Statistical Modeling, Causal Inference, and Social Science: Antony Unwin's graphs for autism data

Antony Unwin's graphs for autism data

In response to this query on how to reexpress Venn-diagram data graphically, Antony sends along this picture:

and writes:

Nathan

I'm going to have to agree with you, Andrew. I've never been a big fan of the mosaic for lots of categories. Much the same problem of pie charts. Information gets confusing in a hurry.

Antony Unwin

Mosaicplots are best used interactively, rather than for presentation, and there is rarely a good alternative. Pie charts are best rarely used and there is usually a better alternative.

I'm curious to see Andrew's line plots, how will he show the sizes of the groups?

Igor Carron

Antony,

I like it too, it conveys in a forceful manner that unless you have the trained clinician in the loop you chance close to a coin flip to be right seven years later.

One of the commenter in my blog reminded me of something else, there are values outside of all the circles. It is the number of kids that have been identified by clinical services as having a problem but for which no tests were performed at age 2 yet showed a percentage of them having autism at age 9. Some of these numbers are interesting on their own.

Igor.

Antony Unwin

Igor,
Your point about the clinician is a good one. Analysing data in a vacuum (i.e. without a domain expert to interact with) is always second best. We want to put statistical insights in context -- and find out if they are of any practical use.

If the additional data you mention are comparable and available, it would be easy to add them to the plot.

Igor Carron

Antony,

I don't have any additional data, the data I am mentioning is on the graph. They are the numbers outside of the circles.

These numbers tell another story: these kids pass through the sieve of having first been referred clinically at age 2 but for some reason, they could not take the tests or were not deemed (by the clinical staff or the family) in such a condition that they should take the test. Yet, 14 % of them had a diagnosis of autism at age 9. I would say this is odd.

Igor.