Dynamic modelling of Nonresponse in Business Surveys

It is well-known that nonresponse affects the results of surveys and can even cause bias due to selectivities if it cannot be regarded as missing at random. In contrast to household surveys , response behaviour in business surveys has been examined rarely in the literature. This paper is one of the first which analyses a large business survey on micro data level for unit nonresponse. The data base is the Ifo Business Tendency Survey , which was established in 1949 and has more than 5,000 responding firms each month. The panel structure allows to use statistical modelling including time-varying effects to check for the existence of a panel fatigue. The results show that there are huge differences in business characteristics such as size or sub-sector and that nonresponse is more frequent in economically good times. 1 Introduction Data collection is the essential tool and fundamental for all empirical studies. In the socioeconomic sciences the methods used are mostly surveys. These surveys often face the problem of nonresponse, i.e., partial or complete drop out of information. In contrast to one-time studies, in panel surveys nonresponse is much more problematic, because the same units are analysed over time. Since panel studies with sociological or economic background mostly base on household surveys, a large literature exists about techniques for reducing the effect of nonresponse. Only less is known about the processes and reasons for participation and responding behaviour in business surveys (Janik and Kohaut, 2009). Although individuals are questioned in the survey, they are representatives of an organisation, so that or-ganisational relationships have to be considered (Tomaskovic-Devey et al., 1995). This paper is one of the first that models unit nonresponse behaviour in business surveys on micro level. To this end we examine the Ifo Business Tendency Survey (Ifo BTS) for unit nonresponse. The most well-known result of this survey is the Ifo Business Climate Index, one of the most prominent economic indicators for the German business cycle. Because the Ifo BTS is a survey performed since 1949 with more than 5,000 respondents each month, it provides a large amount of data with panel structure. For the nonresponse analysis in this paper, monthly data from 1994 to 2009 is used which lead to a total number of observations of more than one million. Former empirical studies on nonresponse mainly focussed on the aggregate response 1 rate, in particular explaining effects of survey characteristics. …