Introduction

The frequency and intensity of disasters caused by natural hazards have almost tripled globally, leading to increased economic loss and damage and affecting more people (Thomas & Lopez 2015). Low-income and lower-middle-income countries in Asia and the Pacific are particularly vulnerable to these disasters due to high poverty rates, inadequate infrastructure, limited technological development, increased risk exposure, and weak disaster governance (Hallegatte et al. 2020; UNISDR 2019; Fankhauser & McDermott 2014; Winsemius et al. 2016). Natural hazard-induced disasters cause damage to life, property, and livelihoods, worsening poverty and inequality (Hannan 2002; Akter & Mallick 2013; Schwab et al. 2007; Akter 2021; Akter & Grafton 2021). The economic loss and damage from disasters as a percentage of the gross domestic product in developing countries far exceeds that of industrialized nations (de Goyet et al. 2006). In addition to economic damage, the disaster-affected population also experiences substantial non-economic welfare losses, such as loss of access to water, sanitation, healthcare, food, and nutrition.

Assessments of disaster impacts often focus on the immediate economic loss and damage caused by the direct effects of the disaster. The non-economic losses of welfare caused by the indirect effects of natural disasters often remain invisible (UNDRR 2023a; UNDRR 2022). According to the United Nations Framework Convention on Climate Change (UNFCCC 2021), economic loss and damage refers to the loss of physical assets, infrastructure, and services typically bought and sold in markets. Non-economic loss encompasses items not commonly traded in markets, such as loss of life and health and loss of access to water and sanitation. The medium-term non-economic welfare losses resulting from a catastrophic event are seldom considered in policy and program development, especially in the case of large-scale earthquakes, which cause massive infrastructure damage (UNDRR 2023b). It is important to consider these medium-term non-economic welfare losses in disaster risk reduction policymaking to count the true cost of disasters and to prevent affected communities from slipping into a worse socio-economic condition than they were in before the disaster (UNDRR 2022).

Recovery following a disaster comprises various stages, including emergency response and short-term, medium-term, and long-term recovery efforts (UNDP 2019). The emergency response stage focuses on immediate lifesaving and relief efforts, such as search and rescue operations and providing emergency shelter and medical care. The short-term recovery phase, also known as early recovery, involves restoring essential services, clearing debris, and providing temporary housing for displaced individuals. The medium-term recovery phase concentrates on addressing the ongoing social and economic impacts of the disaster. Lastly, the long-term reconstruction phase, or the development phase, aims to rebuild infrastructure, homes, and businesses. There is no consensus on the specific time periods that define these phases (Contreras 2016; Onuma et al. 2021). The duration of each phase depends on the severity of the disaster, the context of the affected area, and the progress made during each phase (Contreras 2016; Onuma et al. 2021). However, several studies and policy documents in the literature have considered 1 to 5 years after a catastrophic earthquake as the medium-term recovery phase (see Wang et al., 2020; Platt et al. 2020; UNISDR 2015; Weiss et al. 2014).

In the field of earthquake recovery assessment, many studies, particularly those focusing on medium-term recovery, prioritized examining economic loss and damage (Aguirre et al. 2023; Chatterjee et al. 2018; Zülfikar et al. 2017; Yong et al. 2001). Previous earthquake recovery assessments relied on data from a single year (Platt et al. 2020) or on macroeconomic indicators such as GDP (Zhu et al. 2018) and recovery indices based on spatial factors like changes in building conditions and building alterations (Contreras et al. 2018), data from social media (Contreras et al. 2022), time-series analysis (Karatani & Hayashi 2007) or a life recovery index (Karatani & Hayashi 2004). Non-economic welfare loss assessment of earthquakes in the medium-term is an understudied area (UNDRR 2023b).

Our research utilizes two publicly available data sources to conduct a medium-term assessment of the non-economic losses caused by a series of devastating earthquakes in Indonesia. These data sources include (a) slow-moving socioeconomic welfare data obtained from an administrative source and (b) high-frequency night-time illumination data. Night-time illumination, or nightlights, offer valuable insights, particularly for nations with limited socioeconomic data collection capabilities or outdated administrative survey data (Qiang et al. 2020; Xu & Qiang 2021; Gao et al. 2020; Li et al. 2018). Nightlight data is a useful and reliable proxy for human activities (e.g., GDP and electricity usage) where household data collection is challenging and time-consuming (Gibson et al. 2021; Beyer et al. 2021; Chen & Nordhaus, 2011; Chen & Nordhaus, 2019). It is recognized for its enormous potential to replace slow-moving and costly survey data, guiding policy decisions and program designs to facilitate post-disaster recovery (Yabe et al. 2022).

