Following resettled people over time: the value of longitudinal data collection for understanding the livelihood impacts of the Three Gorges Dam, China

Abstract

The long-term effects of dam-induced displacement and resettlement are poorly understood. This paper presents a rare longitudinal assessment of the livelihood outcomes of people resettled as a result of a major dam project – the Three Gorges Project in China. It signposts pitfalls that practitioners need to be aware of when planning their own longitudinal monitoring and evaluation (M&E) projects, such as those relating to (i) sample and control group selection, (ii) deciding the timing and frequency of data collection, (iii) mitigating attrition and (iv) accounting for the broader social and policy changes that occur over longer periods when analysing results. The paper argues that longitudinal research is important not only for assessing resettlement outcomes accurately but also for ensuring that ongoing government interventions are adequate and not withdrawn prematurely.

Keywords:

• Longitudinal research
• Three Gorges Dam
• China
• dam-induced displacement and resettlement
• social impacts
• livelihoods

Introduction

Now completed, the Three Gorges Project (TGP) on the Yangtze River in China is the largest hydroelectric dam in the world in terms of installed capacity and annual average power generation volume (Chen et al. 2014). Conversely, in hydrological terms, the Three Gorges Dam is small with a storage capacity of only 9% of the mean annual flow of the upper basin, and 4% of the mean annual flow of the whole basin (Chen et al. 2014). Given the size of the reservoir – it flooded 632 km² – the impacts were extensive and presented the Chinese Government with many challenges, including the resettlement of more than 1.13 million residents. International NGOs (Human Rights in China 1998; Probe International 1998; International Rivers Network 2003) and many academics (Fearnside 1988; Chau 1995; Jun 1997; Stein 1998) anticipated severe social impacts and widespread impoverishment. The government was adamant it could successfully resettle this population by approaching the resettlement programme as a development project in its own right. As resettlement projects around the developing world still struggle to avert impoverishment, such an outcome would be significant.

The Three Gorges Project Construction Committee (the representative body of the Central Government) put in place long-term economic development plans to improve the livelihoods of the people in the Three Gorges. Under the 1993 Regulations (revised in 2001), The Regulations on Resettlement for the Construction of the Three Gorges Project on the Yangtze River, two key programmes drove the effort. The first, the Partnership Support Programme, provided funding to improve infrastructure and create employment opportunities within new enterprises. The second, the Development Assistance Fund, channelled revenue generated by the operational dam into six main areas: (1) production; (2) living standards; (3) education; (4) health services; (5) subsidies for resettled people; and (6) vulnerable groups (the poor, the elderly and the disabled). Funds were also directed into key industries such as oranges, animal husbandry, tourism and fishing.

Planning a resettlement as a development opportunity in its own right is rare and, as impoverishment is a common outcome of resettlement (Cernea 1997), the question of whether a development approach could improve livelihoods is important. Indeed, during the 1990s and early 2000s, there were a number of researchers drawn to the region to investigate the effects of resettlement. These researchers reported adverse effects for different sections of the population, including: risks of impoverishment due to a shortage of economic resources, environmental constraints and mismanagement (Wilmsen et al. 2011a); high levels of dissatisfaction among those displaced by the project (Jackson & Sleigh 2001); the feminisation of agriculture (Tan et al. 2005); and high levels of outmigration (Li et al. 2002). Moreover, in 2011, the Chinese Government made the surprising move of admitting that there were problems with the TGP resettlement and set about addressing those issues with a new working plan for post-project support (Duan & Wilmsen 2012).

After this initial flurry of research, in the English language literature, at least, there have been few studies of the longer term effects of the TGP on resettled households. Doing such long-term research is problematic as the Chinese Government’s development plans span many decades. Furthermore, given the extended upheaval caused by the construction of the project, it was important to look at longer time frames before making substantive observations about resettlement outcomes. Therefore, there was a need to follow up with the resettled households after the dam became operational in 2009. To this end, this paper presents a rare case study of a longitudinal assessment of the livelihoods of people displaced and resettled by a major dam project. In doing so, it demonstrates the importance of longitudinal resettlement research for assessing resettlement outcomes, and discusses the key decisions that need to be made when planning a research project. It should be noted, however, that this paper does not seek to provide a comprehensive review of methodological best practice in this area. Rather, it draws some practical lessons from our personal experience, whilst signposting some common pitfalls practitioners might encounter in their own work.

