With coastal populations growing and sea levels rising, reconstruction decisions after coastal disasters are increasingly consequential determinants of future societal vulnerability and thus the sustainability of development. The humanitarian sector tends to favour rebuilding in-place to avoid the social disruptions of mass relocation, yet evidence on what affected people want is mixed. Using the case of post-tsunami Banda Aceh, Indonesia, we investigate whether a policy to rebuild in-place in the disaster-affected area suits an urban population that was previously unaware of the hazard. We show that following the tsunami, a substantial proportion of the population prefers to live farther from the coast. This has caused a new price premium for inland properties and socio-economic sorting of poorer households into coastal areas. These findings show that offering reconstruction aid predominantly within a hazard-exposed area can inadvertently transfer disaster risk to the poor.
Beyond the immediate destruction and loss of life, disasters can influence long-term social and economic development1,2,3,4,5, with the well-being of the poor affected most severely6. The international disaster-risk-reduction community has long argued that reducing societal vulnerability to hazards is a key component of sustainable development7, a view repeatedly endorsed by United Nations member states in the Hyogo and Sendai Frameworks8. In post-disaster contexts, the widely adopted ‘build-back-better’ approach promotes sustainable development through integrating a wide range of vulnerability reduction measures into reconstruction and rehabilitation efforts8.
However, measures intended to reduce people’s vulnerability to natural hazards may entail difficult trade-offs against other factors that influence people’s vulnerabilities to a wider set of shocks. Such trade-offs are particularly salient in the decision of where to rebuild after a disaster. Although rebuilding in areas less exposed to hazards reduces vulnerability to those hazards, mass relocation projects often have negative social impacts on people’s livelihoods, land rights and community cohesion9,10,11,12,13. Given this difficult trade-off, the humanitarian sector has come to favour rebuilding in-place in order to avoid the social disruptions of mass relocation14,15, while trying to reduce vulnerability to natural hazards through other means. In addition, political, economic and logistical practicalities tend to make rebuilding in-place the more expedient option.
Yet while mass relocation can be problematic, it is not clear that a general policy to rebuild in-place is always consistent with the preferences of people affected by disasters. Studies have shown both that affected people tend to return after disasters16,17,18,19 and that some people prefer relocation20,21,22,23. Differing preferences may be expected, as people have differing experiences, priorities, livelihoods and attachments to place and community16,17,24. Especially after a disaster or provision of new information about hazards, housing markets show price discounts for properties closer to earthquake faults25,26,27,28 and on floodplains29,30, indicating some preference to live in areas less exposed to hazards. This in turn can lead to socio-economic segregation. Following Hurricane Andrew in the United States, middle-income households tended to move to less exposed areas whereas lower-income households moved into more exposed areas because of reduced property prices31. Especially in developing countries, locations that are exposed to frequent disasters tend to be inhabited by lower-income households32,33 (for an exception to this, see Supplementary Note 1). Such socio-economic sorting is a major social driver of vulnerability34,35, as lower-income households that are least able to cope with shocks become disproportionately exposed to natural hazards36.
Thus, it is not clear where it is best to rebuild, who should decide, nor how these decisions influence societal vulnerability in the longer term. To shed more light on this, we examine long-term outcomes of the reconstruction of Banda Aceh, Indonesia, following the 2004 tsunami. Unlike Hurricane Katrina, after which lower-income households faced substantial barriers to return37, the tsunami on 26 December 2004, which destroyed roughly half of the city of Banda Aceh38, provides an opportunity to observe the long-term impacts of a humanitarian policy to rebuild predominantly in-place in coastal areas that were affected by the disaster and remain exposed to coastal hazards in the future (Supplementary Note 2). Tsunami risk was largely unknown to the population of Banda Aceh before the 2004 tsunami (Supplementary Note 3), therefore this case provides important insights for post-disaster reconstruction efforts in situations in which a disaster generates new local knowledge of a hazard. Such situations are made more likely as growing coastal populations face rising sea levels and the potential for more intense storms39. If sustainable development is in part contingent upon reducing disaster risk, then it is critical to know how post-disaster resettlement policies endorsed by the international humanitarian sector influence societal vulnerability.
