Metropolis approximation to assess dependence between fixed and random effects in a count model for overdispersed data

Simple item page
dc.contributor.advisorMüller, 1963 August 9-, Peter
dc.contributor.committeeMemberLin, Tse-min
dc.creatorMaresca, Philip Joseph
dc.creator.orcid0000-0002-3387-8317
dc.date.accessioned2016-12-19T23:28:59Z
dc.date.accessioned2018-01-22T22:31:12Z
dc.date.available2016-12-19T23:28:59Z
dc.date.available2018-01-22T22:31:12Z
dc.date.issued2016-05
dc.date.submittedMay 2016
dc.date.updated2016-12-19T23:29:00Z
dc.description.abstractThis analysis will provide inference on extra-Poisson variation in annual tornado counts from 12 US states, under the assumption that the counts arise from a Poisson distribution. Parameterizing the rate of occurrence as a function of time varying covariates including Sea Surface Temperatures yields a fixed effects Poisson model. We use such a model as a vehicle to test counts for equal mean and variance-a condition called equidispersion. The rejection of equidispersion prompts the introduction of state-specific random variability in the Poisson regression model to get a negative binomial model. We find that in the presence of the random effect, Sea Surface Temperatures becomes insignificant while Year remains significant. Assessing the relation between Year and the state-specific random effects proceeds by estimating the joint posterior full conditional of the regression parameters for different levels of the parameter representing state heterogeneity.
dc.description.departmentStatistics
dc.format.mimetypeapplication/pdf
dc.identifierdoi:10.15781/T2KW57M9H
dc.identifier.urihttp://hdl.handle.net/2152/43993
dc.subjectNegative binomial
dc.subjectMetropolis
dc.titleMetropolis approximation to assess dependence between fixed and random effects in a count model for overdispersed data
dc.typeThesis
dc.type.materialtext

Files

Collections

University of Texas at Austin