Subjective measures of household resilience to climate variability and change : insights from a nationally representative survey of Tanzania

Promoting household resilience to climate extremes has emerged as a key development priority. Yet tracking and evaluating resilience at this level remains a critical challenge. Most quantitative approaches rely on objective indicators and assessment frameworks, but these are not fully satisfactory. Much of the difficulty arises from a combination of conceptual ambiguities, challenges in selecting appropriate indicators, and in measuring the many intangible aspects that contribute to household resilience. More recently, subjective measures of resilience have been advocated in helping to overcome some of the limitations of traditional objective characterizations. However, few large-scale studies of quantitative subjective approaches to resilience measurement have been conducted. In this study, we address this gap by exploring perceived levels of household resilience to climate extremes in Tanzania and the utility of standardized subjective methods for its assessment. A nationally representative cross-sectional survey involving 1294 individuals was carried out by mobile phone in June 2015 among randomly selected adult respondents aged 18 and above. Factors that are most associated with resilience-related capacities are having had advance knowledge of a previous flood, and to a lesser extent, believing flooding to be a serious community problem. Somewhat surprisingly, though a small number of weak relationships are apparent, most socio-demographic variables do not exhibit statistically significant differences with regards to perceived resilience-related capacities. These findings may challenge traditional assumptions about what factors characterize household resilience, offering a motivation for studying both subjective and objective perspectives, and understanding better their relationship to one another. If further validated, subjective measures may offer potential as both a complement and alternative to traditional objective methods of resilience measurement, each with their own merits and limitations.


