Why did CDC growth charts compress the upper weight values?
By Steven B. Halls, MD, FRCPC
While Dr. SB Halls and J Hansen were re-analysing the NHANES III data on the pediatric height and weight distribution in the American population, a slight difference was discovered between our results and the results plotted in the standard CDC growth charts.
The CDC weight charts show the 95th percentile line at a lower weight than the halls.md weight charts do. This is illustrated in the graphic below. The Purple Lines is the CDC chart for Boys Weight ages 2 through 20, and the chart with a light blue background and Red lines is the halls.md chart of Boys Weight for White race/ethnicity, superimposed so you can compare them easily.
Notice that the 95th percentile line (the highest red line) is higher on the halls.md chart than the CDC chart. The CDC’s value at age 18 years is 202 pounds, whereas the halls.md chart shows a value of 230 pounds. That’s a significant difference.
In other words, the NHANES III survey data showed more childhood obesity than the CDC charts show.
Why did this happen? The explanation could be found in the document describing the CDC methods ( http://www.cdc.gov/nchs/data/series/sr_11/sr11_246.pdf ) of creating the CDC growth charts. In brief, the CDC combined data from the recent NHANES III study, with data from several large older studies from the 1960s and 1970s. ( NHES II, NHANES I and II ) They did this in order to create a very large data sample, to create enough statistical power to allow beautiful smooth curves to be mathematically proved1 to a high degree of certainty.
The drawback of combining data from new and older surveys, is that the older data showed less obesity. This lowers the CDC’s overall numbers for the heaviest percentiles.
Since the CDC charts are superior from a statistical point of view, you might ask why I didn’t remove the halls.md charts, after the CDC published their charts? The explanation is, the halls.md charts express a different opinion than the CDC.
A benefit of the world-wide-web, is that different opinions and points of view can be expressed, which benefits everyone. – Steve Jobs
The CDC point of view:
The CDC has a greater responsibility to the public. The CDC properly chose to use rigorous scientific statistical methods for curve fitting. This choice unfortunately required extremely large data samples, so data was combined from all race/ethic groups, and older surveys were combined with newer data1. Their choice was "statistics & science" over "race-specific & most recent".
The halls.md point of view:
The NHANES III survey was very big, and it is the most up-to-date snapshot of America’s heights and weights available. It is possible to plot this data on charts, although admittedly, the extreme percentiles lines were curve-fitted by "eye-balling" (empiric methods) more than with numeric "regression" methods. The halls.md charts show how height and weights ARE recently, with their averages and current variances plotted. Some web researchers are genuinely interested to find the most recent data, or to compare the heights and weights of different race/ethnicity groups.
If you had an overweight child, the CDC charts would suggest that your child is more overweight than the recent NHANES III population survey indicated.
If you were an obese teenager looking at the CDC charts, they might cause you to feel very isolated and freakish, when in fact you are not alone.
- Guo SS, Roche AF, Chumlea WC, Johnson C, Kuczmarski RJ, Curtin R. Statistical effects of varying sample sizes on the precision of percentile estimates. Am J Human Biol 2000 Jan;12(1):64-74