Introduction

Poverty is defined as the inability to obtain sufficient resources or income to meet basic human needs, such as food, shelter, and health care (Callan and Nolan 1991). Poverty is one of the top threats to global development. Eradicating poverty worldwide in all its dimensions is the first United Nations Sustainable Development Goal (SDG-1) and the prime objective of national and international development organizations (Alkire and Kanagaratnam 2021; World Bank 2017). Despite global poverty-reduction efforts, approximately 1.3 billion people worldwide are multidimensionally poor, and most are in developing countries. Of those 1.3 billion people, nearly 1 billion are unable to afford secure housing, adequate sanitation, or solid fuels for cooking, and 568 million lack access to adequate drinking water (OPHI 2021). Health deprivation is a common problem in developing countries. Energy deprivation, which significantly affects the living standards of the poor, is also a key obstacle to developments goals (Chishti et al. 2023b). The magnitude and intensity of poverty remain severe in many regions, particularly in the developing countries of Asia and Sub-Saharan Africa, which are hotspots of poverty (Sharma 2019). The youth “bulge,” gender inequalities, poor reproductive health, lower employment levels, and high fertility rates pose challenges for the development process in these regions.

Pakistan, the seventh-poorest country in South Asia, is attempting to cope with the multiple dimensions of poverty through numerous initiatives. Since 2001, Pakistan has moved approximately 32 million people out of monetary poverty, and the poverty headcount rate (measured by consumption-based poverty estimates) dropped from 64.3% in 2001 to 24.3% in 2015 (Sharma 2019). However, the poverty rate in Pakistan remains significantly high, and a large portion of the population is vulnerable, living at or close to the minimum poverty line. Pakistan faces an energy shortage that is impacting the living standards of households. Almost 55% of households are energy-deprived in Pakistan (Qurat-ul-Ann and Mirza 2021).

In 2015, the government of Pakistan announced its long-term growth strategy, “Vision 2025,” designed to overcome economic, social, governance, and security challenges and provide citizens with the ability to improve their choices and standards of living. Vision 2025 stresses a more comprehensive definition of poverty that includes education, health, and other dimensions, together with monetary indicators. Acknowledging the importance of addressing multiple dimensions of poverty for achieving SDG-1, the Pakistani government launched the first national multidimensional poverty index (MPI) in 2016. According to the National MPI (2016), almost 39% of Pakistanis are multidimensionally poor. Hence, the measurement and identification of poverty are key areas of interest to researchers and policymakers. A close look at the available empirical literature on poverty in Pakistan leads to several broad conclusions: most empirical studies focus on the measurement, identification, and determinants of monetary poverty (Ahmad and Faridi 2020; Cheema and Nadeem 2020; Chishti et al. 2022; Sadiq 2010) few empirical studies have examined multidimensional poverty (Debela and Tamiru 2016; Khan et al. 2014; Padda and Hameed 2018; Saboor et al. 2015; Saleem et al. 2021; Saleem and Bilal Khan 2018); and few studies have measured poverty by combining both approaches.

Understanding the determinants of poverty is a crucial step toward effective poverty eradication. While recognizing and estimating poverty dimensions are essential, identifying the factors that contribute to poverty at the household level is equally important (Alkire et al. 2017; Cappellari and Jenkins 2002). A significant body of research has focused on understanding poverty at individual and household levels, with a growing body of literature investigating its determinants. The literature has extensively explored how traditional factors such as household head characteristics (age, education, marital status, health, employment status, gender, occupation), region of residence, remittances received by households, and ownership of land and livestock have shaped household poverty (Alkire et al. 2017; Chen et al. 2019; Lekobane 2021; Meng et al. 2007).The typical household structure and family setup in Pakistan (i.e., joint family system, large household size, large share of dependents, and lower engagement of females in income-generation activities) are crucial factors in determining household deprivation according to the multiple dimensions of poverty. Therefore, this study seeks to clarify the relationship between household demographic characteristics and multiple dimensions of poverty. The study analyzes household-level data from a large-scale survey to investigate how changes in household demographic characteristics impact multidimensional poverty levels. Specifically, the study examines the effects of dependency ratio, family type, changes in family size, employment, and age composition on poverty while controlling for other traditional factors that have been explored in the literature. Policymakers require a comprehensive understanding of these factors to develop effective poverty-reduction strategies, and the findings of this study provide new insights into the complex interplay between household demographic characteristics and poverty dynamics.

