Bias Diagnosis in Digital Economy Data: A Reverse Regression Approach

This study explores reverse regression (RR) as a diagnostic tool for detecting measurement error and endogeneity in digital economy datasets, where traditional Ordinary Least Squares (OLS) estimates are often biased. Motivated by the growing use of digital platform-generated data, the study employs simulated data calibrated to real-world benchmarks for medium-sized enterprises. The simulation introduces controlled measurement error, heteroscedasticity, and platform drift to replicate noisy typical of digital reporting environments. Forward OLS Regression and RR estimators are compared across multiple scenarios, including variations in error variance, distributional assumptions, and sample size. Results show that RR consistently achieves lower root mean squared error (RMSE) than forward OLS Regression, with the performance gap widening under greater measurement error or non-Gaussian noise. The forward–reverse coefficient gap is proposed as a bias indicator, revealing attenuation bias in forward OLS estimates. These findings have practical implications for businesses and policymakers, as uncorrected bias may lead to underinvestment in advertising or poorly calibrated digital incentives. The study introduces RR as a low-cost diagnostic tool for detecting bias in digital economy data, especially when traditional instrumental variable methods are infeasible. The study recommends a triangulated diagnostic framework that combines RR with instrumental variables (IVs) or generalized method of moments (GMM) to address endogeneity and concludes with recommendations for future research, including dynamic settings and multi-platform validation.