I am analyzing some data which consist of cell counts per length of tissue that express a protein and I have a concern about zero values in my data set. The zeroes are a result of the loss of a cell type during development and thus do not have the protein; they are not a result of my treatments.
For example, I have 10+/- SEM cells in group A, 12 +/- SEM cells/length in group B, and 0 cells/length in group C. Because these are normalized based on the length of the tissue, they are not true counts from my understanding. ANOVA analysis doesn't seem appropriate due to the heterogeneity issue (group C has no variation). I'm assuming that non-parametric analysis might be the best option.
However, if this is a loss of that cell type, is including group C even appropriate/relevant to the statistical analysis? In my opinion, this comes down to the hypothesis, which is the number of cells that express this protein at the different developmental stages. This leads to me to ask if the values of group C are truly 0 or are they "no data/n.d." because those cells simply don't exist at that stage. I lean toward considering group C as truly 0 and doing non-parametric analysis. Feedback of my thought process and outcome would be appreciated - thank you!
1 n.e. 0, 0 is n.e. doesn't exist. none of this has to do with nonparametric statistics.. now you grab the ball and run with it. Best wishes David Booth
I don't entirely understand, but it sounds like you don't have data for Group C. It may be best to document this fact in the results, and continue with the analysis of A and B.
The easy approach is to drop Group C and use a standard method to analyze the remaining data. You may use parametric or nonparametric statistics as appropriate, but that is a different issue unrelated to the outcome in Group C.
A harder approach is to assume that the sample size in Group C was insufficient to detect some rare event. Group C is not a true zero, but the best estimate of 0.0000000005. You might need more data or at least discuss how this might influence the interpretation of your data. You could address the problem through simulation: analyze the outcome of 100,000 simulated experiments (of your original sample size) and discuss how rare events would influence a specific research outcome.
- Badges
- Science topic