Name: Juliana Bottoni de Souza
Type: MSc dissertation
Publication date: 30/04/2013

Namesort descending Role
Valdério Anselmo Reisen Advisor *

Examining board:

Namesort descending Role
Antonio Carlos Monteiro Ponce de Leon External Examiner *
Neyval Costa Reis Jr. Internal Examiner *
Valdério Anselmo Reisen Advisor *

Summary: This dissertation uses two statisticals tools, Principal Component Analisys (ACP) and Generalized
Additive Model (GAM), jointly, to estimate the eect of the association between atmospheric exposure
of PM10, SO2, NO2, O3 and CO and the number of admissions due respiratory diseases in children
less than 6 years in the Regi~ao da Grande Vitoria, Brazil.Usually the atmospheric pollutants are considered
the explanatory covariables in MAG, but since they have an autocorrelation structure, they
must be used with caution. The PCA technique provides a new set of orthogonal variables, these variables
are linear combinations of environmental variables.Therefore, We use this approach in MAG,
hereafter denoted by GAM-PCA. However, the principal components obtained through the matrix
of variance / covariance applied to processes indexed by time also exhibit the properties of temporal
correlation. A countermeasure to attenuate the temporal correlation of the components is use the ltering
method to transform the data in an atmospheric white noise process. The residual matrix is used
to obtain these components and applied to the model MAG - method here called VAR-GAM-PCA.
The empirical results show that this model removes the autocorrelations of the main components and
indicates signicant estimates of relative risk (RR) for each pollutant. The results conrm the hypotheses
established, the main components have selected correlation and the autocorrelation lags. To
adjust the GAM-PCA model, an ARMA(p,q) model was used in the residual matrix since that structure
carried autocorrelation from the original data. The VAR model-MAG-ACP, besides producing
more signicant in RR estimates, generated best t residuals. Compared to the usual modeling MAG,
the two strands proposals presented better results, both in estimating the RR and the quality of the t.

Access to document

Acesso à informação
Transparência Pública

© 2013 Universidade Federal do Espírito Santo. Todos os direitos reservados.
Av. Fernando Ferrari, 514 - Goiabeiras, Vitória - ES | CEP 29075-910