Fig 1.
Flow chart for the selection of county-utility service areas.
Additional zero observations removed from primary and secondary analysis as described in S1 Table. Raw unprocessed PowerOutage.us data included 22 utilities.
Fig 2.
Mean daily log of SAIDI values for each county-utility service territory (n = 31).
Areas shaded in white were not included in the PowerOutage.us data or were excluded from all analyses. Washington county boundaries were provided by the Washington State Department of Natural Resources [67]. Utility service territories provided by the Oak Ridge National Laboratory were modified based on maps provided on utility websites [46].
Table 1.
Description of medically relevant/major event definitions vs. non-zero outages.
Table 2.
Summary table of ZALN GAMM for SAIDI for the primary analysis.
Fig 3.
Smooth effect on the log-odds of outage absence.
Partial effects from the fitted GAMM model predicting the log-odds of a power outage absence for 23 county-utility areas as a function of function of poverty (%), disability (%), square root of the % of limited English, unemployment (%), minimum temperature (°C), maximum wind (m/s), and precipitation (mm). The shaded areas represent the 95% confidence interval for the partial effects, the solid lines represent the smooth fitting curves of outage absence, and the x-axis represent the measured values of the explanatory variables. Rug marks along the x-axis represent data points from the original dataset (n = 31,714) to indicate the distribution of observations.
Fig 4.
Smooth effect on log of SAIDI in minutes.
Partial effects from the fitted GAMM predicting daily mean log-transformed SAIDI for 23 county-utility areas as a function of poverty (%), disability (%), square root of the % of limited English, unemployment (%), minimum temperature (°C), maximum wind (m/s), precipitation (mm) for the effects of social vulnerability and weather on mean daily log-transformed SAIDI. The shaded areas represent the 95% confidence interval for the partial effects, the solid lines represent the smooth fitting curves of outage absence, and the x-axis represent the measured values of the explanatory variables. Rug marks along the x-axis represent data points from the original dataset (n = 23,597) to indicate the distribution of observations.
Fig 5.
Short-term seasonal effects on log of SAIDI in minutes for county-utility service areas.
Partial effects from the fitted GAMM predicting daily mean log-transformed SAIDI for each study year for 23 county-utilities and 23,604 county-utility days.
Fig 6.
Short-term seasonal effects on log-odds of outage absence for county-utility service areas.
The top panel includes the partial effects of seasonal effects for the county-utility (n = 5) on the presence/absence of outages. The bottom panel includes the partial effects of seasonal effects for the county-utility (n = 13) on the log(SAIDI). Figure includes only county-utility areas with 95% confidence intervals excluding zero.