A systematic review of employment outcomes from youth skills training programmes in agriculture in low- and middle-income countries

Engagement of youth in agriculture in low- and middle-income countries may offer opportunities to curb underemployment, urban migration, disillusionment of youth and social unrest, as well as to lift individuals and communities from poverty and hunger. Lack of education or skills training has been cited as a challenge to engage youth in the sector. Here we systematically interrogate the literature for the evaluation of skills training programmes for youth in low- and middle-income countries. Sixteen studies—nine quantitative, four qualitative and three mixed methods—from the research and grey literature documented the effects of programmes on outcomes relating to youth engagement, including job creation, income, productivity and entrepreneurship in agriculture. Although we find that skills training programmes report positive effects on our chosen outcomes, like previous systematic reviews we find the topic to chronically lack evaluation. Given the interest that donors and policymakers have in youth engagement in agriculture, our systematic review uncovers a gap in the knowledge of their effectiveness. Youth engagement in agriculture in low- and middle-income countries offers social and economic opportunities, and supports food security. Agriculture skills training is key for youth engagement, yet programmes are not well evaluated for effectiveness in employment outcomes.

positive consequences for the alleviation of hunger, particularly for women, as their empowerment in agriculture improves households' food security and nutrition [16][17][18] .
However, there has been a declining trend of youth participation in agriculture since 2000, mainly in favour of the service sector 6,19,20 , which precipitates migration from rural to urban areas. Increased educational attainment for rural youth coupled with inability to rent or own land is a driver of urban migration 21 . In addition, the increasing ageing farmer population in rural areas exacerbates the demographic pressure on land at the expense of the youth 22 .
A further constraint on youth engagement in agriculture is a lack of education in disciplines related to agriculture or skills training [23][24][25] . A study among Thailand's youth reported that 71% identified knowledge of farming practices as a pre-requisite to setting up a viable farm 23 . In rural Ethiopia, government initiatives to increase skills and productivity, and introduce improved and modern farming methods have generated interest among youth in joining the sector, and in Indonesia, vocational training was noted as increasing the likelihood of a successful career in agriculture 26 . A study in Zambia on rural youth aspirations, opinions and perceptions on agriculture documented high interest among youth in more productive forms of farming, such as the use of draught animals, electricity and the increased application of fertilizers 24 . Such findings challenge an assumption common in policy proposals that youth are not interested in agriculture 25 . Today, with the development of information and communication technology (ICT), young people have more opportunities to strengthen their skills and access relevant information and are therefore well positioned to understand market dynamics, and institutional and financial systems, enabling them to initiate and capitalize on processes of change in the agricultural A systematic review of employment outcomes from youth skills training programmes in agriculture in low-and middle-income countries sector 27,28 . Human capital theory predicts a positive correlation between human capital accumulation and labour productivity. On that basis, skills training can be used to improve agricultural employment outcomes 29 . Where governments and policy interventions support skills training for youth, there is a real possibility for entrepreneurship, a competitive economy and ultimately national growth. But, despite the implementation of skills training interventions, generally via youth employment programmes 30 , few specifically target agricultural skills training in LMIC and very little is known about the effectiveness of youth agricultural interventions 30,31 .
Here we systematically review skills-based training interventions that aim to increase youth engagement in agricultural employment in LMIC to better inform investment decisions made by donors and policymakers. The interventions include agriculture-related courses, on-the-job training, technical or vocational education and training in agriculture, as well as general skills training including entrepreneurship, financial literacy and life skills for engagement in agriculture. The outcomes of interest we started out with were: employment along an agricultural value chain; employment in agribusiness; engagement in contract farming; development of agricultural entrepreneurship; agricultural business performance (productivity, profit, income, marketing rate); involvement in agricultural extension service provision. After data extraction, the outcomes of interest found in the selected studies are jobs created in the agricultural sector, self-employment and entrepreneurship, provision of and employment in extension services, profit/income/ earnings from an agricultural activity or job, farm productivity, and the accessibility of employment opportunities in the sector. These outcomes pertain to the categories of jobs that can be found along the agricultural value chain.
We found among the studies yielded from the systematic literature search that skills training interventions reported employment in agriculture, agribusiness or agriculture-related activities, development of agricultural entrepreneurship, agricultural business performances (productivity, profit, income) and involvement in agricultural extension service provision for young participants. However, we also found a chronic lack of evaluation of the effectiveness of interventions designed to enhance agricultural opportunities and engagement for young people in LMIC, a finding previously shown 31 .

