Data sources

In this study, we use data from the second Congolese household survey to evaluate poverty in 2011 ( CNSEE, 2011) driven by the Congolese National Institute of Statistics (INS). The sampling for this survey was a two-stage stratified survey, with a draw proportional to the size of the enumeration areas, in the first stage, and in the second stage, a simple random sample of households enumerated in each enumeration area. The choice of these is due to the opportunity offered by this survey to analyze poverty. From this base, we selected 929 households whose heads had a university education level. Our target population consists of male and female heads of households with a higher level of education.

Study variables

Several variables will be used in this research. The main (endogenous) variable is poverty, estimated from household expenditure per adult equivalent. The primary variable of interest is the level of university education.

The endogenous variable of our study is household poverty captured by the relative approach, following Haughton and Khandker (2009). This study’s “poverty” variable is obtained from the household’s expenditure per adult equivalent. Thus, we define the monetary poverty threshold as a fixed proportion of the level of expenditure per adult equivalent, i.e., 60 percent of the average. This setting allows us to split our population into two modalities. Thus, households below this line will be considered poor (modality 1). Those above will be non-poor (modality 0).

The primary exogenous variable in our study is household heads’ university education level. Indeed, we seek to know whether, in the context of Congo, higher education helps reduce poverty. This variable is coded as follows: First/second year of university = 31; Third/fourth year of university = 32; Fifth year or more = 33. The sign of the expected coefficients is negative (-). Poverty is expected to decrease each time university education increases.

The exogenous control variables for the poverty-university education relationship are household size, age, gender, marital status, and economic activity status: ➢

Household size: This variable measures the number of people in the household. The expected coefficient sign is positive (+). The larger the household size, the poorer the household.

Ethics statement

The household consumption survey (ECOM 2) was conducted in 2011 by the National Center for Statistics and Economic Studies (currently the National Institute of Statistics-INS) of the Republic of Congo to assess poverty in the country. The INS is the public body responsible for collecting, processing, analyzing, and publishing official statistics.

Law No. 8-2009 of October 28, 2009, in its chapter 2, and article 8 stipulates that: “Natural and legal persons must respond, accurately and within the set deadlines, to censuses and statistical surveys carried out using questionnaires or other forms developed by public administrations of the national statistics system.” The INS regularly collects official data from administrations and households using that disposal of low. In the case of the data from this survey, they were approved by the Minister responsible for statistics on November 11, 2011.

Our research methodology is structured in two points: the presentation of poverty indicators and the method of analyzing the results. This research uses the Lorenz curve, the relative poverty index, and the Gini index, based on the Duclos and Araar (2006) equally distributed equivalent incomes concept. That concept is then applied to distributions whose incomes have been censored at the poverty line.

Presentation of poverty and inequality indicators -

The incidence of poverty

The incidence of poverty apprehended in this study is part of the Foster-Greer-Thorbecke (1984) indices for measuring poverty. These authors developed a family of indices considering the size of the total population N, the poverty level or line z, the income of the individual yi, a parameter α, which is an indicator of a version to poverty and I( yi ⩽ z) a function taking 1 if yi ⩽ z and 0 if not. FGT α = 1 N ∑ i = 1 N ( z − y i z ) α I ( y i ≤ z ) (1)

Modifying the parameter α allows one to move from one index to another. The larger this parameter, the greater the weight given to the poorest individuals in calculating the poverty rate. Where α equals zero, the FGT index divides the number of poor people (people below the poverty line) by a country’s total population. FGT 0 = 1 N ∑ i = 1 N I ( y i ≤ z ) (2)

Thus, when α = 0, the FGT index gives the most straightforward and commonly used poverty index ( Duclos and Araar, 2006). It is called the poverty rate and is simply the proportion of a population that lives in poverty. The shorter term “poverty rate” is sometimes intended to indicate the absolute (as opposed to relative) number of poor people in the population. -

The Lorenz Curve

The Lorenz curve is the most popular tool for visualizing inequality and comparing it across given distributions. For a given percentile pi, the Lorenz curve shows the share of total population income held by the poorest people. Formally, the Lorenz curve in percentile pi is given by: L ( pi ) = ∑ j = 1 i Q ( pj ) ∑ j = 1 n Q ( pj ) = ∑ j = 1 i y j ∑ j = 1 n y j (3)

For several decades, this curve has been the most popular graphical tool for visualizing and comparing income inequality ( Duclos and Araar, 2006). It provides complete information about the entire income distribution relative to the mean. It, therefore, provides a complete description of relative incomes than any of the traditional summary dispersion statistics, and it also provides a better starting point for studying income inequality than calculating the many inequality indices that have been proposed. -

The Gini coefficient

The Gini index or coefficient is a statistical measure used to account for the distribution of a variable (income, wealth) within a population. It measures the degree of inequality in a country’s income distribution. Developed by Corrado Gini in 1912, this index varies between 0 (perfect equality) and 1 (perfect inequality) ( Duclos and Araar, 2006). Alternatively, the Gini index can be defined as half of the average relative Gini difference of the income series: G = E 2 M (4) -

where M is the average of incomes and E, the Gini’s average difference of incomes. That is, if ( x _i ) _{1 ⩽ i ⩽ n} were the n individual incomes: G = 1 n 2 ∑ i = 1 n ∑ j = 1 n | x i − x j |

The method of analyzing the results:

To estimate the effect of university education on poverty, we use a model based on nonlinear binary regression, given the binary nature of our poverty variable (1 for poor and 0 otherwise), such variables can be analyzed using a probit model ( Greene, 2019). In such a model, the expression of the probability of being poor can be written: P r ( y = 1 / x ) = Φ ( β ′ x ) (5) •

Pr is the probability of being poor;

However, when there are measurement errors and omitted simultaneity in econometric models, the explanatory variables can be correlated with random components, and standard estimation methods no longer provide consistent estimates of the model parameters ( Kim & Frees, 2006). Three sources of correlation errors can be encountered ( Kim & Frees, 2006): the omission of certain exogenous variables, measurement errors of the variables, and simultaneity bias. Several studies addressing endogeneity have focused on resolving level 2 endogeneity ( Rabe-Hesketh and Skrondal, 2008; Mboko and Ikiemi, 2021).

In this research, we use the simultaneous equations approach proposed by Steele, Vignoles, and Jenkins (2007) to consider the supposed endogeneity of employment on poverty. Indeed, seeking to estimate the equations individually assumes that the error terms of the different equations are independent of each other (a strong assumption for our estimates). That is why the Full information maximum likelihood method was preferred. The technique used to carry out this method is the so-called “conditional mixed process (CMP)” developed by Roodman (2011). This model allows us to consider the variables’ causality by avoiding the endogeneity bias ( Roodman, 2011) and to study the impact of university education on poverty in sub-Saharan Africa. Thus, the probability for an individual to have a paid job and to be poor is respectively given by the following latent variables: y 1 ∗ = β 1 ′ x 1 + ε 1 , y 1 = 1 si y 1 ∗ > 0 , y 1 = 0 , otherwise , (6) y 2 ∗ = β 2 ′ x 2 + δ y 1 + ε 1 , y 2 = 1 si y 2 ∗ > 0 , y 2 = 0 , otherwise , (7)

where: E [ ε 1 | x 1 , x 2 ] = E [ ε 2 | x 1 , x 2 ] = 0 , Var [ ε 1 | x 1 , x 2 ] = Var [ ε 2 | x 1 , x 2 ] = 1 , Cov [ ε 1 , ε 2 | x 1 , x 2 ] = ρ [ ε 1 , ε 2 ] ∼ Φ 2 [ 0 , 0 , a , 1 , ρ ] , ρ ϵ [ − 1 , 1 ]

Φ ₂ denotes the distribution function for a bivariate standard normal distribution, ρ being the dependence parameter measuring the endogeneity of y 1 in the equation of y 2 ∗ . δ is the coefficient on the outcome variable of the first-step analysis appearing in the second-step equation.

The variables in our model are:

y 1 = Employment

y 2 = Poverty

x 1 = Independent variables in the employment equation

x 2 = Independent variables in the poverty equation

Such a recursive model of simultaneous equations can be estimated using the full information maximum likelihood approach.

Given the probabilities in (6) and (7), the conditional mean of being poor can then be written as: E [ y 2 | y 1 = 1 , x 1 , x 2 ] = Pr [ y 2 = 1 , y 1 = 1 | x 1 , x 2 ] Pr [ y 1 = 1 | x 1 , x 2 ] = Φ 2 ( β 1 ′ x 1 , β 2 ′ x 2 + δ , ρ ) Φ ( β 1 ′ x 1 ) (8)

On the other hand, the unconditional mean function can then be written as E [ y 2 | x 1 , x 2 ] = Pr [ y 1 = 1 | x 1 , x 2 ] E [ y 2 | y 1 = 1 , x 1 , x 2 ] + Pr [ y 1 = 0 | x 1 , x 2 ] E [ y 2 | y 1 = 0 , x 1 , x 2 ] = Pr [ y 2 = 1 , y 1 = 1 | x 1 , x 2 ] + Pr [ y 2 = 1 , y 1 = 0 | x 1 , x 2 ] = Φ 2 ( β 1 ′ x 1 , β 2 ′ x 2 + δ , ρ ) + Φ 2 ( − β 1 ′ x 1 , β 2 ′ x 2 , − ρ ) (9)