Across Aceh province, the 2004 tsunami caused an estimated 160,000 casualties40 and led to an international reconstruction effort costing around US$6.7 billion41. Initial plans by the Government of Indonesia called for housing aid beneficiaries to be offered the choice to return to their previous place of residence or to relocate farther from the coast, according to their wishes42. This was not implemented43. Instead, facing intense pressure to rebuild quickly and difficulties in acquiring land for relocation settlements, aid providers rebuilt housing predominantly within the tsunami-affected area, with few changes to infrastructure that would aid evacuation44. Surviving landowners typically received housing aid only to rebuild on their original plots (Supplementary Note 4). This reconstruction approach provides a direct test of the humanitarian consensus that people want to return and that it is best to rebuild in-place14,15. We are not aware of previous studies that have examined the suitability of such a policy for the affected population, nor the longer-term consequences of this.
To examine the long-term impacts of the policy to rebuild in-place, we collected multiple sets of data at both the household and village (neighbourhood) levels in both the tsunami-affected and inland areas of the city of Banda Aceh (Supplementary Fig. 1) in 2014–2015, 10 years after the tsunami and 5 years after the official end of reconstruction. We interviewed the following groups: a representative sample of households currently living in aid houses in the tsunami-affected area (n = 576); a purposive (selective) sample of tsunami survivors who have moved on their own (without aid) to the inland part of the city (n = 180); village leaders in both parts of the city (n = 88); higher-level officials who had worked in the reconstruction agency or local government during the reconstruction (n = 11); and private property developers (n = 3). We analysed the spatial patterns of post-reconstruction urban development as well as official population and poverty data collected at the village level. We investigated whether aid beneficiaries still live in their aid houses, whether residents place importance on tsunami risk in where they wish to live, and whether socio-economic patterns in the city resemble their pre-disaster state or reflect a new price premium for properties farther from the coast with associated socio-economic segregation (for further details, see Methods).
Our geospatial analysis and field verification show that aid houses were predominantly built back on the same plots where houses stood before the disaster. More than 99% of aid houses were occupied as of 2014 (population has also recovered; Supplementary Fig. 2). However, in our representative survey sample (n = 576), only 40% of aid houses are occupied by aid beneficiaries who now say that they preferred to return to their pre-tsunami place of residence, while 17% of aid houses are occupied by aid beneficiaries who now say that they would have preferred to have relocated outside the tsunami-affected area, had they been given the chance. The remaining 43% of aid houses are occupied by post-tsunami newcomers to these villages, most of whom (85%) are renters. These findings, as well as our interviews with village leaders, indicate that a substantial proportion of tsunami survivors never returned to live in their aid houses, returned and later moved elsewhere, or wish to move elsewhere but are constrained from doing so.
We interviewed a purposive sample of tsunami survivors who moved permanently out of the tsunami-affected area of Banda Aceh and into the inland area (n = 110). When asked why they moved (open-response, unprompted), 43% of survey participants said that tsunami risk was one of the reasons that they moved to their current residence (Supplementary Fig. 3). They often expressed this as concern about a tsunami recurring or mental trauma from the 2004 tsunami; less frequently, they said that being in their former village reminded them of lost loved ones. Other important reasons for moving included family living in their new village (mentioned by 47%), having already owned land in their new village (mentioned by 42%), marriage (mentioned by 21%) and work (mentioned by 19%). Although people’s recollections of their motivations for their move may differ from their motivations at that time, these data provide at least a crude estimate. It is notable that people mentioned tsunami risk as frequently as other reasons for moving. Tsunami risk was also an important motivation for those who moved to the inland part of Banda Aceh from tsunami-affected areas outside Banda Aceh (Supplementary Fig. 3).
To examine the importance that people place on exposure to tsunamis relative to other housing characteristics, we asked survey participants to rank the importance of four housing characteristics if, hypothetically, they were moving to a new house (‘Close to family’; ‘Close to work, market and schools’; ‘Located in an area safe from tsunami’; ‘Size and quality of house’). We asked this question of tsunami survivors who moved from the tsunami-affected area to the inland area, tsunami survivors who still live in the tsunami-affected area and newcomers to the tsunami-affected area. Strikingly, roughly half of each group ranked tsunami safety as first or second importance of the four factors; this proportion is slightly higher for those who moved to the inland part of the city, but this difference is not significant (χ 2(2) = 2.4, P = 0.31; Fig. 1a). In contrast, these groups do differ significantly in proxies for socio-economic status: educational level (χ 2(2) = 20.9, P < 0.001; Fig. 1b) and livelihood type (χ 2(2) = 26.7, P < 0.001; Fig. 1b; further livelihood data presented in Supplementary Table 1). The proportions with university or vocational degrees and with income as civil servants are significantly higher among tsunami survivors who moved inland than among either tsunami survivors or newcomers who currently live in the tsunami-affected area (contrasts significant at least at P < 0.05; Supplementary Table 2). Employment in civil service is particularly relevant to housing, as the reliable salary enables access to home loans. Taken together, this evidence suggests that although tsunami risk is an important motivation to move to the inland part of the city, socio-economic status constrains who is able to do so.
