A well-established risk factor in surveys is social desirability bi as (SOB): the tendency to report about oneself in a favourable manner (Edwards 1957). Stan­dard solutions consist of the inclusion of a scale, e.g. the Marlowe-Crowne scale (Crowne & Marlowe 1960) that should measure the respondent's general SOB. Such solutions, however, have serious limitations and are criticized by many (e.g., Barger 2002). One disadvantage is that a SOB scale does not allow to assess the amount of SOB included in self-reports of specific forms of behaviour since the sea les only measure the respondent's _general_ SOB. As an alternative, re­searchers have proposed using _indirect questioning_ as a means of overcoming social desirability bias (Fisher 1993). Indirect questioning asks respondents about what ,,"other people'''' think about some, potentially sensitive, issue. The idea is that respondents may find it easier to express their own opinions and attitudes on a sensitive issue in response to such indirect and impersonal questions. Indeed it seems to be the case that this approach may help alleviate social desirability bias (Jo 1997). In the literature on online data collection, however, some support can be found that in online surveys respondents feel more anonymous and are there­fore somewhat less susceptible to SOB (Richman et al. 1999, Joinson 1999). It is therefore an open question whether the inclusion of indirect questions is worth its efforts in online surveys.

ln our contribution, we strive at measuring the willingness to contribute to collecti­ve goods in groups by presenting realistic scenarios about group behaviour to the respondent. We compare to which extent the tendency to give socially desirable answers is reduced by using indirect questions. More specific, we analyse under which conditions the approach leads to a reduction of the SOB for a number of behavioural forms. Such conditions include characteristics of the respondent, the question, and the survey topic. For example, we hypothesize that questions de­scribing scenarios should be short enough so that the respondent does not give in to the incentive to read ,,"diagonally"". One hypothesis is that longer ,,"scenarioquestions"" lead to less reduction of social desirability bias (that is, to less of a difference between direct and indirect questioning). We use quasi-experimental data of two online surveys that present similar questions to eBay users and to members of online communities for teachers.