The analysis of discrete states in the psychological sciences are commonplace. One of the most powerful techniques used to analyze transition between states is the Markov model. The Markov model is composed of states and transitions, the transitions measure the probability from moving from one state into another at any point in time. Alternatives to the Markov model include the semi-Markov model, which relaxes some of the assumptions made in the Markov model, and may make it more appropriate for the analysis of behavioral data. As the acquisition of these types of data are now easier given recent technological developments, such as the ubiquity of smartphones, streams of states can now easily be acquired from multiple individuals. Multilevel modeling is capable of pooling across individual's with heterogeneous characteristics to make inferences that are true across the population. The goal of this dissertation is to synthesize three perpetrate streams of modeling together to propose an alternative methodological framework for the analysis of intensive longitudinal data composed of discrete states. The three methodological streams include multilevel modeling, survival analysis, and Markov models. Multilevel modeling provides a framework to make inferences about a population when data are composed of clusters. Survival analysis describes a framework which can be used to estimate when an event is most likely to occur, and incorporates an inferential framework to identify variables which may increase or decrease the likelihood of these events across time. Finally, the Markov model describes a time series analytic framework which is used to identify the probability of a state transition to occur at a specific point in time. By synthesizing these three streams, a multilevel Markov, or semi-Markov model can be estimated in an efficient fashion and can identify predictor variables which influence transition probabilities, and the timing of these probabilities.

In order to showcase the utility of these methods, both a simulation study, and an empirical study are examined. The simulation study is used to examine the resilience of time-to-event models estimated under various simulation factors, and to compare the performance of Markov, semi-Markov, and their multilevel counter parts to estimate the true parameters. The empirical study examines the pre- and post-treatment effects when comparing case versus intervention cohorts and their verbal dynamics during a structured parent-child interaction task. The goal is to examine difference is positive and negative behaviors after the administration of a structured parent child interaction therapy.