Reading Values from a Graph



Introduction

Have you ever download a graph from the internet or a PDF or Word file or even as an image that you want to help with your analysis of the data?

You can now watch this on YouTube: https://youtu.be/Zriz4D__tAc

A graph like this one, for example:


What are the Numbers?

It might not be so clear here so click on the above image to enlarge it. In my Excel file, that graph is a good size and it is very clear. But, where are the numbers? Where are all the values for me to analyse? They are essentially hidden from us.

I know of software and web pages that will read the data from such a graph that you can download and use. The trouble is, the ones I know of are not free. So, I was hoping to find a free application that will tell me the values in the graph. So I asked Claude for Excel and it said, yes, I can do that, providing the image is clear.

Claude, can you do this  ... ?

My prompt was a simple question, can you read and download all of the data from the four curves found on the graph on the current worksheet? This is what it gave me:


There is no significance in the blue banding in that table. It took Claude a minute or so to find those numbers and it volunteered to help me to find the real numbers but I declined.

Yes but it's not Perfect


In fact, I had the data all along, since I was testing Claude's ability to do this. I won't share the data but I am about to share the percentage differences between the Claude data and the actual data:














The final line of the table shows the arithmetic means of the differences of all four variables and they turn out to be about 1% in all cases.

An average 1% difference for a general study and a general review is tolerable, I think. For a deeply scientific study for which the data and the statistical analysis need to be pin point accurate, a 1% difference may be too much.

Conclusions

I would say that is a pretty good result and I shared with Claude what my findings were and it suggested it will try harder next time!

I don't know why but I didn't think of asking ChatGPT to do this but maybe I will.

Duncan Williamson

May 2026

No comments: