Research shows that weight trends associated with Body Mass Index (BMI) perpetuate throughout life-overweight children become overweight adults-leading to high percentages of overweight individuals at all ages. Despite the persistence of high percentages of overweight individuals, research has not yet identified the cause of unhealthy weight in adolescence. A variety of social, environmental, and genetic factors have been linked to overweight, but no study has identified how/ways adolescent and adult weight development differs [1]. This study focuses on identifying whether intrinsic, extrinsic, or both types of factors significantly correlate with Body Mass Index (BMI) and whether these relationships change as adolescents age.

This paper assesses the intrinsic and extrinsic determinants of BMI using a 15-year panel (duration 1997 to 2011) from the National Longitudinal Survey of Youth 1997 (NLSY97). The analysis questions the relative strength of these relationships, how they vary over time, and it consists of three stages. First, analysis of covariance (ANCOVA) tests the strength of the relationship between BMI and selected intrinsic factors. Second, multilevel Generalized Linear Models (GLM) evaluate the relationship between adolescents age 12 to 20, and adults age 21 to 32 using BMI [1] and intrinsic and extrinsic covariates. Finally, multinomial logit regressions test the relationship between weight category (normal weight, overweight, obese) and the intrinsic/extrinsic factors.

Section I discusses what is known about BMI. Section II includes a description of the data and estimation methodology. Section III presents the estimation results from all three analyses. Results are discussed in Section IV, and Section V briefly concludes the primary findings.

BMI is highly correlated with body fat and can be used to classify individuals as underweight, healthy weight, at risk of being overweight, or overweight using a nationally accepted rubric [42]. Among adults, BMI appears to be a satisfactory measure of body fat [43] especially if comparing across race and ethnicity [44]. Among adolescents, the CDC child and adolescent BMI thresholds are used to categorize individuals by weight status. Thresholds, expressed as percentiles, are designed to capture the category the individual would be in upon reaching young adulthood (if staying within the same BMI percentile).

For adolescents under 20, normal or healthy weight status is based on BMI between the 5^th and 85^th percentile. After age 20, BMI is interpreted using standard weight status categories, and normal weight equals BMI between 18.5 and 24.9. The standard weight status categories associated with BMI ranges for adults are shown in the following Table 1 along with the adolescent percentiles.

Fig. (1) shows average BMI for male and female racial and ethnic subgroups by age (measured in months. In the NLSY97, gender is self-assigned by respondents as either “male” or “female” in the first year of the survey assignment remains unchanged The two black lines represent the 5^th and 85^th percentiles for ages below 20 and BMI 18.5 and 24.9 for ages 20 and above. Therefore, the lines are parallel after age 20. The lines are disjointed due to changes in the way the CDC defines healthy weight for adolescents’ and adults’ BMI. Minority groups have higher BMI at all ages. Hispanic males and black females display higher BMI levels and steeper growth (Fig. 2) consistent with other studies [45, 46]. The proportion of underweight has also decreased among all groups, but BMI levels remain persistently high.

To maintain a balanced panel, the sample used in all three analyses only includes respondents with a BMI value in each year of the panel. While measurement and misspecification error is a concern in self-reported data, the data was cleaned to remove errant, inconsistent, and illogical values of height and weight. If BMI values were missing due to omitted height, height was imputed from nearby observations. Full height is achieved at relatively early ages; thus, imputations were unlikely to bias the sample which consisted of 4,205 individuals. Means are listed in Table 2. Minimum BMI is 12.5-underweight-and maximum BMI value is 55-overweight or obese-with an average of 25 and 26 for men and women respectively. BMI increases with growth and weight gain; however, rates differ by race and gender [47].

Fig. (1). Average BMI by Age and Gender. *=Prior to age 20, the CDC defines “normal” BMI as BMI between the 5th and 85th percentiles. These values represent the BMI at each percentage. After age 20, BMI between 18.5 and 24.9 is considered normal weight.

Fig. (2). Average BMI by Race and Gender.

Table 2 also includes mean statistics of intrinsic and extrinsic characteristics. Average respondent age (measured in months) is 272 months (22 years). Ages ranged from 146 to 204 months (12 to 17 years) in 1997 and 300 to 387 months (25 to 32 years) in 2011. Average age is 272 months (22 years). Average household size is 3.5 persons, but it decreases with age. Roughly 75 percent of the sample resides in an urban area, compared to 80 percent of the US population [48]. Dummy variables, northeast and south, control for regional differences, and the income/poverty ratio accounts for income level. Ratios below 1 indicate an income below poverty, while ratios of one or greater indicate income at least at the poverty level. The average ratio in the sample is between five and six-above poverty level. Maternal education is translated into two dummy variables: less than high school and high school completion. Less than 20 percent of mothers have less than a high school education, while 35 percent have completed high school.

