyses you can find here.

With kind regards

Karabiner
Hi!
Although the example in Your link seems quite simple, I had some issues.
As I understand, the step by step solution would be:
1) correlation between GG and FM
2) correlation between GG and SM
3) use the calculator.
As I have only basic knowledge in statistics, I have some questions GG is a discrete variable, with values 0,1,2,3,4,5. How do I correlate it with %?
Where do I get r3 for the calculator?
My apologies for these primitive questions.

obh

Active Member
I will try it!
Final questions:
What software would You advise?
I do not understand what are these values?
Hi Kate,

I just transform the Grade to numbers (GG0=>0, GG1=>1 etc ....) and ran multiple regression with an online calculator.
I suggested running ordinal regression on each mark separately and see which one will give a better result, but in statistics, there is not only "better result" you should also prove it is significant and not just a random result, and that what @Karabiner suggested. So please do as @Karabiner suggested.

Cheers