Nutrition 209/309: Statistical Methods for Nutrition Research

Nutrition 209/309 is a one-year sequence intended primarily for nutritional epidemiology and science students. Its mission is to take students with no prior statistical or computing experience and, in one year, make them capable of carrying out many of their own statistical analyses. In practice, this means teaching students basic statistical concepts, such as confidence intervals and signficance tests, and making them comfortable with t tests, chi-square tests, multiple linear regression, analysis of variance, and logistic regression. Not only must students be familiar with the techniques, but also they must be able to perform them by using a popular statistical software package, such as SAS (SAS Institute Inc., Cary, NC)

The inherent contradiction of Nutrition 209/309 is that it's both a first course and a last course. Most students will be seeing much of the material for the first time. However, despite the novelty of computing and statistical methods, the course must give students enough experience so that they can use the techniques effectively. In many ways, Nutrition 209/309 is a two semester abridgment of what takes three semesters in other universities (two semesters of theory and one semester of statistical computing). However, it is not equivalent to those three semesters. It achieves this coverage by leaving things out, not by trying to do in two semesters what really needs three.

Nutrition 209 focuses on understanding variability and how statisticians use that understanding to develop methods that identify general trends in data despite the presence of variability. Nutrition 309 focuses on using linear models (regression, analysis of variance, analysis of covariance, logistic regression) to describe these general trends.

Even though final grades are awarded at the end of each semester, Nutrition 209/309 is viewed a one-year sequence. Its purpose is to teach students how to apply statistical methods. It is not about covering a specific number of pages or techniques in a fixed amount of time. The number of topics that are covered in each semester will vary year-to-year depending on the particular class. The list that follows is typical of what is covered. Additional topics will be added if time allows. They are likely to be selected from clustering analysis, the pitfalls of factor analysis, and others judged appropriate by the Directors of the Science and Epidemiology programs.

• understand basic ideas of study design--including randomization, (double) blinding, and controls--their importance, and how they relate to "cause & effect".
• understand the importance of audit trails and how to maintain them.
• understand the importance and use of graphical displays such as histograms, box plots, dot plots, scatterplots, and normal plots.
• be able to construct the tables and figures of summary statistics that are commonly used in scholarly publications and reports.
• understand the proper use and interpretation of confidence intervals.
• understand the proper use and interpretation of observed significance levels (P values).
• be able to distinguish between independent and paired samples.
• be able to perform Student's t test (for independent or paired data) to test the equality of two population means and recognize those situations where Student's t test is appropriate.
• be able to perform Pearson's goodness-of-fit test to compare population proportions and to recognize situations where the test is appropriate.
• be able to estimate the sample size required to compare two population means or proportions by using independent samples.
• be aware of other options that can be used when the conditions necessary for the validity of standard procedures are not met. They include the median test and the Mann-Whitney U test (Wilcoxon rank sum test) as alternatives to Student's t test for independent samples, the sign test and Wilcoxon signed rank test as alternatives to Student's t test for paired data, and Spearman's rank correlation as an alternative to the Pearson correlation coefficient.
• understand the use of simple linear regression for prediction and adjustment.
• be able to use a statistical program package such as SAS to perform the tasks listed here.

After taking Nutrition 309, students should

• understand the use of multiple linear regression for prediction and adjustment.
• know how to interpret regression coefficients and associated F and t tests.
• understand how to use residuals to investigate lack of fit of a regression model.
• understand the consequences of collinearity when fitting regression models.
• be aware of stepwise methods for fitting parsimonious regression models and the limitations of these methods.
• understand the use of the extra sum of squares principle to compare two regression models.
• understand how to compare two measuring devices and the way it differs from regression analysis.
• understand the regression effect and the regression fallacy.
• understand single factor analysis of variance and a variety of multiple comparison procedures for comparing factor levels, including Fisher's least significant differences, Tukey's honestly significant differences, Newman-Keuls multiple range test, and the Bonferroni adjustment.
• understand the distinction between crossed and nested factors and between fixed and random factors.
• understand issues peculiar to multi-factor analysis of variance, such as interactions and imbalance, and how to deal with them.
• know how to perform repeated measures analysis of variance.
• be able to use analysis of covariance to compare levels of experimental factors while adjusting for the effects of other predictor variables.
• be able to use logistic regression to model or predict the probability that a subject has a particular characteristic.
• be able to use Poisson regression to model or predict outcomes that are counts.
• be able to use a statistical program package such as SAS to perform the tasks listed here.