Five intrinsic characteristics are used in the analysis-age of first sexual activity, number of years since first tobacco use, desired weight change action, and general health perception. The average age of the first sexual encounter is slightly under 16 years for males and slightly over 16 years for females. The indicatory for tobacco use measures the number of years since the respondent first used tobacco. Responses range from 0 to 11 years with an average between two and three years-implying that most respondents first tried tobacco in their mid to late teens. The NLSY97 captures desired weight action using a respondent selection of 1- Lose Weight, 2- Gain Weight, 3- Stay the same weight, and 4- Not Trying to Do Anything about My Weight. Analysis uses two dummy variables: lose weight and stay the same weight. Thirty percent of males want to maintain weight, and 30 percent hope to lose weight. Over half of females want to lose weight, and 25 percent want to maintain their current weight. Finally, perception of respondents’ health is measured using five descriptive options: 1- Excellent, 2- Very Good, 3- Good, 4- Fair, and 5- Poor. Average male and female response was two-indicating very good health.

These data are analyzed with three different statistical techniques-ANCOVA, GLM, and Hierarchical logistic regressions-each was performed separately for men and women.

Data came from the first 15 rounds of the NLSY97-a longitudinal panel that follows a sample of 8,984 American youth from 1997 to 2011. After 2011, the survey became biennial. While 2013 and 2015 are available, the sample focused only on those consecutive survey years.

BMI-the dependent variable-was calculated from self-reported height and weight. To calculate BMI, respondents needed a height and weight value. To maintain a panel balance, respondents without a BMI in each year were removed. Remaining data was cleaned using a series of flags to indicate errant, inconsistent, or illogical height and weight values. If height was missing, it was imputed from nearby observations whenever possible. Since full height is likely achieved early in the sample for most respondents, imputations did not likely bias data. Summing the flags, and removing errant values, left 4,205 respondents.

It is important to consider the nature of the data when reviewing the results. BMI is based on self-reported height and weight data. Self-reported height and weight data are consistently over- and under-estimated, respectively. With this in mind, the interpretation of multinomial estimates is not entirely intuitive, but the significance pattern of the multinomial logit resembles that of the GLM specification. Age and race/ethnicity are highly correlated with adolescent weight category-with lower probabilities that minority races appear in lower weight status categories. Intrinsic characteristics such as general health perception, years since first tobacco use, age at first sexual encounter, and desire to maintain or lose weight are highly correlated with weight categories. Interestingly, desire to maintain or lose weight are associated with lower probabilities of being underweight and higher probabilities of being overweight and obese. This indicates that respondents were aware of their weight and desired to attain a healthier body size. Finally, obese respondents had lower probabilities of favorable general health score (as would be expected).

While the intrinsic factors remain significant among adults, extrinsic characteristics, including household size, region of residence, income to poverty ratio, and maternal education, appear significant. Overweight and obese respondents are more likely to have large households and lower income ratio than those of normal weight. Higher probabilities of low level(s) of maternal education exist for overweight/obese adults. While having had an eating disorder was insignificant in earlier GLM models, logit estimates for both adolescent and adult females reflect significantly higher probabilities of underweight respondents with eating disorders and significantly lower probabilities of overweight respondents with eating disorder histories. These results appear substantial and significant for both age categories.

Using 15 years of data from NLSY97, longitudinal analysis assesses the impact of extrinsic-environmental, biological, geographic, and household-and intrinsic-sexual activity, substance use, desire to lose weight, etc.-characteristics on BMI. GLMs including fixed and random effects assess the relationship between these factors and BMI separately for both adolescents and adults. Results suggest that race and age are the largest BMI correlates at all ages. Multinomial logit models show that intrinsic factors are highly correlated with adolescent weight category. Overweight and obese respondents are more likely to express desire to lose weight than normal weight adolescents are, but underweight females are more likely to have suffered from an eating disorder than their normal weight counterparts. These intrinsic attributes, along with several household and geographic characteristics, determine adult weight categories.

Intrinsic factors are the most deterministic for adolescent BMI. Age of first sexual encounter, tobacco experimentation, perspective on general health, and weight-related desires to lose or stay the same weight are highly correlated with adolescent BMI and weight classification. As respondents become adults, intrinsic factors remain important, but several extrinsic characteristics also appear significant.

Results indicate that BMI correlates differ for adolescents and adults-suggesting that the BMI development process changes with age. While BMI is only influenced by a small number of factors for youth/adolescents, adult BMI is more complex, and it is influenced by a variety of household, environmental, demographic, and personal characteristics. These results imply that adolescents and adults would require different treatments for being overweight-therefore explaining why conventional policy interventions aimed at high BMI groups have been unsuccessful. More effective programs targeting adolescents should focus on internal, less tangible, characteristics.