The Methods and Research meetings will take place at 12:30 in room 5799 of the Geopolis building in the Mouline quarter. Also see here
Presentation on 25.09.2012
Model-based Clustering and Classification ; presentation of the Mixmod software
[Gilles Celeux (INRIA, France)]
This talk is devoted to the presentation of the mixture model for cluster analysis, semi supervised and supervised classification. The advantages of this model to answer the selection issues involved in clustering and classification with continuous, discrete or mixed data will be highlighted. EM-like algorithms to derive the parameter estimates will be presented. The numerous versatile and parsimonious useful mixture models will be described. Some emphasis will be placed on the relevant model selection criteria, depending of the focus of the analysis, ground in this model-based setting. The features of the Mixmod software (www.mixmod.org) and its recent interface RMixmod with R (cran.r-project.org/web/packages/Rmixmod/index.html) will be presented.
Presentation Gilles Celeux
Presentation on 30.10.2012
The R-indexes as a complementary measure of data quality ? Application to the Swiss ESS 2010 data
[Caroline Roberts and Caroline Vandenplas (FORS/UNIL)]
The only indication of data quality found in many reports is the response rate. It is however known that response rates are a poor estimator of non-response bias and this practice is more and more criticized. Therefore, researchers in the Netherlands have developed an alternative quality measure : R-indexes or R-indicators where “R” stands for representativity. These measures are based on variables known for respondents and non-respondents (typically socio-demographic variables). Since 2010, some FORS surveys have benefited from the Swiss Federal Statistical Office register as a sampling frame. This register contains some information like gender, age or marital status that are thus available for all sampled individuals, respondents and non-respondents. Some of the advantages and remaining issues of this new measure will be illustrated using ESS 2010 data.
Presentation Caroline Vandenplas
Presentation on 27.11.2012
First steps towards a software framework for handling life course survey data in R
[Emmanuel Rousseaux (University of Geneva)]
Population studies strongly rely on survey data and much time is needed to prepare the data. The job is even harder for longitudinal data and network data. From a general point of view, the Dataset project aims at facilitating the management of survey data by providing researchers in social sciences with high-level tools for storing, sharing, exploring and recoding survey data in a secure and efficient way. The toolbox comes as a series of R packages. The software allows to store detailed descriptions for each variable and variable values and then can generate a full codebook directly exported as a PDF file that proves particularly useful for sharing data with others. As key functionalities, the toolbox accepts user-defined missing values, natively account for weights, helps to search for specific variables across the whole data base, and process automatic checks to prevent the loss of representativeness when filtering out cases. The software also provides some specific data-analysis tools such as bivariate association measures, front-end to tree-based methods and a logistic regression method. For all methods provided the toolbox generates nice summary tables easily exportable in PDF format.
In this presentation we introduce the main functionalities of the Dataset toolbox, and discuss the functionalities forthcoming for specific handling of panel data, especially for life course analysis. We will use data from the Swiss Household Panel.
Presentation Emmanuel Rousseaux
Presentation on 11.12.2012
Etudier l’engagement militant à l’aide de récits de vie. Quelques réflexions tirées d’une recherche sur l’UDC
[Philippe Gottraux (UNIL)]
Sur la base d’une recherche achevée consacrée à l’engagement militant au sein de l’Union démocratique du centre (Philippe Gottraux et Cécile Péchu, Militants de l’UDC. La diversité sociale et politique des engagés, Lausanne, Antipodes, 2011), nous reviendrons sur diverses opérations de recherche (construction de l’objet ; types de données et modalités d’analyses ; réflexivité, etc.) et accessoirement sur quelques résultats. L’idée sous-jacente sera alors de montrer les apports d’une démarche qualitative pour l’étude d’un tel objet politique, apports trop souvent impensés ou sous-estimés dans la science politique suisse « classique ».
Presentation Philippe Gottraux
Presentation on 29.01.2013
Spatially weighted context data with the R package spacom :
Studying the indirect impact of war on well being of young adults in ex‐Yugoslavia
[Sandra Penic (UNIL)]
In multilevel models, geographic space is divided into discrete, often arbitrarily defined units with fixed boundaries. Such a practice ignores the fact that there are usually important social, economic, and institutional ties that link residents from different units. The closer that these units are and/or the stronger the links between their residents, the more they are likely to influence each other. In order to address this issue, a new approach to contextual analysis has been developed : the spatially weighted context data approach. The approach complements classical multilevel analyses, by allowing for the study of spatial dimensions of contextual influences. The R package spacom has been developed by an interdisciplinary team in Lausanne to facilitate the use of the spatially weighted context data approach. Spacom’s functions can be used to construct and describe spatially weighted context data, to introduce spatially weighted contextual indicators in multilevel models (and estimate models through bootstrap procedures that provide robust point estimates and standard errors), and to diagnose spatial dependency in residuals from multilevel models. In my talk, I will illustrate the use of spatially weighted context data with spacom by the study of the impact of collective war victimization and economic exclusion on life satisfaction of young adults in post-war Yugoslavia.
Presentation Sandra Penic
Presentation on 26.3.2013
Cost-benefit estimation when choosing an apprenticeship offer
Evidence from a choice experiment
[David Glauser (University of Berne)]
Despite the fact that a high portion of juveniles attending vocational education and training (VET) after compulsory school in German speaking cantons, little is known about why juveniles opt for certain apprenticeships and how they balance pros and cons when they can choose between several apprenticeship offers. The aim of the presentation is to shed light on these questions using a survey-based choice experiment. Data from the DAB-Panel study (http://www.berufswahl.unibe.ch) on vocational and educational decisions at the end of compulsory school is used. The effects of the attributes used in the choice experiment are estimated with conditional logit models.
Presentation David Glauser
Presentation on 30.4.2013
Child development and child care : Marginal treatment effects
[Rafael Lalive (UNIL)]
Many countries are currently expanding access to child care for young children. But are all children equally likely to benefit from such expansions ? We address this question by adopting a marginal treatment effects framework. We study the West German setting where high quality center-based care is severely rationed and use within state differences in child care supply as exogenous variation in child care attendance. Data from the German Socio-Economic Panel provides comprehensive information on child development measures along with detailed information on child care, mother-child interactions, and maternal labor supply. Results indicate strong differences in the effects of child care with respect to observed characteristics (children’s age, birth weight and socio-economic background), but less so with respect to unobserved determinants of selection into child care.
Underlying mechanisms are
a substitution of maternal care with center-based care, an increase in average quality of maternal care, and an increase in maternal earnings.
Presentation Rafael Lalive