CODEWITHSUNDEEP
rpoolimputationp-value
thdr
thdr

Reputation: 11

p-value using pool.scalar in R

How can I compute the p-value of the model coefficients of a multiple imputation model using the pool.scalar function? When using the pool function I can just do summary(pool(imputed_model)) and get the model coefficient and their p-values. However, I am fitting a quantile regression model and the pool function does not support rq models. Therefore, I am using pool.scalar to apply Rubin's rule for the model coefficients. Is there is any way to get the corresponding p-values?

I have calculated the confidence intervals using the qbar and t outputs of the pool.scalar function. Can I do it in a different way?

Upvotes: 1

Views: 64

Answers (1)

A. Bohyn
A. Bohyn

Reputation: 187

Following equation 2.34 of section 2.4.2 In Stef van Buuren's book, the pooled estimate Q_bar follows a t-distribution since the total variance T is not known a priori.

The mice::pool functions gives you the total variance T and the pooled estimate Q_bar. Therefore, using R t-distribution's distribution function pt(), you can obtain the p-value as follows:

p <- mice::pool(Q, U)
t <- p$qbar / sqrt(p$t)
pvalue <- 2*pt(q = t, df = p$df, lower.tail = FALSE)

Note that lower.tail=FALSE means we keep the area under the curve on the right, so we need to double it to obtain the p-value for a two-sided t-test.

Upvotes: 0

Related Questions