What is a statistical graph?

A good statistical graph is, most often, a way of summarizing a lot of numeric information in a relatively small space.

What should you think about before starting work on a statistical graph?

There are at least five things to think about before working on a statistical graph. The first ting to think about is what to graph. The second thing is why do you want to graph it. The third thing is who will see it. The fourth thing is how are they going to look at it. And the fifth thing is when do you need it done?

Each of these requires some thought.

“What to graph” is the most basic question, but also the most difficult. You need to think about how many graphs you will use, how many variables each will have, what relationships you are trying to highlight, and so on.

The question of why you want to graph it is also key: Is the graph part of a long modeling process? Is it a ‘quick and dirty’ graph to show to a colleague? Will it be part of a scientific paper? Are you going to use it in a formal presentation? Are you trying to convince someone of something? Are you trying to summarize a lot of information, or highlight a relationship, or both?

The next two questions are related to the previous one. Who will see it? How much do they know about graphs? Are they your colleagues, your supervisors, your subordinates, or your rivals? Might they look at it for flaws? Just as you would speak differently to each audience, so you may wish to make different graphs for each audience.

How will they look at it? In a paper? On a screen? As part of a presentation? If so, who will control the pace of the presentation? Will they be able to ask questions? Will you be able to answer them?

Finally, the question of when you need it done influences the whole process. If you need a graph done quickly, it had better be one you know how to produce quickly.

Can you lie with statistical graphs?

Of course you can. Just as you can lie with words, you can lie with numbers – and, since a good graph represents words and numbers, you can lie with a graph. But, because most people are not used to looking at graphs (but they are used to listening to words), you can lie more easily – unless, of course, the people who are reading your graph know how to look at one. It’s easier to lie to someone who isn’t fluent in the language you are speaking; it’s harder to lie to an expert.

How should you read a statistical graph?

The first thing to ask, here is “Do I understand this graph?” If the answer is no, stop. There’s no shame here. There are at least two possible reasons you might not understand a graph: First, it might be a really bad graph. Second, it might be a type of graph you don’t know about. The first is like the situation where you are reading a passage that is so poorly written that it is incomprehensible; just as there are people who write this way, there are people who make this kind of graph. The second is like listening to a language you don’t understand.

Then you should carefully examine the axes and legends, noting any oddities. Then you can scan over the whole graph, to see what it is saying, generally. Finally, if necessary, you can look at each portion of the graph. Then go back to the first question and ask it again: Do I understand this graph? And, if so, what is it saying? And, if not, why not?

What next?

In future articles, I will cover:

What is a good statistical graph?

How to rate statistical graphs?

What is a bad statistical graph?