Methodology of correcting non-response bias: introducing another bias?the case of the Swiss innovation survey 2002

The non-response in a survey can lead to severe bias. In order to manage this problem, it is usual to make a second survey by a sample of non-respondent. This allows us to test if there is a significant difference in the key variables of the survey between respondents and nonrespondents and, if yes, to take it into account. But, the risk is great to introduce another bias depending on the mode (mail vs phone) of survey. The KOF industrial economics group is exploring for many years the innovation behaviour of Swiss firms using a mail survey addressed to almost 6600 panel firms of the industrial, construction and service sector. We use since some years the data of a second survey by nonrespondents to correct non-response bias. Contrarily to the first survey, this one is made by phone. One can suspect that the personal interaction with the person(s) calling may be introducing another bias. In order to investigate this question, in the case of the ETH Zurich's innovation 2002 survey, we decided next to the regular non-respondent-phone-survey, to conduct a similar phone survey by a subsample of the respondent-group. Thus, we dispose of data for the same variables coming from the two modes of survey and allowing us to show if there is a difference or not in the response behaviour. We use different statistical approaches to investigate this issue, considering x2-test and Logit models. Our results show that data collection method may influence the response.