What led me to this project was my desire to assist the local community by turning health data into usable knowledge, as well as my aspiration to gain statistical consulting experience while working with a large real-life data set.

The response variable analyzed was food consumption and the explanatory variables were: sex, race, quarter, food group, stress, exercise, BMI, sleep quality and quantity. These variables were chosen based on interest in how they could relate to the change in dietary patterns over the first year of college.

After investigating multiple methods, a split-split plot design was used to determine a significant difference in food consumption between fall 2009 and spring 2010. This was done using PROC MIXED in SAS 9.2 ®. The results provided evidence that dairy consumption is significantly different for males compared to females in fall 2009. There was also evidence that dairy and fruit consumption significantly differ for males compared to females in spring 2010. However, this is still a work in progress and many issues were encountered during the analysis.

My project analyzed the FLASH data to investigate the relationship between various perceived variables and actual health measures for Cal Poly freshmen. The motivation for this analysis was an interest in both diet and exercise and its impact on an individual’s overall health. In particular, my interest lies in what a person perceives as their diet and exercise regimen and how that relates to overall health. To assess overall health, I examined both the Body Mass Index (BMI) and blood pressure of students. BMI was computed using the standard formula involving height and weight. Blood pressure was classified by using both systolic and diastolic blood pressure.

Conventional wisdom states that proper diet and exercise leads to better overall health. I was interested in the following research question: “Can we simultaneously model college students’ BMI and blood pressure using various lifestyle variables?” The response variables chosen were BMI and blood pressure and the explanatory variables examined consisted of various lifestyle variables such as diet preference, activity level, marijuana use, cigarette use, and alcohol use. Before I simultaneously modeled BMI and blood pressure, I created several models that had univariate responses.

My goal was to simultaneously model college students’ BMI and blood pressure using different lifestyle variables. Using the FLASH data containing the first time physical assessment with its survey from that corresponding quarter, I was able to investigate this question.

In my investigation, I used Discrete Multivariate Analysis to compute two separate generalized logit functions for each response, BMI and blood pressure, and Cluster Analysis to group lifestyle variables by their similarities to each other. By using Discrete Multivariate Analysis, I was able to take into account the relationship that existed between BMI and blood pressure. In each model, sex was a significant explanatory variable. Cluster Analysis illustrated that while certain variables can be grouped together, many lifestyle variables are different from each other.

While an incredibly useful method, the sample size limitations that exist make it difficult to create models with multiple explanatory variables. For future analysis, it would be interesting to see the association of the overall health measures, BMI and blood pressure, with other lifestyle variables. Additionally, with further data cleaning, it might be interesting to add more lifestyle variables into the cluster analysis to see if more clusters form.

Unfortunately, there are several challenges to using R in the classroom. Primarily, the user interface is designed around coding. This abstracts the important statistical concepts that students are trying to learn. Many times, students try to learn the code rather than learn the relevant statistics and end up not learning from the technology. This was the key motivation for my project. The idea was that I would build a graphical user interface (GUI) on top of R to allow students to explore statistical concepts and be exposed to R.

As such, I created five applets to allow students to explore simulation and distribution based tests found in introductory statistics courses:

1. 2x2 Table
2. 2x3 Table
3. rxk Table
4. Regression
5. One-Way Analysis of Variance

These applets have the benefit over other similar approaches in that they allow students to see the animation of the effect of each repetition on simulation of no association, students do not need to have any knowledge of code or programming to utilize the applets, and most importantly, students and instructors have the ability to use their own data sets.

After utilizing several of these applets in introductory statistics courses, students responded very positively. Specifically, a survey was distributed to students following use of the ANOVA applet on a lab and on a homework assignment. The survey indicated that the project was a resounding success. Further, all students recommended use of the applets in future statistics courses. Specifically, students felt that it was helpful in furthering their knowledge of ANOVA. Students felt that it was particularly helpful because of the step-by-step animation showed students the simulation as it was happening. It allowed students to understand what was occurring in a much more visual fashion. Some students even preferred the applets over other packages like Minitab that students were more familiar with due to the straightforward and visual nature. Finally, students also recommended that statistics professors use the applet in future courses, as well.

Adaptive design methods have been in practice since the 1970s, but have become increasingly complex ever since. One type of adaptive design is adaptive randomization. This is where the researcher makes changes in the way patients are randomized to treatment groups based on information gathered so far in the trial. Adaptations can be made to either trial procedures (eligibility criteria, study dose, treatment duration, study endpoints, laboratory testing procedures, diagnostic procedures, criteria for evaluation, and assessment of clinical responses) or statistical procedures (randomization, study design, study objectives/hypotheses, sample size, data monitoring and interim analysis, statistical analysis plan, and methods for data analysis).

The biased coin design, adaptive biased coin design and covariate adaptive randomization are specific types of adaptive randomization designs which are aimed at balancing certain aspects of a clinical trial. The biased coin design and the adaptive biased coin design aim to balance treatment group sample sizes by assigning the next patient to the group with the smaller sample size with higher probability. The biased coin design uses a fixed probability to achieve this whereas the adaptive biased coin design determines the severity of the imbalance between treatment groups using the total sample size to scale the difference in treatment group sample sizes. The probability of assignment to a specific group in the adaptive biased coin design thus depends on the ratio of difference in sample sizes between treatment groups to total sample size. Covariate randomization designs aim to balance the covariates across the treatment groups by assigning the next patient to the group that causes the smallest maximum imbalance across the covariate groups.

SAS® Macros to perform the adaptive biased coin design and covariate adaptive randomization are available upon request.

Factor analysis was used to categorize 33 food frequency variables into two categories – junk food and healthy food. Then, stepwise selection in a general linear model was conducted to identify lifestyle and demographic variables associated with these two food categories.

Statistically significant predictors of junk food consumption were sex, stress, average hours of sleep on the weekend, the interaction between sex and average hours of sleep on the weekend, and eating preference (vegetarian vs. non-vegetarian). When predicting the consumption of healthy food, the variables sex, days of moderate physical activity, days of vigorous physical activity, eating preference, and whether one perceives oneself as an early bird or a night owl were statistically significant. The study results show that how much junk food and healthy food a Cal Poly freshman student consumes per month is related to stress, sleep, and exercise, as well as the student’s sex and eating preference.