Model Building. In order to assess the estimates used from logistic regression, a stepwise selection method was used for each outcome of interest (respiratory, gastric, or either reason), each time period (1, 2, or 3) and region of interest (overall, Brevard, Volusia, St Lucie, and Xxxxxx counties). While different covariates were significant in each model, there were overall trends that can be seen in Tables 27-29 in the Appendix. Gender was found to be significantly associated with admission for both gastric and respiratory reasons in most models. Those models where gender was not included were all modelling gastric admission except for one – respiratory admission in Brevard County during the third period (Table 29). In every case, the odds of admission to the hospital was higher for females as compared to males for both gastric and respiratory reasons. Race was categorized as white, African American, and other – white was treated as the reference category. When modelling to examine the odds of admission for gastric reasons, in models where race was significant, the odds are lower for African American individuals as compared to white individuals. For the first two periods, the odds of admission were higher for individuals in the other race category – when the association was significant. For respiratory reasons, in the second and third periods, the direction of the association shifted and the odds were higher African Americans and lower for those in the other race category. When looking at the age categories that were previously used, the age group 0-9 was used as the reference category. For young adults (aged 20-24), the odds of admission were lower for every single model regardless of time, area, or reason for admission. For teens/older children (aged 10-19), the odds of admission were lower as compared to young children for gastric reasons but as higher for respiratory reasons. When looking at the months of each bloom, the seasonal pattern that we observed earlier appears again. Monthly patterns were examined with the first month being used as the reference (May, January, and June for each respective period). For the blooms that took place during spring/summer/fall months (the first and third periods), rates were higher for the reference month in all cases when the covariate was significant, except in two models examining the odds of gastric admission during the first period. For our second bloom period, during winter months, the odds of admission for respiratory r...
Model Building. Prior to conducting the analysis, data were cleaned using an iterative three phase process of screening, diagnosing and editing (Xxx xxx Xxxxxx et al., 2005). Screening involves using univariate and bivariate analyses and graphs to identify relationships among variables and “lack or excess of data, outliers, strange patterns in joint distributions and unexpected analysis results and other types of inferences and abstractions" (Xxx xxx Xxxxxx et al., 2005; Wilkinson, 1999). Subsequently, we diagnosed potential issues (e.g., data is outside or inside the range of possible measurements) and treated the issues by correcting, deleting, or leaving the values and documented our choices. We also examined focal policies of interest for multicollinearity by calculating the Variance Inflation Factor (VIF). A VIF of 10 or greater was considered indicative of a potential issue suggesting that TANF policies should be examined in separate models. The VIF values ranged from 1-3.5, suggesting no such issue.
