World Development Perspectives 31 (2023) 100529 Contents lists available at ScienceDirect World Development Perspectives journal homepage: www.sciencedirect.com/journal/world-development-perspectives Research paper Ecological shocks and children’s school attendance and farm work in Ghana Edward Martey a,1,*, Prince M. Etwire a,2, Jonathan Mockshell b,3, Ralph Armah c,d,4, Eli Akorsikumah e,5 a Socio-Economic Section, CSIR-Savanna Agricultural Research Institute, Ghana P.O. Box TL 52, Tamale, Ghana b International Center for Tropical Agriculture (CIAT), Apartado Aereo 6713, Cali, Colombia c Institute of Statistical, Social and Economic Research (ISSER), University of Ghana, Legon, Accra, Ghana d Environment for Development (EfD), Ghana e University for Development Studies, Nyankpala, Tamale, Ghana A R T I C L E I N F O A B S T R A C T Keywords: Accelerating the education of children and reducing child labor in agriculture remains an important development Ecological shocks pathway to preventing intergenerational poverty and achieving the sustainable development goals. While several School attendance studies have analyzed the impact of ecological stressors on yield, income, and food security, there is limited Farm working hours understanding of the linkages of prevailing ecological shocks to child education and farm work. In this paper, we Ghana examine the effect of ecological shocks of pest and weed invasion on children’s school attendance and working hours on the farm using the seventh round of the Ghana Living Standards Survey (GLSS). We employ a multi- nomial endogenous switching regression (MESR) model that corrects for selection bias and endogeneity origi- nating from both observed and unobserved heterogeneity. The results show that double shocks (pests and weeds) reduced the number of children attending school by 11% and increased children’s on-farm working hours by 0.75 h. Comparatively, the decline in the number of children attending school due to weed invasion (0.88) is higher than the decline due to pest invasion (0.43). Furthermore, weed invasion increases children’s on-farm working hours by 0.05 h while pest invasion reduces children’s on-farm working hours by 0.08 h. Increasing access to improved agricultural technologies bundled with credit and policies are critical to reducing the threats from ecological shocks and improving farmers’ welfare. To avert the decline in school attendance and children’s working hours requires training farmers to reduce the practice of continuous cropping and to embrace crop rotation and fallow to reduce the spread of pests and weeds. 1. Introduction Humphreys, 2022). With the renewed global interest in promoting children’s education and reduction of child labor in response to In developing economies, family farms play a critical role in ensuring achieving Sustainable Development Goal 4 (SDG 4) (United Nations, food and nutrition security of households and contributing to economic 2015), there is a consensus among stakeholders to reduce child labor in growth. Children contribute labor to agricultural production and related agriculture. Through the SDGs, world leaders agreed to achieve uni- off-farm activities. Economic burden, social norms, and ecological versal primary, all-inclusive, and equitable education. Governments shocks are major drivers of the forms and types of activities that children have made considerable efforts in promoting children’s education across participate in the farm. Often, there is a trade-off between children’s Sub-Saharan Africa (SSA). The case of Ghana is peculiar and charac- farm work and spending time pursuing formal education (Dunne and terized by several educational reforms, such as the recent free senior * Corresponding author. E-mail addresses: eddiemartey@gmail.com (E. Martey), etwiremaxwellprince@gmail.com (P.M. Etwire), J.Mockshell@cgiar.org (J. Mockshell), rnaarmah@ug. edu.gh (R. Armah), akorsikumah@gmail.com (E. Akorsikumah). 1 ORCID: 0000-0002-6933-3685 2 ORCID: 0000-0002-6533-2538 3 ORCID: 0000-0003-1990-66570 4 ORCID: 0000-0003-4552-971X 5 ORCID: 0000-0002-3871-7184 https://doi.org/10.1016/j.wdp.2023.100529 Received 25 June 2022; Received in revised form 27 February 2023; Accepted 16 August 2023 Available online 25 August 2023 2452-2929/© 2023 Elsevier Ltd. All rights reserved. E. Martey et al. W o r l d D e v e l o p m e n t P e r s p e c t i v e s 31 (2023) 100529 high school (FSHS) program that seeks to provide free secondary edu- shown that the type of work children (boys and girls) engage in affect cation for all Ghanaian children (Chanimbe and Dankwah, 2021). their schooling. The greater need for care-giving and completion of Beyond school enrolment and infrastructural provisioning, time spent in household chores by female children influence their punctuality at school is critical to achieving educational outcomes. But economic, so- school (Dunne and Ananga, 2013). The literature also highlights that cial and ecological drivers impact the achievement of child education school may not necessarily improve learning (Tafere and Pankhurst, outcomes (Dunne and Humphreys, 2022). While the literature has 2015; Humphreys et al., 2015) and prevent child work given that tasks examined child education as it relates to economics and social factors, assigned to children at school may be demanding which expose them to the exploration of effects of ecological shocks on child education is several risks (Adonteng-Kissi, 2018; Bakari, 2013; Dunne et al., 2005). recent. We aim to contribute to this knowledge gap by examining the Finally, the proponents of the edu-workscape suggested that social do- effects of ecological shocks on children’s school attendance and farm mains must be discussed within the wider economic, social, temporal, work based on a novel framework, “edu-workscape.6” and spatial contexts to understand the compromises that households Dunne and Humphreys (2022) developed the edu-workscape undertake regularly regarding education and work (Dunne and Hum- framework to explain the interaction between households, children’s phreys, 2022; Boyden et al., 2021). work, and schooling. Unlike the conventional approach which focused In this paper, we define ecological shocks as the spread of field insect only on the direct link between child work and schooling, the edu- pests and invasive weed, Striga hermonthica (witchweed) which has been workscape emphasize the interaction between the domains by contex- shown to significantly reduce crop productivity, adversely impact the tualizing the child in several ways. In the framework, the child is sustenance of people and significantly reduce food security and increase considered to be an active social subject with multiple identities which poverty (David et al., 2022; Gowda et al., 2021; Dawud et al., 2017). The are likely to influence their experiences of work, education and harm. invasion of this parasitic weed thrives in places with poor soils and The conceptualization has several implications for research approach intensive crop cultivation with poor management practices (Gowda and policy implications. In the scientific literature, the education et al., 2021). Similarly, the insect pests can cuase damage to crop both attainment agenda has been tremendous, attracting several empirical on the field and during storage. In the advent of ecological shocks, studies on children’s education to guide policy (Agamile & Lawson, children may be kept on the farm or tasked with off-farm labor as a 2021; Björkman-Nyqvist, 2013; Duryea et al., 2007; Martey et al., 2021; coping strategy to reduce the effect of the shocks and contribute to Tabe-Ojong et al., 2021). Many of these studies have revealed important household income. These strategies potentially reduce the contact hours empirical setbacks at the micro level that compromised the attainment of children with teachers, causing reduction in school outcomes with of children’s educational outcomes. For instance, Agamile & Lawson subsequent long-term effects on the intergenerational transfer of poverty (2021) demonstrated that farm households’ exposure to rainfall shocks (Kes and Swaminathan, 2006). significantly reduces the attainment of education by children in Uganda. Recent studies on income shocks have examined the impact of Martey et al. (2021) used nationally representative data to show that agricultural and ecological shocks on educational attainment, cognitive food hardship consistently reduces children’s school attendance in skills, and aspirations (Baker et al., 2020; Tabe-Ojong, 2022; Tabe-Ojong Ghana. Glick et al. (2016) highlighted concerns about health and eco- et al., 2021). However, there are limited studies on the combined effects nomic shocks on children’s schooling. Their study finds that a health of pests and weeds on school attendance and children’s farm working shock, measured as parental death or illness, increases the likelihood of hours. A recent study, Tabe-Ojong et al. (2021), demonstrated that dropping out of school among children in Madagascar. Additionally, ecological shocks (pest and weed invasion) reduce households’ aspira- economic shocks in the form of parental unemployment and lack of tion capacity. Using a partial analysis of the edu-workscape framework, assets raises the probability of dropping out in Madagascar. our study investigates the link between ecological shocks (pests and In addition to economic and climate shocks (Masih et al., 2014), weeds) on children’s schooling and working hours, using Ghana as a farmers in developing economies are affected by ecological shocks. case study. Furthermore, the study expands the immediate impact of Ecological shocks include the invasion of pests and weeds that often off- ecological shocks on schooling to include both public and private school set economic and environmental conditions (Mbaabu et. al. 2020; Tabe- attendance. We also quantified the effect of the ecological shocks on Ojong 2022). Pest invasion reduces crop yields and the incomes of children’s on-farm working hours. This design contributes to the farmers (Oliveira et al., 2014) causing losses on field (pre-harvest losses) uniqueness of this study. Our results show that double shocks (pests and and in storage (post-harvest losses) (Oerke, 2006). The indirect effect of weeds) reduced the number of children attending school and increased weed invasion is experienced through competition with crops over re- children’s working hours on the farm. The result remains robust to an sources, serving as host for pests inter alia (Zimdahl, 2018). The invasion alternative estimation method that corrects for endogeneity. of pests and diseases contributes to wide-spread farm income reduction, The rest of this article is structured as follows. Section 2 presents the uncertainty in farming, and farm households’ vulnerability. For farmers Ghanaian context on the environment, school attendance and child work in an imperfect economy where there is a near absence of farm insur- while Section 3 presents the conceptual framework that provides the ance, obvious strategies to adapting to farm shocks can involve both on- basis for establishing the link between shocks, schooling and child work. farm and off-farm diversification (Dimova et al., 2015; Asfaw et al., This is followed by the presentation of data and descriptive statistics in 2019; Bandyopadhyay and Skoufias, 2015). Other studies (Tafere, 2014; Section 4, empirical strategy in Section 5, results and discussion in Rose and Dyer, 2008) have shown that environmental shocks such as Section 6, and conclusions in Section 7. drought or flooding is associated with difficulty in regular school attendance. In contrast, some studies in Ghana (Ananga, 2011), Ethiopia 2. The Ghanaian context - environment, school attendance, and (Boyden et al., 2021), and Madagascar (Moreira et al., 2017) have child work School attendance is a precursor for achieving educational outcomes 6 such as academic performance. The urge to promote children’s educa-The edu-workscape is a triangular matrix which includes domains - work- tion globally and specifically in Ghana has received much attention as places, the school, and the household with the child as the central focus of decision-making through the interaction of the three domains. The matrix manifested in the Sustainable Development Goal Four (SDG 4). The highlights the fact that children’s experiences are shaped by social relations SDG4 focuses on education and aims to “ensure inclusive and equitable within and between home, school and work in specific contexts. The edu- quality education and promote lifelong learning opportunities for all.” workscape foregrounds the dynamic interplay within and between each insti- Specifically, the goal seeks to promote children’s school enrolment and tution and emphasises how the child navigates and experiences this nexus attendance by addressing challenges such as the lack of suitable school (Dunne and Humphreys, 2022). infrastructure, shortage of teachers and teaching materials, and high 2 E. Martey et al. W o r l d D e v e l o p m e n t P e r s p e c t i v e s 31 (2023) 100529 school fees, which some parents cannot afford (Boissiere, 2004; DeJae- tests the hypothesis that ecological shocks (insect pests and invasive ghere et al., 2006). weed) reduce school attendance and increase child farm work. Ghana has long introduced the free compulsory universal primary education (FCUBE) policy which mandates government to make basic 3. Conceptual framework linking ecological shocks, children’s education fee-free and compulsory. The policy also provides books and schooling, and work other learning materials that pupils needed for school. Subsequently, the government introduced free school meals and free school uniforms This section provides the concept underpinning ecological shocks, (Salifu et al., 2018). This policy also sought to improve the quality of children’s schooling, and farm work among farm households (Fig. 3). education through expansion of education infrastructure while building Our conceptual framework is based on the tripartite “edu-workscape” the capacity of more teachers at the basic level (Salifu et al., 2018; framework proposed by Dunne et al., (2021) with slight modification to Akyeampong, 2009). Assessment of the policy show a steady increase in suit agricultural households. The framework provides insight to the primary school attendance for both boys and girls (Akyeampong, 2009). complex household decisions on education and work. We apply this Fig. 1 shows the gross enrolment ratio before and after the imple- framework to establish the link between ecological shocks and chil- mentation of the FCUE. The highest gross enrolment ratio pre-FCUE dren’s schooling and work. The framework shows the “dynamic inter- policy is 80%. The gross enrolment ratio increased to 103% post-FCUE play within and between the institutions and the ways in which the child policy implementation. Generally, gross enrolment ratio has been navigates and experiences the social landscape” (Dunne and Hum- increasing after the policy. Similarly, the number of children attending phreys, 2022; pp.7). In the framework, the child is put in three over- primary school increased after the implementation of the FCUE policy lapping social domains (the household, the school, and the workplace (Fig. 2). (s)) and shows the multiple tensions between these domains. For Despite the significant gains made in terms of reducing child labour7 example, the school domains have a broad and rigid education systems (International Labour Organization [ILO], 2018) and increasing enrol- which do not recognize community constraints (Bourdillon et al., 2015). ment, increasing ecological shocks induced by insect pests and invasive Households are dynamic and belong to the communities where the weed, poses a serious threat to child education by increasing child work. school operate and have multiple and often changing family structure Although the public cost of education has declined, the private cost of (Thorsen and Yeboah, 2021). The workplace exists in several locations schooling is considerably high due to the long travel distance to and as communities, households, and schools. The social domains operate from school (Afoakwah and Koomson, 2021) and costs of supplies. High within an economic, social, temporal and spatial contexts (Dunne and demand for household chores, competing interest of farming and Humphreys, 2022). The interplay between and within the social do- schooling, and household poverty have contributed to increase in child mains are influenced by several external factors such as culture, envi- labour and reduction in school attendance (UNICEF Data Warehouse, ronmental shocks, political forces, and economic policies. 2022). The UNESCO estimate that about 21 percent of children between Within the African context, low school attendance may be due to the ages of 5 to 17 years are involved in child labour (for example, environmental shocks, such as drought or flooding (Rose and Dyer, agriculture, fishing industries, prostituting and trafficking, and street 2008; Tafere, 2014), migration and mobilities (Martey and Armah, work) and 14 percent are engaged in hazardous forms of labour (UNI- 2021; Robson et al., 2006), and poverty and family crises (Thorsen and CEF, 2022; Ortiz-Ospina and Roser, 2016). UNESCO estimates approx- Yeboah, 2021; Hamenoo et al., 2018; Jonah and Abebe, 2019; Bandara imately two million children to be involved in child labour (Ghana et al., 2015). As household income falls below a threshold, children are population, 2022). According to ILO and UNICEF, the global increase in forced to work for the household to meet the basic subsistence re- child labour is partly attributed to the novel COVID-19 pandemic which quirements through alternative income source (Bandara et al., 2015). led to lockdowns, school closures, and economic struggles, thus forcing The type of work undertaken by children largely affects their schooling. children to work (Child Labour: Global Estimates, 2020). For the purpose of our study, we focused on how ecological shocks (pest In a typical agricultural household, male children are mostly infestation and weed invasion) shapes the intercourse between and engaged in one form of an agricultural activity due to high demand for within the three social domains. Given that the households are agrarian labour which creates a heavy domestic workload for girls (UNICEF, and vulnerable to climate shocks (Serdeczny et al., 2017; Mathenge and 2022) and thus reduce school attendance (Boateng and Dako-Gyeke, Tschirley, 2015), pest infestation and weed invasion are likely to restrict 2022). Low use of labour-saving agricultural technologies and contin- farm households from achieving optimal productivity. Ecological shocks uous cropping among agricultural households increase pest infestation widen the gap between attainable and potential crop yields across many and weed invasion which requires high cost of control or abandoning developing nations (Kassie et al., 2018: Tadele, 2017). These shocks farms. In view of this, any agricultural shock may increase labour time contribute to significant crop losses in Africa given the conditions (i.e., for male child or increase labour time for both male and female child in temperature and humidity) that enable these shocks to thrive (Tadele, non-agricultural activities and subsequently reduce school attendance 2017). Reduction in crop productivity due to ecological shocks in Africa and learning outcomes (Boateng and Dako-Gyeke, 2022). The positive ranges between 30 and 60% (Oerke, 2006). effect of ecological shocks on school attendance