Chapter 37 Other Biases

In econometrics, the main objective is often to uncover causal relationships. However, coefficient estimates can be affected by various biases. Here’s a list of common biases that can affect coefficient estimates:

What we’ve covered so far (see Linear Regression and Endogeneity):

  1. Omitted Variable Bias (OVB):

    • Arises when a variable that affects the dependent variable and is correlated with an independent variable is left out of the regression.

  2. Endogeneity Bias:

    • Occurs when an error term is correlated with an independent variable. This can be due to:

      • Simultaneity: When the dependent variable simultaneously affects an independent variable.

      • Omitted variables.

      • Measurement error in the independent variable.

  3. Measurement Error:

    • Bias introduced when variables in a model are measured with error. If the error is in an independent variable and is classical (mean zero and uncorrelated with the true value), it typically biases the coefficient towards zero.

  4. Sample Selection Bias:

    • Arises when the sample is not randomly selected and the selection is related to the dependent variable. A classic example is the Heckman correction for labor market studies where participants self-select into the workforce.

  5. Simultaneity Bias (or Reverse Causality):

    • Happens when the dependent variable causes changes in the independent variable, leading to a two-way causation.

  6. Multicollinearity:

    • Not a bias in the strictest sense, but in the presence of high multicollinearity (when independent variables are highly correlated), coefficient estimates can become unstable and standard errors large. This makes it hard to determine the individual effect of predictors on the dependent variable.

  7. Specification Errors:

    • Arise when the functional form of the model is incorrectly specified, e.g., omitting interaction terms or polynomial terms when they are needed.

  8. Autocorrelation (or Serial Correlation):

    • Occurs in time-series data when the error terms are correlated over time. This doesn’t cause bias in the coefficient estimates of OLS, but it can make standard errors biased, leading to incorrect inference.

  9. Heteroskedasticity:

    • Occurs when the variance of the error term is not constant across observations. Like autocorrelation, heteroskedasticity doesn’t bias the OLS estimates but can bias standard errors.

In this section, we will mention other biases that you may encounter when conducting your research

  • Introduced when data are aggregated, and analysis is conducted at this aggregate level rather than the individual level.
  1. [Survivorship Bias] (very much related to Sample Selection):
  • Arises when the sample only includes “survivors” or those who “passed” a certain threshold. Common in finance where only funds or firms that “survive” are analyzed.
  • Not a bias in econometric estimation per se, but relevant in the context of empirical studies. It refers to the tendency for journals to publish only significant or positive results, leading to an overrepresentation of such results in the literature.