We characterize the asymptotic bias that arises in ordinary least squares regression when control variables have nonlinear effects on an outcome variable, but are assumed to enter the regression equation linearly. We show that this bias can be larger than that from omitting such variables altogether, or may vanish entirely despite the nonlinearity. We find that under a natural assumption an upper bound to the magnitude of the bias may be estimated from the data, and consider examples of the bias through Monte Carlo simulations.
The Bias from Misspecification of Control Variables as Linear
Working Paper by Leonard Goff — Nov. 19, 2014Download
Common Resources — Apr 24, 2020
Preparation Now for Environmental Opportunities in Fiscal Stimulus Programs
Coronavirus will motivate some fiscal stimulus spending proposals that could benefit the environment, as long as proponents of environmental programs are prepared for the opportunities.
Common Resources — Mar 18, 2020
Local Economic Impacts of Federal Protected Lands: National Monuments in the Mountain West
The prospect of lost livelihoods can produce conflict over the limits that national monuments place on land use, but creating monuments can also create value by growing new industries related to recreation and tourism.
Resources Articles — Mar 18, 2020
It’s a Good Time for Women to Win the Nobel Prize
Catherine Wolfram discusses last year's historic Nobel Prize in Economics, reviews the significance of randomized controlled trials, and shares wisdom for women in economics.