is more noticeable According to Mathenge and Tschirley (2015), households may among households where children travel long distance to school and participate in off-farm work as a long-term strategy to cope with external incur high extra costs of education due to transportation, feeding, and shocks. Alternatively, households are likely to increase their labour for other supplies. Children in such households are pushed into the labour on-farm activities to reduce the effects of the shocks therefore, reducing force at a younger age to provides support to their parents on the farm labour time for domestic and voluntary activities. However, the allo- and thus reduce school attendance (Boateng and Dako-Gyeke, 2022; cation of time resources or trade-offs among competing tasks largely Hamenoo et al., 2018). Anecdotal evidence suggests that within the depends on household demographics, socioeconomic characteristics, rural agricultural setting, children engaged in farming are viewed as a social and cultural norms, and labor availability (Kes and Swaminathan, positive experience for acquiring valuable skills, cultural identity, and 2006). The structure of the households is likely to influence labor sense of belonging within the community (Krauss, 2013). This study availability among competing tasks. The extent to which males and fe- males are deployed for farm work is influenced by labor availability and social and cultural norms. The redistribution of labor among competing 7 Child labour is defined as the employment of children that violates state, tasks directly affects crop productivity. Male children are more likely to federal, or international laws because of the type of work performed or the age spend long hours in paid work while female children undertake more of the child involved (Merriam-Webster, s.v. “Child Labor,” Accessed July 1, domestic unpaid work. Children undertake paid work to complement 2022, https://www.merriam-webster.com/dictionary/child labor.). household income and support schooling. Similarly, children may work 3 E. Martey et al. W o r l d D e v e l o p m e n t P e r s p e c t i v e s 31 (2023) 100529 Fig. 1. Primary Gross Enrolment Ratio in Ghana, 1971–2020 Source: World Data Bank, 2023 (https://data.worldbank.org/indicator/SE.PRM.ENRL?locations=GH). Fig. 2. Primary education, pupils in Ghana, 1971–2020 Source: World Data Bank, 2023 (https://data.worldbank.org/indicator/SE.PRM.ENRL?locations=GH). long hours on-farm to control the spread of the shocks. The long hours in sand for school construction project) which may be hazardous and thus paid and unpaid work may likely affect school attendance. reduce learning outcomes (Dunne and Humphreys, 2022). The com- The intersectoral trade-off is particularly severe among income-poor bined effects of these arduous physical work in school and the burden of households with few assets and less available labor, which may affect domestic and farm works can reduce school attendance and learning food security, child nutrition, health, and education (Kes and Swami- outcomes. Our study focuses on a partial analysis of the framework and nathan 2006). To minimize crop losses, household labor may be shifted test two main hypotheses: (1) ecological shocks reduce children’s school towards agricultural production. It is typical among farming households attendance and (2) ecological shocks increase children’s workload. that children have limited or no decision-making power when it comes to the use of their time or labor; such decisions are taken by parents or 4. Data and descriptive statistics adult household members. Thus, children may be directed to carry out farm work, which reduces their school attendance or hours spent in 4.1. Data school. However, within the framework, the school domain can induce child labour effect. For example, children in rural schools are likely to We used the seventh round of the Ghana Living Standards Survey engage in hard physical work (toiling on school farm, fetching water and (GLSS), a nationally representative household-level dataset that assesses 4 E. Martey et al. W o r l d D e v e l o p m e n t P e r s p e c t i v e s 31 (2023) 100529 Fig. 3. Conceptual model of ecological shock, schooling, and child work Source: Adapted from Dunne and Humphreys, (2022). the living standards of Ghanaians. The data was collected between 22 several pests with considerable crop loss (Obeng-Ofori & Amiteye, October 2016 and 17 October 2017. Details of the sampling approach 2005). Weevils are common pests that attack crops both on the field and are adequately described in the GLSS 7 main report (Ghana Statisitical in storage with an estimated damage of 8–15% (Williams 2010). Our Service [GSS], 2018). Our analysis focused on farm households in the study focuses on weevil and FAW infestation. Northern, Upper East and Upper West Regions, which together consti- The data shows that farmers who experienced only pest invasion and tute northern Ghana.8 The total sample used for the analysis is 3,251 only weed invasion spend about GHS 123 (USD 29)9 and GHS 120 (USD households. 28) annually to control the shocks, respectively. Farmers who experi- Our study focuses on the section of the data that captures household enced both pest and weed invasion spend GHS 204 (USD 48) annually to demographics, socioeconomics, assests, child education, child farm control the shocks. work, as well as institutional and geographic controls. Finally, we computed ecological shocks based on incidence of weed or pest invasion 4.1.2. Children’s school attendance and farm work on farmers’ farms within the year preceding data collection. In this study, we consider children below 15 years of age to be at the basic school level. School attendance is measured as the number of 4.1.1. Ecological shocks children in a household attending school (private and public), which is a Our primary measure of ecological shocks is insect pest infestation form of human capital accumulation (Martey et al., 2021). We separated and invasion of a parasitic weed. The most dominant invasive weed in private and public school as they reflect differences in the resource northern Ghana is Striga hermonthica (witchweed): a parasitic weed that outlays of parents. Children’s farm working hours are computed as the thrives in moisture and limited fertilizer/organic matter content as well number of hours children spend working on a farm either alone or as when when poor crop management practicies are undertaken. Maize together with household members. grown in Striga-prevalent soils has a yield loss potential of up to 100% when improperly managed or controlled (David et al, 2022; Teka, 2014; 4.2. Descriptive statistics Atera et al., 2013). Monocropping, relative to rotation and intercropping systems, increases the prevalence of Striga in smallholder farming sys- Fig. 4 shows the relationship between ecological shocks and working tems (Badu-Apraku & Fakorede 2017). The farming system in northern hours. Consistent with the conceptual framework, pest and weed shocks Ghana is characterized by continuous cropping with little or no fallow: a are positively associated with working hours. Farmers who experienced practice that increases Striga invasion in most soils, leading to farmers both weed and pest invasion worked the most hours followed by farmers abandoning their fields. who experienced only weed and only pest invasion. Fig. 5 highlights the Pests are biotic stressors that increase crop losses (Oerke, 2006). Pest relationship between shocks and school attendance and child working attacks occur at different stages of crop growth (stem, leaves, flowers) hours. With reference to child working hours, farmers who experienced with varying levels of economic damage (Day et al., 2017). Studies have both pest and weed invasion recorded the highest child working hours shown that a fall armyworm (FAW) attack can lead to 100% destruction followed by farmers who experienced only weed invasion and only pest of crops (Kasoma et al., 2021). In Ethiopia and Ghana, FAW may lead to invasion. However, the relationship between shocks and school atten- 30% and 45% yield reduction, respectively (Assefa & Ayalew, 2019; Day dance is mixed. Farmers who experienced both weed and pest invasion et al., 2017). The climate of Ghana provides favorable conditions for recorded the highest school attendance followed by farmers who expe- rienced only pest invasion and only weed invasion. Compared with 8 Northern Ghana is now made up of Northern, North East, Savannah, Upper East and Upper West Regions. The GLSS 7 data was captured before the 9 The Bank of Ghana exchange rate at the time of the survey (2017) was GHS administrative restructuring of the regions from three to five. 4.271 to USD 1 at the end of January 2017. 5 E. Martey et al. W o r l d D e v e l o p m e n t P e r s p e c t i v e s 31 (2023) 100529 Table 1 Ecological shock packages. Ecological shocks Choice Binary double Pest Weed Frequency (j) package (%) P1 P0 W0 W1 1 P0W0 √ √ 46.05 2 P1W0 √ √ 8.92 3 P0W1 √ √ 28.55 4 P1W1 √ √ 16.49 Notes: The binary double package represents the possible ecological shock combinations. Each element in the shock package is a binary variable for ecological shocks (pests (P) and weeds (W)). The subscript “1″ represents experience of shock and 0 if otherwise. communities, approximately 9% had access to a credit facility, and about 69% of the farmers reported having access to a motorable road. Fig. 4. Committed time by ecological shocks. Additionally, farmers have an average of about 26 neighbors within their network who have access to agricultural extension services, while only 6% of farmers had access to irrigation services. Also, about 41 neighbors of farmers experienced some form of ecological shock on their farms during the time of the study; and the majority (89.5%) of farmers are members of rural farming households. Regarding regional distribu- tion, 31.8%, 34.4%, and 33.9% of the sampled farmers reside in the Northern, Upper East and Upper West Regions, respectively. 