+ Reply to Thread
Results 1 to 8 of 8

Thread: comparing models

  1. #1
    Super Moderator
    Points: 31,766, Level: 100
    Level completed: 0%, Points required for next Level: 0
    bugman's Avatar
    Posts
    2,255
    Thanks
    290
    Thanked 324 Times in 265 Posts

    comparing models




    This might be a "101" question but hey, The Ecologist" encouraged me to ask questions , lots and lots of stupid questions.

    So when I have two competing models and I use
    Code: 
    anova(m1,m2)
    to test whether a reduction of the residual sum of squares is statistically significant or not, I understand that it is based on a chi-squared test, but I read that it only makes sense if the models are nested.

    Why does it only makes sense if the models are nested?
    The earth is round: P<0.05

  2. The Following User Says Thank You to bugman For This Useful Post:

    gianmarco (08-21-2015)

  3. #2
    TS Contributor
    Points: 12,227, Level: 72
    Level completed: 45%, Points required for next Level: 223
    rogojel's Avatar
    Location
    I work in Europe, live in Hungary
    Posts
    1,470
    Thanks
    160
    Thanked 332 Times in 312 Posts

    Re: comparing models

    hi,
    just a quick idea, before coffee: in anova the F test uses the ratio of the variance of the effect and the residuals. So, to compare the two models one needs a common basis for the residual noise. If the two models were completely independent that would not work.

    does this make sense?

    regards

  4. The Following User Says Thank You to rogojel For This Useful Post:

    bugman (08-21-2015)

  5. #3
    Super Moderator
    Points: 31,766, Level: 100
    Level completed: 0%, Points required for next Level: 0
    bugman's Avatar
    Posts
    2,255
    Thanks
    290
    Thanked 324 Times in 265 Posts

    Re: comparing models

    So in this context the term nested is not referring to nested models as in random effects nested in fixed effects, we are saying the models must be subsets of the models we are testing?

    As in:

    m1 <- aov(y~x1+x2+x3+error)

    vs

    m2<-aov(y~x1+x2+error)?

    Am I on the right track?
    The earth is round: P<0.05

  6. #4
    TS Contributor
    Points: 12,227, Level: 72
    Level completed: 45%, Points required for next Level: 223
    rogojel's Avatar
    Location
    I work in Europe, live in Hungary
    Posts
    1,470
    Thanks
    160
    Thanked 332 Times in 312 Posts

    Re: comparing models

    Yes, I think so.

    regards

  7. #5
    Points: 657, Level: 13
    Level completed: 14%, Points required for next Level: 43

    Posts
    57
    Thanks
    1
    Thanked 9 Times in 8 Posts

    Re: comparing models

    Nested models can be several things in the world of statistics. However, with regards to model comparison, then this refers to the larger model containing all the terms of the smaller model (as you represented in your equations).

  8. #6
    Super Moderator
    Points: 31,766, Level: 100
    Level completed: 0%, Points required for next Level: 0
    bugman's Avatar
    Posts
    2,255
    Thanks
    290
    Thanked 324 Times in 265 Posts

    Re: comparing models

    Thanks evelyn13,

    yes, looks like I was getting the terminology confused. Actually I was aware that it only makes sense to be comparing full and reduced models, it was the use of the word nested that threw me.
    The earth is round: P<0.05

  9. #7
    TS Contributor
    Points: 12,227, Level: 72
    Level completed: 45%, Points required for next Level: 223
    rogojel's Avatar
    Location
    I work in Europe, live in Hungary
    Posts
    1,470
    Thanks
    160
    Thanked 332 Times in 312 Posts

    Re: comparing models

    Hi,
    this seems to be sort of accepted:

    E.g. in this book they use the term "nested model" as you use it.

    http://www.amazon.de/gp/product/0387...h_gw_p14_d0_i1

  10. #8
    Omega Contributor
    Points: 38,303, Level: 100
    Level completed: 0%, Points required for next Level: 0
    hlsmith's Avatar
    Location
    Not Ames, IA
    Posts
    6,993
    Thanks
    397
    Thanked 1,185 Times in 1,146 Posts

    Re: comparing models


    Just stumbled on this thread wjhile looking for another, they typically have to be nested so you have a degree of freedom to look up in the Chi sq dist. So if one model had 3 terms and the other 2, you woulud be on the Chi sq dist with 1 degree of freedom.

    And as mentioned they are nest within each other (saturated and reduced) .

+ 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