+ Reply to Thread
Results 1 to 1 of 1

Thread: a question about estimates in mixed models

  1. #1
    Points: 889, Level: 15
    Level completed: 89%, Points required for next Level: 11

    Thanked 0 Times in 0 Posts

    a question about estimates in mixed models


    I was hoping to be able to find all of the estimates of covariates and of the categories with a number of factors. I found a similar study to mine that took advantage of mixed models to assess which covariates most affected foraging behaviour in piping plovers and as I have similar data thought that this may be a good approach.

    jpkelley on here was the first to suggest that models would be the best approach due to a small sample size (but lots of data) but I really struggled with the concept.

    I am down to the wire now and this past week was able to set up all appropriate data in a csv file and I design a global model. I ran dredge in MUMIn to find which models had the smallest deltas and then wanted to average the parameters through model.avg.

    The original paper seemed to use these techniques and ended up averaging two models. They then showed the results in two tables, one listing the top nine models and their AIC values etc. and then another table showing the results of the model averaging (a list of Estimates and CIs for the Intercept, the covariates and also each category with the fixed-effects factors.

    The issue is that when I run a model, I only get Estimates for the second and greater categories within a factor e.g. if a category has two categories, it only shows the one for the second one). I was advised to run the model with -1 to remove the intercept, but although this worked to get the Estimates for the first factor, it continued to only give me the second for the other factor categories. In addition, the distance between the estimates within a factor do not only change directly, the size of the estimates do (greatly) and also the distance between different covariates also change, along with the z scores and p-values (however the relative weights of importance do not).

    The original paper shows an intercept score AND then an estimate for each level of each factor. Does anyone know how this is possible? This is really the last issue I am dealing with and with so much help on here already given and many hours of trudging through statistics, I would very much appreciate help with this.
    If there isn't a way, would it be acceptable just to show the top models and the parameter weights?
    Thank you.
    Last edited by Dragonwalker; 03-28-2012 at 10:43 AM.

+ 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