Leaving No One Behind: Multidimensional Child Poverty in Botswana

Child poverty measurement is vital for informing policies and for improving children’s lives. Nevertheless, efforts to measure (child) poverty remain dominated mainly by monetary approaches, and many countries fail to monitor multidimensional child poverty. Using the 2015/2016 Botswana multi-topic household survey, this study developed a child-centred, individual-level and composite measure that offers nationally relevant and context-specific insights into the magnitude and depth of multidimensional child poverty in Botswana. In particular, it did so through the lens of Leave No One Behind (LNOB) by zooming in on demographic, economic and geographical characteristics that may be associated with greater vulnerability or marginalisation using both descriptive and regression analysis. Results point towards a relatively high incidence and depth of multidimensional child poverty in Botswana. Results show that disabled children, orphans, children living in larger families, families headed by unmarried couples and living in rural areas are more likely to be multidimensionally poor.

direct environment for the provision of basic needs that are essential for their development (Trani and Cannings 2013). Second, children experience poverty and are affected by it differently compared to adults (Leu et al. 2016). Third, a child-centred approach to poverty measurement is vital for ensuring that commitment to children's rights is monitored (Leu et al. 2016).
Following widespread acknowledgement of these arguments, there is now a robust literature on multidimensional child poverty measurement. The pioneering crosscountry study by Gordon et al. (2003) gave rise to country-level studies (e.g. Amarante et al. 2010;Roche 2013;Roelen et al. 2010) in the early 2000s and ultimately paved the way for UNICEF's Multiple Overlapping Deprivation Analysis (MODA) (de Neubourg et al. 2012). More recently, the increasingly adopted global measure of multidimensional povertythe Multidimensional Poverty Index (MPI)has been disaggregated and adjusted to analyse the situation through a child-focused lens (Alkire et al. 2017. At present, much debate regarding multidimensional child poverty measures focuses on the comparative merits of the MODA and MPI approaches (e.g. Hjelm et al. 2016;Vaz et al. 2019). MODA places the child at the centre of analysis by including individual-level indicators and incorporating the child as a unit of analysis. MPI allows for the calculation of a composite index that offers insights into the scale and magnitude of the issue of multidimensional child poverty.
Notwithstanding these debates and widespread efforts to measure poverty from a multidimensional perspective, monetary measures remain dominant, and child poverty receives relatively limited attention (Global Coalition to End Child Poverty 2019). Botswana is a case in point. Poverty in Botswana has been almost exclusively measured using the traditional monetary approach, and little attention has been paid to child poverty in specific. However, in alignment with the Sustainable Development Goals (SDGs), Botswana's new Poverty Eradication Policy and Strategy (BPEPS), includes recommendations to reduce the intergenerational transmission of poverty through the eradication of severe multidimensional child poverty (Republic of Botswana [RoB] 2018). 1 This commitment is further underpinned by the principle of LNOB, thereby highlighting the need to include all children in efforts to reduce poverty. The commitment to the eradication of multidimensional child poverty in conjunction with the LNOB principle calls for a comprehensive country-and context-specific measure of multidimensional child poverty for Botswana.
The objective of this study is twofold. First, it aims to extend the current field of child poverty measurement by building on two dominant approaches, namely MODA and MPI. It builds on the comparative advantages of these approaches by developing a child-centred, individual-level and composite measure that offers nationally relevant and context-specific insights into magnitude and depth of multidimensional child poverty in Botswana. Second, and relatedly, it seeks to provide empirical insights into the state of multidimensional child poverty in Botswana through the lens of LNOB. Findings will serve as a baseline study in Botswana to track progress towards SDG 1 and national development plans regarding eradication of multidimensional child poverty and LNOB.
The remainder of the paper is organised as follows: Section 2 offers an overview of multidimensional child poverty measurement, including background on the situation of children and poverty in Botswana. Section 3 presents data and methodology. Section 4 provides results, and Section 5 presents conclusions and policy implications.
2 Multidimensional Child Poverty Measurement Gordon et al. (2003) pioneered the first global study on child poverty to compare multidimensional child poverty across developing countries. The approach is also referred to as the Bristol approach. Since then, the importance of measuring child poverty from a multidimensional perspective has been recognised (e.g., Roelen and Gassmann 2008;Roelen et al. 2009Roelen et al. , 2010. The global significance of child poverty has also been recognised in the SDGs. Specifically, SDG target 1.2 calls for "reducing at least half the proportion of men, women and children of all ages living in poverty in all its dimensions according to national definitions" by 2030 (UN 2015). This target presents a significant step forward in the fight against child poverty in three ways: firstly, it explicitly recognises children; secondly, it acknowledges the multidimensional nature of poverty; and thirdly, it highlights the importance of national definitions.