An increased population preference to live in the inland part of Banda Aceh should lead to higher rates of post-reconstruction urban development and a price premium for properties relative to the tsunami-affected area. Our analysis of post-reconstruction urban development shows that from 2009 to 2013, the number of houses increased by 9.5% in the inland area but by just 4.2% in the tsunami-affected area. This pattern is striking because as of 2009, there was 3.3 times as much undeveloped land in the tsunami-affected area as there was in the inland area. Our interviews with property developers indicate that they are building new housing developments mostly, or exclusively, outside the tsunami-affected area. They say their customers wish to live outside the tsunami-affected area because of safety from tsunamis for themselves and their families, to have a safe investment to leave to their children and because of remaining trauma from the 2004 tsunami. Most of their customers are middle-class or wealthy.
We find that before the tsunami there were no significant differences in prices of land, houses or house rental between the inland area and what would become the tsunami-affected area (respectively: t(31.0) = −0.4, P = 0.68; t(31.7) = −1.4, P = 0.17; t(39.9) = −0.6, P = 0.59; Fig. 2, Supplementary Fig. 4, and Supplementary Table 3). Today, in contrast, we find that land, house and house-rental prices are significantly higher in the inland area than in the tsunami-affected area (means 2.1, 2.4 and 1.8 times higher, respectively; P < 0.001 for each), with large effect sizes (r = 0.70, 0.88 and 0.78, respectively; Fig. 2, Supplementary Fig. 4 and Supplementary Table 3). We asked village leaders (unprompted) what factors influenced property prices in their villages; tsunami risk was mentioned as an important factor by village leaders in 19 of 26 studied villages in the tsunami-affected area and in 11 of 18 studied villages in the inland area. In these open responses, village leaders did not attribute changes in patterns of property prices to other factors such as job prospects.
Associated with these changes in patterns of property prices, we find that before the tsunami (2003), poverty rates were slightly lower in what would become the tsunami-affected area than in the central or inland areas (Fig. 3, Supplementary Table 4). Following the reconstruction (2011 and 2014), however, the pattern had reversed: the poverty rate was much higher in the tsunami-affected area than in the central or inland areas.
Aside from reactions to tsunami risk, it is plausible that long-term effects of losses from the disaster and/or an incomplete recovery of employment opportunities within the tsunami-affected area may also have contributed to observed changes in socio-economic patterns in the city. Evidence suggests, however, that these factors are unlikely to fully account for the observed changes. The tsunami-affected and inland areas have similar commuting distances to the city centre, where many people work. Post-disaster aid provided abundant housing and many livelihood rehabilitation and training programmes. Among our sample of tsunami survivors who moved from the tsunami-affected area of Banda Aceh to the inland area, 19% said work was an important factor, whereas 43% said tsunami safety was an important factor. Many newcomers to the tsunami-affected area themselves came to the city for work. Finally, in open-response questioning of residents and village leaders, local job prospects did not emerge as a major reason for changes in the socio-economic characteristics of these communities.
The reconstruction of Banda Aceh, where beneficiaries typically were offered housing aid only to return to their pre-disaster place of residence, provides a direct test of the humanitarian consensus that people generally want to return and that it is best to rebuild in-place. Our findings contradict this consensus. We show that although many tsunami survivors did wish to return, many other tsunami survivors and newcomers to the city prefer to live outside the tsunami-affected area. As a result, a policy to rebuild mostly in-place has had unintended negative consequences. Only about half of aid houses are still occupied by aid beneficiaries. Many tsunami survivors, particularly those of higher socio-economic status, moved to the inland part of the city; safety from tsunamis was an important motivation for many of them. The resulting large supply of unwanted aid houses has acted as a magnet, drawing lower-income newcomers to live in the hazard-exposed part of the city, even though some of them would prefer to live in less-exposed areas. Some tsunami survivors who still live in the tsunami-affected area would like to live farther from the coast but cannot afford to do so, as population preferences have led to a price premium for inland properties. Before the tsunami, there was no socio-economic segregation according to tsunami exposure. Now, in stark contrast to the humanitarian ideal to ‘build back better’, lower-income residents disproportionately live in areas exposed to tsunamis and other coastal hazards. This combination of reconstruction in-place, population preferences for living in less-exposed areas and consequent population movements has transferred disaster risk to the poor.
