======= Date Modified 04 05 2009 15:05:23 =======
Hi everyone,

I'm stuck in statistics hell and am hoping one of you clever people can help me get out of it (and yes that was shameless flattery ;-) )

basically, i am attempting to compare two groups of data that look a bit like this:

drug observed value predicted value
drugA 3 3.2
drugB 6 7.8
drugC 25 20 etc etc

basically, my sup has suggested i use a t test, but being a complete stats numpty, i have no idea which one to use! am i right in thinking that because the data is related in some way (i.e. there are observed and predicted values for each drug individually) that i need a paired t test? i am confused, because although the data is related in this way, it is also unrelated, e. the observed value has no influence on the predicted one, if that makes sense! (not like say taking someone's blood pressure before and after an event, where the two measurements are taken from the same person)

hope someone can clear this up for me....

thanks! :-)

The right test will depend on many things:

What exactly do you want to compare- differences among the drugs or between the observed and predicted?
Where did you get the predicted values from? Are they from a statistical model?
Are these means or raw values?
How many drugs are there?
How many individuals within each drug?
What is the distribution of the values.

The paired t-test (or any t-test) assumes that the values are normally distributed and should only be used if you have at least 30 pairs of observations. Alternatively, the Mann-Whitney test could be used, which doesn't require those assumptions.

May the force be with you!