results
Sixteen studies were identified for review based on a priori inclusion and exclusion criteria ( Fig. 1) detailed in our Preferred Reporting Items for Systematic Reviews and Meta-Analyses Protocol, PRISMA-P (Supplementary Material 1, summarized in Methods and published on Open Science Framework, https://osf.io/bhegq//). Table 2) was used to document all information of interest from each of the 16 studies, overviewed in Table 1.
Nine of the studies solely targeted youth [32][33][34][35]37,38,43,45,46 , and seven targeted mixed groups of youth and others 36,[39][40][41][42]44,47 . In fourteen studies, the participants were from all genders. In nine of the studies, Studies meeting the inclusion criteria and targeting mixed group (youth and other demographic groups) were also retained in the search strategy. A double-blind title and abstract screening were performed on 4,789 articles that were uploaded to systematic review software, Covidence, for title and abstract screening. each article was reviewed by two independent reviewers and discrepancies were resolved by a third independent author within the team. After title and abstract screening, 261 articles remained. From title and abstract screening, 16 articles met a priori inclusion criteria. when youth programmes are increased by one unit (one programme) youth are eight more times likely to take the offer of participating in agriculture programmes.
• A total of 33% of respondents believed that they will be self-employed.
• A total of 18% of respondents stated that they will earn money by selling agricultural products, 15% believed that participation will lead to a permanent job, 13% were of the position that agriculture will alleviate poverty in their families. The findings of the survey revealed that youth perceive agriculture as a bad career.
• When programmes available are increased by a unit, the odds ratio is 18 times higher and therefore youth are likely to participate in agricultural activities when they are more programmes available for agriculture.
• engagement/ entrepreneurship in agriculture/contract farming/agribusiness Singh et al. • The continued adopters of beekeeping and mushroom growing had increased their family income by 49% and 24%, respectively.
• Profit/income/ earning of the farm/ agriculture-related • Yields of major crops and income of farmers were slightly higher among the participants than the non-participants.
•84.2% of beneficiaries achieve yields that exceed one tonne per hectare for maize in Nigeria.
• Only 15.8% of the participants obtained yield below one tonne per hectare against 34% among non-participants. Mixed group (youth and others) • The ICT-enhanced participatory radio approach has the potential to enhance food security of smallholder farmers in sub-Saharan Africa (SSA).
• The participatory strategies can help to engage audiences, increase knowledge of agricultural improvements and innovations, and contribute to higher levels of adoption than result from listening alone.
• The interactivity has the potential to contribute to women's empowerment.
• In most cases, significantly more listeners adopted the agricultural practice (growing orange-fleshed potatoes) that was the focus of the participatory radio campaign than non-listeners.
• Youth only (students in the national high schools) • The experiment showed that these students are willing to engage in farming. Also, the paper showed that about 68% of students from Bayanihan National High School and 85% of students from Maria Aurora National High School intended to stay in agriculture as farmers.
• Self-employment in agriculture/ agribusiness/ agriculture-related • The majority (52%) of graduates were engaged in extension work, 13% in business organizations, 14% in research, 1.7% self-employed, 43% in agriculture, 29% in consultancy and 28% in other forms of enterprise.
• 42% worked in rural areas, 36% worked in semi-urban areas and 22% worked in urban areas. Most graduates were satisfied.
• The majority find the training and preparation adequate for the labour market and the skills relevant to the requirements of their jobs.
• 95% of employers find the required skills in graduates.
• Job creation in agriculture • engagement/ entrepreneurship in agriculture/contract farming/agribusiness Mixed group (youth and others) • On average, eight individuals were employed daily in bone-craft production and seven in soap production.
• The intervention created a total of 120 direct jobs that were involved in soap and bone-craft production.
• Job creation in agriculture • The programme helped youth engage in agro-allied businesses and earn income.
• The youth training farm helped increase productivity of local farmers.
• Job creation in agriculture • engagement/ entrepreneurship in agriculture/contract farming/agribusiness Studies were identified for review as per the PrISMA-P protocol. Data were extracted using a template available as Supplementary Table 2. Studies are described here primarily according to their design: quantitative, qualitative and mixed methods. DID and PSM are also impact evaluation methods that use a treated and control groups approach to assess the effectiveness of an intervention.

