Which test do I use? Urgent thesis help needed!

#1
Hi all,

I have decided to register to Talk Stats, as I am experiencing quite some stress writing my thesis, mainly because I'm struggling with statistics! I hope you can help me out with a small difficulty I am facing.

I have measured children's liking-rates of four different brands. These children were divided over four conditons (so, in each condition, four brands had to be rated). Now, I want to conduct a randomization check, and find out whether there are significant differences of liking-ratings of the four brands between the groups, but also within the groups. So, I want to know whether the four groups differ in their ratings of each brand, but I also want to know if the four brands differ significantly on liking-rates.

Several tests have crossed my mind, such as a Mixed Anova, and a MANOVA, but I am getting a little overwhelmed and blind to my own reasoning.

So, my question is: which test is most appropriate?

Thank you so much!
 
#2
Don't make it so complicated. Do a pairwise t-test for each of two brands (A versus B, A versus C, A versus D, B versus C, .... etc).

Make the pairing based on the childrens levels.

And cool down! :)
 
#3
Thank you so much! This has helped me a lot! It's not new for me to think too difficult about things, haha.

Another thing that I stumbled upon, is that I want to see whether two groups have different rates on multiple dependent categorical variables. I have measured whether they saw a specific brand in four images, and asked whether they have seen this brand after each image (yes/no). This leads to having four categorical variables. What test can be used for this? I was thinking about something like the MANOVA, but as dependent variables need to be continuous in this case, I should probably go for another test but I can't seem to figure out which one.

Thank you so much for your help!
 
#4
I was thinking about something like the MANOVA,
Again, complicated!

I guess that you want to compar brand A versus brand B etc.

You could calculate the proportion (say p_A) of all children that have seen brand A after one image. (I believe that you know how to compute the standard error for that.)

Then test if the proportions differ: p_A - p_B

And I guess you know how to get the standard error for that difference.

(It culd maybe be better with a logistic regression with mixed model for the child effect. But, don't make it too complicated.)