The project is put forward by Researchers who come from both the Academia and National Research Institutes and are divided into two research units (University of Perugia, UNIPG, and University of Pisa, UNIPI). Researchers in the two units come from different Universities and each one has been involved because of his/her experience and skills to deal with particular challenges of the project. Most of the Researchers involved have already successfully worked together and produced important contributions to estimation from sample surveys. UNIPG hosts the fundamental contribution of those Researchers from ISTAT who are in charge of the production of estimates for small areas from EUSILC and LFS, and of a Researcher from the Bank of Italy who is in charge of the design and the estimation strategy for SHIW. The PI has already collaborated with the Researchers from ISTAT and the Bank of Italy on estimation issues linked to small area estimation from the Italian LFS and non-sampling errors from SHIW (see the PI list of papers). UNIPG will also hire a PostDoctoral fellow to work on the project and envisions involvement of PhD students from the PhD school in Mathematics and Statistics at the Departments of Economics, Finance and Statistics of the University of Perugia. As mentioned earlier, the activities of the project are summarized in two Work Packages (WPs) that will be dealt with by an intense collaboration between the two research units. Each research unit will be in charge of particular tasks within WPs and also envisions partnerships with international universities where outstanding expertise on the topics of the project can be found and collaborations are already established (Prof. Ray Chambers, Un. of Wollongong, Australia; Dr. Nikos Tzavidis, Southampton Un., UK; Dr. Alina Matei, Un. of Neuchatel, Switzerland; Proff. Jean D. Opsomer and F. Jay Breidt, Colorado State Un., US; Prof. Peter G.M. Van der Heijden, Utrecht Un., The Netherlands; Dr Fabian Sobotka e Prof. Thomas Kneib, Georg-August-Universitat Gottingen, Germany).
WP1 deals with the estimation of labor market indicators (BES-3 indicators in particular) from the Italian LFS for planned/unplanned domains, with particular attention to the situation of young people. First, details on the LFS and on the precision of the direct estimates will be studied (UNIPG via the ISTAT personnel). Both units will then focus on a review of small area estimation (SAE) methods that borrow strength from auxiliary information to produce model-based estimates of labor market indicators (see Task 1.1 and Task 1.2 below). The units will then work on developments for different approaches - based on mixed models (UNIPG) and on M-quantile models (UNIPI) - to SAE to suit the LFS challenges (Task 1.3) and will consider parametric and non-parametric regression models as well as the inclusion of time or spatial correlations. Robust and Bayesian methods will also be investigated (UNIPI). As many survey variables in the LFS are categorical, empirical best predictors based on generalized linear mixed models (GLMMs) should be used. However, estimates of GLMM parameters can be very sensitive to outliers or departures from distributional assumptions. Hence, robust SAE based on GLMMs in the frequentist framework will be developed (UNIPI). A new approach to SAE for categorical variables based on M-quantile modeling will be studied to allow for simultaneous estimation of unemployment, employment and inactivity rates/totals. SAE based on finite mixtures models will be considered in which a discrete nonparametric distribution is assumed for the area effects that allows for an automatic clustering of the small areas. In addition, the project aims at developing SAE methodologies that account for the complex sampling design used in the LFS (UNIPG, Task 1.3). The issue of benchmarking will be considered by developing methods for estimating indicators for unplanned domains that are coherent with direct design based estimates released at higher level planned domains (UNIPG, Task 1.3). Using information from the different waves of the LFS, latent variable models for longitudinal data can be employed to investigate the evolution of this classification over time and transitions of one area from a cluster to another (UNIPG, Task 1.3). When dealing with area level models, direct estimates can be considered as variables observed at consecutive time occasions depending on unobservable characteristics and their evolution can be described by hidden Markov models (UNIPG, Task 1.3).
WP1: Analysis of geographical and temporal patterns of labor market indicators, with a particular focus on youth unemployment.
Task 1.1 (Month 1 - Month 4) Overview on the features of the Italian LFS.
- Analysis of the details of the survey design employed and the microdata available
- Preliminary analysis on the level of error for direct estimates of a selection of indicators for sub-regional local levels (BES-3 indicators)
Task 1.2 (M5-M12) Overview on existing small area methodologies for labor force indicators. Thematic literature review and outline of small area models, and of statistical and computational approaches including robust and Bayesian methods.
Task 1.3 (M13-M32) Modeling labor force indicators.
- Development of new models for SAE of BES-3 indicators
- Development of methods for estimating indicators for unplanned domains that are coherent with direct design based estimates released at higher level planned domains (benchmarking)
- Developments of models for estimating unemployment rates that account for spatial and temporal correlation
- Software development
UNIPG will be main actor for Task 1.1, while UNIPI for Task 1.2, whereas models will be developed (Task 1.3) jointly by UNIPI and UNIPG also in collaboration with researchers from University of Southampton, University of Wollongong and Colorado State University.
The purpose of WP2 is to analyze and evaluate indicators at the local level on social sustainability, focusing on living conditions, income and poverty (BES-4 indicators). The first contribution of WP2 is to review the precision of the available indicators in official statistics at local level in Italy (UNIPI, Task 2.1). Then, new SAE methods will be studied developing ad-hoc statistical models for the variables available from EUSILC and SHIW involved in poverty and inequality indicators based on GLMMs, M-quantile (UNIPI) and latent variable models (UNIPG) - Task 2.2.
Estimation of indicators based on variables related to income and wealth suffers from the presence of possible bias, because households may refuse to disclose such information (nonresponse) or, when collaborating, it usually happens that rich ones tend to understate it so as to reduce their tax liability, while it may happen that the poor tend to overstate it for sense of shame or to avoid controls (measurement error). In the presence of such sensitive variables, the process that describes nonresponse is said to be non-ignorable as it depends on the variables of interest. The project intends to introduce new methods able to use all auxiliary data available for the use of latent variable models to measure the propensity to response (UNIPG, Task 2.3). The possibility of collecting and using information on those subsets of the population which are known to have high non-response rates (elusive populations) using link-tracing sampling methods will also be pursued (UNIPI, Task 2.3). Finally, in order to increase respondent's collaboration and obtain truthful data on sensitive variables, the use of Randomized Response Techniques will be explored and tailored to the estimation of the aforementioned indicators of economic inequality and poverty (UNIPG, Task 2.3).
WP2: Mapping and monitoring households economic wellbeing.
Task 2.1 (M1-M4) Critical review of the precision of the available indicators (BES-4 indicators) in official statistics at the local level: - Critical analysis of the data available in EUSILC and SHIW
- Poverty indicators Social cohesion/sustainability (BES-4 indicators)
Task 2.2 (M5-M30) Modeling poverty, inequality and income indicators:
- Development of models for SAE of BES-4 indicators
- Development of models to estimate cumulative distribution function and quantiles
- Development of SAE methods based on latent variable models
- Software development
Task 2.3 (M13-M36) Treatment of non sampling errors.
- Development of methods to deal with nonignorable nonresponse
- Development of generalized calibration estimators that use latent constructs as instrumental variables
- Development of randomized response methods to increase data quality
- Development of finite mixture models for SAE methods that account for measurement error
- Software development
Tasks 2.1 will be coordinated by UNIPI. Both partners will be concentrated on the development of the methodological issues of Task 2.2. Tasks 2.3 will be coordinated by UNIPG also in collaboration with the personnel at the University of Granada, University of Neuchatel and Utrecht University.
