Statistics

B

Hi all,

Does anyone have any suggestions on how to investigate mediation over time? I have basic statistical knowlegde and im really struggling. The research question is:

Does fatigue mediate the relationship between drug treatment and cognitive functioning over time.

Any help would be greatly appreciated

:-(

B

Hi there, I'm a bit confused about what the question is? Are you asking for help designing a study to investigate this, or do you have the data but you don't know what to do with it?

B.xx

B

Hi,

Ive got the data just not sure what to do with it - I seem to be going round in circles x

B

I'm guessing you have a measure of cog. func, a measure of fatigue and some indication of drug treatment, all taken at different points in time?

I'm a bit confused as to how your in this situation, as the design of your study should directly inform your stats? What does your sup. say?

B.xx

M

So your dependent variable is cognitive functioning, and you're looking for an interaction between drug treatment, fatigue and time?
If those are your 3 independent variables, how many levels do they have and what kind of variables are they (between subjects or within subjects)?
Sounds like it could be a factorial ANOVA, but you'd need to explore the data and check that it meets the assumptions.

M

Quote From bobby:

I'm guessing you have a measure of cog. func, a measure of fatigue and some indication of drug treatment, all taken at different points in time?

I'm a bit confused as to how your in this situation, as the design of your study should directly inform your stats? What does your sup. say?

B.xx

Ah of course, fatigue would be more likely to be a DV than a grouping variable. Bloody multivariate stats then :p

Buttercup - you should definitely discuss this with your supervisor, you could do some robust stats (bootstrap the data), regression as well cos I guess this is a clinical study where you'd want to be able to predict cognitive functioning. I reckon your supervisor would have an opinion on what would be best for your design.

S

======= Date Modified 09 Nov 2009 19:42:51 =======
Hello ButterCup,

Could you be a little clearer of what is the problem?

statistics is very tricky and one of the most popular questions on viva day, it is better to ask an expert.

I had problem with statistics, I asked someone who has been working on it for three years, but something told me his method is not correct. I asked someone else they told me a different story.

what I did, I sat down and tried to understand the problem as much as I can.

Then our university had a statistic course, and the tutor was willing to help with any research questions, so I emailed her, and she solved it for me so nicely.

am telling you all that to show what steps might help you, first : try to read some basic statistics ( I used prism software and I read its guide very carefully) secondly ask someone who is doing statistic in your area.

Read some basic statistics like t test, and ANOVA, I think if it is two variables you do a t test, if more you do other tests for larger populations.

Good luck

======= Date Modified 09 Nov 2009 20:42:04 =======
In my field, mediation is most often tested with regression. See http://davidakenny.net/cm/mediate.htm

I am not sure how you would include the 'over time' bit. it may be that you can run the regression twice (time 1, time 2) and then compare beta values.

A

Find a local friendly University statistician and buy them a huge mug of coffee and they are generally (in my experience) the nicest people ever.

A

p.s. - I'd use regression too.

B

As others have said here, based on the info given, my gut reaction is multiple regression. However, this sounds like too important an issue to leave it to this - you need to have a chat with your sup (or at worst, someone else in your lab) as they will have a clearer idea of what the norms are in your area.

I also second the ask a stats person! I've ALWAYS found them completely charming and helpful!!

Is there some reason why you haven't brought this up with your sup though?

B.xx

B

I haven't read all of the replies but.... if you are looking at something over time then it may be worth looking at Survival Analysis/Event History Analysis..... I am not a stats student, however, I am conducting some quantitative research as part of my PhD. I have been on a training course for the above - it will allow you to look at the data from a particular time to a particular event. Not sure if this will help or not.

B

Hi all,

Thank you so much for your help and suggestions. I have made contact with a statistician who is lovely. I hate statistics so much!

x