This study makes three valuable contributions to the multidimensional poverty literature on Pakistan. First, it constructs an MPI at the national, regional, and provincial levels using the Alkire and Foster (AF) methodology, which combines the non-monetary dimensions of education, health, and living standards with monetary poverty. Few empirical studies have measured acute poverty across the country by combining traditional absolute monetary poverty and non-monetary dimensions in a single index using the latest HIES 2018-19 dataset. By doing so, this study adds to the limited body of literature on multidimensional poverty in Pakistan. Second, the study employs a logit model to shed light on the demographic characteristics that influence multidimensional poverty at the household level, such as family, population, and employment structures, which have received little research attention. Third, the study offers practical implications that can be used in policies for eliminating multidimensional poverty based on empirical research and regional and provincial heterogeneities.

The remainder of this paper is organized as follows. Section 2 provides a detailed account of the relevant literature. Section 3 explains the study’s data and methodology, including the data sources and statistical description, MPI construction, estimation technique, and descriptive details for the explanatory variables. Section 4 presents the estimation of multidimensional poverty, logistic regression findings, empirical results, and robustness checks. Section 5 presents concluding remarks and policy recommendations derived from the findings.

Literature review

Poverty can be attributed to the scarcity of human well-being. Traditional poverty analysis is limited to a single dimension based on monetary indicators (e.g., income or consumption expenditure) of well-being (Gamboa et al. 2020; Jolliffe and Prydz 2016). Sen is one of the pioneers who changed this conception of poverty. According to him, well-being is affected by multiple non-monetary dimensions (basic human capabilities), which are essential in determining poverty status (Braybrooke 2008; Sen 1985). Focusing solely on the monetary dimension is thus insufficient (Kim 2019; Najitama et al. 2020; Salam et al. 2020). The concept of “multidimensional poverty” emerged from Sen’s “capability” approach, which inspired the multidimensional measurement of global poverty. Researchers agree that a multidimensional approach is more appropriate than using a single dimension for quantifying poverty (Alkire and Fang 2019; Alkire and Foster 2011; Dutta et al. 2021).

The United Nations Development Program (UNDP) developed the Human Poverty Index (HDI) in 1997 to measure poverty multidimensionality at national and macro levels. The Oxford Poverty and Human Development Initiative (OPHI) and UNDP jointly introduced the global MPI in 2010 (Alkire and Santos 2010, 2014). Since 2010, the MPI has been considered a significant measure of the non-monetary dimensions of poverty, especially in developing countries (Alkire and Kanagaratnam 2021; Alkire and Foster 2011). The global MPI measures multiple dimensions of poverty by using the household as the unit of analysis, which is then aggregated at the national level. The global MPI was revised in 2018 to align with the SDGs (Alkire et al. 2020; Alkire and Jahan 2018; Alkire and Kanagaratnam 2021). This index covers three dimensions of poverty (i.e., education, health, and living standards) with 10 indicators. In 2018, the World Bank officially adopted the multidimensional poverty concept to monitor global poverty. In the last decade, a scholarly consensus has developed that addressing poverty holistically requires that both of its dimensions be considered (Belete 2021; Bersisa and Heshmati 2021; Mothkoor and Badgaiyan 2021; Santos and Villatoro 2018; Wang et al. 2016). Multidimensional poverty is more common than income poverty in poor households, and worsens more quickly (Liu et al. 2023).

Researchers have shifted from measuring monetary poverty to measuring the multidimensional aspects of poverty in order to advance towards sustainable development goals. For instance, most researchers have measured deprivation in the health dimension (Nawaz and Iqbal 2021). The deprivation of energy sources affects households’ living standards. Energy source deprivation has been identified as an aspect of poverty (Al-Tal et al. 2021; Chishti and Dogan 2022), and researchers are trying to identify its determinants at the national (Xia et al. 2022; Chishti et al. 2023a) and international (Murshed and Ozturk 2023; Chishti and Sinha 2022) levels. Furthermore, researchers have highlighted that achieving green development is an important aspect of economic development (Chishti et al. 2023a; Murshed et al. 2023). Technological and financial innovations are crucial for improving environmental quality and achieving development goals (Chishti and Sinha 2022; Dogan et al. 2022; Murshed et al. 2022).

Household demographic characteristics are crucial factors for assessing poverty status (Ahmad and Faridi 2020; Bayudan-Dacuycuy and Lim 2013; Libois and Somville 2018; Munoz Boudet et al. 2018). The nature, intensity, and dynamics of poverty are associated with individual personal characteristics and household composition. Artha and Dartanto (2018) used the education, health, and quality of life dimensions to measure multidimensional poverty in Indonesia by applying the AF approach. They found that around 61% of monetarily non-poor people were multidimensionally poor, and that most had deprivation-level health indicators. They also found that the household head’s education, household size, house ownership, and location were significant determinants of multidimensional poverty. Chen et al. (2019) applied the AF approach to estimate multidimensional poverty in Taiwan. They identified significant correlations between marital status, age, socioeconomic status, household size, and multidimensional poverty in both Taiwan and China. Similarly, Boudet et al. (2021) conducted a comprehensive analysis of household demography using a dataset of 91 countries, finding that the poverty status of households changed along with changes in their demographic composition, and that demographic characteristics such as household type, age, and sex pattern were important determinants of poverty in poor households.

