Publication Type: Paper/Poster Proposal
Abstract: The increasing uncertainty of the US public’s reaction to survey requests has led to heightened awareness of the interplay of costs, nonresponse rates, and nonresponse errors in large scale household surveys. Groves and Heeringa (2004) outline an approach to survey design that directs orderly changes to key features of the recruitment protocol of a survey, based on real-time analysis of the incoming survey data. These so-called “responsive designs” identify a set of alternative key statistics, callback alternatives, and incentive options prior to the start of the data collection. Cost and error-related models are measured during the early phases of a survey, and then used to determine final design features, more nearly cost-optimal. In short, these designs adapt or respond to real-time information about the performance of the survey. A key tool in this responsive process is the use of propensity models on the sample case level, estimating the probability that an active case will be interviewed.

Propensity models are also used after the survey data collection period has been completed. These propensity models are often used to form weighting class adjustments in an attempt to reduce unit nonresponse error (Little, 1982). These models identify groups that have higher or lower likelihood of being measured, based on all knowledge available at the end of the data collection.

This paper addresses whether the predictors of propensity during data collection heavily overlap those available at the end of the data collection period. It relates this analysis to efforts during the data collection to attempt to achieve the most cost efficient acquisition of completed cases. It then studies how the estimated propensities of cases correlate with key statistics in the survey, among respondents. Conclusions are drawn about the relative utility of observational and process data predictors for response propensity in adjustment models.

Publication Type: Conference Paper/Unpublished Manuscript
Abstract: This study is a theoretical review about message discrepancy and a proposal of a new model of message discrepancy. This study analyzed four mathematical models of message discrepancy (Anderson & Hovland, 1957; Fink, Kaplowitz, & Bauer, 1983; Fishbein & Ajzen,1975; Laroche, 1977). A new mathematical model was proposed by solving a differential equation that derived from on a biological metaphor and existing models. A new model predicts a monotonically increasing function of message discrepancy.

Publication Type: Conference Paper/Unpublished Manuscript
Abstract: This paper provides a model to foresee some potential problems of lay participation in the current on-going justice system reform in Japan. In order to achieve the popular base of the justice system, a new lay-participation system, Saibanin-seido (civil judge system), is to be introduced in criminal proceedings in 2009. In this system Japanese lay judges will serve a term of only one case; however, they will deliberate and decide the case together with professional judge(s). As the deliberation body will generally consist of three professional judges and six lay judges, the actual and active participation of civil judges may be difficult to achieve.
The model called Linguistic Legal Model, which is a combination of Layer Analysis and Register Analysis, is intended to explain the discrepancy between legal experts and lay people. Layer Analysis frames a theatre-type concept to gain different types of perception affected by a variety of norms of differing layers. Register Analysis aims to identify linguistic characteristics of one legal genre, judgment paper. Layer Analysis elucidates the structure of law and society, whereas Register Analysis examines the structure of legal language. The model is like a two-wheeled vehicle of Layer and Register Analyses, in which the two wheels are indispensable for the operation of the vehicle. With the movement of the two wheels Linguistic Legal Model can illuminate the complex interaction among law, language, and society.

Publication Type: Conference Paper/Unpublished Manuscript
Abstract: Historically, sociologists focused almost exclusively on Durkheim’s idea that social integration is associated with suicide. Gabriel Tarde provided an opposing view suggesting that social behavior is adopted through imitation. This paper examines the impact of imitation on suicidal behavior in a concentrated urban setting. Data were drawn from official police statistics from the Toronto Police Service and included all occurrences of suicidal behavior over a 60-month period in 16 police divisions located in Canada’s largest city. Contextual characteristics derived from census data were constructed using factor analysis and revealed four independent factors related to neighborhood stability, material deprivation, ethnic diversity and dependency. Using dynamic simultaneous equation models we estimated the temporal effects of suicide rates between first-order (common boundary) spatial neighbors. This method allowed us to establish the direction of the association between suicide rates in one geographical area on rates in adjacent areas. Findings show significant temporal and spatial-temporal effects across police divisions. These relationships remain significant despite adjusting for contextual characteristics. The methodology employed in this paper provides a statistical approach that enables us to model simultaneous temporal and spatial patterns in suicidal behavior. Findings suggest a potential imitative or ripple effect in suicidal behavior across police divisions. Unraveling the social factors that lead to imitation may be important for suicide prevention programs.

Publication Type: Conference Paper/Unpublished Manuscript
Abstract: Choice set models assume that voters first narrow down the universe of alternatives (parties) to a smaller subset and then select one alternative from within this set. We laid out this general idea and its statistical implementation in a previous paper. In the current paper, we extend the basic model in two ways. First, we allow for different screening rules. Some of these rules are compensatory, in that they allow for trade-offs between attributes of the parties (e.g. distances from the voter on issues). Other rules are non-compensatory, in that alternatives are eliminated as soon as they fail on one attribute. Our second extension is to allow for voter heterogeneity in the thresholds of acceptability of parties, using Dirichlet process priors. These extensions render the basic choice set model even more flexible, albeit also more difficult to estimate. We apply the model to the 2002 Dutch parliamentary elections. The goal is to unravel how Dutch voters determined their vote choice in 2002 and whether certain groupings of voters can be found in terms of thresholds of acceptability. Apart from the methodological innovation in this paper, we are the first to apply choice set models to vote choice in the Netherlands.