63

Beware. And again, I say: beware.

No website knows what your research question is. So only you will be able to match your analysis to the research question. Years ago, I did a paper on this. A researcher came to me with data on patients attending for in-vitro fertilisation. She wanted to know the correlation between their age and the chances of a successful outcome. What test would you use?

Well, you can do a correlation. But it tells you nothing.

You can also compare the ages of those who did and did not become pregnant using a t-test. It tells you that those who became pregnant were younger.

You can use logistic regression to calculate the odds ratio per 1-year increase in age. It also tells you that the chances are falling as you get older.

But what the researcher really wanted, when we talked about it, was to be able to advise couples on the chances of success. So what she needed was something easily calcualated from logistic regression: the probability of success at age 25, 30, 35, 40 and 45.

So which was the right statistical procedure? No website could have told me. All of them were possible, but only one of them answered the researcher's question. And I needed to get her to articulate the study question clearly before I knew which one to use.

So go back and state your study question. The stats will fall into place.

I recommend this  website,  which is a part of Interactive Statistical Calculations Pages :

http://statpages.org/#WhichAnalysis

Thanks so much! I'll immediately go to check for that. 

This could be helpful:

http://books.google.de/books?hl=de&lr=&id=kBUBRh1AWG4C&oi=fnd&pg=PR6&ots=Mho2KfjGxk&sig=H3lonaT81-obRLDmuzuLCJRxXwg&redir_esc=y#v=onepage&q&f=false

63

Beware. And again, I say: beware.

No website knows what your research question is. So only you will be able to match your analysis to the research question. Years ago, I did a paper on this. A researcher came to me with data on patients attending for in-vitro fertilisation. She wanted to know the correlation between their age and the chances of a successful outcome. What test would you use?

Well, you can do a correlation. But it tells you nothing.

You can also compare the ages of those who did and did not become pregnant using a t-test. It tells you that those who became pregnant were younger.

You can use logistic regression to calculate the odds ratio per 1-year increase in age. It also tells you that the chances are falling as you get older.

But what the researcher really wanted, when we talked about it, was to be able to advise couples on the chances of success. So what she needed was something easily calcualated from logistic regression: the probability of success at age 25, 30, 35, 40 and 45.

So which was the right statistical procedure? No website could have told me. All of them were possible, but only one of them answered the researcher's question. And I needed to get her to articulate the study question clearly before I knew which one to use.

So go back and state your study question. The stats will fall into place.

Hi Gippo,

I have a table that i give to my students, it helps if you know a priori exactly the types and number of predictor and response variables. It is in Portuguese but I am sure you can understand. Let me know if you have any questions

Ronan Michael Conroy has it nailed.  Statistics is not just a science, but an art, and I am very leery of non-statisticians using a "decision tree" to decide between statistical procedures that they do not understand, especially when they struggle to articulate their own hypothesis.

I do not doubt that the people who have assembled some of these helpful tools know what they are doing and mean well.  I do fear people who don't understand basic terms trying to use the decision tree un-aided and ending up taking the wrong path.

Thanks everyone! I'll keep on studying... ;-D

Thanks Sofia!

It is always advisable to consult a statistician, however, you can take some questions on the UCLA website, in the book: "Motulsky H. Intuitive Biostatistics Oxford University Press Inc in 1995.",  etc..

As a general resource, I'd recommend the site mentioned by Sofia Cividini.

A somewhat simpler, though maybe not as clear, site is:

http://www.biostathandbook.com/testchoice.html