Poverty Reduction through Growth, Redistribution and Social Inclusion in Times of COVID-19: Kenyan Evidence on the Underlying Mechanisms

Abstract

The paper looks at the nexus between growth, poverty, inequality and redistribution in Africa, using Kenya as a case study. The existing literature shows a strong causal link from growth to poverty reduction. This link is the basis for the pro-poor poverty reduction strategy. There is evidence from the AERC studies that, poverty reduction in a given period is associated with higher growth rates in successive periods that are inequality-reducing and conceptually long lasting. This virtuous spiral of poverty reduction, higher growth and less inequality over time, is the basis for the pro-growth poverty reduction strategy that has recently been emphasized in the literature ( Thorbecke and Ouyang, 2022). The two poverty reduction strategies, a pro-poor strategy and a pro-growth poverty reduction one, complement each other, sustaining household escapes from poverty over time. The paper provides evidence from Kenya showing that human capital formation is the key mechanism underlying the virtuous spiral of lower poverty, higher growth and less inequality as the economy progresses through time. A perspective on robustness of the virtuous spiral in the context of COVID-19 and other pandemics is offered in the concluding section of the paper.

1. Introduction

Recent AERC Framework Papers on growth, poverty, inequality and redistribution show that Sub-Saharan Africa is home to countries with the highest poverty and inequality levels in the world. The framework papers further show that inclusive growth reduces both poverty and inequality. Moreover, a dynamic virtuous spiral of lower poverty, higher growth and less inequality is critically discussed. The policy measures associated with this spiral are commonly referred to in the literature as pro-growth poverty reduction strategies (Eastwood and Lipton, 2000, p. 41; Perry et al., 2006, p. 6; Thorbecke and Ouyang, p. 2; 2022). Thorbecke and Ouyang note that the pro-growth poverty reduction strategy is rarely mentioned in the literature because strong empirical evidence in its support is lacking. They discuss the complementarity between pro-growth poverty reduction strategy and the conventional pro-poor strategy in the fight against poverty. Perry et al. (2006, p. 6) elaborate on this complementarity as follows: ‘…smart investments in the poor can lead to virtuous circles and that the issue of ‘pro-growth poverty reduction’ should perhaps be as important a policy concern as traditional concerns with ‘pro-poor growth.’ They conclude that investments that reduce poverty are good business for society as a whole, not just for the poor.

Since the main asset of the poor is their own labor, the bulk of ‘smart investments’ must be directed at improving the productivity of this asset. Broad-based labor productivity can be enhanced by providing human capital inputs that are affordable by the majority of the population. Moreover, investments are needed in social infrastructure to complement the human capital (health, nutrition, general education and technical training) in raising total factor productivity of the population, not just that of labor.

An important question about the virtuous spiral is how it can be initiated and sustained. The human capital literature in poverty analysis (see, e.g., Eastwood and Lipton, 2000) suggests that large scale poverty reduction can be triggered by targeted transfers to the poor and by establishment of inclusive social protection programs. In the same vein, the reduction can be sustained by public investments that promote entrepreneurial activities, thus creating opportunities for self-employment. In addition, such investments should spur wage employment because wage-work is a derivative of entrepreneurial activity. The private initiatives, which could be set off by government policies, can play a major role in sustaining robust reductions in poverty and inequality. Kamgnia and Ahouré (2020) show that inclusive business partnerships can increase productivity and profits of small firms and thus protect them from an abrupt fall into poverty. They further provide suggestive evidence that education, technical skills and business experience of enterprise owners are preconditions for success in such circumstances. It is worth noting that inclusive business partnerships and social protection programs, which provide conducive environment for the formation of human capital are variants of pro-growth poverty reduction strategies.

Ideally, the production of human capital is a cooperative effort between households and governments. Grossman (1972) provides a framework for understanding human capital formation by households using their own resources, including time, combined with publicly provided semi-public goods inputs, such as education, curative care, vaccinations and sanitation infrastructure. In the context of a disease pandemic, such as the COVID-19, this framework is useful in the design of policies that are needed to change individual and collective behaviors to avoid infection, to provide emergency care to those infected and, generally, to protect lives and livelihoods.

The remainder of the paper is organized as follows. Section 2 presents a model of a worker's economic well-being, driven by human capital formation, plus a brief overview of the data used to estimate the model. Section 3 provides descriptive statistics on the data and also presents the estimation results. Section 4 discusses the findings, while Section 5 concludes the paper highlighting its policy applications.

