Wildre-mitigating power shut-offs promote household-level adaptation but not climate policy support

Unmitigated climate change threatens to disrupt energy systems, for example through weather- and wildﬁre-induced electricity shortages. Public responses to these energy crises have the potential to shape decarbonization trajectories. Here, we estimate the attitudinal and behavioral eﬀects of Californian power shut-oﬀs in 2019, intended to reduce wildﬁre ignition risks. We use a geographically targeted survey to compare residents living within outage zones to matched residents in similar neighborhoods who retained their electricity. Outage experience increased respondent intentions to purchase gas or diesel generators and home battery systems, but reduced intentions to purchase electric vehicles. Respondents blamed outages on their utility, not local, state, or federal governments. However, outages did not change climate policy preferences, including willingness-to-pay for either wildﬁre or climate-mitigating reforms. Our ﬁndings show that, in reaction to some climate-linked disruptions, individuals may undertake adaptive responses that, collectively, could exacerbate future climate risks.

tomers across 35 counties. 1 Another 177,000 customers were de-energized during a second event 23 between October 23 and October 25, 2 followed by two successive outage events beginning on 24 October 26 and 29 that impacted another 941,000 customers. 3

25
Here, we report the results from a new, high spatial resolution survey of Californians fielded 26 in the immediate aftermath of these widespread outage events. Unlike traditional surveys that 27 rarely achieve geographic resolution below the ZIP code, we use a mail-to-web recruitment 28 strategy that allows fine-grained spatial control (see Methods for details). Briefly, we used the 29 spatial boundaries released by PG&E to generate a sample of addresses subject to at least one 30 outage during October 2019, oversamples of addresses within 1km inside or outside the outage 31 boundaries, and targeted samples of non-outage addresses that were otherwise similar to outage 32 zone addresses. We visualize this sampling frame as Figure 1. All sampled addresses were mailed 33 a letter inviting resident participation in a web-based survey in the second week of November, 34 and then sent a postcard reminder in the first week of December. In total, we received complete 35 survey responses from 890 Californian households (see SI Figure A1 for map of respondents' 36 addresses).
The utility also de-energized 11,300   We can also exploit spatial variation in the distribution of PG&E power outages to evaluate how 39 exposure to energy infrastructure disruption shaped the public's adaptive behaviors and climate 40 attitudes. In general, the boundaries of PSPSs are a function of local transmission networks 41 that remain opaque to most residents. Because transmission networks play a negligible role in 42 structuring where people choose to live, we can estimate the effect of outage exposure on public 43 attitudes by matching respondents within outage zones with otherwise similar respondents just 44 outside outage zones. Overall, we find that outage experience increased respondent intentions to 45 purchase gas or diesel generators and home battery systems, but reduced intentions to purchase 46 electric vehicles. At the same time, outages did not change climate policy preferences, including willingness-to-pay for either wildfire or climate-mitigating reforms. Broadly, our findings sug-48 gest that, in reaction to some climate-linked disruptions, individuals may undertake adaptive 49 responses that, collectively, can exacerbate future climate risks.  an optional question asking how much money they spent on preparations. The average reported 69 amount was $327, with responses ranging from a low of $0 to a high of $5000.
In sum, respondents self-reported that the PSPSs had psychological and economic effects 71 on their households. We now evaluate whether these experiences shaped respondent's behav-72 ioral intentions and climate attitudes. Simple comparisons between individuals exposed and not 73 exposed to outages would likely produce biased estimates; respondents "treated" with outages 74 may differ systematically from non-exposed respondents, including as a result of differences in 75 neighborhood characteristics. 4 Accordingly, we combine our spatially targeted sampling with 76 matching algorithms to match treated respondents with similar unexposed households to esti-77 mate the causal effect of outage exposure (see Methods for details).

78
The effects of outage experience on behavioral intentions 79 We first consider behavioral intentions related to adaptation. Here, we examine whether re-80 spondents planned to take the following actions over the subsequent year: 5 1) change home 81 landscaping to reduce wildfire risk, 2) upgrade home building materials to reduce wildfire risk,

82
3) install a home battery system, 4) install gas or diesel backup generation, 5) move, 6) purchase 83 additional food and water supplies to prepare for future shutoffs, 7) and install solar panels. 6 We 84 also asked respondents whether they thought the next car they purchased would be an electric 85 vehicle (EV) which is, instead, a mitigation behavior that could help reduce the risk of future 86 climate-related hazards. The perceived benefits of EVs might also be affected by reliability of 87 electric service. In the outage-exposed area, 50 percent of respondents reported plans to pur-88 chase additional food and water, 24 percent reported plans to install backup generation, and 89 another 19 percent reported plans to change home landscaping. On the other hand, just 4 per-90 cent reported plans to upgrade home building materials, 9 percent reported plans to install a 91 home battery system, and 7 percent reported plans to install solar panels.

92
When we compare average household-level adaptation outcomes between matched respon-93 dents inside and outside outage areas, we find that outage exposure shaped certain adaptation 94 outcomes, but not others. 7 As demonstrated by Figure 3, exposure to an outage had the strongest 95 effect on respondents' plans for installing a backup gas or diesel generator; individuals exposed 96 4 While outage boundaries may be exogenous, topographical differences still create differences between neighborhood types, property values and other characteristics within 1km of outage boundaries. 5 We excluded respondents who had already taken the activities prior to outage onset. 6 We recognize that the ability to take these actions can depend on home ownership and income. Since the great majority of the sample owned their homes (86 percent), we were unable to estimate heterogeneous effects by home ownership. We also do not find statistically significant differences in adaptation behaviors by income (see SI Section 3). 7 We present differences in means in the matched sample in the main text, with estimates from covariateadjusted OLS regression in the SI Section 2.  Figure presents proportion of matched respondents in outage and non-outage zones stating their intention to adopt a given behavior. Bars are 95 percent confidence intervals. Stars represent significance of difference-in-means between outage and non-outage sample for given behavior; **p<.05 ; *p<.1.
to outages were 16 percentage points [SE=.03, p<.01] more likely to plan generator installa-97 tion. Outage-exposed respondents were also more likely to say they planned to install a home 98 battery system, but only by 4 percentage points [SE=.02, p<.1]. In addition, outage-exposed 99 respondents were 7 percentage points more likely to report that they planned to change their 100 home landscaping to reduce wildfire risk [SE=.04, p<.1]. Finally, outage-exposed respondents 101 were 7 percentage points less likely to report that they planned to purchase an EV as their next 102 car [SE=.04, p<.05]. We do not find statistically significant effects for other household-level 103 adaptations including building upgrades, plans for rooftop solar installations, plans to move, or 104 preparing for future outages by buying additional food and water. On balance, we find that 105 outage-exposed respondents tended to focus on their individual-level adaptive needs. Collec-106 tively, at least some of these behaviors (an increase in fossil fuel generator purchases and a 107 decrease in EV purchases) might inadvertently exacerbate the climate risks that contribute to 108 power outage events. Respondent openness to installing home battery systems suggests one 109 potential countervailing measure that might support climate change mitigation. by whether respondents accept that global warming is caused mostly by human activities. We 112 generally do not find major differences in the effects of outages conditional whether respondents 113 accept climate science. The one exception is with respect to electric vehicle uptake, where we 114 find that the negative effect of outage exposure on EV uptake is concentrated among those who 115 deny anthropogenic global warming. We do not find significant differences in adaptive responses 116 to outages by partisan identification (see SI Section 3).

117
The effects of outage experience on utility and government trust 118 Our second set of outcomes concern respondents' attitudes with respect to electric utilities and 119 government officials. Since outage decisions were made by electric utilities (principally, in this 120 case, PG&E), we would suspect that being exposed to outages might affect respondents' attitudes 121 towards their electricity providers. We measured respondents' trust in their electric utility, the 122 degree to which respondents held their utility responsible for power shut-offs, whether they held 123 PG&E liable for damages from their equipment, and whether they thought PG&E's corporate 124 governance should be restructured as part of its bankruptcy proceeding. Overall, respondents 125 held negative attitudes toward their utility provider. The average level of trust (across outage-126 exposed and non-outage areas) was "somewhat," more than half of respondents felt that PG&E 127 was "completely" responsible for the shut-offs, and 80 percent agreed that PG&E is liable for 128 wildfire damage caused by their equipment. Just 23 percent of respondents felt that PG&E 129 should continue to operate as a privately-owned utility.

130
These attitudes were amplified by outage exposure. As shown in Figure 5, outage-exposed 131 individuals reported statistically significantly lower levels of trust towards their electric provider 132 than individuals in the control group. 43 percent of outage-exposed respondents reported they 133 completely distrusted their utility, compared to 29 percent in the non-outage-exposed area 134 [SE=.04, p<.01]. They also were more likely, by nearly half a standard deviation, to hold their 135 electric utility responsible for causing the planned power shut-offs. 70 percent of outage-exposed 136 respondents reported that the utility was completely responsible, compared to 58 percent else-137 where [SE=.04, p<.01]. However, we do not find that outage exposure was causally associated 138 with respondents agreeing that utilities should be liable for the damage from wildfires caused by 139 their equipment, nor with respondents advocating for a major restructuring in PG&E's corpo-140 rate governance. These latter results may stem from limited variation in the outcome measure 141 (even in non-outage-exposed areas, 79 percent of respondents reported holding PG&E liable, 142 and 77 percent advocated for a major restructuring).

143
The strong effects of outage exposure on electric utility attitudes contrasted with minimal 144 overall effects on attitudes towards politicians. In Figure 6, we do not find evidence that exposure 145 affected overall attitudes towards former President Trump, California Governor Gavin Newsom, 146 or local politicians. However, when we split the sample by partisan identification, we find 147 some evidence that outage exposure affected politician approval among political Independents.    associated with increased concern about global warming. 8

158
In addition to evaluating the degree to which outage exposure affected behaviors and atti-159 tudes, we also leveraged the survey to evaluate respondents' willingness to pay (both financially 160 and in terms of days without power) to reduce fire risk and make the electricity system more 161 stable in California (see Methods for details). First, we estimated that the median respondent 162 was willing to live without electricity for 6.7 days a year to reduce fire risk-6.6 days in the outage 163 area and 6.9 days outside it (no statistically significant difference). We also estimate that the 164 median respondent would be willing to pay a surcharge of just $4.19 per month to avoid future 165 planned power shutoffs-$2.19 in the outage area and $7.89 outside. Again, this difference was 166 not statistically significant due to large standard errors in the willingness to pay analysis. This  the spatial outage boundary (SI Section 5). One exception is the finding that outage exposure 181 reduced intention to purchase an EV, which weakens in some alternative specifications.

182
Broadly, public responses to these power outages reflected households' short-term and prox-183 imate needs-maintaining power and reducing fire risk-rather efforts to climate change, a sys-184 temic but indirect driver of the energy system disruption. Moreover, outage-exposed respondents 185 tended to blame their utility, who made the proximate decision to implement the outages, rather 186 than the politicians who could potentially be held accountable for the policies that may reduce 187 climate change risk through mitigation and adaptation. Our findings trouble assumptions that 188 individuals will change their attitude and behaviors if simply informed about the ways that 189 climate change will personally affect them or if they experience a climate-related hazard event, 190 particularly when -as was the case with the 2019 Californian outages -climate change was not 191 portrayed as a major event driver. Efforts to decarbonize our energy systems cannot assume 192 that all climate-linked disruptions will mobilize the public in support of clean energy reforms.

194
Our data collection protocol began with creating a spatially disaggregated sampling frame that 195 allowed us to target individuals who experienced at least one PSPS as well as groups of otherwise 196 similar residents. During the Fall 2019 PSPS events, we collected spatial polygon files publicly 197 shared online by PG&E for each successive shut-off event. We intersected all outage polygons 198 to define the spatial extent of Californians who were projected to experience one or more PSPS 199 events in the PG&E service area during October and November 2019. We also recorded the 200 number of overlapping projected outages experienced in each part of the service area. 201 We then defined a series of additional spatial zones using buffering methods. First, we 202 defined a spatial zone containing all areas within California located between 0 and 1 km inside 203 the projected outage zone boundaries. Second, we defined a spatial zone containing all areas 204 located between 0 and 1 km outside the outage zone boundaries. Third, we defined a spatial  illustrates these different spatial zones that structure our survey sampling frame.

209
Using the WorldPop gridded 100 meter population dataset as a probability surface, we gen- address (if available) for each point and a label indicating whether the address was a "premise" 221 (Google's label for a dwelling unit). We then subset reverse geocoded points to only those with 222 street addresses identified as premises and removed duplicates. Finally, we randomly subset 223 3000 addresses in each zone, except for the full outage zone, where we selected 6000 addresses 224 to sample. In Figure 1 we also visualize local-scale sampling points in the East Bay Area. 225 We also generated a list of control addresses in Southern California that were as closely percent who drive alone to work; percent who use public transit; and percent who work in-state. 246 We then randomly sampled, with replacement, from these census tracts, using the entropy 247 balance weights as sampling weights. This produced a list of census tracts. We then randomly 248 sampled points within each census tract in this list, and reverse geocoded these points as before. 249 We included the first 3000 addresses identified as premises.  On November 14, 2019, we mailed a customized letter to each of these 18000 addresses, 257 inviting one resident from each household to participate in an online survey on California's 258 electricity system (see SI Figure A14 for example recruitment letter). Each letter contained 259 a customized URL so that we could identify the spatial location for every survey response.

260
Respondents who completed our survey received a $5 digital gift card by email that they could 261 redeem at dozens of different online retailers, or that they could donate to a charity of their choice.

262
As a result of our initial letter, we received 565 complete survey responses. On December 3rd, we 263 sent a follow-up letter to all individuals who had not completed the survey, again inviting them 264 to participate. This generated an additional 325 survey responses. In total, we received 890 265 complete response, a 4.94% response rate. In Table 1  outage, a majority were without power for three or more days.

272
Respondents living in an outage zone differed systematically from respondents who were not 273 exposed to outages, as demonstrated by particular, those exposed to outages were more likely to identify as Democrats, were more liberal,

277
were more likely to identify as female, and were older. As a result, we should suspect underlying 278 differences in attitudes and behaviors when making naive, direct comparisons between these 279 groups.

280
To address these possible underlying variations between treated untreated groups, we used a and Sekhon 2012) via the Matchit package in R to identify a set of individuals that were not 284 exposed to outages that are otherwise comparable to the individuals exposed to outages. 9 In this 285 way, our spatially resolved sampling helps us to identify high quality likely matches for treated 286 respondents; likewise, the quasi-arbitrary nature of outage boundaries reduces somewhat the risk 287 of persistent unobserved confounders. The matching algorithm identified 678 respondents (of 288 890 in the full sample) for whom we were able to achieve balance on key covariates. 485 resided 289 in areas that spatial data provided by PG&E indicate were exposed to outages, while 193 resided 290 in areas that were, according to the PG&E data, unaffected. This is reflected in survey responses 291 to questions about respondents' experience of power outages. In total, 66 percent of respondents 292 9 An alternative approach is to compare individuals on either side of the boundary between outage-exposed and non-outage areas through a geographic regression discontinuity design (Keele and Titiunik 2015). If the boundary is randomly placed, we would expect, within a small geographic window around the boundary, no systematic differences between treatment and control groups. The problem with this approach in our case is imprecision in the spatial data specifying the outage-exposed areas. Only 25 percent of respondents living between 0 and 1000 meters on the inside of an outage zone reported exposure to planned outages-while 12 percent of respondents living between 0 and 1000 meters on the outside of an outage zone reported exposure. Given this imprecision, the matching design provides much greater leverage for estimating the effect of outages exposure. Notes: Ideology was measured using a standard 7-point Likert scale (1 is most conservative, 7 most liberal). Education was measured on a 5-point scale (less than high school, high school diploma or GED, some college, associates degree, bachelors degree of higher). Income was measured using a 4-point scale (less than $40,000, $40,000 to $100,000, $100,000 to $250,000, over $250,000). Smoke level is 4-point measure of degree to which smoke has made air quality in respondents' community worse since beginning of October, 2019.
in the matched treatment group reported that they experienced a planned outage, compared to 293 just 11 percent of respondents in the matched control group. 10 294 Respondents are indexed by i. T i denotes outage exposure and X i is a matrix of demographic 301 covariates measured the respondent level. α is an intercept, and ε i represents standard errors. 11