In summary, extensive empirical analyses have been conducted to identify and estimate poverty. Despite their differences in scope and methodology, the studies all found that alleviating poverty in the monetary and non-monetary dimensions always furthers the achievement of sustainable development goals, and that changes in household demographic characteristics affect multidimensional poverty levels. Hence, an in-depth understanding of poverty and its determining factors at the household level is crucial for developing an appropriate framework and strategy for poverty eradication and sustainable development.

Data and methodology

Data sources and sampling

This study used the eleventh round of nationally representative HIES data collected by the Pakistan Bureau of Statistics (PBS) of the government of Pakistan in 2018–19. The HIES contains a sample of 24,809 households from all four of Pakistan’s provinces (Khyber Pakhtunkhwa [KPK], Punjab, Sindh, and Balochistan). It collects nationally representative data on a wide range of social indicators, including information on the sociodemographic profiles of households, health, education, population welfare, housing characteristics, water and sanitation, income, and consumption.

The PBS employs a stratified two-stage sampling framework. The framework is based on the 2017 population and housing census to ensure sufficient provincial and regional representation. The country is divided into enumeration blocks, which were selected as primary sampling units using a probability proportional to the size sampling technique. Households in specific primary sampling units were selected as the secondary sampling units. The detailed sampling procedure and data are presented on the official website of the PBS.Footnote 1

This study selected the HIES 2018-19 dataset for its empirical investigation for two main reasons. First, the PBS revised the survey questionnaire and added innovative modules to monitor the social and economic indicators of the SDGs (Awan et al. 2022). Second, it provides thorough information on both monetary (consumption expenditure) and nonmonetary indicators, such as household health, education, and living standards, to enable a multidimensional measure of poverty in Pakistan. This satisfies the AF methodological requirement that the data for each multidimensional poverty indicator be derived from the same survey and be linked to the household (or individual) level (Alkire and Foster 2011).

Methodology

Multidimensional poverty: Construction of MPI

This study employs Alkire and Foster’s (Alkire and Foster 2011) method, which is based on Sen’s capability approach, to estimate multidimensional poverty. The AF method utilizes a comprehensive set of poverty measures based on the FGT class’s poverty measures (Foster et al. 1984) and the counting approach (Atkinson et al. 2002), enabling the consistent measurement of both cardinal and ordinal data.Footnote 2 This method is simple and flexible enough to permit a detailed examination of multiple aspects of the multidimensional concept of poverty, which includes monetary poverty as one of several poverty dimensions (Evans et al. 2020).

This methodology constructs the MPI through a dual-cutoff (deprivation and poverty cutoff) approach.Footnote 3 In a population with n persons, the primary information is usually represented by an n*d dimensional matrix X[xij], where xij denotes the achievement of person i in dimension j. The d-dimensional row vector z = (z1, .…, zd) is used to produce the deprivation cutoff zj. The person i is considered non-deprived i dimension j if the person’s achievement xij ≥ zj. Then, a deprivation matrix \({g}^0\left[{g}_{ij}^0\right]\) is obtained from matrix X and vector Z, where person i is considered non-deprived if xij ≥ zj and \({g}_{ij}^0=0\) and is considered deprived if xij ≤ zj \({g}_{ij}^0=1\). The d-dimensional weight vector w = (w1, .…, wd) reflects the relative importance of each dimension\(\left({\sum}_{j=1}^d{w}_j=1\right)\). The weighted deprivation score of person i is denoted as ci, computed as \({c}_i=\left({\sum}_{j=1}^d{w}_j{g}_{ij}^0\right)\) and commonly known as the “deprivation score” of person i. If a person is deprived of all dimensions, then ci = 1; similarly, ci = 0 when person i is not deprived of all dimensions.

