1

I have a question regarding the interpretation of a log independent variable in a linear-probability model.

For example: I have $\log(GDP)$ as my independent variable and the coefficient is 0.35. Can I then interpret the result as follows: A one UNIT increase in the log of GDP is associated with a 3.5% increase in the probability that $Y = 1$?

I know that this question was answered elsewhere, however, people talked about a PERCENT instead of unit increase and this confused me.

Stephan Kolassa
  • 95,027
  • 13
  • 197
  • 357
Julian
  • 11
  • 1
  • 2

1 Answers1

3

If your $y$ variable is binary, i.e. 0 or 1, then one interpretation of your coefficient can be is as follows: a one unit increase in log GDP would increase $y$ by .35.

But let's clarify your other area of concern. Let's express this model as follows $$y_{i} = \beta_{0} + \beta_{1}log(GDP_{i}) + u_{i}.$$

If $y_{i} \in [0,1]$, then we can think about the dependent variable as measured in terms of percentage. Right? Recall $\Delta{log(x)} \approx \%\Delta x$. What happens when we add a little extra to GDP? Let's call the new dependent variable $\tilde{y}_{i}$. Hence, we can write the following $$\tilde{y}_{i} - y_{i} = \beta_{1}\left\{log(GDP_{i} + \delta) - log(GDP_{i})\right\}.$$

Hence,

$$\beta_{1} = \frac{\tilde{y}_{i} - y_{i}}{log(GDP_{i} + \delta) - log(GDP_{i})}$$.

Therefore, if the change in $y$ is very small, we can think of numerator as a percentage point change in $y$. Thus, it has the interpretation of elasticity, i.e. $\beta_{1}$ is the percent point by which the dependent variable responds to a given percentage change in the independent variable. Hopefully that clarifies both areas of confusion for you!

JuliusBilly
  • 507
  • 3
  • 20
  • Thank you very much for your answer. Just to get that right. Since I have a binary outcome in my linear proability model I can interpret the coefficients (i.e 0.35) as 0.35/100 increase in the probability of Y=1. Moreover, when I have the log of an independent variable, I also talk about percent increase instead of unit increase when it comes to the independent variable. Hence, a one percent increase in the log of GDP is associated with a 0.35/100 unit increase in the probability that Y=1? Is that correct? – Julian Jul 17 '18 at 20:24
  • 1
    @Julian not quite. Please carefully read my answer. $\hat{\beta}_{1}$ in this case is the elasticity measure in itself. Hence, a 1% increase in GDP elicits a 0.35% change in $y$. The other interpretation is a one unit increase in log of GDP increases $y$ by 0.35. – JuliusBilly Jul 18 '18 at 01:30
  • 1
    Can you clarify why the dependent variable is measured in terms of percentage when it is binary? The numerator seems to be in percentage points, rather then percentage. – dimitriy Oct 15 '19 at 03:04
  • This sounds more like a semi-elasticity to me rather than an elasticity. I also think you might have to divide by 100, like I did [here](https://stats.stackexchange.com/a/326138/7071). – dimitriy Oct 15 '19 at 03:59