The Dataset Projectemmanuel.rousseaux.me/Rousseaux2012a.pdf · Introduction The Dataset package...

IntroductionThe Dataset package

Perspectives

The Dataset Project

Emmanuel Rousseaux

Institut d’études démographiques et du parcours de vieUniversité de Genève1211 Genève 4, Suisse

[email protected]

Rousseaux E. – The Dataset Project May 24, 2012 1/30

Perspectives

Outline

Introduction

The Dataset package

Perspectives

About the speakerNCCR LIVESGoals of the thesis projectWhat will be presented this day

Introduction

The Dataset package

Perspectives

Teaching and Research Assistant at the Departement ofEconomics, SES, Unige

PhD directed byI Gilbert Ritschard, iDemo, UnigeI Michel Léonard, ISS, Unige

Perspectives

(some) Research interestsI Decision TreesI Association rules miningI Nature-based optimization algorithmsI Health sociologyI Cognitive psychology

Perspectives

NCCR LIVES "Overcoming vulnerability: life courseperspectives"

I work within the IP14 "Measuring life sequences and thedisorder of lives" leaded by Gilbert Ritschard

Aims at providing ad hoc methods for life course analysis inorder to have more insight about dynamics of vulnerability

Perspectives

OverviewI Providing a software framework for handling survey data

in RI Providing a software framework for handling life courses

as a wholeI Providing new mining tools for rare events

I Decision trees for the discovery of vulnerable profilesI Multi-channel association rules mining

I Apply these tools for getting new insight on poor healthsituations

Perspectives

Motivation

I Currently no framework to handle survey data in RI Possible in SPSS, SAS, Stata

However

I In theses commercial softwares, no rigourous consistencytest

I No real standard for sharing datasetI A lot of methods in the state-of-the-art are provided on R

Perspectives

Motivation

However

Perspectives

Motivation

However

Perspectives

Motivation

However

Perspectives

Motivation

However

Perspectives

Motivation

However

Perspectives

Motivation

However

Perspectives

Motivation

However

Perspectives

I store and manipulate life courses dataI sophisticated managment of missing valuesI automatic consistency testsI representativity of the initial population testsI user-oriented functionsI automatic summaries

Perspectives

Representativity is central. Efforts have to be made for helpingthe user

I real structure for handling weights in the databaseI representativity checks on each variableI compute new weights to correctly balance a subdataset

Perspectives

Proposal: the ’Dataset’ project

I 2 librairies in R:I ’Dataset’: for cross-sectional survey dataI ’stDataset’: for spatio-temporal survey data

I Full S4I https://r-forge.r-project.org/projects/dataset/

Perspectives

IntroductionObject VariableObject DatasetPerforming a preprocessing stepNative analysis tools provided

Introduction

The Dataset package

Perspectives

In the Dataset package we mainly definedI the Variable object: store one measure on individualsI the Dataset object: store the full output of the survey

Perspectives

The Variable object is represented byI codes: vector of codes for each individualsI missings: vector specifying the coding of missings valuesI values: vector specifying the coding of valid casesI description: a label

Then the Variable is declined into different kind of measures

Perspectives

Variable (V)

Quantitative (V)

Scale svar()

Weighting wvar()

Timestamp tvar()

Categorical (V) cvar()

Ordered (V) Nominal nvar()

Ordered Binary bvar() Ordinal ovar()

Figure: Class diagramme of objects inheriting of the Variable class.

Perspectives

Standard builders for VariablesI svarI cvarI ovar

Perspectives

Demo 1

Working with Variable objects

Perspectives

The Dataset object is represented byI variables: list of variablesI name: name of the datasetI description: a long labelI row.names: names for rowsI weights: a variable used for weightingI control: some control variablesI infos: a list for storing other information the user want to

Perspectives

How to create a Dataset objectI from an existing data.frameI from a SPSS fileI by hand (as a list of Variable objects)

Perspectives

Demo 2

Preprocessing step: starting from a data set in a spss file, wewant get the data in R, et create our data set for our study

importing - recoding - exporting

Perspectives

Native operations on Dataset/Variable objectsI recoding VariablesI bivariate analysisI logistic regression

Perspectives

Demo 3

Launching analysis

bivariate analysis - logistic regression

Perspectives

For the Dataset packageReleasing the stDataset packagePresentations forthcoming

Introduction

The Dataset package

Perspectives

Finish up implementingI Weights managment toolsI Basic spatial toolsI Easy data capture of a surveyI (Detection/correction of bugs)

Perspectives

Taking spatial data into account

Figure: Poor/Good SRH ratio. PSM 2011, wave 2010 (no weighted)

Perspectives

stDataset: handling longitudinal survey dataI Same design as the Dataset packageI Longitudinal summary (in PDF)I Manipulating trajectories directlyI Construction of an object "life course" ready for analysis

Perspectives

Presentations forthcomingI (LACOSA) The ’Dataset’ Project (poster)I (SRSSS) Manipulating panel data with stDatasetI (SRSSS) Handling weights with Dataset and stDataset,

illustration with the PSM

Perspectives

Elements

of bibliography

Perspectives

Elements of bibliography I

[Voorpostel et al.] Voorpostel, M., Tillmann, R, Lebert, F., Weaver, B., Kuhn, U., Lipps, O., Ryser,V.-A., Schmid, F., Rothenbühler, M., and Wernli, B. Swiss Household Panel Userguide(1999-2010), Wave 12. Lausanne: FORS.(October 2011).

Perspectives

Thank you for your attention

Any question?

The Dataset Projectemmanuel.rousseaux.me/Rousseaux2012a.pdf · Introduction The Dataset package...

Documents

Transcript of The Dataset Projectemmanuel.rousseaux.me/Rousseaux2012a.pdf · Introduction The Dataset package...