2. Wages, human capital and the COVID-19 context

2.1. Model

This section shows that household income, in form of wages, is linked to human capital status of a worker. Thus, policies that promote human capital formation or that protect it from disease pandemics are poverty-alleviating. Furthermore, from the perspective of the virtuous spiral (Thorbecke and Ouyang, 2022) such policies are inequality reducing.

We use the standard wage function (Mincer, 1974; Buchinsky, 1994), augmented with COVID-19–related variables, to estimate wage effects of education capital and health human capital components. The effects are estimated using re-centered influence regressions (Firpo et al., 2009) to facilitate measurement of human capital on observed wages at different quantiles.

Where,

Log_biwekly_wage = log of a fortnightly wage;

d = demographics (gender, age, marital status, place of residence);

e = education dummy (1 = attended school as a child);

H = health dummy (1 = consulted a doctor for treatment when ill, proxy for high quality care), an endogenous regressor;

i.county = county-specific dummies, with Mombasa being the comparison county;

c19 = COVID-19 vulnerability index during the period of pandemic;

gr = generalized residual constructed from the probit regression of H on a constant, d, e and c19 (all are exogenous covariates in Equation 1; see Wooldridge, 2015, pp. 427–8); and u₁ and u₂ are the usual disturbance terms, uncorrelated with exogenous explanatory variables in Equation 1.

2.2. Data

Equations (1) and (2) are estimated using panel data from five waves of high-frequency phone-based surveys conducted by the World Bank in collaboration with the Kenya National Bureau of Statistics and the University of California at Berkeley (see, Manda et al., 2022). The data description provided below follows closely the text in Manda et al.

Briefly, Equations 1 and 2 were estimated using panel data from a rapid response phone survey that collected panel data on members of households over roughly a 2-year period, 2020–21. The data set contains information from five waves of the COVID-19 rapid response survey, conducted at the time (2020–21), which was part of a bi-monthly panel survey that targeted Kenyan nationals. The durations of each of the five waves and the sample sizes involved are as follows.

• Wave 1:

  May 14 to July 7, 2020; 4,063 households;

• Wave 2:

  July 16 to September 18, 2020; N = 4,504;

• Wave 3:

  September 18 to November 28, 2020; N = 4,993;

• Wave 4:

  January 15 to March 25, 2021; N = 4,860;

• Wave 5:

  March 29 to June 13, 2021; N = 5,854.

The initial sampling frame comprised 92,999,970 randomly ordered phone numbers assigned to three networks: Safaricom, Airtel and Telkom. An introductory text message was sent to 5,000 randomly selected phone numbers to determine if the numbers were in operation. Out of these, 4,075 were found to be active and formed the final sampling frame. There was no stratification, and individuals that were called were asked to provide information about the households they lived in. The survey started in May 2020 and ended June 2021. The same households were interviewed every two months, with interviews conducted using Computer Assisted Telephone Interviewing (CATI) techniques. The data set contains information from two samples of households. The first sample is a randomly drawn subset of all households that were part of the 2015/16 Kenya Integrated Household Budget Survey (KIHBS) Computer-Assisted Personal Interviewing (CAPI) pilot study. The second was obtained through the Random Digit Dialling (RDD) method, by which active phone numbers created from the 2020 Numbering Frame produced by the Kenya Communications Authority were randomly selected.

The samples covered urban and rural areas and were designed to be representative of the Kenyan population that was using cell phones at the time, which of course was not a random sample of the total Kenyan population. All waves of this survey included information on household background, service access, employment, food security, income loss, transfers, health and COVID-19 knowledge. The data is in three files. The first is the household file, which contains household level information. The household file uniquely identifies all households. The second file is the adult level file, which contains data at the level of adult household members. Each adult in a household is uniquely identified. The third file is child level file, which contains information for every child in the household. Each child in a household is uniquely identified by the child-specific identification number. However, the analysis is based on a sample of adults of age 18 to 64 years. The data set on adults provides information on age, gender marital status, region and county of residence. This information is combined with the information provided at the household level on COVID-19 virus and COVID-19 diseases and symptoms. This sample provided unbalanced panel with a total of 55,621observations on individuals of whom only 6,119 to 6,654 reported having received wages over the five waves.