No research has been conducted to examine the ability of nightlight products to accurately represent the non-economic welfare of disaster survivors, despite the increasing interest in using them for disaster recovery assessments. There is growing interest in using freely available, near-real-time data for natural hazard damage and recovery analysis due to the high cost and time-intensive nature of socioeconomic data collected through primary surveys. Therefore, it is crucial to understand the relevance and reliability of these data (Yabe et al. 2022).

Our study offers two key contributions to the disaster risk reduction literature. Firstly, we are the first to utilize administrative survey data to assess the medium-term non-economic welfare losses following a major earthquake. This provides crucial evidence highlighting the enduring impact of non-economic welfare losses and emphasizes the necessity of considering them in policy design. Secondly, our study is pioneering in testing the applicability of high-frequency nightlight data, which is increasingly used in post-disaster needs and recovery assessments, for non-economic welfare loss assessments. We do this by comparing the findings of the administrative survey data with the findings of the nightlight data.

The paper proceeds as follows. Section "The 2016 Aceh Earthquake series" describes the features of the Aceh earthquake series. Section "Methodology" covers the data and our methodology. Section "Results" presents our regression results, Section "Discussion" discusses the results and lessons learned, and Section "Conclusions" concludes.

The 2016 Aceh Earthquake series

Nanggroe Aceh Darussalam is located at the tip of Sumatra, Indonesia's northernmost and westernmost island. With approximately 5.2 million people, Aceh ranked 20th in population density across Indonesia in 2019 (Statistics Indonesia 2019). The population density is low as the Acehnese are spread across 18 districts (kabupaten) and five municipalities (kota).

On Dec 7, 2016, Aceh was hit by a series of devastating earthquakes. The earthquake hit three times on the same day, starting with the lowest magnitude of 4.5 MW, followed by 5.2 MW, and ending with 6.5 MW. The last one was the strongest and the shallowest earthquake. The epicenter of the last earthquake was located near Reuleuet village in the rural Pidie Jaya district (see Fig. 1), which is southeast of the province's capital city, Banda Aceh (U.S. Geological Survey 2016). With over 100 people dead, 1,000 injured, 11,000 buildings destroyed, and 45,000 people displaced (WHO 2016), the series of earthquakes was the deadliest in Aceh since the 2004 earthquake, a 9.1 Mw megathrust followed by a tsunami that killed more than 167,000 people (U.S. Geological Survey 2004).

Fig. 1
figure 1

A flowchart of the conceptual framework and methodology. Notes: Villages within a 20 km radius of the earthquake epicenters were considered earthquake-affected villages, and villages within a 20 to 40 km radius of the earthquake epicenters were considered unaffected villages. ⟺ refers to comparability

Full size image

The provincial government declared a state of emergency two weeks after the earthquakes. The President, who visited earthquake-hit areas three two days after the impacts, mentioned that accelerating recovery in Aceh was a priority of the national government. The President’s directive required recovery to be completed by 2018, which included rebuilding housing, physical infrastructure and public infrastructure (e.g., health, education, and livelihood sectors) to revive the local economy (Reliefweb 2017). Coordination centers were set up, and search and rescue teams were assembled and deployed to the hardest-hit areas (Kholid 2016). As local hospitals were overwhelmed, severely injured victims were evacuated to Banda Aceh, which is located more than 170 km away (Ibrahim 2016). Temporary shelters, clothing, food, mental health and trauma support, and financial compensation for the victim's families were provided. Help from local governments came in the form of medicine, food, and search and rescue personnel, along with international aid from neighboring countries and international organizations, as well as aid from the private sector. On 12 December, 2016, the National Search and Rescue Agency confirmed that their operations officially ended (Masriadi 2016), and the emergency response period ended on 20 December, 2016 (WHO 2016).

