|
on Discrete Choice Models |
| By: | Ben Dwamena (University of Michigan Radiology and VA Nuclear Medicine Service, Ann Arbor, Michigan) |
| Abstract: | We have previously demonstrated Stata implementation of bivariate ramdom effects meta-analysis of the sensitivity and specificity of a single binary diagnostic test by means of the midas module (Dwamena NASUG2007; Dwamena WCSUG 2007). In this presentation we extend our work to paired-comparison studies of two binary diagnostic tests. Using a dataset of studies comparing the accuracy of positron emission tomography(PET) and x-ray computed tomography (CT) for staging lung cancer, we compare the fit(deviance) and complexity (BIC, AIC) and test performance estimates (sensitivity, specificity, dignostic odds ratios nand likelihood ratios) of 4 multivariate models : (1) bivariate binomial mixed model with test type as fixed-effect covariate; (2) bivariate binomial mixed model with test type as random-effect covariate; (3) independent test-specific bivariate binomial mixed models; and (4) correlated test-specific bivariate binomial mixed models. Estimation is performed with the Stata-native procedure xtmelogit using both the default adaptive quadrature method and its laplacian approximation (nip=1). Results are then compared with those from the user-written gllamm command (Rabe-Hesketh et al.) |
| Date: | 2008–07–29 |
| URL: | http://d.repec.org/n?u=RePEc:boc:nsug08:11&r=dcm |
| By: | Högberg, Hans (Centre for Research and Development, Uppsala University and Country,Council of Gävleborg, Sweden); Svensson, Elisabeth (Department of Business, Economics, Statistics and Informatics) |
| Abstract: | Rating scales for outcome variables produce categorical data which are often ordered and measurements from rating scales are not standardized. The purpose of this study is to apply commonly used and novel methods for paired ordered categorical data to two data sets with different properties and to compare the results and the conditions for use of these models. The two applications consist of a data set of inter-rater reliability and a data set from a follow-up evaluation of patients. Standard measures of agreement and measures of association are used. Various loglinear models for paired categorical data using properties of quasi-independence and quasi-symmetry as well as logit models with a marginal modelling approach are used. A nonparametric method for ranking and analyzing paired ordered categorical data is also used. We show that a deeper insight when it comes to disagreement and change patterns may be reached using the nonparametric method and illustrate some problems with standard measures as well as parametric loglinear and logit models. In addition, the merits of the nonparametric method are illustrated. |
| Keywords: | Agreement:ordinal data; ranking; reliability.rating scales |
| JEL: | C14 |
| Date: | 2008–08–08 |
| URL: | http://d.repec.org/n?u=RePEc:hhs:oruesi:2008_006&r=dcm |
| By: | Högberg, Hans (Centre for Research and Development, Uppsala University and Country,Council of Gävleborg, Sweden); Svensson, Elisabeth (Department of Business, Economics, Statistics and Informatics) |
| Abstract: | Subjective assessments of pain, quality of life, ability etc. measured by rating scales and questionnaires are common in clinical research. The resulting responses are categorical with an ordered structure and the statistical methods must take account of this type of data structure. In this paper we give an overview of methods for analysis of dependent ordered categorical data and a comparison of standard models and measures with nonparametric augmented rank measures proposed by Svensson. We focus on assumptions and issues behind model specifications and data as well as implications of the methods. First we summarise some fundamental models for categorical data and two main approaches for repeated ordinal data; marginal and cluster-specific models. We then describe models and measures for application in agreement studies and finally give a summary of the approach of Svensson. The paper concludes with a summary of important aspects. |
| Keywords: | Dependent ordinal data; GEE; GLMM; Logit; modelling |
| JEL: | C14 |
| Date: | 2008–08–08 |
| URL: | http://d.repec.org/n?u=RePEc:hhs:oruesi:2008_007&r=dcm |
| By: | Tu Thuy Anh (Foreign Trade University, Hanoi, Vietnam); Dao Nguyen Thang (National Economics University, Hanoi, Vietnam); Hoang Xuan Trung (National Economics University, Hanoi, Vietnam) |
| Abstract: | <p>In this paper, we explore employment decision of Vietnamese farmers as having five choices: staying on the farm exclusively, staying in the village but partially engaging in local off-farm activities, and working outside the home region for a certain period, in which destination options are Hanoi, Ho Chi Minh City and Other which combines the remaining places. This choice model departs from the existing literature in several aspects. Firstly, previous papers focused mainly on the population that takes off-farm jobs or migrate, that are dichotomous employment choice. More importantly, most existing papers using the random utility model ignore factors in the destination areas. They assume implicitly that either migrants choose their destination randomly or that all migrants face exactly the same migration choices. In our paper, we allow multi-destination possibility, and examine impacts of distance, wages and social network on migrants' decisions. The indirect utility of a given migration option is modeled as a function of choice attributes and individual specifics. Choice attributes for each migration option include wage in destination area, transport between origin and destination area which is proxied by the corresponding distances, and social network of the migrants, while those for farm and non-farm option mainly include agricultural prices and local job creation opportunities. Individual specific include age, education, gender, marital status, share of children and elderly in the household.</p><p> The data used in this research are the Vietnam Living Standard Survey (1998) which is until now the only available data set that provides information on the migrant destinations. We start by estimating determinants of wage in destination areas using full information maximum likelihood to overcome selection bias. Then, we predict wages of those who do not currently work for wage. Finally, we run a conditional logit estimation with predicted wage being one of the explanatory variables to examine probability of migration to each location choice and of taking off-farm employment. Our results show that wage and network have significantly positive effects on all migration choices, while distance negatively affects them. Impact magnitude however differs across destination locations.</p> |
| Keywords: | Migration, choice attributes, off-farm employment, random utility model, conditional logit |
| Date: | 2008 |
| URL: | http://d.repec.org/n?u=RePEc:dpc:wpaper:2208&r=dcm |