Furthermore, a poverty cutoff of k(0 < k ≤ 1) and an indicator function ρk(xi, z) are used to identify the multidimensional poor. A person is considered poor if ci > k and ρk(xi, z) = 1; otherwise, ci < k , ρk(xi, z) = 0. Thus, the multidimensional poverty headcount (H) can be presented as\(H=\left({~}^{q}\!\left/ \!{~}_{n}\right.\right)\), where \(q=\sum_{i=1}^n\rho k\left({x}_i,z\right)\), which indicates the proportion of the poor population. H explains the incidence of multidimensional poverty but does not represent each dimension’s contribution to poverty.Footnote 4 To address this issue, Alkire and Foster (2011) proposed several other indices. They defined dimension j’s censored headcount ratio as\({h}_i(k)=\frac{1}{n}{\sum}_{i=1}^n{g}_{ij}^0(k)\), where \({\mathrm{g}}_{ij}^0(k)={\mathrm{g}}_{ij}^0(k)\times \rho k\left({x}_i,z\right)\) denotes the censored deprivation matrix. The intensity of poverty (A) in the poor population is given by\(\mathrm{A}=\sum_{i=1}^q{~}^{{c}_i(k)}\!\left/ \!{~}_{q}\right.=\sum_{i=1}^q{\sum}_{j=1}^d{w}_j{g}_{ij}^0(k)\), where \({c}_i(k)=\sum_{j=1}^d{w}_j{g}_{ij}^0(k)\) is the censored deprivation score of person i. The aggregation phase in the AF methodology employs an adjusted headcount ratio (M0), also known as the MPI. The MPI (M0) combines two types of information: the multidimensional headcount ratio (H) and intensity of poverty (A):

$$MPI={M}^0=H\times A=\frac{1}{n}{\sum}_{i=1}^n{c}_i(k)=\frac{1}{n}\sum_{i=1}^n{w}_j{h}_i(k)$$
(1)

where H represents the percentage of multidimensionally poor individuals in the total population, and A is the average proportion of indicators in which poor people are deprived. Here, MPI can be defined as the proportion of deprivation experienced by a multidimensionally poor person as a share of the deprivation that would be experienced if the entire population were multidimensionally poor and deprived of all the indicators considered.

Dimensions and indicators of MPI

The selection of dimensions and indicators for developing an MPI is a normative decision and an initial step in the empirical valuation of poverty (Alkire and Seth 2015; Firdausy and Budisetyowati 2022). With this in mind and following the literature on multidimensional poverty measures (Alkire and Santos 2014; Alkire and Foster 2011) we selected the education, health, living standards, and monetary poverty dimensions to construct our MPI. The living standards, education, and health dimensions are similar to those of the traditional global MPI; however, some indicators have been adjusted to allow for data availability. The global MPI is sufficiently flexible to include additional dimensions, indicators, and changes to the weighting scheme. This allows a detailed analysis of multidimensional poverty, particularly by counting country-specific dimensions and indicators. We included monetary poverty as an additional dimension, assuming that monetary deprivation combined with multidimensional deprivation captures a comprehensive picture of overall deprivation (Evans et al. 2020). The MPI consists of four dimensions and 11 indicators, including two for education and health, six for living standards, and one for monetary poverty.

The education dimension comprises the household members’ number of years of schooling and the children’s school attendance. These two indicators reflect a household’s fundamental educational ability and attainment (Alkire and Santos 2014; Yu 2013). This education dimension is consistent with the global MPI (Alkire and Santos 2010). Education level is a key dimension of multidimensional poverty, and numerous studies have found that education level positively influences poverty reduction (Debebe and Wuletaw 2022; Tran et al. 2020).

This study used child mortality and the availability of health service indicators to measure household deprivation in the health dimension. The association between health status and poverty has been extensively discussed in the empirical literature. Health has been identified as one of the top contributing factors to multidimensional deprivation (Chen et al. 2019) because physical vulnerability may limit a person’s freedom and functional scope (Callander et al. 2013). In empirical research, health-related indicators, such as nutrition, child mortality, access to basic health services, and health insurance, have been extensively employed as health deprivation measures with a major impact on economic welfare (Alkire and Santos 2010; Cheema 2021; Cheema and Nadeem 2020).

This study considered five indicators (availability of safe drinking water, sanitation facilities, housing quality, cooking fuel, electricity, and small asset ownership) to assess household living standards. For the living standards dimension, most studies have used indicators such as electricity, sanitation, drinking water, cooking fuel, flooring, and assets to assess deprivation (Alkire and Jahan 2018; Alkire and Santos 2010). Possession of durable assets (such as televisions, radios, animal carts, telephones, bicycles, motorbikes, and refrigerators) has been used as a poverty indicator in several empirical studies (Alkire and Santos 2014; Pham and Mukhopadhaya 2022).

The fourth dimension comprised information on the deprivation of the households’ monetary well-being. The World Bank considers the monetary dimension, a common and robust measure of well-being, a principal measure of poverty. The study considered only one indicator, household per-capita consumption expenditure (the sum of food and non-food expenditures), in this dimension. Household per-capita consumption is a significant indicator of monetary well-being (Bersisa and Heshmati 2021). Based on the internationally defined poverty line for developing economies, households unable to meet a per-capita consumption expenditure equal to $1.90 per day were considered deprived in the monetary dimension (Jolliffe and Prydz 2016).

