Databases
StackOverflow2013
Postsdbo
POUsing clustered covariance matrix in predict.lm()
 

Note that there are some explanatory texts on larger screens.

plurals
  1. POUsing clustered covariance matrix in predict.lm()
    primarykey
    data
    text
    <p>I am analyzing a dataset in which data is clustered in several groups (towns in regions). The dataset looks like: </p> <pre><code>R&gt; df &lt;- data.frame(x = rnorm(10), y = 3*rnorm(x), groups = factor(sample(c('0','1'), 10, TRUE))) R&gt; head(df) x y groups 1 -0.8959 1.54 1 2 -0.1008 -2.73 1 3 0.4406 0.44 0 4 0.0683 1.62 1 5 -0.0037 -0.20 1 6 -0.8966 -2.34 0 </code></pre> <p>I want my lm() estimates to account for intraclass correlation in groups and for that purpose I am using a function <code>cl()</code> that takes an <code>lm()</code> and returns the robust clustered covariance matrix (original <a href="http://people.su.se/~ma/clustering.pdf" rel="noreferrer">here</a>):</p> <pre><code>cl &lt;- function(fm, cluster) { library(sandwich) M &lt;- length(unique(cluster)) N &lt;- length(cluster) K &lt;- fm$rank dfc &lt;- (M/(M-1))*((N-1)/(N-K-1)) uj &lt;- apply(estfun(fm), 2, function(x) tapply(x, cluster, sum)); vcovCL &lt;- dfc * sandwich(fm, meat = crossprod(uj)/N) return(vcovCL) } </code></pre> <p>Now, </p> <pre><code>output &lt;- lm(y ~ x, data = df) clcov &lt;- cl(output, df$groups) coeftest(output, clcov, nrow(df) - 1) </code></pre> <p>gives me the estimates I need. The problem now is that I want to use the model for prediction, and I need the standard error of the prediction to be calculated with the new covariance matrix <code>clcov</code>. That is, I need</p> <pre><code>predict(output, se.fit = TRUE) </code></pre> <p>but using <code>clcov</code> instead of <code>vcov(output)</code>. Something like a <code>vcov() &lt;-</code> would be perfect. </p> <p>Of course, I could write my own function to do predictions, but I am just wondering whether there is a more practical method that allows me to use methods for signature <code>lm</code> (like arm::sim).</p>
    singulars
    1. PO
      singulars
      1. PTAnswer
    1. This table or related slice is empty.
    1. PTQuestion
    1. USgriverorz
    1. USgriverorz
    plurals
    1. This table or related slice is empty.
    1. This table or related slice is empty.
    1. This table or related slice is empty.
    1. PO
      singulars
      1. PTAnswer
    2. PO
      singulars
      1. PTAnswer
    1. VO
      singulars
      1. POUsing clustered covariance matrix in predict.lm()
      1. This table or related slice is empty.
      1. VTUpMod
    2. VO
      singulars
      1. POUsing clustered covariance matrix in predict.lm()
      1. This table or related slice is empty.
      1. VTUpMod
    3. VO
      singulars
      1. POUsing clustered covariance matrix in predict.lm()
      1. This table or related slice is empty.
      1. VTUpMod
    1. COYou need to specify a bit more. What is that cluster function to start with? Why are the standard errors coming out of lm() not valid? I can't really follow what you try to do. It might very well be you need a more generalized model, eg glm, glmm or gam/gamm. There is very little left to do on standard errors of simple lm functions, unless you use them in a completely different context. But then we need the context...
      singulars
      1. POUsing clustered covariance matrix in predict.lm()
      1. USJoris Meys
    2. CO@Joris I have edited the question. Hope it is clearer now. Please, note that I am explicitly avoiding a `glmm` model.
      singulars
      1. POUsing clustered covariance matrix in predict.lm()
      1. USgriverorz
 

Querying!

 
Guidance
A row detail

Detail views are divided into sections. All the information in the data section comes from columns in the selected row. The other sections display data from other, related rows.

Related data can be related in a to-one or a to-many fashion. Captions of data related in a to-many fashion link to a list view showing a filtered view of the table.

Try moving around until you find a non-empty to-many entry and click on the label to get to one. You can move back to the root by clicking on the database name in the header.

SQuiL has stopped working due to an internal error.

If you are curious you may find further information in the browser console, which is accessible through the devtools (F12).

Reload