Missing Data, Speculative Reading

Abstract

In this article we use an approach we term “speculative reading” to explore gaps in Sylvia Beach’s lending library records and the *Shakespeare and Company Project* datasets. We recast the problem of missing data as an opportunity and use a combination of time series forecasting, evolutionary models, and recommendation systems to estimate the extent of missing information and speculatively fill in some gaps. We conclude that the datasets include ninety-three percent of membership activity, ninety-six percent of members, and sixty-four percent to seventy-six percent of the books despite only including twenty-six percent of the borrowing activity. We then treat Ernest Hemingway as a test case for speculative reading: based on Hemingway’s known borrowing and all documented borrowing activity, we generate a list of books he might have borrowed during the years his borrowing is not documented; we then verify and interpret our list against the substantial scholarly record of the books he read and owned.