The paper is structured as follows. The next section provides a brief overview of how monitoring and evaluation (M&E) is conducted in China. Following this, we describe our research methods, which were designed to monitor the livelihoods of the people resettled over time. We reflect on the practical constraints of conducting a longitudinal evaluation in China whilst pointing out what practitioners should aim for under ideal circumstances. We then provide a brief overview of our empirical results, paying particular attention to how an assessment of resettlement outcomes can change dramatically depending on the choice of time frame. The final sections of the paper discuss the value of gathering longitudinal data and highlight some particular issues practitioners must consider when conducting evaluations spanning several years.

Monitoring and evaluation in China

Although M&E is a routine part of resettlement practice in China, it is short-term in scope and rarely independent. M&E begins during the project implementation phase and ends when the final financial milestone is met (Xiao & Arthur 2015). An external agency, such as a university or research institute, is often employed to conduct the M&E programme. This approach was introduced in 1991 at the Xiaolangdi Dam resettlement at the insistence of the World Bank and has since become a general requirement of Chinese resettlement practice (Habich 2016). However, as the research institute conducting the work is contracted by the implementing agency, it is not strictly independent (Xiao & Arthur 2015). This undermines the reliability of M&E reports and particularly claims of resettlement successes.

In China, a variety of methods are used to collect baseline data for planning and benchmarking. The monitoring agency usually applies a mixed-methods approach which includes detailed surveys, interviews with a range of stakeholders and document analysis (Xiao & Arthur 2015). Despite using a range of methodologies, the focus tends towards collecting data to calculate compensation entitlements rather than understanding the broader context of resettlement. Zaman and Gonnetilleke (2015, p. 172) call such approaches ‘rush in social data collection’. It is required to meet the tight time frames of project preparation, but does not capture the sociocultural milieu or the political organisation that underpins pre- and post-displacement lives (Zaman & Gonnetilleke 2015). Indicators useful for gaining a broad understanding of how people are faring through time are not a central consideration in many survey designs. Given the short time frames, lack of independence and preoccupation with counting material losses, M&E methods often lack the sensitivity to gauge whether resettlement was successful or detrimental in any meaningful way.

Designing a longitudinal evaluation of the TGP resettlement

Evaluating the TGP is a complex task, not only because of the size of the resettlement – the largest organised resettlement ever undertaken – but also because of the sensitivity of the project. Entry into the field had to be carefully negotiated with a Chinese partner university (see McDonald-Wilmsen 2009). Anthropological methods that might deepen understandings of the community were deemed too risky as they could attract undue attention from the authorities. Although the risks to researchers were minimal, we were concerned about the respondents being harassed by the authorities (McDonald-Wilmsen 2009; Wilmsen & Webber 2016). Therefore, the evaluation methods had to be designed in line with the ethics approval of La Trobe University that aims to protect research participants, whilst striving for a holistic evaluation of changing livelihoods after displacement. To this end, the aim of the research was to understand the reconstruction of livelihoods as a transformative process that extends beyond the early years of displacement. The methods described in the following sections highlight some of the complexities of conducting this research. The goal here is to forewarn practitioners of potential pitfalls and provide ideas about how they might respond when the on-the-ground reality precludes the use of ideal methods.

Our research involved a survey in 2004 and 2012 and in-depth interviews that took place between 2004 and 2016. The first point of data collection occurred soon after submergence (which reached 153 m in 2003). The 2004 survey asked displaced households to reflect on their living standards and livelihoods pre-resettlement and in 2003. Relying on memory is not ideal – respondents may be unable to recall the details of their livelihoods five (or more) years ago as reliably as the current state of their livelihoods. As such, data about pre-resettlement livelihoods cannot be considered as accurate as data pertaining to livelihoods in 2003. The second round of data collection took place in 2012. It was concerned with living standards and livelihoods in 2011. In this sense, this study can be regarded as longitudinal. The dam was officially completed in 2009, so unlike many M&E studies organised by an implementing agency, this study extended well beyond the end of the project cycle.

Optimal practice for M&E is to survey and interview households prior to displacement and resettlement. For the practitioner, baseline data collected before resettlement is crucial as it provides a benchmark against which post-resettlement changes can be quantified (Souksavath & Nakayama 2013). However, gaining site access far ahead of time is often difficult, even for M&E practitioners employed by the project proponent. Proponents often limit public disclosure in the project planning phase as the ‘very mention of a project may affect people’s actions and responses from that point onward’ (Rowan 2009, p. 187). Once a project becomes public, land speculation, influxes of people claiming rights to compensation, social unrest and resistance may ensue (Vanclay 2012). To compound matters, independent researchers are often reliant on external funding; a competitive process that may not allow for the optimal timing of data collection.

