$X$ is democracy level

$Y$ is infant mortality rate (IMR)

In my model, 1 unit increase in $X$ predicts a $20\%$ decrease in $Y$

Does this mean 5 unit increase in $X$ will eradicate IMR? In other words, can I conclude that 5 unit increase in democracy predicts $100\%$ decrease in IMR?


The regression models of positive outcomes usually consider multiplicative decrease in outcome and additive change in predictor. Assuming model $\log Y=\beta X+\epsilon$. Then an additive change in $X$ by one unit changes $Y$ by a multiplicative factor $e^\beta$, which can be translated into percent change with $(e^\beta-1)\cdot 100$.

Increasing $X$ by five units i.e. $X=x_0+5$ gives us $Y=y_0 \cdot \exp(\beta)^5$ with $y_0=\epsilon \exp(\beta x_0)$. Assuming that

1 unit increase in X predicts a 20% decrease in Y

then $\exp(\beta)= 1-20/100=.8$ and for 5 units increase in $X$, $Y$ decreases by a factor $\exp(\beta)^5=0.8^5=0.33$. That is approx. $67\%$ decrease.

It is sometimes recommended to approximate the logarithm with $\log(z)=z-1$ which is what You implicitly do when You claim that the percantage change is equal to $\beta$. The approximation is only valid for small changes in $Y$. Thus, with a change in $X$ by one unit, the approximation gives $-19.5$ for a true value of $(e^\beta-1)\cdot 100=-17.71\%$ which is reasonably accurate. However, when $X$ changes by five units, the approximation fails. Approximation suggests a change by $-97.5\%$ while the true value is $(e^{5\beta}-1)\cdot 100=-62.28\%$. When in doubt avoid the approximation and perform the exact calculation.

  • $\begingroup$ Hi Matus, thanks for answering. could you explain a bit more please. i am trying to find the relationship between democracy level and infant mortality rate (IMR). I log-transformed the dependent variable (IMR). my equation is now: log Y= 4.107-0.195*democracylevel. The interpretation is: one unit increase in democracy level will predict 19.5% reduction in IMR. but how do i find the effect if democracy level is increased by 5 units? $\endgroup$ – Hans Apr 18 '17 at 18:34
  • $\begingroup$ @Hans I added more details. Let me know if additional clarification is needed. $\endgroup$ – matus Apr 18 '17 at 23:41

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.