Confirmation Bias, Data Manipulation and the Kuznets Curve: Original, Environmental and Financial

Abstract

Empirical work in economics and finance involves data manipulation in ways that make it possible to confirm prior beliefs. The results typically turn out to be sensitive to model specification, sample period, variable definitions and estimation methods. The status quo provides the motivation for dealing with this problem using three versions of the Kuznets curve for the purpose of illustration. The underlying hypotheses are represented by quadratic functions of income per capita, with a different dependent variable for each version of the Kuznets curve. Both time series and cross-sectional data are used to estimate the equations. The results turn out to be highly sensitive to a number of factors, which provides an incentive for being selective in the reporting of results while exhibiting confirmation bias. To overcome the model uncertainty problem one can resort to the use of one of several methods that are based on the reporting of the distribution rather than the point estimation of the coefficients.