Just like TNT, the Second Circuit sure knows drama. After years of protracted litigation, the Second Circuit finally put an end to an attempt to recuse a judge for knowing too much about eDiscovery and predictive coding. On April 10, 2013, in an incredibly brief order most likely meant to send a message deeper than its two sentences, a Second Circuit Judge denied a request for the recusal of Judge Andrew J. Peck from an ongoing employment discrimination case. According to Judge Jane A. Restani, “Petitioners have not ‘clearly and indisputably demonstrate[d] that [Magistrate Judge Peck] abused [his] discretion’ in denying their court recusal motion… or that the district court erred in overruling their objection to that decision.” The contentious attempts to recuse Judge Peck stemmed from a discovery dispute after Judge Peck ordered the parties to use a method of predictive coding during discovery. Although the parties seemed to agree that predictive coding should be used, they could not agree on the methods of predictive coding that would be implemented. The plaintiffs believed that Judge Peck favored the defendants in his order, and therefore they moved to recuse the judge because of his established history with eDiscovery and more specifically, his history of actively advocating predictive coding. Judge Peck has a long history of participating in eDiscovery conferences and was considered one of the Court’s “experts in e-discovery.” National Day Laborer Organizing Netwrok v. U.S. Immigration and Customs Enforcement Agency, 2012 WL 2878130, 11 (S.D.N.Y. 2012). Judge Peck was even involved in one of the first cases to order the discovery of electronic data. Atlantic-Monopoly, Inc. v. Hasbro, Inc., 958 F.Supp 895 (S.D.N.Y. 1995). Despite the strong undertones of the order’s brevity, the plaintiffs continued to fight this seemingly uphill battle and later filed a cert petition to the Supreme Court. Rather than attacking Judge Peck’s background and connections to the eDiscovery community, the plaintiffs in this case should have instead accepted that judges need to actively participate in conferences and seminars to better understand the technology implicated in eDiscovery. Just as attorneys can no longer ignore the ramifications of eDiscovery, judges too must enhance their knowledge to further develop this complicated area of law and readily adapt it to continually changing technology. Judges should not be punished or accused of bias for engaging in programs geared towards teaching them about technology and its implications on eDiscovery. If this were at all all permitted, judges would be afraid to participate in seminars and review panels, which would stagnate the development of the law, a process that is already far-behind the rapid progress experienced by technology. Jeffrey, a Seton Hall University School of Law graduate (Class of 2014), focused his studies primarily in the area of civil practice but also completed significant coursework concerning the interplay between technology and the legal profession. He was a cum laude graduate of the University of Connecticut in 2011, where he received a B.S. in Business Administration with a concentration in Entrepreneurial Management.
By the time In re Biomet made it in front of a Seventh Circuit Judge for a ruling, 2.5 million documents and attachments were produced to the plaintiffs in this large class action case against Biomet. The plaintiffs wanted the judge to order the discovery of electronically stored information. The plaintiff’s Steering Committee was unhappy with the amount of documents produced and claimed that it should have been almost five times that amount. The plaintiffs challenged the electronic discovery procedure that Biomet had undertaken. Specifically, the plaintiffs wanted the judge to make a ruling that the defendant’s process was “tainted” by their use of keyword culling. The judge disagreed and refused to make such a ruling, which would have thrown Biomet back to almost square one. Biomet went through an extensive process to cultivate documents to produce for the plaintiffs. Biomet first used “electronic search options, then predictive coding, and finally personal review.” The plaintiff’s issue was primarily the first step that defendants irrevocably ruined the rest of their document production from the get-go. To first identify what documents would be relevant the defendant used a “combination of electronic search functions” which included keyword culling. The defendant’s original pool consisted of 19.5 million documents and attachments which the first step narrowed down to 3.9 million (eventually getting to 2.5 million documents). The plaintiffs thought they should have produced 10 million documents and said their keyword searches were the problem. The plaintiffs cited to a New York Law Journal article that said that keyword searches were “only 20 percent” responsive. According to the plaintiffs, Biomet’s approach was flawed because it used the “less accurate” method of keyword search in the beginning instead of predictive coding. They asked for the judge to rule that the defendants had to go back to the first step and use predictive coding with “plaintiffs and defendants jointly entering the ‘find more like this commands’. The judge found that the plaintiff’s journal article and one cited search claiming that Boolean and keyword searches are less effective at producing relevant documents were insufficient in proving that Biomet did not meet its discovery obligations. Instead, the judge found that its procedure did comply with FRCP 26(b) and 34(b)(2). The judge also refused to read into the rules that Biomet had to allow the plaintiffs to sift through possibly privileged documents with them. The judge also found that Biomet fulfilled their federal requirements as proven through their statistical sampling and confidence tests that they ran over their documents. This sampling found that less than 1.34 percent of the documents that weren’t selected would be responsive and that between 1.37 and 2.47 of the original 19.5 were. Biomet’s process singled out 16 percent out of that original. The judge cited heavily to FRCP 26 (b)(2)(C) and said that Plaintiffs’ request did not comport well with proportionality. Biomet had already spent $1.07 million and “will total between 2 million and 3.25 million.” Were Biomet to go back to their original bank of ESI, it would cost them in the low seven-figures. The judge said that it would not make Biomet do that just to test the plaintiffs’ theory that more responsive documents would be found through predictive coding instead of keyword searches.
