Poverty and Its Correlates among Kenyan Refugees during the COVID-19 Pandemic: A Random Effects Probit Regression Model

Abstract

Poverty remains a major problem among refugees, and the COVID-19 pandemic seems to have exacerbated its incidences. In Kenya, although refugees ordinarily face serious economic conditions, COVID-19 worsened their economic status. The objective of this paper was to analyze the determinants of poverty dynamics among Kenyan refugees during the COVID-19 pandemic. The data were the COVID-19 rapid response panel data that were collected between May 2020 and June 2021 by the Kenyan National Bureau of Statistics (KNBS), the United Nations High Commissioner for Refugees (UNHCR) and the University of California, Berkeley with technical assistance from the World Bank. The random effects probit regression model was used for data analysis using the absolute and relative poverty lines. The results showed that, using the Kenya’s national poverty lines, 73.03% of the respondents were poor across time, while there was a steady decline in poverty incidences from 76.55 in July–September 2020 to 68.44% in March–June 2021. The results further showed the presence of significant heterogeneity, thereby justifying the panel estimation approach. Poverty significantly declined (p < 0.05) with receipt of food assistance, remittances, gifts, amount of loan, amount realized from sale of assets and agricultural enterprises, while it increased with education, household size, non-farm enterprises, residence in urban areas, and at the Kakuma, Kalobeyei and Shona camps. It was concluded that welfare deprivation among refugees during COVID-19 is pathetic, and post-COVID-19 recovery should, among other things, take cognizance of place and camp of residence, and access to some form of socioeconomic support.

1. Introduction

COVID-19 presents a double tragedy to refugees across the world given the implications of development policies that were initially channeled at redefining the speed of achieving equilibrium between the ubiquitous trios of social, economic, and environmental sustainability [1]. Globally, the pandemic is of significant socioeconomic and public-health concern [2]. The refugee population has received some international attention given the diversity of migration restrictions that were implemented by different countries at the onset of the pandemic [2]. It should be emphasized that, beyond the socioeconomic inequity that was promoted by COVID-19, structural policy changes that were promulgated by some governments were prescriptively against the rights of many refugees and asylum seekers. Therefore, the economic impacts of the pandemic on the refugee population have been significantly amplified by pre-pandemic exposure to weak social-support systems, health inequity, poor housing, engagement in indecent jobs, and exposure to economic exploitation and social injustice [3]. The welfare losses among refugees due to economic lockdowns and the sudden meandering of health and social policies that deprioritize the human rights of refugees went a long way in redefining the speed of economic recovery among these extremely vulnerable groups [4,5].

The desirability of poverty reduction as one of the major development goals of the twenty-first century has been emphasized in the Sustainable Development Goals (SDGs) [6,7]. Projections by the World Bank indicated that, in 2020, between 119 and 124 million people would fall below the poverty line due to various economic impacts of the pandemic [8]. Worried about persistent poverty in Africa, the international communities have identified a reduction in exposure to risks and social conflicts as one of the major pathways for addressing socioeconomic deprivations and persistent vulnerability among the poorest segments of the population [9]. This is fundamentally relevant because the displacement of a human population has always constituted significant barrier to the achievement of steady economic growth and poverty reduction. At the end of 2020, some statistics revealed that the East and Horn of Africa, and Great Lakes regions, play host to about 4.5 million refugees, with Uganda, Sudan and Ethiopia hosting about two thirds of them [2]. The past two years have witnessed a significant increase in the number of people requiring humanitarian assistance [10,11], while COVID-19 destabilizes global humanitarian efforts to conserve the socioeconomic progress that had been achieved from the Millennium Development Goals (MDGs) [12,13].

