1 Introduction

An asset-based approach to poverty analysis is relatively nascent in the development literature and grew out of the dissatisfaction with the income-/expenditure-centric approach to poverty measurement. The chief opposition to the income/expenditure approach as embodied in the Foster–Greer–Thorbecke (1984) measures is that the traditional indices, by focussing on the outcome rather than the cause, ‘cannot address the dynamic, structural and relational factors that give rise to poverty’ (Harriss 2007). The failure of well-intentioned policies in effecting a significant reduction in rural poverty in Third World countries is attributed to this ‘technology focus’ on the part of governments and development agencies which treats all poor as a homogeneous group (FAO 2005). The basic contention of the new paradigm is that household income poverty is an observed phenomenon whose incidence and character can be explained in terms of the quantum, quality and types of assets that households own. Further, income data also fail to capture the full amount of available resources as households can rely upon—productive, financial and, to an extent, even social assets—to tide over income shortfalls. Although the theoretical underpinnings of this strategy can be traced to Sen’s (1981) work on famines, entitlements, assets and capabilities and also in the writings of Chambers (1995) on risk and vulnerability, the earliest elaborate attempt to integrate assets into the poverty framework in the context of urban poverty reduction was made by Moser (1998). Subsequently, Sahn and Stifel (2000) employed asset indices for undertaking inter-temporal and inter-regional comparisons of poverty situations across Africa. The asset-based approach came into sharper focus as a result of a seminal paper by Carter and Barrett (2006) who adopted the concept of asset poverty to analyse the existence of poverty traps in post-apartheid South Africa. The asset-based approach has since then gained wide popularity in development discourse and its conceptual framework and scope has further been refined in the works of Carter and Barrett (2006), Moser and Felton (2007), Booysen et al. (2008), Liverpool and Winter-Nelson (2010), Brandolini et al. (2009) and Barrett and Carter (2013).

While asset-based poverty indices are being increasingly employed to gain deeper insights into the nature of deprivation afflicting poor households in developing countries, another strand of inter-related literature, often referred to as the livelihood framework, has examined the relation between assets and the choice of livelihood (Azzarri et.al. 2009; Winters et al. 2002; Nielsen et al. 2013). The idea espoused by the livelihood approach is that the choice of a livelihood strategy is a function of household assets and liabilities. Asset ownership influences the range of livelihood options open to households which in turn give rise to different livelihood outcomes across the economy. Hence, identification of asset-based livelihood strategies in location-specific/regional contexts and linking them to observed welfare outcomes in terms of income, food security and social claims has been the central focus of many individual and institutional research studies in recent times. However, such studies are almost non-existent in India. An attempt at decomposition of poverty into its structural and stochastic components was made by Dutta (2013) on the basis of NCAER data; however, the study stopped short of evaluating the role of assets in livelihood choices and their impact on realised incomes of households. Besides, a limitation of some of the earlier studies is that while assessing the relation between assets, livelihood and welfare indicators, these studies do not take into account the endogeneity resulting from the simultaneous nature of livelihood strategy adoption and income realisation (Winters et al. 2002). Given that households with low incomes self-select themselves into specific livelihood strategies, our study uses a combination of cluster analysis and propensity score matching methods to understand the link between assets, rural livelihood and income poverty in the distinct agro-climatic zone comprising of the Barak Valley, located in the southernmost part of the north-eastern state of Assam, India. The objectives of the paper may be stated as follows: a. To understand the nature of rural poverty in the study area in terms of its structural and stochastic components; b. to identify the livelihood strategies adopted by rural households in the region and to evaluate the role of assets in the choice of livelihood strategies and further to map poverty typologies by livelihood strategies; and c. to assess the impact of alternative livelihood strategies on income poverty status of rural households.

2 Background of South Assam

The Barak Valley, located in South Assam, India, constitutes one of the six agro-climatic zones of Assam and comprises the three contiguous districts of Cachar, Karimganj and Hailakandi (Figure 1). The region (bordering Bangladesh) is located between longitude 92° 15/E and 90° 15/E and 24° 8/N and 25° 8/N and covers a geographical area of 6922 km2. It derives its name from the Barak River that originates in the hills of Manipur and flows through the region before entering Bangladesh, where it is known as Surma. The valley is enclosed by hills on the northern, southern and eastern parts. The topography of the valley consists of hills, wet lands and plains. Nearly one-third of the geographical region is covered by dense forests. Almost the entire valley is dotted with scattered hillocks known in local parlance as ‘tillahs’. The chief commercial crop is tea, which is grown in the hill areas of the region. Tea gardens are run by the corporate sector with the help of hired labour. The staple crop in the plain region is rice, which is chiefly grown on a subsistence basis. Even in rice cultivation, the scales are tipped overwhelmingly in favour of autumn rice. The percentage of GCA devoted to rice has remained at around 93% over the period of 1991–1992 to 2015–2016, reflecting a lack of crop diversification among field crops, as compared to the rest of the state, where the share of rice in GCA is around 80%. In the absence of irrigation facilities, monocropping is a rule rather than an exception. According to the Population Census 2011, the rural work participation rate in the region is around 33% as compared to 39% for the rest of Assam and 41% for the country as a whole. However, a classification of rural workers by occupation reveals that in 2011, only 26% of workers were engaged in cultivation activities as opposed to 40% for the rest of the state. Given these peculiarities of the regional economy, it is interesting to undertake an investigation into the levels of well-being of rural households in the region and the livelihood patterns adopted by them. This study, in particular, analyses the role of assets in explaining both the character of rural poverty in the region and in the choice of the livelihood strategies. In attempting to do so, it offers a new perspective of the dynamics between assets, income and livelihood in the context of the regional economy of South Assam.

Fig. 1
figure 1

Source: Borah et al. (2016)

Location of the study area

Full size image

3 Conceptual Framework

A detailed overview of the evolution of the asset-based approach to poverty analysis is included in Carter and Barrett (2006) who asserted that the fundamental limitation of both static and dynamic approaches based on income/expenditure methods lies in their inability to distinguish between structural and stochastic poverty and the transitions therein. They therefore propose the reformulation of poverty in asset space and suggest the adoption of the asset poverty line ‘as a natural extension of the familiar flow-based concept of an expenditure or income poverty line’. An asset is generally defined as the ‘stock of financial, human, natural or social resources that can be acquired, developed, improved and transferred across generations’ (Ford Foundation 2004). Assets include both tangible and non-tangible resources and are usually classified into five categories, viz. physical, natural, human, financial and social. The amount and relative importance of each type of asset vary between communities and between wealthy and poor households within the same community. Households can be ranked in asset space using a suitable asset index. The asset poverty line is then a specific value of the asset index which is used to distinguish between the stochastic poor and the structural poor. As shown in Fig. 2, the relationship between assets and income, expenditures or any other flow measure of well-being is illustrated by the (expected) livelihood function. The asset poverty line is then simply the value of the asset index that predicts a level of well-being equal to the income poverty line u. Then, assuming for expositional purposes that the livelihood function does not change over time, in any time period, a household is stochastically poor if it holds assets worth at least A yet its realised income or expenditure falls stochastically below u. Conversely, the household is structurally poor if its stock of assets is less than A and its realised income or expenditure level falls, as expected, below u. Greater availability of panel data has also facilitated further refinements in the asset poverty framework through the introduction of the dynamic asset poverty line that can help distinguish households caught in a long-term structural poverty trap from those expected to follow an upward trajectory, that is, those who enjoy structural economic mobility. In Fig. 2, a household that moved over time from above the income poverty line to below could be said to have made a stochastic transition back to its expected status if the household’s assets still were mapped into an expected standard of living below the poverty line (shown by movement from C to \(\hat{u}\left( {A^{\prime}} \right)\). Alternatively, a household that moved from \(\hat{u}\left( {A^{\prime\prime}} \right)\) to \(\hat{u}( {A^{\prime}})\) would have made a structural transition below the poverty line due to a loss of assets from A″ to A′. Based on context and data availability, the formulation proposed by Carter et al. has been subsequently adapted by various studies to identify different poverty and also vulnerability typologies—Chiwuala et al. (2011), Nguyen et al. (2013), Cahyadi and Waibel (2016), Amare et al. (2015), Angelson et al. (2018) among others.

Fig. 2
figure 2

Source: Reproduced from Carter and Barrett (2006)

Single-period income and asset poverty lines.

Full size image

In the context of the rural economy in developing countries, household assets, livelihood and poverty status taken together constitute the overall livelihood framework (Fig. 3). Livelihood comprises capabilities, assets and activities required for a means of living. Livelihood are invariably intertwined with the asset base from which they are derived and to which they contribute. Livelihood strategies are the range and combination of activities and choices that people make in order to achieve their livelihood goals. Livelihood outcomes are what household members achieve through their livelihood strategies. Such outcomes may be positive or negative: food and income security, health, well-being asset accumulation and high status in the community are manifestations of positive livelihood outcomes. Unsuccessful outcomes include food and income insecurity, vulnerability to shocks, loss of assets and impoverishment. However, it is to be noted that the causality between the trinity of assets, livelihood and outcomes is hardly one-dimensional; rather it runs in both directions. In other words, not only do assets affect livelihood strategies and livelihood outcomes, they are themselves affected and altered by the latter. Thus, while positive livelihood outcomes facilitate asset accumulation, negative livelihood outcomes result in asset depletion. Moreover, policies and institutions as well as the vulnerability contexts related to shocks, seasonality and trend changes exert further influence the interactions within the constituents of the livelihood framework. The vulnerability contexts refer to unpredictable events such as weather-related shocks and natural calamities, pests and disease epidemics and economic shocks that can undermine livelihood and cause households to fall into poverty. Again, it is important to emphasise that a double causality exists between vulnerability context and asset ownership. On the one hand, assets determine the vulnerability contexts in which households operate as households with many assets are better able to withstand idiosyncratic and covariate shocks than households with fewer assets. On the other hand, shocks cause people to lose their assets. Within the assets–livelihood framework, policies and institutions also influence the range of livelihood options open to different categories of people besides influencing access to assets and vulnerability to shocks. An enabling policy and institutional environment makes it easier for people—poor and less poor—to gain access to assets they need for their livelihood. A disabling policy and institutional environment may discriminate against the poor, thus making it difficult for them to get access to land, livestock, capital and information (FAO 2005).

Fig. 3
figure 3

Source: FAO (2005)

The livelihood framework.

Full size image

4 Data and Methods

4.1 Data

The study is based on primary data obtained from 394 rural households across 18 villages in the region. The administration of the Barak Valley is organised in three districts comprising of Cachar (which is the largest district in terms of both land area and population) and the two smaller districts of Karimganj and Hailakandi. The districts are further subdivided into development blocks, with 15 blocks in Cachar, 7 in Karimganj and 5 in Hailakandi. For the selection of the sample, a multistage stratified mixed sampling technique was adopted. In the first stage, seven blocks were randomly selected from the three districts on a proportionate basis in the ratio 4:2:1. In the second stage, villages were purposively selected on the basis of their distance from the nearest urban centre to get a representative sample. Further, three villages were selected from the blocks having larger geographical area and two villages from the relatively smaller blocks. This yielded a sample of 18 villages. In the third stage, 10% of the households in the selected villages were randomly surveyed to get the final sample. The field survey was conducted from January 2016 to June 2016.

4.2 Econometric Specifications and Definitions

A convenient and often adopted approach for integrating assets into welfare analysis is the construction of asset indices for households. An asset index is a composite indicator such that the underlying indicators on which it is based reflect a household’s ownership (or lack thereof) of a range of assets. The use of asset indices in development studies has become increasingly popular due to the availability of data on asset ownership from various demographic and health surveys across countries. However, a considerable amount of debate exists on (1) the assets to be included in the index and (2) the method of compilation (Johnston and Abreu 2013). Based on a review of extant literature (Moser and Felton (2007); Liverpool and Winter-Nelson (2010); Johnston and Abreu (2013) and Damien 2011), 29 variables, whose definitions and indicators are outlined in Table 1, were employed for construction of the household asset index using principal component analysis (PCA). The choice of PCA was dictated by the fact that certain variables used for the compilation of the index were continuous while others were binary or categorical. However, the village infrastructure index, used subsequently for regression analysis, was compiled using multiple correspondence analysis (MCA), as all the indicators were binaryFootnote 1.

Table 1 Components of household asset index
Full size table

Evaluating the nature of poverty in terms of its structural and stochastic components requires the fixation of both income poverty and asset poverty lines. In India, the poverty line is interpreted in terms of the income/expenditure required to access a pre-defined basket of goods and services which are considered essential for sustenance. A household that fails to meet the minimum expenditure on these goods and services on a monthly basis is categorised as poor. The poverty lines are periodically revised by the government agencies (formerly Planning Commission and now NITI Ayog) for all the states in the Indian union on the basis of nationwide surveys on consumption expenditure conducted by the National Sample Survey Organisation (NSSO). Given the differences in cost of living between rural and urban areas, the income/expenditure poverty line is given separately for rural and urban areas in each state. The last official estimate of the poverty lines for Indian states based on the recommendations of the Expert Group was announced by the Planning Commission in 2011–2012. Since the survey for the present study was conducted in 2016, the poverty line (z) of INR 1066.66 per capita for rural Assam was revised using the consumer price index (CPI) for rural areasFootnote 2. Thus, a rural household with a monthly per capita income (MPCI) of less than Rs. 1166.97 was designated as poor. Further, for the choice of the asset poverty line (A), on the basis of a survey of the relevant literature in this field, the methodology proposed by Sahn and Stifel (2000) was adopted whereby households with asset index scores(Ai) of less than the lowest 40th percentile of the asset index distribution were considered as asset poor. Subsequently, using standard terminology in the literature, the households were grouped into four categories, each corresponding to a specific poverty typology, as given in Table 2.

Table 2 Poverty typologies and definitions
Full size table

For the second objective, i.e. evaluating the relation between assets and the choice of livelihood strategies, the households are classified into six livelihood clusters according to their sources of income, using k-means clustering technique. Here, cluster analysis is a data reduction technique that allows a relatively large number of sample observations to be assigned to a smaller number of groups or clusters on the basis of some common latent factors. This involves random assignment of observations to each of the k clusters and then reassigning them through an iterative procedure so that within-cluster variation is minimised and between-cluster variation is maximised. Convergence is achieved when any further relocation of observations among groups would increase within-cluster variance. However, the central issue in k-means clustering is determining the optimal number of clusters to which observations have to be assigned. A widely practised rule of thumb consists in applying the ‘elbow method’ which looks at the percentage of variation explained as a function of the number of clusters. Thus, working simultaneously with a number of predetermined clusters, the optimal number of clusters was determined at the point where further addition of a cluster did not significantly lower the total within-cluster sums of squares. Subsequently, logit regression was employed to ascertain how asset ownership/access influences the choice of livelihood strategies.

Following the identification of livelihood strategies, the impact of alternative livelihood strategies on household welfare has been assessed using treatment effect model (TEM). The use of TEM is dictated by the fact that the choice of the livelihood strategy by a household is critically linked to the level of assets owned by it. In other words, a significant entry barrier exists for households when it comes to adopting specific livelihood strategies. Hence, in order to evaluate the impact of a given livelihood strategy on household income status, a balance has to be achieved with regard to the asset base of households and their demographic and social characteristics. This is done through inverse probability weighting (IPW) within the framework of TEM. The variables on which the treatment and the control groups were matched along with suitable robustness tests have been mentioned in suitable places.

5 Findings and Discussion

5.1 Mapping Poverty Typologies in the Study Region

As the present study uses a combination of both income poverty and asset poverty measures to understand the nature of deprivation faced by rural households, it is intuitive to begin by assessing how households fare independently on these two dimensions. Classifying sample households only on the basis of income poverty, temporarily sidelining asset poverty, reveals that out of the 394 households, 39.59% were below the income poverty line (BPL). However, reclassifying the households on the basis of the asset poverty line, the proportion of households that were asset poor was found to be 40.10%Footnote 3. Though the estimates of income poverty and asset poverty appear similar, there are vital differences in the two estimates. This is because of the fact that households that are income poor are not necessarily asset poor and vice versa. An overview of the poverty situation prevalent in the study region is shown in Table 3. It is observed that out of the 394 sample households, 18.53% are categorised as structural-chronic poor; in other words, these households are afflicted with both asset poverty and income poverty. Further, 21.57% of the households are classified as structural-transient poor, implying that these households are asset poor but not income poor. Likewise, 21.07% of the sample households are stochastic-transient poor, which means that although they are not asset poor their realised income falls below the specified income poverty line. Lastly, 38.83% of 394 rural households surveyed are never poor, indicating that their assets as captured by the asset index as well as realised incomes are both above the cut-off for asset poverty and income poverty. Here, the situation of stochastic-transient poverty can be attributed to negative shocks such as death, disease, family break-up, increase in input prices and decrease in output prices. In contrast, structural-transient poverty may be the outcome of some positive shocks such as implementation of Mahatma Gandhi National Rural Employment Schemes and availability of work in local construction projects which enables asset-poor rural households to support their livelihood through participation in these schemes/projects, thereby lifting them temporarily out of the income poverty trap. Thus, if these positive shocks to income were withdrawn, there is every possibility that these households would slip into income poverty. Lastly, the structural-chronic poor comprise the most vulnerable group as they are the severely deprived. Thus, considering both structural and income poverty, it is observed that the vulnerability among the sample households is quite high given that about 61% of the households are exposed to some kind of vulnerability or the other. However, as the nature of deprivation is different in different households, the policy interventions and institutional approach towards overcoming these vulnerabilities would also have to be different. This is in fact the central argument around which the present study is developed.

Table 3 Sample households by poverty typologies.
Full size table

Table 4 provides a deeper understanding of the income poverty situation of the sample households. Thus, among the non-BPL households, only 64.29% are above the risk of poverty as their asset holdings are above the asset poverty line. However, more than one-third of these households, though not income poor, are essentially asset poor which makes them vulnerable to income poverty shocks in future. Expectedly, for the non-BPL households, both stochastic-transient poverty and structural-chronic poverty are non-existent. As for the BPL households, a larger share of the income shortfalls (53.21%) can be attributed to negative shocks to the income stream than the lack of assets (46.79%). This finding is important in itself as it highlights that the stochastic element in income poverty is quite significant and needs to be tackled differently from structural poverty. Again, for the category of BPL households, the proportions of never-poor and structural-transient-poor households are zero, as per the definitions used for the classifications.

Table 4 Income poverty by asset structure.
Full size table

Having taken stock of the poverty situation in the study region on the basis of its structural and stochastic components, it is pertinent to ask how asset poverty and income poverty correlate at the micro-level. As illustrated in Fig. 2, in a dynamic sense the relation between assets and income is two way. This is because while the stock of assets owned by households determines their income flows, the level of income can itself alter (add or reduce) the tangible and intangible assets at the disposal of a household. Given the duality of this relation, we refrain from undertaking a regression analysis of assets on income. Instead, Spearman’s rank correlation test is employed for determining the strength of the association between household assets (captured by the asset index) and realised income (measured in terms of MPCI). The value of Spearman’s rho is positive and highly significant, which reveals that households with a higher rank in the asset space also rank high in the income space; in other words, assets and income have a strong positive correlation at the household level.

Table 5 depicts the ownership pattern of a dozen selected assets across the four pre-defined poverty groups in the study area. A perusal of the figures presented in the table enables us to make three succinct observations. First, significant variations exist in the ownership of assets among the four poverty groups as reflected by the highly significant value of the Chi-square statistic for all assets. Second, inequality in asset ownership is more pronounced in the case of certain assets while they appear less glaring for others. In particular, very high rates of deprivation are observed in access to basic services such as electricity for asset-poor households. Thus, while two-thirds of households belonging to the never-poor and stochastic-transient households had electricity connection, in the case of the structural-transient and structural-chronic households, the figures were less than 10%. Here, all the 18 villages in the sample had electricity connections which points to the inability of the structural-poor households to avail of such services despite their availability at the local level. A similar kind of inequity is observed in access to communication facilities measured in terms of access to mobile telephones and also high school education. On the other hand, disparities are less pronounced in terms of ownership of bank/post office accounts which may be attributed to the Jan Dhan Yojana, a programme of financial inclusion implemented by the Government of India on a mission mode since 2015. Disparities also appear to be less pronounced in terms of ownership of agricultural land although the average size of land owned falls monotonically from .45 to .31 hectares, 0.27 hectares to only .15 hectares for never-poor, stochastic-transient, structural-transient and structural-chronic households. Likewise, given the government’s efforts in improving the access to and quality of housing in rural areas under the Indira Awas Yojana, nearly 59% of the structural-transient poor and 69% of the structural-chronic poor had access to improved/finished roofs for their dwelling units. This demonstrates how institutional delivery of services can serve to ameliorate poverty and reduce both the inequity and the extent of deprivation. Third, the variation in asset ownership among the poverty groups also demonstrates the success of the asset index in identifying the poverty typologies and segregating households according into various groups, thereby throwing significant light on the character of poverty afflicting rural households in the study region.

Table 5 Asset ownership by poverty groups (percentage of households).
Full size table

The composition of rural poverty by social class is demonstrated in Table 6. It is found that Tea Tribes (TT) constitute the most vulnerable group in the region as more than 90% of these households were found to be afflicted with one type of vulnerability or the other. Only a little over 9% of the TT households belong to the never-poor category compared to 48.4% of General (GEN) category households, 33.78 for Scheduled Caste (SC) households, 42.86% for Scheduled Tribe (ST) households and 35.53% for Other Backward Caste (OBC) households. TT households were found to be particularly deprived in terms of access to human capital. Only 4.78% of adult family members had completed at least 10 years of school education as compared to 36.17% for the general-category households. Moreover, working members of TT households were employed in tea gardens on a contractual basis at low wage rates and suffered fairly long spells of seasonal unemployment. All these factors explain why vulnerability is so high among this group of households. Interestingly, the proportion of ST households that are structural-transient poor is very high at 50%. This implies that although these households are asset poor, they are not income poor. An explanation for this outcome lies in the fact that several households who come under this poverty configuration were found to be self-employed in skilled and semi-skilled activities such as drivers, plumbers, masons and carpenters. As subsequent analysis will show, both these types of livelihood strategies have an alleviating effect on income poverty. Vulnerability was observed to be quite high among SC and OBC households at 66.21 and 64.47%, respectively.

Table 6 Poverty typologies by social groups.
Full size table

5.2 Assets, livelihood and Welfare Outcomes

The sample survey yielded detailed information on assets, income structure, demographic and other characteristics of the 394 sample households. However, on account of missing observations on certain variables, three households were excluded from the livelihood analysis. Thus, the cluster analysis for identification of livelihood strategies was carried out for 391 households. The raw data yielded information on 17 different income sources for the study households. However, the cluster analysis of the sample households identified six distinct livelihood strategies. Here, the choice of the number of clusters in k-means clustering is somewhat arbitrary and often depends on the judgement of the investigator. On the basis of the elbow plot (Fig. 4), we negotiated between five and six income clusters for the sample households. As the ratio of between-cluster sum of squares to within-cluster sum of squares for six clusters (75.4%) was higher than that for five clusters (67.6%), the income data were grouped into six clusters, viz. (1) salaried employment in the private sector (PVT); (2) mixed livelihood (cultivators, income from government jobs, household industry and trade and business (MXDIN); (3) agricultural labour (AGRILAB); (4) self-employed in skilled and semi-skilled activities (SSS);Footnote 4 (5) Non-agricultural Labour (NONAGRILAB); and (6) Elementary Non-farm Occupations (ENF)Footnote 5.

Fig. 4
figure 4

Source: Primary data

‘Elbow Plot’ of livelihood clusters.

Full size image

Table 7 shows the composition of income across the six livelihood clusters. Thus, 75% of the income of the households belonging to PVT is derived from salaried employment in the private sector. The share of income derived from other individual sources of income for this cluster is 5% or less. The households belonging to MXDIN derive 19% of their income from government salaries, 18% from cultivation, 17% from self-employment in trade and business, 13% from household industry, 11% from pensions and the remainder from other sources. This cluster has therefore been named as the mixed-income cluster. Although households in this cluster have varied income sources, functionally they are distinct from the other clusters as none of the other clusters have an income structure that is similar to this group. Households in the third cluster, viz. AGRILAB, derive 82% of their income from agricultural labour. Hence, they have been categorised as agricultural labour households. Likewise, households belonging to the fourth cluster (SSS) derive 76% of their income from employment in skilled and semi-skilled jobs. In the fifth cluster, viz. NONAGRILAB, income from non-agricultural wage labour is the dominant income source as 84% of the income is obtained from casual wage-based labour in the non-agricultural sector. The last cluster has been labelled as ENF as nearly 70% of the income is procured from these occupations.

Table 7 Composition of income across livelihood clusters.
Full size table

The distribution of households across the six income clusters is given in Table 8. It is observed that agricultural labour is the most dominant livelihood strategy in the study region, as one-fourth of the total sample households are found to be pursuing this strategy. This is followed by casual wage-based employment in the non-agricultural sector with nearly 20% of the households belonging to this category. Nearly 19% of the sample households are found to be adopting the mixed-income strategy. A little more than 15% of the households derive their incomes predominantly from skilled and semi-skilled employment, while around 10% of the households realise the major share of their income from salaried jobs in the private sector. Elementary non-farm occupations constitute the livelihood strategy of the remaining 10.74% of the rural households. The predominance of casual wage-based livelihood in the form of agricultural and non-agricultural wage labour may be taken as an indication of rural distress in this remote southern region of Assam.

Table 8 Distribution of sample households by income clusters.
Full size table

Table 9 profiles livelihood strategies of rural households in the region by the size class of land holding. It is observed that landless rural households predominantly are involved in NONAGRILAB and AGRILAB. Besides, more than 20% of the landless households pursue SSS as a livelihood strategy, indicating that these households have substituted the lack of this vital rural asset by the acquisition of various skills. In fact, the proportion of households adopting this strategy is the highest for this group among all land-owning classes. A disproportionately large share of marginal landowning households are found to be pursuing NONAGRILAB, which reveals that meagre earnings from land force them to supplement their earnings by participating in casual wage-based labour in the non-agricultural sector. For small and semi-medium landowners, AGRILAB is the dominant strategy. Lastly, for medium and large landowners, the most commonly pursued strategy is MXDIN which, among other things, also includes income from cultivation. This is a pointer to the fact that relatively richer households follow a diversified pattern of livelihood by combining cultivation with trade, business, household industry or even public sector jobs. To test the robustness of this conclusion, a Simpson diversification index (SDI) was computed to measure the extent of livelihood diversification among households, by taking into account the percentage of income derived from individual sources (without reference to the results of the cluster analysis). The average value of the SDI was found to be lowest for landless households, while it was quite large for medium and large landowning households. Thus, the results obtained from the cluster analysis are substantiated by the SDI.

Table 9 Livelihood strategies by the size class of land owned.
Full size table

We next focus our attention on how possession of productive assets, both tangible and intangible, influences the adoption of livelihood strategies in the study area. Table 10 reports the coefficients of the logit regression which shows how the log-odds of selection of livelihood strategies are affected by households’ ownership of assets. For ease of exposition, the role of specific assets in influencing the choice of livelihood strategies is summarised as follows:

Table 10 Household assets and choice of livelihood strategy: logit results.
Full size table
  1. a)

    Natural capital It is observed from the table that an increase in per capita ownership of agricultural land (PCLANDO) significantly increases the log-odds of participating in MXDIN. On the contrary, as per capita availability of agricultural land increases, the probability of participating in AGRILAB, SSS, NONAGRILAB and ENF declines. This is indicative of the fact that rural households are pushed towards the adoption of these strategies due to inadequate opportunities in agriculture. In the case of PVT and ENF, this asset has not been found to be significant.

  2. b)

    Human Capital Four indicators have been used in the regression to capture the human capital base of rural households, viz. (1) percentage of household members aged over 15 who have studied up to the primary level (EDUPS); (2) percentage of household members aged over 15 who have studied up to middle school (EDUMS); (3) percentage of household members aged over 15 who have studied up to high school (EDUHS); (4) percentage of household members aged over 15 who have studied up to senior secondary level and above (EDUSS). It is observed that an improvement in the human capital base of households increases the probability of participation in PVT as well as MXDIN. In contrast, the log-odds of a household adopting AGRILAB, SSS and NONAGRILAB decline with an increase in human capital assets. Interestingly, the probability of households adopting ENF is also significantly enhanced with an increase in educational base. This is explained by the fact that the educated population in rural areas is wary of participation in livelihood involving strenuous physical labour and prefers to pursue less productive elementary activities in the non-farm sector in the absence of better opportunities elsewhere.

  3. c)

    Social Capital Social capital in the idiosyncratic sense is captured by membership in SHG (MEM) and receipt of remittances from kin residing elsewhere (REMIT), while covariate social capital is measured by the village-level infrastructure index (INFRA). It is interesting to note that membership in SHG and other community-level associations increases the likelihood of AGRILAB being adopted as a livelihood strategy while reduces the probability of choosing NONAGRILAB. This situation is perhaps explained by the fact that AGRILAB cluster is dominated by tea tribe households who are members of cooperatives for tea garden workers. Since the MXDIN cluster also includes trade and business, a positive and significant impact of community membership is observed on the probability of participation in this strategy. Adoption of SSS is positively impacted by a rise in INFRA, indicating that these activities are more likely to be found in relatively developed areas. Incidentally, adoption of AGRILAB also increases with an increase in the infrastructure index. This is again explained by the fact that tea-growing areas are endowed with relatively better infrastructure to facilitate production.