BACKGROUND AND CONTEXT
Resilience has its roots in several different disciplinary fields, ranging from mechanics to ecology and psychology (see Alexander 2013 for a comprehensive historical overview).However, the term's more recent adoption across the sustainability sciences has bolstered its popularity amongst academic and policy communities alike.Indeed, resilience now forms a central pillar in many key international policy frameworks such as the United Nations' Agenda 2030, Sendai Framework for Disaster Risk Reduction, and Paris Agreement (United Nations 2015a,b,c).Though the utility of a broad resilience framing has brought many benefits, it has also contributed to definitional and conceptual inconsistencies in its use (Olsson et al 2015).For example, early framings of resilience in describing social systems leant heavily on ecologic frameworks and revolved around the capacity to absorb change and disturbance in order to maintain core functions (Holling 1973, Walker et al. 1981, Odum 1985).Subsequent thinking within the field of social-ecological systems has challenged these frames when applying resilience to understand human responses to climate risk, encouraging greater recognition of the ability of social groups to adapt and change their core structure and functions (Berkes et al. 2002, Walker et al. 2004, 2006, Folke 2006, Bollettino et al. 2017).In some cases, it is argued that the complete transformation of a system may be a necessary component of a resilience process (Kates et al. 2012, Aldunce et al. 2015).
Nevertheless, there is broad agreement across disciplines that resilience comprises a range of evolving capacities and processes rather than constituting a static state (Maguire and Cartwright 2008).For example, in the context of community resilience to disaster risk, Norris et al. (2008) propose that resilience can be broken down into three core capacities: robustness (the strength of a system's resources); redundancy (the extent to which elements are substitutable in the event of disruption or degradation); and rapidity (how quickly the resource can be accessed and used).Béné et al. (2012) propose a framework consisting of the capacities to absorb, adapt, and transform.Many other such frameworks exist.However, agreement over the exact characterization of resilience is missing amongst the wider literature.These conceptual distinctions matter not only because the capacities needed to support them are different, but because they present a fundamentally different conceptualization of what a resilient system constitutes: "Such wide meanings may end up being contradictory as in the notion of 'restoring equilibrium and getting away from it by moving to a new state'" (Alexander 2013, as cited in Olsson et al. 2015:22).
As such, properties such as coping, adaptive and transformative capacities are often used in different ways and in different combinations when framing resilience (Bahadur and Pichon 2017).Indeed, they are frequently used interchangeably with resilience itself (Olsson et al. 2015).This lack of definitional and conceptual agreement presents challenges when seeking to track and measure resilience, but these difficulties notwithstanding, a wide range of measurement toolkits have emerged in recent years (Constas and Barrett 2013, FAO 2016, Frankenberger et al. 2014, Béné et al. 2016a).
Most efforts to measure climate resilience use objective criteria, socioeconomic indicators and processes that are considered to support a household's ability to deal with risk (Schipper and Langston 2015)."Objective," in this context, tends to denote framings of resilience that are based on external judgement and verification (Maxwell et al. 2015).Such approaches tend to be guided by an overarching conceptual framework, usually designed by technical experts or those external to the individual or household themselves, though sometimes these draw on qualitative inputs from intended communities or the piloting of survey instruments (Jones and Tanner 2017).Yet there is no universal acceptance on how resilience can and should be measured and hence a plethora of different objective frameworks and indicator lists exist, some contrasting markedly (Constas et al. 2016).
One widely used example is the United Nation's Food and Agriculture Organization (FAO) Resilience Index Measurement and Analysis model (RIMA), which combines socioeconomic variables from five dimensions: access to basic services; assets; adaptive capacity; social safety nets; and sensitivity to shock.These are then further broken down into dozens of individual indicators (D'Errico and Giuseppe 2014).Given that RIMA is centered around a predefined framework primarily based on wider resilience literature, has a list of expert-derived indicators for each characteristic, and relies on extensive surveys to collect externally verified information on household socioeconomic conditions (FAO 2016), the approach can be readily classified within the objective camp of measurement tools.