Sample selection

Before selecting a sample, practitioners need to define the affected population of the project. This requires that the direct and indirect effects of a project – and their spatial extent – be determined (Rowan 2009). The full effects of a project are often difficult to predict in advance, making the task of defining the ‘affected population’ difficult. Impacts not only occur at the scale of the individual or household, but also extend into surrounding communities and regions through economic and social connections (Vanclay 2002). Further, defining the boundaries of communities can be complex as they often overlap and merge seamlessly (Rowan 2009).

Even when geographic boundaries can be decided with some confidence, it remains a challenge to quantify the number of people to be displaced. Researchers often rely on statistics produced and provided by the state, which is often also the project proponent. For example, at the Three Gorges Dam, the number of people to be displaced was initially stated at 725,000 people based on the 1991 census. Andrews-Speed and Ma (2013) believe this was deliberate under-reporting to give the impression that the social and financial costs were lower than in actuality. At a restricted meeting of the State Council in 1992, a figure of 1,980,000 displaced people was given (Webber 2012), which is close to the 2 million predicted by Gleick (2009). The most recent figure stated by the official press agency of the Chinese Government, Xinhua News Agency, is 1.3 million (Xinhua 2016). The lesson here is that some caution is required when relying on official statistics for the purposes of planning longitudinal research.

Ideally, a researcher should select a control group against which the affected population can be compared. As with defining the affected population, identifying a reliable control group can be challenging. If the control group is too close to the dam site, questions may be raised as to whether the dam project also impacted this group through spill-over effects. If the control group is too far from the dam site, questions can be raised about whether the social, cultural and economic characteristics are similar enough to the affected population to act as a reliable control. There are no simple answers to these dilemmas, of course. The practitioner must simply be aware of these issues and take account of the on-ground complexities to minimise problems wherever they can.

In this case, in lieu of a control group, the nearest prefecture level city, Yichang city, was used as a point of comparison. The dam did not directly displace Yichang city residents, which is downstream of the construction. However, this is not to say the city was unaffected. Considering the influx of capital and skilled workers associated with the dam, Yichang residents may have experienced some economic and social flow-on effects from the dam. Nonetheless, Yichang city provided a good reference point for assessing changes in livelihood indicators observed within the sample group, such as whether income growth in the study area was exceptional relative to regional economic performance.

As a reliable sampling frame for the affected population was not available (e.g. there was no detailed list of all households displaced by the TGP), it was necessary to employ a more spatially targeted sampling strategy to assess resettlement outcomes. In our research, we were particularly interested in the effects of different levels of state-led investment and different forms of resettlement on household resettlement outcomes. Within Hubei province, two counties were selected based on their level of economic development and different levels of development intervention – Zigui and Badong (Figure 1). Within each of these counties, three sites were selected to capture different forms of resettlement using changes in a household’s registration status (hukou) – rural-to-rural resettlement, rural-to-urban resettlement and urban-to-urban resettlement. Within these six sites, only the people who had already moved or lost land were targeted.

Figure 1. Location of sample sites. Source: Mcdonald et al. (2008, p. 89).

A total of 521 households participated in the 2004 survey, which asked respondents about their livelihoods in 2003 and prior to resettlement. In 2012, we conducted a follow-up survey, which focused on livelihoods in 2011. Three hundred and fifty-one people responded to this second survey – 67% of the original sample. Both surveys were supplemented by interviews with households, local officials, county and provincial government representatives, businesses and academics. In 2014 and 2016, interviews were also conducted with central government representatives and NGOs.

Attrition

With any longitudinal study, attrition of the sample population is an issue. Some attrition is unavoidable because participants do not provide reliable contact details, decline to be involved in further research, move or pass away. Where attrition occurs, it is necessary to understand who has been missed in the subsequent surveys and to ask friends and neighbours where they are so as to establish the reasons for their absence. At the TGP, a follow-up study of those who were missing from the 2011 survey found that 55 households had moved away permanently and only returned at Spring Festival; 31 households were not at home but still resided in the same house as in 2003/2004 survey; 16 householders had passed away and other family members had moved on; 41 households provided no contact details in their 2003/2004 survey as they wished to remain anonymous; 8 households refused to participate (mostly due to illness); 8 were part of the same family and now live together and 11 were unaccounted for.

Statistical analysis can also be conducted to test whether there were any systematic causes for non-response. We conducted a regression analysis of non-respondents in 2012 based on their 2003 responses. Independent variables tested included household income (pre-resettlement and in 2003), the number of household members, household registration (hukou), average household age, education attainment, compensation received, social well-being and the level of savings. None of these variables were correlated with non-response in 2012, effectively ruling out some important explanations for observed attrition; for example, poorer households being forced to leave the area.

