Indicators for risk of selection bias in non-probability samples

In recent years, in surveys, is increased the relevance of the non-probability sampling with which the units included in the sample are determined by an unknown self selection process. The self-selection of the units in the sample raises concerns about the potential for biased estimates of population characteristics and consequently it originates the need of indicators for assessing the degree of bias and for identifying potential adjustment/weigting methods for reducing it. Here, we are motivated by a real problem: the risk of bias in a non-probability web survey on the customers’ satisfaction with the services provided by the municipal library system in Florence. We suggest and discuss the combined use of three measures, each one for a specific aspect of the data that impacts on the risk of biased estimates.