At 33% as of 2024 Q1, South Africa has the highest unemployment rate in the world, with the rates doubling (61%) for youth aged 15-24. This is an unequal crisis on multiple dimensions. Rates are overwhelmingly higher among the Black African population (36%), followed a far second colored population (23%). Women face a 13% higher likelihood of unemployment than men.

Over three million young people in South Africa are not in employment, education or training (NEET). The same patterns of unequal outcomes persist: young South Africans who are not in employment, education or training comprise of a majority black (88%) urban (59%) and women (54%). These individual who are NEET are particularly vulnerable to long spells of unemployment and insecure work because they do not possess traditional qualifications and experiences that are observed by firms to determine skills and experiences.

More than 80% of the youth who are NEET have never been employed and have been searching for employment for more than 1 year while 20% have been searching for 5+ years. Low enrolment rates in tertiary education (25.2% as of 2021), is a key factor behind this widespread and persistent unemployability of youth. The lack of tertiary education is further compounded by the apartheid legacy of legal discrimination and segregation that generated large wealth and spatial inequalities across the country. This means, for many of these already disadvantaged jobseekers, available jobs are also located prohibitively far away.

Together, these factors characterize a labor market facing a vast disconnect between labor demand and labor supply sides. On the supply side, a large pool of jobseekers struggle to signal their skills to prospective employers and face inherently high search costs due to socioeconomic and spatial inequalities. On the demand side, firms appear uncertain about the quality of the applicants they attract with many reporting that the skills they are interested in are not emphasized or tested within the education system. This in turn leads to firms relying on trial-and-error recruiting tendencies, creating insecure entry level jobs. Experimental research found firms more willing to hire when given legal advice on how to fire entry-level workers.

Inaccurate beliefs about skills lead to high prevalence of misdirected job search. Job search support interventions have proven useful in improving employment outcomes. Studies have tested and found positive results for job search support through 1) behavioral encouragement interventions such as planning support, social support; 2) better utilization of job search platforms such as training jobseekers in how to use LinkedIn and 3) endorsement mechanisms such as provision of reference letters and low cost skill testing and certification to job seekers. When implemented at scale, better job search infrastructure has the potential to bridge the gap between what firms need versus what jobseekers can offer through increased visibility of the labour market and pathways within it for both sides.

Spatial mismatches between jobs and jobseekers, combined with high search costs, further deepen inaccurate beliefs about the job market among youth. Overly optimistic beliefs about the job market lead to young jobseekers under searching and holding out for "better jobs". On one hand, transport subsidies and cash transfers have been found to increase job search activity, but these search efforts do not necessarily translate to better employment outcomes on average. One study found a negative information shock to such subsidies leading to lower reservation wages and acceptance of lower paying jobs.

Further readings:

Abebe, G., Caria, A.S., Fafchamps, M., Falco, P., Franklin, S. and Quinn, S., 2021. Anonymity or distance? Job search and labour market exclusion in a growing African city. The Review of Economic Studies, 88(3), pp.1279-1310.

Abebe, G., Caria, S.A., Fafchamps, M., Falco, P., Franklin, S., Quinn, S. and Shilpi, F.J., 2023. Matching frictions and distorted beliefs: Evidence from a job fair experiment (No. 958). working paper.

Abel, M., Burger, R., Carranza, E. and Piraino, P., 2019. Bridging the intention-behavior gap? The effect of plan-making prompts on job search and employment. American Economic Journal: Applied Economics, 11(2), pp.284-301.

Abel, M., Burger, R. and Piraino, P., 2020. The value of reference letters: Experimental Evidence from South Africa. American Economic Journal: Applied Economics, 12(3), pp.40-71.

Afridi, F., Dhillon, A., Roy, S. and Sangwan, N., 2023. Social Networks, Gender Norms and Labor Supply: Experimental Evidence Using a Job Search Platform (No. 677). Competitive Advantage in the Global Economy (CAGE).

Banerjee, A. and Sequeira, S., 2023. Learning by searching: Spatial mismatches and imperfect information in Southern labor markets. Journal of Development Economics, 164, p.103111.

Beam, E.A. and Quimbo, S., 2023. The Impact of Short-Term Employment for Low-Income Youth: Experimental Evidence from the Philippines. Review of Economics and Statistics, 105(6), pp.1379-1393.

Bertrand, M. and Crépon, B., 2021. Teaching labor laws: Evidence from a randomized control trial in South Africa. American Economic Journal: Applied Economics, 13(4), pp.125-149.

Bhorat, H., Köhler, T. and de Villiers, D. (2023). Can Cash Transfers to the Unemployed Support Economic Activity? Evidence from South Africa. Development Policy Research Unit Working Paper 202301. DPRU, University of Cape Town.

Branson, N., DeLannoy, A., & Kahn, A. (2019). Exploring the transitions and well-being of young people who leave school before completing secondary education in South Africa. Working Paper Series Number 244, NIDS Discussion Paper 2019/11 Version 1.

Carranza, E., Garlick, R., Orkin, K. and Rankin, N., 2022. Job search and hiring with limited information about workseekers’ skills. American Economic Review, 112(11), pp.3547-3583.

Chakravorty, B., Bhatiya, A.Y., Imbert, C., Lohnert, M., Panda, P. and Rathelot, R., 2023. Impact of the COVID-19 crisis on India’s rural youth: Evidence from a panel survey and an experiment. World Development, 168, p.106242.

Fields, G.S., 2011. Labor market analysis for developing countries. Labour economics, 18, pp.S16-S22.

Franklin, S., 2015. Location, search costs and youth unemployment: A randomized trial of transport subsidies in Ethiopia.

Hardy, M. and McCasland, J., 2023. Are small firms labor constrained? experimental evidence from ghana. American Economic Journal: Applied Economics, 15(2), pp.253-284.

ILO., 202. The impact of COVID-19 on the informal economy in Africa and the related policy responses.

ILO., 2023. African youth face pressing challenges in the transition from youth to work.

Jones, S. and Sen, K., 2022. Labour market effects of digital matching platforms: Experimental evidence from sub-Saharan Africa.

Kiss, A., Garlick, R., Orkin, K. and Hensel, L., 2023. Jobseekers’ beliefs about comparative advantage and (mis) directed search. Available at SSRN 4593303.

Loiacono, F. and Silva Vargas, M., 2023. Improving access to labor markets for refugees: Evidence from Uganda.

McKenzie, D., 2017. How effective are active labor market policies in developing countries? a critical review of recent evidence. The World Bank Research Observer, 32(2), pp.127-154.

Mudiriza, G., De Lannoy, A. (2023). Profile of young people not in employment, education or training (NEET) aged 15-24 years in South Africa: an annual update. Cape Town: Southern Africa Labour and Development Research Unit, University of Cape Town. (SALDRU Working Paper Number 298).

Wheeler, L., Garlick, R., Johnson, E., Shaw, P. and Gargano, M., 2022. LinkedIn (to) job opportunities: Experimental evidence from job readiness training. American Economic Journal: Applied Economics, 14(2), pp.101-125.

World Bank (2023), “Skills and Workforce Development,” worldbank.org. Hensel, L., Tekleselassie, T., Isphording, I., Radbruch, J. & Witte, M. 2024. Demand for Feedback and Job Search. Working Paper.

When adapting ESCO, a European labor taxonomy, to the South African "seen" economy, the first step involved asking if occupations in Europe - and the skills associated with them - are similar to those in South Africa. The answers to this were approached from two directions.

First, do occupations with the same titles across Europe and South Africa substantively entail the same tasks and require the same skills? For instance, being a "specialize seller" or "taxi driver" in Europe (and hence in ESCO), rarely involves having to negotiate prices or bargain. However, these are critical skills for such occupations in most developing countries, including South Africa. These gaps in occupational definitions and skills requirements are larger for more informal occupations, especially in a gig economy like South Africa's.

