Using Administrative Data to Evaluate Sampling Bias in a Business Panel Survey

Abstract

We examine two sources of bias for the Bank of Italy’s panel business survey of Industrial and Services Firms: 1) the bias caused by panel attrition; and 2) the bias created by delays in the distributional data on the reference population, needed for computing the survey weights. As for the first source of bias, the estimates strongly dependent on big firms’ values are less affected by panel attrition than those representing firms’ average behavior, independent of their sizes. Positive economic results make it easier to enroll new firms in the survey, in order to replace firms dropping out because of bad economic performances. However, the economic results of new entrances become more aligned to those of the population, once they enter the sample. A very different result emerges for the second source of bias, since, when the population size is highly variable, the information delays produce a bias for the estimates influenced by the contribution of great firms, but the effect is negligible for the estimates not dependent on firm size.

Workforce Aging, Pension Reforms, and Firm Outcomes

“…and Viviano, 2018). We restrict the analysis to firms that employed at least 50 employees in the year they first appeared in the sample; we implement this restriction for three reasons: i) balance sheet information can be patchy and variables can show very volatile dynamics in smaller firms, ii) measures of the age shares of employment crucial for our identification strategy are also likely to be volatile for smaller firms, and iii) since the INVIND survey covers 20+ firms only, there could be non random attrition of shrinking firms as they cross the 20 employees threshold from above and of course such phenomenon is more likely to happen in smaller firms (D'Aurizio and Papadia, 2019). We statistically test whether the reform we study had an impact on firm exit and entry into the sample.…”

Workforce Aging, Pension Reforms, and Firm Outcomes

“…We restrict the analysis to firms that employed at least 50 employees in the year they first appeared in the sample; we implement this restriction for three reasons: i) balance sheet information can be patchy and variables can show very volatile dynamics in smaller firms, ii) measures of the age shares of employment crucial for our identification strategy are also likely to be volatile for smaller firms, and iii) since the INVIND survey covers 20+ firms only, there could be non random attrition of shrinking firms as they cross the 20 employees threshold from above and of course such phenomenon is more likely to happen in smaller firms (D'Aurizio and Papadia, 2019). We statistically test whether the reform we study had an impact on firm exit and entry into the sample.…”