Both Quirin's and Nunamaker's titles were excellent references. I favor Nunamaker's pure statistics approach, however, unless prose is necessary to explain any new or complicated concepts (such as Quirin did with his speed points or speed figures), so that readers know how to calculate and apply them; or to clarify the source of data and explain briefly which races in a horse's record a statistic represents, or which data source a statistic is drawn from (as Nunamaker did in the brief forward to
Modern Impact Values).
I would especially favor statistics if inclusion of any prose would somehow limit or cut into the comprehensiveness or depth of the statistics presented. (For example, Nunamaker presented a more thorough breakdown of his statistics than Quirin did, by age, running surface, sex, distance, and maidens vs. previous winners, which, to me, made his data more useful.) Also, Quirin's findings were based on much smaller sample sizes (even though he applied statistical tests to project whether particular positive or negative findings constituted an enduring trend, rather than just a random fluctuation). I preferred Nuamaker's greater size and scope (even though I imagine it required much more number-crunching).
Finally, Nunamaker indicated that he somehow weighted his impact values to reflect the field size in which each horse ran. I recalculated all his impact values, doing them Quirin's way (by dividing percentage of winners in the total sample by percentage of all horses in the sample, irrespective of individual race field sizes), and I don't recall that there was too big a difference in the results. To me, Quirin's method is simpler and adequate.
Just one player's opinion (and apparently literally so, judging from the poll responses to this point).