Second, do Europeans and South-Africans refer to the same occupations in the same terms? For example, is someone who conducts data collection called an data collector in both regions? Locals in South Africa use the term "fieldworker" more commonly. ESCO provides space to link "alternative titles" to every occupation it lists, so as to reflect this diversity of names given to the same jobs. It is also translated in every European language and reflects the conceptual diversity of ways languages refers to the same occupations. Expansion of ESCO beyond the EU, particularly in relatively data-poor developing contexts, requires careful groundwork for the new countries and languages.

Tabiya approached answers to these questions in two ways:

1. Using existing databases and taxonomy: We used a combination of Harambee's rich database of employment data, and SA's pre-existing local labor taxonomy in testing methodologies to systematically add South African alternative titles to ESCO.

2. Directly ground truthing the applicability and usability of ESCO for microentrepreneurs: We conducted a survey with young micro-entrepreneurs registered with Harambee in trying to assess whether: i) respondents could readily find their work on a list of ESCO occupations ii) skills under selected ESCO occupations fully captured the skills that a microentrepreneur thinks they have/need.

Alternative titles for occupations allow us to make sure that there is no "one best way" to refer to a job, an informal activity, or a hobby. To make sure every one feels represented on the Harambee platform, we included idiomatic occupation titles, and allow for an adaptative listing of alternative occupation titles.

Why are alternative job titles important?

In any economy, individuals who are otherwise employed to do the same job may be hired under different titles. The ESCO taxonomy attempts to address this concern by providing a list of alternative titles that are associated with the different occupations captured in its database. However, regional variations in job titles may exist, and so while the ESCO list of so-called “alternative titles” or “alternative labels” for occupations may be sufficient for the European labour market, the same cannot necessarily be said for the South African labour market.

Localizing ESCO to the South African Labor Market

In localizing ESCO to South Africa's context, we aimed to identify to what extent ESCO's existing titles and alternative titles captured the jobs done by participants in this labor market. The exploration and subsequent alternative title additions were completed in two blocks of work: 1) a matching exercise using the EJ data entries and ESCO titles 2) a cross-referencing exercise using SA's local labor taxonomy, the Organizing Framework for Occupations (OFO), with ESCO.

1. Matching with EJ Data

To get a sense of the need to add alternative titles for ESCO occupations, we first mapped 500 formal, seen occupations from the EJ data to the ESCO taxonomy. The first attempt was an automated text-matching exercise between job titles and descriptions in the EJ data and the ESCO taxonomy. Given the diversity of job titles and job descriptions on both sides, automating the text-matching proved to be a significant challenge. Consequently, a manual review of Harambee's EJ data needed to be undertaken as an initial proof of concept before large-scale data collection and analysis could be undertaken.

Random samples of EJ entries from both formal sector workers and the microentrepreneurs were analyzed. A preliminary analysis of a random sample of 50 EJ entries from each category was followed by a scaled up analysis of 500 entries from each category. Results were consistent across both subsamples. For brevity, the results presented in the following section are based on the random subsample of size 500.

To match the job descriptions provided to by young job-seekers to ESCO occupations, we manually applied the following rules:

This manual mapping resulted in a total of 279 individual EJ entries being matched to 73 unique ESCO occupation codes. At the occupation level, 51 occupations required no additional alternative titles to be added, with 39 providing exact matches to individuals’ user-provided titles, and 12 providing root-word matches to user-provided titles. The remaining 22 occupations potentially require the addition of alternative titles to ensure cohesiveness between South African occupation titles and ESCO occupation titles.

At the individual-level, we find that 224 individual user-provided occupation titles were matched to ESCO occupation titles or alternative titles via either an exact match or root word match. This accounts for 80.3% of the total matched sample of individuals. In other words, only 19.7% of individuals in the matched sample entered occupation titles that may require the addition of alternative titles to the ESCO database. This information is summarized in Table 5.

The exercise resulted in a 80.3% match (exact + root-word matches) at the individual level and a 70% match at the occupation level. Table 5 summarizes the findings.

Table 5: Summary of user occupation title match to ESCO occupation title, by occupation group.

Source: Harambee EJ data and own calculation

The South African Organizing Framework for Occupations

In our EJ individuals sample, we find a total of 55 people who do not have exact matches to an ESCO occupation title or alternative title. In a further attempt to match these occupations, we opted to explore using the South African Organizing Framework for Occupations (OFO) as a cross-referencing tool for potential alternative titles that would need to be added to the ESCO framework.

Table 6: List of ESCO occupations with suggested additional alternative titles (based on sample of 500)

Source: Own construction based on Harambee EJ data and European Commission (2022).

The above exercises demonstrated that for the most part, ESCO titles and alternative titles are able to capture most of South Africa’s occupations, and the results can be improved by additionally using the OFO framework.

We then chose to merge the 2021 OFO list of occupations, specialisations and alternate titles to ESCO. Among individuals in our EJ data sample who provided sufficient information to enact a mapping, 89% of these can find their occupation description in some combination of ESCO and OFO titles. This makes a substantive case to use an OFO-ESCO merge as a baseline from which to build our taxonomy for the South African labour market.

The matching from ESCO to OFO was done in two steps. First, straightforward matches were made thanks to a simple matching algorithm the following two rules:

End product: OFO-ESCO merge

The key to maintaining a time-relevant labor taxonomy is to build systems which are consistently taking feedback from users. Using a fixed taxonomy runs the risk of future users not finding occupations and skills because of both new kinds jobs being created and existing jobs being rebranded. To manage these risks, Harambee continues to include a free text option for job descriptions/titles on the platform. This option doubles input and feedback for the research team at Tabiya to continue adding missing occupation titles and skills to the taxonomy.

Time-use surveys worldwide aim to capture how individuals spend their day. The International Classification of Activities for Time Uses Statistics (ICATUS) is an international standard that provides a framework and classification for these time-use activities. The taxonomy encompasses a comprehensive set of activities, including those considered typically productive such as doing a job or working for pay, as well as unproductive activities like sleeping or watching television.

The ICATUS taxonomy offers a more inclusive approach to defining what is productive. It encompasses activities that fall outside the production boundary set by the System of National Accounts. These include services rendered without pay for own and others' final use, such as cleaning one’s own dwellings or cooking one’s own food. These activities are considered productive insofar that they have paid equivalents (for instance, a professional cleaner or a cook). Therefore, they engage human capital and enable human capital formation.

We identify four types of such activities in ICATUS, which we call “unseen” activities.

1. Unpaid domestic services for household and family members (Div. 3)

2. Unpaid caregiving services for household and family members (Div. 4)

3. Unpaid direct volunteering for other households (Div. 51)

4. Unpaid community- and organization-based volunteering (Div. 52)

Highlighting this human capital in the South African first entails uncovering through which activities it might be formed. The South African time use survey from 2011 is the latest national wide source of data on how South Africans spend their time on the daily. When it comes to unseen activities, caring for someone in one’s household and maintaining one’s household takes up 19,9% of young South African’s time (15-34 years old). On the other hand, community service for non-household members such as volunteering (at Church, for instance) takes up 0,4% of young South Africans’ time.

What is the EJ data?

Harambee, through SAYouth, regularly collects updated employment information from youth enrolled on their platform. This information, collected over XX years, produces a rich database of activities and occupations held by enrolled youth across time, highlighting the diversity of employment journeys (EJ) among young South African job seekers. For each activity and occupation held by young job-seekers, the database contains titles, a brief job description and additional information on conditions of employment/engagement such as contract duration and permanence, business classification (product/service/both) etc.

What do employment journeys teach us?

Employment experiences of young South African job seekers are a mix of formal, informal, and experiences.

We first look at the distribution of different types of livelihoods in the EJ data. As a first, cut, the data is decoded into three subsets for those who report any occupation or activity:

1. Those who report working for someone else, i.e. as being “employed” or having an “employer” are considered as a sample of the formal employment sector.

2. Those who report working for themselves i.e. "self-employed" are classified as microentrepreneurs and capture a sample of individuals from South Africa’s informal economy.

3. Volunteer work represents unpaid activities reported by jobseekers, and this includes a range of individual or community level unpaid work starting from apprenticeships to community service.

The formal and informal sectors together represent what we are calling the seen economy, while volunteer work and the 57.7% unemployed and unspecified sample represent the unseen economy. Strikingly, among those who have been involved in some work in the past 30 days, informal work and volunteer work make up 60% of cases. This confirms that, in the South African case, recognizing and highlighting skills acquired by young job-seekers in the unseen and informal economies is crucial for inclusivity.

Table 1: Distribution of Harambee EJ data across livelihood types

Source: Harambee EJ Data

Noting data limitations and sample comparability with the SA QLFS

It is important to note that the EJ data, while a rich source of information, is available for a very specific selection of individuals: South African youth who have registered as jobseekers on the Harambee platform, with a disproportionate number of individuals located in and around the province of Gauteng in South Africa (see Table 2). Having an overrepresented urban sample could introduce biases in our analysis, as our assessment of whether ESCO is applicable to our context will be based on livelihood descriptions provided by these individuals.

To avoid such biases, we are making sure that the Harambee platform (version 0.1) allows users to use free text to describe their experiences. Thanks to this, all livelihoods types can be taken into account, and the new information provided can enrich Harambee's list of activities and occupations.

Table 2: Provincial breakdown of youth from Harambee EJ data relative to South African Quarterly Labour Force Survey data

Source: Own calculations using Harambee EJ Data and Statistics South Africa (2023)

Note: 1. The National Youth Commission (1996) defines youth as all individuals between the ages of 15 and 34 (inclusive). Due to the age of majority in South Africa being 18, we restrict our analysis to youth aged between 18 and 34 (inclusive). 2. Numbers from QLFS weighted using sampling weights.

Analyzing job transitions shed light on how moves are influenced by past experiences; impact of job moves on economic stability, mobility and fragility; and employer preferences.

About the data and analysis: The job transition analysis is based on SAYouth placement data of 164,527 unique youth who report multiple opportunities on the platform between 2012 through May 2024. These youth make up 4.3% of the 3.8 million youth who are registered on the platform, and 18.6% of the 884,000 unique youth that report at least one opportunity on the platform (with 1.2 million total opportunities).

The analysis examines the first transitions made by these youth. This means, opportunities are only included if they have a valid start date and a valid opportunity type namely Public Employment Program (PEP), Formal Sector (FS), or Make Your Own Money (MYOM). Among youth who report multiple opportunities, 80% record two so we capture their full journey in this sample. For the remaining 20% with more than two reported opportunities, we examine only the first two.

A total of 329,054 placements in the data have been classified into one of three types of opportunities: PEP, MYOM/microenterprise, and formal sector. Of the three, the sample is PEP-heavy.

The analysis examines opportunities starting as early as 2012, but the majority fall within the 2020-2022 period. For 27% of transitions, the time between start dates of 1st & 2nd opportunities is less than 100 days. For 71% of transitions, this time between is less than 400 days.

Learnings on flows: Youth zigzag across many types of opportunities, but the likelihoods of different transitions are influenced by and in most cases, mirror their previous experiences.

Transitions overall, across PEP, FS and MYOM, are more likely to be within groups (60.3%) than between groups. PEP participants tend to transition to another PEP 71% of the time, and youth in the formal sector tend to transition to another role in the formal sector 54% of the time. Meanwhile, youth in MYOM opportunities are much less likely to transition to another MYOM opportunity (23%) than to the formal sector or PEPs.

• Transitions from/to PEP: PEPs are the most common opportunities, with DBE as the single largest opportunity. 70% of youth in this sample participate in a PEP as a first and/or second opportunity. On net, there is transition out of PEP. However, 54% of all youth in the sample transitioned to a PEP, composed of 71% youth with prior PEP experience, 36% from MYOM and 29% from the formal sector.

• Transitions from/to formal sector: The formal sector is the second most common type of opportunity overall, with 17% of all youth in this sample staying in the formal sector across both opportunities. On net, there is more transition into the formal sector. 32% of all youth in the sample transitioned to the formal sector, composed of 71% youth with prior FS experience, 41% from MYOM and 18% from PEP.

• Transitions from/to formal sector: On net, more youth move into MYOM than moving out of. However, importantly, youth who were engaged in MYOM were nearly twice as likely to transition to the formal sector or PEPs than to a second MYOM opportunity. Of the 14% that transitioned into MYOM, 23% were previously engaged in MYOM, 17% from FS and 11% from PEP.

Why South Africa?

South Africa faces high youth unemployment rates calling for action into improving intermediation between young job-seekers and employers. Despite being a problem shared with many other countries in the region, the South African case is of particular interest because of the diversity of large scale private and semi-governmental undertakings meant to bring answers to these challenges. Harambee and SAYouth were products of these efforts, and the large database of users, network of stake-holders, and expertise hosted by these organizations present unique opportunities to explore and test a breadth of technical solutions to labour market intermediation.

Our Partner: Harambee Youth Employment Accelerator

Founded in 2011, Harambee Youth Employment Accelerator is a not-for-profit social enterprise working to find solutions to youth unemployment in South Africa, and having expanded operations to Rwanda as of 2018. The organization is an anchor partner for SAYouth, South Africa's national network for youth employment pathway management. Within SAYouth, Harambee operates the multi-channel, tech-enabled SA Youth Platform (sayouth. mobi), which connects a network of 3.8 million jobseeking youth with over 1,300 private and public sector employer partners to facilitate earning and learning opportunities.

The optimal way to use our inclusive taxonomy to help young jobseekers is not obvious. For instance, it could be used as the basis of a static survey in which users pick occupations and skills manually, or in a more flexible way by using AI. Although certainly complementary, we hope to test these two approaches separately to identify the advantages and disadvantages of each approach.

Changing Narratives

Highlighting the skills young job-seekers gain in the unseen economy is only the first step to allowing them to mobilize their full human capital. Changing narratives surrounding skills aquired in the unseen economy is a necessary step to ensuring that job-seekers' skills are not only visible, but also credible in the eyes of their future employers. Indeed, skills acquired in the unseen economy typically suffer from negative biases from employers. Namely, a skill aquired in the unseen economy is typically not deemed as credible or mastered by young job-seekers as a skill aquired in the seen economy.

Changing narratives involves working on two fronts:

1 - Producing statistical evidence: By collecting data on the Harambee platfom and through (quasi)experiments, we aim building compelling, genuine pieces of evidence about the quality of skills aquired in the unseen economy, compared to skills aquired in the seen economy.

2 - Advocacy: By sharing success stories and making scientific evidence understandable to employers, we aim to challeging their biases regarding the skilfulness of job-seekers from the unseen economy.

How Skilled are Unseen Workers ?

Including the unseen part of the economy in an useable taxonomy of occupations and skills comes with numerous challenges linked to the difficulty to grasp the skill content of daily tasks and to the diversity of daily habits among young job-seekers. We identified 3 key challenges associated with skills acquired in the unseen economy:

• A transferability issue: skills acquired in the unseen economy may not be directly useable in formal jobs.

