Policy Research Working Paper 9960 Impact of COVID-19 on Labor Market Outcomes of Refugees and Nationals in Kenya Mirko Vintar Theresa Beltramo Antonia Delius Dennis Egger Utz Johann Pape Poverty and Equity Global Practice March 2022 Policy Research Working Paper 9960 Abstract This paper investigates the labor market outcomes for refu- 40 percent of nationals. Using a panel setup with wave and gee and urban national communities in Kenya during the location fixed effects, the analysis finds that the recovery COVID-19 pandemic, using five waves of a novel high-fre- in the share of employed, hours worked, and household quency phone survey collected between May 2020 and June incomes was slower and often stagnant for refugees com- 2021. Even after conditioning on age, gender, educational pared with the recovery of nationals. These differences attainment, and area of living, only 32 percent of refugees cannot be explained by demographic factors, living in were employed in February 2020 compared with 63 percent an urban or camp environment, having been employed of nationals. With the onset of the pandemic in March previously, or sectoral choice, suggesting that a third, unob- 2020, the share of employed for both refugees and nationals servable “refugee factor” inhibits refugees’ recovery after a fell by around 36 percent, such that in May-June 2020, only major shock and aggravates preexisting vulnerabilities. 21 percent of refugees were still employed compared with This paper is a product of the Poverty and Equity Global Practice. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The authors may be contacted at upape@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team IMPACT OF COVID-19 ON LABOR MARKET OUTCOMES OF REFUGEES AND NATIONALS IN KENYA1 Mirko Vintar Theresa Beltramo Antonia Delius (World Bank) (UNHCR) (World Bank) Dennis Egger Utz Johann Pape (UC Berkeley) (World Bank) JEL codes: J4, I15, O2, O15 Keywords: Labor, Refugees, Covid-19 1 We would like to thank Florence Nimoh (UNHCR) and Eugenie Rose Fontep (World Bank) for their input to this paper. This work is part of the program “Building the Evidence on Protracted Forced Displacement: A Multi- Stakeholder Partnership''. The program is funded by UK aid from the United Kingdom's Foreign, Commonwealth and Development Office (FCDO), it is managed by the World Bank Group (WBG) and was established in partnership with the United Nations High Commissioner for Refugees (UNHCR). The scope of the program is to expand the global knowledge on forced displacement by funding quality research and disseminating results for the use of practitioners and policy makers. We further thank FCDO for additional funding support through its Knowledge for Change (KCP) program. This work does not necessarily reflect the views of FCDO, the WBG or UNHCR. 1. Introduction The COVID-19 pandemic has caused unprecedented global employment losses of 114 million jobs worldwide in 2020 compared to the previous year (ILO, 2021). Kenya, as well as other parts of the world, has been strongly affected as restrictions on personal mobility have severely disrupted economic activities. 2 Although losses have been alleviated through welfare support programs, hard-hit groups remain: children, youth, women, low-paid and low-skilled workers (FAO, 2020; Hill & Narayan, 2020; Chetty, Friedman, Hendren, Stepner, & Opportunity Insights Team, 2020; ILO, 2021; Josephson, Kilic, & Michler, 2020). Displaced populations are particularly vulnerable as they arrive in host destinations without initial assets or established business connections and often lack access to official documentation to establish themselves in formal markets or access formal financing. Refugees living in Kenya face working and movement restrictions which severely impact their ability to participate in the labor market and limit their livelihood opportunities (Zetter & Ruaudel, 2016). Refugees were experiencing high levels of poverty, food insecurity as well as low school attendance and labor market participation rates.3 Yet despite their large numbers and contributions to the local economy, quality data on refugee communities is scarce. 4 In particular, while qualitative assessments or cross-sectional quantitative analyses have been conducted in the past (MacPherson & Sterck, 2021; Sterck & Delius, 2020; Alix-Garcia, Walker, Bartlett, Onder, & Sanghi, 2018), reliable longitudinal studies on refugee job market outcomes are not readily available, severely restricting the ability to assess the impact of a crisis like COVID-19 on refugees. This study uses the Kenya COVID-19 Rapid Response Phone Survey (RRPS) to investigate the socioeconomic implications of the COVID-19 pandemic on labor market outcomes of urban national and refugee communities in Kenya. Refugees in Kenya live in densely populated refugee camps inhabiting more than 200,000 inhabitants and in urban areas, primarily in Nairobi. Available job-market opportunities and labor market dynamics are therefore comparable to urban nationals rather than rural nationals who predominantly engage in agriculture, an economic activity refugees cannot pursue as they do not have the legal right to own land. From May-June 2020 until April-June 2021 we interviewed 6,343 households consisting of both refugees and Kenyans over five survey waves. Our data is unique in at least four dimensions: i) it leverages the high cell phone penetration and coverage throughout the country, including the refugee communities, to reach households during lockdowns when face-to-face interviews are impossible to conduct; 5 ii) its longitudinal nature allows not only to assess the first order impact of the COVID-19 shock but also its longer-term implications for recovery; iii) interviews cover refugees and nationals over the same period and are 2 Khamis et al. (2021) estimate that the work-stoppage rate in Kenya reached up to 62 percent compared to before the pandemic. 3 Results from socioeconomic surveys carried out by UNCHR and the World Bank in Kalobeyei settlement in 2018 and in Kakuma camp in 2019 show that 65 percent of Kalobeyei refugees and 68 percent of Kakuma refugees are poor, while at least 7 in 10 of them are highly food insecure (UNHCR & World Bank, 2020; UNHCR & World Bank, 2020). 4 Kenya hosted around 530,000 refugees in August 2021 which makes it the second largest hosting country in Africa after Ethiopia (UNHCR Kenya, 2021). Alix-Garcia et al. (2018) estimate that the presence of Kakuma refugee camp has increased demand for low-skilled jobs and wage labor and led to favorable price changes for local producers which on aggregate contribute to a 25 percent increase in household consumption in areas in close proximity to the camp compared to surrounding areas. 5 Eighty-six percent of Kenyan households own a mobile phone (KNBS, 2016). Among refugees, 69 percent of Kakuma refugees own a mobile phone (IFC, 2018) and 99 percent report having access to a mobile phone (Hounsell & Owuor, 2018). Access is harder to estimate for urban refugees living in Nairobi, as no official survey numbers are available (Eppler, et al., 2020). However, given the high access to technology in the capital, the mobile phone ownership share is expected to be at least at the national average level (Eppler, et al., 2020). 2 conducted in the same modality, allowing for a comparison between both communities; iv) the survey is nationally representative of both the national and the refugee population. Since few refugees were employed before the pandemic and most faced restrictions on their mobility and the types of work they could engage in much before COVID-19, we expected the influence of lockdowns on refugee labor outcomes to be less pronounced than for nationals. However, using a panel setup with fixed effects for waves and locations as well as controls for gender, age and educational attainment, we find that refugee outcomes significantly worsened during the pandemic. Even before the COVID-19 virus reached Kenya and lockdown restrictions were put into place, only 20 percent of refugees were employed or self-employed in agriculture or in household businesses. In contrast, the share of working Kenyan nationals was almost 43 percentage points higher, at 63 percent. After March 2020 and the onset of the pandemic, less than 1 in 10 refugees remained employed compared to 40 percent of nationals. When controlling for demographic characteristics and county fixed effects, the initial level-difference in employment between refugees and nationals remains large, at 31 percentage points (32 percent compared to 63 for nationals). After the initiation of a nation- wide lockdown in March 2020, the employment share of both refugees and nationals dropped by a similar margin of approximately 36 percent such that even when refugees had the same educational attainment, were of the same gender and age and lived in the same areas as nationals, only 21 percent of them were employed in May-June 2020 as compared to 40 percent of nationals. When initial restrictions were lifted later in 2020, refugees experienced a more stagnant recovery than nationals. While average hours worked for nationals remained on an upward trajectory from 11 to 23 hours between May-June 2020 and January-March 2021, average hours worked for refugees remained close to constant at a very low level (7 hours a week) until March 2021 and only improved afterwards. For household incomes, we observe a large and significant difference in income levels at the beginning of the pandemic which narrows in January-March 2021 but widens in April-June 2021. The different rates of recovery lead to a significant and widening difference in the household per capita income and the total number of hours worked towards the end of the panel. These results cannot be explained by demographic and location differences, previous employment status, living in cities as opposed to camps or sectoral choice. This suggests there is a third, unobservable ‘refugee factor’ which inhibits recovery after a major demand-supply shock and therefore aggravates pre-existing vulnerabilities in times of need. In particular, components of the refugee factor which we cannot control for are the