Date: Thursday June 3rd
Visualizing Multivariate Hypothesis Tests: HE Plots and Reduced-rank Displays for Multivariate Linear Models
Speaker: Michael Friendly
Multivariate response designs are common, yet there are few methods available for visualizing the impact of explanatory variables on more than a single outcome. This talk describes graphical methods for multivariate response data, aimed at visualizing hypothesis tests in the multivariate linear model (MLM). In particular, I describe Hypothesis-Error (or "HE") plots, a new class of visualization methods designed to show the "size" and "shape" of covariation against a multivariate hypothesis (H), relative to covariation due to error (E) by data ellipses. For more than two response variables, these relations can be visualized in 3D HE plots, HE plot matrices, or in reduced-rank spaces that correspond to biplots and canonical discriminant spaces. These methods apply to all cases of the classical MLM, including multivariate analysis of variance (MANOVA), multivariate multiple regression (MMRA) and mixed cases including both categorical and continuous predictors (e.g., MANCOVA).
Date: Friday June 4th
Canadian Families’ Strategies for Employment and the Care of Pre-School Children
Using the 2006 Canadian Census we develop a typology of opposite-sex families' strategies for employment and the care for pre-school children. Multinomial regression is used to estimate the effects of the two partners' age and education on the strategy they choose. When the two partners differ in age or education, the effects are not simply additive and the direction of the difference -- whether the father or mother is older or has more education -- strongly affects the family's strategy. This is more consistent with the gendered perspective on the family than rational choice models.
Date: Saturday June 5th
Individual-Level Correlates of Physical Activity Behaviour Change among Family Heart Health Program Participants
Speaker: Dana Riley
The principal objectives are to determine whether the Family Heart Health intervention causes self-reported physical activity (PA) to increase over the 12-week intervention period, and to investigate the factors at an individual level that are associated with achieving PA targets during this period. Independent variables include demographic characteristics, height, weight, body mass index (BMI), and waist circumference. The covariates include intentions, perceived behavioural control, social support and normative beliefs, cigarette smoking, anxiety and depression. Individual-level data from all participants (both intervention and control groups; recruitment target n=450) will be analyzed. T-tests and chi-squared tests will be used to test for differences between groups (i.e., intervention versus usual care) as appropriate. Zero-order (bivariate) correlations will be performed between each demographic variable and PA at each time point (baseline and 3-month) to identify potential confounding variables. Multivariate regression techniques will be used to explore the predictors of achieving PA targets between baseline and 12 weeks. Univariate analyses will be performed to determine the predictors of PA using variables that have been identified as correlates of PA. Separate analyses of covariance (ANCOVA) will be conducted to investigate differences between intervention groups in PA at week 12, controlling for baseline scores.
The Effects of Child Care on Indicators of Child Development in Canada: Can We Produce Credible Estimates Using Non-Experimental Data?
Speakers: Gordon Cleveland and Barry Forer (University of British Columbia)
Using the National Longitudinal Study of Children and Youth (mostly Cycle 7), we find (in OLS regressions) that type of care, hours of care and use of multiple types of care have a pattern of significant effects on measures of cognitive, social, emotional and health status for preschool children. These appear to be different from the results found by a number of other researchers using the same data set. Given the concern that selection effects are likely to bias estimates of child care’s impacts, we are looking for suggestions about ways to test and establish the credibility of our findings.
Early Marriage Experiences and Later Life Satisfaction: The Case of the Chinese Oldest-old
This study examines early life experiences and their impacts on people’s life satisfaction after age 80 in the Chinese context. The dataset is the three cycles of the Chinese Longitudinal Study of Longevity Survey 1998 -2002. The early life experiences include birth place, years of education, adequacy of medical services, number of biological siblings, birth order, starvation in childhood, caregivers when sick, and living arrangements. In addition, we particularly focus on early marriage experiences like age at marriage, number of marriages and educational differences between spouses, and present marital status. A Generalized Estimate Equations model is applied in the separate analysis of male/female and rural/urban samples. Results from preliminary analysis indicate clearly the gendered patterns of the relationships between early marriage experiences and later life satisfaction: (1) those who remain married for their whole life without divorce are associated with higher levels of life satisfaction for women but not for men; (2) those females who married earlier are associated with lower levels of life satisfaction for women but not for men; (3) men, not women, who have less or equal years of schooling than their wives, are associated with higher levels of life satisfaction. Tentative results also indicate similar patterns for urban and rural samples.
Date: Monday June 7th
It Takes a Village to Improve Health: Methods for Increasing Our Focus on the Village in Maternal and Child Health Research
Speaker: Patricia O'Campo
Consideration of residential community context in our public health and epidemiologic research has emerged as common practice only within the last two decades. Recent advances in statistical software, which accommodate analyses of nested data and data at multiple levels (e.g., individual and neighbourhood), have enabled the widespread use of multilevel modeling to facilitate inclusion of area level data in regression models. This presentation will focus on the use of multilevel modeling on issues of maternal and child health. In particular, the application of multilevel modeling to issues related to intimate partner violence and pregnancy outcome will be presented. While methodological innovations have been critical to progress in the field, further advancement will require additional attention to theory and alternative approaches to modeling, both of which will be discussed in this presentation.