• A proficiency issue: the level of skill acquired in the unseen economy might compare quite poorly with the average level os skill of someone performing the same tasks in a formal job.

• A likelihood issue: performing a given unseen activity does not imply using all skills potentially involved in this activty.

Because of these three separate issues, skills acquired in the unseen economy lack overall credibility - or, economic terms, signaling value-. One qualifies a skill acquired in the unseen economy as "credible" if employers believe the claim of a young job-seekers declaring that they have acquired that skill in the unseen economy. This issue of credibility is not as foceful in the seen economy - although it is still problematic - as employers are less likely to cast doubts on skills acquired in a formal setup - say, a firm - or in a very identificable informal occupation - such as informal specialized seller -.

Addressing the issue of the credibility of skills acquired in the unseen economy - and therefore issues of transferability, proficiency, and likelihood - is essential to Tabiya's goal of changing narratives around unseen activities. On the one hand, our task is to highlight the skills acquired in the unseen economy. On the other hand, acting as if skills acquired in the unseen economy were as credible as skills acquired in formal jobs would do disservice to youg job-seekers from the unseen economy. Indeed, employers ultimately take iring decisions based on their own beliefs and perception of the credibility of skills.

To assess the applicability of the ESCO taxonomy to the informal micro-entrepreneur economy of South Africa, we conducted a series of primary data collection exercises with Harambee users. The core objective of this data collection was to finally address two key questions:

1 - Can all activities in the informal micro-entrepreneur economy in South Africa be associated to at least one ESCO occupation?

2 - Does the ESCO taxonomy effectively encompass both occupation-specific skills and entrepreneurial skills prevalent in the informal micro-entrepreneur economy?

Activity 1: Focused Group Discussions to identify a relevant list of general entrepreneurial skills

Microentrepreneurs constitute a unique segment within the workforce, as they possess two distinct sets of skills. First, they possess occupation-specific skills directly related to their trade; for instance, a hairdresser possesses skills such as hair washing and styling. Second, microentrepreneurs also exhibit general entrepreneurial skills required for self-employment or own-account work, such as identifying suppliers, resource management, and understanding customer needs.

Harambee facilitated focused group discussions (FGD) with 18-35 year-old, prior or current informal micro-enterprise owners within their network. These FGDs were designed to identify the second set of generic entrepreneurial skills that might be relevant to microentrepreneurs in South Africa

A sample of 60 participants including replacements was selected, balanced on gender, age group and type of entrepreneurial venture reported. Ten percent people with disabilities were included in the sample. To encourage participation, a circular was disseminated, inviting eligible individuals to join the FGDs, with selected participants offered R450 for their involvement

Collecting descriptions of micro-entrepreneurial activities: online survey

The second step of our work with Harambee consisted in gaining insights into the income-generating activities undertaken by young individuals.

To do so, we developed a short online survey for individuals within the SA Youth network. The primary aim of the survey was to obtain accurate descriptions of the main micro-entrepreneurship activities individuals engaged in to earn income in the past 30 days. To achieve this, we included the following question: “Tell us about the main way you make money. What things or services do you sell? Write a short sentence or two explaining what you do”. The survey also gathered information on current contact details, the number of income-generating activities, time allocation to different activities, and a brief title for the respondents' primary activity.

5.2.3 Matching descriptions of micro-entrepreneurial activities to ESCO occupations

The third step of the work on microentrepreneurs was to match individual's descriptions of their activities with existing ESCO occupations.

One option would be to perform this matching exercise manually, similar to the process for the formal economy. However, this would be a very labour-intensive exercise, as we had over 3,300 individual descriptions and 3,008 potential ESCO occupations. Furthermore, manual matching at this scale would be rather prone to human bias and error. To streamline the process and mitigate the risk of bias, we opted for a two-part, sequential matching procedure leveraging artificial intelligence.

In the first phase, we designed an algorithm capable of identifying up to three relevant ESCO occupations for each provided description. In the second phase, we reviewed these matches manually to evaluate the accuracy of these matches. Adjustments were made if a match was deemed inadequate or if we believed that an apparent occupation match had been overlooked.

In total, 3,236 individuals (99%) were successfully matched to at least one potential ESCO occupation. There were 23 individuals for whom no applicable ESCO occupation match was identified. These individuals engaged in activities related to:

• Informal lending of money for interest,

• Running errands for people (e.g., renewing licenses, fetching medication, arranging home affairs appointments),

• Making photocopies of documents for people in the community.

Distribution of occupation matched to respondents' answers

Assessing relevance of ESCO occupations and skills: phone survey

Finally, we assessed the applicability of the ESCO taxonomy for informal microentrepreneurs. To achieve this, we designed a phone survey with three primary objectives:

• Verify if the ESCO occupations accurately represent individuals' work and if they would willingly identify themselves with these labels.

• Evaluate the relevance of the ESCO skills associated with each matched occupation.

• Assess the relevance of the identified list of general entrepreneurial skills.

To address the first objective, participants were presented with each matched ESCO label and its description. They were then asked to consider whether these applied to their main way of making money. For the second objective, we randomly selected five essential knowledge and skills and and competences tags associated with each occupation, maintaining proportional allocation between knowledge and skills/competencies tags. Lastly, to test the third objective, five general entrepreneurial skills identified during the previous process were randomly selected for each participant. The complete questionnaire can be found here.

For each skill or knowledge tag, we designed several questions in order to determine its relevance to individuals’ work. These included whether individuals considered the skill as essential or optional, their perceived proficiency in the skill, their perception of how well other entrepreneurs in similar roles perform the skill, the importance of the skill to their own work, its importance to other entrepreneurs engaged in similar activities, and the perceived importance of the skill in a formal sector job akin to their work.

In addition, we incorporated a section aimed at identifying potential skill omissions from the ESCO taxonomy for informal micro-entrepreneurs. This section prompted participants to indicate if there were any other skills that were essential for their work that we had not inquired about.

Results of the Phone Survey

The results of the microentrepreneurship survey seem to corroborate our approach, especially by highlighting the absence of systematic bias regarding own skill evaluation between different subgroups in the data. The full results of the survey can be found here.

Importanly, young South African on average consider that they are more skilled than the rest of job-seekers. This is positive, as it means that they are confident in their own skills.

Contrary to what one may expect, the data does not highlight any statistically significant difference in the estimation of respondents' own skills between males and females. If anything, it shows that female respondents tend to deem their own skills as more advanced, compared to their male counterparts.

Finally, there is no statistically significant difference between respondents' own skill evaluation based on wether or not they matriculated (finished high school).

Why a Panel ?

When it comes to assessing the transferability and credibility of skills acquired by job-seekers in the unseen economy, employers are likely to display strong biases. Namely, when a young job-seekers declares that he possesses skill A, an employers is likely to consider that skill A acquired in the unseen economy is not equivalent to skill A aquired in the seen economy, both in terms of the level of skillfulness (credibility) and nature of the skill (transferability).

To build an inclusive taxonomy, our challenge is, therefore, to try to get an unbiased overview of the skills that could be associated with unseen activities defined in the ICATUS framework. There exist no data systematically estimating the skillfulness of workers and people involved in the unseen economy, although some work has been conducted to assess the acquisition of basic skills (literacy, numeracy) among harambee users (cf. Kate Orkin). Our approach therefore relies on an attempt to oppose different biases in order to get a sense of the credibility of skills associated with unseen activities. Our guess is that, by making people who have different views on the unseen economy discuss, we may obtain a clearer picture of the value agents give to skills acquired in the unseen economy.

