|
Article Excerpt
As part of the National Nutrition Monitoring and Related Research Program, each of the U.S. Department of Agriculture (USDA) food and nutrient intake surveys provides a snapshot of the food choices made at a given time by the population of the United States. Information about trends in food and nutrient intakes by adults age 20 years and over and by children age 6 to 11 years has been published (Enns, Goldman, & Cook, 1997; Enns, Mickle, & Goldman, 2002). This article focuses on trends in intakes by adolescents age 12 to 19 years.
To examine whether adolescents' food intakes have changed over time, we compared nationally representative estimates from the most recent USDA survey of dietary intakes with similar estimates from two previous USDA surveys. The three surveys were the Continuing Survey of Food Intakes by Individuals (CSFII) 1994-96, (1) CSFII 1989-91, and the Nationwide Food Consumption Survey (NFCS) 1977-78 (Tippett et al., 1995; USDA, 1983, 1999, 2000a). The estimates reported in this study are of food intakes, the percentages of individuals consuming foods, and nutrient intakes for girls and boys age 12 to 19 years during all three periods. In the discussion of diet quality in the most recent survey, we cite information on intakes stated in terms of Food Guide Pyramid servings (USDA, 2000b).
Design and Methods
The Three Surveys
The CSFII 1994-96 was the most recent source of information on adolescents' intakes in the evolving series of USDA food and nutrient intake surveys that also includes the two earlier surveys (Tippett, Enns, & Moshfegh, 2000). Differences among the three surveys in sampling and methodology are discussed briefly in the following paragraphs. More information on methods in the NFCS 1977-78 and the CSFII 1989-91 is available elsewhere (Tippett et al., 1995; USDA, 1983).
The target population covered all 50 States in 1994-96 versus the 48 conterminous States in 1977-78 and 1989-91. In 1989-91 and 1994-96, the low-income population was over-sampled. In 1977-78 and 1989-91, all adolescents in sample households were eligible for inclusion in the survey; in 1994-96, selected individuals within each household were eligible. The number of adolescents age 12 to 19 years and the all-individuals Day-1 response rate, respectively, for each survey are 5,890 and 56.9 percent (NFCS 1977-78), 1,627 and 57.6 percent (CSFII 1989-91), and 1,469 and 80.0 percent (CSFII 1994-96).
In 1977-78 and 1989-91, dietary data were collected on 3 consecutive days by using a 1-day dietary recall and a 2-day dietary record. In 1994-96, the number of days was reduced to two, partly to reduce respondent burden (Tippett & Cypel, 1998). Both days of CSFII 1994-96 dietary data were collected with 1-day dietary recalls; interviews were on nonconsecutive days, 3 to 10 days apart, to ensure that nutrient intakes on the 2 days would be statistically uncorrelated. Between the earlier surveys and the CSFII 1994-96, the 1-day recall was modified to include multiple passes through the list of all foods and beverages recalled by the respondent, with the goal of improving the completeness of the data collected (Tippett & Cypel, 1998).
The USDA Survey Nutrient Database was updated on an ongoing basis to incorporate additional nutrients and improved nutrient values as well as to reflect changes in foods on the market (Tippett & Cypel, 1998; Tippett et al., 1995; USDA, 1987, 1993).
Presentation of Estimates
Because the number of survey days and the method of data collection on Day 2 differed among the surveys, tables comparing food and nutrient intake estimates among the surveys are based on only Day-1 data collected from each individual. Using these data maximizes comparability among surveys. One-day data are appropriate for comparisons of group means. All estimates are weighted to he nationally representative.
Mean food intakes are presented "per individual," meaning intakes include those by both consumers and nonconsumers of the food group. To calculate "per user" intakes of foods, researchers may divide the mean intake of a food group by the percentage of individuals using that food group, expressed as a decimal. Because only selected food subgroups are presented, subgroup intakes will not sum to the food group total. (2) Food mixtures were not broken down; mixed foods reported by respondents were grouped by their main ingredient. (3) One effect of this method of classifying food is the inflation of some food groups or subgroups (e.g., meat mixtures) and deflation of others (e.g., sugars and sweets) relative to the amounts they would contain if all ingredients were disaggregated.
Estimates based on a small number of observations or on highly variable data may tend to be less statistically reliable than estimates based on larger sample sizes or on less variable data. Standard errors may he used to calculate a measure of the relative variability of an estimate called the coefficient of variation, the ratio of the standard error to the estimate itself. Because the CSFII has a complex sample design, sampling weights and procedures for specialized standard error estimation were used in computing the estimates and standard errors (USDA, 2000a, documentation section 5). SAS version 8.2 (1999) and SUDAAN version 7.5.1 (Shah, Barnwell, & Bieler, 1997) were used for statistical calculations.
In the tables, we flagged estimates that are potentially less reliable because of factors such as small sample sizes or large coefficients of variation. The guidelines that were used for determining when a statistic may be less reliable involve the use of a variance inflation factor in the role of a broadly calculated design effect. Those guidelines have been described in detail elsewhere (USDA, 1999, appendix B). The variance inflation factors used in this study were 1.19 (1977-78), 2.26 (1989-91), and 1.41 (1994-96).
Approximate t tests were performed to determine whether food and nutrient intakes and the percentages of individuals using foods were significantly higher or lower in 1977-78 versus 1989-91, 1989-91 versus 1994-96, and 1977-78 versus 1994-96. All told, some 460 pairs of estimates were compared. Because the analysis involved such a large number of comparisons, we used conservative criteria for significance. When significant differences are discussed in the text, they may be referred to either as "changes" (or values may be said to have risen/fallen or to be higher/lower in 1994-96 than in 1977-78) or as "trends."
The term "change" is used only if intakes (or percentages using) in 1977-78 and 1994-96 were different when p was less than 0.001. The term "trend" is used only if two criteria were met: (1) mean intakes (or percentages using) either rose or fell progressively from one survey to the next (e.g., intake X rose between 1977-78 and 1989-91, then rose again between 1989-91 and 1994-96), and (2) p was less than 0.05 for both comparisons. For each trend, the level of significance noted in the tables (< 0.05 or < 0.01) is the one that is true of both the 1977-78 versus 1989-91 t test and the 1989-91 versus 1994-96 t...
|