Child-focused approaches are, therefore required to adequately analyse and investigate the issue of child poverty (Roelen 2010). Several approaches to measuring multidimensional child poverty have been employed in the empirical literature. Dominant approaches include the MPI (Alkire and Foster 2011) and MODA (de Neubourg et al. 2012). The MPI is theoretically premised on the capability approach and is methodologically grounded in the Alkire-Foster (AF) approach (Alkire and Foster 2011). The AF approach is flexible as it allows for the inclusion of different dimensions, indicators, cut-offs and dimensional weights that reflect the relative importance of each dimension (Alkire et al. 2015) and reflects deprivations into a single measure (Maasoumi and Yalonetzky 2013). The global MPI represents an application of the AF approach using three dimensions and ten indicators, adopting equal weighting at dimension level and using a proportion of weighted deprivations as the cut-off for being multidimensionally poor (Alkire and Jahan 2018). Crucially, the indicators within the global MPI are all household-level indicators. Estimates of multidimensional child poverty are based on a simple decomposition of overall poverty estimates for children (see Alkire et al. 2017). Many studies of multidimensional child poverty are methodologically premised on the AF approach (e.g., Roche 2013;Trani and Cannings 2013;Roelen 2010).
The MODA approach was developed by UNICEF (de Neubourg et al. 2012) to facilitate the analysis of inequities and to provide instruments to identify deprived children and is rooted in the rights-based framework of the Convention on the Rights of the Child (CRC) (Hjelm et al. 2016). The approach builds on the Bristol and AF approaches (Hjelm et al. 2016;de Neubourg et al. 2012) as it combines both household-and individual-level indicators and considers the interaction and depth of deprivations across indicators and dimensions. Notably, the MODA approach does not advocate for the construction of a composite index. Instead, MODA presents poverty figures using all possible cut-offs based on the number of dimensions across specific age categories. Its primary focus is more on overlap analysis than aggregating into a composite index. This approach has been implemented to study multidimensional child poverty across countries (de Milliano and Plavgo 2014;Chzhen et al. 2016) and within countries (e.g. de Neubourg et al. 2015;Chzhen and Ferrone 2017).

Children and Poverty in Botswana
Botswana is an upper-middle-income country that has witnessed rapid economic growth for most of its post-independence period since 1966 . Economic growth went hand-in-hand with declining poverty rates. Monetary poverty rates, based on national poverty lines, reduced from 59% in 1985/1986to 16.3% in 2015, 2018. 2 Impressive as they are, these figures only paint a partial picture of the poverty situation in the country. With respect to children, poverty incidence is higher than it is for adults. According to Statistics Botswana, child poverty stood at 20.1% compared to 13.8% for adults (SB 2018). However, these monetary figures do not depict the multifaceted deprivation suffered by children.
Other indicators suggest that many of Botswana's children experience deprivation in one or more areas of their lives. Although infant and child mortality rates have dropped considerably in the last four decades (SB 2016), HIV/AIDS infection rates continue to increase over time (SB 2014). Similarly, malnutrition and maternal mortality rates remain at high levels. Chronic malnutrition (stunting) and wasting stood at 30% and 11.6% in 2013, while the maternal mortality rate was 147/100,000 live births in 2012 . With respect to education, Botswana's enrolment rates have been consistently below those of other upper-middle-income countries, such as Algeria, Mauritius, Namibia and South Africa (World Bank 2015).
As noted above, Botswana has made strong commitments to eradicating child poverty, including multidimensional forms of child poverty. In line with the SDGs, Botswana developmental initiatives, such the National Development Plan 11, Vision 2036 and the BPEPS, articulate the need to eradicate multidimensional poverty, including multidimensional child poverty (MFED 2016;RoB 2016RoB , 2018. Despite these commitments, limited efforts have so far been undertaken to gain insight into issues of multidimensional (child) poverty. In 2015, UNICEF published a study of multidimensional child poverty based on the MODA approach (de Neubourg et al. 2015). The study offered critical empirical insights and confirmed the notion that deprivation is widespread among children in Botswana. However, it did not provide a composite measure of multidimensional poverty for all children in the country. In this study, in addition to providing detailed analysis at indicator level, we provide an aggregate estimate of the proportion of multidimensionally poor children at the national level and across demographic, geographical and economic variables in line with LNOB.