Participants to the Panel

Ideally, the panel would have included a dozen participants representing different interests. We expect members to represent:

• Intermediation experts: Such as Harambee's employees, who work on the daily with both young jobseekers and employers. Their approach to skills is likely to be the most realistic, as it reflects both the will to improve the inclusivity or the South African labour market, and the realism acquired through their work with the main employers

• Employers: Employers are likely to display negative biases regarding the transferability and credibility of skills acquired in the unseen economy. However, they are obviously the most important player when it comes to recruitment. Understanding their their take on the credibility of skills acquired o the unseen economy is essential to building an intermediation tool that successfully promotes the skills of job-seekers from outside any labour market (formal and informal).

• Unseen economy workers looking for jobs: They are a precious source of information, not only because they are those who Tabiya and Harambee aim to help, but also because they have clear idea of what they are capable of. However, they have an evident incentive to report possessing as many skills as possible and a high degree of skillfulness in order to get hired. This upward biases is also likely gendered, as sociological studies have shown that women tend to understate their skillfulness, relative to men.

• Unseen economy beneficiaries: They encompass all individuals that might benefit from a job-seekers involvement in unseen activities. This includes children, dependent and non-dependent adults, relatives operating household firms and receiving unpaid help from job-seekers... These participants are likely to have a positive take on skills acquired in the unseen economy, and play a critical role in changing narratives around them.

Mobilizing such a large and dispersed panel of individuals is complex and difficult to implement. Specifically, it was deemed close to impossible to mobilize employers to participate in such an exercise within a reasonable timeframe. Looking back after completing the exercise, it would have indeed been extremely difficult given the length of the exercise. As a result of the different constraints, the panel was held with 5 to 7 members of Harambee, the number of panels percipients varying between the phases of the exercise and in-between online surveys. While the limited panel participants made it difficult to aggregate answers, it made the exercise more manageable, especially when it came to ensuring that responses from panellists within the given timeframe and ensuring attendance at the online panel discussion. All that being said, the experience gathered during the Harambee panel may help us to bring to the table more diverse actors in the future.

Methodology of the Panel

The OMS research team’s initial proposal for the organisation of the panel was not possible to implement due to the lack of time of panellists and the overall length of the exercise. Adapting the exercise to the constraints gradually discovered by the team allowed it to come up with a more realistic methodology that is replicable, albeit improvable. This document presents the new methodology used after the initial conception of suggested skill lists by the OMS team.

Initial Training of Participants

Before starting the panel, the Tabiya team aimed to train panel participants in using concepts used to describe the signalling value of skills and experiences. To this end, participants were required to read a prompt on their own before starting to answer the online surveys. One may find the prompt below:

Because the forms were completed asynchronously and as a result of the large data on the survey resulting in technical issues, this phase of the process survey took a long time to complete. To facilitate panellists' work, it was decided midway to hold a two and a half hours meeting, during which the research team reiterated the concepts of sof signaling value and discussed examples with the participants before filling the surveys independently. This session was deemed useful by participants.

Online Surveys

Given that the initial plan of an in-person survey was quickly proved to be unrealistic, we decided to organise this survey asynchronously through Google forms to facilitate a more efficient live discussion online. This online Google form survey was made up of 6 sub-surveys covering all ICATUS activities under the three categories of focus for the unseen economy, which are, group 3: Unpaid domestic services for household and family members, group 4: Unpaid caregiving services for household and family members, and group 5: Unpaid volunteer, trainee and other unpaid work. Within each ICATUS category, we exclude activities that are considered “other”, for instance “other unpaid domestic service for household and family members” as the skills and activities are identical to the rest within each ICATUS category. In each of 6 sub-surveys, about 8 ICATUS activities were presented to participants, and for each the activities, all of the suggested skills were shown. To make the process easier and faster in the forms, the research team pre-organised the skills into bigger groups of similar skills, such as “communication skills” or “management skills”. For each skill, respondents were prompted to assign a signalling value score ranging from 0 “no signalling value” to 3 “high signalling value”. 1 meant that they deemed the signalling value to be low, while 2 meant that they deemed the signalling value to be medium.

Because the forms were completed asynchronously and as a result of the large data on the survey resulting in technical issues, this phase of the process survey took a long time to complete. To facilitate panellists' work, it was decided midway to hold a two and a half hours meeting, during which the research team reiterated the concepts of sof signaling value and discussed examples with theel participants before filling the surveys independently. This session was deemed useful by participants. By the end of this phase, which took place from April to May, the number of responses was deemed final for the first version of skills associated with ICATUS activities even though some panellists were not able to complete the full 6 forms within the given timeframe. The final number of answers was 7 for the first two surveys, 6 for surveys 3 and 4, and 6 for surveys 5 and 6. Although low, these numbers were deemed satisfactory for a piloting exercise and did not seem to hinder discussions.

Virtual Live Survey

As a final step of the exercise, the research team decided to organize an “in-person” panel on Teams on May 30th. The two- and half-hour session included all online survey participants except one. First, the research team re-introduced the stakes and concepts used in the exercise. Then, participants were asked to discuss the signalling value of skills the answers of which had a high variance during the online Google forms survey. Participants were asked to focus their discussions on likelihood, transferability, and proficiency at first, then open their discussions to other considerations. Finally, at the beginning of the last hour of the in-person panel, the participants were asked about their views on the exercise and potential improvements.

Panel participants were presented with a supporting booklet organised in the following way. Each ICATUS activities were described (including definition, tasks included, and tasks excluded) and accompanied by a temptative list of skills. Participants were then presented with the following prompt:

The Harambee in-person panel was the last step to associate final lists of ESCO skills to ICATUS activities, using their signalling value as a discriminating characteristic. It followed the online panel that took place between March and April 2024 and required panel participants from Harambee to score skills independently. As discussed previously, the in-person panel was meant to fulfil three goals. First, overcoming the issue of the high variance in answers for certain skills and occupations, which suggested misunderstandings rather than the diversity of opinions. Second, the panel was meant to elicit the underlying decision-making rules panel participants had been using to score skills. Namely, it aimed to highlight how signalling-value sub-concepts (likelihood, proficiency, transferability) were used by intermediation professionals in the South African context, and if other considerations entered their subjective signalling value scoring function. Third, the panel format was a first for Tabiya, and it thus served as a benchmark for future work on country-specific taxonomies of occupations and skills.

Results of the Panel

The Harambee panel was deemed successful overall, although it posed multiple technical and timeline issues. Not only did it make it possible to identify final lists of ESCO skills for ICATUS activities, but it also allowed the team to adapt the list of unseen activities to the South African context. Finally, it allowed the research team to disentangle the concepts used by labour market intermediation professionals to think about the signalling value of skills, some of which the teams had not previously identified. This experience was also a useful proof of concept for a qualitative approach to taxonomy building and suggests that reproducing this methodology in other national contexts should be fruitful.

Scoring High-Variance Skills

For each activity in ICATUS, skills were scored by up to 7 Harambee members based on their signalling value over the course of the online panel. However, aggregating these scores raised questions as to the relevance of majority rules or averaging the results. Given the low number of panel participants, summing scores or selecting scores with the most “votes” would have created spurious precision. It was thus decided by the team sum the results of skills for which the answers are similar. More precisely, we defined thresholds such that skills the answers of which display a higher variance shall be discussed again to avoid spurious precision. For each skill/occupation dyad, we computed the variance between the answers of participants. Then, for each survey we selected the skills with the highest variance to present them to panel participants again. In practice, this means that all skill/occupation dyad for which the answers were half 0 “no signalling value” and 3 “high signalling value” were included in the in-person panel.

During the panel, participants were prompted the discuss the concepts defined by our team (likelihood, proficiency, transferability) before agreeing on a final signalling value score. This exercise proved to be successful. It revealed that skills for which answers displayed a high variance were typically understood in different terms by participants, or that the way the signalling value concepts were used differed. For all these skills, the panel managed to agree on a final signalling-value score, which suggests at least a minimum agreement on an underlying signalling value function. To access further information on the distribution of signalling value scores and examples of skills disused in the panel, one may refer to [insert Tina’s doc name].

Learnings about subjective signalling value functions

Despite the initial instruction for the panel to use likelihood, transferability, and proficiency as the signaling value pillars to score skills by, there was frequent debate about other components that the panel included in the scoring. The first and most prevalent other component was South African context specific considerations. The panel often relied on their understanding of the South African labour market to assign signaling value scores. In fact, they relied on this as much (in frequency terms) as the other likelihood, transferability, and proficiency indicators. The other concept that frequently came up during the panel was the panel’s expertise on the expectation of the response of the average South African entry-level employer to a given skill. Given the Harambee panel’s labour market expertise, they were particularly well placed to infer how a representative employer would react to a skill. Thus, the panel used this inference in deciding which high-variance skills to assign no, low, medium, and high relative signaling value strengths to.

The virtual workshop provided the first opportunity for panellists to engage live with one another over the survey and skills selection process. Hence, two ad hoc decisions we made by the panel. First, the panel elected to remove the term “social service users” from ESCO skills. Due to the term’s ambiguity and incompatibility with the South African context, the panel elected to have it replaced with a more intuitive term such as “people” or “dependents”. The final decision implemented by the research team has been to replace ESCO skills comprised of the term ““social service users” with similar ESCO skills that do not include the term. For example, the ESCO skill “Assist social service users with physical disabilities” will be replaced with the ESCO skill “assist individuals with disabilities in community activities”. This rule has been adopted for uniformity in all skills in the localised taxonomy related to social service users.

In addition, the ICATUS Group Level activity entitled “Tending furnace, boiler, fireplace for heating and water supply” was removed from the localised taxonomy because of the inapplicability of the activity and its associated skills to the South African context. The closest version that could replace “Tending furnace, boiler, fireplace for heating and water supply” that is found in the South African Time Use Survey at the Group Level is “Chopping wood, lighting fire and heating water not for immediate cooking purposes”. Despite the existence of this potential replacement, the panel elected that this replacement was not worth including in the finalised taxonomy due to its likely inapplicability and scarcity in the South African labour market.

During the virtual workshop, the panel expressed that it found transferability, likelihood, and proficiency to be good signaling value indicators. They found that these indicators were especially enhanced with their knowledge of the South African context. The panel members cited lengthiness and repetitiveness as the main criticisms of the process. Adding page numbers to each survey was suggested as a way to manage survey length expectations and, in turn, better allocate time to completing the survey.

Types of arguments

a. Likelihood

The argument of the likelihood that a jobseeker has a skill was used very frequently during the discussions. The arguments relating to the likelihood of a jobseeker having a skill were based varied among examples. In most, they were based on the knowledge respondents have about the South African context. For instance, “teaching young horses” was deemed unlikely in link with the “daily pet care” activity, with respondents acknowledging that they would expect the likelihood to be higher in a European context. Another example is that driving skills were deemed to have a low likelihood in link with activities consisting in accompanying dependent adults or children, as participants stressed that moving around entailed less car rides than it would in Europe. In some other cases, the likelihood that a jobseeker has a skill based on an activity is deemed to be low because the skill tag is conceptually remote from the activity title. For instance, although a young jobseeker volunteering in a church could possibly participate in accounting activities, accounting skills seem to be conceptually to far remote from the idea an employer may have of volunteering in a church. Finally, respondents stressed that to answer they had to assume the exact activity young jobseekers were doing. For instance, volunteering in another household’s firm is not precise enough for an employer to infer what skills the jobseeker may have. This calls for a further improvement of the unseen taxonomy.

b. Transferability

Transferability was another of the main concepts presented to panel participants. Therefore, the concept was also widely used during the conversations. Sometimes, the concept was used in the right way. For instance, preserving food has been deemed to be weakly transferable, as professional food preservation relies on scientific principles and rules that one does not typically use in the unseen economy. However, the concept of transferability was also mixed up with other concepts multiple times. For instance, it was mixed up with likelihood, when participants stated that a given experience “did not transfer” to a skill. It was also used concomitantly with proficiency. When is skill was deemed to be basic, for instance tending grass, then it as deemed transferable. Overall, panel users seemed to deem skills to be on average more transferable than likely, but the two concepts appeared as equally important.

c. Proficiency

The concept of proficiency was almost not used by panel participants. More precisely, it was only invoked through the transferability of skills, when participants guessed that basic skills were more transferable and thus had more signalling value. Otherwise, discussions around how good a jobseeker might be able to get at given skills through given activities was not discussed.

d. Labor market demand

Discussions around labour demand were rather unwanted by the team prior to the panel, as the research team hoped that “what employers want” would not influence the making of skills lists for ICATUS activities. However, some interesting conversations happened where panellists stressed that some unseen experiences could have a very signalling value for very specific employers. They thus deemed it important to keep certain skills visible on our skills lists, to allow jobseekers to signal them. On the contrary, some skills were left aside as participants deemed the demand for it to be low.

e. South African context

As described before, the South African context was mobilised multiple times along with the concept of proficiency, usually to argue that a skill or occupation was not relevant to the South African context. The later was unexpected, as the panel decided to delete “Tending furnace, boiler, fireplace for heating and water supply” and “Accompanying non-dependent adults” from the list of unseen activities altogether. Considerations around the South-African context fed into discussions about other concepts such as likelyhood and transferability, for instance through qualifications. Indeed, some skills were deemed not to be transferable on the ground that they require qualifications in South Africa.Such an account for the South African context is perfectly aligned with what the research team hoped the panel would discuss.

f. Stretch between the activity title and the skill

Participants stressed multiple times that although skills relating to an activity might be likely and transferable, they are sometimes too far remote from each other. Therefore, an employer would be unlikely to think about skill A if a young jobseeker came with the unseen experience X. Conversely, some individual skills were deemed very valuable in themselves, but the panel thought that it is unlikely that a user picking the activity they are associated with would do it to pick them. Namely, some skills are unlikely to be picked, not because they are not valuable or transferable, but because picking them would require picking activities that users are unlikely to choose. This finding was of great relevance for the work of Tabiya, as it suggests that a taxonomy like ESCO may not be flexible enough for jobseekers to surface their human efficiently.

g. Naming of skills

Finally, multiple comments from participants referred to the naming of skills, that was deemed to make skills un-pickable for young jobseekers. In particular, many skills refer to “social services users”, for instance “assist social service users with physical disabilities” or “social service users to live at home”. Although the human capital signalled by these skill-tags was deemed to be relevant by panellists, they noted that the added precision of “social service users” made the skill less relevant, as it refers to a much more formal environment than the ones young jobseekers from the unseen economy are working in. In practice, skills like “assist social service users with physical disabilities” were deemed to have a low signalling value, although it was recognized that a skill like “assist individuals with physical disabilities” would have received a higher score. Yet, these skills do not have neutral equivalents that do not refer to “social service users”. Therefore, it was suggested to create new skills.

Takeaways about the signalling value function

The in-person panel proved useful to understand how Harambee workers think about the signalling value of skills in relation with occupations. Although we imposed three concepts (likelihood, proficiency, transferability), panellists appeared to find them useful, although proficiency was discussed less than the two others. Interesting remarks that feed into our concept of signalling value include the importance of qualifications (a skill may have a proficiency, transferability, and likelihood score but send a weak signal because it typically entails formal training). Second, it was interesting to see that skill tags influence the signalling value scores when they become too specific to a given context (social service users).