By June 2017, 18 percent, or a total of Rp 626.3 billion (equivalent to USD 43.1 million) out of the committed Rp 3.4 trillion (equivalent to USD 234 million), was disbursed by the National Disaster Management Agency for rehabilitation and reconstruction in the three affected districts (Reliefweb 2017). The funds were allocated from the national budget and grants, the local budget, international aid, and corporate social responsibility finances. A portion of the total budget was added to the provincial and district government budgets to accelerate the rehabilitation and reconstruction of public facilities and houses.Footnote 1

Methodology

This section describes the methodology and conceptual framework used for the assessment. Figure 1 presents a flowchart summarizing the methodology and conceptual framework.

Data

Primary datasets utilized for analyzing socio-economic welfare are Village Potential Statistics (Podes) 2014 and 2018 (Statistics Indonesia 2014; 2018). The Podes dataset is collected by the Government of Indonesia at regular intervals, once every three years. The survey gathers data from village authorities. Upon completion, the questionnaire is signed by the village heads. It encompasses a wide range of village characteristics of almost 80,000 Indonesian villages (desa), including village identification numbers and various measures of the availability and access to services and infrastructure. In Nanggroe Aceh Darussalam province, the Podes dataset includes all 9,000 villages for the two periods, i.e., before (2014) and after (2018).

We used nightlights or night-time radiance from two sources: (i) VIIRS-DNB cloud-free monthly average radiance composites (version 1) made available by the Earth Observation Group at the National Geophysical Data Center of the National Oceanic and Atmospheric Administration (Earth Observation Group, 2022) and (ii) the annual nightlight luminosity scores obtained from the harmonized global nightlight dataset prepared by Li et al. (2020). The data from the VIIRS satellites have a resolution of 15-arc seconds (0.5 km × 0.5 km tiles near the equator). The units of VIIRS images are nanoW/cm2/sr. We utilized monthly nightlight data from before and a few years after the series of earthquakes (January 2014 – December 2021) to compare the temporal trends of monthly changes in nightlights in villages affected by the earthquakes and those that were unaffected. We used annual nightlight data of 2014 and 2018 for regression analysis to ensure their comparability with the Podes data. Since the Podes survey is conducted every three years, the annual nightlight data was most suitable for comparing their findings.

Third, we used the data from UNOSATFootnote 2 to measure the distance of villages from the earthquake epicenters. UNOSAT used geospatial information technologies (GIT) and applied a big data approach for real-time earth observation in ASEAN member countriesFootnote 3 (UNITAR 2019). In cases of earthquakes, UNOSAT used high-resolution satellite imaging to map the locations of the earthquake epicenters. It then conducted population exposure analyses and damage mapping based on the earthquake intensities, number of populations, and presence of infrastructure in the area.

Measurements

We measured non-economic welfare losses in three domains. The domains were access to healthcare, incidence of poverty and access to socioeconomic services. Table 1 summarizes the domains, indicators, their measurements and associated codes. The first domain, access to healthcare, captures one of the essential impacts of a disaster on household non-economic welfare covered in the literature.

Table 1 Indicators and measurement of household welfare
Full size table

We measured welfare in this dimension using accessibility (i.e., change in an intensive margin) rather than the presence or absence of a healthcare facility (i.e., the change in an extensive margin). More specifically, it was unlikely that a new healthcare facility would be built or an existing one would be closed after an earthquake. Instead, we hypothesized that an improvement or deterioration of access to healthcare facilities would be observed due to poor or good condition of infrastructure (e.g., roads, bridges, and culverts). The Podes survey asked whether a healthcare facility (e.g., a hospital, a polyclinic, or a doctor’s clinic) was available in a village. If a healthcare facility was unavailable, which was the case for 94%, 76%, and 62% of villages for hospitals, polyclinics and doctor’s clinics, respectively, survey respondents were asked about accessibility to the nearest healthcare facility. Accessibility was measured as 1 = very easy, 2 = easy, 3 = difficult, 4 = very difficult. For our analysis, we used a four-point ordinal scale where the highest value represented the highest ease of access and the lowest value referred to the greatest difficulty.Footnote 4

