1
$\begingroup$

I have modeled counts of health factors at the census tract level geography. When I plot these counts against each other (example: smoking vs. heavy drinking) I get a weird scatterplot output. The results seem to be bi-linear or dovetailing, with one point cloud on a steeper trajectory than the other (see image).

I can't seem to find the right description/definition of this to explore it further. My thought is that one group of census tracts has a steeper correlation and then due to another factor (likely SES or race/ethnicity) there is another lesser correlation.

Does anybody have any information on the shape or pattern of this scatterplot?

Thanks![enter image description here]1

$\endgroup$
  • 1
    $\begingroup$ This is a good candidate for "scatterplot brushing": if those two modes are associated with any other factor in your table, you will learn that within seconds. $\endgroup$ – whuber Feb 8 '17 at 20:44
1
$\begingroup$

One possible next step is to run some panel or trellis plots broken down by state levels. (I'm assuming you're using US data since I have only encountered the term "census tract" in US census data.)

When you look at the prevalence of smoking and binge drinking in the US, they don't fully agree. The northern part of the country seems to favor alcohol (perhaps related to the cold weather), which I suspect could be the lower part of the dovetail; the southern part has higher tobacco use, perhaps accounting for the steeper subset of the pattern.

Other than weather, it can also be related to tax. As each state may have different tax schemes for these products which would ultimately affect price. I would consider political boundary, pricing, and income the next factors to investigate before race/ethnicity.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.