Speaker: Tracy Dong, Graduate Student, UW Biostatistics
Abstract: Count time series are commonly seen in health surveillance systems and social economic studies. Many models have been considered in the literature to adequately analyze this type of data, but little has been done to deal with the problem of under-reporting. We introduce a technique to analyze correlated under-reported count time series through first order integer autoregressive (INAR(1)) hidden Markov chain models. Two methods of parameter estimation are proposed: a naïve method based on autocorrelation functions and the maximum likelihood method based on a revised version of the forward algorithm. In addition, the most-probable unobserved time series are reconstructed by means of the Viterbi algorithm. Two examples of application in the public health field are discussed to illustrate the utility of the models. Some simulation results are also discussed to examine the model setup and parameter estimation procedures. This talk is based on the 2016 paper by Fernández-Fonte.