0
$\begingroup$

I was wondering about this problem while I was running a normal regression. Assume that we have the standard regression setup with one dependent variable $Y$ and an independent variable $X$, and the model is: $log(100/Y_i) \sim N(\beta_0 + \beta_1X_i, \sigma^2), i = 1, \ldots, n$.

Suppose that we now fit a new model $log(150/Y_i) \sim N(\beta_0^{'} + \beta_1^{'}X, \sigma^2)$, what would the relationship between the regression coefficients $(\beta_0, \beta_1)$ and $(\beta_0^{'}, \beta_1^{'})$ be?

While it is easy to derive this relationship in the case of a simple linear regression, i.e., if $Y_i \sim N(\beta_0 + \beta_1X_i, \sigma^2)$, then $Y_i + 100 \sim N(100 + \beta_0 + \beta_1X_i, \sigma^2)$, and $2*Y_i \sim N(2 * \beta_0 + 2 * \beta_1X_i, 4\sigma^2)$, is there a way to derive the relationships between $(\beta_0, \beta_1)$ and $(\beta_0^{'}, \beta_1^{'})$ in a similar way?

$\endgroup$
  • 1
    $\begingroup$ Hint: $\log(150/Y) = \log(150/100) + \log(100/Y)$. Take it from there. $\endgroup$ – whuber Aug 23 '17 at 18:11
  • $\begingroup$ Hmm.. so the intercepts would differ by log(1.5) while the slopes would remain the same... wonder if this would be a same in case it is a mixed effects model with X as the fixed effect and Z as a simple scalar random effect. $\endgroup$ – buzaku Aug 23 '17 at 18:35
0
$\begingroup$

You can open the log term.

log(100/Y) -> log(100) - log(Y)

log(150/Y) -> log(150) - log(Y)

Then the difference between both will be a constant term.

$\endgroup$

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.