Which test should I use?

c33n

New Member
#1
Hello, I'm writing my thesis and I'm clueless about which statistic test I need to apply ( is it T-test, Chi-text, Anova …), I will do my ststistics on SPSS.

In my thesis I have 2 sets (=2 groups) of patients with the same disease, one group gets a specific treatment , while the second group is without that treatment. After a period of time some of the patients have developed a certain complication. Now I need to compare the 2 groups and see if the treatment lowered the risk of that complication appearance -with statistical significance.

Thank you very much!
:eek:
 

Karabiner

TS Contributor
#2
You have 2 categorical variables (Group, and Complication yes/no), so you could use crosstabs with Chi-square test.

With kind regards

K.
 

c33n

New Member
#3
Hello again and thank you.
I have yet another question.

If I have 2 categorical variables - let's say : Group/ Treatment

But now the treatment options is not yes/no.
1. Let's say treatment has 3 or 4 different options ( for example- aspirin / paracetamol / naproxen...).
2. Let's say it has 3 or 4 different options- but the variable is a range (>80% / 60%-80% / <80%).

Which test should I apply (using SPSS)?

Thank you, Chen.
 
Last edited:

hlsmith

Not a robit
#4
As for your first question, if you have a small sample with expected low counts in some of the cells in the two by two table, then Fishers exact test may be appropriate. Though you are better off using logistic regression for all of these questions. It will let you correct for small sample size if needed as well as other variables.

How were treatments assigned? If not randomization then you run the risk of results being confounded, and you would want to control for known observed onfounders and speculated effects of unobserved confounders.
 

Karabiner

TS Contributor
#5
2. Let's say it has 3 or 4 different options- but the variable is a range (>80% / 60%-80% / <80%).
This is seemingly variable on an ordinal scale. With 4-5 groups and an ordinal dependent variable, Kruskal-Wallis H-Test could be considered.

With kind regards

Karabiner
 

c33n

New Member
#6
Hello , thank you for your answer.
For the first answer I've used Chi-square.
Well the treatment wasn't assigned by randomization.
I got a list of a hospital's patients (who are being treated there for a certain disease). In total I have 109 patients.
Those patients were followed up, for different information ( renal injury, pulmonary injury, different treatments, disease duration…).
My work is about digital ulcer at those patients.
Every different info I've looked for is being checked separately if it's causative for digital ulcers (does ulcer is more common in those with a renal injury ? Etc.)
But sometime the info is not just – Injury: YES/NO
It has 3-4 options:
For example- Injury : small/ medium/ sever/ very sever. - and Chi test in not suitable.

I hope you understad...

Thank you!
Chen