2
$\begingroup$

I ran a Poisson regression in SPSS with hypothesis: increase in level of democracy leads to reduction in infant mortality (IMR). So democracy is the independent variable and IMR the dependent.

I got the coefficient of democracy -0.186.

Does this mean, one unit increase in democracy level will lead to a $exp(-0.186)=0.83$ case decrease in IMR?

$\endgroup$
2
$\begingroup$

Not exactly. When using Poisson regression with the default log link function, the prediction is

$$ E(Y | X) = e^{\beta_0 + \beta_1 X_1 + \dots + \beta_k X_k} = e^{\beta_0} e^{\beta_1 X_1} \dots e^{\beta_k X_k} $$

so if, for example, $\beta_1 = -0.186$, then its effect on $X_1$ is $e^{\beta_1 X_1}$. It is not the same as $e^{\beta_1} X_1$, since given the properties of exponents this would lead to $e^{\beta_1 + \log X_1}$.

$\endgroup$
  • $\begingroup$ Tim, thanks for your answer. what about natural log? is it the same as log? $\endgroup$ – hans Apr 25 '17 at 12:23
  • $\begingroup$ @hans yes, in the above answer I used $\log$ for $\log_e$ $\endgroup$ – Tim Apr 25 '17 at 12:46

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.