Weights and poverty cut-off

Selecting weights for the dimensions and indicators is another crucial step in the development of the MPI. However, there is no consensus regarding a single standard method. Three weighting methods—frequency-based, expert opinion, and equal weights—are commonly used in the literature; each method has its advantages and disadvantages (Decancq and Lugo 2010; Deyshappriya and Feeny 2021). Most research on multidimensional poverty assigns equal weight to all dimensions and indicators within each dimension (Alkire et al. 2017; Alkire and Santos 2010). In the absence of a justifiable reason for assigning unequal weights to each dimension, it seems logical to use equal weights to measure multidimensional poverty (Atkinson et al. 2002; Chakravarty et al. 2008). We assigned equal weights (0.25) to each dimension given the equal importance of all dimensions and following the AF (2011) approach. The same method was followed for indicators within the dimensions. The definitions, weights, and descriptive statistics of the dimensions and indicators are presented in Table 1.

Table 1 Definitions, weights, and descriptive statistics of MPI components
Full size table

The poverty cut-off k reflects the lowest level of deprivation that needs to be identified for a household to be considered multidimensionally poor in a particular situation. Two extreme methods are used to identify poverty in the literature: the intersection method (where the household is considered multidimensionally poor if it is deprived in all indicators) and the union method (where the household is considered multidimensionally poor if it is deprived in at least one indicator) (Atkinson et al. 2002). However, the AF dual-cutoff method offers an intermediate choice between the intersection and union methods. This implies an intermediary poverty cutoff (i.e., 0.33) for global MPI (Ambros and Saxena 2018; Alkire and Kanagaratnam 2021). For the AF, the 0.25 poverty cut-off value ranges from 0 to 1, where 0 indicates non-poor households and 1 indicates completely poor households. This was used in this study to identify multidimensionally deprived households. Households were considered multidimensionally poor if they were deprived of 0.25 of weighted indicators within a dimension or across dimensions. This is qualitatively equivalent to a 0.33 global MPI poverty cut-off. At a 0.25 poverty cut-off point, the MPI value identifies poor households in the monetary poverty dimension, in terms of global MPI dimensions, or in both (Evans et al. 2020).

Household demographic characteristics

The key explanatory variables in this study were household demographic characteristics, operationalized as household family, household population, and household employment characteristics. Household demographic characteristics are crucial for assessing households’ poverty status (Ahmad and Faridi 2020; Bayudan-Dacuycuy and Lim 2013; Libois and Somville 2018; Munoz Boudet et al. 2018), which changes along with changes in demographic characteristics (Boudet et al. 2021). Key family characteristics include family size and type (whether the house is separate or joint). Dependency and child ratios are key household population characteristics. The youth employment ratio, female employment ratio, and labor force ratio were used as household employment characteristics.

Specification of econometric model

This study applied logistic regression to analyze the impact of household demographic characteristics on the multidimensional poverty status of households. The poverty status of the household (POVi) is a dichotomous variable indicating that a household is poor if the multidimensional deprivation score ci≥ the selected k, as shown in Equation 2:

$${POV}_i=\left\{\begin{array}{c}=1,\kern0.75em if\ {c}_i\ge 0.25\kern2.25em household\ is\ poor\kern0.5em \\ {}\kern2.25em =0,\kern0.75em if\ {c}_i<0.25\kern1.75em household\ is\ non- poor\end{array}\right.$$
(2)

Logistic estimation is the best method for dichotomous dependent variables (Ambros and Saxena 2018; Greene 2003; Sperandei 2014). Hence, this study used the following logistic regression model to analyze the empirical relationship between household demographic characteristics and multidimensional poverty in Pakistan:

$${POV}_i= In\left(\frac{P_i}{1-{P}_i\ }\right)={\beta}_0+{\beta}_1{Hs}_i+{\beta}_2{Ft}_i+{\beta}_3{Dr}_i+{\beta}_4{Cr}_i+{\beta}_5{Ye}_i+{\beta}_6{Fe}_i+{\beta}_7{Lf}_i+{\beta}_8{X}_i+{U}_i$$
(3)

where POVi denotes the natural log of the odds ratio. A household is identified as multidimensionally poor if Pi = 1 and non-poor if Pi = 0. Seven variables related to family, household population, and household employment characteristics were used to capture household demographic characteristics. Family characteristics were assessed by considering family size Hsi and family type Fti (whether the house is separate or joint). The household’s population characteristics were assessed via the dependency ratio Fti of the family to-child ratio Cri. Household employment characteristics were assessed by considering the youth employment ratio Yei, female employment ratio Fei, and labor force ratio Lfi. Xi denotes the vector of control variables that are potentially correlated with the household poverty status. We controlled for household head characteristics (i.e., age, gender, marital status, education, employment status), ownership of property by household, remittances received by the household, and region and province of residence. β1, β2, β3, β4, β5, β6, β7, β8 are the parameters, and Ui is the error term. The detailed description and summary statistics of variables are presented in Table 2.