Conceptual framework

The sustainable livelihoods approach (SLA) provided the conceptual model used to devise the suite of survey questions. Sustainable livelihoods are, at the most basic level, the various ways in which individuals use resources in order to survive (Johnson 1997). The SLA provides a systematic understanding of the various factors that constrain or enhance livelihood opportunities and the inter-relationships between these factors (Krantz 2001). Key factors include: livelihood assets, policy and institutional processes, livelihood strategies and vulnerabilities. An analysis of these elements allows some conclusions to be made about the sustainability of individuals’ livelihoods (DfID 1999).

In the early 2000s, while designing this study, the SLA was a mainstream international development approach, endorsed by a range of agencies, NGOs and governments, including UNDP, CARE and OXFAM (NSSD 2003). It emerged during the 1990s from a range of academic literatures including participatory development, Sen’s work on capabilities and entitlements and sustainability (Small 2007; Smyth and Vanclay 2017). It was used widely, producing a thick catalogue of livelihood studies around the world. Most significantly, it was applied by the UK’s Department for International Development (DfID). The SLA was highly regarded for its people-centred approach and universality; that is, it could be adapted to different and changing contexts (Carney 1998). Moreover, it focussed on people’s strengths and potentials as well as long-term sustainability (Carney 1998). It was also comprehensible to policy-makers, and provided a mechanism through which to embed social and environmental concerns into their programmes alongside economic ones (Small 2007).

In the 2000s, the SLA fell out of favour and, by the end of the decade, it had been the subject of significant critique (de Haan 2012). It was criticised for its lack of proper incorporation of people and cultural considerations. Moreover, the various actors and power differentials were poorly represented in the framework. The architect of the SLA, Scoones (2009, p. 182), best summarised the key gaps in livelihood studies as ‘four failures to engage – with processes of economic globalisation, with debates about politics and governance, with the challenges of environmental sustainability and with fundamental transformatory shifts in rural economies’. There was disagreement too about whether the SLA’s representation of livelihoods was too simple (Morse & McNamara 2013), overly technical (Frediani 2010), or excessively complex, i.e. not compatible with real-world challenges and decision-making processes (Scoones 2009; Reddy et al. 2015). Practitioners largely took the latter view as they felt it required too many issues to be addressed simultaneously (Reddy et al. 2015).

It is important to note that there were several conceptual frameworks available for use in this study. The first widely recognised model, proposed by Scudder and Colson in 1982, observed a number of dam resettlements and used these to predict how communities, households and individuals respond to resettlement. They identified four stages of resettlement: recruitment, transition, preferential development and handing over/incorporation (Scudder & Colson 1982). The second framework is the impoverishment, risks and reconstruction model (Cernea 1997), which identified eight losses associated with displacement that lead to impoverishment: (1) landlessness; (2) joblessness; (3) homelessness; (4) marginalisation; (5) food insecurity; (6) increased morbidity and mortality; (7) loss of access to common property resources; and (8) community disarticulation. Downing and Garcia-Downing (2009) proposed the contextual theory of ‘routine and dissonance cultures’ which observed the disruption of culture caused by the reordering of space-time relationships that changes social networks and traditional coping mechanisms. These frameworks emphasise different aspects of the resettlement process – from the temporal transition (Scudder & Colson 1982), losses and risks (Cernea 1997) and cultural responses (Downing & Garcia-Downing 2009). A recent framework was developed by Smyth and Vanclay (2017). Whilst most theories describe and capture critical aspects of the resettlement process, our research was concerned with livelihoods outcomes and we thus opted for a framework best suited for framing and capturing changing livelihoods – the SLA.

In order to be implemented, a conceptual framework must be transposed into precise questions and measurable indicators suited to the local context. To this end, it is important to pilot surveys before they are widely deployed. Our chosen conceptual framework lists five livelihood outcomes – ‘more income’, ‘increased wellbeing’, ‘reduced vulnerability’, ‘improved food security’ and ‘more sustainable use of the natural resource base’. As we were still waiting for the application for a research permit to be processed, we were only allowed to complete the questionnaire with six resettled households.

The pilot study indicated that several adjustments to the questionnaire were necessary. For example, the participants identified particular complexities related to the context of their resettlement that were not captured by the SLA. The outcome, ‘more income’, for example, did not reflect the changing nature and location of employment. Some people lost their regular jobs in factories during the resettlement process as the government took the opportunity to shut down ailing state-owned enterprises as part of the broader restructuring of the state-owned sector throughout China. ‘More income’ did not provide a good description of the shift from reliable employment to temporary construction work. In addition, many households migrated to find employment, which had follow-on effects for social networks and family life. It was decided that income change needed to be considered alongside an analysis of types of work, production, location and household migration. The questionnaire was altered accordingly.

