Review of a text that offers a hands-on perspective on an eclectic mix of statistics and computing.

“

Effective data analysis requires a mix of computing skills, an understanding of statistical theory as it relates to the task in hand, use of all available insights on the processes that generated the data, and careful and informed critique. Critical evaluative skills are required both for taking account of relevant work that others (including other team members) may have done, and for review of one's own work. This is the mix that was required for effective data analysis fifty years ago, and it is the mix that is required now. What has changed, with the extent of change dramatically different between different areas of data analysis, are the individual components of the mix. The introduction draws attention to the extent to which, in this century, "data are arriving by firehose and without design", and often with very limited information content relative to questions that may be of interest. Surely this makes careful and critical evaluation more important than ever, calling on all available sources of insight, to do a really effective job. I am then concerned when the authors claim that while "statistical science overlaps with data science with respect to the objective of extracting information from data, the statistical aspects of the problem pale in comparison with the algorithmic and computational considerations." If they have in mind algorithmic and computational considerations that will assist in getting a really good handle on what the data may have to say, well, that is indeed a large component of what is required. Think tabular and graphical summary, model fitting, model diagnostics, the many uses of simulation, and graphical summary of results. The text (Part I especially) is too overly focused on the matrix manipulation that underly multiple regression. Yes, the linear algebra conversation, whether implemented using Python or R or other code, is an important conversation. But it needs to more than a demonstration of the matrix operations required when there is no strong dependence between explanatory variables, and it needs a strong focus on the use of tools for model evaluation and criticism. This text never gets much further than the matrix algebra required for regression in a standard, fairly straightforward case, and pays inadequate attention to model diagnostics.

”