quality of education and training received, the special institutional framework which limits refugees’ rights to move, settle and work freely in Kenya, the restricted access to formal permits, documentation, and social safety-nets or variation in the provision of aid-distributions. Our results indicate that these unobserved characteristics translate to observed effective differences between refugees and nationals, which opens avenues for policy programs to level the playing field through inclusive evidence-based measures. This study contributes to the emerging literature investigating socioeconomic impacts of the COVID-19 pandemic with special focus on labor-market impacts which remain under- investigated in the African context as post-COVID-19 data is often scarce (Khamis, et al., 2021; 3 Egger, et al., 2021). Apart from Egger et al. (2021) who include refugee samples in their analyses, no study to the best of our knowledge has empirically investigated pre- and post- pandemic labor market trends for refugees and hosts in Kenya. Although Janssens et al. (2021) and Kansiime et al. (2021) include pre-COVID-19 trends in their analyses, the authors only consider the impact of the pandemic on Kenyans. In addition, Janssens et al. (2021) conducted their analysis in rural Kenya while Kansiime et al. (2021) compared the situation between Kenya and Uganda, neither of which provides a holistic picture of the situation of Kenyan nationals together with refugees living in both urban and rural areas of Kenya. The current study aims to fill these gaps and contributes to the literature that examines the labor market implications of the COVID-19 pandemic around the world in general and in developing countries in particular. The paper also contributes to the growing literature on refugee economies. While most rigorous empirical examinations in this field have considered the impact of refugee migration to the host community (Alix-Garcia, Walker, Bartlett, Onder, & Sanghi, 2018; Depetris- Chauvin & Santos, 2018; Maystadt & Duranton, 2019; Verme & Schuettler, 2019), or have evaluated policy changes within specific refugee contexts (MacPherson & Sterck, 2021; Sterck & Delius, 2020) this study uses a comparative design to investigate the connection between a global demand-supply shock and labor outcomes of both the refugee and the national community, with results being representative at the national level for both communities. The paper is structured as follows: In the next section we briefly describe the refugee context and COVID-19 situation in Kenya. In section 3 we elaborate on the novel high-frequency phone data used in the paper and present our empirical strategy. In Section 4 we present our main findings on how refugees experienced the pandemic over time as compared to nationals. In Section 5 we discuss the results and their policy implications. 2. Background 2.1 The Kenyan refugee context Kenya is the second largest refugee-hosting country in Africa after Ethiopia, with an estimated 534,622 refugees and asylum seekers as of September 30, 2021. Refugee inflows started in the early 1970s and reached a peak in the 1990s with refugees today living in Kenya primarily stemming from Somalia (53%), South Sudan (25%), the Democratic Republic of Congo (9%), and Ethiopia (6%) (UNHCR Kenya, 2021). Refugees are hosted in three main locations: Dadaab camps in Garissa county (43%), Kakuma camp and Kalobeyei settlement in Turkana county (41%) and urban areas, mainly Nairobi (16%). As more refugees have entered the country, camps have grown to considerable sizes, with Kakuma and Dadaab camps hosting more than 200,000 refugees each in July 2020 (UNHCR Kenya, 2021; UNHCR Kenya, 2021). In both counties the camps serve as commercial hubs in an otherwise semi-arid infertile place and thus have grown to fulfill a central cultural, social, and economic role, for both the refugees and the national community. With the influx of refugees in the 1990s, Kenya changed its aid-delivery approach to one based on temporary protection and containment. Instead of integrating refugees into the 4 national society and service structure, refugees were offered temporary protection in purpose-built encampments in proximity to the Kenyan border (Betts, Sterck, & Omata, 2018). The initial goal was to resettle refugees once conflict ceased, therefore employment and agricultural opportunities as well as access to government services were limited, freedom of movement restricted, and service delivery taken over by UNHCR and implementing partners. 6 However, with conflicts ongoing in the East Africa region, the refugee camps continue to persist and to grow. With some of the oldest parts of Kakuma and Dadaab complex being older than 30 years in 2021, encampments have turned from a temporary to a permanent solution. Refugees continue to face legal hurdles that make them more vulnerable during socioeconomic shocks. First, refugees do not have the right to settle freely in Kenya nor to own land. Once registered with UNCHR in one of the camps, refugees are allotted a portion of land that contains a small shelter. The allocated land cannot be sold or purchased through official means, which poses an administrative hurdle in relocating to a different area or expanding agricultural production. Except for small plots that are allocated within Kalobeyei settlement, most allotted land is infertile and does not permit agricultural production (Betts, Omata, Rodgers, Sterck, & Stierna, 2018). Second, refugees are not allowed to leave the encampment unless a travel-permit is acquired in advance which requires an application process that can be lengthy. Income generation strategies that rely on mobility are therefore limited. In addition, refugees forfeit their right to food and cash assistance if relocating to an urban area where assistance from NGOs is very limited. Third, refugees do not have the official right to work outside the camps, unless a work permit is obtained in advance for which a recommendation from a prospective employer must be accompanied by a letter from the RAS confirming refugee status leading to in practice permits being rarely issued (UNHCR & World Bank, 2020). This results in many refugees taking low- paying jobs in the informal sector that are more prone to short-term contracts and layoffs without much protection during times of crises (Betts, Sterck, & Omata, 2018). As many refugees flee their country without much preparation, they frequently lack official documents such as birth certificates, IDs or schooling certificates, which are necessary to access many Kenyan services or signal one’s skill level to a prospective employer. Refugee entrepreneurs are allowed to open their own business inside the camps by applying for a business license to the local county government. However, the license needs to be renewed annually with charges depending on business size, effectively making it a ‘tax’ on economic activity (Betts, Sterck, & Omata, 2018). Given the regulatory challenges, only a fraction of refugees in camps were employed before the pandemic. Results from socioeconomic surveys show that in 2019 and in 2018 only around 20 percent of refugees in Kakuma Camp and 39 percent of those in Kalobeyei Settlement were employed in any sort of economic activity. In contrast, more than 60 percent of Kenyans at the national level and in the county-level host community were employed over the same timeframe. Wage work is the most important type of employment for refugees in Kakuma with nearly 50 percent of the 18-64 year old population employed for wages, usually in the 6 While the refugee camps are jointly administered by the Refugees Affairs Secretariat (RAS) of the Government of Kenya and UNCHR, shelter, food and cash assistance, health care, and education services in Kakuma and Dadaab are primarily provided by the international community (UNHCR & World Bank, 2020). 5 informal sector (UNHCR & World Bank, 2020). Within the camps, refugees who are fluent in English and Kiswahili can apply for special employment opportunities (‘incentive work’) with international organizations, which covers translators, enumerators, community mobilizers, cleaners, community health workers, food assistance distributors and others (IFC, 2018). These are often contractual workers who despite working full-time are given reduced ‘incentive pay’ ranging from $40-$55 a month for most workers instead of a full salary, which allows organizations to engage refugees legally (Betts, Sterck, & Omata, 2018). 7 While among the host community self-employment is the most common work activity, only 2 in 10 of employed refugees in Kakuma are self-employed (UNHCR & World Bank, 2020). Apart from the administrative and financial hurdle of obtaining a business license, refugees rarely arrive in camps with any assets and frequently do not have access to external financing to cover startup costs. Additionally, while refugees are allowed to engage in economic activity within the camps, the occupations they can work in are often restricted. In Kakuma and Kalobeyei for example, refugees are not allowed to own livestock or sell firewood and charcoal out of concerns about fueling tensions with the pastoralist host community (Betts, Sterck, & Omata, 2018). Most self-employed refugees therefore work as small shop owners or in the service industry, i.e., small, often informal, household-run businesses with low earning potential and a high risk of shutdown during lockdowns. 