Examining the value of a longitudinal assessment of the TGP resettlement

The previous sections described our research methods and highlighted some potential pitfalls for practitioners to consider when planning a longitudinal evaluation project. In this section, we present a snapshot of the empirical results of our research, focusing in particular on how the assessment of resettlement outcomes can change dramatically depending on the timing of measurements, thereby demonstrating the value of longitudinal data for resettlement research. More detailed accounts of the empirical results from this study along with their theoretical and policy implications can be found in Mcdonald et al. (2008), Wilmsen et al. (2011a, 2011b) and Wilmsen (2016).

Income

Income restoration is a key measure of economic recovery. Table 1 presents the level of income for the three resettlement categories over three time periods. To more accurately compare real changes in income over time, income data were adjusted for inflation.

Table 1. Summary of income and income change using 2003 consumer prices.

Resettlement type	County	Average income per person (RMB, 2003 prices)				N
		Before resettlement	After resettlement (2003)	2011	% increase (2003–2011)
Rural-to-rural	Zigui	2577	1812	9783	440	31
	Badong	2352	2065	6871	233	87
Urban-to-urban	Zigui	3583	3658	11,020	201	44
	Badong	2998	2053	8288	304	73
Rural-to-urban	Zigui	2020	1574	6915	339	62
	Badong	2308	1674	6581	293	46
Total		2602	2103	7943	400	343

Notes: N = total number of households. Sample includes only households that responded to both 2003 and 2011 surveys, and filled out all three income entries.

Source: Adapted from Wilmsen (2016, p. 46).

Table 1 highlights the importance of timing and repeat monitoring beyond the project cycle. An assessment of changes in income six months after the dam was filled to 153 m reveals a decline in average incomes across all resettlement categories compared to pre-resettlement levels. At this time, the whole area was in a state of upheaval due to the recent submergence of the local area. Conversely, between 2003 and 2011 incomes grew substantially. In the rural resettlement site at Zigui, for example, real incomes increased by 440% in real terms.

For comparative purposes, Figure 2 reports incomes as a percentage of the Yichang City urban average. As discussed, it is important to have a reliable comparison group against which to scale observations. Figure 2 presents two important findings. First, the income growth figures presented in Table 1 were remarkable even by regional standards. Significant catch up with Yichang urban residents was observed for all resettlement categories. Second, despite these impressive income growth figures, the resettlement groups exhibited continued disadvantage vis-à-vis Yichang City urban residents, especially with regards to rural people who moved to urban places and those residing in Badong county.

Figure 2. Mean household income (pp) as a percentage of Yichang City urban average, by location and resettlement type, 2003 and 2011. Source: Wilmsen (2016, p. 46).

Notes: In 2003 the average disposal income per person of a Yichang urban resident was 7033RMB. In 2011, this figure was 16451 RMB. Sample includes only households that responded to both 2003 and 2011 surveys.

Average income growth, whilst a useful indicator of outcomes, can occlude what happened to individual households. In this case, for example, a handful of households may have experienced large gains and brought up the average, thus concealing increased income inequality. To address this potential issue, it is useful to track the changes in income quartiles through time. Households were assigned to quartiles based on their pre-resettlement incomes and then tracked through subsequent periods (see Figure 3). The results demonstrate that the experiences across both intervals differed significantly depending on initial household income. For example, the bottom income quartile group gained little ground on the Yichang city urban residents pre-to-post resettlement, but exhibited significant catch up by 2011. In contrast, pre-to-post resettlement, the top three income quartiles lost ground to the Yichang comparison group, before exhibiting significant catch-up through to 2011. The key lesson for practitioners, here, is the need to be sensitive to the possibility that different households, depending on their starting points can experience very different outcomes at different points in time.

Figure 3. Median quartile income as % of average income per person of Yichang urban residents.

Savings and borrowings

Incomes are just one part of the household economy. Households draw down and replenish financial assets in response to their changing circumstances, and so these provide another important perspective on how households are faring. Furthermore, high levels of savings act as an important buffer to shocks, whereas high debt levels can make households more vulnerable. To this end, the savings and borrowing of the households are presented in Table 2.

Table 2. Savings and borrowings.

	Before resettlement	2003	2011
% households with savings	40.3	14.7	26.7
% households with borrowings	29.0	60.8	66.6
Average savings (RMB)	8848	1966	4784
Average borrowings (RMB)	2685	11,556	25,917

Note:

Not adjusted for inflation. Non-respondents from 2011 survey were removed from 2003/before resettlement data. Those with zero savings were included when calculating averages.

Table 2 demonstrates the importance of digging into the detail of the livelihoods of resettled people and the changes in stock variables, such as financial assets. Despite incomes going up and income inequality declining, the proportion of households in debt in 2011 is higher than in 2003, and the amount of that debt rose significantly in the same period. The proportion of households with savings increased somewhat between 2003 and 2011, however, average savings were still half of what they were before resettlement – keeping in mind these figures were not adjusted for inflation. These figures suggest high levels of borrowing were putting significant stress on the households, potentially offsetting the significant income gains observed between 2003 and 2011.

Types of employment

While changes in household income and wealth are a key concern, the types of work people are doing provides a more detailed picture of the strategies households are deploying as they adjust to their new environments. Such information is critical for policy-makers, especially when outcomes are poor and a targeted government response is required. Further, a detailed analysis of employment and occupation type provides an indication of whether observed income and employment rates are likely to be sustainable. For example, dam construction projects often produce a spike in demand for labourers in the local economy. However, this type of work is likely to dwindle when the construction phase of a dam project is completed, leaving those resettled worse off in the long-term. In contrast, jobs created in other sectors, such as private manufacturing sector, might be considered more sustainable in the long-term.

Table 3 illustrates the fluctuation in types of employment through time. In 2003, at the height of construction, 32% of households in the study were engaged in manual labour, but this declined by 2011 with project completion. Moreover, with less land available due to inundation, and increasing urbanisation within the study area, the proportion of households engaged in farming also declined through time. Table 3 also shows an increase in the proportion of people reliant on pensions and government subsidies as their key source of income, which raises long-term livelihood concerns.

Table 3. Types of employment.

Type of employment	Before resettlement	2003	2011
Farming	55%	29%	14%
Manual labour	8%	32%	17%
Retired/pension/subsidies	1%	7%	30%
State-owned enterprise	17%	2%	1%
Small business	4%	12%	22%
N	520	520	345

Adequacy of income

While income growth is a desirable outcome, income figures do not account for increases in the cost of living. For example, farming households that lose land as a result of resettlement may be less able to produce food for their own consumption. Incomes must increase significantly in order to maintain living standards, as more food needs to be purchased in local markets. Such concerns are particularly pertinent for rural households relocated to – or encroached upon by – urban areas. With these concerns in mind, the 2011 survey asked respondents whether their incomes were sufficient to cover extra expenditures. Table 4 summarises these responses by location and resettlement type.

Table 4. Income versus expenditure, by location and resettlement type, 2011.

Location	Resettlement category	Income more than expenditure (%)	Income matches expenditure (%)	Income less than expenditure (%)
Zigui	Rural-to-rural	34	38	28
	Rural-to-urban	5	53	42
	Urban-to-urban	28	46	26
Badong	Rural-to-rural	16	39	45
	Rural-to-urban	4	37	59
	Urban-to-urban	21	37	42
Total		17	42	41

Table 4 shows that for a large proportion of households, income is less than expenditure. Reflecting differences in income, those resettled in Badong were more likely than those resettled in Zigui to report that income did not cover expenditure. People resettled from rural to urban areas were more likely to report insufficient income than rural-to-rural or urban-to-urban resettled households. This is not surprising given this rural cohort became urban residents, yet their incomes did not increase enough to match those of their urban-to-urban counterparts (see Table 1).

To further test whether incomes were sufficient, the surveys asked whether households had enough food to cover their needs. Figure 4 reports the percentage of households stating they had less food than they needed.

Figure 4. Percentage of respondents reporting that they have less food than they need by location and resettlement type. Source: Wilmsen (2016, p. 49).

In both Badong and Zigui counties and across all resettlement types, the percentage of households reporting that they did not have enough food increased dramatically after resettlement. Although the percentage of households reporting insufficient food fell significantly between 2003 and 2011, it remained much higher than pre-resettlement levels. This suggests that a significant proportion of households were still struggling to meet their basic needs. According to this measure, once more, those who moved from rural-to-urban areas fared the worst of the three resettlement categories. Again, this reflects the lower incomes of rural-to-urban households vis-à-vis the urban-to-urban households, and the capacity of rural households to produce at least some of their own food.

This data on food insecurity highlights the importance of staged monitoring for determining whether additional support is required for resettled people. The earlier that food insecurity can be identified, the sooner people can be supported by targeted programmes. In this case, the Chinese Government provided vulnerable households with food subsidies and staples in the months following inundation. That this attribute had improved by 2011 suggests some recovery in livelihoods. Nonetheless, food insecurity remained a pervasive issue for some households. This kind of longitudinal data can help justify the provision of food and income support beyond the end of the project cycle. Likewise, without longitudinal data, there is the risk that social supports and other development expenditures can be cut prematurely (i.e. before sustainable livelihoods have been re-established).

