Thomas Merly-Alpa

Thomas Merly-Alpa, head of the surveys and polls department, answered our questions on how surveys are conducted at INED.

(Interview conducted in January 2022)

A test survey (we prefer to call them survey tests) is a small-scale version of a survey that has not been conducted yet, a means of improving the survey before it is launched. For example, researchers may want to check that the way a particular question is formulated doesn’t make it difficult to understand for the people who will be asked it. Survey tests can also be used to check the data collection protocol; that is, how collection will be organized and how well that functions. For example, for a survey test on inmates just before they leave prison, we needed to check that we had access to visiting room parloirs to conduct the survey, that we would get the list of inmates in time, that the inmates would be permitted to come to the survey location. It is always helpful to do survey tests, even several of them.

There are a thousand ways to do a survey. While surveys often involve long questionnaires, they can also be done through observation or interviews. For example, we conducted several individual interviews with people in Reunion who were caring at home for a family member at the very end of their life. Questionnaire surveys, meanwhile, can be administered by a survey professional either by phone or at the respondent’s home. There are also short questionnaires that respondents can fill out directly on internet, their computer, or by phone. Increasingly, different approaches are combined, allowing respondents to answer using their preferred means of communication. Such surveys are called “multi-mode.” 

INED’s survey department keeps abreast of new developments and methodological innovations in survey procedures (new qualitative survey techniques, response-based collection method evaluation, and others) and contributes to them by presenting its methods and its test and survey findings to the scientific community. 

Finding respondents is an important phase of conducting a survey. In the vast majority of cases, the sample (that is, the list of respondents) is obtained through random selection; the “chance” nature of this method guarantees high quality findings. The larger the sample, the more precise the findings—and the higher the survey cost, of course. To achieve maximum result precision with the smallest possible sample, more sophisticated methods for selecting individuals are needed. INED’s survey department, particularly its statisticians, are becoming expert in using INSEE’s “Fideli” file, containing tax data (income, property tax, etc.); “Fideli” provides a good overall view of the French population from which to extract samples while protecting personal data. But this is not the only way of drawing a sample. INED researchers occasionally use “snowball” methods—when they are interested in “rare” or hard-to-reach groups, for example. This method constructs a sample using a small group of individuals who then provide contact information for people they know, thus “growing” the snowball at each stage. A method from the “snowball” family of methods was used, for example, in the ChIPRe (Chinese Immigrants in the Paris Region) survey: respondents were asked to provide us the contact information of relatives, friends, or coworkers of Chinese origin living in the Île-de-France region. 

Ensuring respondent anonymity is crucial. The most important guarantee lies in the reputation of INED and its researchers, engaged in serious scientific survey production for over 70 years now. Researchers and the Institute need to maintain their relationship of trust with all French people, as any of them might be mobilized to participate in our surveys. Anonymity is also guaranteed by a set of actors who intervene at various levels: e.g., INED’s Delegate in Charge of Data Protection, closely associated with every survey we do; France’s National “Information Technology and Liberty” Commission; and the Statistical confidentiality committee, all of whom verify that all rules have been followed to ensure data security. Another such actor is INED’s DataLab, which checks before survey results are diffused to scientific community that no individual can be identified on the basis of their responses.