A peep into the spectrum of welfare deprivation among refugees reveals that, compared to the national poverty incidence of 37% in 2015/2016, pre-COVID-19 poverty evaluation among some refugees in Kenya revealed that, using USD 1.9 as the poverty line, 65% and 68% of the groups in Kakuma and Kalobeyei camps were, respectively, below the poverty line in 2018/2019 [14]. It was also found that unemployment was a major problem among the refugees due to some stringent policy actions that limited the level of their labour-market participation and professional engagements, while female-headed households were more likely to be poor [14]. Similar findings had been reported among asylum seekers and refugees in the United Kingdom (UK), where it was reported that, due to intentional limitations that refugees faced through some policies that were meant to deepen their socioeconomic destitution and force their return to original homelands, the magnitude of poverty among refugees is sometimes beyond description [15]. Using some out-of-sample predicted poverty rates, Dang and Verme [16] estimated poverty rates of between 52.3% and 53.0% for Syrian refugees in Jordan using a 2014 dataset and the predicted poverty rates were statistically related to the actual rates.

A theoretical review of poverty analyses shows the integration of unidimensional and multidimensional approaches. Although multidimensional poverty evaluates welfare from different non-monetary indicators, this study will use the unidimensional poverty measurement with monthly expenditures taken as the primary indicator of welfare. Poverty analyses have evaluated the role of several demographic factors among the general population and refugees [17,18,19,20]. Specifically, Santorius et al. [21], using a 2001–2007 dataset, analyzed the determinants of asset index among Mozambican refugees who were residing in rural South Africa. It was found that mortality rates, low level of education, poor access to healthcare facilities, and engagement with gainful employment opportunities were the major determinants.

Among some refugees in Jordan, Khawaja [22] used a 1999 households’ dataset and found that 60% of the respondents did not have enough money to meet their expenses. Hejoj [23] also found that 41.8% of refugees in Jordan were living below the poverty line. Using the basic needs approach, 27% of the refugees in Lebanon were reported to be poor by Chaaban et al. [18]. Among Syrian refugees in Jordan and Lebanon, poverty incidences were 90% and 70%, respectively, using the income approach [24]. Alloush et al. [25] found that income poverty among Congolese refugee camps in Rwanda was between 73% and 76%, and poverty was associated with the number of years of migrating to host country, place of residence, the number of persons in the household, age of household heads, the level of education of household heads, and employment status. More importantly, among refugees, residence in rural and urban areas had been associated with lower poverty rates [25,26]. In addition, poverty declined with the years of migration [22], attainment of higher levels of education [22,23] and households’ heads employment status [23]. In addition, probability of being poor was higher among refugees that were aged 60 and older [23], of retirement ages [22], residing in refugee camps or settlements [25], and had six or more children [24].

Poverty is one of the major development indicators and a perfectly effective lens through which the impact of economic shocks such as COVID-19 must be evaluated. One of the major concerns for policy is understanding the levels of exhibited vulnerability and poverty during the COVID-19 pandemic. Therefore, the policy twists and development implications of COVID-19 demand significant, adequate evaluation of the poverty impacts, which are one of the notable Sustainable Development Goals (SDGs). Although there are general consensuses among policy makers on the retrogressive implications of COVID-19 on many development indicators, the case among vulnerable segments of the population, such as refugees, can be significantly pathetic. This study seeks to fill a major gap in the literature, by analyzing the correlates of time-variant poverty among refugees in Kenya during the COVID-19 pandemic.

2. Materials and Methods

2.1. The Data

This study used the datasets obtained from waves one to five of a high-frequency phone survey that was conducted between 14 May 2020 and 29 March 2021 during the COVID-19 pandemic. The surveys were carried out simultaneously among the regular Kenyan population and the refugee camps. The surveys were implemented by the Kenyan National Bureau of Statistics (KNBS), with support from the World Bank, the United Nations High Commissioner for Refugees (UNHCR) and the University of California, Berkeley. The questionnaire, among other things, probed into households’ demographic characteristics, income generating activities, consumption expenditures, behaviour change, impacts of COVID-19, income sources, food security and coping strategies [27].

The sampling frame for data collection comprised of those refugees that were registered with the UNHCR and residing at Kakuma camp, Kalobeyei camp, Dadaab camp, Shona camp and urban areas. The sampling procedures differ across strata with the sampling frame of the recently conducted Socioeconomic Surveys (SES) used for refugees in Kakuma and Kalobeyei, while the UNHCR’s refugees’ register formed the sampling frame for refugees in urban areas, and Dadaab and Shona camps.

The respondents from refugee camps in Kakuma, Kalobeyei, Dadaab and urban areas were selected based on a two-step sampling procedure. The first step involved selection of 1,000 individuals from each of the strata, with each of them sent a text message to verify if the registered phone numbers were still active on mobile networks. The second stage involved stratification of confirmed respondents by gender and age. However, due to small population size, all the households in Shona socioeconomic survey (n = 400) were included as part of the sampling frame. At each of the waves, sampling waves were generated for each household [28]. Approval to have access to the dataset was granted on the 30 August 2021 after application was submitted on the 26 August 2021. The dataset was then downloaded from the UNHCR microdata catalogue website [29].

The households were interviewed using the computer-assisted telephone interview (CATI). The selected households were loaded on the system with adequate information on place of residence and phone numbers. The respondents were adult household members who were 18 years or above and they were randomly sampled to ensure gender balance [30]. During the first wave (14 May–7 July 2020), 1328 households were sampled. During the second (16 July–18 September, 2020) and third (28 September–2 December 2020) waves, 1699 and 1487 respondents were, respectively, sampled. The fourth (15 January–25 March 2021) and fifth (29 March–13 June 2021) waves, respectively, had 1376 and 1562 households [29].

2.2. Estimated Models

Following Gibbons and Bock [31], the random-effects probit regression model was used for data analysis, considering the longitudinal nature of the dataset and the binary nature of the dependent variable. Panel data analysis requires critical consideration of the time-variant impacts. The proposed model estimates the impact of the time-variant trend parameter or heterogeneity as repeatedly measured within the households. The estimated model can be specified as:

$$Y_{it}={\displaystyle \sum }_{k=1}^n{\beta }_k{X}_{it}+v_t+{ϵ}_{it }$$

(1)

The dependent variable ($Y_{it})$ represents the poverty status of the household. This variable was coded 1 for those who were poor and 0 otherwise. This analysis used two poverty lines. The first is the national poverty line for Kenya, which were KES 3252 for rural households and KES 5995 for urban households [32]. The second poverty line was generated with the Foster–Greer–Thorbecke (FGT) approach [33], which is two-thirds of mean per capita expenditure (KES 1636). Individual respondent is denoted by i, while the panel time is denoted by t. In addition, ${\beta }_k$ are the parameters of included exogenous variables ($X_{it})$ [34]; $v_t$ is the random effects parameter that captures existence of heterogeneity across time and ${ϵ}_{it}$ is the stochastic disturbance term.

The explanatory variables were household head age (years), urban resident (yes = 1, 0 otherwise), household size, household head’s gender (Male = 1, 0 otherwise), improved floor materials (yes = 1, 0 otherwise), power-grid connection (yes = 1, 0 otherwise), number of telephones, agricultural activities (yes = 1, 0 otherwise), pastoral activities (yes = 1, 0 otherwise), non-farm enterprises (yes = 1, 0 otherwise), amount sold assets (KES), received food (yes = 1, 0 otherwise), amount of loans (KES), Kakuma camp resident (yes = 1, 0 otherwise), Kalobeyei camp resident (yes = 1, 0 otherwise), Shona resident and other (yes = 1, 0 otherwise), remittance recipient (yes = 1, 0 otherwise), gift recipient (yes = 1, 0 otherwise), government help recipient (yes = 1, 0 otherwise), NGO’s help recipient (yes = 1, 0 otherwise), politician’s help recipient (yes = 1, 0 otherwise), primary education (yes = 1, 0 otherwise), secondary education (yes = 1, 0 otherwise), tertiary education (yes = 1, 0 otherwise), and Madrassa/Duksi trainings (yes = 1, 0 otherwise).

3. Results

3.1. Descriptive Characteristics of Refugees’ Demographic Characteristics

Table 1. Descriptive statistics of refugees’ demographic characteristics.

Variable	Mean	Std. Dev.	Min	Max
Household-head age	36.09036	11.92076	18	96
Urban resident (yes = 1, 0 otherwise)	0.2402851	0.4272853	0	1
Household size	5.067366	3.241683	1	15
Household-head’s Gender (Male = 1, 0 otherwise)	0.5525077	0.4972687	0	1
Improved floor materials (yes = 1, 0 otherwise)	0.4263816	0.4945839	0	1
Power-grid connection (yes = 1, 0 otherwise)	0.4103805	0.4919359	0	1
Number of telephones	1.497647	0.845932	1	10
Agric activity (yes = 1, 0 otherwise)	0.0103536	0.1012315	0	1
Pastoral activity (yes = 1, 0 otherwise)	0.0075299	0.0864536	0	1
Non-farm enterprises	0.0746269	0.2881892	0	1
Amount of sold assets	162.2722	1096.231	0	35,000
Received food (yes = 1, 0 otherwise)	0.072341	0.2590691	0	1
Amount of loans	443.9207	1999.125	0	60,000
Kakuma camp (yes = 1, 0 otherwise)	0.2331585	0.4228708	0	1
Kalobeyei (yes = 1, 0 otherwise)	0.1293532	0.3356131	0	1
Shona and other (yes = 1, 0 otherwise)	0.4786876	0.4995792	0	1
Remittance recipient (yes = 1, 0 otherwise)	0.0681727	0.2520589	0	1
Gift recipient (yes = 1, 0 otherwise)	0.085115	0.2790715	0	1
Govt’s help recipient (yes = 1, 0 otherwise)	0.0517682	0.2215736	0	1
NGO’s help recipient (yes = 1, 0 otherwise)	0.1996773	0.3997846	0	1
Politician’s help recipient (yes = 1, 0 otherwise)	0.0083367	0.0909302	0	1
Primary education (yes = 1, 0 otherwise)	0.0514993	0.2210286	0	1
Secondary education (yes = 1, 0 otherwise)	0.043566	0.204141	0	1
Tertiary education (yes = 1, 0 otherwise)	0.0170768	0.1295663	0	1
Madrassa/Duksi trainings (yes = 1, 0 otherwise)	0.0116983	0.1075313	0	1

contains the descriptive statistics of selected refugees’ demographic variables. It reveals that the average age of the respondents was about 36 years and about 24% were residing in urban areas. The average household size was about five members, and 55.25% were headed by males. Only about 1% and 0.75% were engaged in agricultural and pastoral activities, respectively. Non-farm enterprises were reported by 7.46 percent of the respondents. The average incomes that were realized from the sale of assets and loans were KES 162.27 and KES 443.92, respectively. Remittances were received by 6.82% of the respondents, while 8.51% received some gifts. In addition, about 5.18% of the refugees obtained assistance from government, while 19.96% were assisted by NGOs.

3.2. Poverty Levels across Time and Refugees’ Demographic Characteristics

Figure 1 shows that, using the national poverty lines, 73.04% of the refugees were poor across the different data waves, while 52.52% were poor using the FGT poverty line. The proportions of the refugees that were poor steadily declined from 76.55% during wave 2 (July–September 2020) to 68.44% during wave 5 (March–June 2021) using the national poverty lines. A similar trend was also observed for the FGT poverty line. The results in

Table 2. Distribution of respondents’ poverty status across selected demographic variables.

