1 – Introduction

1 In recent years, poverty has come to be seen as a multidimensional phenomenon that goes beyond the monetary aspect. A number of authors have grasped this multidimensional nature of poverty, including Townsend (1979) and Atkinson (2003), as well as some more recent authors. Studies by Ravallion (1996; 2010), Sen (1976), and Thorbecke (2008) show that poverty can be synonymous with poor health, lack of education, low income, precarious accommodation, hard or unprotected jobs, political disempowerment, and food insecurity.

2 There exist several different theoretical approaches to a multidimensional understanding of poverty, including the basic needs approach and the capability approach. Generally speaking, these approaches define poverty as a lack of opportunities or valued functionings (Alkire and Foster 2007; Sen 1984; 1985; 1987).

3 In Congo, some studies have been conducted using monetary and non-monetary indicators to define poverty. They include works by Ambapour (2006), Backiny-Yetna and Wodon (2009), Notten et al. (2008), Ambapour and Bidounga (2012), UNICEF (2018), and Ouadika (2018). These have all facilitated the analysis of household poverty and child poverty, but they have not provided data on the factors contributing to multidimensional household poverty in Congo. The research described in this article is distinguished from other studies on poverty in Congo due to its use of the Alkire-Foster method to measure the incidence of poverty and its attempt to analyze inequalities and identify the determinants of multidimensional household poverty.

4 The authors used the Alkire-Foster method in the first instance to estimate multidimensional poverty and to decompose the indicators of multidimensional poverty for different groups. Regression modeling was then used to identify the determinants of multidimensional poverty for Congolese households.

2 – Literature review

5 Over the last decade, research on approaches to understanding poverty has grown in width and depth, though to a limited extent. There has been an evolution in approaches to poverty and to conceptualizing, defining, and measuring deprivation. Several theoretical approaches have been used to assess multidimensional poverty, including the axiomatic approach, the dominance approach, the statistical approach, and the fuzzy set approach.

6 The axiomatic approach introduced by Sen (1976) is the norm for measuring multidimensional poverty. The Sen index (Sen 1976), the Sen-Shorrocks-Thon index (Shorrocks 1995), and the Foster-Greer-Thorbecke (FGT) index (Foster et al. 1984) are all traditional methods for measuring poverty that take an axiomatic approach. The criticisms of this approach focus on the identification and aggregation of the poor. Some measures of multidimensional poverty are based on axiomatic properties that extend the framework of unidimensional poverty (Alkire and Foster 2007; Atkinson 2003; Bourguignon and Chakravarty 2003). Alkire and Foster (2007) proposed the Alkire-Foster counting method based on Sen’s capability approach (1976).

7 The dominance approach is used for analyzing unidimensional poverty and inequalities (Atkinson 1987; Foster et al. 1984) and has also been applied to multidimensional poverty (Bourguignon and Chakravarty 2003; Duclos et al. 2006). Furthermore, when the dominance approach is used, poverty comparisons are not dependent on the choice of poverty measures or parameter values (Alkire et al. 2015). However, the dominance approach cannot be used to compare the intensity of poverty between two countries or regions, and the sample size required increases exponentially with the number of dimensions compared (Alkire et al. 2015).

8 The statistical approach is essentially based on descriptive methods and data analysis models. In particular, descriptive methods are used for summarizing a set of data. They include principal component analysis (PCA) and multiple correspondence analysis (MCA). Model-based methods include latent class analysis, factor analysis, and structural equation modeling, and may be used to make inferences. However, some statistical methods are problematic. For example, statistical methods may violate certain axioms such as deprivation focus or monotonicity. It is also difficult to create a poverty cutoff using deprivation scores derived from PCA (Alkire et al. 2015).

9 The fuzzy set approach introduced by Zadeh in 1965 is used to construct a multidimensional poverty index and to conduct decomposition across sub-groups. Many other studies have used the fuzzy set approach to assess poverty, for example Belhadj (2011), Betti and Verma (2008), Costa and De Angelis (2008), and Leu et al. (2016). As with the statistical approach, the fuzzy set approach violates several axioms, such as focus and weak transfer (Alkire et al. 2015).

