Comparative timings

• tables (PDF, 184kB) - overview of package capabilities for some of the common data types the user is likely to use
• datasets used in reviews - description of the data sets used

In the tables next to the estimates (in brackets) is given the standard error of the estimate (one number) or a 95% confidence interval (two numbers). Where packages produce standard deviations rather than variances these have been (approximately) converted, as have corresponding standard errors or confidence intervals and correlations converted to covariances.

The reviews were carried out on different machines running under various operating systems. We have computed approximate timings using conversions based upon running each software package on a standard machine with one model. The smallest value is less than 1 second (< 1s) and the longest is more than 10 hours (> 10h). This machine is a 433 Mhz Pentium II running windows 2000 and timings refer to that. Because of the approximate nature of the timings we present these as grouped values so that they can be utilised as a rough guide by users. In the following models the description of parameters is given in the detailed reviews.

1. Normal linear model: The models are a 2-level variance components (random intercepts) model, with and without random coefficients and complex level 1 variation, and a 3-level variance components model. REML estimates are given, where these are available. The exam data set is used for the 2-level models and the Chemistry data set for the 3-level model.
2. Binary response model: The model is a 2-level random intercept model. Maximum likelihood and quasi-likelihood estimates are given. The Bangladesh fertility dataset is used.
3. Ordered category model: The model is a 2-level random intercept model. Maximum likelihood and quasi-likelihood estimates are given. The social attitudes dataset is used.
4. Cross-classified model: The model is a variance components model. The primary/secondary school dataset is used