5. Empirical strategy The study employed the multinomial endogenous switching regres- sion (MESR) to model the effect of ecological shocks on children’s school attendance and farm work. A farmer in our sample may have experi- enced a single or a combination of ecological shocks leading to four possible outcomes – no shock (P0W0), pest invasion (P1W0), weed in- vasion (P0W1), and a combination of pest and weed invasion (P1W1). Under the assumption of exogeneity of the ecological shocks, the ordi- nary least squares (OLS) method can be applied to generate a consistent estimate of the effect of shocks on the outcome variables. However, farm Fig. 5. School attendance and farm work by shocks Source: Authors’ construct management practices may either reduce or increase the spread of the based on GLSS 7 data. shocks. The management practices of farmers may be influenced by both observable and non-observable factors, which may be correlated with farmers who did not experience any of the ecological shocks, farmers the error term of the outcome model. In such case, claim of strict exo- who experienced pest invasion recorded a decline in female child school geneity in ecological shock is not plausible, thus, leading to a biased attendance. estimate of children’s school attendance and working hours. Second, we Table 1 shows the proportion of ecological shocks experienced by suspect a reverse causality between the shocks and the outcome vari- sample farmers in Northern Ghana. The ecological shocks considered in ables. Ecological shocks may influence schooling and working hours. this study include weed and pest invasion, which leads to four possible Alternatively, a reduction in farm working hours and increase in school combinations of ecological shocks. For the total sample of 3,251 attendance may reduce labor for farm work, which may contribute to farmers, about 46% of the farmers did not experience any of the shocks, the spread of the shocks. while 9% and 29% of the farmers experienced pest only and weed only, The study addresses the endogeneity and selection bias issues by respectively. About 16% of the farmers experienced both pest and weed employing the multinomial endogenous switching regression (MESR) infestation. treatment effect based on the Dubin and McFadden (1984) and Bour- Table 2 shows the descriptive statistics of the explanatory variables. guignon et al. (2007) approaches. The MESR allows for the evaluation of About 18.3% of the sampled farming households in Northern Ghana are alternative combinations and individual shocks. The MESR corrects for headed by females. The average age of farmers is 48 years with 2.39 the selection bias by computing an inverse Mills ratio (IMR) based on the years of formal education. The average number of years of education theory of truncated normal distribution (Malikov and Kumbhakar, 2014; indicates that the sampled farmers have had minimal primary educa- Bourguignon et al., 2007). Second, the MESR allows for the construction tion. Regarding household characteristics, a farming household contains of a counterfactual based on returns to the characteristics of farmers approximately six members. About 79% of household heads are mar- with experienced shocks and those who did not experience any shock ried. On average, about 70.3% of the sample is employed and about (Kassie et al., 2018). Third, the MESR allows for interaction between 8.4% engage in some type of off-farm employment. A farming household choices of alternative shocks (Wu and Babcock, 1998). in the study area owns about six livestock. Regarding access to institu- The MESR involves a two-stage simultaneous estimation technique tional facilities, 32.4% of farmers had extension officers visiting their where the first stage models farmers’ experience of shock using a multinomial logit selection (MNLS) model and accounts for unobserved 6 E. Martey et al. W o r l d D e v e l o p m e n t P e r s p e c t i v e s 31 (2023) 100529 heterogeneity. The second stage is the outcome equation estimated with where ρ is the correlation between εs and μs, and ωs are error terms with the OLS. The IMR computed from the first stage is included as an an expected value of 0. Based on the multinomial choice setting, there additional variable to account for selection bias from time-varying, are J − 1 selection correction terms, one for each alternative package of unobserved heterogeneity. shocks. Following previous studies and drawing from the peer effect litera- 5.1. First stage – Multinomial selection model ture (Verkaart et al., 2017; Tessema et al., 2016; Magnan et al., 2015; Ward and Pede, 2015; Krishnan and Patnam, 2014; Wollni and Ander- As stated previously, farmer i experiences a shock package j, over an sson, 2014; Conley and Udry, 2010), we computed the leave-out-mean alternative shock package, k, that influences outcomes (school atten- of farmers who have experienced shocks as a potential instrument for dance and working hours) such that S > S k ∕= j. The expected ecological shocks. Based on the relevance condition, we propose that ij ik outcome S* derived by the farmer from experiencing shock package j is a farmers are more likely to be influenced by weeds and pests if neighbors ij within the same community experience weed and pest invasion. The latent variable determined by observed household, socio-demographic, credibility of the instrument also relies on the fact that the shocks are institutional and location characteristics (Xi), and unobserved charac- invasive so the possibility of spreading across farms is highly possible. teristics (εij): We conducted a falsification test following the approach of Di Falco S* = X β + ε (2) et al. (2011). The results confirm that the excluded variable has a sig-ij i j ij nificant effect on weed and pest invasion (see Table A2 in the Online Following Khonje et al., (2018) and assuming G is an index that Appendix) but not the outcome variables (see Tables A3-A6 in the On- characterizes farmer’s choice of package, the index can be expressed as: line Appendix). ⎧ ⎪ ⎪ (1 iff S* max S* ) or π 0 5.2. Estimation of average treatment effects ⎪ i1 > ik i1 < ⎪ k∕=j ⎪ ⎪ . ⎪ The average treatment effect is estimated by comparing the expected ⎪ . ⎨ outcomes of farmers who experienced shock and those without shocks in Ω = . ∀ k =∕ j actual and counterfactual scenarios. The actual expected outcomes of ⎪ ⎪ . farmers who experienced shocks are expressed as: ⎪ ⎪ ⎪ . ⎪ ( ) ⎪ ⎪ J iff S * > max S*ij ik or πij < 0 ⎩ k∕=j (3) Table 2 where πij = max(S*ik − S*ij) < 0(Bourguignon et al., 2007). Definitions and summary statistics of the variables used in the analysis. k∕=j Assuming that the household-specific heterogeneity or idiosyncratic Full sample error are independent and identically Gumbel-distributed across all Variable Description Mean SD ecological shock choice sets (Bourguignon et al., 2007), the probability Socioeconomics (P ) that a farmer i will experience shock j can be expressed as: Gender 1 if farmer is female, 0 if otherwise 0.183 0.387 ji Age Age of farmer (years) 48.084 16.185 ( ) ( ) exp X β Education Number of years of schooling 2.386 4.335 Pij = Pr π i j ij< 0|Xi = ∑J (4) (years) k=1exp(Xiβk) Household size Number of household members 5.902 3.404 (number) Equation (4) is the multinomial logit model (Mc-Fadden, 1973) Marital status 1 if currently married, 0 if 0.788 0.409 estimated by maximum likelihood. otherwise The second stage of the MESR shows the relationship between the Employment status 1 if employed, 0 if otherwise 0.703 0.457 outcome equation for each possible regime j (excluding farmers who did Off-farm work 1 if engaged in off-farm work, 0 if 0.084 0.277 not experience any shock as a base category) with selection bias otherwise Household assets correction term: Livestock Number of livestock owned 5.823 11.052 ⎧ Institutional factors ⎪ ⎪Regime 1 : Yi1 = Ziϑ1 + σ1 λ̂1 + ωi1 if P = 1 Extension visits 1 if extension visit community, 0 if 0.324 0.468 ⎪ ⎪ . otherwise ⎪ ⎪ Access to credit 1 if access to credit, 0 if otherwise 0.088 0.284 ⎪ . Access to motorable 1 if access to motorable road, 0 if 0.690 0.463 ⎨ . j 2 3 4 (5) road otherwise = , , ⎪ Network of extension Number of neighbors with 26.223 20.588 ⎪ ⎪ . access extension access ⎪ ⎪ Access to irrigation 1 if access to irrigation, 0 if 0.059 0.236 ⎪ . ⎪ otherwise ⎩ Regime J : YiJ = Ziϑj + σj λ̂ ω if P J Instruments j + ij = Neighbors Number of neighbors who 41.076 21.037 experiencing shocks experience ecological shock where Yjit represents outcomes associated with the selected regime j(j = Geographic controls 0,⋯., J) and observed if only one of possible shocks is experienced, Zi Urban location 1 if residing in urban areas, 0 if 0.105 0.306 represents a vector of explanatory variables, σ is the covariance between otherwise Northern 1 if residing in Northern Region, 0 if 0.318 0.466 εs (first stage) and μs (second stage), λ̂j is the IMR calculated from otherwise estimated probabilities in Equation (4) as: Upper East 1 if residing in Upper East, 0 if 0.344 0.475 otherwise [ ] ∑J p̂ikln(p̂ik) ( ) Upper West 1 if residing in Upper West, 0 if 0.339 0.473 λ̂j = ρj + ln p̂1 − p̂ ij (6) otherwise k∕=j ik Notes: SD is standard deviation. 7 E. Martey et al. W o r l d D e v e l o p m e n t P e r s p e c t i v e s 31 (2023) 100529 ⎧ ⎪E(Yi2|G = 2) = Z ϑ + σ λ̂ due to constraints on land access and lack the financial resources ⎪ i 2 2 2 ⎪ necessary to invest in labor-saving technologies such as pesticides ⎪ . ⎪ (Peterman et al., 2014). Age positively influences the invasion of pests ⎪ ⎪ . ⎨ but is negatively associated with weed invasion (P0W1), suggesting that . a unit increase in the age of a farmer will lead to a 0.1% increase in P1W0 ⎪ ⎪ and a 0.1% decrease in P0W1. The results suggests that older household ⎪ . ⎪ heads are more likely to experience P1W0 but less likely to experience ⎪ ⎪ . P0W1. Older household heads are more likely to be supported by ⎪ ⎩ ( )E Y |G = J = Z ϑ + σ λ̂ household members in the control of weeds but may lack the financial ij i 1 1 j (7) resources to invest in pesticides. An increase in formal education is associated with a 0.3% increase in P1W0 and a 0.3% decrease in P0W1. The expected outcomes of adopters had they not experienced shock The result is consistent with Asfaw et al. (2019) who find that educated (counterfactual) is specified as: household heads tend to explore off-farm income generating activities ⎧ with less commitment to farm activities, thus, leading to the high chance ⎪E(Yi1|G = 2) = Ziϑ1 + σ1 λ̂⎪ 2 of pest invasion on crop farms. Alternatively, education leads to ⎪ ⎪ . ⎪ improved cognitive reasoning (Li et al., 2020; Khonje et al., 2018; ⎪ ⎪ . Dahmann, 2017) that promotes the adoption of enhanced crop pro- ⎨ duction strategies, which has the potential to reduce weed invasions. . ⎪ The likelihood of weed invasion (P0W1) and double invasion of pests ⎪ ⎪ . and weeds (P1W1) increases with household size. A unit increase in ⎪ ⎪ ⎪ . household size is associated with 0.5% increase in P0W1 and P1W1. Large ⎪ ⎩ household size denotes the existence of labor availability for farm work E(Yi1|G = J) = Ziϑ1 + σ1 λ̂j that encourages the cultivation of large areas. The cultivation of large (8) areas exposes farmers to the risk of pest and weed invasion. Farmers that The average treatment effect on the treated (ATT10) is computed as the difference between Equations (7) and (8): Table 3 Marginal effects of ecological shock packages. ATT = E(Yi2|G = 2) − E(Yi1|G = 2) = Zi(ϑ2 − ϑ1)+ λ̂2(σ2 − σ1) (9) P1W0 P0W1 P1W1 The first term on the right-hand side of Equation (9) (ϑ2 − ϑ1) cap- Variables Marginal Marginal Marginal tures the expected change in the mean outcome due to the differences in effect effect effect coefficients of the observed characteristics. The second term (σ − σ ) (Std. Error) (Std. Error) (Std. Error) 2 1 corrects selection bias (Khonje et al., 2018). Gender 0.043*** − 0.007 − 0.035 (0.017) (0.027) (0.025) Age 0.001** − 0.001*** − 0.001 5.3. Lewbel 2SLS regression (0.000) (0.001) (0.000) Education 0.003*** − 0.003* 0.002 (0.001) (0.002) (0.002) We complemented the MESR with Lewbel 2SLS (Lewbel, 2012) to Household size 0.002 0.005** 0.005*** check for the robustness of our estimates of children’s school attendance (0.002) (0.002) (0.002) and working hours. The Lewbel 2SLS is useful when valid external in- Marital status 0.054*** − 0.006 0.028 struments are unavailable or considered potentially weak. This method (0.017) (0.025) (0.023) Employment status − 0.001 0.071*** 0.078*** exploits heteroskedasticity in the data to generate internal instruments (0.011) (0.018) (0.017) that are used to address endogeneity. Off-farm work 0.038** − 0.035 − 0.049* (0.017) (0.030) (0.026) 6. Results and discussion Livestock 0.001 ** 0.003*** 0.002*** (0.000) (0.001) (0.001) Extension visits 0.011 0.017 0.031** 6.1. Determinants of pest and weed invasion (0.012) (0.017) (0.015) Access to credit − 0.027 0.058** − 0.025 Table 3 shows the marginal effects from the multinomial logit esti- (0.019) (0.026) (0.023) ** *** mates of ecological shocks. Farmers who did not experience any Access to motorable road − 0.024 − 0.075 0.034 ** (0.012) (0.019) (0.017) ecological shock constituted the base category. The Wald test shows that Network of extension access 0.001*** 0.001* − 0.004*** the explanatory variables included in the selection model provide a good (0.000) (0.000) (0.000) explanation of pest and weed invasion. The number of neighbors expe- Access to irrigation − 0.082** 0.097*** − 0.101*** riencing shocks is significantly associated with ecological shocks. (0.035) (0.033) (0.037) Neighbors experiencing shocks − 0.001*** 0.005*** 0.002*** Consistent with the a priori expectation, the number of neighbors (0.000) (0.000) (0.000) experiencing shocks increases the probability of a farmer experiencing Urban location − 0.021 − 0.039 − 0.012 weed invasion and experiencing both weed and pest invasion. However, (0.017) (0.026) (0.021) we observed a negative effect on pest invasion. Region FE Yes Yes Yes The gender of the household head is positively associated with the likelihood of experiencing pest invasion (P1W0). The estimates indicated Joint significance of 275.16*** instruments χ2 that, relative to male farmers, females are 4.3% more likely to experi- (6) 2 ence pest invasion. Female farmers often cultivate relatively small farms Wald χ (54) 882.94*** Observations 3,251 3,251 3,251 10 The ATT is computed based on the post-estimation prediction of the actual Notes: The reference category is no shock (P0W0). P1W0– only pest shock; P0W1– and counterfactual expected value of the outcomes for a household that expe- only weed shock; P1W1– pest and weed shocks. *** Significant at the 1% level, ** rienced shock j after estimating the MESR in Equation (6). significant at the 5% level, and * significant at the 10% level. 8 E. Martey et al. W o r l d D e v e l o p m e n t P e r s p e c t i v e s 31 (2023) 100529 are gainfully employed are 7.1% and 7.8% more likely to experience distributions. The results of the study show that on average, farmers who P0W1 and P1W1, respectively. Employment outside the farm shifts experienced shocks realize lower school attendance (public and private) farmers’ committed time on the farm to the labor market, thereby, and children’s farm working hours for all the shocks than farmers who increasing the likelihood of weed and pest shocks. Similarly, house- did not experience any shock. However, the results are only indicative of holds’ engagement in off-farm activities is positively associated with the effects of ecological shocks. The results could be misleading because P0W1 but negatively with P1W1. The result indicates that engagement in they do not account for observed and unobserved factors (selection bias) off-farm activities increases the likelihood of P0W1 by 3.8% but de- that may influence the outcome variables. creases P1W1 by approximately 5%. Ownership of livestock appears to Table 4 presents the conditional average effects of the ecological compete with investments in crop farms, thus, increasing the risk of pest shocks. The estimation of the impact of pest and weed invasion under and weed invasion; whereas a unit increase in household total livestock conditional and unconditional average effects is based on the predicted increases the likelihood of P1W0, P0W1, and P1W1 invasion by 0.1%, outcomes from the MESR. Consistent across all the packages of shocks, 0.3% and 0.2% respectively. we observed a decline in the number of children attending school and Relative to farmers who did not receive extension service, farmers public-school attendance. Farmers who experienced only weed invasion who received extension services are 3.1% more likely to experience recorded the highest decline in the number of children attending school P1W1. The increasing effect of extension service on pest and weed in- (by 0.88) followed by farmers who experienced both pest and weed vasion can be attributed to lack of understanding of extension service invasion (0.75), and those who experienced only pest invasion (0.43). advice and recommendations, non-application of extension advice, and Regarding public school attendance, our results show that farmers inadequate extension visits. Households’ access to credit increases the whose farm plots were invaded decreased their number of children likelihood of P0W1 by 5.8%. Farm credit, often in the form of cash, may attending public school. Comparing the results across the shocks, we be diverted from its intended purpose of acquiring farm inputs to other find that farmers who have experienced double invasion of pests and non-farm activities, thereby, increasing the risk of P0W1. The findings weeds recorded the highest decline in the number of children attending confirm the result of Tabe-Ojong (2022) who finds a positive association public school (by 0.74) followed by farmers who only experienced weed between households’ credit access and ecological shocks. Access to invasion (0.68) and only pest invasion (0.44). In reference to private motorable roads increases households’ likelihood of experiencing P1W0 school attendance, farmers whose fields were invaded by pests recorded and P0W1 invasion but decreases P1W1. Relative to households with an increase in the number of children attending private school (by 0.02). impassable roads, households with access to motorable roads are 2.4% The result is contrary to the a priori expectation that ecological shock and 7.5% more likely to experience P1W0 and P0W1, respectively. Access reduces farm household income and subsequently reduces the ability of to motorable roads enables farmers to access farm inputs that minimize households to accommodate the high cost of private school attendance. their exposure to ecological shocks. In the specific case of double inva- However, weed invasion and double invasion of weeds and pests reduce sion, access to motorable roads had a positive association with P1W1 the number of children attending private school by 0.03 and 0.04, indicating that households that have access to motorable roads are 3.4% respectively. Consistent with the results on public school attendance, more likely to experience both pest and weed invasion. Hu et al. (2019) double invasion of pests and weeds had the highest effect on private explained that good roads increase farmers’ likelihood of working off- school attendance. The result corroborates the findings of Martey et al. farm, thereby, neglecting