2.2 Evolution of the pandemic While Kenya’s first case of COVID-19 was recorded on March 13, 2020, the first COVID-19 case among refugees was registered in Dadaab in May 2020. Since then, total case numbers have continuously increased through the first half of 2021 (see Figure 1) to a total of 173,661 confirmed cases and 3,235 deaths as of June 8, 2021.8 UNHCR reports over 550 confirmed COVID-19 cases among refugees in Kakuma and Kalobeyei and 16 deaths as of end of Q3 2021 since March 2020. The actual number of cases is likely to be much higher than the number of confirmed cases due to limited testing, an overall low vaccination rate (only about 1,000 refugees in both sites having received two doses of vaccine over the same period), and inadequate access to medical facilities and consultations. In response to the pandemic, the government of Kenya implemented an extensive set of containment measures, including closures of all schools, bars, and indoor restaurants, a ban on international flights, a national dusk-to-dawn curfew, restrictions on public gatherings, limits on public transport passenger capacities and public information campaigns (World Bank, 2021). These measures were effective in all of Kenya from the day they were implemented, including in refugee camps. Although some containment measures were subsequently lifted (Figure 1), Kenya had on average consistently more stringent policies in place than other countries in Sub-Saharan Africa (see Figure 5 in the appendix) which are expected to have had a considerable impact on the socioeconomic lives in national and refugee communities. On March 24, 2021, the country enacted a second lockdown in Nairobi, Kajiado, Kiambu, Machakos and Nakuru counties which entailed closures of schools and other 7 For example, in Kakuma and Kalobeyei over 1,200 teachers are employed by UNHCR and partners as incentive workers. They earn five times less than national teachers, while many have similar certification and perform the same job. 8 Data on COVID-19 cases in Kenya can be found on COVID-19 Dashboard from CSSE at John Hopkins University (Dong, Du, & Gardner, 533- 534.). 6 educational institutions, suspensions on in-person meetings and a dusk-to-dawn curfew. The lockdown was subsequently lifted in May 2021. Refugees living in Kenya were subject to the same containment measures as nationals, although refugee camps were in addition closed off from the outside in the very beginning of the pandemic. When the nation-wide lockdown was stepwise lifted in the fall of 2020, permissions to enter the camp for outsiders continued to be strongly limited to protect the refugees living in the densely populated camp settings. Figure 1: COVID-19 cases and RRPS timeline in Kenya Source: Ritchie, et al (2020). Data downloaded on June 14, 2021, here. 3. Data and methodology 3.1 Survey design The Kenya COVID-19 Rapid-Response Phone Survey (RRPS) is structured as a five-waves bi- monthly panel survey that targets nationals, refugees, and the Shona community. 9 The first wave of interviews was administered in May-June 2020 and the last available wave in April- June 2021. The questionnaires capture extensive demographic and socioeconomic data with modules on employment, income, coping strategies, food security, access to education and health services, child labor, subjective well-being, knowledge of COVID-19, changes in behavior in response to the pandemic, and perceptions of the government’s response. Interviews were administered via phone by trained enumerators in the local language of the respondents. The survey uses a multi-frame design consisting of three sampling frames, two for nationals and one for refugees. The first national sampling frame consists of 9,009 households received from households providing a phone number in the 2015/16 KIHBS-CAPI pilot 9 The Shona are a community of 1,670 formerly stateless persons living in Kenya who mostly originate from Zimbabwe. They were granted citizenship on December 12, 2020. In this paper we only investigate the outcomes of refugees relative to nationals in the wake of COVID-19. We do not investigate the Shona stateless sample. 7 survey. 10 In the first round, 3,295 observations completed interviews. This sampling frame is complemented using Random Digital Dialing (RDD) to ensure that all households that existed in 2020 and who had access to a phone are considered, regardless of whether they had a phone in 2015/16. Using RDD, we generate 93 million phone numbers among which a random subset of 5,000 numbers were sent an introductory SMS to confirm operational capability. 4,075 numbers were confirmed active out of which we successfully interviewed 763 households in the first wave. Based on the collected data, we classify respondents from the national samples into three groups: 1) existed in 2015/16 and did not change number; 2) existed in 2015/16 but changed number; 3) did not exist in 2015/16. The KIHBS households are representative of type-1 households while the RDD households are representative of type-1, 2 and 3 households. We merge the two samples by adjusting weights for KIHBS households and RDD type-1 households to account for the possibility of double-selection. 11 The refugee sample was constructed using four strata: Kakuma camp, Dadaab camp, Kalobeyei settlement and urban refugees. For refugees in Kakuma and Kalobeyei, sampling was based on previously conducted representative socioeconomic surveys (UNHCR & World Bank, 2020; UNHCR & World Bank, 2020; UNHCR & World Bank, 2020), while for urban refugees and refugees living in Dadaab no such data was available, and sampling was based on phone numbers obtained from official UNHCR registration records (proGres). For each stratum, 1,000 phone numbers were drawn and sent an introductory SMS. From the list of active numbers 500 households per stratum and 375 replacements were drawn, from which 1,326 households were interviewed in total in the first wave, 1,687 in the second wave, 1,469 in the third wave, 1,357 in the fourth wave and 1,536 in the fifth wave. 12 3.2 Variable definitions We measure the extensive margin by an employment indicator which equals one for any refugee aged 18 to 64 who was employed or self-employed in agriculture or other non- agricultural household businesses and equal to zero if an individual was not working in a given period. To construct a pre-COVID-19 employment indicator we use a set of retrospective questions available from waves 2 to 4 which ask for every household member if they were working in either wage work, household businesses or agriculture in February 2020.13 We also use work stoppages measured by the proportion of individuals who were employed before COVID-19 and who stopped working afterwards. We construct an indicator equal to one if an individual worked in February 2020 and did not work at time the time of interview. Work stoppage equals to zero when an individual was working both before COVID-19 as well as at the time of interview. For measuring the intensive margin of labor supply, we consider hours worked per week and monthly per capita household income. Total hours worked is only available for wave 1 to 5 (without retrospective questions for before wave 1) and defined for each adult member in 10 The Kenya Integrated Household and Budget survey is nationally representative of the population of Kenya and is administered by the Kenyan Bureau of Statistics (KNBS) using computer assisted personal interviewing (CAPI) methodology. 11 We multiply weights with 1/(number of phone numbers the household uses) such that households with more phone numbers, which are more likely to be called randomly in the RDD, receive a lower weight. 12 For a detailed discussion on the construction of survey weights for refugees, please see the appendix. For a discussion on survey design and representativeness, we provide detailed information in the methodological annex of our in-depth survey report (World Bank, 2021). 13 Unfortunately, retrospective questions on agricultural and enterprise work were not available at the household member level in wave 1. 8 the household based on a 7-day recall distinguishing between agriculture, non-agricultural household enterprises and wage work. 14 It is common practice among small business owners to serve customers from their household throughout the day. These businesses operate from early morning until late night. To reflect this, we top code hours at 120 hours per week or 17 hours a day. Based on a 14-day recall window, we also calculate the total monthly income per capita for each household as the sum of wage income, agricultural earnings and enterprise profits across all household members and enterprises. Retrospective questions are used to construct an income measure for February 2020. We only consider households reporting positive incomes and use the inverse-hyperbolic sine (ihs) transformation to rescale the measure. In addition, we winsorize income at the 99th percentile. 3.3 Empirical model To differentiate between observable characteristics and a ‘refugee factor’ we start by inspecting the difference between refugees and nationals through the following specification: = 1 ∙ + 2 ∙ + ′ + (1) where denotes the outcome variable for individual and wave . 15 are wave fixed effects and ′ is a vector of controls. is set to 1 for urban nationals while is set 1 for refugees. Since we include a full set of wave interactions instead of a constant term, the �2 coefficients deliver point-estimates for the expected mean �1 and outcome within the specified group for each wave, evaluated at ′ = 0, instead of making all point estimates relative to an omitted category. The difference, �1 yields the �2 − expected difference in mean outcomes between refugees and nationals, given that observable characteristics in ′ are evaluated at the same values for both groups. We do not include a set of individual or household fixed effects, as they would absorb time-invariant characteristics such as the influence of the institutional setting, which are important determinants of the ‘refugee factor’ we are interested in measuring. Instead, we include a set of county-fixed effects to account for crucial geographic differences between regions. Standard errors are clustered at the household level. For household-member level outcome variables such as total hours worked, employment, and the work-stoppages, the vector of demographic controls includes age, age squared, gender, educational attainment and respondent dummy to account for differences in interview responses.16 For household income per capita, demographic controls include the age, gender, and educational attainment of the household head as well as a dummy indicating whether the household had poor flooring material as a proxy for poverty status. 14 We only consider hours worked for individuals in the relevant working age population (18–64 years) and set hours equal to zero if an individual was not working in a given period. 