0

I have a linear regression model with violated assumptions (linearity and Constant variance), thus I decided to take the log of the dependent variable and this step solved the problem.

Afterwards, the coefficient of the independent variable (BMI) was:

Coefficients:
                     Estimate Std. Error t value Pr(>|t|)    
(Intercept)          1.249998   0.342765   3.647  0.00032 ***
i$BMI               -0.017679   0.011301  -1.564  0.11892

                 2.5 %     97.5 %



(Intercept)          0.57510785 1.92488723
i$BMI               -0.03992972 0.00457222

I exponentiated the coefficient (-0.017679 ) and reported the coefficient as (0.9824764)

the confidence intervals were ( -0.03992972 to 0.00457222) and when I exponentiate them, they became (0.960857 to 1.004583). the new intervals don't include 0 which means it is significant while the p.value is over 0.05. Thus, I assume they are wrong.

please, advise on how to handle this and report the right CIs.

kjetil b halvorsen
  • 77,844
Ram6
  • 1
  • Why do you have the logistic tag? Is the outcome binary? The multiplicative vs additive effects difference has been covered here. – dimitriy Feb 14 '23 at 00:55
  • no, I just removed the tag. – Ram6 Feb 14 '23 at 01:04
  • Hints about how to think about this. 1. Is it possible to exponentiate a number and get a negative number as a result? Think carefully! 2. Let's assume the estimated coefficient of BMI was exactly equal to zero. Would it be significantly different from zero? 3. What is $\exp{0}$? – jbowman Feb 14 '23 at 01:10
  • exactly, that's the challenge. Don't know how to go about that! – Ram6 Feb 14 '23 at 01:17
  • Try answering the questions, that should help! – jbowman Feb 14 '23 at 01:20

0 Answers0