Naked Science Forum
Life Sciences => Cells, Microbes & Viruses => Topic started by: Fibonacci on 20/03/2013 03:48:18
-
I don't understand why most biological experiments in literature are automatically run in triplicate. What's so special about the number 3? Why is it automatically assumed that a study run in triplicate has enough statistical power to make the observations from that study meaningful? Don't you have to run power of analysis first before determining proper sample size? Why is "triplicate" a mindless automatic default for number of times to do an experiment? I even see papers that get published in Nature and Science with experiments "done in triplicate". I don't understand why so much of the literature out there talks about the need to do power of analysis for proper sample size (in fields like clinical medicine etc.), yet many other branches of science can simply get away with the de facto default of "triplicate" and conclude that results from such an experiment are meaningful. Why doesn't all science require power of analysis to design an experiment?
-
Hello,
You are mistaking two different things for the same : the number of samples in the experiment and the conduct of the experiment itself.
The number of samples will determine the statistical power of the study. It depends on a lot of factors.
The number of replicates you run for each measurement/ test/ experience... only determine if you will have a important variation in your technique. This is called the technical replicate, and it is done in triplicate because it is the minimum number to have a standard deviation. This way you ensure that your technique gives the same result each time you test the same sample.
Only when you have satisfactory technical reproductibility will you look into the statistics of your experiments, and the number of samples required. The triplicate only ensure that you will be able to compare your different samples between them.
-
If you do it twice and get different answers you don't know which is right.
If you do it 3 times and 2 agree you can still be reasonably sure that those 2 are right and the third went awry.
However the number of replicates often depends on other factors- the cost and how important the outcome is are two common factors that affect the experiment.
Experimental design is an important subject in its own right.
http://en.wikipedia.org/wiki/Design_of_experiments
-
Hello,
You are mistaking two different things for the same : the number of samples in the experiment and the conduct of the experiment itself.
The number of samples will determine the statistical power of the study. It depends on a lot of factors.
The number of replicates you run for each measurement/ test/ experience... only determine if you will have a important variation in your technique. This is called the technical replicate, and it is done in triplicate because it is the minimum number to have a standard deviation. This way you ensure that your technique gives the same result each time you test the same sample.
Only when you have satisfactory technical reproductibility will you look into the statistics of your experiments, and the number of samples required. The triplicate only ensure that you will be able to compare your different samples between them.
I think this is easier to explain with an example:
Let's say I have cells that I want to test drug X on at concentrations of 0, 25, and 50 uM just for toxicity by just counting the cells after 48 hours after exposing them to the drug. Experiment 1: collect 3 samples of each concentration and count. Experiment 2 (repeat): collect 3 samples from each concentration and count. Experiment 3 (repeat): collect 3 samples from each concentration and count.
In total, my n=3 for each concentration, it is not 9 for each concentration, since technical replicates like you said are only to control for variance, they are not biological replicates. N=3 reflects the number of times I independently ran the experiment. Independence is key. Separate experiments in this case produce independent results, replicates within the same experiment do not boost my value of n.
Power of analysis must be done a priori, BEFORE an experiment is done to properly determine sample size (in this case sample size for each concentration reflects how many times I run the experiment). It is possible to do post hoc power of analysis, but it is much less useful. In fact, many statisticians basically claim retrospective power of analysis essentially worthless.
I just don't understand why many fields of science simply assume n=3 is good enough. Why is this so? Should a priori power of analysis determine sample size? Power of analysis is absolutely required in fields like psychology, sociology, medicine, or when one wants to conduct an animal study. Why do other fields of science basically not even use it?
-
If you do it twice and get different answers you don't know which is right.
If you do it 3 times and 2 agree you can still be reasonably sure that those 2 are right and the third went awry.
However the number of replicates often depends on other factors- the cost and how important the outcome is are two common factors that affect the experiment.
Experimental design is an important subject in its own right.
http://en.wikipedia.org/wiki/Design_of_experiments
But what I don't understand is why is accepted that 3 is good enough? What's so magical about the number 3? 3 measurements or reproducing and experiment 3 times to obtain 3 measurements might not even be good enough if the effect size you're looking for is small or mediocre (but biologically important). Shouldn't power of analysis be used to properly determine sample size, just like how it is used to determine the number of mice that might need to be used to test drug x?
-
Three often isn't accepted as enough, it's typically the fewest that you can get away with.
If the cost of the experiment is high then you don't repeat it more than you need to.