15 For indicators where baseline data is available, we consider February 2020 as the first wave in our panel-setup. 16 Data on educational attainment was only available for the interview respondent in waves 1 and 2. For household members who were not the respondent, we impute wave 1-2 attainment using information on educational attainment provided from waves 3 to 5. 9 4. Results 4.1 Cross-sectional analysis In our analysis we compare refugee groups to urban rather than rural nationals. Refugees are living in camp-environments in areas with high population density and usually more than 200,000 inhabitants. Job market opportunities and dynamics for refugees are, thus, comparable to urban rather than rural areas.17 However, using the rural national strata does not change the main results of the analyses as we observe in a robustness check. Our data highlights large baseline differences between the refugee and national communities that are in line with the findings in previous studies (Betts, Sterck, & Omata, 2018; UNHCR & World Bank, 2020). Refugee households are on average larger (5.6 members compared to 3.8 for nationals) and more likely to be headed by women (48 percent compared to 36 for nationals) both of which are factors that correlate strongly with household poverty levels and vulnerability during shocks (UNHCR & World Bank, 2020; UNHCR & World Bank, 2020; Kumar & Quisumbing, 2014). Refugees are also more vulnerable than nationals. They are on average less educated with 36 percent having completed secondary school compared to 44 percent of urban nationals. They are also more than four times more likely to have their floor made from lower quality material than nationals. 18 Approximately one-third of the refugee sample is engaged in any kind of work compared to almost two-thirds of the nationals’ sample. The large difference is likely due to the lack of required documents to work (birth certificates, IDs and work permits). They are also more likely to be involved in informal employment than urban nationals which tends to be underreported in surveys (UNHCR & World Bank, 2020; UNHCR & World Bank, 2020). Among working refugees, 85 percent work in trade and services while only very few engage in agriculture (8 percent) or manufacturing and construction (7 percent). By contrast, 55 percent of urban nationals engage in agriculture, 37 percent in trade and services and 8 percent in manufacturing. These differences are not surprising, as refugees do not have the official right to own land but are instead allocated a piece of land by RAS/UNCHR. Agricultural engagement is therefore restricted to the few households that had been allotted suitable acres of lands. The share of employed refugees is more than three times smaller than the share of urban nationals before the start of the pandemic, which aligns with previously conducted socioeconomic studies (UNHCR & World Bank, 2020). In terms of income, refugee households earn on average Ksh 2,966 per capita a month before the start of the pandemic. In contrast, urban nationals earn almost two times that amount (Ksh 5,315). With the nation-wide lockdown, the share of working refugees halves to 9 percent while for urban nationals the share falls from 63 to 40 percent, as can be observed in Figure 2. Given the already small share of refugees employed at baseline, these results suggest that after the 17 As a point in case: while more than 80 percent of the rural national sample engage in agriculture, refugees do not have the legal right to own land and hence only few camp-based refugees in our sample engage in agriculture (9 percent). The dynamic behind employment outcomes would be very different between these two groups and looking at the results for rural nationals could lead to false interpretations of the labor-market impacts of the pandemic. 18 For reference, see Table 3 in the appendix which highlights weighted baseline statistics for the first wave of survey collection. 10 onset of the pandemic, refugees’ access to the labor market was severely restricted. Moreover, the observed level gap in the employment share stays consistent throughout all waves, at approximately 39 percentage points. Even when considering the subsample of refugees who had employment before the pandemic, we observe a high fraction who stop working in May-June 2020 (68 percent). While the work stoppage rate improves for urban nationals since the first survey wave, refugees remain almost as likely to be out of work in October-November 2020 than at the beginning of the pandemic.19 In April-May 2021 still half of refugees who were working in February 2020 are out of work, compared to only 23 percent of urban nationals. Figure 2: Weighted means of outcome variables over time Figure 2-A: Employed Figure 2-B: Work stoppage rate Figure 2-C: Total hours worked (past 7 days) Figure 2-D: Monthly household income per capita (ihs- transformed and winsorized at 99th percentile) We observe a similar pattern for hours worked. 20 In May-June 2020, when only a fraction of refugees still managed to be employed, average hours worked across all refugees are close to zero, at 3 hours per week. In comparison, nationals work significantly more at 11 hours per week. Over time, as more nationals resume working, hours improve such that in April-June 2021 average hours worked across nationals were 26 hours per week. In contrast, refugees work the same hours in January-March 2021 than in May-June 2020, at the start of the pandemic 21 and only begin improving in the last wave of sample collection. Nonetheless, 19 Sixty-six percent in October-November 2020 compared to 68 percent in May-June 2020, with a p-value on the difference of 0.613. 20 Unfortunately, the questionnaire does not elicit hours spent in self-employed activities for February 2020. However, when restricting the sample to active wage workers, refugees work significantly less than urban nationals (34 hours compared to 45 in the last week). 21 It was 3.91 hours in January-March 2021 compared to 3.41 hours in May-June 2020, with a p-value on the difference of 0.227. 11 there remains a significant gap in levels in hours worked between refugees and urban nationals at the end of the survey period. For household per capita incomes, we also observe large baseline differences which persist over most waves. Before the pandemic, refugees earn 56 percent less than urban nationals. This difference widens after the onset of the pandemic, as incomes fall by 64 percent for refugees compared to 49 percent for urban nationals between February 2020 and In May- June 2020. Thereafter incomes improve for both groups but at no point do refugees fare better than urban nationals. To the contrary, the pandemic appears to have widened the income gap between refugees and urban nationals. In April-June 2021 the difference in per capita incomes between refugees and urban nationals (Ksh 5,592) is more than twice the size of the corresponding gap before the pandemic (Ksh 2,348). 4.2 Results controlling for key demographics and location In Figure 2 we observed large baseline differences between refugees and urban nationals that not only persisted throughout the pandemic but even widened in some cases towards the end of the panel, highlighting potential vulnerability of the refugee community in the face of a prolonged shock. However, as refugees are very different from nationals in terms of educational attainment, household size, where they live or in what types of work they engage (see Table 3 in the appendix), we cannot rule out whether the observed level differences can be explained by observable characteristics or whether they are caused by unobservable factors that shape the economic lives of refugees. In the first three columns of Table 1 we present the unconditional means for urban nationals and refugees corresponding to Figure 2 as well as the difference between the two. In the fourth column we present the difference �1 from our empirical model, controlling for �2 − demographic characteristics and county fixed effects. Once demographic and location characteristics are accounted for, the difference between refugees and nationals shrinks for most outcome indicators but remains substantial and significantly different from zero in most waves. Refugees remain significantly less likely to be employed in each wave. The difference in employment rates between refugees and nationals in each period remains relatively constant over time and in some cases exceeds the initial difference from February 2020. For work stoppages, adding demographic and location controls shrinks the estimated difference between refugees and urban nationals such that we cannot statistically distinguish it from zero in May-June 2020 and July-September 2020. After October-November 2020, standard errors shrink, and we observe a significant difference in the work-stoppage rates between both groups that widens from wave 3 onward. Similarly for hours worked, the difference between refugee and national outcomes is small in May-June 2020 and widens thereafter until January-March 2021 only to close slightly in April-June 2021. For household income, we similarly observe a smaller difference at the beginning of the pandemic, which widens over time and remains large and significantly different from zero in the last wave of survey collection. Moreover, when using demographic and location controls, the income gap of Ksh 3,737 in April-June 2021 is more than twice as large than the Ksh 1,525 pre-pandemic income 12 gap. Next, we investigate whether the widening of the difference can be attributed to a worsening of outcomes for refugees or rather different speeds of recoveries for both groups. 