Help decide whether I used the right tests

#1
Hi, I have been doing some statistical tests which I'm a little bit out of my depth on for work. The person I'm doing them for knows less than me, and I don't want to screw this up, because no one would ever correct me or her!

Ok, I work in a microbiology laboratory, and the tests I've run were in reference to the occurrence of some cellular proteins, and whether or not they are good indicators of certain things regarding cancer like whether or not the patient survived, length of survival, subsequent cancer, etc.

On to the data: All the tests that I ran were using non parametric tests. Do I use the Chi-square test to see if this is the case? Do I understand that if you get a successful result in the chi square test then your data are not parametric? In any case, I also ran non parametric tests because I've been told to look over a paper that is similar to this one (just testing different variables), and they ran non parametric tests like the Mann Whitney, Spearmans correlation, chi square, Kaplain Meier test etc.
If the chi-square does tell you whether your data is parametric, then most of my variables are (<.05). Am I to understand something would be non parametric if, let's say it had 3 possible values; 0, 1, and 2, indicating three levels of intensity, and it had 12 occurrences of 0, 3 occurrences of 1, and 1 occurrence of 2? In any case, if I am to compare values in where one is parametric and the other is non parametric...what is a more reliable test, spearmans rho or pearsons r?
Originally the investigator wanted me to use multivariate analysis, which I did, but if my data are nonparametric, then I am not able to do this, correct? The reason she wanted multivariate is because some of the dependent variables might exhibit some causation amongst themselves, such as subsequent cancer and length of life. I ended up using the Kruskal Wallis test instead.
I've run many different tests on the data, and a lot of them give very similar findings, which I assume is proper, but some of them do change the findings enough to achieve statistical significance in a certain direction...I want to make sure I can help her find what can and can not be said about the data.

Thank you!

Btw all tests were done using SPSS 17.
 
#2
Hello,
first of all you can not say that data are non parametric. Only the test can be said so. Most of the time you will use a non parametric test when the condition of a parametric test are not satisfied as for example normality of the data. For example the Kruskall Wallis test can be used if the data are not normal (tested with the Shapiro-Wilk test for example) or the variance of the different groups are not equals (Levene test on SPSS) when wanting to do an ANOVA.
It is not clear to see from your request what you are attempting to do and so which test can be used.
 
#3
Hello,
first of all you can not say that data are non parametric. Only the test can be said so. Most of the time you will use a non parametric test when the condition of a parametric test are not satisfied as for example normality of the data. For example the Kruskall Wallis test can be used if the data are not normal (tested with the Shapiro-Wilk test for example) or the variance of the different groups are not equals (Levene test on SPSS) when wanting to do an ANOVA.
It is not clear to see from your request what you are attempting to do and so which test can be used.
How am I sure if the data are normalized?
What am I attempting to do?
I have discrete interval data corresponding to the perceived amount a protein shows up in particular cancerous cells. It is labeled as intensity, and is moderately subjective (the intensity of the light is not measured by machine, but by the user's eye), and scores a 0-3. That persons case is then matched up with other factors, like whether or not they survived, whether they had subsequent cancer, what the length of their survival was, etc. I am trying to find whether or not that intensity of that protein is a good predictor (or at least correlated with) any of those dependent variables. Now, some of those dependent variables affect each other. Clearly someone who has subsequent cancer is more likely to die. I have been asked to do survival plots, find correlations, and basically any other test that can be made to tell us something interesting about the data.

More info:

For instance, if I want to find if the variables are correlated with each other, can I just run a Pearsons correlation between the two variables, then I know? And if they are, I should maybe run a multivariate test?

Also, I ran a Kruskal Wallis test which gives me a chi-square value between the protein and the length of time lived for the individuals, asymp. significance at .058. In the Jonckheere-Terpstra test, it gives the asymp significance at .346. How do I know which test to use? Obviously I would like to use the Kruskal Wallis because it supports the project, but I don't want to use it incorrectly.

Thanks.
 
Last edited: