Understanding the politics of caste, corruption, and wealth is essential for combating poverty in India. However, relatively few studies have systematically analyzed how these factors explain patterns of poverty combining state-level indicators with household and child-level outcomes. Focusing on child poverty as an outcome measure, this paper tests the explanatory potency of John Harriss's typology of state government political regimes, Transparency International India's measures of state corruption, and state-level wealth. Using data on 120,988 children from the third National Family Health Survey (2005–2006) and multilevel models, we find that Harriss's typology of state regimes better explains child poverty differences between states than Transparency International India's corruption index. States whose political regimes are historically dominated by upper-caste groups tend to have an adverse effect on poor children of lower castes, compared to states dominated by lower-caste groups. This adverse effect is amplified in wealthier states.

## INTRODUCTION

Between 1993 and 2013, India's average annual GDP growth was an impressive 8% (Trading Economics 2015). However, during this period the country saw only a relatively small reduction in non-monetary measures of development, such as child malnutrition (Gragnolati et al. 2005) and access to basic services (IIPS and Macro International 2007).

The dual nature of India's economic and social development has generated much debate, with political, administrative, and bureaucratic incompetence, corruption, and inefficiency highlighted. In 2012 a report on administrative governance examined the political and administrative processes which result in the misallocation and misappropriation of public funds. It noted that “Weak governance, manifesting itself in poor service delivery, excessive regulation and uncoordinated and wasteful public expenditure, is one of the key factors impinging on development and social indicators” (Saxena 2012:2–3).

The report also quoted the Second Administrative Reforms Commission (2008) as saying:

The state apparatus is generally perceived to be largely inefficient with most functionaries serving no useful purpose. The bureaucracy is generally seen to be tardy, inefficient and unresponsive. Corruption is all-pervasive, eating into the vitals of our system, undermining economic growth, distorting competition and disproportionately hurting the poor and marginalized citizens. Criminalization of politics continues unchecked, with money and muscle power playing a large role in elections. In general there is a high degree of volatility in society on account of unfulfilled expectations and poor delivery. Abuse of authority at all levels in all organs of state has become the bane of our democracy.

The deficiencies of governance in India, therefore, are clear. While poverty declined, with the national head count falling from 37% to 30% between 2004/05 and 2009/10 (Government of India Planning Commission, 2012) and all-India urban and rural poverty rates also falling, considerable inter-state differences remained (Alkire and Seth 2015; Cain et al. 2010; Cavatorta, Shankar, and Flores-Martinez 2015; Dhongde 2017). Poverty fell by 10% or more in the states of Himachal Pradesh, Madhya Pradesh, Maharashtra, Orissa, Sikkim, Tamil Nadu, Karnataka, and Uttarakhand but rose in the northeastern states of Assam, Meghalaya, Manipur, Mizoram, and Nagaland. Sizeable disparities persist between social groups, with higher-than-expected rates of rural poverty in Scheduled Tribe (47%) and Scheduled Caste (42%) communities. Poverty rates for urban-based Scheduled Caste and Scheduled Tribes communities were at 34% and 30%, respectively, in 2009/10.

Poverty statistics are always contentious and political (Deaton and Dreze 2002; Saith 2005). Explanations of the causes of poverty, and people's (in)ability to extricate themselves from it, lie at all levels, from international/state-level factors to those attributed (or attributable) to individuals and communities (Kim, Mohanty, and Subramanian 2016).

Explaining the different socio-political processes in a country as diverse as India is challenging. Interactions between ethnic groups, religions, and social stratifications associated with caste and tribe, gender, geography, and occupation result in an almost impenetrable “black box” within which decisions are made and resources allocated (Fontaine and Yamada 2014; Harriss 2005; Kaletski and Prakash 2016), and this especially so in cases of resource scarcity and weak governance (Daoud 2010, 2011, 2015b, 2017; Halleröd et al. 2013). The impressive human development outcomes for states like Kerala contrast sharply with those of Bihar and Uttar Pradesh. Much is made of these state-level differences (Datt and Ravallion 1998; Kohli 1989; Nayyar 1991), not least with regard to the issue of corruption. Harriss's (1999, 2005, 2013) work on state political regimes and rural poverty reduction considers these issues in detail, and his typology of state political regimes in the mid-to-late 1990s provides an interesting framework to examine poverty in India. His theoretical framework captures India's complexity in an analytically sharp yet non-reductionist way, although questions remain as to whether his theory has any explanatory power when applied to empirical data.

In this paper, we test Harriss's theory by analyzing the interaction between state political regimes, Transparency International India's (TII's) index of state corruption, and state economic development (wealth), to see to what degree it explains observed differences in the distribution of child poverty across India. We focus on child poverty rather than general poverty for several reasons. Children are most vulnerable to the impacts of poverty and deprivation; they are more dependent on their parents and basic public services. There is also a moral imperative to analyze their situation, and to improve the conditions that affect their survival and development. A good way to assess a society's level of development is to see how it treats its most vulnerable citizens.

## POLITICAL REGIMES AND CORRUPTION

### Corruption as a Determinant of Poverty

The links between political and administrative corruption and poverty in low- and middle-income countries have been studied extensively. The World Bank's 2001 World Development Report showed that the impacts of corruption fall most heavily on the poorest, by affecting access to and the quality of public services on which the poor depend, by diverting funds to unproductive uses and for personal enrichment, and by increasing the costs of capital investment through kickbacks and bribes.

Gupta, Davoodi, and Alonso-Terme (2002) examined the different processes through which corruption affects national poverty and inequality. They showed that corruption slowed economic growth and heightened inequality, both of which contribute to persistent poverty. Maladministration in areas of tax collection or exemption benefited the wealthiest, with the targeting of benefit programs away from the poor and concentration of assets and resources in the elite limiting the effectiveness of anti-poverty policies. The diversion of public funds away from social programs (e.g., public health or education) or toward groups which control the policy process is another way corruption contributes to poverty and inequality (Puaca and Daoud 2011).

The literature on corruption suggests two broad mechanisms, related to economics and governance. Briefly, by reducing economic growth, increasing inequality, and reducing governance capacity, corruption leads to higher poverty (Heidenheimer and Johnston 2002; Nandy, Daoud, and Gordon 2016). Studies point to an inverse relationship between corruption and economic growth, highlighting the issues raised by Gupta, Davoodi, and Alonso-Terme (2002). They also point to lower levels of public investment (Mauro 2002). Studies focusing on the impact of corruption on quality of governance show how biased decision-making and resource allocation in favor of some groups over others can degrade the quality of public services on which the poor, and children in particular, are disproportionately dependent (Kaufmann, Kraay, and Zoido-Lobaton 2000).

In more recent work, Daoud (2015b) assessed the explanatory power of good governance for combating child poverty in India, using governance measures developed by Mundle et al. (2012). He found that governance explained about 60% of the inter-state variation of child poverty.

One of the strongest characterizing features of stable societies is that they often have little institutional corruption (Rothstein 2014). The main mechanism is social trust (Rothstein and Uslaner 2005): the less corruption, the more social trust. Social trust facilitates many positive societal effects, from stronger economic development to better political representation of weaker social groups in governing bodies. When the poor are well represented via a democratic system, they will work to strengthen their entitlements. They will promote the use of pro-poor policies to a higher degree, compared to forms of organization where they are not well represented (Daoud 2007; Koumakhov and Daoud 2016; Kwon and Kim 2014; Ross 2006; Sen 2000). The same logic applies to how social and economic policies are influenced by greater participation of caste and class groupings; people will work to improve the situation and to protect the interests of those of similar background. These ideas have also been expressed by the theory of the median voter (Meltzer and Richard 1981). If most of the electorate is poor, so that the median income is lower than the mean income, elected leaders promote policies which benefit most voters; this can include redistributive policies such as social spending on health and education (Myroniuk, Vanneman, and Desai 2017) and the provision of public goods (Boix 2003; Bueno de Mesquita et al. 2003; Ghobarah, Huth, and Russett 2004). However, these theoretical explanations have found mixed empirical results, and the literature shows that the link between democracy, political accountability, and poverty reduction is not straightforward (Keefer and Khemani 2005). It has also been argued that the median voter theory cannot be easily applied to societies fragmented by language, ethnicity, and religion (Deaton and Dreze 2002; Meltzer and Richard 1981).