	National Poverty Line				FGT Poverty Line
Variables	Non-Poor		Poor		Non-Poor		Poor		Total
Sector ***	Freq.	%	Freq.	%	Freq.	%	Freq.	%	Freq
Rural	1469	26.00	4181	74.00	2343	41.47	3307	58.53	5650
Urban	536	29.99	1251	70.01	1188	66.48	599	33.52	1787
Household Size ***
1	845	60.66	548	39.34	1093	78.46	300	21.54	1393
2	287	46.07	336	53.93	458	73.52	165	26.48	623
3	208	31.56	451	68.44	399	60.55	260	39.45	659
4	219	26.39	611	73.61	440	53.01	390	46.99	830
5	140	17.50	660	82.50	346	43.25	454	56.75	800
6	127	15.17	710	84.83	285	34.05	552	65.95	837
7	69	11.79	516	88.21	181	30.94	404	69.06	585
8	46	8.63	487	91.37	130	24.39	403	75.61	533
9	24	5.83	388	94.17	78	18.93	334	81.07	412
10	13	4.28	291	95.72	52	17.11	252	82.89	304
11	9	4.74	181	95.26	28	14.74	162	85.26	190
12	9	8.04	103	91.96	16	14.29	96	85.71	112
13	6	6.59	85	93.41	18	19.78	73	80.22	91
14	0	0.00	16	100.00	1	6.25	15	93.75	16
15	3	5.77	49	94.23	6	11.54	46	88.46	52
Gender **
Female	888	26.68	2440	73.32	1519	45.64	1809	54.36	3328
Male	1117	27.18	2992	72.82	2012	48.97	2097	51.03	4109
Refugee Camp ***
Dadaab	395	33.45	786	66.55	602	50.97	579	49.03	1181
Kakuma	293	16.90	1441	83.10	514	29.64	1220	70.36	1734
Kalobeyei	101	10.50	861	89.50	221	22.97	741	77.03	962
Shona	1216	34.16	2344	65.84	2194	61.63	1366	38.37	3560
Education Level ***
No education	1780	27.32	4736	72.68	3160	48.50	3356	51.50	6516
Primary	103	26.89	280	73.11	164	42.82	219	57.18	383
Secondary	73	22.53	251	77.47	120	37.04	204	62.96	324
Tertiary	27	21.26	100	78.74	51	40.16	76	59.84	127
Madrassa/Duksi	22	25.29	65	74.71	36	41.38	51	58.62	87
Total	2005	26.96	5432	73.04	3531	47.48	3906	52.52	7437

***—chi-square statistics for both poverty measures show statistical significance (p < 0.01). **—chi-square statistics for FGT poverty measure show statistical significance (p < 0.01).

shows the poverty levels among refugees across selected demographic characteristics. Focusing on poverty estimates that are based on the national poverty line, the results revealed that poverty was slightly higher among rural dwellers (74.00%) than their urban counterparts (70.01%). Based on the two poverty measures, there is a significant difference in poverty incidences across urban and rural areas (p < 0.01). The incidences of poverty also increased from 39.34% for refugees with one member to 95.72% for those with 10 members. Based on the two poverty measures, there is also a significant difference in poverty incidences across household sizes (p < 0.01). Based on gender, poverty was slightly higher among female-headed households (73.32%) than the 72.82% for male-headed households. There is no significant difference in poverty incidences across gender (p > 0.10) based on the national poverty line. However, with the FGT poverty measure, a significant difference was found (p < 0.01). Across the refugee camps, the respondents from Kalobeyei and Kakuma had the highest poverty incidences, of 89.50% and 83.10%, respectively. However, poverty incidences were highest among those with tertiary (78.34%) and secondary (77.43%) education. Based on the two poverty measures, there was a significant difference in poverty incidences across the different educational levels (p < 0.01).

3.3. Determinants of Poverty

The results in

Table 3. The results of random effects probit modelling of the determinants of poverty.

