Writing

[goldma99]

While results are being generated, focus on a following

• [x] Abstract revisions
• [x] Introduction revisions
• [x] #38
• [x] Background
  • [x] Beaver ecology
  • [x] Scotland agriculture industry (special attention to profitability over time and why farms might be vulnerable to beavers
  • [x] Beavers in Scotland (hunting, extinction, return)
• [x] Figures and tables
  • [x] Explanatory figures
  • [x] Pre-trends figure
  • [x] Results table
  • [x] Two panel results table: A=full sample, B=only on-river cells
  • [x] 3 panel results (with full river/landscape sample): A=all cohorts, B=only 2012 and 2017, C=only 2012
• [x] Data
  • [x] Beaver survey (special attention to survey collection methodology, note that its not technically comprehensive, that there may be bias toward areas with reports, potential false negatives in my data. GPS issue in 2017 that forces me to only use changes in extensive margin)
  • [x] Land Cover Maps
  • [x] Hydrometry
  • [x] Elevation, soil, weather(?)
• [x] Main results
  • [x] Show ag results
  • [x] Show soil subsamples
  • [x] Show jackknife
  • [x] Mention river mechanism (point toward appendix)
  • [x] Write table titles
  • [x] Write figure titles
  • [x] Consistent language: during first mention, note that "grid cell" and "landscape patch" are synonymous
• [x] Conclusion
• [ ] Fill in missing citations

• [ ] Mechanisms

  • [x] #44

  • [ ] Robustness check (not main spec): Consider in one spec: entering a interaction of the linear time trend with soil ag suitability (just the interaction term, not the interaction + the time trend alone)

  • [x] Interact post-period dummy with soil class categorical variable (allows different soil categories to have different trends--narrows residual variation to switchers within each soil class)

  • [x] If effect goes away, then maybe AC is just getting more ag land over time and also more beavers over time.

  • Answer: I ran the following spec and the effect only increased (0.037)

    reghdfe ag_share beaver_d $weather_controls t_overall#lccd_mj_dom_e if on_river == 1, ///
                                    absorb(river_id) ///
                                    cluster(river_id) nocons

  • [ ] Run main spec dropping g=1 ("treated" in 2012)

  • [x] Run spaghetti plots of ag_land_share (for all years, disaggregated), colored by control and treatment groups

  • [x] Jackknife leave-one-out exercise (drop grid cells, one at a time), then plot coefficient distribution (and plot preferred spec coefficient)