Migrated from eDJGroupInc.com. Author: Michael Fluhr. Published: 2013-05-02 05:00:10Format, images and links may no longer function correctly. I can’t recall how many times I’ve received demands from opposing counsel for “all” relevant documents in my client’s possession. Indeed, many court opinions support such an entitlement. Yet countless studies (Blair and Maron, TREC, etc.) show that even the best information retrieval technologies and practices fall well short of perfection. Courts have begun to recognize the impossibility of perfection in eDiscovery, though many still seem reluctant to depart too far from an expectation of perfection. But in a freshly released opinion arising in the context of a multi-district litigation, Judge Robert L. Miller, Jr. of the Northern District of Indiana ventures further than other courts, recognizing that parties may satisfy their discovery obligations under the Federal Rules even with productions that fall well short of total recall. Below, I summarize the opinion and explore the implications of imperfection in eDiscovery.
The case is In re Biomet M2a Magnum Hip Implant Products Liability Litigation, No. 12-MD-2391 (N.D. Ind.). The opinion concerned a dispute over the reasonableness of Biomet’s eDiscovery search and retrieval efforts. Biomet began its efforts with 19.5 million documents, which it culled to 3.9 million using full-text search filters. Biomet then proposed to use predictive coding software to locate relevant documents within that set. Analysis of the full-text search filters showed that they captured approximately 60% of the relevant documents in the initial set of 19.5 million (i.e. 60% recall). Assuming that predictive coding would identify approximately 70% of the relevant document within the filtered set of 3.9 million (a reasonable estimate), Biomet’s proposed eDiscovery efforts would result in production of approximately 40% of the relevant documents contained in the initial set of 19.5 million documents (40% recall). Biomet spent approximately $1 million on these efforts, though it claimed that to fully complete the plan would cost between $2 million and $3 million.
Plaintiffs objected on the grounds that full-text searching has somewhat poorer rates of recall than does predictive coding and that the Court should thus order Biomet to apply predictive coding to the initial set of 19.5 million documents (rather than a full-text filtered set of 3.9 million). Biomet responded (correctly) that predictive coding costs significantly more than full-text searching and that considerations of proportionality and reasonableness do not justify the additional expense (added to Biomet’s existing expenditures of over $1 million).
The Court agreed with Biomet. It held that Biomet’s efforts satisfied the proportionality requirements of Federal Rule 26. The Court also offered that if Plaintiffs really wanted Biomet to apply predictive coding tools to the entire set of 19.5 million documents, Plaintiffs could pay for it themselves.
Understandably, many facets of the opinion have already received attention and written analysis. The Court approved the use of predictive coding; the Court quantitatively analyzed the efficacy of Biomet’s full-text filtering; and the Court provided for cost-shifting. All of these merit attention (not to mention praise for Judge Miller). But I find one aspect of the opinion particularly important and worthy of additional discussion.
I will continue this article next week with my thoughts on the outcome of this case.
eDiscoveryJournal Contributor Michael Fluhr*
*The opinions expressed in this commentary are solely those of Michael Fluhr