SLONews
News
Presentation of the project conclusions at the Statistical Office of the Republic of Slovenia
Members of the Centre for Social Informatics (CSI) visited the Statistical Office of the Republic of Slovenia (SURS) on the occasion of the completion of the project Probability web panels in national statistics for persons and households (ARIS code V5-2157, project leader Vasja Vehovar, PhD). The plenary presentation, which took place on 20 November, was attended by Mag. Apolonija Oblak Flander, Acting Director General of SURS, and about 40 methodologists.
During the visit, CSI members made a plenary presentation of the key results of the project, based on the study of the potential of probabilistic web panels in official statistics, with a focus on cost and data quality. A pilot probability panel in Slovenia has shown some potential but also limitations. The advantages to be highlighted are mainly the cost savings and the increased speed and simplicity of data collection. The limitations relate to the bias of certain estimates, non-coverage and low levels of participation. Strategies such as combining the online mode with field or telephone surveys have been proposed as possible alternatives to the probabilistic online panel.
The V5-2157 project dealt with probabilistic web-based survey panels, which involve probability sampling, one-off recruitment (persons give their prior consent to participate in various surveys) and web-based surveys. Standards in official statistics require probability surveys, but probability web panels are not yet used in Slovenia. Project activities to explore the potential of such panels included a literature review, an inventory of survey questions, the implementation of a pilot probability web panel and the preparation of a full feasibility study.
SurveyWeightingGUI: An open source application for weighting survey data
Luka Štrlekar (Master's student of Applied Statistics) and Vasja Vehovar, PhD (UL FDV) developed SurveyWeightingGUI: an open source application (programming language R) for weighting survey data.
In survey research, it is often necessary to weight the data before analysis. This reduces bias in the estimates and ensures that the socio-demographic structure is matched. While procedures for this type of weighting are already developed (e.g. in R, STATA, SPSS, etc.), they assume that the data are fully prepared beforehand.
It is the preparation of the data that is the main and most time-consuming problem in weighting. It is the naming of the control variables, the entry and archiving of the control data, the treatment of missing values, the problem of too few categories or too few units, as well as the issues of cutting and normalising the weights and analysing their effect.
The application solves these dilemmas through a user-friendly web GUI (Graphical User Interface), while also performing and evaluating the weighting itself. The user is not directly involved with the R code of the weighting (Survey, Anesrake). This makes weighting accessible to anyone without technical knowledge of R programming, and for users who do have this knowledge, the application drastically speeds up the process.
The application has proved its worth in practice and has saved a great deal of time in the survey projects of the Centre for Social Informatics, FDV (CDI.si) and the Public Opinion and Mass Communication Research Centre (CJM.si), FDV.
The SurveyWeightingGUI application is available online, as well as locally.
Web Panels and Mixed Survey Modes 2023
On September 28th 2023, the Centre for Social Informatics (CSI) hosted the traditional annual conference Web Panels and Mixed Survey Modes in collaboration with the Statistical Office of the Republic of Slovenia (SURS) and the National Institute of Public Health (NIJZ). The programme consisted of presentations from SURS, NIJZ and CSI, as well as a joint discussion.
The aim of the event was to exchange experiences and insights on the latest trends in the collection, analysis and comparison of survey data in Slovenia and internationally.
Survey implementers highlighted the importance of motivated and engaged interviewers, quality questionnaires and the use of QR codes on respondent letters as good practices to encourage participation in surveys.
On the topic of data provider profiles and (non-)participation in surveys, SURS highlighted traditional written communication (short texts, use of pictograms and clear basic information about the surveys), slow responses to information from the field, usability of the website, organisation of the work and low awareness of SURS among young people as possible areas of improvement for participation in surveys. They also produced an analysis of the persons involved in statistical surveys (the 'personas') and presented for each of them the factors that are important for their participation in the surveys. They highlighted the upgrading of internal processes, adapting communication to the younger generations and the monitoring and analysis of non-response as important guidelines for the future.
NIJZ also shared their experience of conducting research over the last three years. The importance of cooperation and exchange of views and experiences of all those involved, including the sponsor, the contractor, the interviewers and the selected persons, was highlighted. They note that online panels are excellent for quick assessments, that postal surveys are "in" again due to the pandemic; even the youngest groups are keen to participate, but postal surveys are quite expensive. An interesting finding is that participants in health surveys value results and outputs more than practical rewards.
SURS also presented Smart Survey Implementation (SSI), an international project with the strategic goal of innovative and powerful statistical production. The project will test respondents' willingness to use smart functionalities through the People-Centred Data Collection Methods (Načini zbiranja podatkov po meri ljudi - NZP) survey, with results to be available in 2024.
An open-source tool for weighting and analysing survey data has been developed at CSI. The estimates of the variables of the online probability panels were also compared with the official SURS and NIJZ estimates as well as the estimates of the non-probability (commercial) online panels. The main factors that may lead to differences between the estimates are the mode of interviewing, the duration of the interview and the rewards. The cost per unit accuracy of probability panels compared to commercial panels, which are approximately one times cheaper, was also analysed. In general, for larger surveys, non-probabilistic online panels with larger samples (at the expense of lower costs) cannot compensate for bias. It is therefore essential for a good quality non-probability web panel to prevent bias in the estimates. The potential of probabilistic web panels compared to official surveys will be further explored in the project Probability web panels in national statistics for persons and households.