The importance of longitudinal research

In the study of involuntary resettlement, longitudinal research is lacking (Scudder 1997; Cernea 1999; Downing 2002). As Scudder (1997) noted 20 years ago, longitudinal research is essential to understanding the cumulative impacts of resettlement and it was suspected the long-term impacts of resettlement are often underestimated (Colson 1971; Scudder 1993, 1997). Our research demonstrates how resettlement outcomes can vary dramatically depending on when data is collected. The findings of the first survey revealed that resettled households were having a very difficult time in 2003, just after final inundation of their communities. For the majority of the 521 households, incomes had declined significantly. And, while some resettled people reported better housing quality overall, most had taken on considerable debt to fund housing construction and cover basic needs. Moreover, the government’s development interventions had yet to stimulate the regional economy or generate adequate employment opportunities for the majority of households (Mcdonald et al. 2008; Wilmsen et al. 2011a, 2011b). A small number of urban households in Zigui, however, had secured employment in newly established enterprises, and this was correlated with higher incomes (Mcdonald et al. 2008). A range of other studies also reported negative outcomes for other groups displaced by the Three Gorges Dam (Jun 2000; Padovani 2006; Hwang et al. 2007; Tilt 2015). These studies confirm that the early years after resettlement are a time of acute vulnerability and re-iterate the importance of beginning M&E as soon as practicable in the project cycle.

From the research undertaken in the 2000s, the Three Gorges Dam project might have been judged a failure from a resettlement perspective. But as our follow-up research demonstrated, long-term outcomes have proven to be better than expected, at least for the sampled populations. The 2011 data suggested there have been remarkable gains in the livelihoods since 2003. Incomes rose, income inequality declined and reported well-being had improved. This is good news for those who lost so much to the TGP. However, despite these improvements, a significant proportion of households still reported inadequate food and income to cover basic needs and the proportion of households in debt – and the amount of that debt – also rose significantly between 2003 and 2011. The key point here is that measured outcomes can differ dramatically depending on the choice of time frames. It takes time – perhaps years – for economies and communities to settle down after the construction and resettlement phases of a project. As such, more longitudinal research is needed to fully understand the longer term effects and outcomes of resettlement projects.

In addition to deepening our understanding of resettlement processes, longitudinal evaluation plays an important role in reflexive policy-making. Ongoing M&E can indicate whether the aims of the resettlement project are being met and whether laws and regulations are being observed (such as those requiring that people are no worse off as a result of their resettlement). The high level of government investment in the Three Gorges Dam area and the income support mechanisms implemented post-resettlement are attributable, in part, to the feedback processes generated by those early (mostly negative) evaluation studies. Our longitudinal data indicate these government programmes have buttressed incomes, but also point to possible areas of concern in the future. For example, between 2003 and 2011, there was a dramatic increase in the dependence of households on pensions and other state subsidies, while rural households were increasingly reliant on remittances from household members who had moved away to find work. By identifying these vulnerable groups, longitudinal research can help policy-makers respond in a timely and targeted fashion to problems as they unfold.

Longitudinal research is also a critical input for ongoing policy debates. For example, by 2011, those rural household who had been relocated to new farming land were faring better than those rural household moved to urban areas. This finding is relevant to ongoing policy discussions about whether it is better to compensate displaced rural households with land and cash, thereby letting people continue their rural livelihoods in some capacity, or whether it is better to relocate rural households to towns and cities, where jobs are more available. Without good longitudinal data, such debates are difficult to resolve empirically.

The challenges and pitfalls of longitudinal resettlement research

Longitudinal research presents specific challenges and pitfalls that go beyond those encountered when doing a more traditional, before-and-after-type study. Deciding upon the timing of the data collection is a particularly difficult issue. Ideally, as discussed previously, data collection should begin before resettlement and project construction begins in order to establish a reliable baseline and control for the spill over effects of the project on the local and regional economy. Where the collection of pre-resettlement data is not practicable, which is more often the case for researchers rather than practitioners, other methods are available. In our research at the Three Gorges Dam, we surveyed people’s recollections of their pre-resettlement livelihoods. Whilst not without its problems, this data provided valuable information about changes in livelihoods (sources of income, cost of living, etc.) following resettlement and then over the longer term. Where pre-resettlement data cannot be collected, additional work needs to be done, either by identifying and surveying a robust control group or by identifying other sources of data to be used for comparative purposes. In our research, we relied on census data for nearby Yichang city. The critical point, here, for practitioners is the importance of identifying early on what data sources are available.