Learnings for future country-specific taxonomies

The in-person panel served as a proof of concept for further use of this qualitative methodology to adapt ESCO to local taxonomies. The exercise was deemed successful overall, albeit time-consuming. Unexpectedly, the exercise also brought about discussions that seem to call for solutions like Compass to overcome the rigidity of taxonomies.

Iterations of the panel exercise

Discussions surrounding the South African context, rather they touch upon what unseen activities young jobseekers might do or upon qualifications and demand side needs brought important insights used in building the “unseen” part of our localized taxonomy. Therefore, it gave encouraging results when it comes to using the same methodology in our next countries (France, Ethiopia…).

When it comes to organizing the panel, the research team ended up changing its initial plans multiple times. Online surveys proved to be very time consuming for respondents, who explicitly suggested that the research team shortens lists more than it did already to facilitate the work of panellists. The in-person panel on the other hand seems to have been liked by panellists and was fruitful. The suggestions for further panel iterations are the following:

1 – Apply the skills selections rules more stringently before presenting them to the panel, to reduce the time needed to answer the online surveys.

2 – Organise an in-person panel session to answer the first survey, along with the training session that was provided in this iteration. This was suggested by panellists so that the variance in answers is lower.

3 – Panellists should initially be asked which of the ICATUS activities they deem relevant for their local context, to avoid the situation of “tending furnaces”.

4 – Online surveys should include an indication of the overall length of the survey.

Compass

It is to be noted that this panel exercise suggested the need for a solution like Compass. Panellists stressed multiple time the lack of connection between skills and the ICATUS activities that they were linked with. While recognizing that some skills could have a high signalling value and be very relevant in the South African labour market, they considered that young jobseekers were unlikely to be able to highlight them because they were unlikely to pick certain ICATUS activities. This highlights multiple sources of rigidity in localized ESC0 taxonomies. First, our taxonomy is based on ICATUS, which covers all productive non-paid time uses but groups them in buckets that may not speak to platform users. Second, UX constraints call for the lists of skills associated to each ICATUS activity to be limited, which necessarily leaves aside certain skills. Therefore, the unseen taxonomy does not include all the skills that a young jobseeker might want to highlight, which is a common issue with ESC0 and results from the trade-off between a taxonomy that is inclusive (includes many skills for all occupations) and informative (limits the number of skills associated with each occupation). One solution to this is to have a flexible algorithm, allowing jobseekers to find skill through their experiences but also ad hoc, and thus highlight the full range of their human capital.

Online Survey Results

Six Google Forms surveys were administered to elicit a signalling value score from the Harambee panel members. This survey was filled in between the 15th of March 2024 and the 17th of May 2024. Once the survey completion deadline lapsed, we calibrated the answer “no signalling value” as zero, “low signalling value” as one, “medium signalling value” as two and “high signalling value” as three. This calibration allowed us to visualise each survey's distributions and easily quantify which skills sparked the most dissimilar answers from panellists (namely, which were the skills for which the answers had the highest variance). For analysis coherency, Surveys 1 and 2, which roughly (there were at most two ICATUS Group Level Activities that belong to ICATUS Division 3, 4 or 5 that were not found int their respective surveys due to survey capacity. These individual activities did not dramatically influence the shape of the distribution) represents ICATUS Major Division 3 were combined in our analysis. Similarly, Surveys 3 and 4, which roughly represent ICATUS Major Division 4, were combined, as were Surveys 5 and 6, which roughly represent ICATUS Major Division 5. This compartmentalisation is also done for simplicity since Surveys 1 and 2 received seven response submissions, Surveys 3 and 4 received six responses, and Surveys 5 and 6 received five responses.

Distributions

Each survey’s distribution is shown below. Each ICATUS Major Group has relatively different distributions. This is likely driven by the fact that the survey respondents’ sample size changed within each major division in an already small panel population pool. Note that the signal values are the aggregation of individual panel respondents’ answers. Therefore, Figure 1’s signal value axis is larger than Figures 2 and 3, which each had fewer respondents.

Variances

The variance of answers associated with each skill was particularly important because of their applicability to the live panel workshop discussion. The rationale used was that skills with the highest signaling score variances would be the most worthwhile to attain agreement on amongst all the survey respondents, given that this exercise could only be performed on a subset of all the skills chosen during the survey.

We isolated 20 of the skills with the highest variances per ICATUS Major Division. Ultimately, though, the skill with the 20th highest variance within each Major Division shared this variance with at least four other skills. Thus, the list of 20 high-variance skills expanded in each Division. A total of 26 skills ultimately comprised the high-variance skills for ICATUS Major Division 3, 24 for ICATUS Major Division 4, and 47 for ICATUS Major Division 5.

Following the virtual workshop, the high variance skills that were not discussed were assigned inferred signal scores based on the learnings from the panel. These scores were sent to three Harambee panel members for verification before their final incorporation into the taxonomy.

Virtual Group Discussion Results

A virtual two-and-a-half-hour workshop to discuss skills with the highest occurring variances per survey was hosted on the 30th of May 2024. Six of the seven survey respondents joined the session (though some respondents had to move into and out of the workshop). From this session, 12 skills from Surveys 1 and 2 (ICATUS Major Division 3) were discussed, nine skills from Surveys 3 and 4 (ICATUS Major Division 4) were discussed, and nine skills from Surveys 5 and 6 (ICATUS Major Division 5) were also discussed (this does not include skills and activities which need rewording or will be deleted from the taxonomy).

Of the 12 ICATUS Major Division 3 skills discussed, 83.3% were deemed to have no signaling value (a signaling score of zero), none of the skills were deemed to have low signaling value (a signaling score of one), 16.7% were deemed to have medium signaling value (a signaling score of two), and none of the skills discussed obtained a high signaling value. For ICATUS Major Division 4, 55.6% of the total ten skills were assigned a no signaling value score (a signaling score of zero), none were assigned a low signaling value (a signaling score of one), 22.2% were assigned a medium signaling value score (a signaling score of two), and 22.2% of the skills discussed obtained a high signalling value score. Finally, 66.6% of the nine ICATUS Major Division 5 obtained a no signaling value score (a signaling score of zero), and the remaining 33.3% was equally shared between low, medium, and high signaling value scores amongst the discussed skills. Figure 4 depicts the workshop’s signal value discussion over all ICATUS groups. It is clear from the figure that skills that were heavily contested in the survey were ultimately deemed to have no signalling value once panel members had a chance to pose arguments for and against these skills.

Final Skills Selection

Once we received the remaining high variance skills’ scores, we adopted a “traffic light” decision rule. That is, 33.3% of the skills with the lowest signaling value scores within an ICATUS Group (highest level of disaggregation) form part of the red category. These skills are removed from consideration on the SA Youth platform. 33.3% of the skills with the highest signaling value scores form part of the green category. These skills are surfaced at the top of the list of skills options that SA Youth users can select. The mid-range signaling value scores comprise the remaining 33.3% and are in the orange category. These orange category skills are surfaced beneath the full list of green category lists on the SA Youth platform. Below are three diagrams that show the absolute number of skills that will remain due to the enforcement of the traffic light decision rule for each ICATUS Major Division.

Experience has shown that young jobseekers struggle to identify skills that are the most relevant to their professional occupations and hobbies, or the skills at which they are the most proficient. Yet, helping them navigate ill-functioning labour markets entails surfacing the full length of their human capital – namely, the full set of skills they possess -, while helping them to present it in a strategic manner to improve their job seeking perspectives. To help jobseekers do so, Harambee hopes to rely on static surveys prompting young jobseekers to select, for each of their occupation / activity, the five skills that they are the most proficient at. To do so, they will be presented with lists of ESCO skills associated with their activities, or with skills that have been selected through the work of Harambee and the Oxford Martin School (for micro-entrepreneurship and unseen activities). This raises the question of the optimal way to present skills to job seekers, to ensure that 1 – we cover their whole human capital, 2 – they can effectively pick relevant skills to populate their inclusive CVs, 3 – this process is efficient. Yet, evidence has shown that the way individuals are prompted to pick answers out of list of options impact their final decisions. In this note, I summarize some research pieces to inform how the lists of skills submitted to jobseekers by Harambee and OMS may influence the decisions.

