I want to evaluate the effect of a new product on yield components of three cultivars. Accordingly, I have six subjects (3 varieties and their respective controls). I am planning to have only two replications from each subject. I know three replications are the most common and acceptable. I am wondering if there is any impact on the quality of the result (ANOVA) if I take only two replications instead of three.
Thanks
In general, you will get problem with quality in the aspect of "power": Your low replication will it make it much more difficult to find any significant differences between treatments or between trmts and controls. The quantity of loss in power can be calculated, you find many software on the net for this. I guess, the step from 2 to 3 reps would larger than any other step, going up the ladder of increased replication.
I would advice in general to try get the common 3 reps, if possible.
The only effect of use two replications instead three is that you will need more difference in your treatments means for declare significance differences and the probability (p) of this significance is generally p
I agree with both Fredrik and Miguel. You should be most concerned with having sufficient power to detect a difference. If you take too few samples, you may have a type II error (not finding a statistically significant difference when there really is one). Thus, you may wrongly conclude that your product does not work.
If there is no limiting factor in collecting additional samples, then more is always better than less. On other hand, you may find that after two samples you find a statistically significant difference and thus additional sampling is unnecessary.
I think Miguel is saying the same thing as me, and p-values are indeed strongly dependent on replication. You may as well like to know the 'true' effect of a difference you find, w/o considering the influence of replication on p-values.
Then, I suggest to take a look at:
MAHER, J. M.; MARKEY, J. C.; EBERT-MAY, D. 2013: The other half of the story: effect size analysis in quantitative research. CBE-Life Sciences Education 12, 345-351.
Indeed, these postings are all related. I'll add one more small comment. Some disciplines require the reporting of confidence intervals in addition to (or instead of) p-values, specifically because they provide more information about the size of the effect than a nominal p value
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Modeling
- Analyses of Variance
- ANOVA