How can I compare three linear models (fitlme)?

3 次查看（过去 30 天）
David Obert
David Obert 2023-11-9
评论： David Obert 2023-11-19
I have three different models with different random effects and funtions:
modelspec = 'Latency ~ 1 + TestNumber + (1|AnimalID)';
lmeinter = fitlme(data, modelspec);
modelspec = 'Latency ~ 1 + TestNumber + (1 + TestNumber|AnimalID)';
lmelin = fitlme(data,modelspec);
formula = 'Latency ~ 1+ TestNumber + TestNumber^2+ (1 + TestNumber|AnimalID)';
lmequad = fitlme(data, formula);
Now I would like to compare the performance of the different models. They are all nested. Using the "compare" function (Likelihiid ration test), I can compare two, but not three models. Is there a way to run an anova on models?
  2 个评论
Harald
Harald 2023-11-10
Hi,
how about comparing two models, and then comparing the better model of the two to the third one?
Best wishes,
Harald
Star Strider
Star Strider 2023-11-10
One option could be to get the residuals from each regression (as column vectors) and then use the friedman function to see if any are different. Another option might be the multcompare function (there are also other versions of the function, so consider them to see which is most appropriate for what you want to do).

请先登录，再进行评论。

请先登录，再回答此问题。

采纳的回答

Shivansh
Shivansh 2023-11-15
Hi David!
I understand that you want to compare three different linear mixed-effects models.
One approach to do this is by pairwise comparison of the models. You can do this using the ‘compare’ function using the following code.
% Compare the models pairwise
comparison1 = compare(lmeinter, lmelin);
comparison2 = compare(lmeinter, lmequad);
comparison3 = compare(lmelin, lmequad);
% Display the results
disp(comparison1);
disp(comparison2);
disp(comparison3);
If you want to compare more than two models simultaneously, I am not sure about how you want to use anova but you can use the Akaike Information Criterion (AIC) or the Bayesian Information Criterion (BIC). Both AIC and BIC are measures of the goodness of fit of an estimated statistical model and can be used for model selection.
Lower values of AIC or BIC indicate better fitting models. You can implement it by adding the following code:
AIC = [lmeinter.ModelCriterion.AIC lmelin.ModelCriterion.AIC lmequad.ModelCriterion.AIC];
BIC = [lmeinter.ModelCriterion.BIC lmelin.ModelCriterion.BIC lmequad.ModelCriterion.BIC];
disp('AIC:');
disp(AIC);
disp('BIC:');
disp(BIC);
You can refer to the following documentation links for more information:
  1. compare: https://www.mathworks.com/help/stats/linearmixedmodel.compare.html.
  2. AIC and BIC: https://www.mathworks.com/help/ident/ref/idmodel.aic.html.
  3. Information Criterion for model selection: https://www.mathworks.com/help/econ/information-criteria.html.
Hope it helps!
  1 个评论
David Obert
David Obert 2023-11-19
Thank you for the answer! I combined your answer with the comment comment of Star Strider. I use the compare function as well as the AIC (as you proposed). Additionally, I calculated the residual sum of square. So I have three measures, all being consistent and demonstrating that lmequad is the best model.

请先登录，再进行评论。

更多回答（0 个）

类别

在 Help Center 和 File Exchange 中查找有关 Analysis of Variance and Covariance 的更多信息

标签

产品


版本

R2022a

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by