10 At the empirical level, several studies have used the Alkire-Foster method to measure multidimensional poverty using three main dimensions of deprivation, namely education, standard of living, and health (Alkire and Santos 2014; Batana 2013; Alkire et al. 2018). For example, Batana (2013) used assets, health, education, and empowerment to estimate multidimensional poverty among women in Sub-Saharan Africa and noted that there were differences in deprivation by country and that the incidence of poverty was higher in rural areas. Similarly, when using three dimensions of deprivation, Tigre (2018) showed that standard of living contributed the most to multidimensional poverty, while education and health contributed the least. Binary logistic regression showed that higher levels of education, having a bank account, and larger numbers of working-age family members reduced multidimensional poverty significantly in Ethiopia, while higher numbers of children under five years old and a greater dependency ratio increased poverty significantly.

11 Adeoti (2014) also used the Alkire-Foster method and a binary logistic model to measure multidimensional poverty in Nigeria. The results showed that larger household size, the head of the household being female, working in the agricultural sector, and living in the north of the country increased the likelihood of being poor in rural areas of Nigeria, while greater educational attainment, working in the non-agricultural sector, and living in the southeast or southwest reduced it. The health, assets, and education dimensions were shown to contribute the most to multidimensional rural poverty in Nigeria. Khan et al. (2020) noted that between 2010 and 2014, the contribution of the household assets dimension to the multidimensional poverty index (MPI) increased, while the contribution of the health and education dimensions decreased.

12 Working on similar lines, Amao et al. (2017) used the Alkire-Foster method and a logit model in their research. They showed that larger household size, the head of the household being female, and a higher dependency ratio significantly increased the likelihood of being poor in rural areas of Nigeria, while land ownership and non-agricultural wages significantly decreased it. Nevertheless, it was the standard of living and education dimensions that contributed the most, while the health and assets dimensions contributed the least to the MPI for rural areas in Nigeria. Through a binary logistic regression, Michael et al. (2019) showed that household size, age, and marital status had a positive effect on the likelihood of experiencing multidimensional poverty. In addition, the likelihood of being poor increased with being female (rather than male). Multidimensional poverty decreases with the increasing education level of the household head. Their results also show that respondents who undertake more activities are more likely to be non-poor compared to their counterparts with fewer activities. Similarly, increasing farm size improves household food security status. Also, livestock ownership significantly affects the poverty status of respondents. Similarly, remittances showed a significant relationship with poverty, and having access to credit decreases the probability of being poor.

13 Using the Alkire-Foster method and a multilevel modeling technique, Chen et al. (2019) found different multidimensional poverty profiles in the four districts of Taiwan. They also found that microlevel factors such as age, socioeconomic status, marital status, household income, and household size, as well as macrolevel factors such as the level of urbanization and the service-to-manufacturing ratio, correlated significantly with the level of multidimensional poverty. Similarly, White and Yamasaki (2017) noted that age, sex, race and ethnic origin, education, marital status, employment status, household language, and household size were significant determinants of multidimensional poverty in native-born and foreign-born American households.

14 With regard to Congo, Ambapour and Bidounga (2012) analyzed multidimensional poverty using fuzzy logic. Their main conclusions were that sanitation contributed the most to multidimensional poverty in Congo and that such poverty was most widespread in rural areas and among women. Backiny-Yetna and Wodon (2009) analyzed the profile and determinants or correlates of poverty in Congo by using linear regression to analyze the determinants of monetary poverty. Their results showed that poverty is a rural phenomenon. Education and access to infrastructure have a positive effect on households’ well-being, while larger household size aggravates monetary poverty.

15 Ouadika’s doctoral thesis (2018) analyzed multidimensional poverty in Congolese households based on data from the 2011 Congolese household survey, using the Alkire-Foster method. He used three dimensions (health, education, and standard of living) to analyze household poverty. He concluded that multidimensional poverty strikes in a very similar way regardless of place of residence, with a high MPI value in densely populated cities, and that the dimensions of health, level of education, and ownership of durable goods contributed strongly to poverty in Congolese households. Mboko Ibara and Ossouna (2021) also analyzed multidimensional poverty, focusing more on poverty among children under five years of age. These authors first used the Alkire-Foster method and then a regression modeling technique to determine the microlevel factors contributing to multidimensional poverty. They did not focus on factors influencing multidimensional poverty in Congolese households, however.

3 – Methodological approach and data source

16 To conduct this study, we used the multidimensional poverty index (MPI). This index belongs to the family of multidimensional poverty measures developed by Alkire and Foster (2007; 2011), Alkire, Foster, Seth, Santos, Roche, and Ballon (2015), and Alkire and Jahan (2018). A multidimensional approach to measuring poverty is essential for creating policies to combat poverty and inequalities in living standards.

3.2 – Choice of dimensions

18 In general, the construction of a global MPI for measuring multidimensional poverty relies on three basic dimensions: health, education, and household living standards. The MPI uses three equally weighted dimensions and ten indicators. Each household’s deprivation profile is evaluated through these ten indicators. The dimensions, indicators, and different cutoffs included in the global MPI can be modified according to policy priorities and the objectives to be achieved by political decision-makers (Alkire and Santos 2013).

19 Once the MPI was determined, this study used a regression modeling technique to determine the microlevel factors contributing to multidimensional poverty.

20 Specifying a linear relationship between the variable y_i and the explanatory variables X_i causes several problems, including heteroscedasticity of errors. To avoid these problems, the solution used was to assume that y_i is the manifestation of an unobservable continuous variable, itself linked to the explanatory variables X_i. We adopted a logit model to analyze the probability of a household being multidimensionally poor. This model was chosen because the profile variable is a probability. Let Y be the dependent variable representing the occurrence of the event of interest: Y = 1 if the household is in multidimensional poverty and Y = 0 if not, and X₁, X₂, … X_p are the P predictors. To this end, a logit model with the following form can therefore be postulated:

21

22 where the errors u_i are independent and identically distributed, and follow a zero-expectation logistic distribution with standard deviation σ.

23 The above specification, after transformation, is rewritten as follows:

24

25 let’s say: , so the logit model is rewritten as follows :

26

27 β₀; β₁; β₂; β_p are the parameters that we want to estimate from the data.

28 The estimation of the parameter b in this model is done by the Maximum Likelihood (ML) method. The estimation of the final model was done by the “mixed stepwise method” using the likelihood ratio test for the introduction and elimination of variables at the 5% thresholds.

Table 1

Dimensions, indicators, weights, and cutoffs for constructing the MPI

Dimensions	Indicators	Is deprived if	Relative weight
Health
	Infant mortality	A child under five years old has died in the family.	1/6
	Nutrition (Food Consumption Score)	There is at least one undernourished household member.	1/6
Education
	Years of schooling	A household member aged over ten years has an education level below CM1 (cours moyen niveau 1, the expected level for children aged 9–10 years).	1/6
	School attendance	Any school-aged child is not attending school up to the age at which they would complete class 8	1/6
Living standards
	Electricity	The household does not have access to an electricity connection from the Société Nationale d’Électricité.	1/18
	Sanitation	The household’s sanitation facility is not improved.	1/18
	Drinking water	The household does not have access to drinking water.	1/18
	Floor material	The floor of the household’s accommodation is made of earth, sand, or wood.	1/18
	Cooking fuel	The fuel used for cooking is something other than gas.	1/18
	Assets	The household does not own more than one of these assets: mobile phone, television, radio, motorbike, or refrigerator, and does not own a car or truck.	1/18

Dimensions, indicators, weights, and cutoffs for constructing the MPI

4 – Results

5 – Discussion

37 This study used the Alkire-Foster method to estimate multidimensional poverty. It also analyzed poverty inequalities and important determinants that might contribute to multidimensional poverty.

38 The results show that, in the Republic of the Congo, multidimensional poverty is essentially a rural phenomenon. Living in a rural area of Congo contributes more to multidimensional inequality than does living in an urban area. Living in a rural area multiplies a household’s risk of experiencing multidimensional poverty by 1.3 compared with living in an urban area. Taking marginal effects into account confirms this finding. Households living in rural areas increase their likelihood of experiencing multidimensional poverty by 3.3%. This finding corroborates those obtained in most Sub-Saharan African countries, particularly those of Adeoti (2014) and Batana (2013). Several studies identify the fact that poverty is more established in rural areas because of a lack of infrastructure. In the case of Congo, Backiny-Yetna and Wodon (2009) also reported that the probability of remaining poor is greater for people living in rural areas. This is explained by the level of basic infrastructure, which is predominantly concentrated in urban centers.

39 The findings also show that rural women are more likely than urban women to experience multidimensional poverty. Households headed by women have 1.2 times the risk of falling into multidimensional poverty of those headed by men. Households headed by women see their likelihood of experiencing multidimensional poverty rise by 2.5% (Table A3, Appendixes). This could be due to women having less access to productive capital than their male peers. This finding matches the conclusions drawn by Amao (2017) and Adeoti (2014). Household size has a significant positive association with multidimensional poverty—in other words, larger households are more likely than smaller ones to be multidimensionally poor. A similar finding was obtained by White and Yamasaki (2017), who noted that household size was one of the significant determinants of multidimensional poverty among American households. This is also the case in several African countries (Amao 2017), where households need more basic assets to survive. Level of education has a significant positive effect on a household’s probability of experiencing multidimensional poverty. Inequalities between households whose head has not had any education are greater than those among households whose head has completed at least the second cycle of secondary education. Multivariate analysis shows that households whose head has completed at least the second cycle of secondary education are less exposed to multidimensional poverty than those whose head has not had any education.

40 The marginal effects obtained show that the probability of a Congolese household falling into multidimensional poverty increases by 24.2% (Table A3, Appendixes) if the head of household has not had any education. This supports Bastos and Machado’s conclusions (2009) that education increases the stock of human capital, which in turn increases work productivity and pay. This can be explained by the fact that education makes it easier for people to obtain paid work. These results confirm the positive role of education on the likelihood of experiencing multidimensional poverty. Research by Garza-Rodriguez et al. (2015) in Mexico showed that level of education is a fundamental factor in escaping extreme poverty. Wealth quintile also explains multidimensional poverty.

41 In the light of these findings, specific policies are required to deal with living standards and (separately) with health and education. Nevertheless, to reduce multidimensional poverty in the long term, action must not be limited to making more services available to populations; a focus on the quality of the services is also required. Moreover, all the coefficients of the different wealth quintiles are large and significant at the 5% level. However, they are negatively correlated with the probability of a household being poor. This shows that as household wealth increases from one quintile to the next, the probability of a household experiencing multidimensional poverty decreases. Households belonging to the wealthiest quintile have 0.006 times the risk of being exposed to multidimensional poverty of households in the poorest quintile. In other words, poor households have a very high risk of falling into multidimensional poverty. All other things being equal, the age of the head of household has a positive effect on a household’s probability of experiencing multidimensional poverty. Older adults have significantly higher levels of multiple deprivation than younger adults. In Benin, Hodonou et al. (2010) found that the age of the head of household was positively associated with poverty; the older the head of household, the greater the probability that the household would remain in poverty.

6 – Conclusion and recommendations

42 In this study, the authors estimated multidimensional poverty and analyzed inequality in living standards among Congolese households, using a methodology developed by Alkire and Foster, supplemented by inequality analysis tools. Multidimensional poverty was analyzed using a multidimensional poverty index (MPI) based on ten indicators across three main dimensions. Our estimates confirm the multidimensional nature of poverty in Congo. Rural women are more likely than urban women to experience multidimensional poverty. The findings also show that multidimensional poverty is principally a rural phenomenon, although urban areas also experience non-negligible levels of deprivation with regard to access to sanitation facilities, cooking fuel, and electricity. Using the Gini index on the poverty index shows that inequalities are more pronounced in rural areas than in urban ones.

43 Analysis by geographical area shows that the departments of Likouala, Lékoumou, Cuvette-Ouest, Pool, and Kouilou are the most affected by multidimensional poverty as measured by the MPI in Congo. Our estimates also show that under-eighteens experience poverty more acutely than do adults.

44 However, because of a lack of data, it was not possible to include indicators such as employment, security, and governance, which are important in analyzing individuals’ well-being. As a result, there is room for considerable improvement of this index, first by widening the dimensions of employment and social security, and second by making this methodology applicable in fields other than the measurement of well-being, specifically in the follow-up and evaluation of programs to combat poverty and inequality.

45 This study found that household size, the sex of the head of household, age, level of education, place of residence, and wealth quintile are important determinants of multidimensional poverty in Congo. These factors should allow households at risk of falling into multidimensional poverty to be identified.

46 In the light of these findings, policies for reducing multidimensional household poverty should aim to improve access to sanitation facilities, adequate nutrition, cooking fuel, electricity, and education. An effective response to combat poverty in Congo will need to find solutions to these issues and focus on vulnerable populations in rural areas and, in particular, on under-eighteens. For anti-poverty policies to be effective, social assistance will need to be targeted at the departments or groups of households experiencing a very high intensity of poverty, as meeting their basic needs will have the greatest impact on reducing poverty at the national level. It is therefore conceivable that social policies for combating poverty could focus on rural areas, by, for example, creating training centers for rural trades and occupations to encourage poor households living in these areas to take up training so that they can provide for their own basic needs in the domains of education, health, and living standards.