How to appropriately show a small error bar?

Error bars are graphical representations of the variability of data and used on graphs to indicate the error or uncertainty in a reported measurement.

However, sometimes error bars are really short and even smaller than the size of the markers on my plot. In a scientific paper in the experimental sciences, it is still expected to include such error bars on the graphs. Thus, I got an ugly figure like:

Obviously, I'm not an experimentalist, and my questions are:

(1) Are there some convention about such an issue in practice;

(2) If the lengths of error bars are the same for all measuring points with the same marker. Can I just explain in the caption that I have used the size of markers (such as the diameter of a circle and side length of a square) to represent the error bar instead of plotting the error bars actually? If not, could you please suggest me how should I present such a figure better to show the statistic property.

• You need to have your marker as a dot which is as small as possible, it could even be invisible with the markers being a cross with lengths related to the errors. Commented Jan 26, 2018 at 11:13
• For data with very small error bars (eg CMB temperature spectrum, en.wikipedia.org/wiki/Cosmic_microwave_background), I have sometimes seen plots with enlarged error bars (e.g. by a factor of ten or 100). Otherwise, you could just draw the 'error bars', i.e. the intervals, and omit the marler for the points. Commented Jan 26, 2018 at 11:42
• I agree with both comments. Your points seem rather ok as well. Caption can be a crucial part for graphs. Conversely, never understood the trick used in CMB graphing. Commented Jan 26, 2018 at 13:55

It depends on the area, but in general, the appropriate choice for displaying the error depends on the specific purpose of the plot. Here are some examples:

• When the errors are obtained from a quantity such as the standard deviation $\sigma$, it's possible to the define them instead at $3\sigma$ ($99.7\%$ confidence).

• In some circumstances, e.g., when the errors are being used to judge a fit, depicting the residuals instead is usually the best option.

• When it's important to show the relative sizes (for different data points) of the error bars, they can be multiplied by an arbitrary common factor in order to become visible.

• And when the too-small error bars don't play an important role in the plot, it's very usual to simply point out that the "error bars are smaller than the symbol size".

Science writing is for effectively communicating results. You don't need error bars, you just need to communicate your error. If your error bars are too small to be visible in your graph, then the graph is an ineffective way to communicate your measurement error. If the error bars are all smaller than your data point symbol, it may be sufficient to simply state as much and elaborate in the caption or text. If you really need to communicate the variability of the error across data points, there may be other ways to do this. If it's important, perhaps it warrants a separate plot (or an inset). On the other hand, over-communicating useless detail about your measurement error can detract from the presentation. What is the story you want to tell?

Science writing is an ever-evolving discipline that has developed organically. The only rule is, and has ever been, to communicate your results effectively.

• Hi Gilbert, thank you so much! I want to show my temperature data to compare with computed results from a model. Your answer has taught me on a higher dimension, i.e., scientific writing, although I accepted @stafusa's answer since that one has focused on the method of presentation in the special situations. Thank you all!
– jsxs
Commented Jan 27, 2018 at 3:32