Discussion: View Thread

  • 1.  Negative -2 log likelihood in Proc Mixed

    Posted 05-02-2013 15:58
    I am working on data with repeated measures and analyzing using Proc Mixed in SAS 9.3.  I am trying to use the fit statistics to decide on the covariance structure to use but have one concern.  I do know that sometimes the -2 log-likelihood is negative when it should never be negative.  I used to know why it was sometimes negative but have forgotten and a Google search did not yield an answer quickly.  I assume that the constant is not included and were the constant included the value for -2 log-likelihood would be positive.  Is this correct?  If not why is this value negative?

    Thank you for any help.

    -------------------------------------------
    Robert Podolsky
    Georgia Regents University
    -------------------------------------------


  • 2.  RE:Negative -2 log likelihood in Proc Mixed

    Posted 05-02-2013 17:05
    I'll admit it has been a while since graduate school and actually using equations (think pre-internet), but I think it need not be positive.

    For example take the case of the normal distribution with n samples.  I think that

       -2 log-likelihood = positive number  +  n*ln(sigma-hat-squared)

    So if sigma-hat-squared  is enough less than one, then the value can be negative.

    -------------------------------------------
    Michael Morton
    -------------------------------------------








  • 3.  RE:Negative -2 log likelihood in Proc Mixed

    Posted 05-02-2013 17:51

    Robert:

    I agree with Michael. In addition, you may have another problem. If you use random effects, you can employ AIC and similar model selection criteria only to compare models that differ by a fixed effect. E.g., you can't compare a model where Factor A is fixed to a model where Factor A is assumed random using AIC/AICc/BIC and such like. For more, check out my review of this book:

    http://www.amazon.com/Model-Selection-Multi-Model-Inference-Information-Theoretic/product-reviews/1441929738/ref=cm_cr_pr_top_recent?ie=UTF8&showViewpoints=0&sortBy=bySubmissionDateDescending


    -------------------------------------------
    Nikita Tuzov
    Ph D Student
    -------------------------------------------








  • 4.  RE:Negative -2 log likelihood in Proc Mixed

    Posted 05-02-2013 21:29
    Hi All,

    Regarding Robert's original question, others have answered this clearly.  Think of a very narrow normal distribution.  The area under the thing has to be 1, so the peak may need to be very tall.  In a simple case, the likelihood might be a product of such densities, and so can be arbitrarily large.  More generally the likelihood is always positive, and can be above 1 if the corresponding density is sufficiently concentrated. The log can therefore be positive or negative, as can -2 times the log.

    Nikita's comment needs some clarification.  If you are using the default REML algorithm in PROC MIXED to estimate the variance components (i.e., if you have not overridden it with something like METHOD=ML in the PROC statement), then variance components and the likelihood are based on pseudo-data that depends on the fixed effects. Variance components and likelihoods from models with different fixed effects (i.e., different MODEL terms) are therefore based on different pseudo-data and are NOT comparable.  However, models with the same fixed effects produce likelihoods on the same pseudo-data, and comparisons of variance structures using information criteria can be made. 

    If you use full maximum likelihood (METHOD=ML), then you can use information criteria to compare models that differ in fixed and/or random effects.  However, the estimates of the variance components from these models are potentially somewhat biased.  

    This book (http://www.amazon.com/Mixed-Models-Second-Ramon-Littell/dp/1590475003) is indispensable if you want to do modeling using mixed models in SAS.

    -Tom.

    -------------------------------------------
    Thomas Loughin
    Simon Fraser University
    -------------------------------------------








  • 5.  RE:Negative -2 log likelihood in Proc Mixed

    Posted 05-02-2013 17:07


    -------------------------------------------
    Darwin Poritz
    -------------------------------------------
    -2 log L may be negative or positive depending on whether the density function is greater than or less than 1.  The important consideration is the difference in -2 log L for one hypothesis vs another.  A. W. F. Edwards, Likelihood, Cambridge U. P., 1972, pp 174-198, suggests that an absolute difference for log L of 2 or greater is significant, i.e., an absolute difference for -2 log L of 4 or greater.

    Darwin Poritz
    Statistician, JSC Engineering, Technology and Science Contract
    Houston, Texas