Table 2 Description of variables and statistical summary
Full size table

Furthermore, the study investigated the impact of household demographic characteristics on multidimensional poverty independently for urban and rural regions and each province in Pakistan using Equation 3.

Empirical results and discussion

Multidimensional poverty estimation and decomposition

The MPI is equal to the mean ci(k). It satisfies the axioms of weak monotonicity, dimensional monotonicity, population subgroup decomposability, and ordinality (Alkire et al. 2015). Population subgroup decomposition implies that the MPI, H, A, and each dimension/indicator can be disaggregated by any group for which the data represent subgroup poverty levels. The MPI was used to estimate multidimensional poverty in Pakistan and further decompose it according to region, province, and dimension to investigate its contribution to overall poverty.

Table 3 and Fig. 1 present the multidimensional poverty estimates. The MPI value is 0.22 at the national level, indicating that 22% of Pakistani households are multidimensionally deprived in 25% of the selected dimensions (see Fig. 1).Footnote 5 The national headcount ratio (H) is 48.8%, and the intensity of poverty (A) is 45.6% (see Table 4). A close look at Fig. 1 reveals a significant variation in regional multidimensional poverty, as the MPI value is 0.28 in rural regions and 0.10 in urban regions. Furthermore, the provincial results highlight heterogeneities in the MPI values. For example, Punjab is the least multidimensionally poor province, with a 0.16 MPI value. The proportion of multidimensional poverty is highest in Baluchistan, followed by Sindh and KPK, with MPI values of 0.39, 0.26, and 0.23, respectively.

Table 3 MPI estimates at national, provincial, and regional levels
Full size table
Fig. 1
figure 1

National, regional, and provincial multidimensional poverty in Pakistan

Full size image
Table 4 Impact of household demographic characteristics on multidimensional poverty at national, regional, and provincial levels (logistic regression results)
Full size table

The contributions of the poverty dimensions to the national, regional, and provincial MPIs are shown in Fig. 3. The education dimension contributes 28% to national multidimensional poverty, followed by living standards (25%), monetary poverty (23%), and health (22%). The highest share of the education dimension in the MPI is observed at the regional and provincial levels, as shown in Fig. 2. The contributions of education, living standards, and monetary dimensions differ between urban and rural regions and across provinces. For instance, the contribution of the monetary dimension to the MPI is lowest in KPK province, while the share of the health dimension is lowest in Sindh province. Likewise, the living standards dimension was the second-highest contributor to the rural MPI but the lowest contributor to the urban MPI. Interestingly, the dimension contribution patterns are similar between the most-deprived (rural and Balochistan) and least-deprived (urban and Punjab) regions and provinces.

Fig. 2
figure 2

Percentage contribution of dimensions to national, regional, and provincial MPIs

Full size image

Figure 3 shows the percentages of Pakistani households deprived of each of the 11 indicators. Most of the households are deprived of basic health facilities (56.06%), followed by cooking fuel (55.09%), years of schooling (39.51%), and asset ownership (34.91). Cooking with dung, charcoal, wood, or coal produces considerable volumes of indoor air pollution, resulting in health problems, particularly for rural females. The deprivation percentages of households in terms of child mortality (9.13%) and access to electricity (11.04%) were quite low, indicating that Pakistan has improved significantly in terms of child health and the power sector. The percentages of deprived households in urban and rural regions and provinces are presented in Appendix Table 6.

Fig. 3
figure 3

Percentage of deprived households by indicator

Full size image

Demographic characteristics of multidimensional poverty: Logistic regression results

Table 4 presents the regression results for the national sample, urban and rural regions, and the four provinces. The logistic regression coefficients in the estimated models in Table 4 are the natural logarithms of the odds ratios. All the variables presented in Column 1 are statistically significant. The log-likelihood ratio indicates the statistical significance of the model at the 1% level. Household demographic characteristics have a significant impact on defining multidimensional poverty status. The signs of all the predicted coefficients are consistent with traditional economic concepts.

Family characteristics

The estimated results shown in Column 1 indicate that family characteristics are inextricably linked to a household’s multidimensional poverty level. The household likelihood of being poor increases by 0.17% with a unit increase in household size. These findings can be explained in the context of household liabilities, which tend to increase along with the number of household members. These results are consistent with those of previous studies (Chen et al. 2019; Libois and Somville 2018). The dwellings in which people live play a vital role in determining their lives and well-being. The nuclear or separate dwelling setting negatively affects multidimensional poverty; households living in separate dwellings are 0.24% less likely to be poor than coupled/joint households. A joint family system is prevalent in Pakistani culture, with multiple families living together and often sharing their livelihoods. These shared households tend to be overcrowded and have fewer earners, resulting in financial strain and reduced living space. Consequently, the household members often lack access to adequate living facilities. On the other hand, independent householders have better access to basic amenities, such as electricity, sanitation, water, and healthcare, making them less likely to experience multidimensional poverty. These results support the findings of Mahmood et al. (2019).

Population characteristics

The indicators of household population characteristics—the household dependency and child ratios—contribute positively to multidimensional household poverty, consistent with theoretical forecasts. The probability of being multidimensionally poor increases by 0.29% with additional dependents. Having a sizable unproductive population increases a household’s burden, which increases the likelihood of multidimensional poverty. This finding aligns with the results of Salam et al. (2020) and Najitama et al. (2020), who found that the household dependency ratio was a traditional factor in multidimensional poverty. The child ratio is another important determinant of increased poverty in developing countries. Overall, these findings emphasize the importance of considering household demographics when formulating policies intended to reduce multidimensional poverty. The results in Column 1 show that an increase in the child ratio increases the likelihood of households being multidimensionally poor by 0.17%. These findings are similar to those of previous studies, which have suggested that additional children in a household increases the likelihood of the household falling into poverty and fosters the intergenerational transmission of poverty (Fabrizi and Mussida 2020; Munoz Boudet et al. 2018). The empirical results show that households with higher child ratios are more likely to be multidimensionally poor. This finding is consistent with theoretical expectations and is supported by the sociocultural context in Pakistan, where having more children is preferred and early marriage is common. Women in such households have limited autonomy in their reproductive health decision-making and often face health risks due to increased childbearing. Moreover, a high child ratio increases the household dependency ratio, resulting in a greater burden on the household and leading to a higher likelihood of multidimensional poverty. This result is consistent with the findings of other studies, including Salam et al. (2020) and Najitama et al. (2020), who identified the child ratio as a contributing factor to multidimensional poverty. In summary, the empirical results suggest that the high child ratio in Pakistani households exacerbates the risk of multidimensional poverty. Policymakers should address this issue through targeted interventions designed to improve women’s reproductive health, education, and access to employment opportunities.

Employment characteristics

Household employment characteristics were captured by three indicators: youth employment, household labor force, and female economic participation. The results demonstrate that the engagement of young people in economic activities reduces the likelihood of being multidimensionally poor by 0.39% at the household level in Pakistan. Investing in youth through training, education, and job opportunities in both the public and private sectors can be a key strategy for achieving the SDGs and promoting long-term development in Pakistan. The labor force ratio of households reduced the probability of being multidimensionally poor by 0.42%. As an essential demographic factor, the labor force ratio can help reduce the likelihood of multidimensional poverty. Households with a large proportion of working-age members are more likely to escape poverty. In addition, the results highlight that the participation of more female family members in the labor market lowers the probability of being multidimensionally poor by 0.12%. The findings of this study are similar to those of previous studies (Mulugeta 2021; Sohail 2014). Given Pakistan’s male-dominated social structure and the limited opportunities available to females in terms of education, decision-making, and earnings, promoting greater female participation in the labor force could have significant positive impacts on the nation’s economy and society as a whole.

Columns 2 and 3 of Table 4 present the regional results and columns 4 to 7 the provincial results regarding the effect of the household demographic characteristics on multidimensional poverty subgroups. Overall, the results are similar to the main findings, with a few variations. The demographic characteristics of urban and rural regions show effects on poverty status like those shown for the national level. However, family type does not have a significant impact on household poverty in urban regions. The results regarding the family structure, population structure, and employment structure characteristics for the Punjab and Sindh provinces are similar to those shown at the national level. Moreover, the results indicate that all household demographic indicators except the youth employment and labor force ratios are significant contributors to multidimensional poverty in Balochistan province households. The empirical findings suggest that the female employment ratio does not have a significant impact on multidimensional poverty in the KPK. This can be attributed to the conservative cultural and social norms prevalent in these areas, which limit women’s participation in economic activities. In such areas, women’s roles are typically confined to household chores and child rearing, and their access to education and employment opportunities is limited. As a result, the female employment ratio had a negligible effect on reducing multidimensional poverty in KPK. This finding is consistent with the literature on gender inequality and poverty in developing countries, highlighting the need to address cultural and social barriers to women’s economic empowerment in order to promote sustainable and inclusive development.

Most of the control variables—such as household head characteristics, ownership of property by households, remittances received by households, and region and province of residence—significantly affect the multidimensional poverty status of households, consistent with the literature.

Robustness and diagnostic checks

We used alternative multidimensional poverty cut-offs as a robustness analysis to measure the sensitivity of the multidimensional poverty estimates, as shown in Table 5. The results indicate that the overall MPI value and H decreased as poverty cut-offs increased, whereas A increased along with higher poverty cut-offs. This result indicates that MPI, H, and A are sensitive to changes in poverty cut-off values. The poverty ranking between regions and provinces remained robust at various cut-off values. The rural region is poorer than the urban region, and Balochistan is the poorest province, followed by Sindh, KPK, and Punjab at all cut-offs.

