Hi there,

I would appreciate some expert advice on how to best combine 3 replicated experiments into one analysis.

Im trying to check if a treatment effects growth of a plant Lemna minor. I have a highly controlled environment, but because of a space limitations were only able to use 3 replicates of each treatment (control vs. treated). Therefore I repeated the experiment 3 times to gain more data and statistical power.

I get the same effect (trend) in each experiment (inhibition of growth in treated sample), but the difference is significant in only one experiment. I know, that by repeating the effect 3 times, p value overall should change, Im just not sure, what is the best statistical procedure to do this.

I tried proceeding as follows:
Combine all the measurements with added category experiment number.
Run a Univariate GLM analysis:
- dependent variable growth
- fixed factor: treatment
- random factor: experiment
- interaction: treatment*experiment

Since the interaction was not significant; I excluded it from the model and repeated the analysis. In new model random factor - experiment was not significant; therefore I excluded it from the model as well.

At the end I get a simple model with only fixed factor = treatment and a highly significant effect on growth, that explains only 30% of the observed variation.

Is this a correct procedure or should I use a better approach?

Many thanks for your advice.