Has the nutritional content of American food consumption in the 1990s improved since the 1900s, given that consumption habits and food availability have changed considerably? We believe that Americans are eating healthier now, in comparison to 1909, possibly because of technological and educational advances. Millen et al (1997) observed a trend of increasingly nutritious food consumption during recent years. In their study of population nutrient intake level, they conclude that while the consumption of some key nutrients was still below the recommended guidelines, overall diets in 1995 are closer to dietary recommendations than in 1991. (citation 1) We aim to extend Millen et al’s statement and determine if diets in the 1990s are healthier than in 1909.
In order to assess the validity of our belief we took two samples from two assumed-to-be different populations consisting of consumption levels of 17 key nutrients in 1994 and 1909.2 Using nutritional statistics for the nation as a whole, we used a calculator to randomly pick 10 vitamins and minerals. Our study is not completely random because some less-important nutrients were not included in the report from the United States Department of Agriculture (USDA). However, we believe that the exclusion of these “micronutrients” will not substantially impact our conclusions. Certainly, our study would not be random were we to survey only specific groups of vitamins or minerals, such as only B-complex vitamins.
The USDA’s study collected data concerning average consumption of specific vitamins and minerals by Americans during six points in time in the twentieth century. This was calculated by estimating total food supply in one year, determining the total amounts of various nutrients in the food supply, and dividing them by the population.
Although the USDA adjusts for pre-consumer shrinkage, an estimation error could occur because the USDA assumes that all food that reaches the consumer was in fact consumed.3 Our results thus could contain an estimation error, which would be proportional to the amount of food reaching the consumer that is not actually ingested. No gold standard check can be used to correct this possible estimation error.
Our data was provided in the actual units of a vitamin or mineral. In order to accurately compare increases or decreases in consumption, we converted all values to a percentage of the Daily Recommended Intake (DRI), which serves as a baseline. We used “bounds” of 0 and 1000 and re-verified the data to eliminate recording errors.
Validation
of the Inference Assumption
Skewness Test
The skewness test is used to determine if serious outliers exist by measuring how far the mean is from the median, in terms of standard deviations. No serious outliers are detected if the skewness test yields a value less than 0.5. We found no outliers, as the 1909 data had skewness of .13; 1994, .35. Therefore we proceed with the box plot test.
Box-and-Whisker Plot Test
The box plot is used to verify the skewness test in determining the existence of outliers. Any values lying beyond the horizontal line in the box plots below indicates an outlier. These boxplots detected one outlier, which exists in the 1909 data set, identified below, left. No outliers were found in the 1994 data, below right:
The existence of this outlier makes us less confident in our judgements based on our data.
Normal Quantile Plot
A normally distributed population is characterized by 90 per cent or more of sample data points falling within the confidence bounds, indicated below. Since all data points fall within these bounds, the normal quantile plot indicates that the populations from which our samples are drawn are normally distributed.
Ratio of Variance Test
The ratio of variance test determines if the two data sets have deviations similar enough to be useful for drawing comparisons—meaning that they are considered the same statistically. In order to meet this criterion, the larger standard deviation divided by the smaller standard deviation must be less than 3. Since our ratio of variance is 1.39, we believe that inferences can be drawn from the data.
Empirical Rule Analysis
When the statistical description (deviation and mean) of one
data set describes another, the data sets are considered the same for statistical
comparison purposes. The Empirical
Rule is thus used to determine if our two samples are drawn from the same
population. Assuming the population from which the data values are drawn is
normally distributed, the rule states that if 68% of data from one data set
falls within one standard deviation of the mean of the other data set, the
sets are considered to be the same. This
is done by superimposing the 1909 data values on to the 1994 range of 68%
and vice versa. We observe that 60% of 1994 data are above the 1909 range;
therefore, we believe that the data sets are different and that the 1994 data
are generally greater than the 1909 data, suggesting that consumption of nutrients
was higher in 1994 in comparison to 1909.
However, it must be noted that there is one
data point from 1909 which is above the 68% of the 1994 range. This particular point is our outlier identified
earlier, and it mitigates our confidence in our judgements.
Conclusion
We feel reasonably confident that Americans were consuming more nutrients in 1994 than in 1909. However, as the Empirical Rule Analysis above shows, consumption of some nutrients remains near 1909 levels. Nonetheless, consumption of other nutrients rose enough to confirm our initial belief that Americans are indeed eating healthier.
1 United States Department of Agriculture, Center for Nutritional Policy & Promotion, "Nutritional Content of the U.S. Food Supply, 1909-94: A Summary," at URL: <http://www.usda.gov/cnpp/
foodsupp.pdf>. Accessed March 28, 1999.
2 Millen, B.E., et al. (1997). "Population Nutrient Intake Approaches Dietary Recommendations: 1991 to 1995 Framingham Nutrition Studies," 97 Journal of the American Dietetic Association 742-53.
3 USDA ibid.