Objective measurement approaches such as RIMA have many strengths, including the ability to generate composite scores of resilience that can be readily compared across households.However, they also present several clear limitations.Most notably, they rely heavily on predefined resilience characteristics and standardized indicators.This renders it challenging to capture the context-specific nature of resilience: factors that make a coastal fisher in coastal Kenya resilient are unlikely to be the same as those for a pastoralist in the northeastern drylands of Kenya.In addition, objective approaches operate on the basis that resilience https://www.ecologyandsociety.org/vol23/iss1/art9/can be externally determined.Such approaches favor structural determinants at the expense of those based on human agency, which may be harder to understand and measure (Tanner et al. 2015).Crucially, they do not take into account people's knowledge of their own resilience and how they evaluate their lives.Thus, objective evaluations often require value judgements in generalizing the factors that are assumed to make others resilient and simplifying the complex nature of resilience across differing contexts.Ironically, these value judgements mean that many socalled "objective" approaches to resilience fall short on their own terms.Despite the limitations, objective approaches to resilience measurement remain the norm and dictate to a large degree our understanding of resilience processes at all scales.
Though yet to be fully explored in both conceptual and practical terms, subjective methods may offer an alternative and complementary approach to objective assessments of resilience (Marshall and Marshall 2007, Nguyen and James 2013, Jones and Tanner 2015, Maxwell et al. 2015, Béné et al. 2016a,b, Seara et al. 2016).Subjective evaluations are often used to gain bottom up and grounded insights into people's own understandings of resilience and its components.A significant body of literature has sought to understand subjective elements of household resilience ( Twigg 2009, Buikstra et al. 2010, Gaillard 2010, Miller et al. 2010).The vast majority of this work is based on qualitative assessments, typically based on ethnographic case studies, interviews, and focus groups or participatory rural appraisals (PRA).Although these approaches provide tremendous value, particularly in allowing for depth and nuance, they are difficult to use as a basis for measuring resilience at scale or across contexts.We are concerned with a branch of subjective resilience consisting of standardized and quantifiable methods for evaluating perceived resilience.
At its simplest, subjective household resilience relates to an individual's cognitive and affective self-evaluation of the capabilities and capacities of their household, community, or any other social system to respond to risk (Jones and Tanner 2015).If care is taken to design suitable methodologies and survey questions, then a household's subjective resilience can, in theory, be readily quantified.Standardized subjective indicators can be measured in many ways.Perhaps the most evident and practical way of collecting standardized data is through large household surveys.Although open-ended questions might provide rich qualitative detail, closed-ended questions are more likely to enable the aggregation of scorings of resilience capacities and to facilitate comparison across social groups or time (OECD 2013).Indeed, insights and research from related fields, such as subjective wellbeing, risk perception, and psychological resilience suggest that that standardized subjective evaluations may help to capture many "softer" elements of resilience-related capacities; allow comparison across different contexts; and permit individuals' knowledge of the factors that contribute to their own resilience to be incorporated (Jones and Tanner 2015).
Despite its clear potential, relatively little is known about how people evaluate their own resilience using standardized subjective measures, nor whether factors traditionally associated with objective household resilience match those from subjective selfevaluations when assessed at scale.For this reason, in this article we seek insights into these research gaps.
We focus on a subset of resilience: household resilience to climate extremes.Multiple epistemological entry points for climate resilience exist, though most objective measurement frameworks break the concept down into a common series of distinct yet interrelated characteristics (Schipper and Langston 2015).Properties such as the capacity of a household to prepare for and reduce the impact of climate extremes (Bahadur et al. 2015); absorb and cope with disturbances; and modify and adapt structures in accordance with changing climatic stimuli (Jones et al. 2010) each commonly feature within the literature.However, myriad other properties and combinations thereof surface in each framework, and each also relies on different interpretations of resilience and its constituent processes and indicators (Schipper and Langston 2015).Identifying a common set of observable indicators that relate to a household's capacity to recover from climate extremes, or their ability to adapt to ever-increasing climate risk has so far proven difficult (Cutter et al. 2008).This is not least because many factors that contribute to resiliencerelated capacities are process driven and relatively intangible (Jones et al. 2010).For example, self-efficacy, social networks and cohesion, power and marginalization, and risk tolerance each help to determine a household's resilience (Adger et al. 2013, Béné et al. 2016b).For this reason, we believe that a subjective approach may have value in assessing resilience in this specific context.

