Model misspecification and bias in the least-squares algorithm: Implications for linearized isotropic AVO

This paper shows how to calculate the bias due to misspecified models in least-squares parameter estimation. It introduces Omitted Variable Bias (OVB), a technique well known in least-squares analysis in the context of econometric data analysis. OVB is applied to the analysis of linearized isotropic AVO models, both analytically and numerically. For misspecified models, such as two-term AVO fitting with large angle range or with large contrasts, OVB provides relations between the biased and unbiased least-squares model parameters. A Jupyter notebook and binder link let’s the reader apply the ideas presented to their own AVO models.