3 cohorts in the study control (n=19), pre-cancer lesion (n=21), cancer (n=9), and looking at a number of cells in the blood detected via GEDI.
I do not know if the Mann-Whitney U test is appropriate or to use of Welch's t-test, as the sample size between the cohorts varies.
If the cell counts are "large" (>100) and the standard deviation is comparably small, then an ANOVA model is ok (make sure to estimate the variance pooled over all three groups).
If the cell counts are "large" and the variance correlated with the mean, consider analyzing the log-counts instead or use a count model (see below).
If the cell counts are "small", better use a count model. This could be a quasi-Poisson or a negative binomial model.
Hello
Your information about the experimental design and data distribution is incomplete. The decision to which model and which statistical test is needed depends on the assumptions that you did not explain to us. However, my friend Jochen Wilhelm's comments are approved by me
Hi,
Try one-way ANOVA and notice if the standard errors are low on not. If they are large the Non-parametric ANOVA (Kruskal Wallis) can be tried.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Sampling Design
- Sample Size