Using external sources to understand sample survey bias: the case of the Invind survey

We look at two sources of bias in survey estimates of the Bank of Italy�s Survey of Industrial and Services Firms conducted yearly on a panel of enterprises: 1) the bias owing to panel attrition caused by the differences between units entering and exiting the sample and those participating more regularly in the survey; 2) the bias created by delays in the distributional data on the reference population, needed for computing the survey weights. By comparing an array of performance indicators (available in an integrated database) for the firms regularly participating against those participating more erratically, we find that panel attrition has a limited effect on the official aggregate estimates, since they are determined by larger firms, which tend to participate regularly in the survey. Smaller firms� erratic participation, the comparatively worse performance of the units exiting the sample and the higher-than-average age of the firms in the sample call for a careful assessment of the estimates that are not influenced by firm size. Finally, for the less recent years we measure the extent to which the estimates vary if we use the revised information on the reference population, and we find that the delays in updating significantly bias the aggregate estimates only when the population size is highly unstable, with negligible effects on the estimates less dependent on firm size.