View high res version here.
It is often assumed that low-income populations have worse health outcomes. This is correct for many health outcomes, but for one, it isn’t as clear: Body Mass Index (BMI). BMI doesn’t show a clear association with income. Given that we are in the midst of an obesity epidemic, and BMI is the primary measurement of obesity, this is quite interesting. I wanted to take a look at some recent data to visualize the association.
I created a modified violin plot using the 2015-2016 wave of NHANES. Violin plots are great for getting a sense of the distribution of a variable while also looking at the density of the most likely values. I also included the actual data points, which are represented by the circles (4,973 in total). I also added the average BMI within each level of percent of the federal poverty level. It is clear there is little difference in the distribution and average BMI score by FPL category. The exception is for households with incomes 500 percent or more of the FPL. In that group, we see the average drops slightly and there are fewer individuals with extreme values of BMI and more with BMIs below 30 (see how the violin plot is wider around a BMI of 25, indicating a higher density of values there). Here is a table of some basics descriptive stats:
|Percent of FPL||Mean||SD||Median||Min||Max|
|0 - 99||30.0||7.4||28.9||16.2||64.6|
|100 - 199||29.8||7.5||28.7||15.5||67.3|
|200 - 299||29.6||7.3||28.5||16.3||64.5|
|300 - 399||29.4||7.2||28.4||15.1||62.7|
|400 - 499||29.8||7.6||28.7||16.6||63.6|
|500 or more||28.5||6.2||27.3||16.5||54|
Overall, I think this is a neat and easy way to visualize a continuous variable by levels of a categorical variable. A few notes on the analysis:
- This is the unweighted data, so it isn’t nationally representative, but the mean BMIs are quite close to weighted estimates.
- NHANES top-codes age at 80, so anyone 80+ gets coded as 80. For that reason and because there weren’t many observations for those 80+, I only included those aged 18-79.
- Height and weight are measured directly in NHANES, which means reported BMI is accurate (compared to self-reported height and weight).
- Income is self-reported and is top-coded at 500 percent of the FPL. I chose to keep these observations because there are a lot of them and they are an important category of FPL.
- I chose FPL rather than total family income because it adjusts for household size and provides a reasonable baseline (i.e., how much below or above the poverty level).