The second welfare domain was poverty incidence. We used nutritional security (i.e., malnutrition) and standard of living (i.e., access to clean water, sanitation, cooking fuel) to capture the non-economic aspects of poverty before and after the earthquakes. Nutritional security was measured using the number of people suffering from malnutrition in a village. The standard of living indicators were coded as binary variables, i.e., a value of 1 was assigned when the service was present and 0 if the service was absent. The availability of clean water was coded 1 if the main sources of drinking water for most families in a village were bottled water, metered plumbing water, unmetered plumbing water, and borehole or pump water. The availability of clean water was coded zero if the primary drinking water sources for most households in a village were wells, spring, river, lake, pool, and rainwater. The availability of clean sanitation was coded 1 when most families in a village owned a toilet. If most families relied on shared toilets, public toilets or no toilets, this variable was coded as zero. The availability of clean cooking fuel was coded 1 when most families in a village used Liquid Petroleum Gas as the primary source of cooking fuel. The variable was coded 0, where the primary cooking fuel sources were kerosene or firewood.

The third domain, access to socioeconomic services, was measured using four indicators: (i) the availability of early childhood education centres, (ii) village cooperatives, (iii) business credit facilities and (iv) universal healthcare subsidies. The first three indicators were measured as binary variables, coded 1 to indicate the presence of a service and 0 to indicate the absence of the service in a village. The last indicator, the availability of universal healthcare subsidy, was measured as the number of universal healthcare subsidy beneficiaries in a village.

To understand the mechanisms of the non-economic welfare loss before and after the earthquakes, we examined two potential pathways. The first one was road conditions and the second one was public transportation, both reflecting spatial connectivity as a proxy of recovery (Contreras et al. 2013). The non-economic welfare domains discussed above were likely to be directly affected if road conditions and communication improved or deteriorated after the earthquakes. In the study area, the main forms of public transportation were minibuses, motorized rickshaws (becak), and shared vans (angkot). Unlike more developed regions with modern, reliable, and well-maintained public transportation systems, Pidie Jaya's public transportation network was less developed due to limited government and private investment in this sector. Two variables represent the road condition and availability of public transportation in a village. Road condition was coded 1 if most roads in a village were made of concrete and zero otherwise. The availability of public transportation had three levels: 0 (no public transport), 1 (public transport is available but not operated daily), and 2 (public transportation is available and operated daily).

Empirical framework

The measurement of earthquake exposure required determining the proximity of a village to the earthquake epicentres. Villages within a specified distance of the earthquake epicenters were coded 1, while those outside this range were coded 0. According to the WHO (2016), the Pidie Jaya district was the area most severely impacted by the earthquakes. For each village in this district, we calculated the distance from the earthquake epicenters and determined that the mean distances to the first, second, and third earthquakes were 16.4 km, 10.6 km, and 12.9 km, respectively. The minimum and maximum distances were 0.30 km and 23.40 km, respectively. Based on these figures, we designated the earthquake affected (or treatment) group as villages located within a 20 km radius of the earthquake epicenters. Villages located beyond a 20 km radius were considered the unaffected (or control) group (Fig. 2). To ensure a valid comparison, we extended the distance and designated villages within a 20–40 km radius of the earthquake epicentres as the control group. This cutoff ensured that affected and unaffected villages were subject to similar time-varying unobserved factors.

Fig. 2
figure 2

Map of the 2016 Aceh earthquake series and the sampled villages

Full size image

Our null and alternative hypotheses were:

  • H0: Earthquake exposure did not impact non-economic welfare indicators and night-time radiance score.

  • H.A.: Earthquake exposure impacted non-economic welfare indicators and night-time radiance.

We applied the difference-in-differences (DiD) method to estimate the impact of earthquake exposure on non-economic welfare and night-time radiance score. The DiD method is an assessment technique utilized in a quasi-experimental context (Angrist and Pischke 2009; Wooldridge 2012). It draws on a blend of perspectives from before-after and treatment–control comparisons to estimate the causal impacts of a shock. The DiD method hinges on a parallel trend assumption, positing that in the absence of a shock, the treatment group would have exhibited the same temporal pattern as the control group, specifically concerning the outcome variable of interest.

