Inference in Web surveys
Estimation theory provides both the statistical criteria for the transfer of sample data to population parameters (the so-called inference problem) as well as for the measurement of the level of estimate reliability. However, this is only valid if the sample selection is carried out on a random basis. The theory has been developed with reference to traditional data collection tools.
Regarding web based surveys studies on inference problems are in progress and no final results are identified.
Nevertheless some observations can be made and some basic comments given. Above all it should be pointed out that estimation theory to make inference on web collected data is applicable in principle. The problem is that it is not always possible or easy to provide probabilistic samples, above all due to the difficulty in identifying a population frame suitable and similar to population (Pineau & Slotwiner, 2003).
From the point of view of inference, solutions to resolve situations where the sample is not representative of the target population are still being studied. An operative solution could be using a weighting system which takes into account different probabilities in the extraction of persons from a population which could exclude certain categories from being selected by using a specific sampling techniques (post stratified Horvitz and Thompson estimator).
Another possible solution would be to use the weighted system based on the propensity score, this is introduced with the aim of making it possible to apply inference principles to observational data (i.e taken from samples not from random selections) (Biffignandi & Pratesi, 2005, Forsman, 2004).
More:
This article has been read 1007 times