	National Poverty Lines—Absolute Poverty			FGT Poverty Line—Relative Poverty
Variables	Coeff.	Std. Error	Z-Stat	Coeff.	Std. Erro	Z-Stat
Household-head age	−0005962	0.0016641	−0.36	−0.0008124	0.0015416	−0.53
Urban resident	0.5304689 ***	0.0575498	9.22	−0.0621043	0.0559765	−1.11
Household size	0.2539462 ***	0.0091534	27.74	0.221153 ***	0.0073963	29.90
Household-head’s Gender	0.097118	0.0404006	0.49	−0.0584858	0.037108	−1.58
Improved floor materials	−2904145 ***	0.0541267	−5.37	−0.2066656 ***	0.0498143	−4.15
Power-grid connection	−0.5864182 ***	0.0493093	−11.89	−0.5533266 ***	0.0462329	−11.97
Number of telephones	−0.0902328 ***	0.0258999	−3.48	−0.0982884 ***	0.0226127	−4.35
Agricultural activities	−0.3174728	0.1836556	−1.73	−0.6509349 ***	0.1825093	−3.57
Pastoral activities	−0.2578319	0.2070714	−1.25	−0.1562656	0.2020669	−0.77
Non-farm enterprises	0.2038178 **	0.0694801	2.93	0.2104616 ***	0.0606683	3.47
Amount of sold assets	−0.0000846***	0.0000177	−4.78	−0.0001195 ***	0.0000222	−5.38
Received food	−0.7197434 ***	0.0706498	−10.19	−0.7337994 ***	0.0771994	−9.51
Amount of loans	−0.0000643 ***	9.80 × 10−6	−6.56	−0.0000936 ***	0.0000124	−7.58
Kakuma camp	0.321264 ***	0.065773	4.88	0.3445892 ***	0.0592771	5.81
Kalobeyei	0.8357634 ***	0.0857262	9.75	0.7322573 ***	0.0723144	10.13
Shona and other	0.1555248 **	0.0654397	2.38	0.1258085 **	0.0612393	2.05
Remittance recipient	−0.4317669 ***	0.0780547	−5.53	−0.3401173***	0.0815461	−4.17
Gift recipient	−0.3082336 ***	0.0707499	−4.36	−0.4623711 ***	0.0755273	−6.12
Govt’s help recipient	−0.0453486	0.0856754	−0.53	0.0140349	0.0823195	0.17
NGO’s help recipient	0.0508706	0.0486549	1.05	0.0200612	0.0449748	0.45
Politician’s help recipient	0.1223808	0.2064259	0.59	−0.1444377	0.1979488	−0.73
Primary	−0.0352822	0.0860857	-0.41	0.1836341 ***	0.0802406	2.29
Secondary	0.1232455	0.0967543	1.27	0.3807447 ***	0.088656	4.29
Tertiary	0.2053135	0.1529911	1.34	0.3643535 ***	0.1338196	2.72
Madrassa/Duksi	0.0986097	0.1789524	0.55	0.2288418	0.1646444	1.39
Constant	−0.1293588	0.0881372	−1.47	−0.5856023 ***	0.083026	−7.05
/lnsig2u	−2.422676	0.3341362		−2.520044	0.3203641
sigma_u	0.2977986	0.0497526		0.2836478	0.0454353
Rho	0.0814598	0.0250014		0.0744649	0.0220795
Log likelihood	−3138.8213			−3702.1233
Wald chi2(25)	1241.59 ***			1608.80 ***
LR test of rho = 0:	11.70 ***			12.52 ***
No of observations	7437			7437
VIF	1.42			1.42

Note: ***—Statistically significant at 1% level; **—Statistically significant at 5% level.

show the determinants of being poor based on the two poverty measures. The two models produced good fits for the data based on Wald chi-square statistics that are statistically significant (p < 0.01). The very low value of the variance inflation factor (VIF) shows that there was no serious multicollinearity among the independent variables. In addition, the likelihood ratio test of no heterogeneity (rho = 0) reveals that there is heterogeneity. This implies the existence of random effects in the models. Therefore, estimating the models with ordinary probit regression would produce inconsistent parameters.

There is high level of consistency in the signs of some of the parameters for the results on the determinants of poverty using the absolute and relative poverty lines. However, some differences were observed on the levels of statistical significance among the variables between the two models. The results showed that the age of the respondents and gender are statistically insignificant (p > 0.10). The parameter of urban residence in the absolute poverty model is positive and statistically significant (p < 0.01). This shows that refugees that were resident in urban areas had a significantly higher probability of being poor, when compared with their rural counterparts. Household-size parameters are statistically significant (p < 0.01) in the two models. The results imply that an increase in a households’ sizes will increase the probability of being poor.

The results further show that the parameters of improved floor materials are statistically significant (p < 0.01) with a negative sign. These imply that, compared with those with unimproved floor materials, households with improved floor materials had a significantly lower probability of being poor. In addition, the parameter of power-grid connection is statistically significant (p < 0.01) and with negative sign in the two models. The results show that, compared with those without connections to the power grid, households with access to electricity had a significantly lower probability of being poor. The parameters of the number of telephones are with a negative sign and statistically significant (p < 0.01). These results imply that the higher the number of telephones that a household had, the lower the probability of being poor.

The parameter of engagement in agricultural activities is statistically significant (p < 0.01) and with a negative sign in the relative poverty model. This implies that the refugees that were engaged in agricultural activities had a lower probability of being poor. However, the parameter of pastoral activities is not statistically significant in the two models (p > 0.10). The parameter of non-farm activities has a positive sign and is statistically significant (p < 0.01). These revealed that the refugees who were engaged in non-farm activities had a higher probability of being poor. The parameters of the amount of money realized from the sale of assets and loans are both statistically significant (p < 0.01) with a negative sign. These results show that, as the amount realized from the sale of assets and loans increased, the probability of being poor decreased.

