# Talk:Mixed model

WikiProject Statistics (Rated Start-class, High-importance)

This article is within the scope of the WikiProject Statistics, a collaborative effort to improve the coverage of statistics on Wikipedia. If you would like to participate, please visit the project page or join the discussion.

Start  This article has been rated as Start-Class on the quality scale.
High  This article has been rated as High-importance on the importance scale.
WikiProject Mathematics (Rated Start-class, Mid-importance)
This article is within the scope of WikiProject Mathematics, a collaborative effort to improve the coverage of Mathematics on Wikipedia. If you would like to participate, please visit the project page, where you can join the discussion and see a list of open tasks.
Mathematics rating:
 Start Class
 Mid Importance
Field: Probability and statistics

This article redirects to multilevel model, but this is incorrect. A multilevel model may be mixed, but not necessarily so. A mixed model is one in which some of the effects are considered to be fixed and others are considered to be random. Someone with a bit more statistical knowledge than I have could perhaps re-write the mixed-model article. --Crusio (talk) 09:57, 2 September 2008 (UTC)

Done, better late than never. I acknowledged the undoing of the redirect at WikiProject Statistics, so there will be higher visibility. Baccyak4H (Yak!) 02:55, 19 June 2009 (UTC)
##### Computation

There was a nebulous explanation of the details on how the mixed models (linear) are fit computationally. I gave a reference to the Lindstrom Bates article which describes iteratively fitting the linear model and estimating the variance components using the EM algorithm. I also cited R and SAS since this is how they do it. Might be good to format the procedures as computed code, instead of appearing as garbled jargon. My wiki-fu is not good enough to code them as such, yet.Saffloped (talk) 22:00, 12 September 2011 (UTC)

It seems that this sentence was added:

More recently, methods for maximum likelihood estimation of mixed models have become more widely used than least-squares based methods.[7]

This is not right nor is it wrong: it's confusing. Ordinary least squares minimizes the variance of the unweighted residuals. The point of the mixed model is that the variance components are not equal, so a more efficient estimation method is inverse variance weighting. The EM algorithm is necessary because it iteratively estimates the variance, applies the weights, and estimates model parameters. I say either remove the line, or don't speak of methods in terms of their "popularity" but their justification

```Saffloped (talk) 20:30, 27 September 2011 (UTC)
```
##### Definition

Can we add dimensions to all the matrices, i.e.

```* y is a vector of n observations
* beta is a vector of fixed effects of length n
* X is a matrix of ... with dimensions n x p where p is ...
* u is a matrix of ... with dimensions n x q where q is ...
```

I think this would make things way more clear for people seeing things for the first time (fill in the ellipsis's, I don't feel qualified to do so) MATThematical (talk) 06:30, 31 December 2013 (UTC)