Date: Tuesday June 8th
The Limitations of Using School League Tables to Inform School Choice
Speakers: George Leckie and Harvey Goldstein
In England, so-called ‘league tables’ based on examination results and test scores are published annually, ostensibly to inform parental choice of secondary schools. A crucial limitation of these tables is that the most recent published information is based on the current performance of a cohort of pupils who entered secondary schools several years earlier, whereas for choosing a school it is the future performance of the current cohort that is of interest. We show that there is substantial uncertainty in predicting such future performance and that incorporating this uncertainty leads to a situation where only a handful of schools’ future performances can be separated from both the overall mean and from one another with an acceptable degree of precision. This suggests that school league tables, including value-added tables, have very little to offer as guides to school choice.
Date: Wednesday June 9th
Associations between Neighbourhood Cycling Infrastructure, Physical Activity and Obesity
Speaker: Ann Yew
With almost one quarter of the Canadian population being classified as obese, the built environment (the physical environment built and/or modified by humans) has been implicated as a cause of obesity. Promoting active transportation (walking, cycling and using public transit services to get to work or for utilitarian purposes) at the population health level may be one way of increasing physical activity to combat the rise in obesity. As a health enhancing and environmentally friendly mode of transport, the role of how cycling infrastructure in neighbourhoods is related to health outcomes is an important but under-researched area. This study will determine the relationship between the built environment, physical activity and obesity. Changes to physical activity and obesity will be measured over a 3-year period with relationships to the cycling infrastructure of neighbourhoods analyzed using multi-level regression models. Cycling infrastructure will be referred to as 'bikeability', a measure to assess features such as designated bike paths, road networks, steepness and barriers. We hypothesize that people living in neighbourhoods with better bikeability will have increased levels of physical activity and decreased obesity. For this study, 2754 adults (35-70 years) have been recruited from Metro Vancouver communities that vary in income and urbanization. Participants are part of the Prospective Urban and Rural Epidemiologic (PURE) study, an international 12-year investigation of the social and environmental determinants of obesity, diabetes and heart disease. By collecting physical activity and obesity measures over three years, this study will provide new information to assist policy makers in understanding the contributing role that bikeability has on increasing physical activity and decreasing obesity. By studying the communities in which people live, this research also holds potential value for developing obesity prevention strategies.
PISA and the Americas
Speaker: Todd Milford
The Programme for International Student Assessment (PISA) has been developed by the Organisation of Economic Co-operation and Development to provide participating nations with internationally comparative mean literacy scores in reading, mathematics, and science for students nearing the end of compulsory schooling. The number of nations from the Americas that are participating in PISA has been increasing over the past four administrations. This presentation provides an Americas-specific example of the ways in which the PISA dataset can be used for more exploratory analysis through the lens of policy and curriculum suggestions, rather than just league table comparisons. We used multilevel modeling of scientific literacy predicted on a number of school-level background variables. Results point to the importance of socioeconomic background of the student as well as the school in predicting scientific literacy across all nations in the study; however, other school-level significant predictors were more nation-specific. Interpretations of these models as well as next steps are discussed.
Occupational Stress as a Risk Factor for Heart Disease among Paramedics
Speaker: Imelda Wong
The Heart and Stroke Foundation of Canada has identified stress as a potential risk factor for heart disease. Stress in the workplace can be from a combination of environmental and psychological sources. The purpose of my study is to examine if specific factors associated with Paramedic work put them at a higher risk of heart disease. Work history data for 6,500 British Columbia paramedics (from 1940 to 2009) and emergency-run data for all BC stations (from 1989 to 2002) will be linked to records from BC hospitalization data, Medical Services Plan and Vital Statistics. Multiple conditional logistic regression will be used to examine the association of workplace exposure factors with acute and chronic coronary symptoms in separate analyses. Cases will be age and sex-matched to controls. The exposure variables include: job title (e.g., paramedic, dispatcher, other); shift type (e.g., rotating, day time only); type of run (e.g., H,M,L stress based on prior survey results); and workload (hierarchical data). In this model, the level 2 factor is the Station ‘busy-ness’, based on: number of annual runs, geography and population served, number of paramedics per station, shift lengths, time to respond to emergency run), and the level 1 factor is Individual ‘busy-ness’, based on average number of runs per month. Here are my questions: Is it possible to do a multiple conditional logistic regression with hierarchical data? What if the Level 1 and 2 factors are correlated?
Date: Thursday June 10th
Should One Use HLM or SEM for Modeling Growth?
Speaker: Brad Corbett
In 2005 I presented a SPIDA Lunch Talk demonstrating a multi-level growth model of youth smoking behaviours, using data from Canada’s National Longitudinal Survey of Children and Youth. One of the participants asked me how my hierarchical modeling approach compared to the structural equation modeling approach. The simple answer is: It makes no difference at all. Building comparable models yields the same results.
This presentation will reveal how simple growth models are built using HLM for multi-level modeling and AMOS for structural equation modeling. I will conclude with an examination of the model output from both methods. The data are repeated measures taken from a sample of 27 children. The study measured the physical distance between two points on the face of each child when they were eight, ten, twelve and fourteen years of age.