Data Sources
Our analysis utilises the 2015/2016 Botswana multi-topic household survey (hereafter 2015/2016 BMTHS) collected by Statistics Botswana (SB). This survey is a crosssectional and nationally representative survey, allowing for disaggregation at the district level. The survey aims to provide a comprehensive set of indicators designed to produce multidimensional welfare indicators at both household-and individual-level to allow for enriched and in-depth analyses. According to SB (2018), the 2015/2016 BMTHS results will be used for policy formulation, NDPs, Vision 2036 and the SDGs. The 2015/2016 BMTHS will also serve as a baseline to track the progress of BPEPS and poverty eradication programmes implemented by various stakeholders (SB 2018).
The original dataset contains information from a sample of 24,720 individuals from 7060 households surveyed in 2015/2016. The estimated population when using sample weights is 2,073,675 individuals and an estimated 589,909 households (SB 2018). Our analysis is based on a sample of 9718 children aged 0-17 years from a sample of 3770 households. The estimated number of children is 817,843 from a total of 319,705 households, meaning that 54.2% of households had at least one child (Table 1).

Methods
This study combines conceptual and methodological principles from across the MODA and MPI approaches. Following the MODA approach, we seek to include as many child-level indicators as possible to create an individual-level measure and to give full consideration to overlapping deprivation analysis. Following the MPI approach, we build a composite measure that allows for estimating the incidence and depth of multidimensional poverty among children in Botswana. In line with the principle of LNOB, we also ensure that the measure is decomposable for different groups of children, particularly those that may be deemed vulnerable or marginalised.
In keeping with these principles, we adopt the AF approach for identification and aggregation. Identification is a two-step cut-off process (Alkire and Foster 2011). First, at indicator level, a child is considered deprived if their achievement is below a defined cut-off. Second, at aggregate level, the poverty cut-off represents the number of weighted indicators that a child must be deprived on in order to be considered multidimensionally poor (Alkire and Santos 2014). Aggregation is also a two-step process. First, a simple headcount ratio can be calculated that captures the incidence of Source: Authors' estimates based on the 2015/2016 BMTHS data multidimensional poverty among a given population of children. In keeping with the global MPI (Alkire and Santos 2014), we use a poverty cut-off of 33.3% of weighted indicators. Second, the adjusted headcount ratio offers a measure of the intensity of multidimensional poverty by multiplying the average number of weighted deprivations with the simple headcount ratio. These measures satisfy the axioms of population subgroup decomposability and dimension breakdown (Chen et al. 2019), which is useful for policymakers when developing interventions and targeted policies (Alkire and Apablaza 2016). In line with the MPI, we adopt equal weighting scheme at the dimension level.
To check for robustness of our results, we compute poverty headcount ratios (H), intensity (A) and adjusted headcount ratio (M 0 ) considering three different poverty cutoffs (k values). We limit the values of k to a range of 25-40% to conduct restricted tests of dominance (see Alkire and Santos 2014). We then calculate multidimensional poverty measures across age, gender and geography to check if the results are stable. We find that in general H, A and M 0 across age, gender and geographical variables remain stable and consistent. For example, the ordering of districts does not change with Ngamiland West and Kweneng West ranking one and two (respectively) across poverty cut-offs (see Annex - Table 9).

Dimensions and Indicators
The choice of dimensions and indicators is informed by the capability approach in conjunction with the consensus approach (Alkire 2002). The capability approach directs us to look at vital functionings for children. Documents outlining Botswana's policy commitments and development priorities such as Vision 2036, NDP 11, BPEPS and the SDGs ensures that the measure is contextually relevant. The final choice of dimensions and indicators was restricted by data availability.
We include 23 indicators in seven dimensions: (i) Assets, (ii) Housing and living condition, (iii) Water and sanitation, (iv) Food security, (v) Health, (vi) Education, and (vii) Security. The selected dimensions cover most of the indicators and dimensions of the global MPI (Alkire and Santos 2014) and the dimensions proposed in MODA child poverty study for Botswana (de Neubourg et al. 2015). Table 2 presents the proposed dimensions, deprivation indicators, as well as the deprivation cut-offs. It also indicates the age brackets for which these indicators hold and whether the indicators were included in the 2015 MODA study. Below we provide a brief description of each dimension and the corresponding deprivation indicators.
Asset dimension includes indicators referring to possession of household assets that serve as a proxy for household living standards. This dimension comprises four deprivation indicators: information, durable goods, transport and housing tenure. Household durable assets are integral to the functioning and attainment of people's well-being, including children (Lerman and McKernan 2008). Lack of transport can impact negatively on children's access to health or education in cases where the facilities are far (Allendorf 2007). Housing tenure security is considered a right to adequate housing (Espinoza-Delgado and Klasen 2018). In the capability approach, homeownership is vital because it indicates a crucial functioning of "security or protection" (Doyal and Gough 1991).

Health
Health facility Assesses the perceived quality of the nearest health facility The perceived quality of nearest health facility he/she uses is poor and has the following problems: the facility is too far, the facility is not clean or HH 0-17 Y Housing and living condition dimension aims to capture deprivations related to the quality of housing. Children have the right to basic shelter that will enable them to live a dignified life (UNHROHC 1989). Six deprivation indicators are considered for this dimension: overcrowding, cooking fuel, electricity, floor material, roof material and wall material. These indicators are closely associated with child health (UN-HABITAT 2009).
Water and sanitation dimension includes two deprivation indicators: access to safe drinking water and toilet facility. Contaminated water is a huge cause of diarrhoearelated diseases, including cholera and other diseases such as pneumonia, trachoma and skin infections (UNICEF 2010) while lack of toilet facility increases the risk of transmission of diseases (Trani et al. 2016). Access to safe drinking water and clean toilet facility reduce child mortality and morbidity (Trani and Cannings 2013). These two indicators are captured by SDG 6 that calls to ensure availability and sustainable management of water and sanitation for all (UN 2015).
Food security dimension includes five indicators. The first indicator captures access to food insecurity at the household level, using the Household Food Insecurity Access Scale (HFIAS) methodology 3 (Coates et al. 2007). The other four indicators (weightfor-age, height-for-age, weight-for-height and body mass index) capture the functioning of 'being well-nourished' and are derived using anthropometric measure based on WHO methodology (Alkire and Santos 2014;WHO 2006). They are used to measure children's nutritional deficiencies (WHO and UNICEF 2010), that can lead to numerous health disorders (Trani et al. 2016). Food security dimension is captured by SDG 2.
Health dimension refers to access and quality of the nearest health facility and chronic illness. It captures bodily health capability and refers to a lack of resources for children's health (D'Agostino et al. 2018). Access to a health care facility is necessary for promoting children's health (UNICEF 2012). Children with chronic illness are unable to do any kind of work including play or going to school (Beatty and Fothergill 2005), which is vital for children's social development and is a crucial aspect of the human life and healthy growth and well-being (D'Agostino et al. 2018). The health dimension is captured by SDG 3 (target 3.8) (UN 2015).
Education dimension captures children's access to education and literacy. Education is a fundamental right for children (UNHROHC 1989) and plays a vital role in children's lives. We use two indicators: enrolment and literacy. Enrolment captures whether children in school-going age are being exposed to the learning environment (enrolled in school). Literacy captures whether children aged between 15 and 17 are able to read and write. In the SDGs, education has a stand-alone goal, SDG 4 (target 4.1) (UN 2015).
Security dimension is captured using two indicators (safety and crime) to capture the capability of "being able to move freely from place to place" (Nussbaum 2005). It captures the neighbourhood environment (D'Agostino et al. 2018), in recognition that feeling safe is an essential aspect of quality of life (Rees 2019). Literature has shown that violence or crime can obstruct development and can contribute to sustaining poverty traps (see Trani et al. 2016). According to the CRC, children have the right to be protected from all forms of violence (UNHROHC 1989). In the SDGs, this dimension is captured by SDG 16 (target 16.1) (UN 2015).
Correlation analysis (see Annex - Table 7) based on Spearman rank correlation coefficients suggests that association between indicators is generally weak. The result supports the inclusion of all indicators. We find moderate associations between indicators for housing materials and assets. The adoption of equal weighting at dimension level reduces the potential bias that may emerge from these modest associations. Table 3 reveal that 41.7% of children aged 0-17 in Botswana can be considered to be multidimensionally poor. This result shows that the incidence of multidimensional child poverty in Botswana remains a substantial problem. The intensity of multidimensional poverty is estimated at 43.4% at the national level, meaning that, on average, children are simultaneously deprived in at least ten (10) indicators out of the twenty-three (23) indicators considered. The adjusted headcount ratio is estimated at 0.181. In the remainder of this section, we discuss differences in outcomes of multidimensional child poverty for different groups. Differences are statistically significant unless otherwise indicated.

Multidimensional Poverty by Demographic Characteristics
As expected, there are significant differences in poverty according to demographic characteristics (Table 3). In terms of gender, the proportion of children identified as multidimensionally poor is significantly higher for boys than girls. Furthermore, both the intensity and adjusted headcount ratio are higher for boys than girls. These differences are mostly driven by deprivation in relation to nutrition. Across all indicators, boys experience higher deprivation rates than girls do. This finding is in line with studies elsewhere in Sub-Saharan Africa (Wamani et al. 2007). With respect to age, children aged 0-4 experience significantly higher poverty incidences compared to other age groups. The same pattern is observed for the adjusted headcount ratio, while the intensity of poverty showed mixed results. Again, these differences are mostly driven by deprivation in relation to nutrition. As expected, the results reveal that children living with disabilities experience significantly higher levels of poverty than those without any disability. In terms of citizenship, non-citizen children have lower poverty incidences than citizens. The majority of non-citizens live in cities/towns where poverty levels are lower, and employment opportunities exist. The majority of the non-citizen children (60.4%) are from Zimbabwe, 9.9% South Africa, 14.9% other parts of Africa and 12.6% rest of the World.
With respect to living arrangements, children living with both parents experience lower levels of multidimensional poverty than those living with mothers alone or with none of their biological parents. An interesting finding is that children living with their fathers alone have lower poverty incidences than those living with both parents. To investigate this further, we examined the link between living arrangement and household size. The household size differs considerably across living arrangements. In essence, average household size is 5.5 for children living with their fathers alone compared to 7.3, 6.9 and 6.9 for those living with mothers alone, both parents and non-biological parents, respectively. In other words, children living with their fathers tend to live in smaller households, suggesting that resources do not need to be spread as thinly. Indeed, we observe higher poverty incidence for children in larger households compared to those in smaller households. The same results are found for intensity and adjusted headcount ratio. Orphaned children experience higher incidence and intensity of poverty compared to children with both parents alive. The poverty situation is worse for double orphans (those who lost both biological parents) than single orphans. Similar studies confirmed this finding in developing countries. For example, Misinde (2019) found that on average living conditions of orphans were less than the average conditions of nonorphans in Uganda. Similarly, children not staying with their biological parents experience higher levels of poverty than those living with their biological parents.
With respect to the gender of household head, results show that children living in households headed by women experience higher poverty incidence and intensity than those living in households headed by men. Similarly, children residing in households headed by other children and older persons have higher levels of poverty than those in households headed by adults or youth. Compared with children living in households headed by married couples, children residing in households headed by cohabiting couples experience higher levels of poverty followed by those living in households whose heads never married, widowed/widower, separated and divorced. The finding that child poverty is higher among cohabiting partners might be explained by the fact that children in households with cohabiting partners are often biologically related to Significance levels: * p < 0.1; ** p < 0.05; *** p < 0.01 a All percentages are estimated at the population level using sample weights. Sample size: 9718 only one partner (mostly mothers). This scenario sometimes put such children at a disadvantage regarding sharing of resources that are brought in by the non-biological partner. Finally, the incidence of multidimensional child poverty and adjusted headcount ratios decline sharply with improvements in educational levels of household heads. Children residing in households whose heads never attended school experiencing the highest poverty levels; eight times higher than those residing in households whose heads have a university qualification. Table 4 presents results along the lines of economic variables (economic activity and economic status). Employment plays an important role, and multidimensional child poverty significantly varies across the employment status of the household head. We observe that children residing in households whose heads are unemployed experience significantly higher levels of poverty than those whose heads are involved in paid employment, self-employment and own farm employment. Surprisingly, children residing in households whose heads are engaged as family helpers have higher levels of poverty than those whose heads are unemployed. This finding could be linked to lower salaries associated with those engaged as family helpers.

