The Role of Employment Diversification in Reducing Vulnerability to Poverty among Marginal and Small-holder Agricultural Households in India

Abstract

This article combines two relatively nascent and inter-related approaches to poverty analysis and measurement, that is, the asset-based approach and the vulnerability approach to assess the poverty status of various land-owning classes in rural India. Contingent on the finding that marginal and small-holder households constitute a high-risk group in terms of the incidence of current poverty and vulnerability to future poverty, the study explores the role of non-agricultural activity in providing livelihood security and tackling poverty and vulnerability among land-poor rural households. The findings reveal that while most types of rural non-farm employment have significant poverty-reducing effects, human capital constitutes the most potent element in tackling poverty in the target group. We therefore suggest that an emphasis on skill upgradation of marginal and small landholders, coupled with policies directed towards development of non-farm activity, could provide an effective, permanent solution for curbing poverty and mitigating livelihood risks among these households.

JEL Classification: I31, I32, J21, O18

Keywords

Asset Poverty Vulnerability Rural Nonfarm Employment India

1. Introduction

The approach to poverty analysis and measurement has undergone a significant change in recent decades. The theoretical underpinnings of this new outlook on poverty can be traced to Sen’s (1981) seminal work on famines and entitlements, assets and capabilities, as well as Chambers’ (1995) more recent work. Poverty measurement has since moved away from income/consumption-centric approaches towards multidimensional and, more recently, asset-based approaches (Carter & May, 1999). The chief objection by proponents of the asset-based approach against the staple Foster-Greer-Thorbecke (FGT) class of poverty measures is that while the FGT poverty metrics is a powerful mechanism for assessing the extent, depth and severity of poverty, it fails to draw the fundamental distinction between different types of poverty situations, that is, structural poverty and stochastic poverty. They point out that the structural-stochastic distinction is fundamental to the understanding of poverty traps (Carter & Barrett, 2006): the structurally poor lack sufficient assets to generate expected income or expenditure above the poverty line because of random shocks, even if their observed income is above the poverty line (Barret et al., 2006); the stochastically poor on the other hand have sufficient assets to yield an expected non-poor income, but their observed income is below the poverty line, which can again be attributed to random events. The asset-based approach therefore establishes a functional relationship between assets and welfare indicators such as income.

Another strand of inter-related literature has drawn a distinction between current poverty and expected poverty (Chiwaula et al., 2011). While current poverty is essentially a static concept and identifies the currently poor, expected poverty or vulnerability represents a dynamic/forward-looking approach to poverty analysis since a household that is presently non-poor may still run the risk of falling into poverty at a future date, in the event of idiosyncratic and/or covariate shocks. A comprehensive approach to poverty analysis should essentially incorporate both the static and dynamic approaches to poverty, for the design of effective poverty alleviation strategies. Here, it is relevant to emphasise that there is a strong interconnection between the asset-based approach and the vulnerability approach to poverty analysis as household assets have an important role to play not only in determining the nature of current poverty but also the vulnerability to poverty (Echevin, 2011). It is this interrelation between assets, poverty and vulnerability which forms the theoretical foundations for the present study.

Considering agricultural land as the principal physical asset in rural areas, this article assesses how the incidence of poverty and vulnerability varies across different landowning classes in India. Land records reveal a continuous decline in the size of ownership holdings across the country and the proliferation of marginal land holdings. Drawing a cue from the asset-based approach, the persistent decline in the size of ownership holdings can be taken as constituting erosion in the asset base of households, which impacts their present and future welfare. Against this backdrop, this study investigates the role of non-farm employment in mitigating livelihood risks in terms of current and expected poverty among small and marginal landholders in India (henceforth collectively referred to as smallholders), given that such holdings constitute the bulk of the total landholdings in the country, both in terms of owned area and operational holdings. The non-agricultural sector in rural areas is an amalgamation of heterogeneous and disparate activities, so a blanket analysis of the linkages between non-agricultural employment, poverty and vulnerability is likely to be of limited relevance. This article therefore undertakes an in-depth analysis by assessing the role of different types of non-agricultural employment in combating poverty and vulnerability among small-holder households.

Drawing from the latest NSSO¹ survey on Employment and Unemployment (2011–12), this study constructs a vulnerability score for small-holder households based on observed household and socio-economic characteristics; subsequently descriptive statistics and econometric techniques are employed to ascertain the impact of employment diversification on household poverty and vulnerability. The format of the article is as follows. Section 2 contextualises the subject under investigation by giving an overview of the changes in size structure and ownership pattern of landholdings in India. Section 3 introduces the data base and outlines the definitions and analytical tools applied. Section 4 contains the main analytical findings and examines the role of subsidiary occupations in combating risks and vulnerability among smallholders and the final section concludes by drawing implications for policy.

2. Changing Landowning Patterns in India and Implications for Rural Livelihoods

Land data in India are available from two principal sources. The quinquennial Agricultural Censuses conducted by the Ministry of Agriculture provide the most comprehensive information about the changing dynamics of agricultural landholdings in the country. However, a drawback of these censuses is that the segregation of owned area from operated area is somewhat obscure, which makes it difficult to construct a clear picture of landowning pattern in the countryside and changes therein (Government of India, 2014). The second source of information on the structure of landholdings in India is the Land and Livestock Holding Surveys (LHS) of the NSSO.² The NSSO surveys have an edge over the Agricultural Censuses in that data on land use and holding structure are available both in terms of owned area and operated area. Since in the context of the current study, land is seen as constituting the primary asset for rural households, data from the LHS surveys are used to understand the evolving ownership structure of land holdings in India. A longitudinal view of the changes in the pattern of land ownership in rural India since 1971–1972 is shown in Figure 1. Over the past four decades, the average size of agricultural land owned per household has continuously declined for all households, both including and excluding landless households (Figure 1). It thus appears that there has been progressive attrition in the land base of households in the countryside. The implication of this reduction in land base for rural livelihoods will be explored at the end of this section.

Figure 1

Source: Land and Livestock Holding Surveys, NSSO (various years).

The distribution of rural households according to the size of ownership holdings is presented in Table 1. The most recent LHS survey shows that three-fourths of the rural households are marginal landowners, while one-tenth were smallholders in 2013. Thus, more than 85 per cent of the households in the countryside owned only marginal and small holdings compared to 68 per cent in 1971–1972. A sharp decline is visible in medium, semi-medium and large holdings over this period, as the proportion of households belonging to these categories declined from 22 per cent in 1971–1972 to a little over 7 per cent in 2013.

It is important to note that considerable inequality still exists in the distribution of agricultural land in the country. Thus while 85 per cent of the households owned marginal and small holdings, their share in the total owned area was only 53 per cent in 2013. The small and marginal holdings comprised only about 24 per cent of total area owned in 1971–1972. Although the total area under small and marginal holdings has increased over time, the area under medium, semi-medium and large holdings remain disproportionately large with barely 7 per cent of rural households accounting for 47 per cent of owned area in 2013. Similar conclusions can also be drawn from the distribution of net operated area (Table 2).

Table 1

Distribution of Households by Size of Holdings Owned (per cent)

Size of Land Holding	1971–1972	1982	1992	2003	2013
Landless (< = 0.002 ha)	9.64	11.33	11.25	10.04	7.41
Marginal (0.002–1 ha)	52.98	55.31	60.63	69.63	75.41
Small (1–2 ha)	15.49	14.7	13.42	10.81	10
Semi-medium (2–4 ha)	11.89	10.78	9.28	6.03	5.01
Medium (4–10 ha)	7.88	6.45	4.54	2.96	1.93
Large (>10 ha)	2.12	1.43	0.88	0.53	0.24

Source: Land and Livestock Holding Surveys, NSSO (various years).

The proliferation of marginal landholdings has raised doubts about their economic viability. While, technically, net owned area may be different from net operated area, data from Table 2 suggests that there is not much difference in the percentage distribution of owned area and operated area when total area is classified by size of land holdings. The leased area as a percentage of total owned area was 11.62 for the country as a whole although there were considerable interstate differences. Small-holder families still constitute the bulk of India’s hungry and poor (Singh, Kumar, & Woodhead, 2002), and the produce from their limited land resource is hardly sufficient to meet their basic consumption requirements. In this scenario, diversification into off-farm or non-farm activities can help them combat poverty. Sen (1999) noted that in years when rural non-farm employment increases, rural poverty declines and vice versa. The income from off-farm and non-farm sources serves to supplement income earned from operational holdings, and this can have important implications for food security and enhanced well-being. The role of non-farm activities in promoting household welfare is fairly well documented in literature (Barrett, Reardon, & Webb, 2001; Chand, Prassanna, & Singh, 2011; Mahajan & Gupta, 2011). Most of these studies dwell on the positive effects of non-farm income on alleviating current poverty.

Table 2

Distribution of Owned and Operated Area by Size of Landholding (per cent)

Size of Land Holding	Share of Owned Area		Share of Operated Area
Size of Land Holding	1971–1972	2013	1971–1972	2013
Landless (< = 0.002 ha)	0	0.01	0	0
Marginal (0.002–1 ha)	9.76	29.75	9.2	27.71
Small (1–2 ha)	14.68	23.53	14.8	23.44
Semi-medium (2–4 ha)	21.92	22.07	22.6	23.5
Medium (4–10 ha)	30.73	18.83	30.5	19.33
Large (>10 ha)	22.91	5.81	23	6.02

Source: Land and Livestock Holding Surveys, NSSO (various years).

While it is maintained that poverty is a stochastic phenomenon and that the current poverty status of a household may not be a good indicator of its expected poverty in the future, a household that is not income-poor in the current period still faces the risk of falling below the poverty line in the event of shocks such as illness/death of the principal bread winner, crop failure, rising food prices, and so on. The likelihood of a household falling into the poverty trap is referred to as vulnerability to poverty. Thus while poverty per se is an ex-post concept, vulnerability to poverty is necessarily an ex-ante measure.

This distinction between ex-post (poverty) and ex-ante (vulnerability) is crucial for designing appropriate anti-poverty interventions. One can find a rich empirical literature in India directed towards measuring households’ vulnerability to poverty both at the micro and meso levels. Apart from facilitating poverty alleviation, they have emphasised that managing risk and vulnerability will require diversification by rural households into non-farm activities, as earning from these supplement current income from agricultural operations but also provide a hedge against risks from unforeseen events. However, in India, so far, there is no study directed towards understanding the significance of non-farm employment diversification for the economic welfare of small holder households. Besides, as already mentioned, the non-farm economy encompasses a wide range of activities both skilled and unskilled, and it is essential to probe what forms of non-farm activities may be relevant to reducing the incidence of current and future poverty. The analysis in this article is motivated by these considerations.

3. Data Set and Definitions

The study is based on unit-level (household level) data from the 68^th round of the NSSO survey on Employment and Unemployment, which covers the period July 2011 to June 2012. The survey has employment data for 59,700 households, but consumption records for 59,693 households. As data on household consumption expenditure is essential for constructing the poverty and vulnerability profile of households, our analysis dropped the 77 households with missing consumption data. To identify poor households, we adopted state-specific rural poverty lines (based on the Tendulkar Committee methodology given by the Planning Commission for 2011).³ Thus, households whose monthly per capita consumption expenditure (MPCE) is below the minimum threshold are categorised as poor. These are then combined using the sample weights (available in the dataset) to get the poverty head count for different groups and the entire rural population.

Here, it may be mentioned that while current poverty figures can be conveniently compiled from cross-section surveys that become available occasionally, measurement of vulnerability ideally requires panel data sets. An attempt to measure vulnerability on the basis of cross-section data was first made by Chaudhuri et al. (2002) and the methodology has since been widely employed given the paucity of panel datasets in developing countries. Starting with the basic premise that much of the variation in consumption levels across households can be attributed to observable differences in household characteristics, Chaudhuri (2003) estimated a vulnerability index for households by assuming that household consumption is generated by the following stochastic process:

\ln C = X β + ε

(1)

where, C is average monthly per capita consumption expenditure, X is a set of observable household characteristics, β is vector of parameters and ε is a mean-zero disturbance term that captures idiosyncratic shocks.

To account for the possibility that households with low per capita consumption may experience greater consumption volatility than households with a high mean consumption, the variance of the disturbance term is taken as a function of the household characteristics:

σ_{ε}^{2} = X θ + ξ

(2)

The estimates of β and θ were obtained using the three-stage feasible generalised least squares (FGLS). In the first stage, equation 1 was estimated using ordinary least squares (OLS). The estimated residuals obtained from equation 1 were used to run the following regression:

{\hat{ε}}_{O L S}^{2} = X θ + η

(3)

The predictions obtained from equation (3) were used to transform it as follows:

\frac{{\hat{ε}}_{O L S}^{2}}{X {\hat{θ}}_{O L S}} = \frac{X}{X {\hat{θ}}_{O L S}} θ + \frac{η}{X {\hat{θ}}_{O L S}}

(4)

The OLS estimate of θ in equation (4) yielded an asymptotically FLGS estimate of θ, that is, ${\hat{θ}}_{F L G S}$ . Further, $X {\hat{θ}}_{F G L S}$ is a consistent estimator of $σ_{ε}^{2}$ . Equation (1) was transformed as follows:

\frac{\ln C}{{\hat{σ}}_{ε}} = \frac{X}{{\hat{σ}}_{ε}} β + \frac{ε}{{\hat{σ}}_{ε}}

(5)

where ${\hat{σ}}_{ε} = \sqrt{X θ_{F G L S}}$ . The OLS estimation of equation (5) yielded the asymptotically FLGS estimator ${\hat{β}}_{F G L S}$ which in turn enabled us to estimate the expected log consumption and its variance as:

\hat{E} (\ln C | X) = X {\hat{β}}_{F G L S}

(6)

\hat{V} (\ln C | X) = X {\hat{θ}}_{F G L S}

(7)

If we assume that consumption is log-normally distributed, the probability that the household will be poor, that is, the vulnerability index is:

ν = Φ (\frac{X β_{F G L S} - \ln z}{\sqrt[​]{X {\hat{θ}}_{F G L S}}})

(8)

where, Φ represents the cumulative density of the standard normal variate.

Having obtained a vulnerability score, classification of households into vulnerable and non-vulnerable groups depends critically on the choice of a vulnerability threshold. Since the vulnerability index ranges between 0 and 1, if the threshold is set at the lower limit of zero than all households would end up being classified as vulnerable, and if the upper limit is taken as unity, then the proportion of vulnerable households would be nil. Different studies offer varied arguments regarding the selection of an appropriate threshold.

In the present study, the mean population weighted vulnerability score for 59,693 rural households in the sample is 0.399. This value is taken as the cut-off for segregating vulnerable households from the non-risk category. Thus, households with a vulnerability score greater than 0.399 are categorised as vulnerable; households with a vulnerability index between 0.399 and 0.5 are categorised as being moderately vulnerable; while those with a vulnerability index greater than or equal to 0.5 are termed severely vulnerable.

Table 3

Households Characteristics Used to Estimate Vulnerability Score

AGEHH	Age of household head (in completed years)
SEXHH	Sex of household head (0 if female, 1 otherwise)
HHSIZE	Household size
DR	Dependency rate measured by the proportion of non-workers in the household
HC	Household’s human capital measured by the number of people over 15 years of age and have completed at least secondary education
LAND	Amount of land owned per capita
SC	Caste dummy (if Scheduled Caste = 1, 0 otherwise)
ST	Caste dummy (if Scheduled Tribe = 1, 0 otherwise)
OBC	Caste dummy (if Other Backward Caste = 1, 0 otherwise)
T1	Dummy for household type (T1 = 1, if self-employed in agriculture, 0 otherwise)
T2	Dummy for household type (T2 = 1, if self-employed in non-agriculture, 0 otherwise)
T3	Dummy for household type (T3 = 1, if casual labour in agriculture, 0 otherwise)
T4	Dummy for household type (T4 = 1, if casual labour in non-agriculture, 0 otherwise)
T5	Dummy for household type (T5 = 1, others, 0 otherwise)

Source: Author.

The household-level regressors used for estimating household vulnerability scores are listed in Table 3. While the variables are largely self-explanatory, it may be mentioned that the general category households and regular wage/salary-earning households were taken as the base categories for the dummies relating to caste and household type, respectively.

To assess the impact of employment diversification on household vulnerability, the sample of workers were classified into agricultural and non-agricultural workers on the basis of three-digit National Classification of Occupations (NCO, 2004) provided by the Ministry of Labour and Employment, Government of India.⁴ Thus, self-employed cultivators, agricultural labourers and workers in allied agricultural activities were classified as agricultural workers. The residual group comprising non-agricultural workers covered a wide variety of non-farm occupations is the rural areas. To account for the heterogeneity in non-farm employment and to assess its impact on the outcome, that is, poverty and vulnerability of rural households, non-agricultural workers were further regrouped into relatively homogenous categories, mentioned at relevant places in the article.

An examination of the impact of non-farm employment on current and expected poverty requires a suitable econometric framework. As the current poverty status of a household is a binary variable, a logit equation is appropriate, with the following specification:

Y_{1}^{*} = X β + u

(9)

where the latent unobservable outcome variable Y₁^*is assumed to be a function of household characteristics X and an error term u (Maddala, 1999). Here, the observed outcomes corresponding to the latent variable are as follows:

Y₁ = 1(the household is below the poverty line) if $Y_{1}^{*}$ > 0 = 0 (otherwise) if $Y_{1}^{*} \leq 0$

On the one hand, the vulnerability score of households being a continuous variable that can take any value between 0 and 1, a Tobit model was estimated to explore the effect of non-farm employment on expected poverty, that is, vulnerability. The structural equation in the Tobit model is:

Y_{2}^{*} = X χ + u

(10)

where $Y_{2}^{*}$ is a vector of observations on the vulnerability score of small and marginal households, X is a matrix of observations on the explanatory variables, β = a vector of coefficients to be estimated and ε is the stochastic error term. The observed Y₂ is defined by the following measurement equation: ₂₂≤₂₂₂₂

To check the robustness of the estimators, the impact of various types of non-farm employment on poverty and vulnerability was also analysed using the average treatment effect (ATE) model. The basic argument of this model is that households that are exposed to a particular treatment (non-farm farm employment categories in our case) may be fundamentally different from those that do not receive the treatment in terms of their demographic composition and socio-economic criteria. Hence, to correctly ascertain the impact of a treatment on the outcome of choice, households in the treatment group (that receive treatment) have to be matched with those in the control group (that do not receive treatment) to make them comparable. In our analysis, this is carried out through inverse probability weighted regression adjustment (IPWRA). The IPWRA estimators use weighted regression coefficients to compute averages of treatment-level predicted outcomes, where the weights are the estimated inverse probabilities of the treatment. The contrasts between these averages provide the estimated treatment (Stata Corp., 2013). The IPWRA estimators have a double-robust property.⁵ The ATE estimation was also employed to ascertain the impact of subsidiary occupations on current and expected poverty.

4. Results and Discussion

As the central objective of the article is to understand the relation between rural employment diversification and welfare outcomes among smallholders, at the outset it is intuitive to analyse how the occupational pattern among rural workers in India varies by size class of land owned.⁶ Table 4 lists the occupational structure of rural workers by principal industry. For this analysis, rural workers were categorised into six groups according to their usual principal status: skilled agricultural and fishery (SKAGRI); elementary farm occupations comprising agricultural, fisheries and related labourers (EFARM); managerial, professional and clerical (MPC); shop and market sales (SSM); crafts and related trades (CRAFT); plant and machine operators (PMO); and elementary non-farm occupations (ENF). This classification is broadly in consonance with the National Classification of Occupations (NCO) 2004, except that in our analysis managerial, professional and clerical jobs have been clubbed together whereas in the NCO classification, they are treated as separate categories.

Here, it may be mentioned that the NCO 2004 classifies agricultural workers into skilled and elementary agricultural workers; likewise non-farm workers are also divided into various categories, with elementary non-farm occupations constituting a distinct group. In our study, MPC, SSM, CRAFTS and PMO are considered as constituting skilled non-farm employment. It is observed that landless workers have greatly diversified into skilled non-farm occupations comprising MPC, SSM, CRAFTS and PMO, and the four categories account for about 47 per cent of total employment for landless workers. Among these categories, employment in rural crafts was largest, as it supported 18.1 per cent of the rural landless workers followed by employment in MPC. The proportion of workers belonging to landless households and depending on agricultural vocations was about 36 per cent, with most of them engaged in non-agricultural occupations. In contrast, marginal and small land owners were heavily dependent on the agricultural sector for their livelihood, as 57 per cent and 85 per cent of the workers with marginal and small land holdings, respectively, derived their livelihood from agriculture. In particular, the dependence on elementary farm occupations is quite high among marginal land holders as well as landless workers. Clearly, involvement in non-farm occupations declines monotonically as the size of land holding increases. Nevertheless, non-farm occupations are a source of livelihood for 43 per cent and 15 per cent of marginal and smallholders, respectively. Overall, agriculture continues to be the mainstay of rural workers although non-agricultural vocations comprise nearly 37 per cent of the overall employment in the countryside.

Table 4
Occupational Classification of Rural Workers by Size of Land Owned (per cent)

Landless Marginal Small Semi-medium Medium Large Total

i. Skilled Agriculture 8.31 26.70 77.31 83.71 86.29 87.52 39.55

ii. Elementary Farm 27.58 30.15 7.49 3.16 1.17 0.36 23.73

iii. Total Agriculture (i + ii) 35.89 56.85 84.8 86.87 87.46 87.88 63.28

iv. MPC 13.08 8.94 5.16 6.11 6.68 6.68 8.30

v. SSM 6.13 6.00 2.89 1.77 1.53 2.49 5.09

vi. CRAFT 18.1 13.08 2.65 1.78 1.29 1.15 10.54

vii. PMO 9.84 3.57 1.39 1.75 0.87 1.3 3.30

viii. Skilled Non-farm (iv–vii) 47.15 31.59 12.09 11.41 10.37 11.62 27.23

ix. Elementary non-farm 16.96 11.55 3.12 1.73 2.17 0.51 9.49

x. Total non-farm (viii + ix) 64.11 43.14 15.21 13.14 12.54 12.13 36.72

	Landless	Marginal	Small	Semi-medium	Medium	Large	Total
i. Skilled Agriculture	8.31	26.70	77.31	83.71	86.29	87.52	39.55
ii. Elementary Farm	27.58	30.15	7.49	3.16	1.17	0.36	23.73
iii. Total Agriculture (i + ii)	35.89	56.85	84.8	86.87	87.46	87.88	63.28
iv. MPC	13.08	8.94	5.16	6.11	6.68	6.68	8.30
v. SSM	6.13	6.00	2.89	1.77	1.53	2.49	5.09
vi. CRAFT	18.1	13.08	2.65	1.78	1.29	1.15	10.54
vii. PMO	9.84	3.57	1.39	1.75	0.87	1.3	3.30
viii. Skilled Non-farm (iv–vii)	47.15	31.59	12.09	11.41	10.37	11.62	27.23
ix. Elementary non-farm	16.96	11.55	3.12	1.73	2.17	0.51	9.49
x. Total non-farm (viii + ix)	64.11	43.14	15.21	13.14	12.54	12.13	36.72

Source: Based on the author’s calculations from the NSSO survey on Employment and Unemployment, 2011–12.

Before proceeding further, we make a brief digression into a relatively ignored aspect of rural livelihoods, that is, subsidiary occupations. The economic welfare of a household depends not only on the principal occupation of its workers but also on the subsidiary activities. However, subsidiary activities are not as common in rural India as they are expected to be: about 27 per cent of the rural households reported that they had at least one worker employed as per usual principal status (UPS) and engaged in a secondary activity. Nevertheless, households with marginal and small landholdings were more likely to be involved in subsidiary occupations compared to other groups. Figure 2 reveals that 28.5 and 30.23 per cent of households from these two categories, respectively, were found to be pursuing a secondary occupation, whereas for landless, semi-medium, medium and large landholders, the respective percentages were 14.76, 22.94, 17.36 and 13.49 per cent only.

We turn our attention to the incidence of poverty (measured by head count ratio) and the extent of consumption vulnerability for different landowning groups. It is evident from Figure 3 that the incidence of poverty is highest among marginal land-owners and then tapers off with an increase in landholding size. Here, it may be pointed out that these estimates of poverty have been obtained from the NSSO Survey on Employment and Unemployment and hence are not directly comparable with the Planning Commission estimates, which are based on the Consumption Expenditure surveys.⁷ Nevertheless, they provide a good guide on how the incidence of poverty varies for different segments of the rural population. Interestingly, the poverty headcount among marginal landowners is almost 10 percentage points higher than for landless households at 35.87 per cent. Current poverty is also high among smallholders, as almost 29 per cent of the population belonging to this category was found to be below the poverty line. In fact, landless households fare better not only in comparison with marginal landowners but also with small landholders, the poverty head count for the population belonging to this category being 26.79 per cent. The dominance of marginal landowners in the population serves to raise the average poverty ratio in the population to 32.88. It thus follows that tackling poverty among the marginal and small landowners is critical for reducing the overall incidence of poverty among the rural population in India.

Figure 2

Source: Based on the author’s calculations from the NSSO survey on Employment and Unemployment, 2011–12.

Figure 3

Source: Based on the author’s calculations from the NSSO survey on Employment and Unemployment, 2011–12.

The insights obtained from the occupational classification of rural workers seem to point to the fact that the prevalence of higher poverty among marginal and small landowners compared to landless households could be attributed to the higher dependence on agricultural vocations by the former categories of households, while landless households have largely shifted away from agriculture into non-agricultural vocations. Therefore, there is some indication to suggest that non-farm occupations may have a strong poverty-reducing effect. It therefore becomes relevant to analyse whether diversification into non-farm vocations could produce beneficial welfare outcome among marginal and small landowners who comprise the bulk of the rural population in the country.

The assessment of the overall vulnerability situation of households on the basis of the selected threshold reveals that the fraction of the rural population exposed to the risk of poverty in the next period is substantially higher than the current poverty rate (Table 5). More than half the population is above the vulnerability threshold, while the current head count is 32.88 per cent. Marginal farmers form the bulk of the vulnerable population as nearly 57 per cent of them belonging to this category have a high probability of experiencing future poverty. The risk of exposure to poverty is also high among landless and small-holder farmers at 51 per cent and 42 per cent, respectively. Vulnerability is expectedly lower for the larger landholding classes; nevertheless, with the landless, marginal and small-holders comprising the bulk of the population, overall vulnerability in the rural areas is remarkably high.

The findings of Figure 3 and Table 5 apparently represent a paradox: landless households exhibit a lower incidence of poverty than smallholders, but at the same time have a much higher vulnerability. To reconcile this contradiction, we reclassify the sample households according to their occupations and see how current and expected poverty vary by occupation, regardless of the size of land owned. For this, rural households which specialise in specific types of employment are cross-tabulated according to their poverty and vulnerability status. Thus, out of the 59,623 rural households for which both consumption and employment data are available, only 50,874 households are considered, as the remaining 8,749 were found to be pursuing multiple occupations. According to the occupational classifications outlined above, 19,849 of the 50,874 households specialised in agriculture; 8,782 were engaged in managerial, professional and clerical jobs; 6,294 households specialised in SSM; 6,590 households were in CRAFT; 3,972 in PMO; and the remaining 5,747 households were engaged in ENF. It is discernible from Table 6 that households that specialise in crafts have lower poverty levels compared to agricultural households but they have greater vulnerability.

Table 5

Vulnerability Profile of Rural Households by Size of Landholding (per cent)

Vulnerability	Landless	Marginal	Small	Semi-medium	Medium	Large	Total
Not vulnerable (v < 0.39)	48.66	43.44	58.13	64.88	73.76	80.37	48.17
Moderately vulnerable (0.39 ≤ v < 0.5)	31.56	36.06	27.14	21.65	12.10	8.71	32.84
Severely vulnerable (v ≥ 0.5)	19.78	20.49	14.73	13.48	14.14	10.92	19.00
Completely vulnerable	51.34	56.56	41.87	35.12	26.24	19.63	51.83

Source: Based on the author’s calculations from the NSSO survey on Employment and Unemployment, 2011–12.

The upshot of this cross-classification is that craft-based occupations may serve to lower current poverty but at the same time they are also associated with higher vulnerability, probably due to the seasonality of these occupations. Further, ENF occupations are associated with very high vulnerability as more than 62 per cent of the households that specialised in these activities were found to be above the vulnerability threshold. Drawing an analogy between Tables 5 and 6, it is seen that while households specialising in CRAFT and ENF are prone to higher vulnerability, these occupations constitute an important livelihood option for landless households. It may be recalled from Table 4 that 18.1 per cent and 17.96 per cent of workers belonging to landless households pursued CRAFT and ENF, respectively, as their principal occupations compared to 2.65 per cent and 3.12 per cent, respectively, in the case of small-holder households. It thus emerges that the differences in occupational structure between landless households and smallholders explain the observed differences in their poverty status with regard to current poverty and expected poverty.

Table 6

Vulnerability Profile of Rural Households by Occupation (per cent)

	AGRI	MPC	SSM	CRAFT	PMO	ENF
Currently poor	36.28	14.58	20.87	31.17	22.19	41.69
Vulnerable population	51.38	26.36	34.93	52.07	44.92	62.02

Source: Based on the author’s calculations from the NSSO survey on Employment and Unemployment, 2011–12.

In assessing the poverty threat faced by the rural population, it is important to examine the sources of vulnerability, as this would yield valuable information on how vulnerability can be best tackled. This is because vulnerability for some households may be due to low long-term consumption prospects, while for others vulnerability may be associated with consumption volatility. With this objective, we reclassify vulnerable households into two groups, high consumption volatility (HCV) and low mean consumption (LMC). The first group comprises households with a vulnerability score greater than the population mean but an expected mean consumption above the poverty line. The second group comprises vulnerable families for which expected mean consumption expenditure is below the poverty line. Table 7 gives a breakup of various land-owing categories by sources of vulnerability. It is observed that for medium and large land owners, the major source of vulnerability stems from households with low expected mean consumption, whereas for landless, marginal, small and semi-medium categories, the chief source of vulnerability lies in their highly volatile consumption. It thus follows that the nature of vulnerability differs quite significantly among different land-owning groups in rural areas. This finding has powerful implication for policy, as it shows that measures that help to smooth volatility in consumption among the landless, marginal and small-holder households would go a long way in not only reducing vulnerability among these segments of the population but would also serve to lower the overall incidence of vulnerability in rural areas.

Table 7

Sources of Vulnerability by Size of Landholding (per cent)

	Landless	Marginal	Small	Semi-medium	Medium	Large	Total
HCV^a	73.06	64.08	68.34	62.27	33.92	11.18	64.20
LMC^b	26.94	35.92	31.66	37.73	66.08	88.82	35.80

Source: Based on the author’s calculations from the NSSO survey on Employment and Unemployment, 2011–12.

Note: a–high consumption volatility (HCV); b–low mean consumption (LMC).

The foregoing analysis suggests that marginal and small land holders along with landless households constitute the most vulnerable segments of the rural population. We examine one potential solution to this endemic problem, that is, the role of employment diversification in reducing livelihood risks for marginal and small-holder households, which comprise the majority of India’s rural population. The degree of employment diversification within a household is measured by the proportion of principal workers in a household engaged in non-agricultural activities (PERNON). The marginal effects of the logit and Tobit regressions of households’ poverty status on the index of employment diversification and other household characteristics are presented in Table 8. Here, it may be mentioned that out of 59,623 rural households for which employment and consumption records are available in the NSSO Schedule on Employment and Unemployment, 49,178 belonged to the category of small and marginal land owners, of which 160 households had at least one worker whose occupation was unclassified as per NCO, 2004. As the composition of occupations within a household is crucial for our analysis, these households were dropped from our econometric estimation. It was further noted that of the remaining 49,018 households, 2,517 were dependent on transfers and did not have a principal worker, so these were also excluded. Moreover, as subsidiary occupations were likely to confuse and confound the results, 14,282 households in which at least one person had a subsidiary occupation were also not considered. The regression was therefore carried out for 32,219 marginal and small-holder households with 51,885 workers. Here, it may be mentioned that a household income depends, among other things, on the operated area and not on owned area. However, the size of the cultivated/operated area is likely to have a direct impact on the employment structure within a household. This is because there is every possibility that the size of the operated area is highly correlated with the percentage of workers engaged in nonfarm occupations within a household (PERNON). In other words, the larger the size of the operated area, the lower will be the percentage of workers engaged in nonfarm occupations and vice versa. To avoid the multicollinearity problem, the size of operated area has not been explicitly considered in the regression, as its effect is already reflected in the composition of agricultural and non-agricultural workers within the household as measured by PERNON. It is found from Table 8 that for every 1 per cent increase in PERNON, the probability of the household falling below the poverty line in the current period decreases by 5 per cent. However, the largest poverty reduction effect is observed in the case of HC, as a 1 per cent increase in the human capital base of households results in a 26.5 per cent reduction in the probability of current poverty. Both coefficients have been found to be statistically significant at the 1 per cent level. The coefficients of the Tobit regression in Table 8 show the impact of various factors on the likelihood of future poverty. The coefficients of both PERNON and HC in the Tobit regression have been found to be negative and statistically significant, which shows that vulnerability is reduced for every percentage increase in the value of these variables; however, in case of PERNON, the size of the coefficient is quite small, which indicates that all types of nonfarm employment may not be equally potent in reducing vulnerability and their combined effect is quite muted. The likelihood of both current and expected poverty also falls if the household head is male and, further, with every percentage increase in the age of the household head. In contrast, the incidence of poverty and vulnerability increases for every percentage increase in household size and dependency rate, and also if the household belongs to the SC, ST and OBC categories as opposed to general category households. All coefficients of all variables used in the regression are statistically significant.

Table 8

Effect of Non-farm Employment on Current and Expected Poverty

Variable	Marginal Effects on Poverty: Logit Estimates			Marginal Effects on Vulnerability: Tobit Estimates
Variable	dy/dx	z	P>\|z\|	dy/dx	z	P>\|z\|
AGEHH	–0.001	–6.020	0.000	–0.002	–82.040	0.000
SEXHH#	–0.035	–4.150	0.000	–0.042	–38.200	0.000
HHSIZE	0.046	42.360	0.000	0.050	303.370	0.000
DR	0.158	11.160	0.000	0.131	56.340	0.000
HC	–0.265	–36.430	0.000	–0.181	–191.530	0.000
SC#	0.103	12.520	0.000	0.096	99.660	0.000
ST#	0.136	14.410	0.000	0.140	135.970	0.000
OBC#	0.052	8.810	0.000	0.050	63.340	0.000
PERNON	–0.047	–10.500	0.000	–0.005	–6.560	0.000
# dy/dx is for discrete change of dummy variable from 0 to 1

Source: Based on the author’s calculations from the NSSO survey on Employment and Unemployment, 2011–12.

It is thus seen from Table 8 that employment diversification into non-farm activities by marginal and small land-owning households can significantly improve their livelihood outcomes by lowering the risk of current and future poverty. However, in stressing the role of non-farm activities in enhancing household welfare and reducing livelihood risks, it needs to be borne in mind that the non-farm sector encompasses a wide range of high-end and low-end activities which are likely to have a differential impact on poverty and vulnerability profile of households. Hence, to facilitate a deeper understanding of the nexus between poverty and employment diversification, the 32,219 small and marginal landowner households considered for the above regression were further categorised according to their occupations. To clearly measure the effects of different forms of non-farm employment on poverty and vulnerability, 5,800 households whose principal workers were engaged in different types of non-agricultural employment were excluded from the analysis. Of the remaining 26,419 small-holder households, 10,026 were found to be exclusively engaged in AGRI, 5,694 were involved in MPC, 2,804 households were found to be specialising in SSM, 3,594 were engaged exclusively in CRAFT, 1,361 in PMO and the remaining 2,940 households were employed in ENF activities. For the regression equation, households involved in AGRI were taken as the base category.

The marginal effects of the Logit and Tobit regressions obtained from this disaggregated analysis are shown in Table 9. It is observed that most forms of non-farm employment have a significant and distinct effect on poverty outcomes at the household level. Thus, households specialised in MPC, SSM and PMO are less likely to fall into the poverty trap than households exclusively specialised in agricultural activities. The coefficients of the dummy variables corresponding to all these non-farm categories were found to be negative and statistically significant. Although the coefficients for CRAFT and ENF were also found to be negative, they were insignificant, implying that their impact on the current poverty situation of smallholders is uncertain.

Table 9

Effects of Non-farm Employment on Poverty and Vulnerability

Variable	Marginal Effects on Poverty: Logit Estimates			Marginal Effects on Vulnerability: Tobit Estimates
Variable	dy/dx	z	P>\|z\|	dy/dx	z	P>\|z\|
AGEHH	–0.001	–4.940	0.000	–0.002	–75.650	0.000
SEXHH#	–0.033	–3.630	0.000	–0.041	–34.160	0.000
HHSIZE	0.050	40.390	0.000	0.050	269.030	0.000
DR	0.144	9.680	0.000	0.129	52.490	0.000
HC	–0.226	–27.040	0.000	–0.167	–153.450	0.000
SC#	0.098	10.900	0.000	0.094	88.720	0.000
ST#	0.151	14.280	0.000	0.143	128.000	0.000
OBC#	0.047	7.260	0.000	0.048	56.470	0.000
MPC#	–0.099	–18.350	0.000	–0.023	–23.600	0.000
SSM#	–0.045	–6.890	0.000	–0.016	–13.580	0.000
CRAFT#	–0.005	–0.800	0.425	0.004	3.460	0.001
PMO#	–0.069	–9.370	0.000	–0.010	–6.020	0.000
ENF#	–0.005	–0.770	0.441	0.015	12.460	0.000
#dy/dx is for the discrete change of the dummy variable from 0 to 1

Source: Based on the author’s calculations from the NSSO survey on Employment and Unemployment, 2011–12.

The results of the Tobit regression reveal that the effects of different types of non-farm employment on household vulnerability were not uniform. Thus, while MPC, SSM and PMO activities help lower the probability of expected poverty among marginal and small-holder households, ENF activities have the opposite effect. However, as is the case with current poverty, HC has the strongest vulnerability-reducing effect judging from its large coefficient. The coefficient of the dummy relating to CRAFT is positive and significant which implies that households pursuing CRAFT as their sole occupation tend to have greater vulnerability than agriculture-dependent households. This finding is in consonance with our earlier analysis in Tables 5 and 6. The interpretation of the remaining variables is the same as in the first regression and does not warrant repetition.

To check the robustness of the estimates obtained through the Logit and Tobit regressions in Table 9, the impacts of various types of non-farm employment were cross-verified using the ATE model. The ATE coefficients for non-farm occupation types, obtained through the IPWRA method taking AGRI as the base category, are presented in Table 10. Considering the ATE on poverty first, it is observed that after correcting for endogeneity in the choice of occupations, the poverty headcount among small-holder households specialising in MPC is 9.6 percentage points lower than the poverty headcount among those that specialise in agricultural occupations only. Similarly, small-holder households that pursue SSM and PMO as their principal occupations have a poverty headcount that is 6 and 8.9 percentage points lower, respectively, than the poverty head count among agricultural households. The ATE coefficients for all these categories have been found to be statistically significant. In contrast, engagement in craft-related work serves to marginally increase the poverty head count in the current period by 0.2 percentage points although the coefficient is not statistically significant. Likewise, the impact of ENF employment on poverty is not clear as the ATE coefficient for this category is again not significant.

Analysis of the ATE coefficients for vulnerability reveals that the vulnerability scores of households specialising in MPC, SSM and PMO on average are lower than agricultural households by 0.023, 0.019 and 0.007 points, respectively, indicating that these occupation types serve to lower vulnerability. ENF occupations and CRAFT on the other hand are found to be increasing the risk of future poverty among small-holder households by 0.015 and 0.005 points, respectively. The ATE coefficients relating to vulnerability have been found to be statistically significant at 1 per cent for MPC, SSM, PMO and ENF; for CRAFT it has been found to be significant at 5 per cent. The results obtained from the ATE model therefore corroborate the findings from the Logit and Tobit regressions considered earlier. It may be recalled that households with subsidiary occupations have not been considered here.

Table 10

ATE Coefficients for Poverty and Vulnerability (by Type of Non-farm Employment)

Type of Non-farm Employment	Poverty			Vulnerability
Type of Non-farm Employment	Coefficient	z	P>\|z\|	Coefficient	z	P>\|z\|
MPC vs AGRI	–0.096	–13.150	0.000	–0.023	–19.340	0.000
SSM vs AGRI	–0.060	0.009	0.000	–0.019	–13.750	0.000
CRAFT vs AGRI	0.002	0.240	0.807	0.005	3.870	0.000
PMO vs AGRI	–0.089	–6.000	0.000	–0.007	–2.220	0.026
ENF vs AGRI	–0.010	–1.120	0.262	0.015	9.240	0.000

Source: Based on the author’s calculations from the NSSO survey on Employment and Unemployment, 2011–12.

The findings from the above analysis are summed up in Table 11. Thus, smallholder households that specialise completely in MPC, SSM and PMO are less exposed to the risk of current poverty and expected poverty as compared to agricultural households. The impact of CRAFT and ENF on poverty is uncertain; however both result in higher vulnerability.

Any study assessing the relationship between livelihoods and current and expected poverty would be incomplete without considering the role of subsidiary occupations. This is because subsidiary occupations supplement the income from principal occupations as well as cushion the effects of shocks or unforeseen contingencies. The econometric exercise above has not considered subsidiary occupations, as there is an inherent element of endogeneity, given that relatively poorer households are more likely than better-off households to pursue a secondary activity. To determine the effect of subsidiary occupations on household welfare, we again employ the ATE model. The procedure involves two stages: In the first stage, the probability of a household participating in a subsidiary activity is estimated on the basis of observed household characteristics by using a Logit model. In the second state, these probabilities are used in an IPWRA model to ascertain the impact of subsidiary occupations on current poverty and vulnerability.

Table 11

Summary of Regression Results

Non-farm Employment	Impact on Poverty	Impact on Vulnerability
MPC	Reduces	Reduces
SSM	Reduces	Reduces
CRAFT	Uncertain	Increases
PMO	Reduces	Reduces
ENF	Uncertain	Increases

Source: Based on author’s calculations from NSSO survey on Employment and Unemployment, 2011-12, 68^th Round.

Table 12

Impact of Subsidiary Occupations on Poverty and Vulnerability

	Coefficient	z	P>\|z\|
ATE on poverty	–0.009	–2.470	0.014
ATE on vulnerability	–0.003	–6.330	0.000

Source: Based on the author’s calculations from the NSSO survey on Employment and Unemployment, 2011–12.

The results presented in Table 12 show that the poverty head count among small-holder households that pursue a subsidiary occupation is 0.9 percentage points lower than the poverty head count among households without a subsidiary occupation. Likewise, the vulnerability score of households on average falls by 0.003 points if the household has access to income from a secondary activity. However, it should be noted that although the ATE coefficients are statistically significant, the coefficients themselves are fairly small, indicating that subsidiary occupations may not have a substantial impact on the poverty situation in rural areas.

5. Summary and Conclusion

Poverty in rural areas is predominantly a rural phenomenon as vast majority of the poor is found to be concentrated in rural areas (Dey, 2017). The poverty situation of households in the current and future periods is critically linked to the level of assets they own, in particular agricultural land. However, with the progressive decline in the size of ownership holdings, most rural households have been reduced to being marginal landowners. This paper has taken a comprehensive look at how current poverty as well as the future outlook of poverty changes for various rural landowning groups. It is observed that marginal and smallholders are the most vulnerable group in terms of current and expected poverty. Moreover, the major share of the observed vulnerability among the target group can be attributed to the high volatility of consumption. As they account for the majority of rural population, they also serve to push up overall poverty and vulnerability ratios in the rural areas. Overcoming deprivation in this target population is therefore central to tackling the overall poverty situation in Indian villages. While several policy measures can be suggested for tackling this problem, this study investigated one potential solution namely the role of non-farm employment in alleviating poverty for small-holder rural households in the current period as well as in reducing the future risk of poverty.

The study reveals that employment diversification into non-farm activities by smallholders has a significant impact on lowering the prevalence of current and expected poverty. However, all forms of non-farm employment are not equally potent in fighting chronic poverty The impact of craft-based occupations and elementary non-farm occupations on the welfare of households is particularly uncertain. Although the other types of non-farm employment, that is, managerial and professional jobs, service and sales occupations in rural areas as well as machine and plant operations serve to lower the incidence of poverty, it is important to note that of all variables considered in the study human capital has the largest poverty-alleviating effect. This serves as a pointer to the fact that an emphasis on education and skill formation, coupled with a policy of employment diversification, could constitute a near full-proof strategy for tackling poverty and vulnerability among marginal and small-holder households in the rural areas. The National Rural Livelihood Programme (NRLM)⁸ which is the flagship programme of the Government of India aimed at enabling rural poor through sustainable livelihood enhancements programmes is a potentially powerful model to effectively implement these twin strategies to ensure an enduring solution to the problem of rural poverty.

Footnotes

Notes

References

Barret

C. B.

& McPeak

J. G.

(2006). Poverty traps and safety nets. In Janvry

A. D.

& Kanbur

(Eds), Poverty, inequality and development: Essays in honor of Erik Thorbecke (pp. 131–154). New York: Springer.

Barrett

C. B.

Reardon

& Webb

(2001). Nonfarm income diversification and household livelihood strategies in rural Africa: Concepts, dynamics, and policy implications. Food Policy, 26(4), 315–311.

Carter

M. R.

& Barrett

C. B.

(2006). The economics of poverty traps and persistent poverty: An asset-based approach. Journal of Development Studies, 42(1), 178–199.

Carter

M. R.

& May

(1999). Poverty, livelihood and class in rural South Africa. World Development, 27(1), 1–20.

Chambers

(1995). Poverty and livelihoods: Whose reality counts? Environment and Urbanisation, 7(1), 173–204.

Chand

Prassanna

P. A. L.

& Singh

(2011). Farm size and productivity: Understanding the strength of smallholders and improving their livelihoods. Economic and Political Weekly, XLVI(26–27), 5–11.

Chiwaula

L. S.

Witt

& Waibel

(2011). An asset-based approach to vulnerability: The case of small-scale fishing areas in Cameroon and Nigeria. Journal of Development Studies, 47(2), 338–353.

Chaudhuri

(2003). Assessing vulnerability to poverty: Concepts, empirical methods and illustrative examples. Department of Economics, Columbia University. Retrieved from www.econdse.org/wp-content/uploads/2012/02/vulnerability-assessment.pdf

Chaudhuri

Jalan

& Suryahadi

(2002). Assessing household vulnerability to poverty from cross-sectional data: A methodology and estimates from Indonesia (Columbia University Discussion Paper No.0102-52). Retrieved 20 January 2015 , from https://doi.org/10.7916/D85149GF

10.

Dey

(2017). Assessing the poverty impact of India’s largest livelihood security programme: A study based on the 68^th round of the NSSO. Journal of Development Policy and Practice, 2(1), 56–70.

11.

Echevin

(2011). Measuring vulnerability to asset poverty in Sub-Saharan Africa. MPRA Paper No. 35660. Retrieved from https://mpra.ub.unimuenchen.de/35660/1/MPRA_paper_35660.pdf

12.

Funk

M. J.

Westreich

Weisen

& Davidian

(2014). Doubly robust estimation of treatment effects. In Faries

D. E.

Obenchain

Haro

J. M.

& Leon

A. C.

(Eds), Analysis of observational health care data using SAS (pp. 85–103). North Carolina: SAS Institute.

13.

Government of India. (2014). All India report on number and area of operational holdings. New Delhi: Ministry of Agriculture.

14.

Maddala

G. S.

(1999). Limited dependent and qualitative variables in econometrics. Econometric Society Monograhps No 3. Cambridge, UK: Cambridge University Press.

15.

Mahajan

& Gupta

R. K.

(2011). Non-farm opportunities for small holder agriculture. Paper Presented at IFAD Conference on New Directions for Small Holder Agriculture, Italy. Retrieved 18 May 2017 , from https://www.ifad.org/documents/10180/d0b4c8ab-bcf0-4c87-bc4b-6127daf06e78

16.

Robins

J. M.

Hernan

M. A.

& Brumback

(2000). Marginal structural models and causal inference in epidemiology. Epidemiology, 11(5), 550–560.

17.

Sen

A. K.

(1981). Poverty and famines: An essay on entitlement and deprivation. Oxford, UK: Clarendon Press.

18.

Sen

A. K.

(1999). Economic policy and equity: An overview. In Vito

Ke-Young

& Gupta

(Eds), Economic policy and equity (pp. 28–43). Washington, DC: IMF.

19.

Singh

R. B.

Kumar

& Woodhead

(2002). Smallholder farmers in India: Food security and agricultural policy. Bangkok: Food and Agricultural Organization.

20.

StataCorp. (2013). Stata: Release13 (Statistical Software). College Station, TX: StataCorp LP.