Combining the before-after and treatment–control group comparisons, we estimated the impact of earthquake exposure on non-economic welfare indicators using the following equation:

$${Y}_{it}={\beta }_{0}+{\beta }_{1}{Earthquake}_{i}+{\beta }_{2}{POST}_{t}+{\beta }_{3}\left({Earthquake}_{i}\times {POST}_{t}\right)+{\varepsilon }_{it}$$
(1)

In Eq. 1, Yit is an indicator of non-economic welfare; Earthquakei is a binary variable that is coded 1 if a village is located within a 20 km radiuses of the earthquake epicentres. Villages within a 20 – 40Footnote 5 km radiuses of the earthquake epicentres were included in the analysis sample and coded as 0. POSTt is coded 0 for the year 2014 and 1 for the year 2018. εit is an error term clustered at the village level. \(\beta {3}\) is the DiD estimator, which captures the differences in pre- to post-earthquake change in Y between the villages affected and unaffected by the earthquakes.

Results

Non-economic Loss Assessment Using Administrative Data

The findings of the regression models are presented in Tables 2, 3 and 4. The variable of interest in all regression models is Earthquake x Post (2018). The results from the healthcare access domain are presented in Table 2. This domain was measured using four indicators: ease of access to hospitals (with and without inpatient service), polyclinics, and doctors’ clinics. Ease of access was measured using a five-point scaleFootnote 6 where the highest value reflected the greatest ease of access, and the lowest value reflected the least ease of access. All coefficients of Earthquake x Post (2018) presented in Table 2 are negative and significant, at least at the five percent level. These coefficients indicate that access to all healthcare facilities in earthquake-affected villages significantly decreased after the earthquakes.

Table 2 The impact of the earthquakes on healthcare access: Difference-in-differences regression results
Full size table
Table 3 The impact of the earthquakes on poverty: Difference-in-differences regression results
Full size table
Table 4 The impact of earthquakes on social and economic service availability: Difference-in-differences regression results
Full size table

The results for the incidence of poverty are presented in Table 3. The first indicator used in this domain was malnutritional, which was measured by the number of malnourished people in a village. The coefficient of Earthquake x Post (2018) shown in column 1 is positive and significant at the 10 percent level. This suggests that, on average, the number of malnourished people in earthquake-affected villages increased significantly after the earthquakes. When considering indicators reflecting the standard of living, we observed a significant decline in the availability of safe drinking water and clean cooking fuel in earthquake-affected villages (columns 2 and 4). As for access to sanitation, the coefficient of Earthquake x Post (2018) shown in column 3 is positive but not statistically significant at the 10 percent level.

In the next section, we examined the third domain of socioeconomic welfare, which was access to socioeconomic services (Table 4). This domain was represented by four indicators: the availability of a publicly funded early childhood education center, the availability of business credit, the presence of a village cooperative, and the number of universal healthcare beneficiaries. All coefficients of Earthquake x Post (2018) shown in Table 4 are negative and statistically significant, at least at the 10 percent level. These results suggest that the earthquake-affected population experienced a significant decline in access to socioeconomic services after the earthquakes compared to nearby unaffected villages.

Recovery assessment using night-time radiance data

According to Fig. 3, both the earthquake-affected and unaffected villages exhibited a similar pattern of changes in monthly night-time radiance scores prior to the earthquakes in December 2016, thereby satisfying the parallel trend assumption. We confirmed the validity of the parallel trend assumption by conducting an F-test. The F-test hypothesized that the pre-earthquake trends of nightlights in the earthquake-affected and unaffected villages were parallel. This hypothesis cannot be rejected at the 10 percent confidence level (F (1, 774) = 1.34, p > F = 0.2469).

Fig. 3
figure 3

Trends of monthly night-time radiance scores change of earthquake-affected and unaffected villages. Notes: Author’s calculation using UNOSAT and VIIRS data. Nightlight is measured in nanoW/cm2/sr

Full size image

Following the earthquakes, there was a noticeable contrast in the night-time radiance trends between the affected and unaffected villages. Both groups of villages experienced an immediate drop in night-time radiance scores, which further declined a month later. The brightness of the affected villages began to significantly increase two months after the earthquakes, surpassing their unaffected counterparts. This trend persisted until the end of 2021, at which point both groups began to converge as the comparison period ended.