This technique not only has the advantage of taking into account a greater diversity like endogenous variables (discrete, censored, continuous) but is also adapted and appropriate mainly for recursive models insofar as it allows to take into account the relationships that may exist between the different equations (endogeneity problem) and manage selection and simultaneity biases ( Mboko and Ikiemi, 2021). Hence, the systemic estimation of our system will lead to more efficient estimates than the individual estimation of each equation ( Roodman, 2011). This is what characterizes the advantage of this method compared to others, in particular, the two-step Heckman approach, which does not consider simultaneously several equations with dependent variables of different natures ( Mboko and Ikiemi, 2021).

Presentation and interpretation of results

This section is devoted to presenting and interpreting the results of the descriptive and explanatory analyses on poverty. We used the variable of expenditure per adult equivalent to measure household poverty in the population of heads of households with a higher level of education. Poverty was approached from the poverty incidence indicator developed by Foster, Greer, and Thorbecke (FGT) in 1984. For this reason, the poverty threshold retained was set as 60 percent of the average expenditure per adult equivalent of the household. This threshold corresponds to 405109 CFA francs.

The results of the descriptive analyses of poverty show that in 2011 Congo ( Table 1), 33.6 percent of heads of households with a higher level of education were poor. The effects of higher education on poverty are very noticeable. Indeed, the poverty rate is 40.4 percent among heads of households who have not reached the third year of university and only 27.8 percent among those with a master’s degree or higher. Hence, a university education mechanically reduces poverty and improves the population’s well-being.

The analysis of the Gini index ( Table 2) shows a non-negligible level of inequality at 0.37. It is in the class of the most educated heads of household (Master’s degree or more) that we find the most inequality in expenditure (0.41). These results are visible on the Lorenz curve, where we observe a clear demarcation of the curve of the fifth year or more, far from the 45-degree line than the other curves. This distance means that inequalities increase with the level of education in higher education.

Table 3 presents the recursive simultaneous equations model estimates from the conditional mixed model of Roodman (2011), considering the endogeneity of employment on university education. It presents both the coefficients and the marginal effects. The values in parentheses are standard errors.

The Wald test statistics for the overall model fit are significant at the 1 percent level. The atanhrho statistic is non-significant with a positive correlation between the error terms of the two equations of the order of 7 percent. These last two pieces of information on the model fit suggest the absence of endogeneity of the exercise of economic activity on poverty in Congo.

The employment equation analysis shows a non-significant relationship between employment and poverty, all else equal. This result is inconsistent with the work of Siddique (2023), who showed that good-quality jobs helped households escape poverty while poor-quality jobs did not. This author suggested that poverty eradication requires the government to direct labor market policies to focus more on distributing jobs from individuals to households and transforming bad jobs into good jobs rather than simply creating new jobs in the economy. Among the control variables, only household size is significantly related to poverty. Thus, all else being equal, household size increases with the poverty level. In other words, the poorest households are those with many people.

The poverty equation shows that university education has a very significant negative impact on household poverty in Congo. Indeed, even after controlling for economic activity and other characteristics, the coefficients associated with households whose head has a bachelor’s/master’s degree or a master’s/doctorate are significant at the 1 percent level. Hence, the probability of remaining poor for a household whose head is of a higher level decreases significantly with the improvement in the level of university education.

Table 3 also presents the marginal effects calculated from the sample mean. These are the marginal effects on the probabilities of realization (success) of the two simultaneous equations (Work and poverty) retained in the model. The interpretation will be limited to the main variable of interest (Level of university education). University education acts directly on poverty (poverty equation). Compared to heads of households with 1st/2nd year of university education, heads with 3rd/4th year have 12 percent less risk of finding themselves in poverty. Those with 5th year of university education or more have 10 percent less risk of being poor than their counterparts in the reference modality.

The household consumption survey (ECOM 2) was conducted in 2011 by the National Center for Statistics and Economic Studies (currently the National Institute of Statistics-INS) of the Republic of Congo to assess poverty in the country. The INS is the public body responsible for collecting, processing, analyzing, and publishing official statistics.

Law No. 8-2009 of October 28, 2009, in its chapter 2, and article 8 stipulates that: “Natural and legal persons must respond, accurately and within the set deadlines, to censuses and statistical surveys carried out using questionnaires or other forms developed by public administrations of the national statistics system.” The INS regularly collects official data from administrations and households using that disposal of low. In the case of the data from this survey, they were approved by the Minister responsible for statistics on November 11, 2011. The data is collected by the public body responsible for collecting it for the state, so there is no recourse to the ethics committee.