Our study shows that post-disaster reconstruction in hazard-exposed areas can unintentionally amplify societal vulnerability beyond that which existed before the disaster. Although this has not been shown elsewhere, we suspect that the case of Banda Aceh is not unique, as post-disaster research has generally not examined the long-term consequences of rebuilding in-place. Plausibly, similar patterns could occur in other contexts where the risk was unknown or under-appreciated before the disaster and where moving a short distance within an urban area substantially reduces hazard exposure. Research on these issues in other contexts is warranted in order to better understand the conditions under which humanitarian approaches may amplify societal vulnerability and jeopardize sustainable development.
Our findings extend previous research on migration and residential immobility in the face of natural hazards. It has been theorized that lower-income households may become ‘trapped’ in hazard-exposed areas—that lower socio-economic status makes people more vulnerable to extreme events and less able to move away from them36. We show that both disaster survivors and newcomers to the city can be trapped in hazard-exposed areas as a consequence of reconstruction in-place and new knowledge of a risk leading to a price premium for properties in less-exposed areas. An important caveat, also highlighted elsewhere24, is that not all lower-income households living in hazard-exposed areas are necessarily ‘trapped’ there; some households do want to live in these areas, for diverse reasons. Although many economic studies have shown price discounts for properties in hazard-exposed areas25,26,27,28,29,30,31, we show that this is linked to people’s stated motivations and to movement toward less-exposed areas by higher-income households. Finally, although many studies have observed that hazard-exposed areas tend to be occupied by lower-income households32,33,34, we show the origination of this pattern as a result of new knowledge of the risk and an abundant supply of aid housing provided in hazard-exposed areas.
Our findings imply that people should be given a choice in where they are resettled after a disaster, a policy that was initially proposed for the reconstruction of Banda Aceh42 but ultimately not implemented. A choice either to return to their pre-disaster place of residence or to relocate to a less hazard-exposed area should be offered to individual households rather than to a community as a whole, as we show wide variation in these preferences within communities. Such a policy would allow for the difficult trade-offs among hazard exposure, livelihoods and attachments to place and community to be made by the households themselves who must live with the consequences. This has the potential to shift the urban footprint away from hazard-exposed areas, leading to a more sustainable development pathway7,8, without compelling people to resettle in places where they do not wish to live.
Previous research indicates that the relationship between environmental hazards and human migration is complex16,17. This influenced our choice of methodology for this study in two ways. First, we examine several independent lines of evidence in order to examine many facets of people’s preferences and choices of where to live. Second, we include both quantitative and qualitative approaches. This enables us to both compare groups on standardized measures and to examine people’s motivations without unduly biasing or constraining their responses.
We examine whether aid beneficiaries are still living in their aid houses, whether tsunami survivors who remain in the tsunami-affected area do so because this is their true preference, the motivations of tsunami survivors who moved to the inland part of the city, the location of post-reconstruction urban development and the motivations for this, and changes in the housing market and socio-economic patterns in the city from the pre-disaster to post-reconstruction periods. This combination provides several independent lines of evidence at both the household and aggregate levels.
We worked with an experienced team of Acehnese field researchers to conduct these studies. We conducted survey training, pre-testing and a pilot in Indonesian. The field researchers conducted interviews in either Acehnese or Indonesian, depending on the interviewee’s preference. Responses were recorded and transcribed in Indonesian; we cleaned, verified and analysed these data.
The city of Banda Aceh comprises 90 villages. We classify these as ‘tsunami-affected’ areas with moderate to heavy damage38,45; ‘central’ areas with minor damage, including the city centre; or ‘inland’ areas that were not affected by the tsunami (Supplementary Fig. 1).
Household survey in tsunami-affected area
To examine the status and preferences of residents within the tsunami-affected area of the city of Banda Aceh, we conducted a representative household survey (n = 576). We based the sampling frame on numbers of aid houses per village, as reported by the Indonesian Agency for Rehabilitation and Reconstruction (BRR). We sampled 29 random clusters of 20 participants each (n = 580) in 26 of 39 tsunami-affected villages. We examined satellite imagery of these communities from the pre-tsunami and post-reconstruction periods. Using satellite imagery from the post-reconstruction period and field verification, we took a systematic random sample of households within each studied village to participate in the survey. If the interviewer was unable to contact occupants after two attempts, or if the occupants declined to participate, the interviewer moved to the next house in the random list for that village. Our team had 576 completed interviews (45%), 563 unsuccessful contact attempts (44%) and 149 instances in which people declined to participate (12%). There is likely to be survey fatigue in this population from the many surveys done by non-governmental organizations during the post-tsunami reconstruction.
The questionnaire included items about the household as well as about the individual who was participating in the survey. A question on household income was problematic: those with irregular income had difficulty calculating it, and those with higher income tended to under-report. Because of this, for analysis we favour educational level and livelihood type as less biased, if crude, proxies for socio-economic status.
We also individually interviewed 21 village leaders in 13 tsunami-affected villages and conducted focus-group discussions with residents in 16 tsunami-affected villages. With the aid of a village facilitator in each of these villages, we selected 8–12 participants who were original (pre-tsunami) residents of that village. The groups were roughly gender-balanced and included a range of ages and livelihood types.
Survey in the inland part of Banda Aceh
To examine the status and preferences of tsunami survivors who moved from tsunami-affected areas to the inland part of the Banda Aceh urban area on their own (without aid), we took a purposive sample of these tsunami survivors (n = 180) from 21 inland villages that lie outside the tsunami-affected area. We used a purposive sample because no other studies or records have tracked tsunami survivors who moved inland on their own. We selected 21 villages with higher numbers of new houses built between 2009–2013 (from our geospatial analysis described below) in the inland area of the city of Banda Aceh (13 villages) and its immediate urban periphery in Aceh Besar district (8 villages). Village leaders and facilitators helped our team to identify residents who moved to these villages from tsunami-affected areas.
To minimise bias in responses, each interview began with open questioning about the participant’s reasons for moving, without prompting them with tsunami risk or other specific factors. Participants could describe as many or as few reasons for moving as they wished. Later parts of the interview included closed-ended questions. In this sample, 38% had moved to their current village within a year of the tsunami (2004–2005), 34% moved while reconstruction was well underway (2006–2009) and 28% moved after reconstruction (2010–2015). We also interviewed village leaders in each of these villages.
For analysis, we subsample those whose original village is located within the city limits of Banda Aceh (n = 110), to make direct comparisons with our sample from our household survey in the tsunami-affected area (described above) that includes original residents who still live in their aid houses in tsunami-affected Banda Aceh and newcomers to tsunami-affected Banda Aceh. These data are presented in the main text, in Fig. 1 and in Supplementary Fig. 3. Our sample of those who moved from a tsunami-affected village to the inland part of the Banda Aceh urban area also includes people whose original village is located elsewhere in the province of Aceh (n = 70); we present these data in Supplementary Fig. 3.
Post-reconstruction urban development
To compare post-reconstruction housing development in the tsunami-affected versus inland areas, we identified and counted all houses in Banda Aceh and manually identified undeveloped areas using geographical information system analysis with satellite imagery from 2009 and 2012/2013. We also interviewed private housing developers (n = 3) and higher-level officials in the city and provincial governments (n = 11). As stated in the main text, we find that the number of houses increased by 4.2% in the tsunami-affected area, just half the rate in the inland area. The true rate of post-reconstruction housing development inside the tsunami-affected area is lower still, as some of the housing development observed in the tsunami-affected area during this time interval was post-disaster housing aid.
To test for systematic differences in property prices between the tsunami-affected and inland parts of the city, hedonic price modelling would have been ideal but was not possible because detailed data on house sales and characteristics do not exist. Nor was a contingent valuation study possible, because there were no potential respondents with knowledge of the city-wide housing market.