The parameters of receipt of remittances and gifts are with a negative sign and statistically significant (p < 0.01). These show that the refugees that received some food assistance and gifts had a significantly lower probability of being poor. However, in the models, the parameters of receipt of assistance from the government, NGOs, and politicians do not show statistical significance (p > 0.10). Furthermore, out of the education dummy variables, none are statistically significant in the absolute poverty model. In the relative poverty model, the parameters of primary, secondary and tertiary education have a positive sign and are statistically significant (p < 0.05). These results show that those refugees with primary, secondary and tertiary education had a higher probability of being poor. In addition, the parameters of residence in the Kakuma camp, Kalobeyei camp and Shona camp all have a positive sign and are statistically significant (p < 0.01). These results show that, compared with those in Dabaab, those refugees that resided in Kakuma, Kalobeyei and Shona camps all had a significantly higher probability of being poor.

4. Discussion

The refugee population in Kenya comprises of relatively young households’ heads, with an average age of 36 years. In addition, the average household size among refugees is relative high (five) when compared with the general Kenyan population of 3.9 [35,36]. The gender distribution of households’ heads favours males, with 55% as compared to 49% for the general population in 2014 [35]. Urban residents constitute 24%, while the national figure was put at 34% in 2014 [35]. Access to electricity was 41%, which is lower than the 75% of Kenya’s general population in 2021 [37].

The high variations in estimated poverty incidences between the two poverty lines results from the differences in the poverty lines. Based on the national poverty line, the estimated poverty incidences among refugees were a bit high during the COVID-19 pandemic, with an average of seven out of ten households being poor. The results reveal the impacts of inherent vulnerability to shocks in the form of economic lockdowns and other business restrictions during the pandemic. It had been noted that, among the general Kenyan population, about 1.7 million jobs were lost during the first four months of the pandemic and an additional 2 million households slipped below the poverty lines [38]. In addition, the contraction of the Kenyan economy during the pandemic resulted in a general decline in economic growth rates and the first quarter of 2020 witnessed growth of GDP of −4.9% and this further declined to −5.7% during the second quarter [39]. The high poverty rate among refugees during COVID-19 is also a reflection of perpetual pre-COVID-19 deprivations. Specifically, with the national poverty in Kenya projected to have increased by 4% [40], the rate of poverty among the refugees is somehow within expected range given that, in 2018/2019, 65% of Kakuma residents and 68% of Kalobeyei residents were poor [14]. In some other pre-COVID-19 studies, poverty levels among refugees were 60% in Jordan [23], 70% in Lebanon [24], and between 73 and 76% among the Congolese refugees in Rwanda [25].

Refugees in urban areas had a significantly higher probability of being poor. This is a general reflection of the significant impact that the pandemic had on job losses in urban formal and informal employment opportunities. The persistent informality of the majority of the income-generating activities in rural areas underscores the lesser impact of lockdowns and other business-conduct restrictions that were put in place to control the spread of COVID-19. This finding is in alignment with speculation that indicated that the pandemic had the highest impact on poor households in urban areas [41,42,43]. Specifically, engagement in agriculture—which was the least affected sector in most countries during the pandemic—resulted in lesser impacts among those residing in rural areas [42]. The finding agrees with that of Betho et al. [43], but is contrary to the findings of Barletta et al. [44] in Mozambique, where rural residents exhibited a higher poverty rate (50.1%) than urban households (37.4%).

As expected, household size increased the probability of being poor among refugees. The contributions of a large family size to welfare deprivation have been highlighted in the literature [45]. The poverty transmission routes among large family sizes are multidirectional. Specifically, large families need more income to meet their basic needs of food, housing, education, health and other social services. Due to a high dependency rate, large families cannot always cope with financial needs and stay above the poverty line. It has been emphasized that having three or more children can significantly increase the risk of poverty [46,47,48]. It has also been noted that, in comparison with having one or two children, having three or more children increases the likelihood of experiencing hardship by 1.6 times [49]. In addition, the risk of persistent poverty in association with family sizes was examined by Barnes et al. [50]. It was found that poverty risk increases by 12%, 18% and 22% for having one, three and four or more children, respectively.