Table 1 | General overview of selected studies (continued)
participants were a mixed group of those already and not yet engaged in agriculture 32,34,37,39,[41][42][43][44]46 ; in five of the studies, participants were already engaged in agriculture before receiving skills training interventions [35][36][37]45,47 ; there was not enough information to determine whether the participants were already engaged in agriculture in two studies 33,40 . Six of the studies indicated that the participants resided in rural areas [33][34][35][36]46,47 , while participants located in urban areas and in both rural and urban areas were identified in four 32,38,40,45 and five 37,39,41,43,44 of the studies, respectively; there was not enough information to determine the location of the participants in one 42 study. The population targeted in the studies was both educated and non-educated youth. Among the nine studies [32][33][34][35]37,38,43,45,46 that focused exclusively on youth, two targeted youth with a secondary education background 34,46 , one 45 targeted youth with a university background and six 32,33,35,37,38,43 of the studies targeted youth with a mixed educational background.
Risk of bias assessment. We evaluated the risk of bias of the included studies based on a previous approach 48 . The domains of risk retained are (1) the sampling technique used for the study, (2) the type of intervention, (3) the choice of the area of study, (4) the population targeted, (5) the method of data collection, (6) the method of data analysis, (7) the measurement of outcome and (8) the statistical significance of the effect. For each domain of risk, the criteria evaluated were defined and rated by their relevance for assessing the effectiveness of the interventions. Supplementary Table 4 summarizes the criteria of each domain of risk and its assessment and rating. Using this scale, 15% of our included studies are at low risk of bias, 60% at moderate risk of bias and the remaining 25% at serious risk of bias. The outcome of the risk of bias assessment of the included studies in this systematic review is presented in Table 3.
The risk of bias assessment process highlighted differences in focus, methods used and standards of evidence across the included studies. Weaknesses in study design, survey methods and method of evaluation of the impact of the interventions were common in most of the studies (with the exception of the studies ranked at low risk of bias), leading to weak results and limited generalizability.
Effects on youth employment outcomes. The youth employment outcomes of interest to this systematic review are job creation, self-employment, engagement in entrepreneurship, provision of extension services, productivity of the farm/agriculture-related activities, profits/income, and job search or employment opportunity in agriculture-related activities. Here we elaborate on the study design and risk of bias of all studies, and highlight the effects on outcomes of interest for a selection of low and moderate risk studies.
In one quantitative study, deemed at low risk of bias (Table 3), 1,700 workers and 1,500 firms were followed over four years to compare the effects of offering workers vocational training and offering firms wage subsidies to train workers on-the-job (firm training) in Uganda 32 . The results showed that both interventions allowed participants to acquire sector-specific skills and firm-specific skills leading to higher employment rates post-training for each type of worker, but the effect was greater for vocational training compared with firm training (21% versus 14% post-training employment rate) and their total earnings rose by more compared with the firm-training intervention (34% versus 20%). The qualitative studies 41,42 , although not designed to assess the effectiveness of an intervention, highlighted a link between skills training and employment outcome. However, both studies were deemed at serious risk of bias. A mixed-methods study 38 on youth programmes in Ghana showed that about 86.4% of young people still pursued maize farming a year after exiting the Youth in Agriculture Programme (YIAP). This public intervention was implemented to address youth unemployment in Ghana with the goal of getting young people to engage in the agricultural sector. The four main components of the programme were crops/block farm, livestock and poultry, fisheries/aquaculture, and agribusiness. The study focuses on evaluating the crops/block farm component. The crops cultivated under the YIAP include maize (seed and grain), sorghum, soybean, tomato and onion. This study is ranked at moderate risk of bias.
Self-employment in agriculture. Six studies 36,39,41,45-47 indicated that skills training interventions resulted in self-employment in agriculture. Out of these studies, two studies are quantitative 36,45 , three are qualitative 41,46,47 and one is a mixed-methods study 39 .
In one quantitative study 36 , self-employment was stimulated by a skills training radio campaign on growing orange-fleshed sweet potatoes in Ghana, Tanzania, Burkina Faso and Uganda. A survey of the local communities where the radio campaign was run found that households that reported hearing the educational radio campaign in Ghana, Tanzania, Burkina Faso and Uganda were 8.9, 2.3, 1.7 and 1.1 times more likely, respectively, to engage in growing orange-fleshed sweet potatoes, than households that did not. This study is deemed at moderate risk of bias.
Engagement/entrepreneurship in agriculture. Five studies 34,38,39,41,42 showed that skills training interventions encourage youth engagement or entrepreneurship in agriculture. Among these studies, one is quantitative 34 , two are qualitative 41,42 and two are mixed-methods studies 38,39 . In the quantitative study, a youth programme including agriculture content (training in livestock production, crop production and dairy farming) in South Africa indicated that youth engagement or self-employment in agriculture is eight times higher when agricultural programmes that specifically target the youth are implemented compared with when agricultural programmes are not available. This study is deemed at moderate risk of bias. Regarding the mixed-methods studies, one study 38 , deemed at moderate risk of bias with youth programme (YIAP in Ghana) as intervention, showed that after exiting the programme, 86.4% of beneficiaries were still involved in farming within a year. The qualitative studies were deemed at serious risk of bias.
Productivity of the farm/agriculture. Two studies 35,41 found that skills training interventions lead to higher productivity of the farms. One of the studies is quantitative 35 and the other is qualitative 41 . In the quantitative study, estimated to be at moderate risk of bias, the National Agricultural Extension and Research Liaison Services (NAERLS) rural youth extension programmes (RUYEP) helped 84.2% of beneficiaries achieve yields that exceed one tonne per hectare for maize in Nigeria, compared with 66% of non-participants 35 . The qualitative study 41 , outlined in Table 1, is deemed at serious risk of bias.
Profit/income earning of the farm. Ten studies 32,33,35,38,[40][41][42][43][44]47 looked at profit/income earning of the farm as an outcome. Among those studies, five are quantitative 32,33,35,43,44 , three are qualitative 41,42,47 and two 38,40 are mixed-methoda studies. In one of the quantitative studies, the Training for Rural Economic Empowerment (TREE) programme increased beneficiaries' income by US$787 compared with non-beneficiaries over the 2011-2014 programme implementation period 33 . This study is deemed at low risk of bias. Another quantitative study 44 , deemed at moderate risk of bias, found that the continued adopters of beekeeping and mushroom growing had increased their family income by 49% and 24%, respectively. The three qualitative studies, not described here but outlined in Table 1, are deemed at serious risk of bias 41,42,47 . The mixed-methods study 40 showed that the creation of a company that recycled livestock by-product (bone crafts and soap production) allowed vulnerable women and youths to earn an additional US$44.6 from bone crafts and US$50.2 from soap production weekly. This study is at moderate risk of bias.
Job search or employment opportunity. Three studies 39,41,42 investigated the effect of skills training on this outcome. One study is a mixed-methods design 39 and two 41,42 are qualitative. All of these studies, not described here but outlined in Table 1, are deemed at serious risk of bias.