3. Results

Table 1 shows that 67% of the estimation sample is male, suggesting that the sample was not a randomly draw from the general population. If the COVID-19 pandemic affected the human capital of men and women differently, the estimated wage effects of the pandemic are biased by gender-specific factors, such as cultural norms. The same argument applies to the wage effect of urban residence because 61% of the sampled population resided in towns. The COVID-19 vulnerability index for the sampled population was .69, indicating that COVID-19 vulnerability in the sample was nearly 70% of its maximum possible value (100%). That is, the Kenyan population constituting the study sample was highly vulnerable to COVID-19 infections and/or deaths. The mean age of the sample was 36 years, which is consistent with the mean age of the workers covered by the minimum wage policy in Kenya (Muriithi et al., 2020).

Another important feature of the sample is that nearly all the sampled individuals had some level of education as only .63% of the sample had never enrolled in school. In contrast only a tiny fraction of the sample had access to high quality care in the event of sickness, as only 0.56% of the sampled had consulted a doctor. Similarly, labor mobility is limited, since only 0.92% of the sample had migrated from county of birth. Labor mobility is an important dimension of human capital and of the labor market participation because without mobility, workers cannot participate in labor markets that give the highest return to human capital embedded in them. That is, labor immobility is a constraint on poverty reduction and could be inequality-increasing.

Finally, given that the sample mean is 36 years, a marriage prevalence level of 12.6% in the sample is low, considering that the prevalence for the general population is over 70% for women. Marriage is also an important determinant of wages, as labor market participation of married women is constrained by childcare duties, and by labor supply norms, which usually do not apply to single women or men, irrespective of marital status (Heckman, 1974).

Table 2 shows the distribution of wage earners by log wage percentiles. Panel a indicates that the bottom 25% of the workers have very small logs of hourly wages per week, suggesting a high likelihood of extreme poverty among this category of workers. Panel b shows strongly positive median and mean wages, indicating that workers earning these wages are distinctly better off than workers in the lower percentiles. The skewness parameter for the wage distribution bears a negative sign suggesting that the bulk of wage earners are located close to the median wage, which is our relative poverty line. The magnitude of kurtosis (12.4) which is evidence of large outliers in log wages shows that the earnings distribution is not normal but in repeated sampling this issue might not arise.

The re-centered influence regression equations are estimated for the 25th, 50th, 75th and 90th percentiles of the wage distribution to investigate the role of human capital formation (education, health and labor mobility) in the level and distribution of wage incomes. The unconditional quantile regression results are also used to assess the extent to which policies that promote human capital formation can be used to reduce poverty and inequality in Kenya, particularly in the bottom 25% of the population, taking into account the COVID-19 context that prevailed in the country at the time. In addition to the estimation of the behavioral parameters of interest, poverty and distributional statistics are computed to assess the extent of income poverty, and the magnitude of income inequality in Kenya at the time of the COVID-19 period, the sample was collected. In the regressions, effects of COVID-19 context are controlled for via inclusion of COVID-19 vulnerability index in wage equations. The COVID-19 vulnerability index is a statistical aggregate of variables that are positively correlated with the risk of contracting a COVID-19 virus, such as not wearing a mask, not washing hands, and lack of trust in government's COVID-19 control measures.

Table 3 shows that the three estimators yield practically the same results but the unconditional quantile estimator has the advantage that it can be used to estimate parameters of the wage equation at each point of the wage distribution and is thus the preferred procedure (Table 4).

In Table 3, only individual level covariates are included in an implicit multilevel model of log wage regressions to illustrate the three equivalent ways of estimating the parameters of the model (the second level, county variables are captured by county-specific dummies, which are included in the fully specified model; see Table 4). A glance at Table 4 shows that inclusion of survey waves and of county-specific dummies as controls in the estimating equations substantially changes the magnitudes and occasionally the signs of the estimated coefficients.

The baseline results show that human capital variables (education, health and labor mobility) are the key determinants of wage incomes. For example, a unit increase in COVID-19 vulnerability index reduces the biweekly wage by approximately 1.3%. Workers without education have wage incomes that are 39% lower than the earnings of educated persons (the adjusted coefficient is exp(−0.3871) − 1 = 32.1%); see Harvorsen and Palmquist (1980). Persons who were ill prior to the survey and who sought treatment from a doctor (a proxy for high quality care) had wages that were 285% higher than that for workers treated by other health professionals elsewhere outside or inside a hospital setting. The finding suggests that high-quality care is the key to cure after getting ill, and that better health is productivity-enhancing. The large wage gains associated with doctor consultations are consistent with the fact that an illness that is not properly treated can wipe out a worker's income. Thus, effective health care upon illness is an important policy measure for sustaining poverty escapes within and across households. The coefficient on migration from county of birth is associated with 75% higher wages (Table 3), indicating that labor mobility is one of the mechanisms through which the human capital embedded in workers reduces poverty. The finding underscores the important roles of transportation infrastructure and telecommunication systems in enhancing productivity of human capital.

