0
$\begingroup$

I'm running logistic regression for ecological samples using R Deducer.. Specifically, I'm studying the habitat preference of a specific bird species.. There are eight predictor variables believed to affect the presence of this bird species and I'm looking for the most significant habitat (predictor) variable that affects the presence of the species.. Of the eight predictor variables, only one came out as significant, the variable X.21m (trees with height equal or greater than 21m) .. I understand that the relationship of the variable X.21m is positive with the presence of the species but by how much (odds ratio)? Attached is the photo of the results output from R.. I'm not familiar with interpreting the coefficient results, and I would like to ask help in this matter and for suggestions to make it better..

P.S. I'm trying to learn statistics that's relevant in my line of work as I find it really interesting and adds more dimension in what I do.. Thank you in advance for your time..

Logistic regression result in R Deducer

Improve this question
$\endgroup$
6
  • $\begingroup$ Is X.21 a binary variable (either the tree is >= 21m or it is not)? Either way, this is a good resource: stats.idre.ucla.edu/other/mult-pkg/faq/general/… $\endgroup$
    – Michael Webb
    Jul 11, 2017 at 14:24
  • $\begingroup$ Parameter value is 0.49, so the odds ratio is exp(0.49) = 1.63. Additionally, you seem to have several categorical variables relating to height; if you have the precise heights, you could do a more informative analysis by using height as a continuous variable. $\endgroup$
    – mkt
    Jul 11, 2017 at 14:25
  • $\begingroup$ Thank you for the response.. Yes there are 3 categories for tree height but we have lumped then between those ranges and did not do precise height categorization.. So based from the odds ratio (1.63), is it right to say that with X.21 habitat variable, it is 1.63 times more likely that the species would be present? $\endgroup$
    – Andrew Reintar
    Jul 11, 2017 at 15:29
  • $\begingroup$ @AndrewReintar You should probably treat height as one ordinal variable with 3 levels, instead of 3 separate levels. But your statistical interpretation of the present result is correct: the bird is 1.63 times more likely to be present on trees >= 21m. $\endgroup$
    – mkt
    Jul 11, 2017 at 18:01
  • $\begingroup$ @mkt Ok, understood.. Also related to this topic, if in my results, there came out two significant variables (<0.05), how do I interpret that? For example aside from trees with heights equal or greater than 21m (X.21m), Canopy cover was also found to be significant but with a negative correlation.. $\endgroup$
    – Andrew Reintar
    Jul 12, 2017 at 1:50

1 Answer 1

Reset to default
0
$\begingroup$

Quantitative variable are easy to interpret in GLM, categorical ones can be quite tricky without prior experience. As mentioned to you in some comments already, you must always interpret the exponent of the coefficients (e.g. eb0, eb1 etc.) and never the raw estimate.

For continuous variables, the interpretation is always the same: ebi is the average change in the probability parameter of your Bernoulli event for 1 unit of increase in variable xi (ceteris paribus).

Improve this answer
$\endgroup$

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.