Just briefly:
I have WB quantifications from 4 different blots (4 biological repilcates) and I normalized my signal intensity values to the untreated ctrl sample. Therefore, for all 4 replicates the value of the ctrl is 1.0, while the other sample values are numeric values. I want to compare several groups, so I can't use simple t-test. I want to use ANOVA with an appropriate post-hoc test, but I read in some discussions that using normalized data in an ANOVA violates statistical assumptions and is therefore not possible. I tried the analysis anyways with PRISM and it's all fine except the comparison of one of my values with the ctrl (all other comparisons are between values), which gives of course a rediculously high significance (p
That's the wrong question (or the answer is: don't do that).
The right question would be: how do I analyze data with a correlated error structure? (if you take the data from all blots, values will vary a lot, and a large part of that variation is due to the differences between the blots; the values within a blot will thereby tend to vary together)
You can use a hierarchical (mixed) model using "blot ID" as a random factor, or (if your design is balanced across blots) a model including "blot ID" as a fixed factor (for two groups, this would be the equivalent to a paired analysis, with pairing of the values within each blot).
PS: Make sure to use log band densities as your response variable.
That's the wrong question (or the answer is: don't do that).
The right question would be: how do I analyze data with a correlated error structure? (if you take the data from all blots, values will vary a lot, and a large part of that variation is due to the differences between the blots; the values within a blot will thereby tend to vary together)
You can use a hierarchical (mixed) model using "blot ID" as a random factor, or (if your design is balanced across blots) a model including "blot ID" as a fixed factor (for two groups, this would be the equivalent to a paired analysis, with pairing of the values within each blot).
PS: Make sure to use log band densities as your response variable.
You have complicated a simple subject
Why did you normalize the data?
It was not necessary to normalize the data
Prism software analyzes both for normal data (parametric tests) and for abnormal data (non parametric tests).
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Modeling
- Analyses of Variance
- ANOVA