Hi,
I am trying to identify (if any) statistically significant differences between surface marker expression of two unpaired samples performed in replicates of five.
It is expressed as %parent of CD19+ cells into various subgroups and is also split into a CD45+/- group.The data is not, nor do I believe it should be normalised. Even if so, I have had a look and cannot find a way to make it normalised. I have some populations appearing as 0% expression so cannot transform the data traditionally without losing potentially important elements.
I split the CD45 expression as it is very approximately 50/50 in my samples though varies between samples and want to see if it impacts my results.
Should I try a parametric test? I noticed when I did a proportional bar graph that there is quite a bit of heterogeneity across my replicates. TIA
If the data is non-normal you should use a non parametric test. from your description, you could use the Wilcox signed rank test
https://en.m.wikipedia.org/wiki/Wilcoxon_signed-rank_test
DATA of two non-normalized samples may be tested for their significant difference under F test. Recall the general rule that one non-normalized sample is tested under chi square test; two non-normal or two chi square distributed samples may be tested under F test.
Your data is expressed in %. That is not a problem. Treat the percentage values, i.e. 0.10, 0.20, ..., 0.99, as quantitative values capable of being subjected to mathematical calculation.
F = var1 / var2
Use the large value of the variance as numerator. Find the significance level by reading the degree of freedom for sample 1 and cross with the degree of freedom of sample 2 in the F table. Compare the result of the above calculation F(obs) to the value from the table F(*).
If F(obs) = > F(*) then significant. If F(obs) < F(*) then not significant.
Le recomiendo utilizar la prueba no parámetricas a partir de su descripción específicamente la prueba de rango con signo de Wilconxon.