A glance at Table 4 shows that endogeneity is not a problem in all wage quantiles (the coefficients on generalized residuals for the median and 25th percentiles are not statistically significant).

4. Discussion of main results

4.1. Wage effects of COVID-19 vulnerability

The negative coefficients on the COVID-19 vulnerability index in Table 4 are at least two to three times larger than the coefficients reported in Table 3. As already indicated, COVID-19 vulnerability is the risk of workers contracting a COVID-19 virus or illness. If workers are averse to this risk they would be unlikely to travel to work or to step out of their homes in search of employment. The negative coefficient on the vulnerability index captures the income loss associated with workers' risk aversion to a COVID-19 environment. In some counties, people might be more averse to risks of COVID-19 infections than in other counties but this issue has not been investigated. Moreover, the effects of the index vary across the quantiles of the log wage function, with wage residuals at the 25th percentile showing practically no sensitivity to COVID-19 risk.

4.2. Education, health and location effects

The baseline regressions presented in Table 3 show that the effects of education and location on wages are large and statistically significant in all quantitiles, a finding consistent to that reported in previous work using South African data (Schultz and Mwabu, 1998). The coefficients on health care quality (proxied by doctor-consultation dummies) follow the same pattern. However, once the county-specific factors are controlled for (Table 4), the robustness of wage returns to education and health across quantiles weakens considerably. This fact should be considered in the design of policies aimed at increasing human capital investment as a tool to fight poverty and inequality. A general, across the board human capital investment, as in Perry et al. (2006), might not yield the desired effect. The coefficient on the interaction of COVID-19 vulnerability index with a visit to a doctor is negative at all quantiles, indicating that COVID-19 vulnerability erodes productivity gains associated with better health. The size of the average effect of health human capital on wages is driven by returns at upper quantiles. The large size of the health coefficient (5.817 [t = 3.04]) is not implausible because recovery from illness can increase earnings considerably.

The coefficient estimates on the interaction of COVID-19 with education dummy across quantiles are all positive. However, only the coefficient for the top quantile is statistically significant. This finding suggests that uneducated persons at the top wage quantile had a higher propensity of participating in the labor market and therefore earned higher wages relative to educated workers who were more sensitive to the risk of a COVID-19 infection and thus made efforts to avoid it by not working away from home. This finding suggests that uneducated workers might have sacrificed health for a wage, or were more willing than their educated counterparts, to take the risk of an infection in order to participate in the labor market. The wage effects of human capital (education, health and labor mobility) should be interpreted with caution because of the small sample sizes available for dummies that are turned on during estimations.

The coefficients on location in Table 3 show that wage returns to urban location are larger than those for rural areas. Table 4 does not report wage returns to location because these effects are captured by county-specific factors.

4.3. Effects of factors specific to survey periods

The coefficients on dummies for the various periods over which the surveys were conducted show that gains in log wages were higher in subsequent 4 waves relative to the first, but only for the mean regression. In the upper quantiles, the gains are generally higher in the first than in subsequent waves. This finding suggests that wages of workers at different quantiles are affected differently by the same set of period-specific factors, likely due to differences in unobserved personal characteristics or because of the nature of the work they do. This is also true of human capital effects on log wages. As can be seen from the coefficients on the education dummy (at 25th percentile), a person without education has a wage that is 17.03% lower than for an educated person in contrast to a gap of 44.7% at the median distribution. Starting at the median quantile upwards, education investments increase incomes, reduce poverty as well as inequality (because education returns fall with quantiles). In contrast, investments in better health reduce poverty but are inequality-increasing.

4.4. Welfare statistics

A look at the lower panel of Table 4 shows that the mean log wages increase with quantiles, as expected (see Table 2). Poverty prevalence at the time of the survey was quite high at 49% (using the median as the relative poverty line). Poverty inequality and poverty intensity (7.8% and 2.8%, respectively) were also equally large, indicating that in this period, on average, every citizen in the sample had a wage shortfall (the amount below the relative poverty line, in this case the median wage) that was around 8% of the median. Similarly, the per capita poverty severity index was around 3%, an indicator of the intensity of poverty felt by every citizen in the sample.

