# At least, how many times an experiment should be replicated?

Is there a recommendation on the number of times that an experiment should be replicated? As many of you know, is not always possible to make many replicas. What would be the recommended minimum? Is there some references to support it?

In my particular case (animal reproduction), for reasons of seasonality, I can only replicate experiments 3 times and I have sometimes been criticized for the low number of replicates performed. Could be considered appropriate to assess the effect that a parameter measured 3 times in the same individuals have on the performance of these individuals?

• Do you have a three measurements on one individual or three measurements on three individuals? Nov 8, 2010 at 12:08
• I have three measurements in 20 individuals belonging to a small population of about 100 individuals. But my question seeks a more general rule. Csgillespie and you pointed out that the sample size is important when deciding whether or not the experiment should be replicated and how many times. However, it is not always possible to have a sufficiently large sample size, especially if you work with animals. Nov 8, 2010 at 13:06

There is no such thing as a minimum (or maximum) sample size rule. It depends on the size of the effect you are trying to measure. Your description of the experiment is slightly unclear, but consider this example, if you measured blood pressure in three different people, what could you conclude about blood pressure in the population?

Likewise, if you are conducting a clinical trial and it's clear (using statistical arguments) that one of the treatments is harmful, should you continue?

Another comment. In experiments concerning animals/people I would consider it unethical to conduct an experiment that has no chance of success due to low sample sizes. If in doubt, find a local friendly statistician. Most institutions have them somewhere.

• The size of population is important, for sure. Thus, for the blood pressure example three individuals may be unrepresentative. My case is different, there is no a problem of sample size. Imagine you have a sample of fixed size (it is not possible to include more individuals) and you want to assess the effects of a drug on blood pressure. Would you realize the experiment only once or several times? I think that it should be replicated at least two (or three) times in order to consider the individual variability of each subject. Nov 8, 2010 at 12:56
• @Manuel You are correct; you need to assess individual variability. The number of repeated measurements needed depends on the size of that variability, how the variability translates to uncertainty in the inferences about the population, the cost of repeating the measurements, practical constraints such as the time needed for replication, and (perhaps) technical issues like the possibility of positive temporal correlation among the replicate measurements.
– whuber
Nov 8, 2010 at 13:30
• So, the idea could be to replicate the experiment two or three times, evaluate the variability of the estimates and based on this variability decide whether more replicates will be necesary. Nov 8, 2010 at 13:39
• @Manuel: This is a very dangerous strategy and you need to be careful. You can't just carry out a few experiments and stop when you like. Basically you can't change the sample size midway through your experiment unless you are very careful. Nov 9, 2010 at 21:25
• I totally agree with you. We can not decide to stop an experiment because the results are what we wanted or otherwise continue. In my work usually replicate experiments 3 to 5 times depending on the availability of time. I think 3 replicas are sufficient. Also, if the estimates are accompanied by a measure of variability (standard errors and confidence intervals) the reader will have enough information to decide whether these estimates are more or less "reliable". Nov 12, 2010 at 12:32