302
Discussion of covariates included, and estimates from covariate-adjusted models, are provided in 303 SI Section 2. Throughout, all statistical tests are two-side.

304
In addition to using the survey for causal inference, we also leveraged the survey to gain 305 insights about the public's understanding of reasons for the planned electricity outages. For 306 respondents who had reported experiencing a shutoff, we asked: "In a few words, why do you 307 think your electricity was shut off?" For respondents who did not report that their own electricity 308 was shut off, but that electricity of other homes in their communities was shut off, we asked: "In uncertainty as to what caused the shutoffs. We discuss the proportion of responses that fell into 317 each category in the main text.

318
In the main text, we also report median respondent willingness to pay (both financially and 319 in terms of days without power) to reduce fire risk and make the electricity system more stable.

320
To estimate willingness-to-live without electricity to reduce fire risk, asked respondents: "Would 321 you be willing to live without electricity for X days each year to reduce the risk of wildfires in 322 California?" We randomly assigned X from among 1, 2, 3, 4, 5, 7, 10, 14, and 21, and used the 323 function sbchoice from the package DCchoice in R to compute median willingness-to-pay. We 324 conducted similar analysis for the other willingness-to-pay items. To estimate willingness-to-pay 325 a surcharge to reduce future planned power shutoffs, we asked: "Would you be willing to pay 326 a surcharge of $X every month on your electricity bill to avoid future planned power shutoffs?" 327 We randomly assigned X from among 1, 2, 5, 7.5, 10, 15, 20, 30, 40, 50, 75, 100, 150, and 250. 328 To estimate willingness-to-pay to bury power lines underground, we asked: "How much would 329 you support burying power lines in California if it cost you $X more per month on your utility 330 bill for the next 10 years?" 12 For this question, we randomly assigned X from among 1, 2, 5, 331 11 We exclude income from the covariates in regression adjustment because high missingness reduces sample size considerably. Table 2 indicates balance on income. 12 We provided more detail in a prior vignette: "A number of different policy ideas are being discussed to try to make the electricity system in California more stable. One idea is to bury power lines underground. This 10, 25, 50, 75, 100, 110. This study was reviewed and approved by the University of California Office of Research as Demski, C., Capstick, S., Pidgeon, N., Sposato, R. G., and Spence, A. (2017) In the main paper, we present the difference in means within the matched sample between 394 treatment (outage zones) and control (non-outage zones) groups for key outcomes. Here, we 395 present estimates from OLS regression with the outcome on the left-hand side, and treatment 396 (outage zone) and a matrix of covariates on the right-hand side (see Equation 1 above). Overall, 397 we recover consistent results when we include covariates in the analyses.

398
The following covariates were included: age, gender, education, ideology, income, partisan-399 ship, whether employed, whether a language other than English is spoken at home, and whether 400 there were children in the household. Education was measured on a 5-point scale (less than high 401 school, high school diploma or GED, some college, associates degree, bachelors degree of higher).