Even if a political system functions as the median voter theory suggests, and if the governance of a state can be said to be “good” (i.e., effective, unbiased), the question remains of how and to what extent corruption hinders development. Corruption limits the full and effective functioning of the state through the misallocation or misappropriation of public resources. Corruption grows in government organizations that lack transparency and accountability (Rothstein 2014; Rothstein and Uslaner 2005). Khan (2006) has argued that corruption in developing countries is deeply rooted in the local political, social, and economic structures (e.g., patronage). While corruption is almost certainly prevalent in all countries, what matters is its nature and degree. Accordingly, since corruption has been found to explain many of the ills of poor development, a central task of this paper is to analyze the relative importance of political regimes, corruption, and economic development for child poverty in India.

### Political Regimes Determining the Quality of Governance

Expanding on earlier work (Kohli 1989), Harriss (2000) examined the effect of class power in state governance and its impact on performance in rural poverty reduction. He realized the importance of taking into account caste/class distinctions and the impact of “accommodationism” between caste/class groups, which remains a strong element in Indian politics. Harriss's typology attempted to explain differences in the democratic functioning of Indian states and the political, economic, and social forces behind them. Differences in states’ levels of industrial development, for example, would determine the relationships between an industrial bourgeoisie and the working class, which in turn would alter the nature and extent of political mobilization and organization of civil society. These processes are also influenced by caste, class, and other ethnic identities, and when taken together result in quite different political environments and, eventually, economic and social outcomes. Thus, it is expected that in regimes where the strategic interests of the poor (for example, around issues like land reform) are more effectively organized, and where the caste/class balances tilt political power in their favor, one would see better outcomes for them. Harriss concluded, perhaps unsurprisingly, that “the regime differences … distinguished do seem to make sense of some of the variations in the adoption, resourcing and implementation of what can be described as ‘pro-poor policies.’ The structure and functioning of local agrarian power, and the relations of local with state-level power-holders, do vary significantly between states and exercise influence both on political patterns and on some policy outcomes.”

Harriss's typology grouped the largest and most populous states of India as follows:

Type A(i) states, in which upper caste/class dominance has persisted and the Congress has remained strong in the context of a stable two-party system (“traditional dominance” rather than politics of accommodation vis-à-vis lower classes). This group included the states of Madhya Pradesh, Orissa, and Rajasthan.

Type A(ii) states, in which upper caste/class dominance has been effectively challenged by middle castes/classes, and Congress support has collapsed in the context of fractured and unstable party competition (both “dominance” and the politics of accommodation have broken down). This group included Bihar and Uttar Pradesh.

Type B states, with middle caste/class dominated regimes, where the Congress has been effectively challenged but has not collapsed, and there is fairly stable and mainly two-party competition (the politics of accommodation vis-à-vis lower-class interests have continued to work effectively, most effectively in Maharashtra and Karnataka, least effectively in Gujarat). This group included Andhra Pradesh, Gujarat, Karnataka, Maharashtra, and Punjab.

Type C states, in which lower castes/classes are more strongly represented in political regimes and where the Congress lost its dominance at an early stage. This group included Kerala, Tamil Nadu, and West Bengal.

The states in groups A(i) and A(ii) were also “low-income states”; those in group B were “high-income states,” and those in group C were “middle-income states.” Harriss also examined levels of public spending on development and rural poverty alleviation, identifying clear patterns and relationships between state regimes and outcomes for the poor. He noted: “Those states which have most clearly pursued what might be described as a direct approach to poverty reduction, through investments in the key social sectors of education and health, and by means of food subsidies, are those in which there is evidence that lower castes/classes are most strongly represented in the political regime.”

This was not entirely unexpected, as earlier work reached similar conclusions (Kohli 1989). Causation is an essential factor, and Harriss's conclusions could equally (or at least as validly) have been that more direct or effective approaches to poverty reduction were implemented by some state regimes precisely because the poor were most strongly represented. Harriss's work was done at a time when questions were increasingly being asked about the importance and impact of state governance in national development processes, not least in a situation of natural disaster (Daoud, Halleröd, and Guha-Sapir 2016) or economic crisis (Daoud et al. 2017). Defined by the World Bank (1991) as “the exercise of political authority and the use of institutional resources to manage society's problems and affairs,” governance is now considered an essential element of the development process, for both state and non-state actors (Daoud 2015a, 2015b). Major donors increasingly emphasize the importance of good governance, and it is set to be a key issue in the era of the Sustainable Development Goals (UNGA 2010; von der Hoven, 2012).

### Articulating Four Hypotheses

From the theoretical discussion of how political regimes and corruption affect poverty, we derive four hypotheses. The first is that Indian states politically dominated by higher and middle castes/classes (political regime type) will have more child poverty. The reasoning behind this hypothesis is that in a political climate dominated by better-off groups in society, the worse-off will have less say in public discourse and decision-making (Drèze and Sen 2014).

The second hypothesis is that political regime type has a greater influence than corruption, but both will lead to more child poverty. This hypothesis directly tests the relative impact of Harriss's (2013) idea that political regimes matter, versus the established view that it is corruption which matters most for explaining child poverty across states.

The other two hypotheses test the moderating effect of state wealth on political regime type and the perceived influence of corruption. The third hypothesis is that political regime types dominated by higher and middle castes/classes, and as their wealth increases (higher GDP), will result in less poverty compared to those states with less wealth. This hypothesis assumes that higher GDP will have an ameliorating effect on child poverty rates, if trickle-down theory is to be believed—that is, better-endowed state governments will invest public money in such a way as to reduce poverty (e.g., via public education, health care, and food security). The fourth hypothesis is that greater corruption will lead to greater child poverty in states with higher levels of wealth (GDP) compared to those with less wealth. The key assumption of this hypothesis is that the impact of corruption is likely to be greater in wealthier states since there is a greater incentive for capturing economic power (Corbridge, Harriss, and Jeffrey 2013). The underlying mechanism is that of resource scarcity leading to conflict between ethnic and religious groups. When there is some perceived local abundance (wealth) in a context of macro scarcity (general poverty), then the incentive increases to capture that wealth, by political means or by corruption (Daoud 2018).

## METHODS AND DATA

### Child Poverty

We use data from the third round of India's National Family Health Survey (NFHS-3, 2005/06). The NFHS is a nationally and sub-nationally representative household survey which collects a wealth of individual and household-level data about people's living conditions, access to services, health status, and well-being (IIPS and Macro International 2007). The data are similar to the Demographic and Health Surveys used in other low- and middle-income countries to track progress toward international goals like the Millennium Development Goals and the Sustainable Development Goals (Corsi et al. 2012).

Our sample included data on 120,988 children, from birth to 18 years. The main dependent variable is an indicator of absolute child poverty, based on a definition agreed at the 1995 World Summit on Social Development. The governments of 117 countries defined absolute poverty for policy purposes as “a condition characterised by severe deprivation of basic human needs, including food, safe drinking water, sanitation facilities, health, shelter, education and information. It depends not only on income but also on access to social services” (United Nations 1995:57).

This definition of “absolute” poverty remains to this day one of the few internationally agreed definitions of poverty. It requires the operationalization of indicators of severe deprivation for the basic human needs identified, and the so-called Bristol Approach has done this for over 15 years (Gordon et al. 2003; Minujin and Nandy 2012; UNICEF 2007). Children experiencing multiple severe deprivations are classed as living in absolute poverty. These definitions are outlined in Table 1.

TABLE 1.

Absolute child poverty: an aggregate of seven types of individual-level child deprivation

Child Deprivation
Water. Children who only had access to surface water (for example, rivers) for drinking or who lived in households where the nearest source of water was more than 15 minutes away. Children under 18.
Food. Children whose heights and weights for their age were more than 3 standard deviations below the median of the international reference, that is, severe anthropometric failure. Children under 5.
Education. Children who had never been to school and were not currently attending school, that is, no professional education of any kind. Children 7 to 12.
Shelter. Children in dwellings with more than five people per room and/or with no flooring material. Children under 18.
Sanitation. Children who had no access to a toilet of any kind in the vicinity of their dwelling, that is, no private or communal toilets or latrines. Children under 18.
Health. Children who had not been immunized against diseases or young children who had a recent illness involving diarrhea and had not received any medical advice or treatment. Children under 5.
Absolute child poverty: Experiencing two or more of the seven deprivations defined above.
Child Deprivation
Water. Children who only had access to surface water (for example, rivers) for drinking or who lived in households where the nearest source of water was more than 15 minutes away. Children under 18.
Food. Children whose heights and weights for their age were more than 3 standard deviations below the median of the international reference, that is, severe anthropometric failure. Children under 5.
Education. Children who had never been to school and were not currently attending school, that is, no professional education of any kind. Children 7 to 12.
Shelter. Children in dwellings with more than five people per room and/or with no flooring material. Children under 18.
Sanitation. Children who had no access to a toilet of any kind in the vicinity of their dwelling, that is, no private or communal toilets or latrines. Children under 18.
Health. Children who had not been immunized against diseases or young children who had a recent illness involving diarrhea and had not received any medical advice or treatment. Children under 5.
Absolute child poverty: Experiencing two or more of the seven deprivations defined above.

A recent paper (Gordon and Nandy 2016) using this approach with NFHS-3 data detailed the extent of deprivation of basic needs among India's children:

• Over two-thirds (68%, around 300 million) were shelter deprived, living in dwellings with more than five people per room or which had a mud floor.

• Over a quarter of a billion (62%, 272 million) were severely sanitation deprived, lacking any form of toilet facility.

• Over 30 million (7%) were severely water deprived, either using unsafe (open) water sources or having a long walk (over 30 minutes) to collect water.

• 27% of Indian children under five were severely food deprived (severe anthropometric failure).

• 13% of Indian children under five were health deprived, either not being immunized against any diseases or having had an illness causing diarrhea and not receiving any medical advice or treatment.

• 13% of school-aged children (around 34 million) were severely educationally deprived: they reported never having been to school.

• In 2005/06, over half (58%) of India's children (256 million) were living in absolute poverty, that is severely deprived of two or more basic human needs; over 350 million were severely deprived of one or more basic needs.

The indicators were age-appropriate, such that, for example, young, non-school-age children were not considered education deprived. Similarly, anthropometric data are not collected on children six and up, so measures of food deprivation only applied to children five and under. Children of all ages were covered by four household-level indicators: shelter, water, sanitation, and information deprivation. Children under five were also covered by indicators of health and food deprivation, but not education, and older children by an indicator of education deprivation but not food or health deprivation. A more detailed examination of the thresholds and weighting of these indices is provided by Abdu and Delamonica (2018).

### Political Regimes

We derive the political regime measure from Harriss (1999, 2005), as described in the theory section. His typology is one of the few measures describing the interaction of caste and politics in India. A limitation of Harriss's measure is that it does not cover all the states of India. It was developed for the 13 largest states of India, so any statistical treatment effects are limited to a sample average rather than a population (India) average (Morgan and Winship 2014). It is also worth noting the time lag between when Harriss released his typology (in 1999) and the micro-data we are using (2005/06). This lag is useful since the impact of a political regime and the policies it implements takes time to propagate and affect the outcomes (child poverty). Accordingly, our design assumes a five-year lagged effect, where possible anti-poverty policy decisions emerging from the politics of class and caste take around five years to affect outcomes for children. Table 2 shows how the sample in NFHS-3 is populated across Harriss's typology of states.

TABLE 2.

NFHS-3 sample distribution across Harriss's state regime typologies (un-weighted)

No. of children (age under 18)No. of states
A(i): Upper caste/class dominated states 24,791
A(ii): Upper-middle caste/class dominated states 33,880
B: Middle caste/class dominated states 42,476
C: Lower caste/class dominated states 19,841
Total 120,988
No. of children (age under 18)No. of states
A(i): Upper caste/class dominated states 24,791
A(ii): Upper-middle caste/class dominated states 33,880
B: Middle caste/class dominated states 42,476
C: Lower caste/class dominated states 19,841
Total 120,988
Source: National Family Health Survey (NFHS-3).

### Corruption

We derive the two measures of corruption from TII. TII makes sub-national assessments of the perceived extent or degree of corruption in each state and territory of India. We recognize the difficulty of reliably measuring something as latent and contentious as corruption. We therefore use two distinct measures to capture different aspects of corruption. The first is derived from TII's 2005 report (Centre for Media Studies 2005), which is based primarily on the opinions of national and international experts, and their assessments of corruption in India. The second is derived from TII's 2008 report (Centre for Media Studies 2008) and focuses instead on the opinions and experiences of people living below the poverty line. Scholars have shown that expert-based and people-based measures can lead to different results. For example, the OECD Metagora project demonstrated that estimates of corruption based solely on expert opinion often overstated the level of corruption experienced by citizens (OECD 2008).

The TII 2008 measure covers nearly all the states and territories of India (29); the TII 2005 measure covers only 20 states. High scores in the TII 2005 imply greater corruption. Table 3 provides basic descriptive statistics for these key measures.

TABLE 3.

Descriptive statistics for the state variables

TII 2005TII 2008HarrissGiniGDP per capita (Indian rupees)
State cases 20 29 13 29 29
Missing cases 16
Min. 240 0.10 7,914
Max. 695 0.34 76,968
Range 455 0.24 69,054
Median 493.5 0.20 24,086
Mean 488.95 2.34 2.38 0.22 25,025
Std. Dev. 104.77 1.17 1.12 0.07 24,100
TII 2005TII 2008HarrissGiniGDP per capita (Indian rupees)
State cases 20 29 13 29 29
Missing cases 16
Min. 240 0.10 7,914
Max. 695 0.34 76,968
Range 455 0.24 69,054
Median 493.5 0.20 24,086
Mean 488.95 2.34 2.38 0.22 25,025
Std. Dev. 104.77 1.17 1.12 0.07 24,100

Note: Levels of overall corruption in states (involving households below the poverty line) for TII 2008 are 4 = alarming, 3 = very high, 2 = high, and 1 = moderate. Harriss's variable category description is 4 = A(i), 3 = A(ii), 2 = B, and 1 = C. TII 2005 is based on a composition index where a higher number indicates more corruption.

### Statistical Models

We deployed multilevel models to test the four hypotheses. Our data have a hierarchical structure such that children are nested in households, which are nested in National Family Health Health Survey geographical clusters, and finally clusters are nested in states. We used sample weights provided by the NFHS. We fitted both logistic and linear probability models, which yielded similar results. We opted to present the results of the linear probability models, as their interpretation is simpler. We used MLwiN's iterative generalized least-square estimator (Rasbash et al. 2013) and controlled the workflow with R2MLwiN (Zhang et al. 2013) in the R environment (R Development Core Team 2013).

### Procedure and Covariates

We tested the first hypothesis, about the direct effect of political regime on child poverty, in two steps. First, we estimated a bivariate association to identify any potential evidence of the effect of regime on poverty. We then included a set of individual-level covariates to evaluate the sensitivity of this association. These covariates, outlined in Table 4, are standard demographic measures such as child's gender and age, caste of the household (defined as the caste of the head of the household), religion of the household (reported by the head of the household), location (urban or rural) of the household, and adults-to-children ratio (controlling for the size of the household). Our models also include a measure of economic inequality to control for possible confounding between economic and political factors. We produced a Gini coefficient for each state using the household wealth index. The distribution of this Gini variable is outlined in Table 3. If, after this, the Harriss typology still had an effect on child poverty, we concluded that there is statistical evidence relevant to the variations in child poverty across the 13 states of India. We tested the remaining hypotheses in a similar manner. For the interaction hypotheses, we also calculated their marginal effects, to disentangle the direction of effect (Kam and Franzese, 2007).

TABLE 4.

Demographics of the child sample

Overall
n 198,294
Absolute child poverty (mean (sd)) 0.45 (0.50)
sex = female (%) 95,981 (48.4)
caste (%)
Scheduled Caste 34,861 (17.6)
Scheduled Tribe 30,689 (15.5)
Other Backward Class 65,158 (32.9)
none of the above 58,998 (29.8)
don't know 729 (0.4)
not applicable 7,859 (4.0)
religion (%)
nonreligious 102 (0.1)
Buddhism 2,489 (1.3)
Christianity 18,771 (9.5)
Hinduism 136,624 (68.9)
Islam 32,930 (16.6)
Jainism 600 (0.3)
Judaism 8 (0.0)
Sikhism 3,902 (2.0)
Zoroastrianism 5 (0.0)
Don't know or other 2,816 (1.4)
not applicable 47 (0.0)
city or town (%)
countryside 118,329 (59.7)
large city 36,147 (18.2)
small city 12,843 (6.5)
town 30,975 (15.6)
adults-to-children ratio (mean (sd)) 1.21 (0.97)
age (mean (sd)) 8.52 (5.03)
Overall
n 198,294
Absolute child poverty (mean (sd)) 0.45 (0.50)
sex = female (%) 95,981 (48.4)
caste (%)
Scheduled Caste 34,861 (17.6)
Scheduled Tribe 30,689 (15.5)
Other Backward Class 65,158 (32.9)
none of the above 58,998 (29.8)
don't know 729 (0.4)
not applicable 7,859 (4.0)
religion (%)
nonreligious 102 (0.1)
Buddhism 2,489 (1.3)
Christianity 18,771 (9.5)
Hinduism 136,624 (68.9)
Islam 32,930 (16.6)
Jainism 600 (0.3)
Judaism 8 (0.0)
Sikhism 3,902 (2.0)
Zoroastrianism 5 (0.0)
Don't know or other 2,816 (1.4)
not applicable 47 (0.0)
city or town (%)
countryside 118,329 (59.7)
large city 36,147 (18.2)
small city 12,843 (6.5)
town 30,975 (15.6)
adults-to-children ratio (mean (sd)) 1.21 (0.97)
age (mean (sd)) 8.52 (5.03)

## RESULTS

Our intention here is to examine the relationship between different state regimes (using Harriss's typology) and outcomes for children. Table 5 presents descriptive estimates of the proportion of children living in absolute poverty in 2005/06 in each state of India; these are children who experienced severe deprivation of two or more of the basic needs.

TABLE 5.

Absolute poverty among children in India, by state, 2005/06

StateAbsolute poverty (2+ deprivations, %)Included in Harriss's study
Bihar 77.9 ✔
Jharkhand 77.7
Chhattisgarh 77.6
Orissa 71.7 ✔
Rajasthan 66.3 ✔
ALL INDIA 58.1
West Bengal 53.2 ✔
Karnataka 48.6 ✔
Manipur 48.3
Uttaranchal 48.1
Gujarat 47.4 ✔
Haryana 46.2
Assam 45.1
Maharashtra 43.5 ✔
Meghalaya 42.5
Nagaland 41.3
Jammu and Kashmir 38.6
Punjab 29.3 ✔
Tripura 28.4
Sikkim 27.9
Goa 22.0
Mizoram 17.8
Delhi 15.5
Kerala 4.0 ✔
StateAbsolute poverty (2+ deprivations, %)Included in Harriss's study
Bihar 77.9 ✔
Jharkhand 77.7
Chhattisgarh 77.6
Orissa 71.7 ✔
Rajasthan 66.3 ✔
ALL INDIA 58.1
West Bengal 53.2 ✔
Karnataka 48.6 ✔
Manipur 48.3
Uttaranchal 48.1
Gujarat 47.4 ✔
Haryana 46.2
Assam 45.1
Maharashtra 43.5 ✔
Meghalaya 42.5
Nagaland 41.3
Jammu and Kashmir 38.6
Punjab 29.3 ✔
Tripura 28.4
Sikkim 27.9
Goa 22.0
Mizoram 17.8
Delhi 15.5
Kerala 4.0 ✔
Source: Calculated from NFHS3 data.

Figure 1 displays the ordering of states, using the mean number of severe deprivations experienced by children in each state. This ranking is not an unfamiliar one for those who know the Indian context; outcomes are best for children in the southern state of Kerala and worst for children in the more central states of Jharkhand, Madhya Pradesh, and Bihar. States group roughly together into those with low deprivation scores (Kerala, Delhi, Goa, and interestingly, Mizoram), average deprivation scores (Punjab to West Bengal) and high deprivation scores (Jharkhand, Bihar, Madhya Pradesh, Rajasthan, Orissa, and Chhattisgarh).

FIGURE 1.

Mean number of deprivations experienced by children, by state, in India, 2005/06

FIGURE 1.

Mean number of deprivations experienced by children, by state, in India, 2005/06

## DIRECT EFFECTS OF CORRUPTION AND POLITICAL REGIMENS

Table 6 describes the main results of the multilevel model analysis for the Harriss variable. The null model (Model 1) reveals that the total variance of absolute child poverty partitions 11.4% at the state level, 35.5% at the cluster level, 42.0% at the household level, and only 11.2% at the child level.1 An explanation of how these variances are calculated is provided in footnote 1. It is not surprising that such a significant portion of the variance in child poverty is at the household level since much of the NFHS information is collected at this level. This variance partitioning defines how much variance we can expect each set of variables to explain (Steele 2008). Accordingly, the maximum amount of variance we can explain with state-level variables, such as the TII measures or the political regime measure, is 11.7%.

TABLE 6.

Effect of political regimes on child poverty

Model 1Model 2Model 3Model 4Model 5
Intercept 0.37***
(0.03)
0.21***
(0.05)
0.01
(0.20)
−0.02
(0.16)
−0.08
(0.20)
State variables
HarrissB  0.13*
(0.06)
0.14***
(0.03)
0.12***
(0.02)
0.12***
(0.03)
HarrissAii  0.41***
(0.07)
0.22**
(0.07)
0.23***
(0.06)
0.22**
(0.07)
HarrissAi  0.37***
(0.07)
0.08
(0.05)
0.11**
(0.04)
0.09.
(0.05)
GDP20042005   0.00
(0.00)
0.00*
(0.00)
0.00
(0.00)
gini   2.41***
(0.37)
2.41***
(0.28)
2.17***
(0.38)
TII_2008high    −0.09**
(0.03)

TII_2008very_high    −0.05*
(0.03)

TII_2008alarming    −0.01
(0.04)

TII_2005     0.00
(0.00)
Child and household variables
Child sex    0.00***
(0.00)
0.00***
(0.00)
0.00***
(0.00)
casteScheduled tribe   0.05***
(0.01)
0.05***
(0.01)
0.05***
(0.01)
casteOther backward class    −0.09***
(0.00)
−0.09***
(0.00)
−0.09***
(0.00)
casteNone of above   −0.17***
(0.01)
−0.17***
(0.01)
−0.17***
(0.01)
casteDK   −0.06*
(0.02)
−0.06*
(0.02)
−0.06*
(0.02)
religionBuddhism   −0.11
(0.10)
−0.11
(0.10)
−0.11
(0.10)
religionChristianity   −0.05
(0.10)
−0.05
(0.10)
−0.05
(0.10)
religionHinduism   −0.05
(0.10)
−0.05
(0.10)
−0.05
(0.10)
religionIslam   −0.03
(0.10)
−0.03
(0.10)
−0.03
(0.10)
religionJainism   −0.07
(0.10)
−0.07
(0.10)
−0.07
(0.10)
religionJudaism   −0.01
(0.23)
−0.01
(0.23)
−0.01
(0.23)
religionSikhism   −0.12
(0.10)
−0.13
(0.10)
−0.12
(0.10)
religionZoroastrianism   0.05
(0.18)
0.05
(0.18)
0.05
(0.18)
religionDK or Other   −0.03
(0.11)
−0.03
(0.11)
−0.03
(0.11)
citytownLarge city   −0.52 ***
(0.01)
−0.52 ***
(0.01)
−0.52 ***
(0.01)
citytownSmall city   −0.45***
(0.01)
−0.45***
(0.01)
−0.45***
(0.01)
citytownTown   −0.34***
(0.01)
−0.34***
(0.01)
−0.34***
(0.01)
(0.00)
−0.03***
(0.00)
−0.03***
(0.00)
Child age   −0.00***
(0.00)
−0.00***
(0.00)
−0.00***
(0.00)
Var. states 0.03 0.01 0.00 0.00 0.00
Var. clusters 0.09 0.09 0.03 0.03 0.03
Var. housholds 0.10 0.10 0.10 0.10 0.10
Var. children 0.03 0.02 0.02 0.02 0.02
Num. states 29 13 13 13 13
Num. clusters 3850 2278 2276 2276 2276
Num. housholds 80074 49137 48235 48235 48235
Num. children 198294 120988 118796 118796 118796
Deviance stat. 34976.30 6312.15 1419.72 1411.44 1417.72
Model 1Model 2Model 3Model 4Model 5
Intercept 0.37***
(0.03)
0.21***
(0.05)
0.01
(0.20)
−0.02
(0.16)
−0.08
(0.20)
State variables
HarrissB  0.13*
(0.06)
0.14***
(0.03)
0.12***
(0.02)
0.12***
(0.03)
HarrissAii  0.41***
(0.07)
0.22**
(0.07)
0.23***
(0.06)
0.22**
(0.07)
HarrissAi  0.37***
(0.07)
0.08
(0.05)
0.11**
(0.04)
0.09.
(0.05)
GDP20042005   0.00
(0.00)
0.00*
(0.00)
0.00
(0.00)
gini   2.41***
(0.37)
2.41***
(0.28)
2.17***
(0.38)
TII_2008high    −0.09**
(0.03)

TII_2008very_high    −0.05*
(0.03)

TII_2008alarming    −0.01
(0.04)

TII_2005     0.00
(0.00)
Child and household variables
Child sex    0.00***
(0.00)
0.00***
(0.00)
0.00***
(0.00)
casteScheduled tribe   0.05***
(0.01)
0.05***
(0.01)
0.05***
(0.01)
casteOther backward class    −0.09***
(0.00)
−0.09***
(0.00)
−0.09***
(0.00)
casteNone of above   −0.17***
(0.01)
−0.17***
(0.01)
−0.17***
(0.01)
casteDK   −0.06*
(0.02)
−0.06*
(0.02)
−0.06*
(0.02)
religionBuddhism   −0.11
(0.10)
−0.11
(0.10)
−0.11
(0.10)
religionChristianity   −0.05
(0.10)
−0.05
(0.10)
−0.05
(0.10)
religionHinduism   −0.05
(0.10)
−0.05
(0.10)
−0.05
(0.10)
religionIslam   −0.03
(0.10)
−0.03
(0.10)
−0.03
(0.10)
religionJainism   −0.07
(0.10)
−0.07
(0.10)
−0.07
(0.10)
religionJudaism   −0.01
(0.23)
−0.01
(0.23)
−0.01
(0.23)
religionSikhism   −0.12
(0.10)
−0.13
(0.10)
−0.12
(0.10)
religionZoroastrianism   0.05
(0.18)
0.05
(0.18)
0.05
(0.18)
religionDK or Other   −0.03
(0.11)
−0.03
(0.11)
−0.03
(0.11)
citytownLarge city   −0.52 ***
(0.01)
−0.52 ***
(0.01)
−0.52 ***
(0.01)
citytownSmall city   −0.45***
(0.01)
−0.45***
(0.01)
−0.45***
(0.01)
citytownTown   −0.34***
(0.01)
−0.34***
(0.01)
−0.34***
(0.01)
(0.00)
−0.03***
(0.00)
−0.03***
(0.00)
Child age   −0.00***
(0.00)
−0.00***
(0.00)
−0.00***
(0.00)
Var. states 0.03 0.01 0.00 0.00 0.00
Var. clusters 0.09 0.09 0.03 0.03 0.03
Var. housholds 0.10 0.10 0.10 0.10 0.10
Var. children 0.03 0.02 0.02 0.02 0.02
Num. states 29 13 13 13 13
Num. clusters 3850 2278 2276 2276 2276
Num. housholds 80074 49137 48235 48235 48235
Num. children 198294 120988 118796 118796 118796
Deviance stat. 34976.30 6312.15 1419.72 1411.44 1417.72
***

p < 0.001

**

p < 0.01

**

p < 0.05

.

p < 0.1

Political regime accounts for almost 70% of the state-level variance in a bivariate model (Model 2). Seventy percent is a lot, and confirms that political regime is a key explanatory variable of absolute child poverty. All the coefficients are significantly different from zero. This is noteworthy, given the small sample of states (13). This model indicates that children residing in states with regime types A(i) and A(ii)—states dominated by higher caste and class groups—are 37% and 41%, respectively, more likely to be poor compared to regime type C, which is the reference category. Children living in type B states were about 13% more likely to be poor than children in type C states. This finding suggests an explanatory order, with states dominated by higher caste and class groups associated with the most child poverty, followed by states dominated by middle caste and class groups.

These stark regime effects diminish when we control for confounders such as child's age and sex, religion, place of residence, state GDP per capita, and economic inequality (Gini). While the effects of the controls are not the primary focus of this paper, we comment on them briefly. The models (Models 3 to 5) do not detect any significant child poverty differences between religious groups. Children in Scheduled Tribes and Scheduled Castes are, not surprisingly, the most likely to be poor. Children in Scheduled Tribes have a 5% higher probability of being poor compared to the reference group, of Scheduled Caste children (Model 3). The other higher-caste categories are less likely to be poor compared to the reference. The effect of children's sex differences is significant but negligible; boys have on average a 0.4% lower probability of being poor compared to girls. Older children are less likely to be poor than younger children. Within-state economic inequality correlated with prevalence of child poverty throughout all models. State wealth, or state GDP per capita, which is usually considered a strong determinant of poverty, is insignificant in most models. To validate this finding, we conducted a Farrar-Glauber multicollinearity test between GDP, Gini, and the TII 2005 measure. The test showed no excessive correlations.

Both type B and type A(ii) states remain significant after controlling for confounders (Models 3 to 5). Middle caste/class dominated states (type B) correlate with the least detrimental effect compared to states with low caste/class dominated states (type C, the reference). B-type states have about 12% more children in poverty compared to type C, and type A(ii) states (upper-middle caste/class dominated states) have about 22% more than type C. These models cannot detect robust results for upper caste/class dominated states—type A(i). Model 2 finds that as many as 37% poor children live in these states, whereas Model 3 can only detect 8%, a non-significant effect. In summary, this evidence supports our first hypothesis, but with a reservation for the effect of type A(i) states.

With regard to the second hypothesis—whether it is political regime or corruption which is more important for explaining state differences in child poverty—Models 4 and 5 show that the effect of political regime remains stable and dominates over corruption. The TII 2008 measure shows that corruption has some effect, although the direction points in a counter-intuitive direction. We do not regard these effects as stable, as these effects only appear when we control for economic inequality (Gini), indicating a suppressed effect. The regime variable remains stable, however. The TII 2005 measure effect (Model 5) is indistinguishable from zero, and neither does this corruption measure change the political regime effect.

### Interactive Effects of Corruption and Political Regimes

In the last modeling step, we test the interaction hypotheses. This step checks whether regime type and level of corruption, respectively, interact with economic wealth in explaining variations in child poverty.

Table 7 shows the key interaction effects of three models: the political regime measure and the TII 2008 and TII 2005 corruption measures. We assessed the marginal effects of all models to analyze how GDP moderates the effect of regime type and corruption on child poverty (Kam and Franzese 2007).

TABLE 7.

Results of interaction models: political regime, TII 2005, and TII 2008

Political regime
Coeff.[95% conf. interval]Std. err.zPr(>|z|)
Ai −0.48 −0.94 0.03 0.25 −1.84 0.07.
Aii −0.61 −1.18 −0.11 0.27 −2.37 0.02*
−1.03 −1.65 −0.50 0.29 −3.66 0.00***
GDP20042005 0.00 0.00 0.00 0.00 −3.48 0.00***
GDP20042005:Ai 0.00 0.00 0.00 0.00 1.82 0.07.
GDP20042005:Aii 0.00 0.00 0.00 0.00 2.41 0.02*
GDP20042005:B 0.00 0.00 0.00 0.00 3.54 0.00***

TII 2008
Coeff. [95% conf. interval] Std. err. z Pr(>|z|)
alarming 0.35 −0.05 0.75 0.21 1.69 0.09.
high 0.29 −0.14 0.72 0.22 1.33 0.18
very_high 0.62 0.07 1.16 0.28 2.23 0.03*
GDP20042005 0.00 0.00 0.00 0.00 0.77 0.44
GDP20042005:alarming 0.00 0.00 0.00 0.00 −1.51 0.13
GDP20042005:very_high 0.00 0.00 0.00 0.00 −2.27 0.02*
GDP20042005:high 0.00 0.00 0.00 0.00 −1.04 0.30

TII 2005
Coeff. [95% conf. interval] Std. err. z Pr(>|z|)
TII_2005 0.00 0.00 0.00 0.00 −0.23 0.82
GDP20042005 0.00 0.00 0.00 0.00 −1.28 0.20
TII_2005:GDP20042005 0.00 0.00 0.00 0.00 1.13 0.26
Political regime
Coeff.[95% conf. interval]Std. err.zPr(>|z|)
Ai −0.48 −0.94 0.03 0.25 −1.84 0.07.
Aii −0.61 −1.18 −0.11 0.27 −2.37 0.02*
−1.03 −1.65 −0.50 0.29 −3.66 0.00***
GDP20042005 0.00 0.00 0.00 0.00 −3.48 0.00***
GDP20042005:Ai 0.00 0.00 0.00 0.00 1.82 0.07.
GDP20042005:Aii 0.00 0.00 0.00 0.00 2.41 0.02*
GDP20042005:B 0.00 0.00 0.00 0.00 3.54 0.00***

TII 2008
Coeff. [95% conf. interval] Std. err. z Pr(>|z|)
alarming 0.35 −0.05 0.75 0.21 1.69 0.09.
high 0.29 −0.14 0.72 0.22 1.33 0.18
very_high 0.62 0.07 1.16 0.28 2.23 0.03*
GDP20042005 0.00 0.00 0.00 0.00 0.77 0.44
GDP20042005:alarming 0.00 0.00 0.00 0.00 −1.51 0.13
GDP20042005:very_high 0.00 0.00 0.00 0.00 −2.27 0.02*
GDP20042005:high 0.00 0.00 0.00 0.00 −1.04 0.30

TII 2005
Coeff. [95% conf. interval] Std. err. z Pr(>|z|)
TII_2005 0.00 0.00 0.00 0.00 −0.23 0.82
GDP20042005 0.00 0.00 0.00 0.00 −1.28 0.20
TII_2005:GDP20042005 0.00 0.00 0.00 0.00 1.13 0.26
***

p < 0:001

**p < 0:01

**

p < 0.05

.

p < 0.1.

Note: All figures are taken from an interactive fully specified model, with absolute child poverty as dependent variable. Only relevant numbers are shown.

We find that GDP does not moderate the effect of corruption on child poverty. The TII 2005 model shows no significant interaction effect with GDP. The TII 2008 model suggests a statistically significant effect. However, when investigating the marginal effect plot (not shown here), the effect extrapolates beyond our sample of wealth distribution. Therefore, we conclude that this effect is not substantively interesting.

Depicting the marginal effect of political regime reveals two key facts. Figure 2 shows that political regime has a positive marginal effect for all regime types, although with mixed statistical significance. First, the direction of the effect is unexpected, given the assumptions of the trickle-down theory, which underlies the formulation of the interaction hypotheses. A positive marginal effect means the effect of political regime is increasingly adverse as GDP per capita increases. When the effect is negative (y-axis), it implies that political regime has a beneficial effect. However, as states’ GDP per capita rises, that beneficial effect is eroded. When the effect is positive, it implies an adverse effect—that is, as GDP per capita increases, the adverse effect is amplified. Accordingly, an amplified positive effect would mean that higher caste/class dominated regimes (A(i), A(ii), and B) produce more poverty (compared to the reference regime C) as the state's wealth increases. Thus, as the wealth of a state increases, the effect of higher caste/class dominated regimes hampers combating poverty, given that they control access to public services and resources and thus limit what resources can trickle down to the already deprived and marginalized groups in that state.

FIGURE 2.

Marginal effect plots for Harriss (regime type C set as referent in each instance)

FIGURE 2.

Marginal effect plots for Harriss (regime type C set as referent in each instance)

Second, the marginal effect plots show that the interactive effects are mainly relevant for regime types A(ii) and B, in comparison with C. For type A(ii) states, the lower GDP per capita interval is barely different from zero, but for higher incomes (from Rs. 22,000 to just below Rs. 40,000) there is both a moderate statistical and substantive effect. This result implies that children living in richer A(ii) states are more likely to be disadvantaged compared to children living in lower caste and class states (regime C) with the same state wealth level.

For states dominated by regime type B, the result is more pronounced but mixed. On the one hand, lower wealth produces a marginal effect that is beneficial for combating child poverty (negative y-axis values). On the other hand, at wealth levels above Rs. 27,000 this same effect switches into adverse impact. As the wealth of a state increases, it will be less and less beneficial for the poor, eventually even exacerbating their living conditions.

The results are weak for regime type A(i). The lower range of the wealth distribution (GDP per capita Rs. 12,000 to Rs. 30,000) is statistically insignificant, and the upper range of the distribution is significant but extrapolates beyond our sample. The political-regime-and-wealth interaction model explains most of the state-level variance of child poverty: almost 90% compared to the null model. It would be hard to improve the model performance further, with this data. Nevertheless, due to the small state-level sample size, we advise that these effects apply mainly to the sample average treatment effect.

## DISCUSSION

This paper has examined the statistical relationship between Harriss's typology of state political regimes, corruption, and the prevalence of child poverty in India. Using NFHS data, we tested four hypotheses:

1. Indian states politically dominated by higher or middle caste/classes have more child poverty;

2. Political regime type has greater importance than corruption;

3. Political regimes dominated by higher or middle caste groups in wealthier states (i.e., higher state GDP per capita) have less poverty; and

4. More corruption results in more child poverty in wealthier states.

Our bivariate model supports hypothesis 1 across all three regime types, with adverse effects relative to type C states (i.e., in which lower castes/classes are more strongly represented in political regimes and where the Congress lost its dominance at an early stage). In controlled models, these results are robust for two regime types: B (middle caste/class dominated regimes, where the Congress has been effectively challenged but has not collapsed, and there is fairly stable and mainly two-party competition) and A(ii) (in which upper caste/class dominance has been effectively challenged by middle castes/classes). Accordingly, this hypothesis enjoys empirical support in states where the politics of accommodation in relation to lower class interests have continued to work effectively.

Concerning hypothesis 2, the political regime measure is the strongest predictor of child poverty. It remains stable and statistically significant across the two corruption measures. TII 2005 exhibits no effect at all. However, asking the poor about corruption (TII 2008) yielded a different result than asking experts (TII 2005). An explanation for this discrepancy could have to do with the types of corruption these two groups encounter. The poor report the types of corruption they experience in their everyday life: for example, in dealings with patrolling police, administrative officials (e.g., hospitals), and other authorities. We could call this petit corruption (Rothstein 2014). Experts, on the other hand, might be reporting higher-level, institutional types of corruption, which affect the broader functioning of political and economic systems, effects that are less tangible for poor and vulnerable families.

Our findings for the first two hypotheses resonate with Drèze and Sen's (2014) account. In considering the issue of political and bureaucratic accountability in India, they note that “corruption flourishes in informational darkness” (p. 96). Legislative changes in India, such as the 2005 Right to Information Act, in combination with wider social and technological changes, have resulted in more light being shed on previously hidden organizational and bureaucratic processes. The public are increasingly aware of their rights, of instances of corruption, and of where bureaucratic bottlenecks form to block the provision of basic services. Such changes have had a positive impact: the 2011 India Corruption Study showed a decline in the proportion of Indians who felt that corruption had increased, from 70% in 2005 to 45% in 2010; there was near five-fold increase (from 6% to 29%) in the proportion who felt that corruption had decreased. Importantly, the proportion of rural households who reported having paid bribes in the previous year fell, from 56% in 2005 to 28% in 2010. While corruption is still a problem, its apparent retreat in recent years is welcome. However, the politics of caste and class remain, with no significant improvements (Harriss 2005, 2013).

The evidence collected to evaluate hypotheses 3 and 4 demonstrates the depth of the class and caste issues. Cooperation between social strata should lead to higher joint prosperity. Thus, as the theory of trickle-down economics argues, more wealth should benefit all members of society (Corbridge, Harriss, and Jeffrey 2013). Yet the conflicts and dynamics between these social strata demonstrate a different trend in India. We find that political regime types dominated by higher castes, and as state wealth increases, will led to more child poverty. This finding has several possible explanations. One might be that as the wealth of a state increases the poor are increasingly marginalized and less able to make their voices heard on the political agenda. For example, the resurgence of Maoist movements in states like Bihar, Jharkhand, and Andhra Pradesh testifies to such processes. The benefits of economic growth seem to turn into a liability, and the benefits from the years of “India Shining” clearly had not reached large sections of the population by 2005, when the survey data were collected.

Even if they could put forth their interest in politics, another argument says that large economic inequality effectively blocks the lasting social mobility of the poor (Vanneman and Dubey 2014). This explanation is consistent with our statistical findings. Greater economic inequality is strongly associated with greater child poverty. It also a possible confounder of political regime. For it seems unlikely that more economic wealth would cause more poverty—a possible interpretation of our interaction models—rather, greater collective economic wealth implies greater economic inequality between households, and this inequality blocks poorer households from pulling themselves out of poverty (Sonalde and Vanneman 2005). This intimate relationship between political regime, economic inequality, and poverty suggests a line for future research.

For now, based on our findings about 13 Indian states, we suggest that the politics of caste amid inequality trump those of corruption in explaining absolute child poverty (Harriss 2013). Social prejudices continue to divide Indian society into distinct groups—deserving and undeserving. Researchers have shown that children from lower-caste groups are already excluded from the benefits of India's impressive economic growth (Drèze and Sen 2014). What this paper shows is that greater wealth and inequality can exacerbate this exclusion. What is needed are policies based on social justice and equal rights for children, to enable them to flourish regardless of where they are born (Drèze and Khera 2017).

