Thank you very much for you posting. I’m working on a open source software that plots financial graphs but it had a broken log-scale-plotting-function which I had to fix. After a few days of some misleading readings I discovered your explanation and it gave me the clues I needed to solve the problem.

Your writing is really clear. Congrats!

Again, thank you very much.

Mario Marinato, from Brazil

The nature of logarithms is that you can only convert positive numbers to logs, with the resulting logs covering the range of ± infinity.

You can always fake it. I have a number of samples on my main site that show various Custom Axis Scales using Dummy Series. Most involve adding a helper (or “dummy”) XY series to create the axis, then transforming the data to accommodate this dummy scale

You would have to decide where your line of symmetry would be. For example, you could have the positive values go 1, 10, 100, etc, and the negatives go -1, -10, -100, etc, with +1 and -1 occurring at the same point. To get your negative values to plot, you need to transform them using something like -log(-value).

Alternatively, you could select a ± range that will be linear in the middle, with log values extending outward from this range. So the scale would go
-100, -10, -1, 0 ,1 , 10, 100, etc., with the data plotted linearly between -1 and +1.

Basically what I’m looking for is:

Tick marks, each an inch apart, on the positive Y axis, labeled: 10, 100, 1000, 10000, …
Tick marks, each an inch apart, on the negative Y axis, labeled: -10, -100, -1000, -10000, …

It just seems wrong that I’m allowed to scale positive values but not negative ones. Can you help? :)

This is a good point. I noticed the difference in units, but your conversion into similar units tells an even harsher tale of deception. Have you chared this with Nicolas?

Anon – In this case, the values were determined “by inspection”. I had made one or two preliminary charts, which led me to decide these values would work well. Originally I was thinking 2-5-10-20-50-100, but I adjusted as I went along.

In some professional applications, I’ve used algorithms to determine these values. Depending on the spacing and the relevant endpoints, I might choose between 1-5-10, 1-25-50-75-100, 1-20-40-80-100, etc., for the general spacing, and use another algorithm to choose where precisely to start and end.

I have a standard protocol I use for linear axes, which has evolved from a VBA technique of Stephen Bullen’s. I have developed both worksheet function and VBA versions. Stephen has posted his protocol in the newsgroups, but it’s also available in chapter 15 of Professional Excel Development, under “Function to Calculate Reasonable Chart Axes Scales”.

EXCELlent post… as always.