[Albacoide]

Dear users,

I have to present my work in a poster form. I wish to put confidence interval bars in all my graphs.

I'm studying and comparing the degradation of a compound by different types of treatments. I measure the concentration of the compound periodically during each treatment by HPLC.

If I want to calculate a mean and a standard deviation of the concentration (in order to get the confidence interval), with n = 3, should I measure the concentration 3 times in one experiment of each treatment, or should I perform three experiments for each treatment and measure the concentration only once in each one of them?

Thanks in advanced!

[Corribus]

The experiments are both valuable and tell you different things. The first tells you about the error of your measurement method, including the way the sample is prepared for measurement. Ideally if you measure the same sample three times, you will get the same value, but it is rarely the case. There's always some variation. The second would tell you more about the systematic variation around what you're actually studying. Typically the second experiment is more scientifically interesting and is what you would present at a meeting. The first experiment is part of your method development and would only really be scientifically interesting if you were presenting a new analytical technique.  For analysis many instruments implicitly build multiple replicates into the analysis routine and present an average value anyway. If not, measuring the sample multiple times isn't a bad thing to do, but the error associated with multiple measurements of the same sample is almost always much smaller than the variation around multiple independent samples. So I would generally say that once you have developed your method and satisfactorily demonstrated its precision within certain tolerances, measuring multiple completely independent samples is the approach you would want to take.

[Albacoide]

Thank you. There is only one experiment which I have performed three times.

Due to lack of time, is it acceptable to present error bars only for this experiment or should I simply not include them at all?

[Corribus]

I would include the error bar for the experiment you have done multiple times to get an indication of the expected variation, and then present no error bar for the other samples - and note that you have only done one replicate under these conditions. 

Of course, with only a single replicate it's difficult to get a sense of whether your trends are significant or not, particularly if the variation of your data as a function of the test condition is small.