+ Reply to Thread
Results 1 to 2 of 2

Thread: Information Criterion to compare different linear models

  1. #1
    Points: 189, Level: 3
    Level completed: 78%, Points required for next Level: 11

    Posts
    6
    Thanks
    0
    Thanked 0 Times in 0 Posts

    Information Criterion to compare different linear models




    Dear Community,

    In order to perform a price forecast, I built four linear models:1-3 have one independent variable, 4 has two independent variables. n = 28 in all cases
    1. ad. R = 0.91, AIC = -27, BIC = -25
    2. ad. R = 0.95, AIC = -37, BIC = -34
    3. ad. R = 0.97, AIC = -70, BIC = -68
    4. ad. R = 0.95, AIC = -35, BIC = -32

    Which statements can be derived from these results?

    As I understood the issue so far, the lower AIC/BIC score the better. However, both information criteria penalize the usage of more independent variables as the risk of overfitting increases.
    Therefore, I would drive that model 3 is the best model as adjusted R is high as well as AIC and BIC are the lowest of all four models. Further I would prefer 4 over 2 as ad. R is the same and 4 as lower AIC/BIC scores even though it has two independent variables in comparison to one in 2. Model 1 would be the worst performing model.

    Is it possible to make these statements?

    Thank you very much for your clearification.

    Best,

    Anton

  2. #2
    Points: 189, Level: 3
    Level completed: 78%, Points required for next Level: 11

    Posts
    6
    Thanks
    0
    Thanked 0 Times in 0 Posts

    Re: Information Criterion to compare different linear models


    Hi Community,

    I just saw that my post has over 100 views. Has somebody a short comment on my post? I would be really intested whever I understood the theory, especially since I am not quite sure why the BIC values are negative as the formula is as follows:
    BIC = -2 * ln(likelihood) + ln(n)*k
    k = number of parameters estimated
    n = observation

    ln(likelihood) has to be in my oppinion negative, as the likelihood has values from 0 to 1, resulting in a negative value for ln(likelihood). -2 * ln(likelihood) therefore has to be positive; the other term ln(n)*k has to be positive as well, so currently I cannot understand why I received negative BIC scores.

    I would highly appreciate your comments!

    Best,

    Anton

+ Reply to Thread

           




Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts






Advertise on Talk Stats