Methods

The development of a data anonymization methodology enables simplified files to be widely distributed, while ensuring confidentiality. This is a key area of work for the Innovations department. A collaboration with INRIA has been set up, with the recruitment of a specialist in anonymization techniques. The aim is to go one step further in ensuring compliance with individual privacy protection standards, while at the same time maximizing the usefulness of files, in particular by ensuring that they contain as many variables as possible. Once the methodology has been validated, it will be necessary to set up a file production circuit, as well as a file distribution circuit.

More information :

présentation de l’anonymisation des enquêtes dans le cadre de LifeObs (in French only)

Data matching involves bringing together two databases of different origin that share common statistical units but contain different information. A classic case for statisticians is the matching of person files. There are many advantages to this: identifying duplicates, enriching databases, etc. This raises practical issues (identification of a key enabling the same individuals to be identified between databases) and legal issues (declarations to the CNIL, authorisation from producers, etc.).

The enrichment of survey data by administrative sources is a special case of matching. It makes it possible to reduce the number of questions asked by collecting information directly from the sources. Respondents must be informed of the matches made. Data from LifeObs surveys will be enriched by matching tax or social security administrative data.

To find out more, click here (in French only):

Appariements à des données administratives dans le cadre d’Erfi 2 et FamEmp

Regrouper les utilisateurs des données démographiques au CASD

For surveys of households or individuals, including several LifeObs surveys, the sampling frame can be that of tax data (Fidéli database). This sampling frame provides good coverage of the population and access to a large amount of auxiliary information for stratification (type of dwelling, income, geographical location). This sampling frame can only be used once it has been validated by the National Council for Statistical Information (CNIS), via the Official Statistics Label Committee.

Where it is not possible to use this frame, other random sampling methods such as random generation of telephone numbers may also be used for telephone surveys.

To find out more, click here:

Géraldine Charrance, Amandine Stephan et Thomas Merly-Alpa, 2024, "Echantillonner des enquêtes multimode dans Fidéli : retours sur l’expérience de l’enquête Familles et employeurs". 13e Colloque Francophone sur les Sondages - 2024, Luxembourg (Luxembourg). http://hdl.handle.net/20.500.12204/j0RnYpMBGHEcdMY0JAE_

Les coordonnées de contact dans Fidéli-Nautile : Quels enseignements en tirer pour la mise en place de protocoles de collecte ? : Travaux menés dans le cadre d’une convention avec l’Insee et présentés lors de l’Assemblée du 14 octobre 2022, Ined

Optimiser une collecte par génération aléatoire de numéros de téléphones mobiles pour une enquête en population restreinte : présentation dans le cadre de l’Assemblée plénière LifeObs du 14 octobre 2022

For the new round of the Generations and Gender Programme’s European surveys, the switch to filling in questionnaires online has become a necessity. In 2018, faced with the difficulty of conducting face-to-face interviews, rising costs and falling participation rates in the surveys, an experiment was carried out in Portugal, Germany and Croatia to test the feasibility of online surveys and the effectiveness of financial incentives to improve response rates and compensate participants for their time (Emery et al. 2020; Lugtig et al. 2022). Tests have been carried out in a number of countries, including France. The test carried out in France in 2021 for the Erfi 2 survey (compared nine scenarios comparing the organisation of interviews by telephone and the completion of questionnaires by internet, combined with the introduction of financial incentives, at the time of sending the letter (unconditional incentives) or after completion of the questionnaire (conditional incentives). The test showed that unconditional incentives (costly in the event of a low response rate) were necessary but complex and costly, and that many responses were received via the Internet (Bouchet-Valat et al. 2022). After consulting INED’s ethics committee (opinion 35 of 1/3/2024), the Erfi2 project abandoned the use of financial incentives for organisational, financial and ethical reasons (Bouchet-Valat et al. 2022; 2023).

