Explore chapters and articles related to this topic
Evaluation of Water and Its Contaminants
Published in William J. Rea, Kalpana D. Patel, Reversibility of Chronic Disease and Hypersensitivity, Volume 5, 2017
William J. Rea, Kalpana D. Patel
The distributions of manganese concentrations in hair and water, as well as manganese intakes, were considerably skewed. They thus employed log10 transformation to normalize residuals. Similarly, they log-transformed the concentrations of other elements measured in water. Manganese intakes from consumption of water and from the diet were divided by the weight of the child for use in the analyses (micrograms per kilogram per month). They used generalized estimating equations (GEE) to examine relationships between exposure to manganese and children's IQ scores. GEE is an extension of generalized linear models for nonindependent data.321 These analyses were used to account for the community- and family-clustered data in our study. Some of the advantages of using GEE, instead of the more common approach of mixed models with random intercepts, include more efficient estimators of regression parameters and reasonably accurate standard errors (i.e., confidence intervals [Cis] with the correct coverage rates). With GEE, the computational complexity is a function of the size of the largest cluster rather than of the number of clusters—an advantage and a source of reliable estimates when there are many small clusters,322 such as in the present study (i.e., the 251 families). CIs were calculated with Wald statistics. An exchangeable working covariance matrix was used, with a robust estimator providing a consistent estimate of the covariance even when the working correlation matrix is misspecified.
How does the incompatibility of different vehicle types affect the odds of driver injury?
Published in Journal of Transportation Safety & Security, 2021
Hessam Arefkhani, Mohammad mehdi Besharati, Moslem Azizi Bondarabadi, Ali Tavakoli Kashani
Since each vehicle in a crash is considered both from the perspective of the driver’s vehicle type (DVT) as well as the other vehicle type (OVT), the injury outcomes may be correlated. If not appropriately addressed, this may result in biased parameter estimates (Savolainen, Mannering, Lord, & Quddus, 2011). The speed of involved vehicles may be one possible source of correlation between the drivers’ injuries in the same crash (Mannering & Bhat, 2014). To overcome this, a GEE should be used with logistic properties to accommodate the inter-crash injury outcome correlation. The GEE is a model estimation method which offers different approaches to handle correlation (such as independence, exchangeable, dependence, and autoregressive type 1) (Lord & Mannering, 2010). The exchangeable structure approach is used in this research. Interested readers may refer to Gail et al. (2007) for a detailed explanation of this approach. The statistical software SPSS V21 was used to perform statistical analyses and modeling.
Comparison of modelling methods accounting for temporal correlation in crash counts
Published in Journal of Transportation Safety & Security, 2020
Lai Zheng, Qinzhong Hou, Xianghai Meng
Because of the difference in addressing the temporal correlation issue, these five models can be grouped into two classes: subject-specific model and population-averaged model. As the name implies, the subject-specific model, including RENB, RPNB, MLNB, and NM models, is named after a subject-specific random component added to the model. This random component is allowed to vary with subjects but remains the same for observations within a subject. The GEE model is population averaged because the working correlation is assumed to be the same for all subjects, and the correlation reflects average dependence among the observations over subjects. The difference between subject-specific model and population-averaged model has been extensively studied in other disciplines (Hu, Goldberg, Hedeker, Flay, & Pentz, 1998; Hubbard et al., 2010; Stukel, 1993; Ma, Raina, Beyene, & Thabane, 2013). The fundamental difference is the way to interpret estimated coefficients. Coefficients for population-averaged model are effects on average across all subjects, while effects for subject-specific model are conditional on the effect for a particular subject. This difference will be further discussed below with the developed crash prediction models.
The impact of external facial features on the construction of facial composites
Published in Ergonomics, 2019
Charity Brown, Emma Portch, Faye C. Skelton, Cristina Fodarella, Heidi Kuivaniemi-Smith, Kate Herold, Peter J. B. Hancock, Charlie D. Frowd
Correct naming responses were analysed using the Generalized Estimating Equations’ (GEE) function within IBM SPSS Statistics. GEE is a regression-type approach that accommodates the presence of a within-participant factor. This approach is more statistically powerful than ANOVA given that it provides a combined by-participants and by-items model (Ballinger 2004). As the dependent variable (correct naming) is dichotomous, a binary logistic link function was used, and cell frequencies were initially checked as meeting the appropriate assumptions for this Chi-Square type of analysis [f(observed) > 0 and f(expected) < 5 for at most 20% of cells]. We chose a regression analysis that could detect variables whose influence is determined by the presence of other variables (suppressor variables). Therefore, we built a saturated model with predictors subject to backward sequential elimination; at each step, the variable possessing the smallest partial correlation with correct naming (the dependent variable) was removed (based on the established standard criteria of p > .1 and lowest X2). The model was built containing two factors, target type (coded as 1 = unaltered, 2 = internal-features, 3 = hooded top added, and 4 = hair changed) and naming task (1 = spontaneous and 2 = cued naming) and their interaction. For the Working Correlation Matrix, an Exchangeable structure was selected to model repeated observations for the within-participants factor, naming task. When built, standard errors (SE) of Beta (B) coefficients were inspected for markers of model instability (of which none were present).