Conceptual approach
To explore the feasibility of assessing subjective resilience quantitatively and its links to objective characteristics that often feature in resilience measurement, we added a module of closeended questions to a nationally representative longitudinal telephone survey in Tanzania.To narrow the focus, we concentrated our survey questions on household-level disaster resilience, more specifically, resilience to flood risk. [1]Our survey was based on a standardized and widely used framing of climate resilience (Aldunce et al. 2015, Bahadur et al. 2015), comprising of three core capacities.
The first capacity used in our subjective framework relates to a household's ability to prepare, more specifically, to anticipate and reduce the impact of climate variability and extremes through preparedness and planning, often by making use of relevant information and early warning (Bahadur et al. 2015).The second capacity relates to a household's ability to recover.This is primarily associated with its ability to absorb and cope with the impacts of climate variability and extremes, often through maintaining core functions or livelihood activities (Sundres andBirnbaum 2003, Folke et al. 2010).The third capacity relates to a household's ability to adapt, more specifically, to adjust, modify, or change its characteristics or actions to moderate potential damage or take advantage of new opportunities that arise (Jones et al. 2010).
We administered a single question to address each of these three capacities (Table 1).The three items are conceptually distinct, well understood by respondents, easy to use and, as we show below, were found to be relatively independent of one another.Each capacity question used a standardized unipolar Likert scale with four response alternatives.The approach builds on methods used to assess social resilience in individual communities (see Marshall https://www.ecologyandsociety.org/vol23/iss1/art9/as well as similar approaches used to evaluate subjective capacities in related fields such as subjective well-being (OECD 2013), risk perception (Mills et al. 2016), and psychological resilience (Connor and Davidson 2003).
Although each capacity question is administered via a single-item, as per the approach adopted by Béné et al. (2016a,b) [2] , we see resilience as multidimensional; therefore it is not amenable to being measured via a stand-alone question.It is important to underscore that none of the three capacity questions should be interpreted as representing resilience overall; it is for this reason that we refer to them as "resilience-related capacities" throughout.Instead they should be considered to represent several core functions deemed necessary to support household resilience.
This study is primarily exploratory.Although we seek to test a research question, given this is a novel area of academic interest with few other studies applied at the national level, we are primarily interested in understanding the dynamics of subjective resilience and working toward further validation and refinement of the tools applied.We hope that future work can build on these research insights and seek more confirmatory and experimental approaches, including the use of scales with numerous, replicative items (see Jones 2017).

