A latent class approach for allocation of employees to local units

In 2011, the Italian Business Register has been reshaped as a database of single workers microdata. Determining the workplace of each individual provides the National Statistical Institute (NSI) with huge information potential. Unfortunately, the administrative sources at our disposal do not always allow a reliable determination of the workplace of each worker. We present a probabilistic methodology to assign a workplace to each employee by assigning him to one of the local units of the enterprise he works for. We used a Latent Class Model to estimate the probability of each employee to belong to each local unit. We assumed the total number of employees per local unit as a constraint. A computationally intensive optimization problem has been solved for each of the ca. 200 thousands multilocated enterprises. The results refer to year 2011.