I keep hearing that R-squared does not really matter in economics research and that due to the unpredictable human nature, economics research regressions tend to have low R-squared.

But how much is too low for even economics research?

I am asking this because one of the regressions I ran has an R-squared of around 1%. Is it still okay?

And, in general, what is a low R-squared in economics?


1 Answer 1


Disclaimer: this answer comes from a microeconomic research perspective. Time series / macroeconomic specialists will likely have other perspectives.

There is no general rule for what's too low across the entire field of economics. Yes, microeconomic models (i.e., individual-level observations) will tend to give low R-squared values (often in single percentage point digits) because there are so many factors that can affect human outcomes, a lot of which just cannot be observed. But more generally, R-squared will depend on your data and model, specifically the nature of your dependent variable, any transformations you may have applied to the variable, what explanatory variables you include and what variables you exclude.

In Economics research, R-squared is rarely of concern because predictive power is often not the main goal. Instead, economists focus on finding reliable estimates of coefficients for variables of interest and deriving useful inferences from the structure of the model and the estimated values of its parameters. And standard error of the regression is considered a better metric than R-squared because it comes in the units of the dependent variable, provides the absolute measure of the typical distance that the data points fall from the regression line, and it scales the width of all confidence intervals calculated from the model.

Source for some of the points included in this answer: https://people.duke.edu/~rnau/rsquared.htm

  • $\begingroup$ Makes sense. My dependent variable is a normally distributed principal component (PC1) after reducing the dimension of a couple of variables, which is pretty vague, to begin with, and I am working with individual-level data, and which has a lot of variation. $\endgroup$
    – Jerry
    Commented Mar 22, 2019 at 18:35
  • $\begingroup$ Right, so principal components, by construction, already capture a limited percentage of the original variance of the variable, which will drive down the explained variance even with a strong set of explanatory variables. $\endgroup$
    – AlexK
    Commented Mar 22, 2019 at 19:06

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.