Survey instrument
The questions were administered via a nationally representative survey in Tanzania, namely the Sauti za Wananchi (Voices of Citizens) longitudinal survey managed by the Tanzanian nongovernmental organization Twaweza and surveying company Ipsos Synovate.The survey is composed of two phases.First, a baseline survey was carried out through traditional face-to-face interviews using a multistage stratified sampling approach (Twaweza 2013).A sample of 2000 households in 200 enumeration areas were surveyed in October 2012, using a sampling frame designed to be representative of the Tanzanian population aged 18 years and older based on the 2012 Tanzania Population and Housing Census (NBS 2013).At this point, all households were given a mobile phone and solar charger.The second phase consists of a series of mobile telephone surveys with the same sampled households as in the baseline (Details of surveys to date and the datasets are available at http://www.twaweza.org/go/sauti-za-wananchi-english).
In the round associated with this paper's results, the survey focused on assessments of political leadership.Resilience-related questions were included in an add-on module. [3]Respondents were contacted in July 2015 to take part in the survey through a computer aided telephonic interview (CATI) operated via an Ipsos Synovate managed call center in Dar es Salaam.A total of 1335 respondents out of the potential 2000 from the initial baseline completed this wave of the survey. [4]Questions were administered in Swahili and English, with a small financial incentive provided to respondents for their participation (US$0.5 mobile airtime credit).For full details of the sampling procedure, weighting, and data collection see (Twaweza [date unknown]).For 1334 of the respondents, a wide array of socio-demographic data from the 2012 baseline are available, as well as responses to the resilience questions listed above. [5]We removed an additional 40 of these respondents from the dataset because it was not certain that the same person replied as in the baseline, leaving 1294 matched observations. [6]  In the analysis, we describe the characteristics of our sample and then present descriptive statistics on their reported resiliencerelated capacities, followed by multivariate analysis.Because the ordinal variables measuring resilience-related capacities are not normally distributed, we test the equality of proportions rather than means. [7]In the multivariate analysis, we used ordinal logistic models in which we regressed resilience-related scores-the extent to which respondents reported it was likely they could prepare for, adapt to, or transform their livelihoods in the event of severe flooding-on a range of objective controls to test whether these individual variables were independently able to predict levels of perceived household resilience.Independent variables included the age, gender, education, and household size of respondents, whether they were occupied in farming and whether they lived in an urban or rural area; [8] the wealth quintile of the household (using an asset index); and whether the household had previous experience of a flood, believed flooding to be a serious problem for their community, and whether they had known about the last flood that affected them (within the previous two years) in advance. [9]Given the regressors are the same across these models, we use a seemingly unrelated estimation technique to account for the correlation in the error terms (Weesie 1999, StataCorp 2013).

Sample characteristics
Our respondents to the survey were primarily household heads (98%), the majority of whom were male (57%). [10]They were primarily rural (65%) and occupied in farming (also 65%).We defined household wealth status according to an asset index that places households into quintiles. [11]Some 93% of households in the poorest asset quintile were in rural areas compared with about 16% of households in the richest asset quintile.Most respondents had a completed primary education (61%) whereas around 13% had at least some secondary education, 3% had a higher education, and just under 10% had no formal education.The mean age of respondents in our sample was 40 years, 37 for females and 42 for men, with a range of between 18 and 89 years old.

Experience of floods and perception of risk
Respondents were asked to report their previous experience of extreme flooding (see Table A1.1. [12]).Overall, 32% reported having experienced at least one such event in the past two years.Among respondents with recent experience of severe flooding, 26% reported having had advance warning before the flood.
Respondents were also asked how serious a problem extreme flooding was, independently of whether they had recently experienced a flood.Most did not report flooding as a serious concern either for their households (86%) or for their communities (71%; Fig. A1.1).However, respondents from households that had experienced a flood in the previous two years were far more likely to perceive flooding as problematic; close to 40% of the exposed population reported flooding as a serious problem or the most serious problem for their household and over half (54%) reported it as serious or most serious for their community, compared with 2% and 17% of those who had not been exposed to a recent flood, respectively.This is in line with previous studies finding experience of past flooding as a strong contributing factor to risk perception (Mills et al. 2016).There were no other sociodemographic cleavages, except that respondents from asset poor households were more likely to believe flooding was serious for their communities.
Among respondents with recent flood exposure, those who had early warning of that flood were more likely to perceive it as a serious problem, both for their households and their communities, than those who had not (Fig. A1.2).Some 57% and 67% of the population who had received advance warning of a previous flood perceived flooding to be a serious threat to their households and communities respectively, compared with 33% and 49% of those that did not have an early warning.

Perceptions of resilience and association with experience and perception of risk
Respondents assessed their perceived capacities to prepare for, recover from, and change their livelihood strategy in response to an extreme flood event.Most respondents reported a low perceived ability to prepare, recover, or change.Just one-third of the population reported that their households would be prepared in the event of a flood, one-quarter felt their households were capable of recovering fully within a six-month period, and 4 in 10 people felt their households could change their source of income/livelihood, if needed (Fig. 1).

