Spatial Discontinuities in Support for Hydraulic Fracturing: Searching for a “Goldilocks Zone”

Abstract

Recent research suggests that those located closer to energy development are, on average, more supportive of this development. However, case studies in specific locations reveal additional nuance. In a case study of Bakken Shale residents, Junod et al. identified a “Goldilocks Zone” of unconventional oil and gas development (UOGD) acceptance—an area on the periphery of development that is “just right” because residents feel close enough to receive economic benefits but far enough away to avoid negative impacts. We explore whether this Goldilocks Zone extends nationally by combining geocoded public opinion data (N = 23,154) with UOGD locations. Using multilevel regression modeling, we find that respondents located within 115 km of newly active UOGD are more supportive of hydraulic fracturing while those located within 115–305 km are comparatively less supportive. While we do not uncover a national-level Goldilocks Zone, our work highlights innovative approaches for examining spatial relationships in energy development opinion.

Keywords:

• Hydraulic fracturing
• geographic proximity
• public opinion
• unconventional oil and gas development
• multilevel regression modeling

Notes

Acknowledgements

We would like to thank Drillinginfo and University of Texas at Austin Energy Poll for generously providing us with data applied in this research.

Notes

1 We applied the five-category version of the fracking opinion dependent variable in linear multilevel regression analysis and ordinal logistic multilevel regression analysis (Tables S7 and S8, Supplemental Materials). Findings from these models are not demonstrably different from binary logistic multilevel regression results presented in this main text. Linear multilevel regression does not account for the non-continuous, five-category version of the fracking opinion dependent variable, while estimation of ordinal logistic regression models revealed a violation of the proportional odds assumption.

2 We tested a longer time period (two years) between survey administration and first well production and applied this measure in additional analysis (Table S9, Supplemental Materials). Two-year proximity measures have a substantively similar effect compared to one-year proximity measures.

3 We generated a kernel density surface for well locations using three bandwidths: 20 km, 50 km, and 100 km. The density surfaces were generated using the same underlying well location data applied in the proximity measure (see Methods S1, Supplemental Materials).

4 This includes Washington, DC and excludes Hawaii and Alaska.

5 To examine whether patterns were driven by other factors, we tested multiple interactions between proximity and residence in a metropolitan area, percentages of the labor force employed in extractive industries, and respondent’s political affiliation (Table S4, Supplemental Materials). To test patterns within distance thresholds we interacted continuous distance with distance thresholds, and also tested polynomial distance transformations (Table S5, Supplemental Materials).