Multidimensional Poverty by Economic Variables
We also observe a negative relationship between household expenditures and multidimensional child poverty. Children in the poorest quintile (Q1) experience higher levels of multidimensional child poverty, and the incidence declines along the quintiles with those in the richest quintile (Q5) experiencing lowest levels of poverty. The incidence of poverty in Q1 is 23 times higher than in Q5. Similarly, children from Q1 have higher adjusted headcount ratio compared those in Q2-Q5. Similar studies have found declining multidimensional child poverty rates along income quintiles/ deciles. For example, Roelen (2017) showed declining multidimensional incidences of child poverty along income deciles (measured using per capita real consumption), with the richest decile recording the lowest poverty rates in Vietnam. Table 5 considers Botswana's multidimensional child poverty situation across cities/ towns, urban villages and rural areas. Children residing in rural areas experience significantly higher levels of poverty than those in urban villages and cities/towns. For example, the incidence of multidimensional child poverty in rural areas is triple that in cities/towns. Furthermore, both intensity and adjusted headcount ratios are higher in rural areas. Similar findings exist in the literature (see Ferrone and de Milliano 2018). We further explore whether multidimensional child poverty varies across administrative districts. Results reveal that incidence varies significantly across the 26 administrative districts with Ngwaketse West and Kweneng West recording the highest incidences of multidimensional child poverty, estimated at 86% and 76%, respectively. The two districts also recorded the highest adjusted headcount ratios estimated higher than 0.300.

Micro-Determinants of Multidimensional Child Poverty
To complement the descriptive analysis, we employed a logit regression model to investigate the joint correlation of demographic, economic and geographical factors in relation to multidimensional child poverty. Table 6 presents the results showing the estimated coefficients, their robust standard errors and the marginal effects. The log pseudolikelihood ratio test indicates that there is a significant relationship between the probabilities of being multidimensionally poor and the explanatory variables included in the model (p < 0.001).
Findings of regression analysis mainly confirm those of descriptive analysis. Boys, children aged 0-4, citizens and children living with a disability are more likely to experience multidimensional poverty. In terms of living arrangements, children living with mothers alone or with none of their biological parents have higher probabilities of being multidimensionally poor than those living with both parents. The finding that children living with their fathers alone experience lower levels of poverty is also confirmed through regression analysis; they have a lower probability of being multidimensionally poor than those living with both biological parents. Orphans have higher probabilities of being multidimensionally poor than those with both parents alive. Relationship to the household head also matters: grandchildren and children otherwise related to the head of the household are more likely to be poor than sons or daughters of the household head.
With respect to the characteristics of the household head, children living in households headed by men have a higher probability of being poor than those in households headed by women. This empirical evidence from Botswana is contrary to the general belief that female-headed households are more likely to be poor than male-headed households (see Bradshaw et al. 2017). Our results are consistent with the case of Nicaragua (Espinoza-Delgado and Klasen 2018). This finding couldin partbe explained by household composition. In the case of Botswana, households headed by men are characterised by slightly larger household sizes than those that are headed by Significance levels: * p < 0.1; ** p < 0.05; *** p < 0.01 a All percentages are estimated at population level using sample weights. Sample size: 9718 women. For example, in the case of married couples, household size averaged 7.1 for male-headed households compared to 6.2 of those led by women (SB 2018). Children living in households headed by other children and youth have higher probabilities of being multidimensionally poor compared to those children living in households headed by adults. Children residing in households headed by married couples have lower probabilities of being multidimensionally poor, an indication that marriage plays a pivotal role in poverty . Larger households are associated with a greater likelihood of being poor: a unit increase in household size will result in 0.9 percentage points increase in the probability of being multidimensionally poor. The probability of being multidimensionally poor declines with improvements in educational attainment. In terms of employment status, children living in households whose heads are engaged in paid employment have lower probabilities of being multidimensionally poor than those from households whose heads are unemployed. Children living in households whose heads are involved in farming on their farms or land are more likely to be multidimensionally poor than those from households with unemployed heads. This finding is consistent with Lekobane and Seleka (2017) who found that in Botswana, households whose heads worked in their own farms are more likely to be poor (based on monetary poverty) than those households whose heads were unemployed. This finding could be explained by low yields in agricultural production due to low technology adoption, especially subsistence arable agriculture, which is predominant in most rural households in Botswana. Also, farming households comprise of mostly dependents (children and older persons) who are economically inactive.