We found that of all potential informants, village leaders had the best knowledge of property prices. We therefore asked village leaders (n = 88) to estimate average land (village roadfront), house (two- to three-bedroom) and house-rental (two- to three-bedroom) prices in their villages before the tsunami (2000–2003) and after reconstruction (2010–2015). Pre-testing indicated that these house types and time ranges included enough home sales for village leaders to feel that they could make reliable estimates. We asked these questions of two village leaders per village in each of 26 randomly selected villages in the tsunami-affected area of Banda Aceh and 18 randomly selected villages in the inland area of Banda Aceh. To triangulate these numerical estimates, we also asked these same village leaders to rank property prices in their village relative to other villages in Banda Aceh at the time (five-point scale). Where the two informants’ numerical estimates were substantially different within the same village, we removed one if we had reason to think that one informant might have better knowledge than the other, or if ranking data revealed inconsistent responses from an informant. We then averaged retained estimates by village for further analysis. The two methods (numerical estimates and rankings) produced qualitatively consistent results (Fig. 2 and Supplementary Fig. 4). We also asked these village leaders what factors (open-response, unprompted) they thought contributed to changes in property prices in their villages relative to other villages in the city.
For visual comparability on Fig. 2, we adjust pre-tsunami (2000–2003) property price estimates for inflation using consumer price indices from 2003 and 2013 (accessed 4 April 2017)46. This choice of inflation adjustment has no influence on our analysis, which compares price estimates between the tsunami-affected and inland areas during the same time periods.
Population and poverty rates
To determine changes to the socio-economic characteristics of communities, we used village-level population and poverty data (‘Potensi Desa’) from the Statistics Agency of Indonesia (Badan Pusat Statistik) and the Population Agency for Banda Aceh (Dinas Kependudukan). We show population totals by area for the years 2003, 2005, 2011 and 2014 in Supplementary Fig. 2. We examine poverty rates before the tsunami (2003) and after reconstruction (2011 and 2014). The measures that are reported consistently across all examined time periods are the number of households granted poverty status (‘Surat Miskin’) and the total population. Surat Miskin is requested by the families themselves in order to access various forms of social aid; the decision to grant this status is made by the head of the village. We present these raw data in Supplementary Table 4. Because the total number of households is reported for 2003 and 2011 but not for 2014, we cannot exactly calculate poverty rate as a percentage for all time periods. To provide illustrative estimates, we first calculate an estimated average household size by dividing the total population by the total number of households. For the 2003 data, this average is 4.99 persons per household; for the 2011 data, this average is 3.89 persons per household. Because the 2014 data do not include the total number of households, we use the 2011 average of 3.89 persons per household to calculate the estimated poverty rates in 2014. Using these averages, we then calculate estimated poverty rates as the number of households granted poverty status (Surat Miskin) multiplied by the average number of persons per household and divided by the total population. We show this calculation in Supplementary Table 4 and plot the estimates in Fig. 3.
As presented in Fig. 1 and the accompanying text, we compare tsunami survivors who moved inland (n = 110), tsunami survivors who still live in the tsunami-affected area (n = 315) and newcomers to the tsunami-affected area (n = 210). We first test for significant differences among proportions using a chi-square test (stats::chisq.test function in R)48. Where the null hypothesis of no difference is rejected at P < 0.05, we use the Marascuilo procedure47 as described in Supplementary Table 1 to identify which contrasts are significant.
The t-tests reported on property prices (Supplementary Table 2) are two-tailed Welch t-tests, using the stats::t.test function in R.
The data that support the findings of this study are available on request from the corresponding author. These data are not publicly available owing to confidentiality agreements with research participants.
Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
This research is supported by the National Research Foundation Singapore and the Singapore Ministry of Education under the Research Centres of Excellence initiative. This work was funded by the Earth Observatory of Singapore (EOS) and is EOS Contribution Number 132. The International Centre for Aceh and Indian Ocean Studies facilitated this research in collaboration with Syiah Kuala University. Nizamuddin, Ardiansyah and M. Affan carried out geospatial analyses. Hayatullah, N. Anwar, Z. Ak, A. Uzia, C. Murnita, F. Nailufar, Fitriani, I. Fitria, Israyani, Jihan, Safrina and S. Tahir helped to refine research instruments and carried out field research. N. Elviera, I. Fitria, J. Yong, R. Zahara, S. Novita and D. Hundlani assisted with data management and verification. E. Maida, C. Dian Fitri and I. Arisandy provided operational support. P. Adamek provided comments on the manuscript.
Electronic supplementary material
Supplementary Information Notes on methods and additional figures and tables