ELVIRE is a survey in two parts, each with its own self-administered web questionnaire. Conducted by the French National Demographic Institute, it aimed to investigate the languages used by the researchers working in the French public research Institutes and Universities. The rst part of the survey (called Unit) was sent to about 4000 heads of all the public research units. The questionnaire was available on-line for four months (December 2007 to March 2008). The second part (called Individual) was an open survey (without previous sampling) addressed to all researchers and PhD students working in these units. The questionnaire was available between November and December 2008.

The purpose of this paper is to describe the strategies implemented in order to reduce non-response to the Unit part and to measure the impact of non-response bias. As a self-administered web survey, a high non- response rate was expected, all the more because of the duration of the questionnaire (more than 1 hour on average) and the respondents' limited availability (overworked unit heads). However, 54% of the heads lled out the questionnaire. Among other strategies described by the paper, a special eort was made to contact the heads individually several times, alternating a variety of means. After sending one letter and if no answer or no completed questionnaire two emails to each of them, a follow-up phone call was made one month later. The interviewers had to contact the potential respondents and persuade them to ll out the questionnaire or else obtain a reason for their refusal. All information was coded in a data le. This telephone contact not only increased the response rate by 24 percent points (from 30 to 54%) but also oered a source of information on respondents reactions and the reasons for their refusal. To better understand these reactions, two other sources, more qualitative in nature, will be mobilized. First, the e-mails exchanged between the survey team and the respondents. Second, the eld of the questionnaire especially dedicated to respondents comments on the quality of the questionnaire.

Furthermore, the comparison with the Individual part (almost 9000 responses) may give additional clues about the impact of some factors, such as the questionnaire design (more functional and attractive in the second part); its length (20 min instead of 1 hour); or the respondents' status.