We find a strong association with expenditure profiles; children in higher quintiles are all significantly less likely to be poor than children in the lowest quintile. With respect to geography, the results show that ceteris paribus children residing in rural areas are more likely to be multidimensionally poor than those residing in urban villages. Results show statistically insignificant differences in probabilities of being multidimensionally poor between urban villages and cities and towns.
To capture intersecting inequalities, we included interaction terms on the gender of household head and marital status of the household head to capture the joint impact of the two variables. The results show that children living in households headed by women who are separated, widows or single (never married) have, ceteris paribus, higher probabilities of being multidimensionally poor than those living in households headed by men who are separated, widowers or single, respectively. Children living in cities/towns-women led households have higher probabilities of being multidimensionally poor than those living in cities/ towns-men headed households.

Conclusion and Policy Implications
Measurement of multidimensional child poverty in low-and middle-income countries is now relatively widespread with national and context-specific studies existing alongside large cross-country comparisons. Nevertheless, efforts to measure (child) poverty remain primarily dominated by monetary approaches, and many countries fail to monitor multidimensional child poverty. The two main approaches (MODA and MPI) offer opportunities for expanding efforts to monitor multidimensional child poverty. However, these approaches may also detract from other initiatives that aim to build on their respective strengths in offering childfocused analysis based on an aggregate index.
In this article, we developed a child-centred, individual-level and composite measure that offers nationally relevant and context-specific insights into magnitude and depth of multidimensional child poverty in Botswana. In particular, it did so through the lens of LNOB by zooming in on demographic, economic and geographical characteristics that may be associated with greater vulnerability or marginalisation.
Results point towards relatively high incidence and intensity of multidimensional child poverty for an upper-middle-income country like Botswana. More than four out of ten children can be considered multidimensionally poor, and on average they are deprived in almost half of all deprivations. These numbers suggest the importance of multidimensional poverty measurement alongside economic indicators such as economic growth or monetary poverty.
The descriptive and parametric analysis leads to both expected and more surprising findings in terms of which children may be left behind. Children living with disabilities, orphaned children and children not living with their relatives, for example, are more likely to be poor. More counter-intuitively, children who are citizens of Botswana are more likely to be poor than non-citizens. In addition, children living with their fathers alone have lower poverty incidences than those living with both parents. It should also be noted that despite the seeming discrepancy between economic status and multidimensional poverty, children in poorer quintiles (based on expenditures) are more likely to be multidimensionally poor. Children living in rural areas experience higher levels of poverty and have higher probabilities of being poor.
The application illustrates the added-value of building on key components of the MODA, and MPI approaches, namely their focus on individual-level measurement and aggregated index analysis. In doing so, this allows for providing overall estimates of multidimensional child poverty that are based on individual children as units of analysis, which in turn provided for an in-depth investigation into groups of children that are most affected and left behind. Also, the individual-level multidimensional child poverty measure allows for the operationalisation of the LNOB principle in Botswana, which can be applied elsewhere.
These findings have critical policy implications. Various groups of children are at greater of risk of deprivationand therefore to be left behindthan others are. For Botswana to live up to the principle of LNOB as set out in the BPEPS, Vision 2036 and the 2030 Agenda, it will be vital to put in place relevant policies that will take into account the heterogeneity of different groups of children. The adoption of an individual-and context-specific measure of multidimensional child poverty will be vital in guiding those efforts.
Notwithstanding the importance of cross-sectional analysis of child poverty for a one-time point, future work should be conducted to analyse trends in multidimensional child poverty. Lack of more frequent and timely data prevented this study from analysing trends in multidimensional child poverty.
In sum, results in the article illustrate how the combination of strengths of emerging and increasingly prominent approaches in multidimensional child poverty measurement can be combined to offer child-centred analysis that provides insight into incidence and depth of deprivation. Detailed disaggregated descriptive and non-parametric analysis provides critical understandings of groups who may need further consideration in countries'and in this case, Botswanaefforts to achieving SDG 1 against the principle of LNOB.       Source: Authors' estimates based on the 2015/16 BMTHS data a All percentages are estimated at the population level using sample weights. Standard deviations (SD) are reported in parentheses. n is the sample size for different age groups. − The indicator is not applicable to the age group