Chi-square post hoc tests?

#1
Hi there!

So I've been testing a drug to see if it had any effect on the presence of a specific chromosome abnormality. There were 4 drug concentrations so I ran a chi-square of the 4 concentrations against 'number of cells with abnormality vs number of normal cells'. The test showed significance, so concentration had an effect on the number of abnormalities.

Now I want to compare between the groups and see if one had statistically more abnormalities. When I plot the percentage of abnormalities per concentration on a graph there is one group that has almost double the abnormalities! I ran a chi-square pairwise comparison and all the other concentrations showed significance with this particular concentration. Is this enough to tell me that this concentration caused a higher proportion of abnormalities compared to the others? Or is there another test I should run?
The information I have access to is: each concentration was analysed at 4 different time points and I have been given the total number of cells analysed and the number of abnormal cells found at each time point - just in case this changes what tests I can do!

Thanks for any advice :)
 

rogojel

TS Contributor
#2
hi,
was it the same sample that was measured four times for each concentration or separate samples? If the former, you should look at longitudinal studies.

I think you have a lot more information here then and so, better analysis choices then what you can achieve with a chi-squared - like interaction effects between concentration and time, you just need to pick the right method.

regards
 
#3
Yes it was the same sample measured every two weeks. Ah ok so you mean look and see if the number of abnormalities increases with time with each drug concentration? Do you have any suggestions of methods to do this? (I'm completely new to statistics so I don't understand a lot yet!).
 

rogojel

TS Contributor
#4
Yes, exactly.
I would do a graphical exploration first and see if there is some indication of an effect, then look at statistical tests.

regards