Introduction to agridat

Kevin Wright

For a quick overview of the datasets available in the package, type ?agridat at the R prompt.

Comments on the package purpose

When this project was first begun in the early 2000s, electronic versions of agricultural datasets were hard to find. Since then, there has been a revolution in the availability of datasets related to agriculture. See the vignette which describes some data sources.

Box (1957) said, “I had hoped that we had seen the end of the obscene tribal habit practiced by statisticians of continually exhuming and massaging dead data sets after their purpose in life has long since been forgotten and there was no possibility of doing anything useful as a result of this treatment.”

Massaging these dead data sets will not lead to any of the genetics being released for commercial use. The value of this package is: 1. Validating published analyses. 2. Providing data for testing new analysis methods. 3. Illustrating (and validating) the use of R packages.

White and van Evert (2008) present some guidelines for publication of data.

Some of the examples use the asreml package since it is the only R tool for fitting mixed models with complex variance structures to large datasets, and the best option for modelling AR1xAR1 residual variance structures. Commercial use of asreml requires a license from VSN. (Use a search engine to find the latest version).

Comments on the package structure

Many of these datasets appear in electronic form here for the first time.

A tremendous amount of effort has been given to the curating process of identifying datasets, extracting the data from source materials, checking data values, and documenting the data. In effect, to make the data somewhat ‘computable’ (Wolfram 2017).

The original sources for these data use different words to refer to genotypes including accession, breed, cultivar, genotype, hybrid, line, progeny, stock, type, and variety. For consistency, these datasets mostly use gen (genotype).

Also for consistency, row and col are usually used for the field coordinates.

In dataframes, block, rep, and similar terms are almost always coded like B1, B2, B3 instead of 1, 2, 3. This causes R to treat the data as a factor instead of a numeric covariate (which is a good thing).

Almost all of the data are presented as ‘tidy’ dataframes with observations in rows and variables in columns.

Although using data() is not necessary to access the data files, the example sections do include the use of data() because devtools::run_examples() needs it.


G. E. P. Box (1957). Integration of Techniques in Process Development, Transactions of the American Society for Quality Control.

J. White and Frits van Evert. (2008). Publishing Agronomic Data. Agronomy Journal, 100, 1396-1400.

Stephen Wolfram (2017). Launching the Wolfram Data Repository: Data Publishing that Really Works.