  4. d)

    Financial capital Financial capital, measured by access to financial services, increases the probability of participation in PVT, MXDIN and ENF and reduces likelihood of participation in NONAGRILAB. Ownership of jewellery reduces the probability of participation in AGRILAB and increases the probability of participation in MXDIN.

  5. e)

    Household demographic and other characteristics An increase in the age of household head is found to be associated with adoption of AGRILAB as a livelihood strategy while reducing the likelihood of participation in NONAGRILAB. Likewise, if the household head is male, it significantly increases the likelihood of adoption of AGRILAB and ENF. SC households are less likely to adopt PVT, MXDIN and ENF but are more likely to adopt SSS. OBC households are more likely to adopt NONAGRILAB. An increase in the household size (HHS) reduces the probability of a household engaging in AGRILAB, while an increase in the number of adult males significantly raises the likelihood of PVT.

While asset ownership/access has a significant role in determining the choice of livelihood strategies, it is also relevant to examine the welfare outcomes realised from such choices. The average treatment effect (ATE) of participation in a given livelihood clusters on income status of rural households in the study region is given in Table 11. While estimating the impact of livelihood strategies on the income poverty of the sample households, BPL status is considered as the outcome variable. Age of household head, sex of household head, number of adult males in the household, infrastructure, caste and assets are employed as treatment-independent variables. Here, the composite asset index of households, rather than individual assets, was used along with other covariates for the regression.

Table 11 Livelihood strategies and income poverty—average treatment effect.
Full size table

It is observed from Table 11 that on average, a household receiving Treat1 (PVT) has 21% less probability of being below the poverty line compared to households belonging to other livelihood clusters. Likewise, the households that belong to the MXDIN cluster are also found to have lower probability (17%) of falling in the poverty trap. The size of ATE is somewhat lower in the case of Treatment 4 (SSS); however, its poverty-reducing effect has been found to be significant. Further, those households who are pursuing ENF as livelihood strategies have 25% less possibility to fall in income poverty compared to households that do not receive this treatment. On the contrary, households that pursue AGRILAB and NONAGRILAB as their livelihood strategies have a significantly higher probability of falling below the poverty line. In the case of the former, the likelihood of poverty increases by 20.1%, while for the latter, the corresponding probability is 26.9%.

The reliability of the estimates obtained through IPW depends on how well the weighting mechanism has succeeded in balancing the baseline covariates of sample households. A post-estimation test of covariate balance was conducted to check the robustness of the ATE coefficient. In this context, the overid test has been used using STATA 15.0. None of the values of the overid test reported in Table 11 has been found to be statistically significant. This proves that the IPW mechanism of the TEM was able to effectively iron out the differences in household background characteristics and that the estimates obtained in Table 11 are robust. The summary of the treatment effect estimates of the impact of various livelihood strategies on income poverty status of the rural households is given in Table 12.

Table 12 Summary table of treatment effect estimates.
Full size table

The findings of the regression analysis are corroborated by the information conveyed by the box plots of MPCI and asset index (segregated by livelihood clusters) shown in Figs. 5 and 6, respectively. It is evident from Fig. 5 that the median MPCI (shown by the solid horizontal line inside each box) is the lowest for AGRILAB and NONAGRILAB. The median MPCI of ENF, PVT and MXDIN is higher than that of AGRILAB, NONAGRILAB and SSS. Interestingly, the latter three groups also have a lower variation in median. Further, these households also have a lower median value of the asset index, thereby confirming the positive association between assets and income.

Fig. 5
figure 5

Source: Primary data

Box plot of MPCI by livelihood clusters.

Full size image
Fig. 6
figure 6

Source: Primary data

Box plot of asset index by livelihood clusters.

Full size image

Table 13 presents a cross-classification of livelihood strategies by poverty typologies. It is observed that structural poverty is low among households pursuing PVT, MIXIN and ENF, whereas it is remarkably high in the case of AGRILAB and NONAGRILAB. This also translates into very high rates of income poverty rates for the latter group of households as 51% of the AGRILAB and nearly 58% of the NONAGRILAB households were found to be living below the income poverty line. Combining both structural and income poverty, we find that the overall vulnerability for these households is above 80%. Hence, specific attention is required to improving the asset base of these households through carefully designed skill development and other programmes. This is particularly important in view of the fact that nearly 46% of the rural households in the sample are found to be pursuing these two livelihood strategies. Therefore, if the sample can be taken as an adequate representation of the entire population, it can be argued that reducing the levels of structural and income poverty for these households would also make a significant impact upon the overall poverty situation in the region.

Table 13 Poverty typology by income clusters
Full size table

Lastly, Table 14 shows the distribution of livelihood strategies by household caste. It is evident that unlike other groups which derive their livelihood from various clusters, TT households exclusively belong to AGRILAB cluster. This is expected given that the major source of employment for this group of households is in the tea gardens. Over one-fourth of the general-category households belong to MXDIN cluster, while the largest proportions of SC and ST households participate in SSS. OBC households, on the other hand, belong largely to the NONAGRILAB and AGRILAB clusters. When these caste-wise livelihood strategies are examined in the light of the conclusions presented in Table 12, they help to explain the caste-wise composition of poverty presented in Table 6 of the preceding section.

