inst/extdata/README_berchtold_raftery.md

In markovchain: Easy Handling Discrete Time Markov Chains

Data from Berchtold and Raftery (2002)

Two categorical series used in Tables 2 and 3 of

Berchtold, A. and Raftery, A. E. (2002). The Mixture Transition Distribution Model for High-Order Markov Chains and Non-Gaussian Time Series. Statistical Science, 17(3), 328-356.

They are used by the package's unit tests and by the higher_order_markov_chains and an_introduction_to_markovchain_package vignettes to check higherOrderLogLik(), assessIndependence() and fitMTD() against the published log-likelihoods and BIC values.

The files were kindly provided by the authors (Adrian E. Raftery and Andre Berchtold) for use in this package; they stated that the data are not copyright protected.

| file | rows | coding | |------|------|--------| | koeberg_wind.csv | 744 | state, hourly wind direction at Koeberg (South Africa) recoded into 4 directions, states 1-4. The original series is from MacDonald and Zucchini (1997). | | epileptic_seizures.csv | 204 | seizure, daily binary series: 0 = no epileptic seizure, 1 = at least one seizure. |

The authors' original epilepsy file is coded 1/2; it was recoded to 0/1 here (1 -> 0, 2 -> 1), as they indicated.

Convention of the published tables

Berchtold and Raftery condition the log-likelihoods on the first 14 observations, so that every model (independence, Markov chains of order up to 3, MTD models) is evaluated on the same n - 14 observations; the BIC uses n - 14 as sample size, and the number of parameters counts only those that are not forced to zero. Reproducing the tables therefore requires start = 15 in higherOrderLogLik() and fitMTD().

MacDonald, I. L. and Zucchini, W. (1997). Hidden Markov and Other Models for Discrete-valued Time Series. Chapman & Hall.

markovchain documentation built on Oct. 10, 2026, 9:07 a.m.