Assessing the accuracy of non-random business conditions surveys: a novel approach

摘要

A number of central banks and other institutions publish their own business conditions surveys that rely on complex non-probability sampling methods. The results of these surveys influence policy decisions and affect expectations in financial markets. To date, no one has assessed the accuracy of these surveys because their complex (and often unique) sampling method renders this assessment non-trivial. The paper describes a novel approach for modelling unique sampling methods when many constraints (including quota sampling and clustering) are imposed. When no closed form solution exists, we show how to compute the selection probabilities from each firm in the known population and the dispersion of the sampling distribution by using Monte Carlo techniques. This method can also be used to assess the appropriateness of a survey sample size. Our approach is applicable to many major surveys conducted by various central banks and institutes across the Organisation for Economic Co-operation and Development. To demonstrate the feasibility of our approach, we apply it to the Bank of Canada's 'Business outlook survey'. Although the survey's coverage is significantly limited, we find, under certain assumptions, no evidence that the Bank of Canada's firm selection process results in a wider dispersion in the sampling distribution than the stratified random sample.