COMPGROUPS.NET | Search | Post Question | Groups | Stream | About | Register

### Plotting error bars in semilogy scale

• Follow

Hi, I'm a matlab and statistics beginner. I have row vector containing the mean values I need to plot. I have another row vector containing the standard error of the mean corresponding to these means.After using errorbar to plot the means and error bars, I use

set (gca, 'YScale', 'log')

to convert to semilogy scale. As a result, I get error bars that are asymmetric (different in the length between the mean and the upper limit, and between the mean and lower limit).
I have two questions:
1. Is having asymmetric error bars a problem? (Does it lead to misleading interpretation?)
2. If it is indeed a problem, how do I get around it?

I have extensively searched on the Mathworks forum and using Google, but I have found nothing helpful
 0

On 29/12/10 1:00 AM, teacher's pet wrote:
> Hi, I'm a matlab and statistics beginner. I have row vector containing
> the mean values I need to plot. I have another row vector containing the
> standard error of the mean corresponding to these means.After using
> errorbar to plot the means and error bars, I use
> set (gca, 'YScale', 'log')
> to convert to semilogy scale. As a result, I get error bars that are
> asymmetric (different in the length between the mean and the upper
> limit, and between the mean and lower limit).

Correct. Although this startled me when I encountered it a couple of
years ago, with a bit of thought I realized that it was a proper
representation in log space of addative quantities.

> I have two questions:
> 1. Is having asymmetric error bars a problem? (Does it lead to
> 2. If it is indeed a problem, how do I get around it?

Difficulty in interpretation is going to depend on what is customary to
the reader of the plots, and "customary" is going to depend upon their
field of study and experiences. *I* am not accustomed to seeing log
plots with error bars, but I am not a statistician. If the error bars
were symmetrical on a log plot, I would probably misinterpret the plot
as being statistics about the log rather than statistics about the full
value.

The lesson I personally would derive from this is to avoid using log
scales with plots that must have error bars of standard deviations. But
as I indicated, I am not a statistician so I do not know what is
customary in such cases.

Keep in mind, though, that any plot you create for statisticians only
has to be interpreted correctly 19 times out of 20 ;-)

 0

1 Replies
1574 Views

Similiar Articles:

7/21/2012 3:49:14 AM