Error bars for exponentially distributed data


Short version: What is the best and most proper way for displaying error bars on exponentially distributed data?

Long version: I have some data (biological, though that's not important) that we bin to produce a histogram. Within any given bin the data are exponentially distributed - that is, lots of events at low values, and few events at high values.

The problem is this - when submitting for publication or showing these data before an audience, folks generally expect error bars. However, since the standard deviation = the mean for an exponential distribution, high mean values are inevitably associated with large error bars. Further, in an exponential distribution 1 sd will encompass the vast majority of the data points, while in a gaussian distribution it will only encompass 68%. The sd/sem mean fundamentally different things for these two distributions.

This is no surprize, but it doesn't satisfy reviewers and audience members who are used to seeing standard deviations and standard errors superimposed as error bars on everything.

So, what is the best and most defensible way of representing uncertainty ("errors") on graphs of exponentially distributed data?