agricultural production and leading to expo- sure to ecological shocks. Table 4 Network of extension access is associated with a 0.1% increase in the The average effect of ecological shock package using multinomial ESR. likelihood of experiencing P1W0 and P0W1 invasion. On the contrary, Shock status Average network of extension access decreases both pest and weed invasion by Outcomes Shock Shock No shock Treatment 0.4%. The result projects the importance of peer effects and social choice (j = 1,2, (j = 0) Effect learning in terms of adoption of agricultural technologies and risk- (j) 3) mitigation strategies. Access to irrigation has varying effects on P1W0, Children attending school P1W0 3.39 3.82 − 0.43*** P0W1, and P1W1. The likelihood of P1W0 and P1W1 decreased by 8.2% (number) (0.14) (0.25) (0.12) and 10.1%, respectively, for households with access to irrigation facil- P0W1 3.23 4.10 − 0.88*** ities. However, for P0W1, access to irrigation facilities provides a (0.08) (0.23) (0.15) P1W1 3.64 4.39 − 0.75*** conducive environment for the growth and spread of weeds increasing (0.14) (0.28) (0.15) their likelihood of farm invasion. The result demonstrates that access to irrigation facilities increases farmers’ likelihood of experiencing weed Public school attendance P1W0 2.99 3.43 − 0.44*** invasion by 9.7%. Farming households whose neighbors experienced (0.11) (0.21) (0.12) shocks are 0.1% less likely to experience P1W0. Although marginal, the P0W1 2.96 3.65 − 0.68*** negative effect can be explained by the fact that the invasion of pests in (0.07) (0.19) (0.13) neighbors’ farms compelled neighbors to adopt control measures to P1W1 3.16 3.89 − 0.74*** (0.10) (0.23) (0.14) prevent the spread of pests. On the contrary, farming households whose neighbors experienced shocks are more likely to experience weeds and both pest and weed invasions. The positive result is consistent with the a Private school attendance P1W0 1.19 1.17 0.02** (0.01) (0.01) (0.01) priori expectation that ecological shocks tend to spread within P0W1 1.14 1.17 − 0.03*** geographical areas. (0.01) (0.01) (0.01) P1W1 1.13 1.17 − 0.04*** 6.2. Average adoption effects for a combination of ecological shocks (0.01) (0.01) (0.01) This section highlights the effects of ecological shocks on the number Child farm work P1W0 1.08 1.16 − 0.08*** (0.01) (0.01) (0.01) of children attending school, public school, private school, and chil- P0W1 1.28 1,24 0.04*** dren’s farm work under actual and counterfactual conditions after (0.01) (0.01) (0.01) controlling for selection bias. The second stage results are presented in P1W1 1.27 1.23 0.05*** the Appendix of the supplementary material. Table A2 in the supple- (0.01) (0.01) (0.01) mentary material reports the unconditional average effects of ecological Notes: P1W0– only pest shock; P0W1– only weed shock; P1W1– pest and weed shocks on outcome variables derived from the actual and counterfactual shocks. ***Significant at the 1% level. 9 E. Martey et al. W o r l d D e v e l o p m e n t P e r s p e c t i v e s 31 (2023) 100529 (2021). The results of their study show that food hardship reduces the number of children attending school and private school attendance. In another study by Martey et al. (2021), they found that time poverty reduces the number of children attending public school. A similar study on shocks by Agamile and Lawson (2021) in Uganda shows that expo- sure to negative rainfall shocks significantly reduces children’s school attendance by almost 10%. Using longitudinal rural household survey data in Ethiopia, Randell and Gray (2016), find that greater summer rainfall is associated with school completion and attendance at the time of the survey. The results further show a positive relationship between shocks and children’s working hours except for farmers who have experienced pest invasion on their farm plots. As indicated by the results, pest invasion reduces children’s farm work by 0.08 h. In contrast, weed invasion in- creases children’s farm work by 0.04 h, while farmers who experienced both pest and weed invasion increased children’s farm work by 0.05 h. The negative effect of pest invasion on children’s farm work may be because children play a reduced role in terms of pesticide application compared with the youth category of the household. The structure of the household, as highlighted by the conceptual framework, may be ac- counting for farm-committed time allocation among household mem- bers. The positive effect of weed and pest invasion on children’s farm work indicates that weed invasion may be driving the increase in chil- dren’s time spent working on the farm given that weed invasion had a higher school attendance reduction effect than pest invasion. Studies have shown that weed invasion (Striga hermonthica) can reduce yield by up to 100% (Yacoubou et al., 2021), so the household heads are more likely to mobilize household members to participate in on-farm activ- ities to curtail the spread of the weed. The implication of the results is that if interventions that seek to reduce the amount of time children do farm work are not implemented, then the education and labor market outcomes of children in the future will be sacrificed. This situation further leads to intergenerational transmission of poverty (Kes and Swaminathan 2006). We illustrate the effects of ecological shocks (pest and weed inva- sion) on children’s school attendance and farm working hours using the kernel densities of the predicted school attendance and child farm work distributions by shock status (Figure A1 in the supplementary mate- rials). The result is more informative than observed schooling and child work since the values are estimated after controlling for observed and unobserved factors. 6.3. Robustness check – Lewbel 2SLS estimation To test the robustness of our estimates, we perform a sensitivity test using the Lewbel (2012) 2SLS method to examine the sensitivity of our MESR estimation of shocks on child schooling and farm work (Table 5). Comparatively, the magnitude of the effect is relatively higher for the Lewbel 2SLS than the MESR. The first stage results show that the number of neighbors who have experienced ecological shocks is highly corre- lated with farmers experiencing ecological shocks. Consistent with the MESR estimate, the Lewbel 2SLS estimation shows that pest invasion reduces the number of children attending public school and the working hours on the farm by 0.14 and 0.06 h, respectively. Weed invasion in- creases child farm working hours by 0.176 h while double ecological shocks reduce the number of children attending school and public- school attendance by 0.11 and 0.09, respectively, and increased child farm working hours by 0.13 h. The results of the Lewbel 2SLS shows the robustness of our estimates on the effect of shocks on children’s school attendance and farm work. The implication of the findings is that children’s school attendance can be improved with investment in labor-saving technologies or the sub- sidization of chemicals for pest and weed control. 10 Table 5 Lewbel 2SLS estimates for ecological shocks and child outcomes. Schooling Public Private Child farm work Outcomes (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) P1W0 P0W1 P1W1 P1W0 P0W1 P1W1 P1W0 P0W1 P1W1 P1W0 P0W1 P1W1 Pest − 0.061 − 0.139* 0.076 − 0.057* (0.070) (0.077) (0.050) (0.034) Weed − 0.058 − 0.010 − 0.046 0.176*** (0.052) (0.055) (0.039) (0.045) Pest and weed − 0.111** − 0.092* − 0.042 0.129*** (0.049) (0.051) (0.034) (0.039) All controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Region FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes First stage Neighbors shocks 0.001*** 0.007*** 0.004*** 0.001*** 0.007*** 0.004*** 0.001*** 0.007*** 0.004*** 0.001*** 0.007*** 0.004*** (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) F-Statistic 13.77 26.70 44.62 13.77 26.70 44.62 13.77 26.70 44.62 13.77 26.70 44.62 J p-value 0.435 0.164 0.203 0.786 0.571 0.605 0.518 0.060 0.880 0.314 6.7e-07 0.089 Observations 1,787 2,425 2,033 1,787 2,425 2,033 1,787 2,425 2,033 1,787 2,425 2,033 R-squared 0.542 0.510 0.558 0.484 0.473 0.515 0.118 0.119 0.115 0.065 0.064 0.068 Notes: P1W0– only pest shock; P0W1– only weed shock; P1W1– pest and weed shocks. ***Significant at the 1% level, ** significant at the 5% level, and *significant at the 10% level. E. Martey et al. W o r l d D e v e l o p m e n t P e r s p e c t i v e s 31 (2023) 100529 7. Conclusion Data availability Several studies have examined the impact of agricultural shocks on Data will be made available on request. welfare outcomes. However, there is a dearth of studies examining the effect of ecological shocks on children’s school attendance and farm Acknowledgement working hours. This study fills the knowledge gap by employing a partial analysis of the edu-workscape framework using data from the seventh This study is made possible through data support from the Ghana round of the Ghana Living Standards Survey, a nationally representative Statistical Service (SSS). The authors are grateful for financial support data set. We employed the multinomial endogenous switching regres- received by the GSS from the Government of Ghana, Department for sion (MESR) to establish the link and used the Lewbel 2SLS estimation International Development (DFID) and the Dutch Government through technique as a robustness check. the International Development Assistance (IDA) and the Statistics for The results from the multinomial logit selection model indicate that Results Facility Catalytic Fund (SRF-CF) which was managed by the the likelihood of experiencing an ecological shock is significantly World Bank under the Ghana Statistical Development Program (GSDP). influenced by sex, age, household size, marital status, employment The authors express their appreciation for the administrative