Table 14 Livelihood strategies by social groups.
Full size table

In the end, interaction among households through asset transfer primarily in the form of land leasing is also likely to have repercussions for rural livelihood. A scrutiny of the primary data revealed that only 40.74% of the rural households in the region which owned agricultural land leased out their land either partly or fully. However, a classification of households by the size class of land owned showed that 63.44% of the households who belong to the category of semi-medium landowners leased out their land to other households. For medium and small landowners, the corresponding percentages were 16.67 and 14.29, respectively. About 40% of the marginal landowners leased out their land, while for small landowners, the figure stood at 27.91%. Thus, households with semi-medium landholdings were found to be leasing out land in larger numbers. On the flip side, among the households who leased in agricultural land for cultivation in the past 1 year, nearly 43% belonged to the category of landless households. The corresponding percentages for marginal, small, semi-medium and medium landowning households were 7.41, 22.22, 12.96 and 14.81, respectively. Interestingly, none of the large landowning households were found to be leasing in land. However, nearly 42% of the landless households who leased in agricultural land were found to be following either AGRILAB or NONAGRILAB as their livelihood strategies, implying that income earned from cultivation was not sufficient to sustain their consumption requirements. It also follows that medium and large landowners do not engage in the land-leasing market in large numbers and prefer to cultivate their own lands. However, the Pearson correlation coefficient between area cultivated and cost of hired labour at the household level was found to be 0.32 (positive and significant at 1% level of significance), implying that cultivation activities by households that possess and cultivate large tracts of land open up employment opportunities for households that follow agricultural labour as a livelihood strategy. Thus, both land leasing and expenditure on hired agricultural labour are manifestation as to how households interact among themselves to influence livelihood at the local level.

6 Summary and Conclusion

The study has undertaken a comprehensive view of the nature of poverty, livelihood choices and household welfare within a single framework on the basis of a fairly large sample of rural households. It has been found that overall vulnerability (in terms of income poverty, asset poverty or both) is very high in the study area as more than 60% of the rural households were afflicted with one type of vulnerability or the other. However, these households do not constitute a homogeneous group as the nature of vulnerability is diverse and consists of structural-chronic, structural-transient-poor and stochastic-transient-poor groups. Further, it is observed that among the income poor, the incidence of stochastic-transient poverty is quite high, implying that random shocks tend to push those that are not asset poor into income poverty. Therefore, designing programmes that would help households absorb/tide over unforeseen shocks would reduce the incidence of income poverty in the study region. The study also identified casual wage-based livelihood to be dominant among the rural households, which can be taken as a sign of rural distress. Needless to say that these households turn out not only to be asset deprived but a disproportionately large proportion of households that follow agricultural and non-agricultural wage-based labour as their livelihood strategies are also found to be income poor. The findings clearly indicate that there are strong entry barriers to the adoption of relatively lucrative livelihood and these barriers invariably relate to the nature of assets that households own. Interestingly, many asset-poor households have been found to have escaped income poverty by engaging in skilled and semi-skilled occupations. Land-rich households, rather than following cultivation as a livelihood strategy, are found to be adopting diversified livelihood strategies by combining crop production with trade and business; these households also have an edge when it comes to securing government jobs, probably because of access to education and also, information reinforced by their stronger asset base. The structurally poor households, on the other hand, are forced to adopt vulnerable livelihood strategies which ultimately results in income poverty. An interesting finding from the study is that even elementary non-farm activities have a significant poverty-reducing effect. This can be explained by the fact that despite income earned from these activities being low, there are lesser seasonal swings, thereby resulting in a steady flow of income throughout the year. Besides, as observed from the logit regression, improvement in the human capital base of households (in terms of educational accomplishments) enhances the likelihood of adoption of ENF as a livelihood strategy. The situation can be interpreted in terms of the unwillingness of the educated workforce to undertake strenuous manual labour and in the absence of better opportunities in rural areas, perhaps prefer to pursue self-employment activities, which, although less remunerative, are less demanding in terms of physical effort. The study underscores the fact that anti-poverty initiatives should be structured away from the mere identification of the income poor to capturing the fundamental nature of poverty in terms of asset ownership and livelihood strategies. In India, the socioeconomic caste census was conducted in 2011 to generate data on the income composition and asset distribution of households. However, despite data availability, so far no effort is discernible to identify households by their poverty typologies at the micro-level. Profiling poverty in terms of its stochastic and structural components would, thus, pave the way for designing well-structured policy interventions, maximise welfare gains and result in greater efficiency in public spending on anti-poverty programmes in terms of impact at the grass roots. Repeated censuses would also generate valuable panel data on household income and assets that would pave the way for more advanced types of analysis such as identification of poverty traps and 'Micawber' thresholds.