4.3 Rate of recovery Even when demographic and location characteristics are accounted for, refugees are 31 percentage points less likely employed in February 2020 than urban nationals. With the onset of COVID-19, the large initial difference narrows in May-June 2020 due to a relatively larger drop in the share of employed urban nationals. However, after May-June 2020 employment rates of urban nationals recover at a quicker rate, leading to a widening of the employment gap. Even if they had the same demographic characteristics and lived in the same location in April-June 2021, only 40 percent of refugees would be employed compared to 70 percent of urban nationals. Figure 3: Mean for urban nationals and estimated mean for refugees, controlling for demographics and location Figure 3-A: Employed Figure 3-B: Work stoppage rate Figure 3-C: Total hours worked (past 7 days) Figure 3-D: Monthly household income per capita (ihs- transformed and winsorized at 99th percentile) Note: The figures show the weighted unconditional mean for urban nationals and the estimated mean for refugees. The latter is calculated as the sum of the unconditional mean for urban nationals and the estimated conditional difference between refugees and urban nationals. Controls used include age, age squared, gender, educational attainment, a respondent dummy and county fixed effects. Only for income, the set of controls changes to: age, gender, educational attainment of the household head and county fixed effects. For the subset of individuals who were employed before the COVID-19 outbreak, 61 percent of refugees do not work in May-June 2020 which is comparable to the work stoppage rate of 13 urban nationals (53 percent). 22 However, as Kenya lifted its restrictions in fall 2020, the work stoppage rate for refugees remains as high as in May-June 2020 23 and persists above 50 percent for months afterwards. In contrast, the work-stoppage rate of urban nationals improves significantly from the beginning of the panel such that in April-June 2021 it lies at only 22 percent. While the national lockdown was enacted for both refugees and nationals at the same time, the lockdown was enforced for longer in refugee camps than in urban settlements, potentially restricting the recovery in the labor market. For Kakuma and Kalobeyei, which prior to COVID-19 received many external visitors, as a precaution to protect the refugee community, travel was essentially halted with very few external visitors allowed to visit through the end of Q3 2021. Refugees work approximately the same number of hours as nationals in May-June 2020 once demographic and location differences are accounted for. However, while hours for nationals are on an upward trajectory since June 2020, hours for refugees remain at the same level until March 2021 and only improve afterwards. In terms of income, after the significant initial 72 percent drop in income in May-June 2020, refugees record strong income gains until the beginning of 2021. However, between January-March 2021 and April-June 2021 refugee incomes fall by 23 percent leading to a significant widening difference in household per capita income in the last round of survey collection. In Figure 6 in the appendix we assert whether refugees who were employed before the pandemic and therefore had established ties, are more resilient in the face of a shock than when the whole population is considered. 24 We do not find much evidence that having an income earning activity at baseline makes refugees more resilient in a crisis. Refugees and urban nationals who were employed before the pandemic work comparable hours in the first five months after the COVID-19 shock. The recovery in hours after fall 2020 is slow for refugees and much quicker for nationals. This differences in trends widens the observed difference between urban nationals and refugees such that in January-March 2021, one year after the pandemic, refugees work on average 13 hours less a week than urban nationals. In term of income, refugee households with an income earning activity at baseline earn significantly less than urban nationals in every round since the start of the pandemic when demographic characteristics and location are controlled for. They experience a stronger negative income shock in May-June 2020 than urban nationals (outcomes fall by 64 percent for refugees and 45 percent for urban nationals) and recover at a similar rate thereafter but remain consistently 29-61 percent below the income levels of nationals. These results highlight that in addition to facing barriers of entry into the labor market, refugees who overcame these barriers before the pandemic were still significantly negatively affected in its aftermath. Re-entering the labor market once exited proves to be challenging, regardless of previous employment status, resulting in a stretched-out and slow recovery process. 22 Controlling for demographic and location characteristics, the difference is 8.3 percentage points with p-value 0.163. 23 It was 61 percent in May-June 2020 compared to 60 percent in October-November 2020 with a p-value on the difference of 0.829. 24 Refugees with previous employment experience do not only proxy for the importance of business connections but also a set of unobservable characteristics correlated with self-selection into an income earning activity. In this section we assert whether these observable and unobservable factors translate to more favorable labor market outcomes in times of a pandemic. 14 Table 1: Regression results Employed Work stoppage rate Raw mean Raw mean Uncond. Conditional Exp. mean Raw mean Raw mean Uncond. Conditional Exp. mean nationals refugees difference difference refugees nationals refugees difference difference refugees [SE] [SE] [p-value] [p-value] [CI] [SE] [SE] [p-value] [p-value] [CI] (1) (2) (2) – (1) (3) (1) + (3) (3) (4) (4) – (3) (5) (3) + (5) February 2020 0,626 0,200 -0,426*** -0,308*** 0,318 [0.018] [0.026] [0.000] [0.000] [0.21; 0.42] May-Jun 2020 0,395 0,092 -0,303*** -0,188*** 0,207 0,530 0,675 0,145*** 0,083 0,613 [0.017] [0.008] [0.000] [0.000] [0.12; 0.30] [0.023] [0.027] [0.000] [0.163] [0.50; 0.73] Jul-Sep 2020 0,456 0,108 -0,347*** -0,239*** 0,217 0,457 0,550 0,092*** 0,045 0,503 [0.020] [0.006] [0.000] [0.000] [0.12; 0.31] [0.024] [0.024] [0.007] [0.433] [0.39; 0.62] Oct-Nov 2020 0,515 0,095 -0,420*** -0,317*** 0,198 0,364 0,657 0,293*** 0,240*** 0,604 [0.022] [0.006] [0.000] [0.000] [0.09; 0.30] [0.024] [0.022] [0.000] [0.000] [0.48; 0.73] Jan-Mar 2021 0,579 0,129 -0,450*** -0,339*** 0,240 0,243 0,564 0,321*** 0,245*** 0,488 [0.021] [0.007] [0.000] [0.000] [0.15; 0.33] [0.021] [0.023] [0.000] [0.000] [0.37; 0.61] Apr-Jun 2021 0,695 0,280 -0,414*** -0,297*** 0,398 0,228 0,482 0,254*** 0,196*** 0,424 [0.017] [0.009] [0.000] [0.000] [0.31; 0.49] [0.020] [0.023] [0.000] [0.001] [0.31; 0.54] Demographics No No No Yes Yes No No No Yes Yes County FE No No No Yes Yes No No No Yes Yes N (pooled waves) 33,826 21,703 55,529 52050 52,050 17,803 3161 20,964 20,482 20,482 Note: All results are weighted using appropriate sampling weights. Column 4 for each variable conditions on demographic controls and county fixed effects. For employment, the work stoppage rate and hours worked, demographic controls include age, age squared, gender, educational attainment and whether the individual was the respondent. For income, demographic controls include age, gender and educational attainment of the household head as well as a dummy indicating poor flooring material. 15 Table 1: Regression results – continued Hours worked in total (past 7 days) Monthly household income per capita (ihs-transformed) Raw mean Raw mean Uncond. Conditional Exp. mean Raw mean Raw mean Uncond. Conditional Exp. mean nationals refugees difference difference refugees nationals refugees difference difference refugees [SE] [SE] [p-value] [p-value] [CI] [SE] [SE] [p-value] [p-value] [CI] (6) (7) (7) – (6) (8) (6) + (8) (9) (10) (10) – (9) (11) (9) + (11) February 2020 9,271 8,688 -0,583** -0,322 8,949 [0.099] [0.202] [0.010] [0.262] [8.39; 9.51] May-Jun 2020 10,552 3,413 -7,139*** -3,296 7,256 8,602 7,658 -0,944*** -0,945*** 7,657 [0.676] [0.326] [0.000] [0.120] [3.10; 11.41] [0.147] [0.147] [0.000] [0.001] [7.09; 8.22] Jul-Sep 2020 15,445 3,595 -11,850*** -8,285*** 7,160 8,624 8,000 -0,624*** -0,460 8,164 [1.080] [0.266] [0.000] [0.001] [2.44; 11.88] [0.173] [0.116] [0.003] [0.115] [7.59; 8.74] Oct-Nov 2020 19,861 2,996 -16,866*** -12,813*** 7,048 9,415 8,289 -1,126*** -0,692** 8,723 [1.085] [0.228] [0.000] [0.000] [2.23; 11.87] [0.120] [0.123] [0.000] [0.010] [8.19; 9.25] Jan-Mar 2021 22,621 3,914 -18,708*** -14,758*** 7,863 9,400 8,713 -0,687*** -0,208 9,193 [1.303] [0.257] [0.000] [0.000] [3.23; 12.50] [0.085] [0.092] [0.000] [0.378] [8.73; 9.65] Apr-Jun 2021 26,350 9,695 -16,654*** -11,895*** 14,454 9,671 8,449 -1,222*** -0,738*** 8,933 [0.853] [0.369] [0.000] [0.000] [10.00; 18.91] [0.062] [0.050] [0.000] [0.001] [8.50; 9.37] Demographics No No No Yes Yes No No No Yes Yes County FE No No No Yes Yes No No No Yes Yes N (pooled waves) 25,187 16,529 41,716 39,502 39,502 7,700 2,263 9,963 9,828 9,828 16 The labor market and policy environments faced by urban and camp refugees are very different. Urban refugees have on average more dependents, are not eligible for emergency relief from WFP and cannot access incentive jobs that are available in refugee camps (Betts, Sterck, & Omata, 2018). On the other hand, they may have access to a wider array and potentially more productive job opportunities. In Figure 4 we split refugees into camp-based and urban refugees. Urban refugees are on average slightly more likely to be employed and work on average more hours than camp refugees in all waves. However, they are still significantly less likely employed and still work significantly less hours than their national counterparts by an average of 7 hours per week. Interestingly, we observe that despite being more likely employed and working more hours, household per capita incomes for urban refugees are comparable to camp refugees in all waves between July-September 2020 and January-March 2021. 