Provision of agricultural extension service. One study 39 investigated on the effects of skills interventions on provision of agricultural extension service and found that the majority of graduates who benefited from student-farmer attachment and/or the Supervised Student Enterprise Project (SSEP) were engaged in extension work. This study, outlined in Table 1, is deemed at serious risk of bias.
Intervention type and engagement in agriculture. Agriculturerelated courses. Two studies 39,47 used agriculture-related courses as interventions. One of these studies is a mixed-methods study 39 and the other is qualitative 47 . The mixed-methods study investigated several outcomes in agriculture, namely, job creation, entrepreneurship, self-employment, provision of agricultural extension service and job search opportunity, which were found to improve with the skills training interventions. The interventions consisted of introducing innovations in agricultural training curricula (community engagement and agri-enterprise development) at Gulu University in Uganda. The community engagement took the form of a one year (or less) placement of undergraduate students to work with smallholder farmers and farmer groups within a 10 km radius of the university. The agri-enterprise development consisted of having the students design business plans; the best plans were rewarded with start-up capital. The employment rate among the graduates was 84% six months after graduation and increased to 90% after one year; less than 2% of the graduates created their own businesses. The qualitative study 47 investigated two outcomes in agriculture, self-employment and income, which were found to increase after skills training on ready food mixes, maize products and mango products. The two studies are deemed to be at serious risk of bias.
Technical education/training. Four studies 35,41,42,46 used technical education/training as interventions. Only one of these studies is quantitative 35 ; the others are qualitative 41,42,46 . The quantitative study 35 investigated productivity and income of the farm, and found both to increase after the intervention. The NAERLS RUYEP objectives The evaluation of the included studies bias is based on a previous approach 48 . For example, for the domain of risk relating to the sampling technique, three criteria were identified: random sampling, non-random sampling and a mix of the two types of sampling. The maximum rate a study can obtain in this domain is three stars. If the study used a random sampling technique, it gets three stars; if it uses a mix of the two types of sampling, it gets two stars; and if the sampling technique is not random, it gets one star (see Supplementary Table 4 for details on the criteria used).
are to provide technical advisory services to boost agricultural production and raise living standards of the youth. The results showed that the intervention allowed 84.2% of beneficiaries to achieve yields that exceed one tonne per hectare for maize in Nigeria, compared with 66% of non-participants. This study is deemed at moderate risk of bias. Among the qualitative studies, one 46 looked at self-employment as an outcome and found a positive association with the intervention. The other two qualitative studies are deemed of serious risk of bias.
Youth programme. Youth programmes are programmes that target youth and train them in either specific skills (agricultural skills, ICT skills and so on) or broad skills (decision-making skills, business skills and so on) to enhance their employability. These have been used as interventions in three studies 33,34,38 . One of these studies is mixed methods 38 and the two others are quantitative 33,34 . The mixed-methods study 38 investigated the following outcomes in agriculture: job creation, engagement and income; a positive association was found between youth programme and both engagement and income. The results showed that about 86.4% of young people still pursued maize farming one year after exiting the programme and the mean income of GH¢758 obtained by beneficiaries was found to be greater than the national mean annual per capita income of GH¢734. Among the two quantitative studies 33,34 , one investigated the income of beneficiaries 33 and the other 34 looked at engagement in agriculture; both found a positive effect of the intervention on their outcome. The study that investigated the income of beneficiaries as an outcome revealed that the TREE programme increased beneficiaries' income by US$787 compared with non-beneficiaries over the 2011-2014 programme implementation period 33 . In the other study 34 , a youth programme including agriculture content (training in livestock production, crop production and dairy farming) in South Africa indicated that youth engagement or self-employment in agriculture is eight times higher when agricultural programmes that specifically target the youth are implemented compared with when agricultural programmes are not available. Given that all three studies are at moderate or low risk of bias, we can conclude that the findings suggest that youth programmes have the potential to influence youth engagement in agriculture.
On-the-job training. Only one study 36 looked at on-the-job training as an intervention. The outcome investigated is self-employment, on which the intervention had a positive effect. The results showed that households that reported listening to an educational radio campaign in Ghana, Tanzania, Burkina Faso and Uganda were 8.9, 2.3, 1.7 and 1.1 times more likely, respectively, to engage in growing orange-fleshed sweet potatoes, than households that did not. The study was deemed at moderate risk of bias.
Vocational training. Vocational training has been used as an intervention by four studies 37,40,44,45 . Among these studies, three are quantitative 37,44,45 and one is a mixed-methods study 40 . One quantitative study 44 investigated income as an outcome, on which positive effects of the intervention were found in India. The findings indicated that vocational training programmes have resulted in continued adoption of beekeeping and mushroom cultivation enterprises by 20% and 51% of trained farmers, respectively, and increased their family income by 49% and 24%, respectively. The second quantitative study investigated job creation and self-employment as outcomes and found positive links with the training 45 . The results of the study highlighted that vocational training in agriculture in Iran resulted in employment of more than half of graduates. The third quantitative study found a positive effect of the intervention on job creation, the sole outcome it had investigated 37 . The study showed that vocational training for a youth employment programme in Ghana resulted in the creation of 16,383 jobs in agribusiness. All four studies are deemed at moderate risk of bias (Table 3); however, the use of descriptive methods in some of these studies preclude us from concluding that they are effective in improving employment outcomes for youth in the agricultural sector.
Vocational training and technical training. One study 43 investigated the combination of vocational training and technical training as an intervention. The outcomes investigated are job creation and income, on which the intervention had a positive effect. The study indicated that vocational training and technical training in agriculture (poultry technician) resulted in an increase in employment of 34.2% among the 41 beneficiaries who were trained as poultry technicians in Nepal. This study is deemed at low risk of bias, suggesting that combining vocational training and technical training may be a way of improving job prospects and income for youth in the agricultural sector.
Vocational training and on-the-job training. One study 32 investigated the combination of vocational training and on-the-job training as an intervention. The outcomes investigated are job creation and earnings, on which the intervention had a positive effect. The results showed that both interventions allowed participants to acquire sector-specific skills and firm-specific skills, leading to higher employment rates post-training for vocational-trained workers compared with firm-trained workers (21% versus 14% post-training employment rate) and their total earnings rose by more compared with the firm-trained workers (34% versus 20%). This study is deemed at low risk of bias.

