0
$\begingroup$

I know that the log-linear model shows the percent change in y if there is a one-unit change in x but how would you solve for it the other way to show a percent change in x if there is a one-unit change in y?

$\endgroup$
1
$\begingroup$

In principle you can just take the model coefficients and solve for $\ln x$. If you have:

$$ \ln y = \hat{\beta}_0 + \hat{\beta}_1 \ln x \implies \ln x = -\frac{\hat{\beta}_0}{ \hat{\beta}_1} + \frac{1}{ \hat{\beta}_1} \ln y$$

However, this being said even though you can do this mathematically the OLS assumes that the relationship is exogenous, so solving for $\ln y$ would not allow you to interpret the $1/ \hat{\beta}_1$ as a causal effect - it is just inverse of the effect of $\ln x$.

If you believe that there is relationship that goes both ways you should be using some model that allows for that. A vector autoregression (VAR) would be one example of such model. VAR would estimate the relationship going both ways giving you separate set of coefficients when $\ln x$ is the dependent variable. However, this is just one example, there is a cornucopia of models that can help deal with endogeneity. For overview of them you can see Verbeek (2008) A Guide to Modern Econometrics, 4th ed.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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