Statistical effects of varying sample sizes on the precision of percentile estimates

Am J Hum Biol. 2000 Jan;12(1):64-74. doi: 10.1002/(SICI)1520-6300(200001/02)12:1<64::AID-AJHB8>3.0.CO;2-N.


The present study evaluates the precision of outlying percentile estimates, with age- and sex-associated variations and facilitates decisions needed to revise the current NCHS 1977 Growth Charts with regard to 1) the inclusion of 3(rd) and 97(th) percentiles and 2) the selection of survey data for the construction of the revised growth charts. Simulation was performed to obtain data with distribution characteristics similar to those of The Third National Health and Nutritional Examination Survey (NHANES III) (1988-1991) data. NHANES III consists of a two-phase, 6-year, complex stratified multistage probability cluster, cross-sectional survey conducted from 1988 through 1994 to represent the US noninstitutionalized population. Phase I of the survey consisted of 679 boys and 622 girls in age groups 3, 8, 13, and 18 years. Weight and stature, the body mass index (BMI) (weight/stature(2); kg/m(2)) was calculated. The results show that 1) the precision of the percentile estimates is greater for stature than for weight and BMI, 2) percentiles during the pubertal period are less precise than those during the prepubertal and postpubertal periods for weight and BMI but there is little difference for stature, and 3) percentile estimates are more precise for girls than boys for weight and BMI, but not for stature. The present findings suggest that pooling of NHANES III and earlier National Center for Health Statistics (NCHS) survey data is necessary to achieve reasonable precision for the 3(rd) and 97(th) percentile estimates. Am. J. Hum. Biol. 12:64-74, 2000. Copyright 2000 Wiley-Liss, Inc.