## REFERENCES

REFERENCES
Abdu, M., and E. Delamonica.
2018
.
“Multidimensional Child Poverty: From Complex Weighting to Simple Representation.”
Social Indicators Research
136
:
881
905
.
.
2008
.
Public Order: Fifth Report of the Second Administrative Reforms Commission
.
New Delhi
:
Government of India
.
Alkire, Sabina, and Suman Seth.
2015
.
“Multidimensional Poverty Reduction in India between 1999 and 2006: Where and How?”
World Development
72
:
93
108
.
Boix, C.
2003
.
Democracy and Redistribution
.
Cambridge
:
Cambridge University Press
.
Bueno de Mesquita, B., A. Smith, R. M. Silverson, and J. D. Morrow.
2003
.
The Logic of Political Survival
.
Cambridge, MA
:
MIT Press
.
Cain, J. Salcedo, Rana Hasan, Rhoda Magsombol, and Ajay Tandon.
2010
.
“Accounting for Inequality in India: Evidence from Household Expenditures.”
World Development
38
(
3
):
282
97
.
Cavatorta, Elisa, Bhavani Shankar, and Artemisa Flores-Martinez.
2015
.
“Explaining Cross-State Disparities in Child Nutrition in Rural India.”
World Development
76
:
216
37
.
Centre for Media Studies
.
2005
.
India Corruption Study 2005: To Improve Governance
.
New Delhi
:
Transparency International India
.
Centre for Media Studies
.
2008
.
India Corruption Study 2008: With Special Focus on BPL Households
.
New Delhi
:
Transparency International India
.
Corbridge, S., J. Harriss, and C. Jeffrey.
2013
.
India Today: Economy, Politics and Society
,
Cambridge
:
Polity Press
.
Corsi, D., M. Neuman, J. Finlay, and S. Subramanian.
2012
.
“Demographic and Health Surveys: A Profile.”
International Journal of Epidemiology
41
:
1602
13
.
2007
.
“(Quasi)Scarcity and Global Hunger: A Sociological Critique of the Scarcity Postulate with an Effort to Synthesis.”
Journal of Critical Realism
6
(
2
):
199
225
.
2010
.
“Robbins and Malthus on Scarcity, Abundance, and Sufficiency: The Missing Sociocultural Element.”
American Journal of Economics and Sociology
69
(
4
):
1206
29
.
2011
.
Scarcity, Abundance, and Sufficiency: Contributions to Social and Economic Theory
.
Gothenburg
:
Gothenburg Studies in Sociology, Department of Sociology, and Geson Hyltetryck
.
2015a
. “Scarcity and Artificial Scarcity.” In
The Wiley Blackwell Encyclopedia of Consumption and Consumer Studies
.
John Wiley & Sons
. Retrieved
May
7
,
2016
(http://onlinelibrary.wiley.com/doi/10.1002/9781118989463.wbeccs202/abstract).
2015b
.
“Quality of Governance, Corruption, and Absolute Child Poverty in India.”
Journal of South Asian Development
10
(
2
):
1
20
.
2017
.
“A Framework for Synthesizing the Malthusian and Senian Approaches: Exemplified by the 1943 Bengal Famine.”
Cambridge Journal of Economics
42
(
2
):
453
76
doi:.
2018
.
“Unifying Studies of Scarcity, Abundance, and Sufficiency”
.
Ecological Economics
147
(
May
):
208
17
. doi:.
Daoud, Adel, Björn Halleröd, and Debarati Guha-Sapir.
2016
.
“What Is the Association between Absolute Child Poverty, Poor Governance, and Natural Disasters? A Global Comparison of Some of the Realities of Climate Change.”
PLOS One
11
(
4
):
e0153296
.
2017
.
“Impact of International Monetary Fund Programs on Child Health.”
Proceedings of the National Academy of Sciences
114
(
25
):
6492
97
.
Datt, G., and M. Ravallion.
1998
.
“Why Have Some Indian States Done Better than Others at Reducing Rural Poverty?”
Economica
65
:
17
38
.
Deaton, A., and J. Dreze.
2002
.
“Poverty and Inequality in India: A Re-examination.”
Economic and Political Weekly
,
September
7
,
3729
48
.
Dhongde, Shatakshee.
2017
.
“Measuring Segregation of the Poor: Evidence from India.”
World Development
89
:
111
23
.
Drèze, Jean, and Reetika Khera.
2017
.
“Recent Social Security Initiatives in India.”
World Development
98
:
555
72
.
Drèze, Jean, and A. Sen.
2014
.
An Uncertain Glory: India and Its Contradictions
.
New Delhi
:
Penguin
.
2014
.
“Caste Comparisons in India: Evidence from Subjective Well-Being Data.”
World Development
64
:
407
19
.
Ghobarah, H. A., P. Huth, and B. Russett.
2004
.
“Comparative Public Health: The Political Economy of Human Misery and Well-Being.”
International Studies Quarterly
48
:
73
94
.
Gordon, David, and Shailen Nandy.
“The Extent, Nature and Distribution of Child Poverty in India.”
Indian Journal of Human Development
. First published
July
25
,
2016
. https://doi.org/10.1177/0973703016654561.
Gordon, D., S. Nandy, C. Pantazis, S. Pemberton, and P. Townsend.
2003
.
Child Poverty in the Developing World
.
Bristol
:
Policy Press
.
Government of India Planning Commission
.
2012
.
“Press Note on Poverty Estimates, 2009–10.”
Released
March
19
,
2012
(http://www.planningcommission.nic.in/news/press_pov1903.pdf).
Gragnolati, M., M. Shekar, M. Das Gupta, C. Bedenkamp, and Y. Lee.
2005
. “India's Undernourished Children: A Call for Reform and Action.” HNP Discussion Paper,
World Bank
,
Washington, DC
.
Gupta, S., H. Davoodi, and R. Alonso-Terme.
2002
.
“Does Corruption Affect Income Inequality and Poverty?”
Economics of Governance
3
:
23
45
.
Halleröd, B., B. Rothstein, A. Daoud, and S. Nandy.
2013
.
“Bad Governance and Poor Children: A Comparative Analysis of Government Efficiency and Severe Child Deprivation in 68 Low- and Middle-Income Countries.”
World Development
48
:
19
31
.
Harriss, J.
1999
.
Comparing Political Regimes Across Indian States: A Preliminary Essay
.
Economic and Political Weekly
34
:
3367
77
.
Harriss, J.
2000
.
“How Much Difference Does Politics Make? Regime Differences across Indian States and Rural Poverty Reduction.”
Development Studies Institute, London School of Economics.
Harriss, J.
2005
. “Do Political Regimes Matter? Poverty Reduction and Regime Differences across India.” In
Changing Paths: International Development and the New Politics of Inclusion
, edited by P. P. Houtzager and M. Moore.
Ann Arbor
:
University of Michigan Press
.
Harriss, J.
2013
.
“Does ‘Landlordism’ Still Matter? Reflections on Agrarian Change in India.”
Journal of Agrarian Change
13
:
351
64
.
Heidenheimer, A., and M. Johnston.
2002
.
Political Corruption: Concepts and Contexts
.
New Brunswick, NJ
:
Transaction
.
IIPS and Macro International
2007
.
National Family Health Survey (NFHS-3) 2005–06: India: Volume 1
.
Mumbai
:
International Institute for Population Sciences
.
Kaletski, Elizabeth, and Nishith Prakash.
2016
.
“Does Political Reservation for Minorities Affect Child Labor? Evidence from India.”
World Development
87
:
50
69
.
Kam, C. D., and R. J. Franzese.
2007
.
Modeling and Interpreting Interactive Hypotheses in Regression Analysis
.
Ann Arbor
:
University of Michigan Press
.
Kaufmann, D., A. Kraay, and P. Zoido-Lobaton.
2000
.
“Governance Matters: From Measurement to Action.”
Finance and Development
37
(
2
):
10
13
.
Keefer, P., and S. Khemani.
2005
.
“Democracy, Public Expenditures, and the Poor: Understanding Political Incentives for Providing Public Services.”
World Bank Research Observer
20
:
1
27
.
Khan, M.
2006
. “Governance and Anti-Corruption Reforms in Developing Countries: Policies, Evidence and Ways Forward.” G-24 Discussion Paper.
Geneva
:
United Nations Conference on Trade and Development
.
Kim, Rockli, Sanjay K. Mohanty, and S. V. Subramanian.
2016
.
“Multilevel Geographies of Poverty in India.”
World Development
87
:
349
59
.
Kohli, A.
1989
.
The State and Poverty in India: The Politics of Reform
.
Cambridge
:
Cambridge University Press
.
2016
.
“Routine and Reflexivity: Simonian Cognitivism vs Practice Approach.”
Industrial and Corporate Change
26
(
4
):
727
43
.
Kwon, H., and E. Kim.
2014
.
“Poverty Reduction and Good Governance: Examining the Rationale of the Millennium Development Goals.”
Development and Change
45
:
353
75
.
Mauro, P.
2002
. “The Effects of Corruption on Growth and Public Expenditure.” In
Political Corruption: Concepts and Contexts
, edited by A. Heidenheimer and M. Johnston.
New Brunswick
:
Transaction
.
Meltzer, A. H., and S. F. Richard.
1981
.
“A Rational Theory of the Size of Government.”
Journal of Political Economy
89
:
914
27
.
Minujin, A., and S. Nandy (eds.).
2012
.
Global Child Poverty and Well-Being: Measurement, Concepts, Policy and Action
.
Bristol
:
Policy Press
.
Morgan, S. L., and C. Winship.
2014
.
Counterfactuals and Causal Inference: Methods and Principles for Social Research
, 2nd ed.
New York
:
Cambridge University Press
.
Mundle, S., P. Chakraborty, S. Chowdhury, and S. Sikdar.
2012
.
“The Quality of Governance.”
Economic and Political Weekly
47
(
49
):
41
52
.
Myroniuk, Tyler W., Reeve Vanneman, and Sonalde Desai.
2017
.
“Getting a Child through Secondary School and to College in India: The Role of Household Social Capital.”
Sociology of Development
3
(
1
):
24
46
.
Nandy, Shailen, Adel Daoud, and David Gordon.
2016
.
“Examining the Changing Profile of Undernutrition in the Context of Food Price Rises and Greater Inequality.”
Social Science & Medicine
149
:
153
63
.
Nayyar, R.
1991
.
Rural Poverty in India: An Analysis of Inter-state Differences
.
Bombay
:
Oxford University Press
.
OECD
2008
.
“Measuring Human Rights and Democratic Governance: Experiences and Lessons from Metagora.”
OECD Journal on Development
9
(
2
).
2011
.
“Vilja och framtid i frågor kring utbildningsval” [Students’ aspiration and their perception of the future in education choices]
.
Pedagogisk Forskning i Sverige
16
(
2
):
100
.
R Development Core Team
2013
.
R: A Language and Environment for Statistical Computing
.
Vienna
:
R Foundation for Statistical Computing
(http://www.r-project.org/).
Rasbash, J., C. Charlton, W. J. Browne, M. Healy, and B. Cameron.
2013
.
“MLwiN Version 2.28.”
Centre for Multilevel Modelling, University of Bristol.
Ross, M.
2006
.
“Is Democracy Good for the Poor?”
American Journal of Political Science
50
:
860
74
.
Rothstein, B.
2014
.
“What is the Opposite of Corruption?”
Third World Quarterly
35
:
737
52
.
Rothstein, B., and E. M. Uslaner.
2005
.
“All for All: Equality, Corruption, and Social Trust.”
World Politics
58
:
41
72
.
Saith, A.
2005
.
Poverty Lines versus the Poor: Method versus Meaning
.
The Hague
:
Institute of Social Studies
.
Saxena, N. C.
2012
.
.
New Delhi
:
National Social Watch
.
Sen, A.
2000
.
“Democracy: The Only Way Out of Poverty.”
New Perspectives Quarterly
17
:
28
30
.
Sonalde, Desai, and Reeve Vanneman.
2005
.
India Human Development Survey (IHDS)
. Inter-university Consortium for Political and Social Research.
Steele, F.
2008
.
“Module 5: Introduction to Multilevel Modelling Concepts, LEMMA VLE.”
Centre for Multilevel Modelling, University of Bristol.
.
2015
.
“India GDP Annual Growth Rate,”
accessed
June
12
,
2015
UNGA
2010
.
“Keeping the Promise: United to Achieve the Millennium Development Goals Resolution.”
65th Session, United Nations General Assembly.
UNICEF
.
2007
.
Global Study on Child Poverty and Disparity 2007–2008: GUIDE
.
New York
:
Global Policy Section, Division of Policy and Planning, UNICEF
.
United Nations
.
1995
.
“The Copenhagen Declaration and Programme of Action: World Summit for Social Development 6–12 March 1995.”
New York.
von der Hoven, R.
2012
.
“MDGs Post 2015: Beacons in Turbulent Times or False Lights?”
System Task Team on the Post-2015 UN Development Agenda, United Nations.
Vanneman, Reeve, and Amaresh Dubey.
2014
. “Horizontal and Vertical Inequalities in India.” In
Income Inequality: Economic Disparities and the Middle Class in Affluent Countries
, edited by J. C. Gornick and M. Jäntti.
Stanford, CA
:
Stanford University Press
.
World Bank
.
1991
.
Managing Development: The Governance Dimension
.
Washington, DC
.
World Bank
.
2001
.
World Development Report 2000/01: Attacking Poverty
.
Washington, DC
.
Zhang, Z., C. M. J. Charlton, R. M. A. Parker, G. B. Leckie, and W. J. Browne.
2013
.
“R2MLwiN v0.1-7.”
Centre for Multilevel Modelling, University of Bristol.

## NOTE

NOTE
1.
These variances are calculated by dividing the variance of interest (e.g., $σstate2=0.0276$ rounded to 0.03 in Table 6, model 1) by the sum of all portioned variances ($σall2=0.0276+0.09+0.1+0.03=0.248$).