Duration of training.
Ten studies out of the 16 overviewed in Table 1 presented information on the duration of training. Eight of these have programmes that last one year or less. The remaining studies indicated a training duration between two and five years. This suggests that training programmes predominantly have a relatively short-term duration, which is consistent with many interventions taking the form of technical and vocational education/training. The popularity of technical and vocational/education training as a model of intervention may be due to the relatively short-term nature of the training, or due to the nature of technical and vocational training, which is well suited for out-of-school youth, which are found in large numbers in LMIC 49 .

Discussion
Issues facing youth engagement in agriculture today are relatively well documented, including educational attainment, matrimonial status, gender, household size, parental income and occupation, membership in social organization, access to ICT, land tenure system and access to state-run agricultural youth programmes [50][51][52] . This present systematic review, which focused solely on interventions to engage youth in agriculture, yielded a limited set of studies-nine quantitative, four qualitative and three mixed-methods studies-so generalizable conclusions are difficult to draw. The risk of bias assessment yielded three studies 32,33,43 deemed at low risk of bias, nine studies [34][35][36][37][38]40,[44][45][46] deemed at moderate risk of bias and four studies deemed at serious of risk bias 39,41,42,47 .
The results of our systematic review generally are in line with those found by the systematic review of Kluve et al. 53 on interventions to improve the labour market outcomes of youth. That systematic review of 107 interventions, including skills training, in 31 countries, found small positive effects for promoting entrepreneurship and skills training-especially integrated skills training programmes-but not for employment services and subsidized employment.
Our systematic review also demonstrated that in general, skills interventions seeking to motivate youth's engagement in agriculture do not undergo a thorough evaluation for effectiveness, with hard outcomes related to employment. Our selected studies included case studies and qualitative methods, which are not adequate methods of evaluating impact and effectiveness of interventions. Only one study used an RCT 32 . The two studies relying on a quasi-experimental approach used DID and PSM methods 33,43 . Indeed, the results of the risk of bias assessment indicated the studies relying on RCT and quasi-experimental impact evaluation methods were at low risk of bias. However, these study designs are expensive to conduct. We found that of the studies that evaluate interventions, the majority did not use state-of-the-art impact evaluation methods. This has been corroborated by other studies 30,31 , showing a chronic lack of evaluation of interventions that aim to provide agricultural skills to youth.
Training on ICT is an important aspect for attracting and retaining youth in the agricultural sector 46 . ICT offers a method of delivering training to a large number of farmers, which could enhance the performance of the youth already in agriculture and attract new youth to the sector 36 . Radio campaigns have been shown to be effective in spurring adoption and consumption of orange-fleshed potatoes in Ghana, Tanzania, Burkina Faso and Uganda 36 . A study conducted in the Philippines found that ICT training helps motivate secondary school students whose parents are engaged in agriculture to work within the sector, especially when combined with offline activities such as exposure and hands-on experience as well as creative and motivational actitivites 46 .
It is important to note that heterogeneity in gender and education are not accounted for in the analysis of the impacts of education on youth participation in agriculture. Our systematic review revealed that most of the included studies failed to address the effectiveness of targeting the population of interest-educated and uneducated youth. Illiteracy and gender heterogeneity were not addressed in the included studies. Indeed, no studies assessed the effects of training interventions on illiterate youth. This calls for investigations to focus on this vulnerable group of society, which represent about 25% of youth in sub-Saharan Africa and 11% in Southern Asia 54 . Failing to account for such variation in the background of the youth participants limits the ability to assess the effectiveness of skills training interventions.
The absence of robust research and lack of effective evaluation of the available data on the effectiveness of agricultural youth employment interventions has notable consequences on potential investment. Ultimately, the commitment of policymakers is necessary to ensure the sustainability and success of interventions to boost youth's engagement in agriculture. It is encouraging that the majority of interventions (12 studies out of 16) studied originated from public policy, compared with three originating from non-public policy programmes (NGOs, international institution) and one from mixed policies (public and non-public policies). However, to provide a compelling basis on which to convince governments and donors to fund future interventions, as well as encourage young people to partake in training, cost-effectiveness analysis and estimates of returns on investment in training programmes is necessary. Indeed, a 2018 stocktaking of the evidence on the effectiveness of youth employment interventions in Africa found that for the agricultural sector in particular, "there is very little literature and virtually no evaluation evidence to inform policymakers about what types of interventions can improve the prospects of young people in the [agricultural] sector" 31 . Our study supports this conclusion. Moreover, to ensure that the skills training provides long-term opportunities for youth, it is crucial to establish a periodic follow-up to assess how trainees are performing after completion of a training programme. This aspect was missing in most of the interventions reviewed in this systematic review, yet it is important to check that the youth who engage in agriculture after receiving skills training are still involved and thrive in their agriculture-related business in the long term.
In summary, there is a need to foster youth skills training programmes and more importantly to evaluate more rigourously these programmes so that knowledge on good practices may be generated and transferred from one developing country to another. Estimates of returns to investment of agricultural skills training programmes are warranted as they could provide governments and donors with the evidence and cost-based analysis to continue and increase support for such programmes. Interventions also need to account for heterogeneity in gender and educational background of the youth to foster sustainability in agricultural value chains, inform inclusive policy design and ultimately contribute to reducing poverty and food insecurity in LMIC.