Find out more:

• Lugtig, Peter, Vera Toepoel, Tom Emery, Susana L. F. Cabaço, Martin Bujard, Robert Naderi, Almut Schumann, et al. 2022. “Can We Successfully Move a Cross-national Survey Online? Results from a Large Three-country Experiment in the Gender and Generations Programme Survey.” SocArXiv. February 13. doi:10.31235/osf.io/mu8jy. https://osf.io/preprints/socarxiv/mu8jy
• Emery Tom, Detlev Lück, Almut Schumann, Susana Cabaço, Peter Lugtig, Vera Toepoel, Martin Bujard,Robert Naderi, Norbert F. Schneider, 2020, Three‐country Experiment on Modes of Data Collection. Generations and Gender Programme, 26 p. https://www.ggp-i.org/wp-content/uploads/2020/10/Three-country-push-to-web-experiment.pdf
• Bouchet-Valat Milan, Bondon Marine, Breda-Popa Ruxandra, Cochet Paul, Charrance Géraldine, Markou Efi, Toulemon Laurent, 2022, Telephone, Internet, and Incentives: First Lessons from the French GGP2020 Pilot, GGP-Connect Seminar. January 25th https://www.ggp-i.org/ggp-connect-seminar-series/#toc18
• Bouchet-Valat M., Toulemon L., Breda-Popa R, Markou E., Charrance G., Cochet P., Bondon M., 2023, L’enquête test de l’Étude des relations familiales et intergénérationnelles (erfi) 2.0. Enseignements d’une collecte multimode téléphone/Internet avec et sans incitations financières. , 12e Colloque francophone sur les sondages, 24 mars 2023. Session 3 : Enquêtes sociales internationales. https://archined.ined.fr/view/AYciUh0JLg0aT10Rufan

Multi-mode protocols now play a central role in quantitative survey research. They make it possible to adapt to changes in respondent behaviour and to the downward trend in participation rates by offering several modes of collection. These systems introduce specific methodological issues linked to fashion effects, i.e. the influence of the mode of collection on the estimates produced.

Fashion effects can be broken down into two mechanisms:

• the selection effect: this results from the fact that it is not the same people who respond depending on the data collection method. These disparities can be explained by factors such as the contact details available in the field, incentives to respond or the preferences of individuals for a particular collection method. Non-response can be considered ignorable if, taking into account the observable characteristics of individuals, their participation does not depend directly on the variable under study. However, when an unobservable bias simultaneously affects the choice of collection mode and the variable of interest, non-response becomes non-ignorable, which introduces an endogenous selection bias into the results.
• the measurement effect: reflects differences in the answers given by the ‘same’ individuals depending on the data collection method used. Interviewer-led modes can be subject to the phenomenon of social desirability, where respondents adjust their responses according to presumed expectations or implicit social norms. Conversely, self-administered modes, such as online questionnaires, may encourage less involvement on the part of the respondent, corresponding to what is known as ‘satisficing’ behaviour. This results in approximate or ill-considered responses, such as systematic repetition or random choice of answers.

The interaction between these two mechanisms, selection and measurement, is necessary for an analysis of bias in multi-mode surveys. The question is whether the differences observed result from the impact of the mode of collection and/or whether they reflect an endogenous selection effect. This distinction is essential for assessing the reliability of the estimates produced.

In this context, a working group has been set up to conduct methodological research to better understand and quantify these effects. The work carried out as part of the FamEmp and ERFI surveys explores approaches for measuring measurement and selection effects to provide tools for analysing and interpreting the biases associated with multi-mode surveys.

To find out more, click here:

Enquête « CDV-TCM » Votre cadre de vie, Enquête méthodologique en collecte multimode : présentation dans le cadre de l’assemblée plénière LifeObs 2022

Most of the LifeObs surveys are longitudinal, making it possible to follow the trajectories of the same individuals over time. The innovation department is working on ways of limiting or circumventing panel attrition.

Find out more about LifeObs:

Table ronde sur le suivi des panels du 14 octobre 2022

Table ronde sur la restitution des résultats d’enquête aux personnes répondantes du 16 octobre 2024