Association between dimensions of resilience-related capacities
Examination of the relationship between the three capacities reveals that they are only moderately associated.For example, 53% of respondents who said they were likely to be prepared also said they were likely to recover, compared to just 10% of those who were unlikely to be prepared.Odd ratios show that respondents who were likely to prepare were five times more likely to recover compared to respondents unlikely to prepare (Fig. A1.3).Nevertheless, when using the four-point Likert scale, the rank order correlations among these three types of capacity were all less than 0.5 (Fig. A1.4).The highest correlation (0.45) is between reporting being able to prepare for a flood and to recover from it; while the lowest (0.25) is between being able to recover from a flood and to change one's way of life in response to it.We also constructed binary variables (likely/unlikely) and found very similar correlations.We examined whether these items could be combined to form an index of a latent construct of resilience.The three items did not meet the established threshold for internal consistency (Cronbach's alpha is 0.62, below the commonly accepted threshold of 0.7).However, item selection was also tested by principal components analysis that showed that the three items loaded strongly onto one variable with an eigenvalue higher than 1 (the threshold recommended by Kaiser's rule; Fig. A1.5).This gives some support for constructing an index of perceptions of resilience; however, in this paper we focus on analysis of the three components individually to obtain more insights into factors that are associated (or not) with each.We defer discussion of the value of a composite index for future work.

Factors associated with perceptions of resilience
We analyzed the three resilience-related capacities across a range of variables of interest (Figs. 2 and 3, Tables A1.1-A1.6).Male and female respondents provide very similar responses across the board, though this may not be surprising given that the survey deliberately asks respondents to rate household-level capacities, not individual ones.Fewer farmers than nonfarmers (and people in rural versus urban areas) report an ability to recover fully from an extreme flood event within six months.Responses are very similar across occupations and rural/urban zone with respect to the perceived capacity to prepare for and adapt, though a lower share of farmers and rural residents report that it is "extremely likely" that they would adapt to an extreme flood.Education is positively associated with the perceived capacity to recover from a flood but not with the capacity to be prepared or to adapt on average.However, far fewer respondents with a higher education believe it is "not at all likely" they would be prepared for or able to adapt to an extreme flood, relative to those with less education.Wealth quintile is not linked with perceived preparedness but a higher share of respondents in wealthier quintiles report that they could recover and change their livelihoods in response to an extreme flood event (Fig. 3).
Self-reported capacities differ more markedly in line with the recent experience of flooding.Indeed, a higher share of those who had an experience of extreme flooding in the two years prior to the survey reported that they would be likely or very likely to prepare and to recover (but not to change their livelihood).For example, one-quarter of the population with recent flood exposure reported it was "not at all likely" they would be prepared for or recover fully from extreme flooding within a six-month period, compared with over one-third (35-36%) of those who had not experienced a flood.This suggests either that perceptions are influenced by experience of flooding, which in turn builds confidence in their ability to deal with flood risk, or that floods have been experienced in areas where households have higher resilience-related capacities.
Perhaps of most interest, having had early warning is consistently and strongly associated with all three capacities (Fig. 3).For example, 45% of those with early warning of a previous flood reported it unlikely that they would be prepared for extreme flooding, compared with 70% of those who had not had such a warning.For the capacity to recover, the figures were 57% and 79%, respectively, while for the capacity to adapt, they were 40% and 64%.In other words, the differences associated with early warning ranged between 22 and 25 percentage points.The most common source of early warning is through local and national radio (75%) with television (18%) and newspapers (3%) being far less common.It could be that respondents in more resilient households are more likely to obtain information regarding upcoming extreme weather events, or conversely, that the receipt of such information improves household resilience (indeed both mechanisms could be in play, or an unobserved trait could influence both aspects).But given that the provision of early warning information is such an important policy lever, greater exploration of the hypothesis that making information about flooding available improves resilience-related capacities is warranted, in line with similar research on risk perceptions of flooding (Miceli et al. 2008).