Methods
This systematic review was prepared following guidelines from Petticrew and Roberts 55 . The approach comprises five steps: identifying the research question; identifying relevant studies; study selection; extracting and charting the data; and collating, summarizing and reporting the results. The protocol for this study was registered on the Open Science Framework before study selection and can be accessed at https://osf.io/bhegq//. The guiding question for this systematic review was: What are the effects of skills training interventions on educated and non-educated youth employment outcomes in agricultural value chains, agribusiness or contract farming in LMIC? The inclusion and exclusion criteria to identify and then select the relevant studies are shown in Table 4.

Risk of bias assessment.
Regarding the risk of bias assessment, each study was assessed following the criteria of the eight domains of risk of bias we considered. The maximum score a study can obtain in terms of minimizing all domains of risk of bias is 23 stars, which is 100% of the stars. A study is deemed to be at low risk of bias across all domains if its total score is in the interval 75-100%. If the total score is in the interval 50-75%, the study is said to be at moderate risk of bias across all domains. A study is at serious risk of bias if its score falls within the interval 25-50%. When the total score ranges from 0 to 25%, the study is deemed to be at critical risk of bias across all domains. See Supplementary Table 4 for details on the criteria used.
Search strategy. An exhaustive search strategy was developed and tested in CAB Abstracts to identify all available research pertaining to the effects of skills training interventions on educated and non-educated youth employment outcomes in agriculture in LMIC. Search terms were developed to address variations of the key concepts in the research question: skills training, youth, employment or engagement, and agriculture. Searches were performed on 9 May 2019 in the following electronic databases: CAB Abstracts (access via OVID); Web of Science Core Collection (access via Web of Science); EconLit (access via ProQuest); Agricola (access via OVID); and Scopus (access via Elsevier). Full search strategies for each database, including grey literature, can be accessed in their entirety at https://osf.io/xv56k/.
A comprehensive search of grey literature sources was also conducted. A list of the resources that were searched can be found at https://osf.io/xv56k/. The grey literature searches were performed using custom web-scraping scripts. The search strings were tested per website before initiating web-scraping. An existing Google Chrome extension was needed to scrape dynamically generated websites.
The results from the databases and the grey literature searches were combined and de-duplicated using a Python script. Duplicates were detected using title, abstract and same year of publication, where year of publication was a match, where title cosine similarity was greater than 85%, and where abstracts cosine similarity was greater than 80% or one of the abstracts (or both) was empty. When duplicates were found, the results from the databases and the grey literature searches were combined and duplicates were removed.
Following de-duplication, each citation was analysed using a machine-learning model. The model added more than 30 new metadata fields, such as identifying populations, geographies, interventions and outcomes of interest. This allowed for accelerated identification of potential articles for exclusion at the title/abstract screening stage.
Study selection and eligibility criteria. Systematic review software, Covidence, was used for both title/abstract and full-text screening decision-making with two independent reviewers evaluating each item. Citations were included in this study if they met all of the inclusion criteria noted above. Studies that did not meet all the inclusion criteria were excluded. Exclusion criteria were the inverse of the inclusion criteria. Each citation that met one of the exclusion criteria at the title, abstract or full-text screening phases were excluded. Studies included in the full-text screening stage were those that met all inclusion criteria and none of the exclusion criteria, or those whose eligibility could not be established during title/abstract screening. Reasons for exclusion were documented at the full-text screening phase.
The retrieval of hundreds of PDFs for full-text screening was done with a combination of automated and manual methods. For the automated method, a Python script was created that would handle the tasks of PDF discovery, download and file renaming using Google Scholar. The script read the bibliographic data from an Excel spreadsheet and then executed a script to retrieve the full-text PDF. If the article is spotted in the search results, the download link is clicked, and the article will be auto-renamed and marked as being downloaded. Manual methods were employed for those items that were not retrieved using the script.
A total of 245 records were identified for full-text screening. This screening process led to the identification of 16 studies that were considered adequate regarding the content and methodological rigour. The PRISMA flow diagram (Fig. 1) shows the steps followed during the screening process and the number of items that resulted after each step.
Data extraction. Data extraction was based on interventions and outcomes established in the research question and exclusion criteria. The data extraction focused on the outcomes of the studies, the methods used to obtain the outcomes, and the validity and reliability of those methods using a data-extraction form. To reduce risk of bias related to the extracted data, two separate researchers extracted data from each included study in the full-text review step. When disagreements occurred between researchers on data extracted from a study, a third researcher was engaged to resolve conflict by extracting data again from the study and the results were compared with those found previously. In total, 31 conflicts were solved among the 261 reviews. The critical appraisal of individual sources of evidence gave an indication of the strength of evidence provided and informed the standards followed for this systematic review.
Reporting Summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
The data that support the findings of this study are available from the corresponding author upon reasonable request.