The results also emphasized the differences in poverty incidences across the refugee camps with Kakuma and Kalobeyei being worst affected. Although the pandemic instilled new dimensional impacts on the livelihoods of many Kenyan households, its welfare impacts on refugees can be location-specific due to several reasons. It should be noted that, while the Kakuma refugee camp accommodates about 160,000 refugees, this camp and Kalobeyei camp are located in Turkana county, which happens to be among the poorest counties in Kenya [14]. The poverty scenario in these two camps is not too surprising because of their interrelatedness given that Kalobeyei camp was created from Kakuma in 2015 due to overpopulation [51]. More importantly, given the nature of endemic poverty in Turkana West, significant economic interactions would be needed to enhance the livelihoods of residents in the county [52].

Although education is often seen as the cure for poverty, the results in this study are saying contrary. Specifically, educational attainment may not be so rewarding among refugees in Kenya due to some employment restriction laws. However, the situations that prevailed during the pandemic resulted in the loss of several jobs, and some refugees could have been affected. The finding in this study is contrary to that of UNHCR and the World Bank [53], which reported a decline in poverty due to educational attainment among refugees in the Kalobeyei camp.

Social assistance is a critical factor in promoting welfare among refugee populations. However, many refugees do not have consistent access to social protection due to legislative deficiencies within host communities [54]. The findings reveal that receipt of assistance from the government, NGOs and politicians did not significantly influence poverty among the refugees during the pandemic. This may be the case if the gifts that were received are not sufficient to significantly impact consumption expenditures. However, the refugees that received remittance, food and other gifts had a significantly lower probability of being poor. It has been noted that, due to COVID-19, refugees faced restrictions in their movements, and they witnessed significant declines in remittances received from abroad [55]. The wider scope of the pandemic and high number of affected people may also imply diversion of social assistance without showing some cognitive preferences for refugees.

The study found that households with access to improved floor materials, electricity from the grid and telephone have a lower probability of being poor. These findings show that housing attributes are sometimes reflections of inherent consumption poverty. In some empirical studies, the associations between income poverty and multidimensional poverty have been established [56,57]. As money is needed to secure better social services, which constitute welfare indicators in multidimensional settings, income deprivation is often a reflection of multidimensional poverty.

It was also found that income realized from loans and sale of assets reduced the probability of being poor. This is in accordance with expectations, and reflects the role of credit and income from the distress sale of assets during the pandemic. The findings also reflect the dimension of survival strategies that were utilized by some households and their effectiveness during the pandemic. Specifically, asset accumulation is a prerequisite for improved livelihood and exit from the poverty trap [58]. Therefore, the ability to have assets that can be disposed of for cash during a time of crises reflects sound economic stability and survival capability [59]. Similarly, possession of viable assets may underscore less susceptibility to structural and stochastic poverty [60].

5. Conclusions

Although COVID-19 affected the larger populations in Kenya, understanding the impact on poverty among refugees is of significant policy relevance. This study assessed poverty incidences using monthly expenditure data and evaluated its sensitivity to the poverty line by using the national absolute poverty lines as defined for rural and urban Kenya and relative poverty that was derived using the FGT approach. It was found that the dimensions of poverty levels are highly sensitive to poverty lines, although similar parametric implications were observed across the estimated models. The findings reflected the extreme vulnerability of urban refugees to the welfare shocks such as the healthcare emergencies that were constituted by COVID-19. This implies that marginal reforms channeled at refugees in urban areas will go a long way in addressing welfare losses during the pandemic. Specifically, residents in Kakuma and Kalobeyei need some special welfare-enhancement interventions. The findings also underscore the role of adequate social protection for refugees in critical times of crises. More importantly, facilitation of remittance transmission channels for refugees and commitment to food aid and other assistance are of paramount importance for poverty reduction. In addition, refugees should be educated on the need for small household sizes as a way of poverty reduction. This implies the need for proper education on the use of contraceptives as a viable way of reducing maternal fertility.