log graph scale

Or is that the big bung, as practiced by Saudi Arabia for years? Brian

This feeds in to something I never got at school. When there could be - values in a result how do you make a graph for it? Brian

Oh dear, this is really a log artefact surely? its all about getting huge ratios in a display form though. Brian

We had a military environmental test chamber (AKA oven) that had to go from

In a race, Wiggins had on his shorts "WIGG^o" (superscript "o") - seemed right, as his number was 1.

That would have looked even better if you'd followed up my follow up to that post. :-)

I expect you understand it *now* though. :-)

I'm plotting the number of responses from about 100 sources. The number of responses varies from 0 to 867.

Bill

The responses vary from no response (0) to 867 responses. Because a lot of the responses are in single figures and quite a few are in double figures I need to show them accurately. A linear scale of 0 to 1,000 would have all the low responses crammed at the bottom. That's no good because the difference between, say, 2 responses and 7 responses is important.

Bill

Yes it's ten.

Bill

Well yes, but it doesn't seem right. I might have a non-log section at the bottom between 1 and 0, and mark it as such.

Bill

Problem is the bins would have to be just one integer because the sales people want to see the difference between, say, two and three referrals.

Bill

Yes that would work.

Bill

Others have suggested two perfectly adequate fixes. But if you really insist on putting zero on a truly logarithmic scale then it *has* to be at minus infinity.

Yes I know, I've seen them, but even so I thought it courteous to reply to your question.

I don't insist on anything. I posed the question because my maths knowledge is poor and I wondered what the maths experts here would make of it. They have as you say provided several solutions. So everybody, except maybe you, is happy.

Bill

... which is pretty foolish IMO, if the number of referrals covers the range you suggest. They want to look at the data in far finer detail than it warrants.

Surely you don't really need to put a mark on the graph to represent zero, so the issue of where to put the mark doesn't arise? Just make a note that the least amount recorded anywhere is one, and maybe represent that by a change in the colour of the x axis, so a black axis line and no red bar above it represents zero, while a red axis line but no red bar above it represents a single response.

If you were doing it in Excel then adding 1 to your raw values would cure it so that you can take logs and plot it. I find the can't plot zero warning in Excel quite annoying since I know that but a lot of experiments involving high dynamic range pulse counting data frequently do have zeros in where there is no signal. I catch the error and silence it but that requires a bit of tricky programming.

The other way would be add a small amount to your data like 0.5 and plot the zero responses below the "1" line making a mental note that they should actually be at -infinity rather than where they are plotted.

The other option is to use a semilog type response. Abd constrct a transfer characteristic that is part linear part log.