Biased and Inefficient

A quadratic form in \(n\) Gaussian variables is \(Z^TAZ\) where \(Z\sim N(0,B)\) is a Normal \(n\)-vector and \(A\) is an \(n\times n\) matrix.

The svylme package does maximum (weighted pairwise) likelihood.

Back when I was a PhD student working on generalisations of GEE, I was interested in ‘sparse’ correlations, defined by ‘most’ small sets of observations being independent.

The announcement There’s now a preprint describing the svylme package and svylme::svy2lme function for fitting linear mixed models to complex samples.

The most important part of this post library(svylme) It exists!

Thanks to Luis Apiolaza on Mastodon, I was looking at some old R mailing list messages.

Someone asked on StackOverflow how to compute the kurtosis and its standard error using the survey package.

One of the problems with object-oriented programming for statistical methods is that inheritance is backwards.

So, I’m working on svylme again, for linear mixed models under complex sampling.

Version 4.2-1 of the survey package has hit CRAN!

Someone asked on CrossValidated about signed rank tests for complex sample data.

Medical training, of necessity, over-represents unusual medical conditions.

I’m at ENAR, and there are talks with asymptotic theory1.

This post is partly because I think the result is interesting and partly to see if anyone will tell me an original reference.

Last week I needed to know how big determinants of correlation matrices could get1, and I couldn’t immediately find a proof on Google, so I worked one out.

Demetri Pananos asked a question on Stack Overflow and then on Twitter about the behaviour of the sandwich package for aggregated count data, and Achim Zeileis pinged me1.

You used to see a lot of the incorrect and misleading description of the Wilcoxon rank-sum and Kruskal-Wallis tests as comparing medians.

The proportional odds model The proportional odds model is a generalisation of the logistic model.

For a while, I’ve been thinking about multiple imputation and its extension to record linkage.

It’s intuitively plausible that pairwise independence of a set of variables implies joint independence.

From time to time, I get asked about estimating effect sizes in the survey package.

As every schoolchild know, you can derive the Student \(t\)-test as a linear regression with a single binary predictor.

As I said last time, there have been quite a lot of bus cancellations in Auckland recently.

A lot of buses are being cancelled in Auckland at the moment.

Back in the early 2000s, when I was working on air pollution epidemiology, I wrote some code to fit polynomial distributed lag models.

Attention Conservation Notice: I’m writing this down because I just spent too long trying to find a citation for it to give a student.

Attention Conservation Notice: NZ alternative history.

As you may have noticed, I’ve been writing software for handling multiple-response data: a package for data manipulation and another one for modelling.

One of the checks in R CMD check is for non-ASCII strings in data objects.

I’ve been working on the rimu package recently, which handles multiple-response categorical data.

Sometimes you want to do inference that combines a real probability survey sample with some other data.

We have a paper coming out in Biometrics describing a reasonably complicated example of validation sampling from electronic health records.

I’m reasonably often asked how to take the value of a character string variable and use it as a variable name in, eg, the survey package.

They’re back.

A survey package user asked me (and StackExchange) how to do tests of independence in contingency tables when there’s a zero cell in the table.

Character string variables have suddenly become much more common in R, with the default stringsAsFactors=FALSE.

Attention Conservation Notice: there’s not actually anything new here A few years ago, in a (mostly positive) review of Proschan & Shaw’s Essentials of Probability Theory for Statisticians, I wrote As some statistics students notice, there’s a bit of a ba…

In ordinary statistics, there are three popular types of model-based tests: score tests, Wald tests, and likelihood ratio tests.

The second paper from Tong Chen’s PhD thesis with me has just been published at Statistics in Medicine.

This was roughly my plenary talk at Herenga Delta, the 13th Southern Hemisphere Conference on the Teaching and Learning of Undergraduate Mathematics and Statistics.

From 2020:, the different ways the term “weights” is used in statistics and when it matters.

As part of adding crossed correlations (and other sparse correlation) to the survey package, I was writing code to test whether a user-supplied adjacency matrix had only 0 and 1 values.

This week, I was prompted to find some old R code xeffect.glm<-function(glm.obj,g1,g2){ if (!

I’ve just been adding score tests for generalised linear models to the survey package.

Ok, this is another attempt at clarifying my thinking about what is and isn’t problematic with ordinal data.

We’re in Covid lockdown again in Auckland, and it’s a weekend, so I happened to be on the Stats stackoverflow site.1 Someone was trying to use R to check MLE calculations for a few papers that propose new parametric models – the “overgeneralised Gamma dis…

When there are problems with public transport it’s much more salient than when things are going well, so I think it’s helpful to have consistent data available on what the problems actually are.

Data is the New Oil is a seriously underrated metaphor: started off as just a curiosity for nerds creates new companies and industries and vastly improves the productivity of some existing ones control over sources or extraction/transport infrastructure m…

The Twitter bus bot, @tuureiti, which I wrote about before, is back!

A question I’ve now answered enough times that it’s worth turning into a blog post: when it is ok and not ok to take a subset of a probability sample?