25 At the same time, camp refugees are at least as likely or slightly more likely to have lost their jobs in terms of work-stoppage rates. Despite high work-stoppage rates, camp refugees are able to retain incomes possibly because of the prevalence of aid- delivery and incentive work available in camps. For urban refugees we observe that the employment share in May-June 2021 (48 percent) increased substantially from one wave prior such that it is now 12 percentage points above the pre-pandemic share from February 2020 (36 percent). In comparison, the share of employed camp refugees is only 7 percentage points larger in May-June 2021 compared to its pre-pandemic level from February 2020 (27 percent). To understand the potential drivers of the increased employment share, a recent in-depth report on Kenyan nationals provides relevant insight. In the report we observed higher employment rates among nationals in June 2021 than before the pandemic. At the same time, the share of individuals working in more than one type of employment has doubled over the duration of the pandemic (from 7 percent to 15 percent) and more individuals reported looking for additional income generating activities as a coping mechanism (World Bank, 2021). In addition, the share of wage workers in high quality jobs declined consistently throughout 2021 26 and job quality was particularly worse among urban populations. These points point to the fact that individuals are seeking additional income generating activities to cover losses induced by the pandemic. The higher employment share is more reflective of a higher willing to take on more (wage) jobs, resulting in higher engagement in less-paid and less secure lower quality jobs. This effect is particularly prevalent among urban populations who cannot substitute for subsistence farming and are thus more likely engaged in wage work where job quality on average worsened through the pandemic. We expect this increased willingness to take on jobs to also apply to refugees who predominantly live in urban areas, are severely impoverished and continue to face a reduction to their food ration entitlements of up to 50 percent (WFP, 2020). 25 The sample of urban refugees is small, ranging from 32 observations in May-June 2020 to 175 in April-June 2021. It is possible that with larger samples and more precision we would observe significant differences between urban and camp refugees in other waves. 26 Job quality was measured by 4 dimensions: income, benefits, satisfaction, and stability. If an individual’s salary does not exceed the poverty line, the index was set to 0. For a detailed description of the indicators used for each dimension, see Annex F.5 in World Bank (2021). 17 Figure 4: Splitting refugees into urban and camp-based refugees Figure 4-A: Employed Figure 4-B: Work stoppage rate Figure 4-C: Total hours worked (past 7 days) Figure 4-D: Monthly household income per capita (ihs- transformed and winsorized at 99th percentile) Due to frequent restrictions on the type of work refugees can engage in, 27 many refugee businesses are limited to the service industry which we expect to have been affected strongly during lockdowns (81 percent work in services as compared to 41 percent of nationals). In Figure 7, we plot the distribution of refugees and urban nationals across the most important work sectors in the first wave of survey collection as well as the distribution over sectors in the period before respondents lost their job. 28 Refugees indeed work in different sectors than urban nationals with a stronger emphasis on services. Urban nationals by contrast rely more strongly on agricultural production. However, refugees are not overrepresented in sectors that are particularly prone to layoffs. While refugees are more likely to lose their jobs if they work in other services, as household employers or as health and social workers than urban nationals, a larger share of refugees work in these sectors to begin with. In fact, there is no significant discernable difference between the distribution of jobs refugees are employed in and the distribution of jobs they are laid off from. 29 27 For example, work is restricted to within refugee camps. For any work outside the camps, a work permit needs to be acquired which is costly both in terms of time and money. Further, refugees are often not allowed to engage in agriculture, keep livestock or engage in certain industries due to fears of competition with the native pastoralist community (Betts, Sterck, & Omata, 2018). 28 Sector information is not available at the household-member level for agricultural work and work in household businesses for the baseline period, hence we use wave 1 data for the first available sector. 29 Except for ‘other services’ which was among the five most important layoff sectors over all waves but was not strongly represented among the employed in wave 1. 18 Since the composition of sectors differs between refugees and urban nationals and different sector types are likely to be affected differently during COVID-19, we re-run our main analysis controlling for whether an individual worked in agriculture, manufacturing and construction or trade in services. We group sectors into three larger categories according to ISIC Rev. 2-4 codes to ensure sufficient observation size within each refugee-wave-sector cluster. Results are presented in Figure 7 in the appendix. Despite controlling for sectoral occupation, we still observe a substantive difference between refugees and nationals in the latter months of 2020 and the beginning of 2021, suggesting that the largest part of the difference cannot be attributed to sectoral choice only. 4.4 Robustness We repeat the main analysis with rural nationals as the reference category (Figure 9 in the appendix). Overall, the results are very similar to the results presented with urban nationals. We observe a significant level-difference in employment for refugees even when compared to rural nationals and a widening gap in terms of hours worked driven by a slower recovery for refugees than for nationals. Between June 2020 and November 2020 we also observe a slower rate of recovery for refugees who worked in February 2020 but are not working at the time of the interview compared to urban nationals leading to a widening of the gap between both groups. Refugees continue to register a stronger negative income shock in May-June 2020 than urban nationals. Incomes improve afterwards but worsen at the end of the panel, leading to a significant income gap in April-June 2021. Since each household was targeted in each wave irrespective of previous interview status,30 the composition of the sample can vary substantially between waves. Around a third of the sample in each wave consists of households that have not been interviewed one wave prior. To ensure our results are not driven by the specific compositions of the sample, we repeat our main analysis including only households that have been tracked throughout all five waves and adjust weights to reflect a balanced panel. We present results in Figure 10 in the appendix. 31 Results are comparable to the main findings. While the differences in outcomes between refugees and urban nationals are less pronounced in the first months of the pandemic, we continue to observe a widening of the gap in total hours worked and employment rates after fall 2020. For household incomes, we now observe a large and statistically significant income gap in July-September 2020 that narrows in the latter months of 2020 and widens in April-June 2021. 32 5. Conclusions Even if refugees had the same age, gender, educational attainment, respondent status or lived in the same location as urban nationals, they had significantly worse employment outcomes already before the start of the pandemic. The recovery for refugees was in most cases slower and often stagnant. Nationals recovered quickly in terms of hours worked and 30 Unless the household explicitly asked not to be contacted in future interviews. 31 We do not report results for work-stoppages since sample sizes of refugees who worked in the baseline period and were included in all waves of survey collection is very small and per-period point-estimates are not reliable. 32 For May-June 2020 our refugee sample comprises only 38 observations. That we do not observe a significant difference to urban nationals is thus most likely the product of reduced precision caused by the small sample size. 19 employment rates, while outcomes for refugees remained mostly constant for several months after the initial shock and only improved in recent months. In terms of income, refugees experienced a large negative shock in May-June 2020 leading to a large income gap between refugees and urban nationals at the beginning of the pandemic. The gap narrows in 2 out of 5 survey rounds but widens again in April-June 2021. At no point do refugees earn significantly more than urban nationals. To the contrary, in most waves, refugees earn 19-61 percent less than urban nationals. Continuously low income levels coupled with a slow and often stagnant recovery in the labor-market (Figure 3) highlight the fragile job-market situation refugees face. Despite controlling for demographic and location characteristics in our main specification, or for any of our alternative specifications (Figure 6 – Figure 8 in the appendix), we are unable to explain a large fraction of the observed difference in employment outcomes between urban nationals and refugees. Refugees fared worse than urban nationals before the COVID- 19 outbreak and even diverge from nationals in later rounds of the panel due to exhibiting a slower rate of recovery. Several factors can contribute to explaining this difference. Refugees do not have a universal legal right to work in Kenya; often lack the documentation needed to acquire formal permits (movement passes and work permits), access security nets or formal financing; face strict movement restrictions; and are subject to labor-market discrimination. Also the quality of education can play a role as we only observe formal educational attainment. It is crucial for policy makers to keep in mind that refugees lived in very vulnerable circumstances even before the pandemic. Recently conducted socioeconomic studies show that about six in ten refugee households living in Kakuma and Kalobeyei did not have access to food in the right amounts and quality (World Bank, 2021; UNHCR & World Bank, 2020). With the closure of public spaces, business restrictions and closures of refugee camps to the outside world, most of the few remaining economic opportunities available to refugees disappeared, leaving most households without an income earning activity and thus fully aid- dependent in the first months after the initial shock. Even after a recovery could be recorded for nationals with the lifting of restrictions in fall 2020, outcomes for refugee communities remained stagnant. We observe that regardless of educational attainment, skills (proxied by age), gender or location, once driven out of the labor market, refugees find it exceedingly hard to navigate a tight job market and build new livelihoods. This prolonged recovery translates into a prolonged period of high labor market insecurity and by extension a high risk of continued food insecurity, especially in times when funding shortages have tightened budgets of relief organizations and aid frequently cannot be delivered at full capacity (WFP, 2020). 