Multivariate analysis
To understand better factors associated with the perceived capacity to be prepared for, to recover from, and to adapt to extreme flooding, and how they relate to one another, we conducted seemingly unrelated regressions using ordinal logistic models with the capacity variables as the dependent variables.Across all the models, it is immediately apparent that the regressors have negligible explanatory power, explaining at most 2% of variation in these capacities. [13]Very few variables display a statistically significant association with any of the capacities (Table A1.7).
Across all the regressions, the only consistent explanator was not having known about the previous flood; in all cases, this was associated with lower reported capacities and the coefficients were strongly statistically significant.Examination of the marginal effects reveals the extent of these gaps (Fig. 4).Interestingly, in all cases, predicted probabilities for respondents who had not experienced a flood or had experienced a flood but did not know about it in advance were very similar (the differences were not statistically significant).Meanwhile respondents who had had advance knowledge of a previous flood were more likely to report preparedness and the capacities to recover and adapt.
Other positive (and statistically significant) relationships were found between having a higher education and both preparedness and the capacity to recover; between household size and the capacity to recover; and between wealth quintile and the capacity to adapt (Table A1.7).The effect of age is negatively associated with reporting preparedness until the age of 35 and positive thereafter.None of the covariates variables had an equivalent effect to having known about a previous flood, with the sole exception of being in the top wealth quintile on the perceived capacity to adapt.

DISCUSSION
We have proposed a tool to explore perceived levels of household resilience to climate extremes in Tanzania, and how objective characteristics map onto these assessments.We seek to contribute to a nascent body of literature on the measurement of resiliencerelated capacities.The results of the nationally representative survey point to a logical association between the previous experience of flooding and perceived severe risk associated with flooding at the household and community levels.Respondents who had received advanced warning of flooding in the past were more likely to perceive it as a serious problem, both for their households and their communities, than those who had not.This association could be a function of the severity of the previous flood-more efforts are likely to be taken to warn people of more extreme events in areas where floods are more severe-but it could also reflect that people who believe flooding is serious are more likely to seek out advance warning.
The survey results also suggest that low resilience-related capacities appear to be a concern in Tanzania where most households reported limited capacities to be prepared for, respond to, or change their livelihood strategies in response to an extreme flood.The scores across the three capacities were fairly similar; around one-third of respondents felt they were likely to be prepared in the event of a flood, one-quarter felt they could recover fully within six months, and 4 in 10 felt they could change their livelihood if needed.It is intriguing, however, that a greater proportion of respondents felt able to change their livelihood strategies than to prepare for, and to a much greater extent, cope with, an extreme flood.This may suggest that perceived levels of adaptive capacity are higher relative to the other two resilience capacities.The share is somewhat lower among people with less education and fewer assets (though only the wealth differences are statistically significant); however, fully 30% of respondents without education and one-third of those from households in the poorest asset quintile felt that they would be able to adapt.The results could, in part, be explained by increasing levels of livelihood diversity and flexibility with regards to sources of income and livelihood among Tanzanians (Hedges et al. 2016).
In future work, it would be advisable to probe understandings of the adaptive capacity-related question, including whether people associate it with longer term change rather than short-term coping strategies, and the sorts of livelihood strategies people feel they can adopt.
The correlations among responses to the three questions were positive but lower than expected (less than 0.5), reflecting considerable diversity among households with respect to the three capacities.These moderate correlations (and the relatively low Cronbach's alpha of 0.62) also point to a lack of internal consistency, though principal components analysis showed that they loaded strongly onto a single factor.To better understand these three components, we treat them separately and defer the question of whether an index of resilience-related capacities could be useful to follow-up research.
What is perhaps most interesting is that though a small number of weak relationships are apparent, most socio-demographic variables do not exhibit statistically significant differences with regards to perceived resilience-related capacities, e.g., age, education, occupation, wealth status, and place of residence.This is important given that these factors feature in objective assessments and are typically assumed to be strongly associated with household resilience (see FAO 2016).In this, our findings align with those of Béné et al. (2016b) who find that among coastal fishing communities in Ghana, Fiji, Vietnam, and Sri Lanka, none of the demographic characteristics that they analyzed apart from assets had a demonstrable impact on subjective resilience, which they characterize as "individual perception and selfconfidence about their own ability to handle future events" (Béné et al. 2016b:21).It is perhaps notable that the strongest relationship we find among the demographic variables is between belonging to the upper wealth quintile and the perceived capacity to adapt.Moreover, our data show that relationships between socioeconomic variables and perceived resilience-related capacities are broadly similar among respondents who had recently experienced a flood, i.e., in the previous two years, and those that had not.https://www.ecologyandsociety.org/vol23/iss1/art9/Several areas are worth considering.On the one hand, these results could indicate that traditionally measured objective characteristics do not have a strong influence on individual perceptions of a household's ability to prepare, recover from, and adapt to climate risk.If replicated in other areas and through different means, this could in turn cast doubt on the suitability of objective characteristics as effective measures of household resilience overall (Levine 2014).On the other hand, a subjective approach to assessing household resilience may be a poor reflection of overall resilience: those with a low resilience may perceive themselves to be more resilient than they are, and vice versa.
Part of the difficulty in establishing which of these two positions is applicable is that there is no present means of validating one or the other.Both objective and subjective measures are approximations of a somewhat intangible, contextual, and evolving concept.This is similar in many ways to difficulties faced in defining and measuring concepts such as well-being, risk perception, and happiness (Deeming 2013).More needs to be done to examine the effects of different characterizations and framings of resilience within subjective survey modules so as to establish the robustness of comparative subjective scores.Additional considerations relate to the validity of the survey questions themselves, response structures, or the means of administering the survey by telephone (see Leo et al. 2015).Each may have affected the results of the survey and explain several of the counterintuitive findings.In addition, several of the variables focus on individual characteristics such as gender and education, making it difficult to differentiate between personal and household-level dynamics.To confront this subjective-objective mismatch, further qualitative work is recommended, to seek to establish which factors are more closely associated with community-wide assessments of preparation, recovery, and adaptation.Ultimately a long-term cohort study and natural experiment may be needed to effectively assess how well proposed measures of resilience, objective and subjective, measure actual resilience as demonstrated in the face of natural disasters.
It is striking, however, that this study finds that the strongest and most consistent relationship is with having had advance knowledge (presumably through some form of early warning system) of a flood's occurrence.The mechanisms are unclear and will warrant further exploration, but these variables suggest potentially valuable policy levers to enhance resilience through the provision of early warning and (to a lesser extent) by raising awareness about the potential severity of extreme flooding.Most importantly, the results provide some confidence to the considerable investments that have gone into early warning systems (EWS) as a means of supporting disaster risk reduction and resilience regionally and globally (Sorensen 2000, Basher 2006, Miceli et al. 2008).In particular, our findings underscore the centrality of radio in sharing weather-related information and warnings among East African communities.
We believe that our study also demonstrates the ability to use subjective measures of household resilience at scale, i.e., in a national survey, and to administer them using mobile phone technology, which confers notable advantages in terms of cost, frequency, and accessibility.The latter characteristic could be especially useful following a severe climatic shock.
It remains to be seen whether standardized scaled-up subjective approaches to resilience measurement can be used for crosstemporal and cross-cultural comparison.Although the former may be relatively straightforward, particularly if panel data are in use, the latter may require considerable thought and validation, both empirical and qualitative, to assess whether internalized notions of resilience mean the same thing across contexts, and if data collected at such scales are meaningful.However, in related work on subjective well-being, there is some, albeit mixed, evidence that cross-cultural measures are meaningful and valid (Jorn and Ryan 2014).If this holds for resilience, then subjective tools may offer some promise in tracing progress in resiliencebuilding over time and across contexts at local, national, and international levels.

