Which Statistical test to use? - (Jul/06/2014 )

Hi all, I have a question regarding what type of statistical test to use, I haven't done stats for many many years and need some help!

I'm currently working on a project which is assessing the best culture conditions to grow mesenchymal stem cells in. The experiment runs as follows:

We have 4 media conditions: 

1) DMEM with 5% FBS 

2) DMEM with 5% FBS + FGF 

3) DMEM with 10% FBS 

4) DMEM with 10% FBS + FGF. 

Each of these conditions are being tested for growth in: 

1) 21% oxygen 

2) 2% oxygen  

3) 2% oxygen enclosed environment 

We have 6 flasks in each set (except for the 10% FBS + FGF group being grown in the 2% oxygen group - we lost 2 of those to infection, and we lost all of the 10% flasks in the 2% oxygen enclosed environment) so in total we have 60 flasks.

The only data I have to work with is % estimated confluency for some of these flasks, % viability and colony counts. My question was, what sort of test do I need to do on these? I'm assuming it would be chi-squared as I'm testing for a difference between counts, but because I have so many groups would I go for an ANOVA test?? I haven't done this in so long and any help would be much appreciated!

We're seeing what effect different concentrations of FBS have, what effect FGF (or absence thereof) has and finally which oxygen conditions are best for growth. Also is anyone using the program Prism 6? I'm just trying to get to grips with it and can't decide which table to use, currently it's between using 'Grouped Tables' and 'Contingency Tables'

Many thanks in advance!

MANOVA perhaps if the data are normal distributed (and the other data requirements of such a test)

Thanks for your reply! I'm having real trouble with this normalisation business! I tried to do a Shapiro Wilk normality test but it said n was too small (n=6) does this mean I should still normalise my data or just use a non-parametric test?

Kudu97 on Mon Jul 7 20:59:56 2014 said:

Thanks for your reply! I'm having real trouble with this normalisation business! I tried to do a Shapiro Wilk normality test but it said n was too small (n=6) does this mean I should still normalise my data or just use a non-parametric test?

Well Anovas are quite robust, but actually I'd use boxplots to see if your data are skewed or not...and if you can transform them with several possibilities which depend on type of data and the problem they have (have a look in a good statistics book or google it, it is quite well explained).

And yes a non-parametric test is also possible, though not sure which for this setup.

The most sophisticated approach would be to use GLM where you can modify the basic distribution of the data and which factors you use for model building, therefore there would be no need for data transformation, normalisation etc.