33 During a shock such as COVID-19, existing vulnerabilities were therefore aggravated over an extended period. It is therefore crucial for policy makers to build resilience among refugees and remedy existing frictions in the labor market. Refugees for example do not have the legal right to own land, work outside refugee camps or move and settle freely in Kenya. Current restrictions limit the viable economic opportunities refugees have legal access to and therefore their ability to find alternative income sources in times when a large fraction of jobs cannot be performed 33 Food assistance has been reduced to 60 percent of a full food ration in 2021 and the distribution cycle extended from monthly to bi- monthly distributions. In October 2021 the ration size was further reduced to only 52 percent of a full ration. 20 because of pandemic-related restrictions. Resilience-building in the face of a crisis is a crucial next step to ensure that refugees can maintain their livelihoods in times of need. Ongoing efforts will also be necessary on the research frontier, especially in providing high- quality socioeconomic data that include refugee populations and are able to address existing data gaps and contribute to the design of inclusive evidence-based measures. With the continuous spread of the pandemic, new methods of collecting data such as phone surveys need to be evaluated relative to traditional face-to-face methods and improved upon. One particularly interesting strand of future research concerns the measurement of income which is notoriously difficult to capture precisely in in-person interviews due to the complexity of income streams in developing countries as well as recall and social desirability bias. It is likely these biases are aggravated during phone interviews. To evaluate the full extent of over and underreporting in phone surveys, more research will be necessary in the future. References UNHCR Kenya. (2021). Dadaab Refugee Complex. Retrieved from https://www.unhcr.org/ke/dadaab-refugee-complex UNHCR Kenya. (2021). Kakuma Refugee Camp and Kalobeyei Integrated Settlement. Retrieved from https://www.unhcr.org/ke/kakuma-refugee-camp IFC. (2018). Kakuma as a Marketplace. A Consumer and Market Study of a Refugee Camp and Town in Northwest Kenya. World Bank. (2021). How Covid-19 continues to affect lives of refugees In Kenya. Ritchie, H., Mathieu, E., Rodés-Guirao, L., Appel, C., Giattino, C., Ortiz-Ospina, E., . . . Roser, M. (2020). Our World in Data. Retrieved from Coronavirus Pandemic (COVID-19): https://ourworldindata.org/coronavirus Dong, E., Du, H., & Gardner, L. (533-534.). An interactive web-based dashboard to track COVID-19 in real time. The Lancet infectious diseases, 2020. UNHCR & World Bank. (2020). Understanding the Socioeconomic Conditions of Refugees in Kenya. Volume B: Kakuma Camp. Results from the 2019 Kakuma Socioeconomic Survey. UNHCR & World Bank. (2020). Understanding the Socioeconomic Conditions of Refugees in Kenya. Volume A: Kalobeyei Settlement. Results from the 2018 Kalobeyei Socioeconomic Profiling Survey. UNHCR & World Bank. (2020). Understanding the Socioeconomic Conditions of the Stateless Shona Community in Kenya. Results from the 2019 Socioeconomic Survey. WFP. (2020, December). WFP Warns of Critical Food Shortages for Refugees in Kenya. Retrieved from https://www.wfp.org/news/wfp-warns-critical-food-shortages- refugees-kenya Himelein, K. (2014). Weight Calculations for Panel Surveys with Subsampling and Split-off Tracking. Statistics and Public Policy, 40-45. Alix-Garcia, J., Walker, S., Bartlett, A., Onder, H., & Sanghi, A. (2018). Do refugee camps help or hurt hosts? The case of Kakuma, Kenya. Journal of Development Economics, 130, 66-83. MacPherson, C., & Sterck, O. (2021). Empowering refugees through cash and agriculture: A regression discontinuity design. Journal of Development Economics, 149. 21 Sterck, O., & Delius, A. (2020). Cash Transfers and Micro-Enterprise Performance: Theory and Quasi-Experimental Evidence from Kenya. CSAE Working Paper Series, 2020-09. ILO. (2021). ILO Monitor: COVID-19 and the world of work. Seventh edition, Updated estimates and analysis. International Labor Organization. Hill, R. V., & Narayan, A. (2020). Covid-19 and inequality: a review of the evidence on likely impact and policy options. Centre for Disaster Protection, Working paper, 3. Chetty, R., Friedman, J. N., Hendren, N., Stepner, M., & Opportunity Insights Team. (2020). How did COVID-19 and stabilization policies affect spending and employment? A new real-time economic tracker based on private sector data. NBER Working Paper, 1-109. Josephson, A., Kilic, T., & Michler, J. D. (2020). Socioeconomic impacts of COVID-19 in four African countries. World Bank Policy Research Paper, 9466. Khamis, M., Prinz, D., Newhouse, D., Palacios-Lopez, A., Pape, U., & Weber, M. (2021). The Early Labor Market Impacts of COVID-19 in Developing Countries: Evidence from high-frequency phone surveys. World Bank Jobs Working Paper, 58. Zetter, R., & Ruaudel, H. (2016). Refugees’ right to work and access to labor markets – An assessment. KNOMAD Working Paper. Eppler, M., Gaetani, S., Köllner, F., Kuhnt, J., Martin-Shields, C., Mebrahtu, N., . . . Preiß, C. (2020). Information and communication technology in the lives of forcibly displaced persons in Kenya. Discussion Paper, 15/2020. Hamilton, Z., Casswell, J., & Alonso, A. (2020). The digital lives of refugees and Kenyans with disabilities in Nairobi: A human-centred design approach to identifying mobile- enabled opportunities. London: GSM Association. Egger, D., Miguel, E., Warren, S. S., Shenoy, A., Collins, E., Karlan, D., . . . et al. (2021). Falling living standards during the COVID-19 crisis: Quantitative evidence from nine developing countries. Science Advances, 7(6). Janssens, W., Pradhan, M., Groot, R. d., Sidze, E., Pythagore, H., Donfouet, P., & Abajobir, A. (2021). The short-term economic effects of COVID-19 on low-income households in rural Kenya: An analysis using weekly financial household data. World Development, 138. Kansiime, M. K., Tambo, J. A., Mugambi, I., Bundi, M., Kara, A., & Owuor, C. (2021). COVID- 19 implications on household income and food security in Kenya and Uganda: Findings from a rapid assessment. World Development, 137. Depetris-Chauvin, E., & Santos, R. J. (2018). Unexpected guests: The impact of internal displacement inflows on rental prices in Colombian host cities. Journal of Development Economics, 134, 289–309. Maystadt, J.-F., & Duranton, G. (2019). The development push of refugees: Evidence from Tanzania. Journal of Economic Geography, 19(2), 299-334. FAO. (2020). Impact of COVID-19 on informal workers. Food and Agriculture Organization of the United Nations. KNBS. (2016). Basic Report: 2015/16 Kenya Integrated Household Budget Survey (KIHBS). Nairobi: Kenya National Bureau of Statistics. Hounsell, B., & Owuor, J. (2018). Innovating mobile solutions for refugees in East Africa. Opportunities and barrier to using mobile technology and the internet in Kakuma refugee camp and Nakivale refugee settlement. Nairobi: Samuel Hall Consulting. Kumar, N., & Quisumbing, A. R. (2014). Gender, shocks, and resilience. 2020 conference briefs 11. International Food Policy Research Institute (IFPRI). 22 Betts, A., Omata, N., Rodgers, C., Sterck, O., & Stierna, M. (2018). Self-Reliance in Kalobeyei? Socio-Economic Outcomes for refugees in North-West Kenya. Oxford Department of International Development. Oxford: Refugee Studies Centre. Betts, A., Sterck, O., & Omata, N. (2018). Refugee Economies in Kenya. Oxford Department of International Development. Oxford: Refugee Studies Centre. Verme, P., & Schuettler, K. (2019). The Impact of Forced Displacement on Host Communities : A Review of the Empirical Literature in Economics. Policy Research Working Paper, 8727. UNHCR Kenya. (2021, September). Kenya: Registered refugees and asylum-seekers. Maffioli, E. M. (2019). Relying solely on mobile phone technology: Sampling and gathering survey data in challenging settings. Working paper. Himelein, K., & Kastelic, J. G. (2021). The socio-economic impacts of Ebola in Liberia: Results from a high frequency cell phone survey, round 5. Working paper. Lau, C. Q., Lombaard, A., Baker, M., Eyerman, J., & Thalji, L. (2019). How representative are SMS surveys in Africa? Experimental evidence from four countries. International Journal of Public Opinion Research, 31(2), 309–330. World Bank. (2018). Listening to Tajikistan - Household survey: Background, implementation, and methods. Washington, D.C.: World Bank Group. World Bank. (2021). How Covid-19 continues to affect livelihoods in Kenya. 23 Appendix Additional notes on response rates Our sample size increases over waves due to an increase in the tracking rate and the share of successfully interviewed households over time (see Table 2). We kept the set of phone numbers targeted the same across waves. 