Whoever thinks that the legal world does not involve math is proven wrong through the Special Master’s analysis in Dornoch Holdings Int’l, LLC v. Conagra Foods, Lamb Weston, Inc. The heart of the opinion involves a percentage breakdown of search terms and their correlation of precision in regard to privileged documents. In Dornoch, the defendants objected to the privilege log of documents for three reasons: 1) the documents on the privilege log, except for communications between the plaintiffs and their outside litigation counsel dated after March 22, 2010, have not been established by the plaintiffs to be privileged; 2) The privilege log was created using overly broad search terms and has not been substantively reviewed, thus, the log contains numerous non-privileged documents; and 3) Non-correspondence documents listed on the privilege log are not privileged. In response to this objection, the court allowed the Special Master to make a recommendation on these objections, specifically allowing the Special Master to review “a statistically significant number of randomly selected documents to confirm the accuracy of the screening method.” The privilege documents log was assembled using search terms created and limited by plaintiff’s counsels and an eDiscovery technology consulting firm. And so, the Special Master did as the Court requested and took a sampling from the log to determine the effectiveness of the screen’s search terms. The consulting firm determined that “1,740 documents would need to be human reviewed” to determine whether the log was effectively precise. The Special Master decided to review 1,813 documents just to ensure it was an effective review. After explaining that Idaho law regarding attorney-client privilege and work product doctrine apply, the Special Master reviewed the documents and determined that 1,249 were not privileged documents and 564 were privileged. The Special Master also went into much detail about the effectiveness of the specific search terms that were used. Specifically, the Special Master determined that 73 percent of the search terms were highly correlated to actual privileged documents. Additionally, the Special Master determined that “those terms which identified a correlation with actual privilege of 59 percent or greater, showed a strong correlation with privilege.” Once the Special Master completed this analysis, the Special Master recommended that the documents that fall below that 59 percent correlation should be released and not kept private. Then, the plaintiffs could also decide to conduct another review of the remaining privileged documents to figure out if more should be released. Finally, the Special Master noted that it does not matter whether documents are listed as “correspondence” or “non-correspondence” for them to be determined to be privilege or not. These documents should be reviewed just as the others. Overall, the Special Master recommended that the court sustain the first objection, and overrule the third objection. As to the second objection, the court recommended the following: “(1) Concur with the selection of a 59% or greater correlation of search term precision for a document to remain withheld as privileged; (2) Allow Defendants the opportunity to further challenge the assertion of privilege above that 59% threshold, if they so choose, by requesting that the Special Master conduct a further targeted review for privilege and release any non-privileged documents discovered. The Defendants will be responsible for cost of this further analysis, if requested; (3) Release the documents associated with the less precise terms that fall beneath the 59% correlation threshold and remove them from the privilege log; (4) Prior to that release, allow Plaintiffs the opportunity to conduct a privilege review of all or a portion of the population to be released and create a supplemental privilege log. The Plaintiffs will be responsible for cost of this further analysis, if Plaintiffs chose to conduct it.”
collapse footer
expand footer