The results of the DiD regression models using annual night-time radiance data for the years 2014 and 2018 for both the earthquake-affected and unaffected villages are shown in Table 5. The table includes three columns, each presenting regression coefficients estimated using different definitions of earthquake-unaffected villages. In column 1, the analysis sample included villages located within a 20–40 km radiuses of the earthquake epicenters, which were coded as earthquake-unaffected villages, while villages outside the 40 km radius were excluded from the analysis. In column 2, villages within a 20–50 km radius of the earthquake epicenters were included in the analysis sample and treated as earthquake-unaffected villages, with the remaining villages excluded from the analysis. In column 3, the analysis sample included villages within a 20–60 km radiuses of the earthquake epicentres, which were treated as earthquake-unaffected villages.

Table 5 The impact of the earthquakes on night-time radiance scores: Difference-in-differences regression results
Full size table

All DiD coefficients, specifically the Earthquake x Post (2018) coefficients presented in Table 5, are statistically significant at the one percent level. These results align with those shown in Fig. 3, which utilized monthly changes in nightlight data in villages affected and unaffected by earthquakes. They indicate that, on average, the changes in night-time radiance scores in earthquake-affected villages in 2018, relative to 2014, are greater than those in earthquake-unaffected villages. In terms of magnitude, the luminosity scores in earthquake-affected villages were 33%, 45%, and 48% higher than those in earthquake-unaffected villages in columns 1, 2, and 3, respectively.

Robustness checks

To assess the reliability of the main findings, we estimated Eq. 1 by keeping the earthquake-affected villages constant and expanded the scope of the earthquake-unaffected villages. This was done to determine whether our main findings were influenced by the selection of the earthquake-unaffected villages. We re-evaluated the parallel-trend assumption tests by expanding the distance of the control villages. Our results indicated that the parallel trend assumption held when comparing the affected villages (< 20 km) to unaffected villages located within 20–50 km and 20–60 km of the earthquake epicenters (refer to the results from the parallel-trend assumption tests in Appendix figure 4 and 5). However, when we set the distance for unaffected villages to > 20–60 km, the parallel trend assumption was no longer met (F (1, 1214) = 10.05, p > F = 0.0016). This suggests that villages further away from the earthquake epicenters did not compare well with the earthquake-affected villages. We then re-estimated all regression models using all villages located within the 20–50 km and 20–60 km radiuses from the earthquake epicenters as unaffected villages.Footnote 7 Our findings remain consistent for all regression results presented in Tables 2–5 except for the malnutrition indicator.

Mechanisms

First, we checked the changes in road conditions and public transportation availability in earthquake-affected and unaffected villages by estimating two regression models (Table 6). The results show that both the physical condition of the road (i.e., concrete versus mud road) and the availability of public transportation deteriorated in the earthquake-affected villages relative to the unaffected villages in 2018.

Table 6 The mechanisms of non-economic welfare loss: Difference-in-differences regression results
Full size table

Discussion

The objective of this study was to assess medium-term non-economic welfare losses after a catastrophic earthquake using administrative survey data and high-frequency nightlights. Our results reveal a substantial lossin non-economic welfare across all domains and indicators (except for sanitation), pointing to a widespread negative impact on various aspects of life in the earthquake-affected villages. Our findings thus present strong evidence supporting the “invisible” cost of disasters, as highlighted in the Global Assessment Report (GAR, 2022) of the UNDRR (2022).

Although our data cannot be disaggregated by gender, from the nature of non-economic welfare loss indicators, it seems likely that women and girls bear the burden of welfare loss more heavily than men. For example, a lack of access to clean water and fuel in the earthquake-affected villages implies a high burden of unpaid household work for women and girls since they are primarily responsible for water and fuel collection in rural Indonesia (Irianti & Prasetyoputra 2019). Further, a lack of access to early childhood education implies a loss of human capital for young children in their formative age and a high burden of unpaid care work for mothers (Halim et al. 2022). The challenges in accessing healthcare services after earthquakes are also likely to have a greater impact on the well-being of women compared to men. Coupled with pre-existing gender inequality in healthcare access and utilization, the decreased accessibility to healthcare facilities following an earthquake may exacerbate the risk of poor maternal and neonatal health outcomes (Sajow et al. 2021).

