turn on suggestions

Auto-suggest helps you quickly narrow down your search results by suggesting possible matches as you type.

Showing results for

- JMP User Community
- :
- Discussions
- :
- Discussions
- :
- Log transformed data - linear graph axis - Use JSL...

Topic Options

- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Permalink
- Email to a Friend
- Report Inappropriate Content

Aug 8, 2017 6:17 AM
(847 views)

I have biological data that is best displayed with a log-Y axis, the problem is that this causes asymmetric error bars which can extend very misleadingly below the data (if any layer has near zero data).

The work around is to create a column to log transform the data, plot on a linear axis and tell the labeller to use a custom formate that is = 10^(:value).

The new problem then is the minor axis ticks, which are now on a linear scale and not log.

Two questions:

1) Is it fair statistically to do this?

2) How to use JSL to draw log ticks when the axis is linear?

3 REPLIES

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Permalink
- Email to a Friend
- Report Inappropriate Content

Aug 8, 2017 1:28 PM
(831 views)

What quantity do your error bars represent? (One standard deviation of the data? One standard error of the statistic? The confidence interval of a statistic?)

Why is the asymmetry undesirable? (On the same scale, the confidence interval of the mean is symmetric but the interval of the standard deviation is not.)

Why did you expect the error bars to be symmetric?

Learn it once, use it forever!

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Permalink
- Email to a Friend
- Report Inappropriate Content

Aug 9, 2017 6:09 AM
(810 views)

Hi Mark,

To answer your questions: I would like it to be one SD, asymmetry is undesirable because the SD on the non-log transformed data goes below the limit of detection and the graph is misleading.

Here’s an example of when I need to use the log transform, these are the same data. Usually I would just do a custom format on the log transformed axis to get the labels to be the same as above:

[cid:image001.png@01D310EF.255E6DD0]

[cid:image002.png@01D310EF.255E6DD0]

To answer your questions: I would like it to be one SD, asymmetry is undesirable because the SD on the non-log transformed data goes below the limit of detection and the graph is misleading.

Here’s an example of when I need to use the log transform, these are the same data. Usually I would just do a custom format on the log transformed axis to get the labels to be the same as above:

[cid:image001.png@01D310EF.255E6DD0]

[cid:image002.png@01D310EF.255E6DD0]

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Permalink
- Email to a Friend
- Report Inappropriate Content

Aug 9, 2017 7:16 AM
(802 views)

Also, I would prefer to use the Engineering SI format.

What I have now is:

ScaleBox,

{Format( "Custom", Formula( 10 ^ value ), 10 )

This displays a log scale in the right place (without the log minor ticks which I still would like to know how to include). But what would be really helpful is to be able to run the Custom format through the Engineering SI format so that the transformation of the axis labels goes 3 - > 1000 - > 1K. Is this possible?

Thanks!