Technical, design, and cultural aspects of developing general-purpose software for mixed models

Background/Question/Methods

The lme4 package for R is one of the most popular tools for fitting mixed models, an important class of statistical models in ecology. lme4 is actively developed; has a very wide spectrum of users, from naive to sophisticated; and attempts to provide a simple but powerful interface to advanced methods that are not completely understood even by statisticians. As such, it provides a useful case study for software development in ecology. As well as technical and software engineering challenges, the development of the package has presented interesting problems that are more social, political, and cultural.

Results/Conclusions

The most interesting challenges have been:

• calibrating advisory, warning, and error messages so that they give appropriate guidance while allowing flexibility
• providing a two-tiered interface that includes both a relatively easy-to-use system for end users and appropriate hooks for developers
• deciding whether to provide functionality that is popular, but of debatable statistical value
• serving a large and extremely heterogeneous user base