34 In Wave 1, 43 percent of all targeted households could be reached and 78 percent of them accepted to be interviewed. In comparison, in wave 5, 52 percent of targeted households could be reached and almost all (99 percent) agreed to the interview. We explored this dynamic extensively with the survey firm. With time the survey team became more experienced with scheduling call backs at different times of the day (in particular evenings) and on weekends which resulted in a higher response rate. We were also able to build increased trust and confidence from the respondents due to consistently honoring the promised respondent gift. In addition, during the earlier waves, at the onset of the pandemic, there was some level of fear in the population and respondents were not very receptive. In wave 1, many households never picked up their phone when called for the survey or had it switched off altogether resulting in a lower tracking rate. This general reluctance, which was also experienced in other projects, reduced after the first lockdown was lifted. Finally, quality control and remote supervision improved over time, such that enumerators negligent with the number or time of callbacks were more easily spotted which improved both tracking and interview rates. We record an average consent rate across all five waves of 42.4 percent which is comparable to phone surveys carried out during the Ebola crisis, a situation similar to the COVID-19 pandemic. In Liberia phone surveys using a similar methodology to ours had a response rate of 46 percent over five rounds of collection (Himelein & Kastelic, 2021). Another Ebola survey in Liberia reports a response rate of 51.9 percent using random digital dialing and interactive voice response (Maffioli, 2019). SMS phone surveys in Africa have found even lower response rates of 12 percent in Kenya and less than 1 percent in Nigeria and Ghana (Lau, Lombaard, Baker, Eyerman, & Thalji, 2019). Response rates in non-crisis phone surveys tend to be slightly higher, in part because many of surveys first conduct an in-person baseline interview followed by a phone survey. For example, in Tajikistan, 25 percent of targeted households did not join the phone panel compared to 66 percent in our survey (World Bank, 2018). However, conditional on joining the panel, less than 7 percent of the sample attritted in any of the later survey waves which is consistent with our finding that agreement to participate in the surveys increases as the same respondents are interviewed over time. To account for the higher nonresponse rate in phone surveys, survey weights were adjusted. 34 Households who did not provide interview consent in earlier waves were called again unless they have explicitly stated they would not like to be called again in the future. 24 Table 2: Sample sizes by wave. Source: (World Bank, 2021) Wave 1 Wave 2 Wave 3 Wave 4 Wave 5 Data collection May 14 to July 7, July 16 to September September 18 to January 15 to March March 29 to June 13, 2020 18, 2020 December 2, 2020 25, 2021 2021 KNBS Sample 3,294 3, 664 3,982 4,060 4,710 RDD Sample 769 840 1,011 846 1,164 UNHCR Sample 1,326 1,687 1,469 1,357 1,536 Total Sample 5,389 6,191 6,462 6,263 7,410 Consent rate 33.8% 39.3% 42.9% 44.4% 51.6% Tracking rate 43.2% 46% 46.6% 44.4% 52.3% Interview rate 78.2% 85.4% 95.7 % 96.5% 98.7% Notes: The consent rate is the number of households successfully interviewed with consent as a proportion to all targeted households in a given survey wave. The tracking rate is the number of households successfully reached in each wave as a proportion to the total number of targeted households. The interview rate is the number of households successfully interviewed with consent in each wave as a proportion of all households that were successfully reached in that wave. Additional notes on the construction of survey weights For refugee strata based on previous surveys, we utilized the available data to model the probability of selection into our sample by fitting a logistic regression on a set of covariates for each stratum, 35 and calculate a propensity score for each observation. 36 The initial survey weights are scaled by the inverse of the winsorized scores such that observations with a higher estimated probability of selection are given a smaller weight. For the urban refugee and Dadaab samples without previous socioeconomic data, we instead use a cell- weighing approach. Within each stratum, we split the sample into cells based on gender and age group of the household head. We calculate the population shares of each cell from official UNHCR records and scale sampling weights by the inverse of the population share of the relevant cell. We use a set of adjustments to the cross-sectional weights to account for different sampling challenges. First, to account for differential response rates, phone-coverage and attrition from samples for which baseline data was available, we post-stratify cross-sectional weights using the 2019 Kenya Population and Housing Census for the nationals’ sample and official UNHCR proGres records for the refugee sample. This step makes our final sample cross- sectionally approximately representative of all individuals living in Kenya who have an active phone subscription in an area with network coverage. Second, as all households in the sample were targeted in each wave independent of whether they were reached in a previous wave, the composition of households in each cross-sectional sample varies over time. When considering balanced panel data in the analysis, we adjust the cross-sectional weights following the attrition adjustment outlined in Himelein (2014), to address potential bias due 35 The set of explanatory variables included dummies for the country of origin, number of children, dependency ratio as well as the gender, literacy and employment status of the household head. 36 To reduce outlier-risk, we order observations into deciles by their score, calculate a mean score for each decile and winsorize the decile- scores at the 99th percentile. 25 to attrition across waves. Lastly, while enumerators were randomly allocated to interviews, we observe high variability in employment responses. To reduce inconsistencies and obtain unbiased labor statistics, interviews collected by enumerators where the share of households without any employment lies above the 95th percentile, are dropped from the analysis. Weights for the remaining households are adjusted accordingly. 37 37 This concerns 596 of the 6,192 national households in wave 2 and 1,109 of the 6,462 national households in wave 3. No observations were dropped for the refugee sample. 26 Table 3: Descriptive statistics for wave 1 (May-June 2020) of survey collection (1) (2) (3) Refugees Urban t-test nationals Difference Variable N Mean/SE N Mean/SE (1)-(2) Age 3000 32.779 5504 33.698 -0.919 [0.402] [0.416] Gender 3089 0.503 5555 0.531 -0.028 [0.013] [0.018] Age of household head 1160 39.127 2251 37.812 1.315 [0.623] [0.662] Gender of household head 1160 0.481 2251 0.364 0.116*** [0.021] [0.029] Number of people in household 1160 5.604 2251 3.756 1.848*** [0.161] [0.128] Poor flooring material 1097 0.652 2239 0.152 0.500*** [0.019] [0.016] Educational attainment None or pre-primary 2482 0.305 4896 0.061 0.244*** [0.014] [0.008] Primary 2482 0.272 4896 0.256 0.016 [0.013] [0.016] Secondary 2482 0.362 4896 0.439 -0.077*** [0.014] [0.019] Tertiary 2482 0.060 4896 0.243 -0.183*** [0.006] [0.018] Employed in February 2020 5188 0.200 8591 0.626 -0.426 *** [0.026] [0.018] Main sector of work (by hours worked) Agriculture 286 0.082 2605 0.546 -0.463*** [0.020] [0.027] Manufacturing and Construction 286 0.072 2605 0.081 -0.009 [0.019] [0.017] Trade and Services 286 0.846 2605 0.373 0.473*** [0.027] [0.027] A household is considered to have poor flooring when the floor was made of earth sand or dung instead of wood, cement, ceramic tiles, or other higher quality materials. Household head variables, flooring material and household income are measured at the household level. The remaining variables are measured at the household member level. The value displayed for t-tests are the differences in the means across the groups. Standard errors are robust. ***, **, and * indicate significance at the 1, 5, and 10 percent level. 27 Figure 5: Google Mobility Data and Oxford Stringency Index 30 100 Lockdown Cessation of Movement Imposed Lifted 90 20 80 10 Change compared to baseline % 70 0 Stringency Index 60 -10 50 40 -20 30 -30 20 -40 R1 R2 R3 R4 R5 10 -50 0 1-Feb 1-Apr 1-Jun 1-Aug 1-Oct 1-Dec 1-Feb 1-Apr 1-Jun Kenya Stringency Index Other SSA Stringency Index Grocery and Pharmacy Transit Stations Residential Workplaces Note: Dotted lines indicate Google Mobility Data and bold lines indicate Oxford Stringency Index Source: Oxford Stringency Index, Google Mobility Data. Figure 6: Results for observations who worked before or had an income source in February 2020 Figure 6-A: Work stoppage rate Figure 6-B: Total hours worked (past 7 days) Figure 6-C: Monthly household income per capita (ihs- transformed and winsorized at 99th percentile) 28 Figure 7: Distribution of sectors in which individuals were working at baseline (W1) and distribution of sectors in which individuals did not find work in the following survey period 100% 90% Percent of working population 80% 70% 60% 50% 40% 30% 20% 10% 0% Refugees Urban nationals Refugees Urban nationals Baseline sector Layoff sector Agriculture, forestry, fishing Construction Wholesale and retail trade Transport and storage Accommodation and food service Education Health and social work Other services Activities of households as employers Other 29 Figure 8: Results with baseline sector controls Figure 8-A: Employed Figure 8-B: Work stoppage rate Figure 8-C: Total hours worked (past 7 days) Figure 8-D: Monthly household income per capita (ihs- transformed and winsorized at 99th percentile) 30 Figure 9: Results with rural nationals as the reference group Figure 9-A: Employed Figure 9-B: Work stoppage rate Figure 9-C: Total hours worked (past 7 days) Figure 9-D: Monthly household income per capita (ihs- transformed and winsorized at 99th percentile) 31 Figure 10: Results using only observations available in all 5 waves and at baseline Figure 10-A: Employed Figure 10-B: Total hours worked (past 7 days) Figure 10-C: Monthly household income per capita (ihs- transformed and winsorized at 99th percentile) Note: Survey weights incorporate the panel-attrition adjustment in Himelein (2014). Missing information on employment in household enterprises is imputed from previous waves, which increases the share of employed by 1-3 percentage points. The adjustment does not affect hours and income. 32