Consistent with the evidence presented by previous research on disaster risk reduction (Laili et al. 2022; de Haen et al. 2007), our results shows that earthquakes caused malnutrition in the study area. This is due to disruptions in the supply chain, which lead to higher food prices and the loss of income and livelihood, reducing access to nutritious food. Although malnutrition data is not broken down by gender, the evidence available in other empirical studies indicates that women and girls are at a higher risk of malnutrition than men and boys (UNICEF, 2023). This is primarily due to the discriminatory gender norms associated with food allocation  in rural societies. These norms often result in women and girls having less access to nutritious food and healthcare compared to their male counterparts, leading to a higher prevalence of malnutrition and disease burden among females.

The deterioration of road conditions and resultant deficiency in public transportation following the earthquakes partly contributed to the poor welfare outcomes. Difficulties in transport and poor road conditions exacerbate challenges in mobility and access to essential services, creating impediments to recovery efforts. Poor infrastructure limits physical access to crucial resources like healthcare and markets and also disrupts economic activities by restricting movement and trade. This relationship between infrastructure quality and welfare outcomes points to the necessity of prioritizing road and transportation system repairs in post-disaster recovery plans.

The comparison of findings using nightlights presents contrasting results. Nightlights are commonly used as a proxy for measuring economic loss and damage (Qiang et al. 2020; Xu & Qiang 2021; Gao et al. 2020; Li et al. 2018). However, when it comes to assessing non-economic welfare losses, this data appears to be a less reliable proxy for socio-economic indicators. This may be due to the reliance on natural resources in rural economies, for which nightlights are not a suitable indicator (Chen and Nordhaus, 2019; Gibson et al. 2021). Previous studies, such as Gao et al. (2020), similarly found a surge in nightlights in earthquake-affected areas due to increased reconstruction activities. Based on the infrastructure conditions and accessibility to healthcare and other services documented by the Podes 2014 and 2018 data, there is insufficient evidence to support the hypothesis that an increase in night-time radiance was driven by reconstruction work in our study area. This suggests that while nightlight data can offer a dependable proxy for economic welfare in developed countries (Chen and Nordhaus, 2011; Chen and Nordhaus, 2019), as well as timely and broad overviews of economic loss and damage in countries with limited socioeconomic data collection capabilities (Qiang et al. 2020; Xu and Qiang 2021; Gao et al. 2020; Li et al. 2018), it may not effectively capture the nuanced and multifaceted aspects of non-economic welfare losses in a rural setting.

Conclusions

Our findings provide strong evidence of the non-economic welfare losses caused by large-scale earthquakes that often remain “invisible”. We have demonstrated that by utilizing routinely collected administrative data in developing countries, these “invisible” losses can be made visible in a cost-effective manner. The multidimensional approach that we employed in our study serves as a valuable tool for conducting a comprehensive assessment across various domains, enabling a more nuanced understanding of the non-economic welfare losses from earthquakes. While high-frequency, low-cost nightlight data is becoming increasingly popular for assessing economic damage and losses from disasters, it is not a suitable proxy for ground-level survey data. Therefore, it is essential to exercise caution when relying solely on nightlight data to assess the non-economic losses of earthquakes.

Future research should focus on examining how non-economic welfare losses are distributed among different groups, particularly marginalized and vulnerable populations. Previous studies have revealed that the impact of natural disasters disproportionately affects marginalized groups facing gender-based inequalities and discrimination. Due to the absence of gender-disaggregated data, our study could not analyze the varied distribution of welfare losses across different socio-economic and demographic groups. Where gender-disaggregated data is available, it is crucial for assessments to consider the diverse distribution of non-economic welfare losses. Additionally, administrative data collection should be designed to encompass the welfare impacts on gender, age, and other socio-economic categories.

In future research, it is important to delve into the potential of high-frequency nightlight data in accurately predicting non-economic welfare outcomes in the realm of disaster recovery. This entails pinpointing the specific economic activities and infrastructure that can be effectively gauged using nightlight data, while also differentiating these from aspects of non-economic welfare that require alternate assessment methods. By integrating nightlight data with rapid, short-term ground-level data collection, a more comprehensive understanding can be attained. Conducting swift surveys to capture real-time conditions can complement the high-frequency nightlight data, resulting in a broader comprehension of economic and non-economic recovery. This holistic approach will elevate the precision and dependability of disaster assessment frameworks, ultimately strengthening the formulation of effective and targeted policies and programs.