For many people, statistics is about simplifying systems so that some standard test (e.g., the t-test) can be applied. Over the last decade, new approaches to doing statistics have emerged that can respect the intrinsic complexity of systems. I'll introduce a couple of these approaches that focus on networks of influences: multiple regression with large numbers of explanatory variables and the topological analysis of causal networks. You don't need to know much about conventional statistics to be able to follow the talk.
This talk presents data from several experiments in neurobiology. In one type of experiment, neurobiologists insert small electrodes into the brains of living animals. The electrodes are fine enough to record electrical signals from individual neurons, so we can record when each neuron fires. Such data are called spike trains. I will show what spike trains look like, how they can be used to learn how neurons work together and how they can be used to learn how the brain encodes various sensations like taste. In another type of experiment, neurosurgeons working on human epileptic patients take a series of digital photographs of the brain. I will show how these photographs can be used to follow the path of an electrical signal across the surface of the brain and determine which regions of the brain do and don't respond to the signal.
Atul Gawande (surgeon at Brigham and Women's Hospital and faculty member at Harvard School of Public Health) asserts that research on our health care system can save more lives in the next decade than bench science, research on the genome, stem-cell research, cancer vaccine research, and everything else we hear about on the news." Janet Rosenbaum will discuss how public health addresses virtually every medical and social issue, and it does so at the population level where it can have the greatest impact. The speaker will outline problems in public health, and give examples from her work on adolescents' lying about sex and virginity pledges, and using tobacco taxes to reduce smoking in the developing world and New York City.
The US spends $200 million annually on abstinence programs, including virginity pledges. Using data from Add Health waves 1-3, adolescents reporting a wave 2 virginity pledge (n=289) were matched with non- pledgers (n=645) using exact and nearest-neighbor matching on wave 1 factors including pre-pledge religiosity, attitudes towards sex and birth control, and family context. Wave 3 outcomes were compared. Five years post-pledge, 84% of pledgers reported having never taken a pledge; pledgers and matched non-pledgers did not differ in premarital sex, sexually transmitted diseases, and oral and anal sex. Pledgers had 0.1 fewer past year partners, but the same number of lifetime sexual partners and same age of sexual initiation. The sexual behavior of virginity pledgers do not differ from that of comparable non- pledgers, but pledgers were less likely to use birth control and condoms in the past year and birth control at last sex. Virginity pledges may not affect sexual behavior, but may decrease the likelihood of taking precautions during sex.
The objective was to identify predictors of becoming eating disordered among the 9,038 females and 7,843 males, aged 9-15 years at baseline, in the ongoing Growing Up Today Study. Participants were sent five self-report questionnaires over seven years. Participants were included in the analysis if they completed at least two of the questionnaires, did not report engaging in purging or binge eating at baseline, and their mother provided information on her history of an eating disorder, thus leaving 6,916 females and 5,618 males for analysis. The outcomes were the onset of starting to binge eat or purge (i.e., vomiting or using laxatives) at least weekly. During seven years of follow-up, 4.3% of females and 2.3% of males started to binge eat and 5.3% of females and 0.8% of males started purge to control their weight. Few participants started to both binge eat and purge. Rates and risk factors varied by gender and age group (< 14 vs. ≥ 14 years). Females less than 14 years of age whose mothers had a history of an eating disorder were nearly three times more likely than their peers to start purging at least weekly (odds ratio, 2.8, 95% confidence interval, 1.3-5.9), but maternal history of an eating disorder was unrelated to risk of starting to purge or binge eat among older adolescent females. The impact of missing data was explored by evaluated models with imputed missing values and comparing the results to those restricting the analysis to participants with observable data. In both sets of models, the same predictors were identified, although the odds ratios were generally slightly larger in the models using imputed data.
Antiretroviral therapy (ART) is becoming increasingly available in resource-poor settings, turning what used to be a certain death sentence into a manageable chronic disease. The emergence of drug resistant HIV, however, has the potential to deeply compromise the benefits of ART programs. In the United States, resistance assays are typically conducted prior to initiation of therapy, allowing physicians to tailor treatment regimens to patients' resistance profiles. Unfortunately, drug resistance genotyping is too expensive for widespread clinical use in the developing world. Instead, in many resource-poor countries, all patients receive a standardized ART cocktail. In these settings, population-level surveillance of drug resistance is needed to characterize the prevalence of resistance mutations and to enable ART programs to select the optimal regimen for the local population. Conducting resistance assays on pooled sera samples increases the accuracy of drug resistances prevalence estimates while simultaneously decreasing the cost of surveillance. We present a Bayesian hierarchical model for pooled-testing data which accounts for uncertainty about the inter-subject heterogeneity of resistance levels as well as for the measurement error of resistance assays.
Jane Garb's current focus is on GIS applications in healthcare, particularly the relationship of social and demographic factors to health as well as the use of census data in healthcare preparedness. Her work further encompasses extensive research in cancer, cardiovascular and trauma epidemiology.
Thanks to the Smith College Lecture committee, Department of Mathematics and Statistics, and Office of the Dean of the Faculty for support of the series.
Applied Statistics Lecture series (2006-2007)
Other 5 college seminars of interest:
University of Massachusetts Statistics and
Probability Seminar Series
Organized by Nicholas Horton.
Last updated April 7, 2009