engleski

Multi-level Analysis of Survey Data Falsificatio

In the housing needs assessment in Croatia, we tried to apply different approach to detect data falsification. Survey was done by methodology applied in previous European Values Study (EVS) wave in 2008. The three-stage probability sample (settlements, households, respondent) was used. Random walk and last birthday method were applied giving enough space to deviate from procedure. Problems of application of fieldwork guidelines and obtaining high response rates were analyzed prior to that survey using metadata (Rimac et al., 2010.). Difference in approach was that very extensive fieldwork control was applied. It included: sending of locked random- walk list prior to interviewing, review of interviewed households list, GIS data and phonebook checking of step in random-walk procedure, verification of respondent choice and telephone contact with respondent in which ownership of contact number was checked by the telephone book. Result of control of 2220 interviews gives 76, 3% of correct interviews, 0, 2% of questionnaires left to respondents and picked afterwards, 16, 9% of substitution by household member and 4, 2% by neighbor, 1% of fully fake interviews with given contact of person that have to approve interviewing and 1, 4% of fully falsified interviews. Results support the conclusion that falsification of survey data can’t be analyzed without considering interviewers motivation, cognitive skills and pattern of falsification. That gives idea that only multilevel approach to data analysis can lead to conclusions. This is especially present in indicators that address less variable answers in falsified data, those are strongly masked by individual differences of interviewers’ approach to falsification.

Multi-level analysis; Data falsification; Face to face survey; Fieldwork observations

nije evidentirano