405
This section presents a number of analyses of heterogeneous effects of exposure to outages based 406 on respondent-level covariates (e.g. income, partisanship). We first explore heterogeneous treat-407 ment effects of outage exposure on household-level adaptation. Figure A4 presents analysis of 408 individual-level adaptation responses to outage exposure by income, recognizing that income 409 may moderate the ability of respondents to adapt. However, we do not estimate heterogeneous 410 treatment effects when we split the sample into two income groups (over / under $100K). We also, as demonstrated by Figure A5, do not estimate significant heterogeneous effects of 412 outage exposure on adaptation responses by partisan identity. 413 Figure A5: Estimated effects of outage exposure on adaptation, by partisan identity. Estimates from OLS regression on matched sample. Bars are 95 percent confidence intervals.
Household-level adaptation might also plausibly depend on distance from outage zones for 414 those in the control group, and relatedly, whether respondents have a close friend or family 415 member who was exposed to outages. However, Figure A6 indicates no statistically significant 416 differences among those in the control group by distance to the outage zone (splitting the control 417 group by the median distance to the outage zone). And Figure A7 indicates no statistically 418 significant differences among control respondents by whether they have a friend or family who 419 was exposed to outages. 420 Figure A6: Adaptation intentions by distance to outage zone. Far from outages indicates more than 1048 meters (median distance in the control group). Estimated proportions drawn from matched sample. Bars are 95 percent confidence intervals. Figure A7: Adaptation intentions by whether friends or family exposed to outages. Estimated proportions drawn from matched sample. Bars are 95 percent confidence intervals.
On the other hand, Figure A8 indicates that household-level adaptation did depend on the 421 length of outage exposure. In particular, those exposed to longer outages were more likely to 422 express an intention to install backup gas or diesel generation. They were also more likely to 423 report plans to change home landscaping to reduce wildfire risk, and less likely to report that 424 they planned to purchase an EV as their next car.  Figure A10: Effect of outage exposure on household-level adaptation and purchasing intentions, including Southern California sample Figure presents proportion of respondents stating intention to adopt behavior in outage-exposed zones and matched non-outage exposed zones. Bars are 95 percent confidence intervals. **p<.05 estimated treatment effect. In the second broad robustness check, we excluded individuals within 1km of the boundary 437 between outage and non-outage exposed areas, lowering the matched sample size from 678 to 438 426. This robustness check is meant to account for the fact that treatment close to the boundary 439 was fuzzy (e.g. some respondents in outage areas did not report experiencing outages). We, 440 again, recover broadly consistent results, except when it comes to estimating the effect of outage 441 exposure on plans to purchase an EV. 442 Figure A12: Effect of outage exposure on household-level adaptation and purchasing intentions, excluding observations close to boundary Figure presents proportion of respondents stating intention to adopt behavior in outage-exposed zones and matched non-outage exposed zones. Bars are 95 percent confidence intervals. **p<.05 estimated treatment effect, *p<.1.

Public Opinion Research Survey on California's Power System
Dear fellow Californian, As you know, there have been a lot of power outages across California over the past month. We hope that you or other members of your household were not harmed by the power outages or fires. We are inviting you to participate in an online survey to help understand the views and experiences of Californians about these important recent events. There is a lot of confusion about how people were affected, so your participation will help California officials make informed policy decisions.
Your address was randomly selected from a public list of California addresses, and we would invite anyone age 18 o o in your household to complete the survey. We expect this will take between 8 and 12 minutes. Responses are voluntary and will be kept confidential.
Summaries of our research findings will be made available to the public, the media, and to policymakers in California. Also, as a small token of our appreciation, we will send you a $5 digital gift card for completing the survey, redeemable at over 100 online vendors like Amazon or iTunes.
By taking a few minutes to share your thoughts and you will help us understand what Californians want. The survey is available now. We would appreciate if you would respond by November 27 .
We hope you enjoy completing the questionnaire and look forward to receiving your responses. Many thanks,

Matto Mildenberger
Professor of Political Science, University of California Santa Barbara   Since the beginning of October, did any of your close friends or family experience a planned California more stable.

700
One idea is to bury power lines underground. This would likely cost $3 million per mile. 701 Currently, California has over 175,000 miles of overhead power lines. This means that burying