Hi everyone!
I investigate the result of knockout of a particular gene on the disease severity. I take WT or KO mice, induce disease, and record readings. So, the genotype is the experimental factor.
I cannot use too many mice each time (just 3-4 per group; the disease induction is tricky and laborious) and the model establishment has a considerable variance itself, so I need to replicate this experiment several times. I have performed 2 experiments already and each experiment gives a clear trend (the disease severity in KO is lower), but p-value is above 0.05 in each individual experiment. Pooling 2 experiments together is not reasonable because there is a clear between-experiment difference (there is a significant impact of the experimental procedures (cell preparation, machine settings, etc).
I`d like to account for the between-experiment variance.
As I know, making "Experiment" as another between-subject factor is not reasonable as it is clearly a random factor.
Making nested design ANOVA (1. subject (mice) nested within a combination of experiment and genotype, 2. experiment, and 3. genotype with 1. and 2. as random factors) would be quite good, I think, but it is quite tricky as well. I use STATISTICA (Statsoft) package. It can do this kind of analysis, but I am not sure about myself :) Also, as I am not a big specialist, it would be difficult to explain this method to my supervisor.
Is it possible to use a somewhat simplified model? For example, a design with experiment and genotype as 2 separated factors without interaction with each other (like a 2-way ANOVA, but interaction and error terms are combined). So, I will see effects of the genotype (factor A) and experiment (factor B) only. Like, X = A + B + error without AB term.
Is it reasonable?
Your approach sounds resonable but in your case I would recommend checking out multilevel models because you already identified random terms and multilevel modelling accounts for these.
Both approaches can be accomplished using R programming language and there are many tutorials (I prefer Bodo Winter's).
The whole point of ANOVA is to test variance between experiments, so put all the data in one big file and run one ANOVA on it, assuming you have checked for normality. Do not have separate ANOVAS. Mice is not factor (it's the subject), so cannot be nested, but experiment is a fixed factor and genotype is the independent response variable. A simple one way ANOVA will probably work :)
Ah, Andrew Paul McKenzie Pegman , but the genotype is an independent variable itself and the dependent variables are: cell quantity and subpopulations and disease severity.
So, basically, the suggested approach is reasonable, right?
Okay. But still have one data file and perform analysis on that to test all interactions
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Modeling
- Analyses of Variance
- ANOVA