Pre-printing Effects in Official Statistics: An Experimental Study
In surveys where respondents are contacted repeatedly, information from previous data collections may be used in the subsequent data collections. The responses then become dependent on the presentation and the quality of that information. Normally, the presented information is historical data concerning earlier reference periods, though relevant to a current reference period, and the respondents can verify and (if necessary) change it. The reasons for presenting historical data are: 1) it increases the efficiency of the data collection; 2) it can correct previous errors; 3) it reduces the response burden and sometimes, it is believed it reduces measurement errors and spurious response variability. The possible drawbacks are that it can conserve errors rather than correct them and it might lead to underreporting of changes from one period to another. Here, we focus on methodological issues of pre-printing historical values on self-administered (electronic or paper) questionnaires for business establishments. A planned experiment conducted in an ongoing large-scale survey indicates that on several aspects of data quality, questionnaires with pre-printed historical values outperform questionnaires without them. We present the main results of the experiment, as well as a general discussion of pre-printing experiences at Statistics Sweden.
Use of historical data, experimental design, measurement errors, response variability