Help with choosing right statisical method.

Hi I am new at statistic and am writing my bachelor thesis.
I am not really sure how to interpret the statistical significance of my final experiment.
I have ELISA readings from supernatants derived from culture plate wells with no hybridoma growth. I need to compare my control with readings from a group(1) of supernantants(non-specific antibodies) derived from hybridoma wells - and a group(2) of supernantants(specific antibodies) derived from hybridoma wells. I need to establish wether the group we think contains different amounts of specific antibodies really is specific.

Until now I have tried different t-tests - but my antibody groups are not really following gaussian distribuation. So I don't think that is valid. Controls do pass the normality test - but are both skew and kurtotic.
Have also tried non-parametric tests as Mann-Whitney - but in all comparisons I get unbelievable low P-values : P<0.0001 even though my control an 1 are close in values. I use Graphpad Prism for the calculations.


Less is more. Stay pure. Stay poor.
Can you determine the median for the full sample (both groups) then in a bar graph with two bars (group 1 and group 2) graph the proportion of group 1 values above and below the median and the same thing for group 2. Maybe don't need to do the graph but it helps in understanding the Mann-Whitney.

Also, how are you testing the normality of the dependent variable using the full dataset, and how are you specifically deciding it is not normally distributed? Also, what is your sample size?
If I do that with group 1 and 2. Then the mean of group 2 is lower than the total mean and the mean of group 1 is higher. But how does that relate to my control?
No, I did'nt do that. I am not sure if the control should be part of that. My control group is 40 wells, and I have 228 hybridoma wells.


Less is more. Stay pure. Stay poor.
Your dependent variable passes normality test (not sure which test) and you have > 30 observations in both groups.

That seems to be enough to move forward with the ttest. What reservations do you have? Normality tests the residuals in the dependent variable, so if the histogram, etc for the variable's raw data do not seems gaussian, you may still be fine, since that is not exactly what you should be concerned with.