Table 5 Multidimensional poverty estimates at different poverty cut-offs
Full size table

We evaluated the accuracy of our estimated model using the receiver operating characteristic (ROC) curve diagnostic test, which is a graph of sensitivity versus one minus specificity, to differentiate between sensitivity and specificity (Zou et al. 2007). Here, sensitivity is the likelihood of accurately identifying positive multidimensionally poor households that are correctly classified, and specificity is the likelihood of correctly identifying true negative multidimensionally non-poor households. The ROC curve test results in Fig. 4 show that the area under the ROC curve for logistic regression is 0.84, which is sufficiently large to imply that our model correctly classified the negative and positive outcomes. Hence, the model is deemed sufficiently accurate for analysis.

Fig. 4
figure 4

ROC for the estimated model

Full size image

Conclusion and policy recommendations

This study measured poverty in Pakistan from multiple dimensions and analyzed the impact of household demographic characteristics on multidimensional poverty using the HIES 2018-19 dataset. The study developed an MPI for Pakistan to capture the non-monetary and monetary dimensions of household poverty using the AF methodology. The MPI comprises four equally weighted dimensions: education, health, living standards, and monetary poverty. One of the unique benefits of the MPI is that it enables a joint estimation of the monetary and nonmonetary dimensions of household poverty.

This study provides fresh estimates, finding that 22% of Pakistani households suffer multidimensional poverty in 25% of the selected poverty dimensions, as reflected by monetary and non-monetary indicators. Moreover, the study finds heterogeneity in multidimensional household poverty status between regions and across provinces. The poverty ranking between regions and provinces is robust to various poverty cut-offs. The MPI was significantly higher in rural regions, indicating that poverty is more evident in rural areas. Baluchistan has the highest poverty level among the provinces, which calls for urgent action. The contribution of the education dimension to multidimensional poverty is highest at the national, regional, and provincial levels. Most households are deprived of basic health facilities, indicating that the health system in Pakistan is inefficient and cannot provide sufficient healthcare facilities (Kurji et al. 2016; Farooq and Ali 2020).

Further, logistic regression was employed to explore the impact of household demographic characteristics (i.e., family size, family type, child ratio, dependency ratio, youth employment ratio, female employment ratio, labor force ratio) on multidimensional household poverty. The results show that large demographic characteristics have significant effects on households’ multidimensional poverty status. The evidence suggests that households with greater household sizes, dependency ratios, and child ratios are more likely to be poor. The findings also indicate that female economic participation reduces the probability of households being multidimensionally poor. Similarly, the engagement of young people in employment activities significantly reduces multidimensional poverty.

The empirical results offer important implications useful for policies aimed at eliminating multidimensional poverty. First, the MPI findings suggest that policymakers should consider both monetary and nonmonetary dimensions when attempting to address multidimensional poverty. Government institutions should also carefully consider regional disparities when designing policies intended to alleviate poverty. One approach could be to conduct a comprehensive analysis of poverty and its determinants at the sub-national level, identifying the areas and populations most in need of assistance. Based on this analysis, the government could allocate resources more effectively and implement targeted poverty-reduction programs in these areas. The government could also prioritize infrastructure development in the regions most affected by poverty. This would include improving access to basic services, such as education, health care, clean water, and sanitation. Governments could also work to enhance transportation networks, communication systems, and electricity supplies, which are critical to economic growth and poverty reduction. Moreover, the government should undertake initiatives to expand access to primary and secondary school education for poor households across the country. Health infrastructure should be strengthened through the construction of more hospitals and by increasing the number of medical personnel to ensure sufficient basic health facilities for poor households. Social protection programs targeting vulnerable households, such as cash transfers, can also help reduce monetary poverty. Second, this study finds that households with higher dependency and child ratios are more likely to be multidimensionally poor. These findings suggest the need for effective policies aimed at controlling population growth, such as introducing social awareness programs to educate about family planning services, particularly in rural areas. Subsidies for family planning services should also be provided to low-income households. Third, the study finds that female and youth participation in the labor market is associated with lower multidimensional poverty. This finding suggests the need for more employment opportunities and the creation of an environment conducive to increasing female and youth labor market participation. Policies for promoting gender equality, such as providing equal access to education and job opportunities for women, can also help reduce poverty. Women’s economic participation can be enhanced by investing in female human capital, enacting gender-sensitive laws, and improving social norms.

This study is limited in terms of its survey time, as its dataset was drawn from a single survey period. The use of panel survey data could improve research in this field by assessing dynamic changes in multidimensional poverty and exploring how it is impacted by household demographic characteristics.