support status, off-farm work, number of livestock owned, extension visits, credit offered by the CSIR-Savanna Agricultural Research Institute. Finally, we access, access to motorable road, network of extension access, irrigation wish to express our profound gratitude to the anonymous reviewers access, and number of neighbors experiencing shocks. The findings from whose contribution has led to a major improvement of this manuscript. this study provide the basis for formulating policies to guide develop- ment practitioners to address the social factors accounting for pest and Appendix A. Supplementary data weed invasion. The significant role of peer effect suggests the need to build the capacity of farmers on effective pest and weed control to Supplementary data to this article can be found online at https://doi. reduce their spread. org/10.1016/j.wdp.2023.100529. The results of MESR show that double invasion (pests and weeds) reduced the number of children attending school (both public and pri- vate) and increased children’s on-farm working hours. Farmers who References experienced only pest invasion witnessed a decline in the number of Adonteng-Kissi, O. (2018). Parental perceptions of child labour and human rights: A children attending public schools, but an increase in the number of comparative study of rural and urban Ghana. Child Abuse & Neglect, 84, 34–44. children attending private schools. Children’s working hours on the Agamile, P., & Lawson, D. (2021). Rainfall shocks and children’s school attendance: farm declined for farm households that experienced only pest invasion. Evidence from Uganda. Oxford Development Studies, 49(3), 291–309. Weed invasion reduced the number of children attending school (public Ananga, E. (2011). Dropping Out of School in Southern Ghana: The Push-Out and Pull-Out Factors. Research Monograph No 55: CREATE Pathways to Access. and private) but increased children’s on-farm working hours. Compar- Afoakwah, C., & Koomson, I. (2021). How does school travel time impact children’s atively, double ecological shocks had the most negative effect on public learning outcomes in a developing country? Review of Economics of the Household, and private school attendance and the highest positive effect on chil- 1–21. Akyeampong, K. (2009). Revisiting free compulsory universal basic education (FCUBE) dren’s working hours on the farm. in Ghana. Comparative Education, 45(2), 175–195. The findings of the study fits within the broader edu-workscape Assefa, F., & Ayalew, D. (2019). Status and control measures of fall armyworm framework and have three main implications. First, the negative effect (Spodoptera frugiperda) infestations in maize fields in Ethiopia: A review. Cogent Food & Agriculture, 5(1), 1641902. of ecological shocks on children’s school attendance implies that chil- Asfaw, S., Scognamillo, A., Di Caprera, G., Sitko, N., & Ignaciuk, A. (2019). dren’s human capital formation is likely to be sacrificed in the presence Heterogeneous impact of livelihood diversification on household welfare: Cross- of ecological shocks. This may impact children’s school performance country evidence from Sub-Saharan Africa. World Development, 117, 278–295. Atera, E. A., Ishii, T., Onyango, J. C., Itoh, K., & Azuma, T. (2013). Striga infestation in and future labor outcomes. Given that the cost of hired labor is relatively Kenya: status, distribution and management options. high, household heads may principally rely on family labor for most Badu-Apraku, B., & Fakorede, M. A. B. (2017). Maize in Sub-Saharan Africa: Importance agricultural activities. To avert this situation, there is the need to pro- and production constraints. In Advances in genetic enhancement of early and extra-early maize for Sub-Saharan Africa (pp. 3–10). Cham: Springer. mote labor-saving technologies to reduce human drudgery and channel Bakari, S. (2013). Making gender sense on schools: Nigeria. Unpublished report (2nd households’ saved time into other productive activities. Second, an in- edition) based on the Commonwealth study for E. Page & J. Jha (2009) Exploring the crease in children’s working hours on the farm implies a trade-off be- bias: gender and stereotyping in schools. London: Commonwealth Secretariat. tween present gains and intergenerational transfer of poverty. Baker, R. B., Blanchette, J., & Eriksson, K. (2020). Long-run impacts of agricultural shocks on educational attainment: evidence from the boll weevil. The Journal of Improving the current input subsidy programs to include pesticides and Economic History, 80(1), 136–174. weedicides will have a far-reaching impact on reducing children’s time Bandara, A., Dehejia, R., & Lavie-Rouse, S. (2015). The impact of income and non- spent doing farm work and subsequently improve their wellbeing. income shocks on child labor: Evidence from a panel survey of Tanzania. World development, 67, 218–237. Further investment in nature-based solutions and bio-control in- Bandyopadhyay, S., & Skoufias, E. (2015). Rainfall variability, occupational choice, and novations will yield sustainability benefits in the long run. Third, welfare in rural Bangladesh. Review of Economics of the Household, 13(3), 589–634. development practitioners must consciously sensitize farmers on the Björkman-Nyqvist, M. (2013). Income shocks and gender gaps in education: Evidence from Uganda. Journal of Development Economics, 105, 237–253. deteriorative effects of continuous cropping and encourage them to Boateng, A., & Dako-Gyeke, M. (2022). Child Labor in Ghana: Current Policy, Research, embrace crop rotation, crop diversification, fallowing and other crop and Practice Efforts. Child Behavioral Health in Sub-Saharan Africa: Towards Evidence management practices to limit pest and weed invasion on their farms. In Generation and Policy Development, 265-281. Boissiere, M. (2004). Determinants of primary education outcomes in developing countries. conclusion, our results fit within the educ-workscape framework which Independent Evaluation Group (IEG), Washington, DC: World Bank. highlights the interaction between household, child work, and schooling Bourdillon, M., Crivello, G., & Pankhurst, A. (2015). Introduction: Children’s work and and has implications for other countries with similar challenges. current debates. In M. Bourdillon, G. Crivello, & A. Pankhurst (Eds.), Children’s Work and Labour in East Africa: Social Context and Implications for Policy (pp. 1–19). Addis Ababa: Organisation for Social Science Research in Eastern & Southern Africa. Declaration of Competing Interest Boyden, J., Porter, C., & Zharkevich, I. (2021). Balancing school and work with new opportunities: Changes in children’s gendered time use in Ethiopia (2006–2013). Children’s Geographies, 19(1), 74–87. The authors declare that they have no known competing financial Bourguignon, F., Fournier, M., & Gurgand, M. (2007). Selection bias corrections based on interests or personal relationships that could have appeared to influence the multinomial logit model: Monte-Carlo comparisons. Journal of Economic Surveys, the work reported in this paper. 21, 174–205. Chanimbe, T., & Dankwah, K. O. (2021). The ‘new’Free Senior High School policy in Ghana: emergent issues and challenges of implementation in schools. Interchange, 52 (4), 599–630. 11 E. Martey et al. W o r l d D e v e l o p m e n t P e r s p e c t i v e s 31 (2023) 100529 Child Labour: Global Estimates (2020), Trends and the Road Forward,“ United Nations Li, L., Gow, A. D. I., & Zhou, J. (2020). The role of positive emotions in education: A Children’s Fund, Accessed February 21, 2023, https://data.unicef.org/resources/ neuroscience perspective. Mind, Brain, and Education, 14(3), 220–234. child-labour-2020-global-estimates-trends-and-the-road-forward/. Magnan, N., Spielman, D. J., Lybbert, T. J., & Gulati, K. (2015). Leveling with friends: Conley, T. G., & Udry, C. R. (2010). Learning about a new technology: Pineapple in Social networks and Indian farmers’ demand for a technology with heterogeneous Ghana. American Economic Review, 100(1), 35–69. benefits. Journal of Development Economics, 116, 223–251. Dahmann, S. C. (2017). How does education improve cognitive skills? Instructional time Malikov, E., & Kumbhakar, S. C. (2014). A generalized panel data switching regression versus timing of instruction. Labour Economics, 47, 35–47. model. Economics Letters, 124(3), 353–357. David, O. G., Ayangbenro, A. S., Odhiambo, J. J., & Babalola, O. O. (2022). Striga Martey, E., & Armah, R. (2021). Welfare effect of international migration on the left- hermonthica: A highly destructive pathogen in maize production. Environmental. behind in Ghana: Evidence from machine learning. Migration Studies, 9(3), 872–895. Challenges, 100590. Martey, E., Etwire, P. M., & Atinga, D. (2021). To attend or not to attend: Examining the Dawud, M. A., Angarawai, I. I., Tongoona, P. B., Ofori, K., Eleblu, J. S., & Lfie, B. E. relationship between food hardship, school attendance and education expenditure. (2017). Farmers’ production constraints, knowledge of striga and preferred traits of International Journal of Educational Development, 80, Article 102304. pearl millet in Jigawa state, Nigeria. Global Journal of Science Frontier Research: D Masih, I., Maskey, S., Mussá, F. E. F., & Trambauer, P. (2014). A review of droughts on Agriculture and Veterinary, 17(3), 1–7. the African continent: A geospatial and long-term perspective. Hydrology and Earth Day, R., Abrahams, P., Bateman, M., Beale, T., Clottey, V., Cock, M., et al. (2017). Fall System Sciences, 18(9), 3635–3649. armyworm: Impacts and implications for Africa. Outlooks on Pest Management, 28(5), Mathenge, M. K., & Tschirley, D. L. (2015). Off-farm labor market decisions and 196–201. agricultural shocks among rural households in Kenya. Agricultural economics, 