Association for Public Policy Analysis & Management

Boilerplate language--standardized, repeated blocks of text with minor modifications--appears frequently in environmental regulatory documents, contributing to the ballooning length of many regulatory texts. Boilerplate has competing impacts. On one hand, literature on modularity suggests that boilerplate enhances the efficiency of document preparation, freeing up staff time and resources to focus on other tasks. On the other hand, literature on scientific readability suggests that boilerplate makes documents less comprehensible to the public, which is particularly problematic for documents that are intended to increase the accessibility of government decision-making processes. Despite these tensions, there is little research assessing the use of boilerplate in regulatory texts.

In this research, we evaluate the use of boilerplate in Environmental Impact Statements (EIS) produced under the National Environmental Policy Act (NEPA) for projects built, funded, or approved by a federal agency. EIS documents review the potential impacts of the project, and evaluate how these impacts change under different construction and implementation alternatives. EISs are purely procedural documents – NEPA requires an EIS to identify the least harmful course of action, but it does not require federal agencies to select this alternative. Instead, the rationale behind NEPA is that providing information about environmental impacts will lead to better projects and enhance opportunities for public participation in the relevant decisions. This makes EISs a particularly fruitful case to explore the use of boilerplate, as they embody the tension between public communication and agency efficiency. In addition, courts have repeatedly and explicitly condemned the use of boilerplate in EISs.

To evaluate trends in boilerplate usage, we apply a variety of computational text analysis techniques to a dataset of EIS documents issued nationwide since 2010. We assess two primary research questions: 1) how frequently is boilerplate language used in individual EIS documents?; and 2) how does usage vary by federal agency, project type, project location, and whether a consulting firm was used? In the context of a large EIS study, boilerplate is likely to reoccur in specific sections and segments rather than an entire document. Thus, to measure boilerplate language, we use a combination of supervised and unsupervised text mining techniques to break each document into subunits at an appropriate scale for identifying boilerplate. Pre-trained Natural Language Processing tools are first used to process EIS pdf documents into machine readable text, identify unique sections, and isolate individual paragraphs. We then use three different text similarity measures--Jaccard distance, edit distance, and cosine distance--to quantify paragraph similarity in each document.

By understanding when and where boilerplate language is used, this research is the first step in evaluating the broader effects of boilerplate on the effectiveness, efficiency, and equity of environmental regulatory processes.