Post-resettlement, the number of data collection points is another crucial consideration. Experienced consultants (Reddy et al. 2015) recommend that M&E be conducted monthly (internally) and quarterly (externally). However, practical constraints (financial resources, site access issue, labour time, etc.) affect the timing of post-resettlement data collection and practitioners may be forced to make compromises. Further, project proponents often have little interest in monitoring long-term resettlement outcomes, and practitioners should be prepared for this. Not only is long-term M&E expensive, it may be seen to increase the risk of complaints and compensation claims in the future. Reflecting on our research, three main data collection points – pre-resettlement, post-resettlement and then several years later – provided valuable data about how livelihoods had changed for a diverse group of resettled households. The key lesson here is that when deciding upon a data collection schedule, taking into account practical and financial constraints, priority should be given to collecting data well beyond a project’s official completion date, ideally several years.

Another complication of longitudinal research is accounting for ongoing changes in the economic and policy environment. An understanding of this changing context is important when interpreting and explaining changes in livelihoods observed over the course of several years or more. In the TGP case, an understanding of ongoing reforms and policy changes shaped the direction of the in-depth interviews in 2012, 2014 and 2016, and also the analysis of the final round of survey data. For example, land reforms that were underway in China between 2003 and 2011 led us to interview agribusinesses about their evolving contractual relationships with rural people who had been resettled. Moreover, in analysing the survey data, particular attention was paid to the changes in land use and the extent of land transfers, both formal and informal.

Matters were further complicated by the fact that longitudinal research can – and should – feed back into ongoing policy decisions. These impacts must be considered when analysing data in later periods. In the case of the TGP, the government implemented a long-term post-resettlement support programme to address the problems. Taking these developments into consideration, in our interviews and data analysis we explored whether households were becoming more dependent on various transfer payments (including pensions and remittances) and found this to be the case.

Out-migration is another problematic issue for longitudinal research, especially resettlement research spanning more than several years. When compensation or new job opportunities are inadequate – or a resettlement programme is entirely absent – many households will simply leave. As such, by the time researchers arrive, many households may have already left. This reiterates the importance of sampling as early as possible in the project cycle. Even where pre-resettlement surveying is feasible, the challenge is identifying those who left the sample due to out-migration as opposed to other common causes of attrition (e.g. not at home, refusal to participate, death of respondent). Households can be re-visited or neighbours interviewed to understand the reasons for non-response and to estimate the volume of out-migration. Local officials may also have a good understanding of the reasons driving out-migration, in particular whether people are leaving as a result of the particular project (a push factor) or because better opportunities exist elsewhere (pull factors). This leaves unanswered the question of whether out-migrants fared better in the long-term than those who stayed behind – a critical policy question. The practical lesson here is that significant advanced planning is required to ensure household members can be contacted in the years after baseline samples are collected. The simplest precaution is to collect as many varieties of contact information as possible (mobile phone numbers, emails, social media identifiers, etc.) whilst taking into consideration the usual privacy and ethical concerns. But with the increasing pervasiveness of communication technologies, even in developing countries, the task of locating out-migrants should become a less daunting task in the future than it has in the past.

Conclusion

Longitudinal research is a critical but underutilised element of resettlement research and practice. This paper demonstrated the importance of longitudinal research, especially in the years after the construction phase of a project is officially complete. Early research conducted in the area of the Three Gorges Project reported mostly negative outcomes for resettled households. This study shows that the picture had changed dramatically by 2011, at least for this study population, with significant increases observed in average incomes and a range of other livelihood indicators.

By monitoring the same households over time, intricate information about livelihood resources, adaptive strategies and outcomes can be gained. If this information is provided to a responsive and committed project proponent, targeted remedies can be implemented in a timely fashion, greatly improving outcomes for those involuntarily displaced by a project. In the Three Gorges Dam case, ongoing research helped assess the Chinese Government’s efforts to address the initial failings of their resettlement programme.

Longitudinal research spanning several years is also essential for addressing broader policy debates, such as those about the most suitable form of compensation (e.g. land vs. cash), the most effective interventions (e.g. for generating employment), and whether displaced rural households are better off being compensated with farmland or being relocated to urban centres, as is currently being discussed in China (Cernea 2016). However, longitudinal research also presents particular challenges that practitioners must be aware of when planning their research and analysing results; such as deciding the frequency and timing of data collection, taking steps in advance to minimise attrition, accounting for the feedback effects of monitoring and evaluation on government policy, and accounting for broader social changes that occur over long time periods when interpreting results.

Disclosure statement

No potential conflict of interest was reported by the authors.

Funding

This work was fully supported by the Australian Research Council [grant number DE120101037].

Acknowledgements

The author would like to acknowledge the helpful comments of three anonymous reviewers and the editors of the special edition, Professor Frank Vanclay and Professor Deanna Kemp on previous drafts and generous funding from the Australian Research Council: DE120101037.