46(5), DeJaeghere, J. G., Chapman, D. W., & Mulkeen, A. (2006). Increasing the supply of 603–616. secondary teachers in sub-Saharan Africa: A stakeholder assessment of policy options Mc-Fadden, D. (1973). Conditional logit analysis of qualitative choice behavior. In in six countries. Journal of Education Policy, 21(5), 515–533. P. Zarembka (Ed.), Frontiers in Econometrics 1973 (pp. 105–142). New York: Di Falco, S., Veronesi, M., & Yesuf, M. (2011). Does Adaptation to Climate Change Academic Press. Provide Food Security? A Micro-Perspective from Ethiopia. American Journal of Moreira, C. N., Rabenevanana, M. W., & Picard, D. (2017). Boys go fishing, girls work at Agricultural Economics., 93(3), 829–846. home: Gender roles, poverty and unequal school access among semi- nomadic fishing Dimova, R., Gangopadhyay, S., Michaelowa, K., & Weber, A. (2015). Off-farm labor communities in South Western Madagascar. Compare, 47(4), 499–511. https://doi. supply and correlated shocks: New theoretical insights and evidence from Malawi. org/10.1080/03057925.2016.1253456 Economic Development and Cultural Change, 63(2), 361–391. Obeng-Ofori, D., & Amiteye, S. (2005). Efficacy of mixing vegetable oils with pirimiphos- Dubin, J., & McFadden, D. (1984). An econometric analysis of residential electric appliance methyl against the maize weevil, Sitophilus zeamais Motschulsky in stored maize. holdings and consumption., I(52), 345–362. Journal of Stored Products Research, 41(1), 57–66. Dunne, M., & Ananga, E. D. (2013). Dropping out: Identity conflict in and out of school in Oerke, E. C. (2006). Crop losses to pests. The Journal of Agricultural Science, 144(1), Ghana. International Journal of Educational Development, 33(2), 196–205. 31–43. Dunne, M., & Humphreys, S. (2022). The edu-workscape: Re-conceptualizing the Oliveira, C. M., Auad, A. M., Mendes, S. M., & Frizzas, M. R. (2014). Crop losses and the relationship between work and education in rural children’s lives in Sub-Saharan economic impact of insect pests on Brazilian agriculture. Crop Protection, 56, 50–54. Africa. World Development Perspectives, 27, Article 100443. Ortiz-Ospina, E., & Roser, M. (2016). Child labor. Our World in Data. Accessed February Dunne, M., Humphreys, S., & Szyp, C. (2021). Education and Work: Children’s Lives in 21, 2023. https://ourworldindata.org/child-labor. Rusal Sub-Saharan Africa. ACHA Working Paper 9. Brighton: ACHA, Institute of Peterman, A., Behrman, J. A., & Quisumbing, A. R. (2014). A review of empirical Development Studies. evidence on gender differences in nonland agricultural inputs, technology, and Dunne, M., Leach, F., Chilisa, B., Maudeni, T., Tabulawa, R., Kutor, N., et al. (2005). services in developing countries. Gender in agriculture, 145–186. Gendered school experiences: The impact on retention and achievement in Botswana and Randell, H., & Gray, C. (2016). Climate variability and educational attainment: Evidence Ghana. London: DfID. from rural Ethiopia. Global environmental change, 41, 111–123. Duryea, S., Lam, D., & Levison, D. (2007). Effects of economic shocks on children’s Robson, E., Ansell, N., Huber, U. S., Gould, W. T., & van Blerk, L. (2006). Young employment and schooling in Brazil. Journal of development economics, 84(1), caregivers in the context of the HIV/AIDS pandemic in sub-Saharan Africa. 188–214. Population, space and place, 12(2), 93–111. Ghana Population. (2022) (Live). World Population Review, Accessed February 21, Rose, P. M., & Dyer, C. (2008). Chronic poverty and education: A review of literature. 2023, https://worldpopulationreview.com/countries/ghana-population. Chronic Poverty Research Centre Working Paper, 131. Glick, P. J., Sahn, D. E., & Walker, T. F. (2016). Household shocks and education Salifu, I., Boateng, J. K., & Kunduzore, S. S. (2018). Achieving free compulsory universal investments in Madagascar. Oxford Bulletin of Economics and Statistics, 78(6), basic education through school feeding programme: Evidence from a deprived rural 792–813. community in northern Ghana. Cogent Education, 5(1), 1509429. Gowda, M., Makumbi, D., Das, B., Nyaga, C., Kosgei, T., Crossa, J., et al. (2021). Genetic Serdeczny, O., Adams, S., Baarsch, F., Coumou, D., Robinson, A., Hare, W., et al. (2017). dissection of Striga hermonthica (Del.) Benth. resistance via genome-wide Climate change impacts in Sub-Saharan Africa: From physical changes to their social association and genomic prediction in tropical maize germplasm. Theoretical and repercussions. Regional Environmental Change, 17, 1585–1600. Applied Genetics, 134, 941–958. Tabe-Ojong, M. P., Jr (2022). Ecological shocks and non-cognitive skills: Evidence from Hamenoo, E. S., Dwomoh, E. A., & Dako-Gyeke, M. (2018). Child labour in Ghana: Kenya. Ecological Economics, 194(C). Implications for children’s education and health. Children and Youth Services Review, Tabe-Ojong, M. P., Jr, Heckelei, T., & Baylis, K. (2021). Aspiration formation and 93, 248–254. ecological shocks in rural Kenya. The European Journal of Development Research, 33 Hu, Y., Li, B., Zhang, Z., & Wang, J. (2019). Farm size and agricultural technology (4), 833–860. progress: Evidence from China. Journal of Rural Studies. https://doi.org/10.1016/j. Tadele, Z. (2017). Raising crop productivity in Africa through intensification. Agronomy, jrurstud.2019.01.009 7(1), 22. Humphreys, S., Moses, D., Kaibo, J., & Dunne, M. (2015). Counted in and being out: Tafere, Y. (2014). Children’s experiences of household poverty dynamics in Ethiopia. Young Fluctuations in primary school and classroom attendance in northern Nigeria. Lives. International Journal of Educational Development, 44, 134–143. Tafere, Y., & Pankhurst, A. (2015). Children combining work and school in Ethiopian International Labour Organization. (2018). Decent Work and the Sustainable communities. Children’s Work and Labour in East Africa: Social Context and Development Goals: A Guidebook on SDG Labour Market Indicators, Department of Implications for Policy, Addis Ababa: Organization for Social Science Research in Eastern Statistics (STATISTICS), Geneva. https://www.ilo.org/ilostatfles/Documents/ and Southern Africa (accessed 29 January 2021). Guidebook-SDG-En.pdf. Accessed February 23 2023. Teka, H. B. (2014). Advance research on Striga control: A review. African journal of plant Jonah, O. T., & Abebe, T. (2019). Tensions and controversies regarding child labor in science, 8(11), 492–506. small-scale gold mining in Ghana. African Geographical Review, 38(4), 361–373. Tessema, Y. M., Asafu-Adjaye, J., Kassie, M., & Mallawaarachchi, T. (2016). Do Kasoma, C., Shimelis, H., & Laing, M. D. (2021). Fall armyworm invasion in Africa: neighbours matter in technology adoption? The case of conservation tillage in Implications for maize production and breeding. Journal of Crop Improvement, 35(1), northwest Ethiopia. African. Journal of Agricultural and Resource Economics, 11 111–146. (311–2016-5659), 211–225. Kassie, M., Stage, J., Diiro, G., Muriithi, B., Muricho, G., Ledermann, S. T., et al. (2018). Thorsen, D., & Yeboah, T. (2021). Mobility and the rural landscape of opportunity. In Push–pull farming system in Kenya: Implications for economic and social welfare. Youth and the rural economy in Africa: hard work and hazard (pp. 78–91). Wallingford Land Use Policy, 77, 186–198. UK: CABI. Kes, A., & Swaminathan, H. (2006). Gender and time poverty in sub-Saharan Africa. Unicef. (2022). United Nations Children’s Fund Accessed February 21, 2023, https: Gender, time use, and poverty in sub-Saharan Africa, 13. //www.unicef.org/about-unicef. Khonje, M. G., Manda, J., Mkandawire, P., Tufa, A. H., & Alene, A. D. (2018). Adoption United Nations. (2015). Transforming our world by 2030: A new agenda for global action and welfare impacts of multiple agricultural technologies: evidence from eastern Zero. Draft of the outcome document for the UN Summit to adopt the Post-2015 Zambia. Agricultural Economics, 49(5), 599–609. Development Agenda. New York: United Nations. Krauss, A. (2013). Understanding child labor beyond poverty: the structure of the Verkaart, S., Munyua, B. G., Mausch, K., & Michler, J. D. (2017). Welfare impacts of economy, social norms, and no returns to rural basic education. World Bank Policy improved chickpea adoption: A pathway for rural development in Ethiopia? Food Research Working Paper, (6513). Policy, 66, 50–61. Krishnan, P., & Patnam, M. (2014). Neighbors and extension agents in Ethiopia: Who Ward, P. S., & Pede, V. O. (2015). Capturing social network effects in technology matters more for technology adoption? American Journal of Agricultural Economics, adoption: the spatial diffusion of hybrid rice in Bangladesh. Australian Journal of 96(1), 308–327. Agricultural and Resource Economics, 59(2), 225–241. Lewbel, A. (2012). Using heteroscedasticity to identify and estimate mismeasured and Williams, I. H. (2010). The major insect pests of oilseed rape in Europe and their endogenous regressor models. Journal of Business & Economic Statistics, 30(1), 67–80. management: an overview. Biocontrol-based integrated management of oilseed rape pests, 1–43. 12 E. Martey et al. W o r l d D e v e l o p m e n t P e r s p e c t i v e s 31 (2023) 100529 Wollni, M., & Andersson, C. (2014). Spatial patterns of organic agriculture adoption: Yacoubou, A. M., Zoumarou Wallis, N., Menkir, A., Zinsou, V. A., Onzo, A., Garcia- Evidence from Honduras. Ecological Economics, 97, 120–128. Oliveira, A. L., et al. (2021). Breeding maize (Zea mays) for Striga resistance: Past, Wu, J., & Babcock, B. A. (1998). The choice of tillage, rotation, and soil testing practices: current and prospects in sub-saharan Africa. Plant Breeding, 140(2), 195–210. Economic and environmental implications. American Journal of Agricultural Zimdahl, R. L. (2018). Fundamentals of weed science. Academic press. Economics, 80(3), 494–511. 13