The inequality in income distribution during the survey period (May 2020 to June 2021) was quite high. The Gini for all wage earnings, including zero wages, was 93.3%. The Atkinson's Inequality Index (inequality aversion parameter, ε = 2) was 91. 2%. These exceedingly high inequality measures are not implausible because they account for the fact that many workers had zero wages, occasioned partly by the pandemic and partly by the control measures associated with it. These inequality measures change drastically when estimated using a sample of workers with positive wages. It is worth noting that out of a total sample of about 56,000 individuals, only 6,119 to 6,654 reported positive wages.

The Gini for wage inequality among the labor market participants (those with positive wages) was 39.1% and the corresponding Atkinson's inequality index (ε = 1.5) was 53%. The Atkinson's inequality measure is preferred because it takes into account extreme incomes at either end of the wage distribution. In the Atkinson's measure, such incomes are appropriately weighted by, ε (inequality aversion parameter) to reflect the social disutility associated with inequality.

5. Conclusion and policy implications

The paper has shown that human capital formation is the main driver of labor productivity, as proxied by wages in labor markets. It has also demonstrated that an increase in labor productivity reduces poverty but its effect on inequality is ambiguous. However, to the extent that productivity gains from human capital accrue more to better-off workers (at higher wage quantiles), human capital formation worsens the income distribution and can eventually increase poverty. Thus, the existence of a virtuous spiral critically depends on how poverty and inequality change as the economy grows. That is, a pro-growth poverty reduction strategy can generate growth that is incompatible with a virtuous spiral.

More importantly, a negative shock to an economy, such as the one exerted by the COVID-19 pandemic in 2020–21, can severely disrupt human capital formation—one of the mechanisms that sustains the spiral. Another spiral-augmenting channel that was shut-down by COVID-19 virus through lockdowns was labor mobility. Labor mobility is a form of human capital because human capital is part and parcel of human beings (Schultz, 1961). We have seen that labor mobility is associated with large wage gains at all quantiles so that its restriction means income loss. School lockdowns had similarly deleterious effects, some of which will be felt by the economy many years to come. Entrepreneurship (business-creating ability embedded in people) is another form of human capital that was affected by lockdowns. Business lockdowns had immediate and long-term adverse effects on labor income and its distribution. The high wage inequality measure (> 90%) on both the Gini and the Atkinson inequality scales can, to a large extent, be attributed to business closures.

The findings of the paper can be used to identify population segments to which human capital investments should be targeted to reduce both poverty and inequality and set the stage for a virtuous spiral of falling poverty and robust growth, plus perhaps, an optimal distribution of income (Tinbergen, 1973).

Using Kenyan data the paper has shown that COVID-19 reduced labor income and slowed down human capital formation. However, the pandemic's erosion of health stocks is evident only indirectly from the analysis of the wage effects of the interaction between ‘COVID-19 vulnerability’ with the quality care provided by doctors. This interaction reduced log wages, suggesting that part of the health capital gained from treatment by doctors was lost to the pandemic or the opportunity to put the capital into productive use was impossible during the COVID-19 times.

An important contribution of the paper is to show that the inequality in wage income was very high (Gini > 0.90) during the COVID-19 times so that the cash transfers targeted to the poor and the vulnerable persons during the pandemic were justified. However, the tax reliefs for income tax-payers were not properly targeted, as in all likelihood they increased the income inequality because they benefited relatively more the workers at higher wage quantiles. In similar circumstances in the future, tax-payers at higher income quantiles should not be exempted. Instead, the tax revenues collected from such tax payers should be used to increase cash transfers to the poor. The statistical approach used in the paper is general enough to be applied to assess effectiveness of fiscal measures and other public policies in reducing poverty and inequality at particular points of the income distributions. The approach is useful in pinpointing segments of the population to which anti-poverty and re-distributive measures can be targeted to achieve the desired social welfare goals.

Acknowledgement

I am very grateful to the African Economic Research Consortium (AERC) for funding this study.

Funding

This study was funded by the African Economic Research Consortium (AERC), Grant No. RP 21503.

Data availability

The data set used in the paper is available from the author on request. However, the Kenya National Bureau of Statistics (KNBS) must authorize the release of the data to a third party.

References