Hello! Here is a brief overview of my experiment and problem.
I am doing Propidium Iodine staining on organotypic brain slice cultures to test effect of alcohol and inhibitors on cell death. We take fluorescent pictures, and then using threshold tool in ImageJ, threshold based on the slice with the most staining (most dead) and then get the percent area of of the remaining slices with pixels that fall in or above that threshold value.
My problem exists in that comparing between multiple experiments (i.e. multiple preparations of slices/staining/imaging) results in the need for different threshold values. Alternatively, if I keep the threshold values the same the variability in %area within threshold is extremely high (which is understandable as everything was done on a different day). HOWEVER, the same trend between treatment groups in each individual experiment is clearly evident (and if I change the threshold on each day it works). But, I do need to do statistics on this data and am unsure how to to best represent the data to do this and to show that the trend exists. Thoughts?
I am not aware of the past literature in your research area. Nonetheless, these are my thoughts.
1. Ideally, such a threshold may be selected based on an a priori criterion from past literature.
2. If such a criterion is not available, and you have to use different thresholds for different days/slices/chemicals/etc., you may consider adding the threshold value as a covariate in your statistical analysis. This may be possible with an ANCOVA or linear regression model.
By entering the threshold as a covariate, you will be setting up the statistical model to control for the variance in the outcome measure added by the varying threshold across days/etc.
Hope it is helpful.
Hello,
adding to what Usman Rashid mentioned, I think ANOVA, ANCOVA analysis of variance is adequate for this type of data, you can create a contingency table in which you put the group (type of treatment), the day, the experience performed and the dependent variable "percentage". this statistical method could give you answers about the variability between groups, experiences and evolution over time.
Best wishes
I think this could be addressed with a regression model (of which ANOVA and ANCOVA are just special cases). Potentially you could treat this as a multilevel regression model - treating variation experiment as level 2 with experiments nested within that at level 1. This would allow you to model random variation between experiments separately from error within experiments and have predictors at either level (or both). I think you'd want fair few experiments say 20-30 to go down the multilevel route. With smaller you can model between experiment variation with dummy variables (though that's less efficient in general).
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS