On the use of error bars
Tags:
2D,
Black/White,
error bars,
statistics
21 ago 2009
This and other interesting figures and tables illustrate an informative paper by Cumming et al. on to use error bars in scientific plotting.
Reference: Cumming, G., Fidler, F. and Vaux, D.L. (2007) Error bars in experimental biology. The Journal of Cell Biology, 177: 7–11. (download pdf) (JCB website)
1 comentarios:
This is a nice visual comparison of different kinds of error bars. However, their verbal description of the CIs is not correct. Frequentist 95% CIs cannot be interpreted as being a range of values that will include the "true" parameter value with probability 0.95 based on a single sample. Instead Frequentist CIs make an appeal to (usually hypothetical) repeated sampling. Their proper interpretation is that under repeated, independent sampling where a CI is constructed for each sample, x% (where x is the confidence level specified by the user, ie. 95%, 90%, etc) of the CIs would contain the true value of the parameter. However, in any particular sample, the probability that the CI covers the true value is either 0 or 1. If one wants to interpret a confidence envelope in the manner they do above, then one needs to calculate Bayesian credible intervals, which are based only on the sample at hand and the prior distribution of the parameter. I should add the caveat that I haven't read the original article, only the parts that have been posted here. I apologize if this issue is discussed in their paper.
Publicar un comentario