Hi,
I'd love some statistics expertise. I have completed an experiment using mice. I have four groups in total: 1. Control non-knockout 2. Control knock out 3. Treatment non-knockout 4. Treatment knockout. I have analyzed for 5 different cytokines in 3 different body tissues. I initially performed two-way ANOVAs on my data (treatment x knockout) for each tissue type and cytokine. I think that maybe a 3-way anova (treatment x knockout x tissue type), or perhaps a MANOVA may be better. The only catch is for one of the groups my N value is 3, while all other groups have N value of 8. Having an N = 3 is inhibiting me from performing a MANOVA without breaking the rules. I do not have access to more mice, and therefore I cannot increase my sample numbers. Can someone help by suggesting the best or most appropriate statistical test(s) to analyze my data? Thank you!
MANOVA is different and in order to perform it you need to have multiple dependent variables. You seem to have one dependent variable, though you did not specify.
So first question what is your dependent variable(s)? If you have only one dependent variable, then you could use ANOVA. In order to find number of factors to be used in ANOVA, we should make clear them. So you have a treatment factor which has 2 levels control and treatment. I guess you have a knockout factor with 2 levels knockout and nonknockout. You have a tissue type factor with 3 levels. Do you have also a cytokines factor? If so you would have 4way ANOVA. A factor in ANOVA is a blocking variable.
ANOVA or MANOVA might depend on equal cell sizes which is absent in your case (you have cells with 3 observations while others have 8). You had better increase the number of subjects in this group to 8 or near 8. You can also try using a mixed effects model if your dependent variable is unique.
If your response is "concentration" (of 5 different cytokines) make sure to model the log-concentrations.
If the log conc of the 5 cytokines are well correlated it might be simpler to model the "average cytokine response" rather than each cytokine individually. If this makes no sense, you can go for MANOVA, but to dissect the individuals cyctokines you will anyway have to go back to the models for the individual cytokines.
A 3-factorial model (genotype x treatment x tissue), to my opnion, would be better than 3 individual (genotype x treatment)-models (one per tissue type).
I don't see that a mixedmodel would be an option. The only possible random factor could be the tissue type (as I see this), but here you have only 3 different types what is too little to make modelling as random factor advantagous.
Thank you for your responses.
Mehmet - to clarify for you the cytokines values are my measured responses.
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Modeling
- Analyses of Variance
- ANOVA