So i have two groups (treatment and control) and I image them and from each cell imaged, I assign particles within that image to three pre-defined categories (for example, if a vesicle is in the nucleus, in the endoplasmic reticulum or in the cytoplasm or in the mitochondria).
Thus overall, I end up with distributions or fractions (i.e. in cell 1 of control it is 25% 25% 50%, cell 2 it is 30%, 30%, 40% etc... cell 1 of treatment it is 0%, 5%, 95% etc...). What statistical test would be best to test wether the difference between control and treatment is significant or not.
Would greatly appreciate help!
I would suggest you compare the mean proportion of nucleus vesicles( across treatment cells ) with the mean proportion of nucleus vesicles(across control cells) by unpaired t test (f normallly enough distributed across cells.). Then do the same for proportions of endoplasm and proportions of mitochondrial vesicles.
Thus you do three separate unpaired t tests.
Because response within cell would be correlated, another alternative would be to model it as split-plot with cell as main plot and position (nucleus, endoplasmatic reticulum, etc) as subplot.
Thank you for your answers so far.
Unpaired t-test or a nonparametric test was my first thought also.
Regarding chi-square I am little bit hesitant as I do not know how that test would allow to account for differences between the individual cells (etc. cell 1 in control and cell 2 in control). Since for each cell there would be the distribution into those three categories and this distribution will vary from cell to cell, giving some indication of how much variability there is. If I performed cross tabulation and chi-square I guess I would have to add the individual values (ie number of vesicles in nucleus, er, mitochondria) from all the cells in control and treatment and in the end I would end up with three values for control and treatment. Within these values however there is no indication of how much it changes from one cell to another (ie no error bars) so I don't know how that would make sense.
I have attached imaginary example of this problem.
Hi, Veijo,
For a cross-tabulation, you should group your data into categories, for example 10 cells with 1-10% cover by vesicles, 20 cells with 10-20% etc. You have to do this for each cell category (control and experiment) separately and you end up with something like this:
size cat1 size cat 2 size cat3
experiment 20 10 5
control 10 30 4
The numbers in a crosstabulation have to be counts, especially if you want to use chi-square test. Please be aware, that you need at least 5 counts in the "expected" crosstab to apply chi-square (SPSS gives you a warning about this).. If you have less, Fishers test has to be used.
kind regards
Irene
- Badges
- Science topic