CONCLUSION
The research presented in this paper represents one of the first efforts to collect nationally representative data on subjective aspects of resilience, namely perceptions of the capacity to prepare for, to recover from, and to adapt to an extreme flooding event.We also explore the potential for collecting such data via a mobile phone survey, taking advantage of an ongoing panel survey in Tanzania.
We find that while some factors traditionally associated with household resilience such as asset wealth are strongly associated with subjective assessments, others like levels of education, livelihood types, and degree of urbanization have weak and in some cases nonexistent statistical relationships to subjective resilience.However, receipt of advance knowledge of flood risk appears to be one of the strongest predictors of a household's perceived ability to deal with risk, with notable policy relevance.Above all, the research underlines the need for greater recognition of subjective elements of resilience, not only with regard to how psychological and socio-cultural factors may contribute to a household's ability to deal with climate risk, but also factoring in people's knowledge of their own resilience.
While the work we have presented suggests the approach we adopt is potentially useful, it is necessarily far from indicative or comprehensive at this stage.Further testing of this instrument and of other efforts to measure perceptions of resilience, alongside objective indicators, is warranted.This includes examining the implications of different definitions and framings of resilience on subjective scores; exploring different multi-item scales to measure subjective resilience; assessing how subjective resilience changes over time and across contexts; and establishing the effects of various cognitive biases.This research also draws attention to a more acute issue facing the study of resilience and resilience-related capacities, namely the lack of a gold standard of what constitutes resilience against which attempts at its measurement could be triangulated.That such a standard might itself be context specific adds yet an additional layer of complexity to this multilayered, yet vitally important, concept.
__________ [1] Flood risk was chosen specifically given that it is a rapid onset shock that is easily communicable and defined in a survey context.In addition, flooding is a hazard that affects large areas of Tanzania with recovery typically occurring immediately after the https://www.ecologyandsociety.org/vol23/iss1/art9/cessation of a flood.It is worth noting that extensive flooding had occurred two weeks prior to the survey (May 2015) affecting areas of Dar es Salaam, Arusha, Kilimanjaro, Tanga, and Kagera.
[2] Indeed, Béné et al. (2016a,b) apply a single-item question to represent household resilience in its entirety.Here we choose to disaggregate further by examining distinct resilience-related capacities. [3]Financial support was provided by the Global Resilience Partnership. [4]The individuals and households who participated in this round were assigned "weights" to adjust for nonresponse and design error (Twaweza [date unknown]).The resulting data are intended to be representative of the adult population of mainland Tanzania not including Zanzibar (Twaweza 2013). [5]For one respondent, baseline information was not available and so the corresponding data were removed. [6]Because the Sauti za Wanachi survey is administered by phone, each time it is conducted, the respondent is asked to give their name.In this round, 8 respondents gave a different name than in the baseline and 32 respondents did not provide a name.We removed all 40 responses. [7]The Wilcoxon-Mann Whitney statistic (for two groups) and the Kruskal-Wallis test (for more than two groups) were selected as the nonparametric test best suited to ordinal responses (following Marusteri and Bacarea 2010), although it does not permit incorporating the complex stratified survey design.Nonparametric tests do not make assumptions about the underlying distribution of a variable but are less powerful than parametric tests. [8]The sample size is not large enough to permit analysis by subregion apart from by urban-rural zone (Sana Jaffers, personal communication). [9]The belief that flooding posed a problem to the community and the household were highly correlated; we chose to include the former because the bivariate analysis revealed stronger relationships with the resilience-related capacities.We restricted the focus to flooding occurring in the last two years to ensure a relatively recent and consistent frame of reference. [10]Because almost all respondents were household heads, we focused our analysis on the gender of the respondent rather than female versus male headship. [11]The wealth index was generated by principal components analysis using the following household assets: radio, mobile phone, fridge, TV, sofa set, electric/gas cooker, motor vehicle, livestock, and water pump (Twaweza, personal communication). [12]The statistics presented here are for the population, i.e., they incorporate the complex sampling design, while Appendix 1 presents the unweighted data and test statistics for these data.In practice, the differences between the averages derived from weighted and unweighted data are very slight. [13]It is not feasible to compute a measure of goodness of fit for the ordinal logit that takes into account complex sampling design in STATA.To give an indication of the fit, we computed the Pseudo R2 for unweighted specifications of these regression, which yields values of about 2%.
Enumerator introduction: "Finally, I'm going to ask you about your household's experience of flooding over the last two years."Severity In the last two years, how serious a problem has flooding been to your household?4-point scale: (1) The most serious problem; (2) One of the serious problems of many; (3) A minor problem; (4) Not at all a problem In the last two years, how serious a problem has flooding been to your community?4-point scale: (1) The most serious problem; (2) One of the serious problems of many; (3) A minor problem; (4) Not at all a problem Early warning Please think about the last extreme flood that affected your household.Did you know about it in advance?3-point scale: (1) No; (2) Yes; (3) Household not affected by a flood in the last two years and Marshall 2007, Nguyen and James 2013, Seara et al. 2016),

Fig. 1 .
Fig. 1.Subjective self-evaluations of resilience-related capacities to extreme flooding in Tanzania.

Fig. 4 .
Fig. 4. Predicted probability of capacity to prepare, recover, and adapt to an extreme flood event, based on early warning of that event.Marginal effects computed on the basis of the separate ordinal logistic regressions.

Table 1 .
Resilience-related questions administered through the national survey.

Table A1 .
1. Respondents' experience of flood in previous two years and whether they knew of it in advance.

Table A1 .
3. Respondents' perceptions of flood severity for their communities.

Table A1 . 4 .
Perceived capacity to be prepared for an extreme flood by respondent characteristics.

Table A1 .
5. Perceived capacity to be recover fully from an extreme flood by respondent characteristics.

Table A1 .
6. Perceived capacity to change livelihood strategy by respondent characteristic.