3
$\begingroup$

If I have a level-log model, such as
y= β0+β1log(x), I know the interpretation is "If we increase x by one percent, we expect y to increase by (β1/100) units of y."

So, my questions are

  1. If y is a percentage variable, what is the correct interpretation of the resulting regression? Is it still the same but has a unit of percentage instead of unit, so the interpretation becomes "If we increase x by one percent, we expect y to increase by (β1/100) percent of y"? Or we don’t have to divide β1 by 100, so the interpretation becomes "If we increase x by one percent, we expect y to increase by β1 percent of y"?

  2. Can I convert my variables (ex: interest rate and inflation rate) to logarithm if it has a unit of percentage? If it can be done, can I change my model into log-log model, although my variable y has a unit of percentage, so the model becomes log(y)= β0+β1 log(x) ? What is the correct interpretation of this model?

$\endgroup$
3
$\begingroup$

General note: Distinguish between percent and percentage points.

Answer for Question 1: If y is a percentage your model will be: $$(\%)y=\beta_0+\beta_1 log(x)$$

$$\frac{d(\%)y}{d log(x)}=\beta_1 $$ $$\frac{d(\%)y}{d log(x)}=\frac{d(\%)y}{dx/x}=\beta_1$$ Dividing both sides by 100: $$\frac{d(\%)y}{d log(x)}=\frac{d(\%)y}{100 \times dx/x}=\frac{\beta_1}{100}$$

Your interpretation is: when X increases by 1 percent, y increases by $\beta_1$ percentage points.

Answer for Question 2: If you have any variable in percentage you can keep it in percentage. It is meaningless to transform a variable that is already in percentage into logarithm. You will be questioned with the zero problem: log (1)=0. I totally do not recommend this.

$\endgroup$

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.