class: center, middle, title-slide # Significance, Thresholds<br/>and Decision-Making ## Joe Thorley ### 2019-05-09 .pull-right[  ] --- ## Uncertainty Our understanding of nature is incomplete. --  .footnote[ © NASA ] -- And always will be. --- ## Decisions Yet we must act. -- background-image: url(19-significance_files/blindfolded-158204_640.png) ---
---
--- ## Posterior Probability Distribution <img src="19-significance_files/figure-html/unnamed-chunk-3-1.svg" style="display: block; margin: auto;" /> -- The uncertainty is fully described by the posterior probability distribution ---
--- ## Posterior Probability Distribution <img src="19-significance_files/figure-html/unnamed-chunk-5-1.svg" style="display: block; margin: auto;" /> The posterior is typically summarized by a 95% confidence/credible interval (CI) -- which is typically summarized in terms of its significance (is/isn't) -- in this case it **isn't**. --- ## Significance <img src="19-significance_files/figure-html/unnamed-chunk-6-1.svg" style="display: block; margin: auto;" /> -- Sample size determines significance. -- Impact assessment equates non-significance with no-effect. -- Violates precautionary principle. -- Data collection is disincentivized. ---
-- > 'uncertainty laundering' Gelman (2016) --- ## Power Analysis Calculate the expected uncertainty in the estimated effect. <img src="19-significance_files/figure-html/unnamed-chunk-8-1.svg" style="display: block; margin: auto;" /> -- Set a minimum sample size. -- Implicit threshold. ---
--- ## Prior Information Incorporate existing knowledge that impact likely to be negative. <img src="19-significance_files/figure-html/unnamed-chunk-10-1.svg" style="display: block; margin: auto;" /> --- ## Prior Information Incorporate existing knowledge that impact likely to be negative. <img src="19-significance_files/figure-html/unnamed-chunk-11-1.svg" style="display: block; margin: auto;" /> -- Incentivizes informative data collection. ---
--- ## Precautionary Principle Test whether under a threshold. -- <img src="19-significance_files/figure-html/unnamed-chunk-13-1.svg" style="display: block; margin: auto;" /> ---
--- ## Statistical Decision Theory Choose option that maximizes expected net benefit given the uncertainty. -- Requires loss function (challenging to develop) -- but criteria are explicit. -- and decisions optimal. --- <img src="19-significance_files/figure-html/unnamed-chunk-15-1.svg" style="display: block; margin: auto;" /> Expected loss = 0.73 - 0.4 = 0.33 <img src="19-significance_files/figure-html/unnamed-chunk-16-1.svg" style="display: block; margin: auto;" /> Expected loss = 0.07 - 0.4 = -0.33 ---
---  --- ## References Amrhein, V., Greenland, S., and McShane, B. 2019. Scientists rise up against statistical significance. Nature 567(7748): 305–307. doi:10.1038/d41586-019-00857-9. Diaz, S. 2019. Summary for policymakers of the global assessment report on biodiversity and ecosystem services of the Intergovernmental Science-Policy Platform on Biodiversity and Ecosystem Services. Gelman, A. 2016. The problems with p-values are not just with p-values. The American Statistician 70: 10. Williams, P.J., and Hooten, M.B. 2016. Combining statistical inference and decisions in ecology. Ecological Applications 26(6): 1930–1942. doi:10.1890/15-1593.1. ---- .footnote[ Thorley 2019. Significance, Thresholds and Decision-Making. A presentation at Regulated Rivers II: Science, Restoration, and Management of Altered Riverine Environments. Columbia Mountains Institute. Nelson, BC. <https://www.joethorley.io/slides/19-significance#1>. ]