Code availability
The code used in this study is available upon request.

nature research | reporting summary
April 2020 Corresponding author(s): NATFOOD-19120581B Last updated by author(s): Jul 24, 2020 Repor�ng Summary Nature Research wishes to improve the reproducibility of the work that we publish. This form provides structure for consistency and transparency in repor�ng. For further informa�on on Nature Research policies, see our Editorial Policies and the Editorial Policy Checklist.

Sta�s�cs
For all sta�s�cal analyses, confirm that the following items are present in the figure legend, table legend, main text, or Methods sec�on.

n/a Confirmed
The exact sample size (n) for each experimental group/condi�on, given as a discrete number and unit of measurement A statement on whether measurements were taken from dis�nct samples or whether the same sample was measured repeatedly The sta�s�cal test(s) used AND whether they are one-or two-sided Only common tests should be described solely by name; describe more complex techniques in the Methods section.
A descrip�on of all covariates tested A descrip�on of any assump�ons or correc�ons, such as tests of normality and adjustment for mul�ple comparisons A full descrip�on of the sta�s�cal parameters including central tendency (e.g. means) or other basic es�mates (e.g. regression coefficient) AND varia�on (e.g. standard devia�on) or associated es�mates of uncertainty (e.g. confidence intervals) For null hypothesis tes�ng, the test sta�s�c (e.g. F, t, r) with confidence intervals, effect sizes, degrees of freedom and P value noted

So�ware and code
Policy informa�on about availability of computer code Data collec�on An exhaus�ve search strategy was developed and tested in CAB Abstracts to iden�fy all available research pertaining to effects of skills training interven�ons on educated and non-educated youth employment outcomes in agriculture in developing countries. Search terms were developed to address varia�ons of the key concepts in the research ques�on: skill training, youth, employment or engagement, and agriculture. Searches were performed on May 9, 2019 in the following electronic databases: CAB Abstracts (access via OVID); Web of Science Core Collec�on (access via Web of Science); EconLit (access via ProQuest), Agricola (access via OVID), and Scopus (access via Elsevier). Full search strategies for each database, including grey literature, can be accessed in their en�rety at h�ps://osf.io/xv56k/. The results from the databases and the grey literature searches were combined and de-duplicated using a Python script. Duplicates were detected using �tle, abstract and same year of publica�on, where year of publica�on was a match, where �tle cosine similarity was greater than 85%, and abstracts cosine similarity greater than 80% or one of abstracts (or both) was empty. When duplicates were found, the cita�on priority order was Scopus, CAB Abstracts, Web of Science, Agricola, EconLit, and followed by grey literature sources.

Data analysis
Systema�c review so�ware, Covidence, was used for both �tle/abstract and full-text screening decision-making with two independent reviewers evalua�ng each item. Cita�ons were included in this study if they met all of the inclusion criteria noted above. Studies that did not meet all of the aforemen�oned inclusion criteria were excluded. Exclusion criteria were the inverse of the inclusion criteria. Each cita�on that met one of the exclusion criteria at the �tle, abstract, or full-text screening phases were excluded. Studies included in the full-text screening stage were those that met all inclusion criteria and none of the exclusion criteria, or those whose eligibility could not be established during �tle/abstract screening. Reasons for exclusion were documented at the full-text screening phase. A total of 261 records were iden�fied for full-text screening. This screening process led to the iden�fica�on of 20 studies which were considered adequate regarding the content and methodological rigor. The data extrac�on focused on the outcomes of the studies, the methods used to obtain the outcomes, and the validity and reliability of those methods using a data extrac�on form. Data extracted from the 20 included studies were analyzed using descrip�ve sta�s�cs methods and cross tabula�on analysis. In addi�on to exis�ng tables in the first version of the submi�ed manuscript, cross tabula�on analysis between outcomes of interest and types of interven�ons was added. In the same vein cross tabula�on analysis between methods of analysis used in the included studies and the types of April 2020 publications was also added. A risk of bias assessment was conducted to determine the rigour of methodology of the included studies. The rating of the studies was done by considering several domains of risk of bias suggested by Sterne et. al (2016).
For manuscripts utilizing custom algorithms or software that are central to the research but not yet described in published literature, software must be made available to editors and reviewers. We strongly encourage code deposition in a community repository (e.g. GitHub). See the Nature Research guidelines for submitting code & software for further information.

