Science Practice 5: Means, Standard Error, and Error Bars

Two bars on a graph can look different just because of random variation in the samples, not because the underlying groups are actually different. Standard error of the mean (SEM) gives you a way to judge whether a gap between bars is real or just noise. The formula is simple: SEM = standard deviation / √n, where n is the number of replicates. Because n sits under a square root in the denominator, increasing it shrinks the SEM, but with diminishing returns: to cut SEM in half, you need four times as many samples, not twice as many.

Error bars on a graph are usually drawn as plus-or-minus one SEM around each mean. When you compare two bars, look at whether the bars overlap. If the error bars overlap substantially, the two means are statistically indistinguishable; you cannot say the groups differ. If the bars do not overlap, that is good evidence the means are genuinely different, though it is not an ironclad statistical test by itself. A rough rule used at this level: non-overlapping SEM bars support a real difference, heavily overlapping bars do not.

This connects directly to experimental design across the required investigations: more replicates give tighter error bars and more trustworthy conclusions, which is why good lab write-ups report sample size alongside the mean. It also sets up the distinction between describing a single sample (standard deviation) and estimating how well that sample represents the true population (standard error) - a distinction that shows up whenever you are asked to defend whether a pattern in data is meaningful or just scatter.