LUCIFERASE ASSAY Standard Error bar ??
ucgatan at socrates-a.ucl.ac.uk
Fri Jan 30 05:27:35 EST 2004
On Thu, 29 Jan 2004, Obaid Khan wrote:
> I want to know what is the proper method of reporting the standard
> error on luciferase assays. We generally do 2 replications of an
> assay, and from my knowledge of stats, you cannot calculate variance
> std. dev. on a sample size of 2.
AIUI, that's correct. I mean, you _can_ calculate a standard deviation,
and moreover, a 95% confidence limit, but it won't be very meaningful with
only two samples.
> I will appreciate if anyone can let me know what to do in my situation
> and if anything is available in this issue on any website.
I see two obvious options: either you do more replicates and calculate a
confidence limit as normal (you'll need at least 10, i think, and
preferably 20), or you stick with a small number of replicates (3-5) and
don't calculate an error parameter, but rather use nonparametric
statistics to compare your control and experimental results; the
particular test you want is the Mann-Whitney U-test. Nonparametric
statistics are strictly less powerful than the parametric alternative, but
are more robust with small numbers of samples. If you must have an error
range on a graph, you can use the interquartile range or something.
Tom Anderson, MRC Laboratory for Molecular Cell Biology, UCL
More information about the Methods