Data
Policy information about availability of data All manuscripts must include a data availability statement. This statement should provide the following information, where applicable: -Accession codes, unique identifiers, or web links for publicly available datasets -A list of figures that have associated raw data -A description of any restrictions on data availability The data that support the findings of this study are available from the corresponding author upon reasonable request.

Field-specific reporting
Please select the one below that is the best fit for your research. If you are not sure, read the appropriate sections before making your selection. All studies must disclose on these points even when the disclosure is negative.

Study description
This study is a systematic review that analyzes the effect of skills training on youth engagement in the agricultural sector in developing countries. it uses systematic method to identify, appraise, and synthesize all relevant studies related to skills training interventions on youth and their engagement in agriculture. The data used in this study include both qualitative and quantitative informations.

Research sample
The targeted population is both educated and non-educated youth from developing countries. The review is on interventions that consist of skills training including agriculture-related courses, general education, on the job training, technical or vocational education and training, as well as general skills training including entrepreneurship, financial literacy, and life skills. The outcomes of interest are all employment outcomes related to the agricultural sector. These outcomes are jobs created in the agricultural sector, self-employment and entrepreneurship, provision of extension services, differences in profit/income/earnings from an agricultural activity or job, farm productivity (including yields), and accessing employment opportunities in the sector.

Sampling strategy
Inclusion criteria was based on studies that include: -Educated and non-educated youth (young men and women); -Youth employment in agriculture; -Job creation in agriculture -Agriculture, Agricultural productivity, Agricultural value chain, Agribusiness, Contract farming, Agricultural entrepreneurship; -Skill training, Occupational Skills Training, On-the-job training, Enterprise skills, Business skills, Agricultural training programs; -Published from 1990 to 2019 in English and French; -Original research and/or review of existing research -International institution reports -Youth participation/involvement in agriculture Data collection An exhaustive search strategy was developed and tested in CAB Abstracts to identify all available research pertaining to effects of skills training interventions on educated and non-educated youth employment outcomes in agriculture in developing countries. Search terms were developed to address variations of the key concepts in the research question: skill training, youth, employment or engagement, and agriculture. Searches were performed on May 9, 2019 in the following electronic databases: CAB Abstracts (access via OVID); Web of Science Core Collection (access via Web of Science); EconLit (access via ProQuest), Agricola (access via OVID), and Scopus (access via Elsevier). Full search strategies for each database, including grey literature, can be accessed in their entirety at https://osf.io/xv56k/. The results from the databases and the grey literature searches were combined and de-duplicated using a Python script. Duplicates were detected using title, abstract and same year of publication, where year of publication was a match, where title cosine similarity was greater than 85%, and abstracts cosine similarity greater than 80% or one of abstracts (or both) was empty. When duplicates were found, the citation priority order was Scopus, CAB Abstracts, Web of Science, Agricola, EconLit, and followed by grey literature sources. Systematic review software, Covidence, was used for both title/abstract and full-text screening decision-making with two independent reviewers evaluating each item. Citations were included in this study if they met all of the inclusion criteria noted above. Each citation that met one of the exclusion criteria at the title, abstract, or full-text screening phases were excluded. Studies included in the full-text screening stage were those that met all inclusion criteria and none of the exclusion criteria, or those whose eligibility could not be established during title/abstract screening. A total of 261 records were identified for full-text screening. This screening process led to the identification of 20 studies which were considered adequate regarding the content and methodological rigor. The data extraction focused on the outcomes of the studies, the methods used to obtain the outcomes, and the validity and reliability of those methods using a data extraction form.

Timing
Searches were performed on May 9, 2019 Data exclusions Studies that did not meet all of the aforementioned inclusion criteria were excluded. Exclusion criteria were the inverse of the inclusion criteria. Each citation that met one of the exclusion criteria at the title, abstract, or full-text screening phases were excluded.

Non-participation
The abstract review process started with 4,789 articles. After full-text screening, 261 remained. During full-text review we excluded 241 records to retain 20 articles.

Randomization
Using Covidence, candidate studies have been randomly assigned and double blinding screened by the team members. Each study was reviewed for relevance by two independent reviewers. Each study that meets all the inclusion criteria was included. All conflicts between reviewers were resolved in Covidence by a third, independent reviewer.
Reporting for specific materials, systems and methods We require information from authors about some types of materials, experimental systems and methods used in many studies. Here, indicate whether each material, system or method listed is relevant to your study. If you are not sure if a list item applies to your research, read the appropriate section before selecting a response.

Recruitment
There are a variety of types of bias to be considered in a systematic review including publication bias, reporting bias and included study bias. In order to reduce risk of bias two separate researchers extracted data from each included study in the full text review step. When disagreements occurred between researchers on data extracted from a study, a third researcher was engaged to resolve conflict by extracting data again from the study and the results were compared to those found previously. In total, 31 conflicts were solved among the 261 reviews.

Ethics oversight
Ceres2030, Cornell University, USA Note that full information on the approval of the study protocol must also be provided in the manuscript.