Choice Overload and Underload

These notes are based on « Cognitive and Affective Consequences of Information and Choice Overload” by Reutskaja et al.

Harambee has expressed concerns that the number of options presented to job seekers may overwhelm them, making their experience on the platform anxiety inducing and the results inefficient. These concerns are aligned with insights from behavioural science, focusing on the concept of “choice overload”. Intuitively, individuals presented with too many options to pick from may find it harder to pick ad experience anxiety, unsatisfaction, loss of motivation to pick, stress, or even choice paralysis. It is to be noted nevertheless that the reverse phenomenon, choice underload, may lead to frustration as well, as the desired options may not be presented to individuals. Behavioural theory suggests that the level of satisfaction felt by individuals prompter to pick a choice out of several options follows an inverted U-shape. Namely, the optimal number of options to give individuals is not the maximum.

The intuition behind this is the following. When presented with an increasing number of options, the needs of people are more likely to be satisfied. In our work, more options means that Harambee’s young job seekers are more likely to find the skills that they are the most of proficient at among the ones we present them with. This abundance of choices can also be beneficial from a psychological point of view. In our case, it may comfort users into the idea that they have large sets of skills, and thus allow them to have a more positive approach in their job search. On the other hand, large sets of options have psychological and time costs. Studies have shown that our minds do not account for the full sets of available information when taking decisions, but only subsets (Wedel and Pieter, 2007). Following Reutskaja et al’s terminology, this “inattentional blindness” means that jobseekers presented with long lists of skills risk ignoring skills that they would have picked in smaller sets of options. In the case of Harambee users, a large set of options may also induce stress, especially linked to the sentiment of forgoing important skills. Worse, it may lead to a loss of motivation to answer the prompt correctly, and to users selecting skills by default. This evidence suggest that Harambee is right to worry about the number of skills presented to young jobseekers being too large.

Ordering of Skills

Once the lists of skills presented to young jobseekers have been narrowed down, the order in which the skills are presented is likely to impact their picks. As Harambee and the teams in Oxford are considering ordering the skills based on concepts such as transferability (essentially, the frequency with which a skill is associated with an occupation in ESCO), understanding how that order may influence users’ pick is critical. Intuitively, one could think that users tend to pick the first skill presented to them, especially when they must take an action (click “see more”, for instance) to see other options. Studies researching the effect of the order of questions in online surveys suggest that order does matter. A study conducted in the US in 2008 found that low-education respondents who answer surveys quickly – which is likely to be the case of Harambee users – a most prone to primacy effects (bias toward selecting earlier response choices). A study conducted in Xiao et al in 2024 also observes such bias, while finding that it is inexistent in “objective questions”, for instance those pertaining to demographics.

The implications for the Harambee platform are not straightforward. On the one hand, Harambee may seek to nudge job seekers into picking skills that Harambee’s counsellors consider as better picks. This would imply putting these skills on top of the list of options. On the other hand, Harambee may choose not to try and influence users’ picks. However, a random order would effectively lead to answers biased toward the first options presented. Countering this bias is not evident, which may be another argument to choose to hierarchize skills.

Prompt

The evidence previously quoted finally suggest that choice overload is different from information overload. Namely, the sentiment of discomfort felt by individuals in choice overload does not come from the number of options only, but also from being prompted to choose a subset of options. Intuitively, the disproportion of the option set with the allowed number of picks – too many options, too few picks – makes choosing uncomfortable. This is likely to be especially true in the case of Harambee users selecting skills, as choosing skills to populate a CV is consequential. Being prompted to choose 5 skills is likely to create two issues: 1 – Encourage people to select 5 skills although they are not proficient in some of them. Picking less skills may make users lose confidence by suggesting they have few skills. 2 – Make users feel like they are not able to surface enough of their human capital, which may create frustration.

However, it is unrealistic to populate CV’s with more than 5 skills per occupation. Moreover, the occupation title also sends signals to employers about jobseekers’ skill sets. Therefore, we only recommend adjusting the prompt from “Pick the five skills that you are the most proficient at” to “Pick up to five skills that you are the most proficient at”.

All in all, evidence appear to justify narrowing down the lists of options presented to young jobseekers on the Harambee platform. It also gives evidence that the order of options matters, which may inform decisions taken by Harambee. Finally, it suggests choosing a prompt carefully, so as to not induce stress or demotivation on the platform’s users.

Bibliography

Neil Malhotra, Completion Time and Response Order Effects in Web Surveys, Public Opinion Quarterly, Volume 72, Issue 5, December 2008, Pages 914–934, https://doi.org/10.1093/poq/nfn050.

Reutskaja, E., Iyengar, S., Fasolo, B., & Misuraca, R. (2020). Cognitive and affective consequences of information and choice overload. In Routledge handbook of bounded rationality (pp. 625-636). Routledge.

Thaler, R. H., & Sunstein, C. R. (2008). Nudge: Improving decisions about health, wealth, and happiness. Yale University Press.

Wedel, M., & Pieters, R. (Eds.) (2007). Visual Marketing: From attention to action. Hove: Psychology Press.

Xiao, Yaqing and Yan, Hongjun and Davidson, Erik, Response Order Biases in Economic Surveys (February 2, 2024). Available at SSRN: https://ssrn.com/abstract=3786894 or http://dx.doi.org/10.2139/ssrn.3786894.

Presentation of ICATUS

ICATUS stands for International Classification of Activities for Time-Use Statistics (ICA-TUS). It classifies "all the activities on which a person may spend time during the 24 hours that make up a day" (UNSTATS). Therefore, it allows for nation-wide and cross-country comparisons of time-uses.

ICATUS activities are meat to partition the univserse of conceivable daily activities in an exclusive maner: each activity can only fall into one ICATUS category, and one only. Therefore, Tabiya and Harambee chose to rely on the ICATUS taxonomy in order to channel it's strong conceptual structure into their work on the unseen economy.

However, ICATUS activities are not meant to be consistent with ESCO or ISCO. Namely, they are not directly associated with ESCO occupations, nor skills. The objective of Tabiya's work on ICATUS activities was to match the latter to ESCO skills. This would allow Harambee's users to select their main time-uses and be presented with candidate skills to highlight on their CVs and inform our matching algorithm.

Assigning ESCO Skills to ICATUS Activities

In order to match ICATUS activities with ESCO skills, the team was presented with two options: matching ICATUS activities directly to ESCO skills using natural language processing, or matching ICATUS activities to ESCO occupations to derive candidate skills from these matches. Given the scarcity of the readily available data on platform users' description of their daily tasks, the second option was selected.

ICATUS activities were thus each matched to up to 4 ESCO occupations manually by a team of 3 researchers. These matches then allowed the team to derive a candidate list of skills for each ICATUS activity by the ESCO taxonomy.

Assigning ESCO occupations to ICATUS activities was based on similarity between their definitions. For instance, "preparing meals and snacks" in ICATUS involves - according to its definition - tasks that are those of "cooks" and "kitchen assistants" in ESCO. Therefore, the complete list of ESCO skills associated with "cooks" and "kitchen assistants" was assigned to "preparing meals and snacks" in ICATUS.

Reducing the Candidate List of Skills

The taxonomy obtained following this method is hardly useable. Indeed, each ICATUS activty ended up being associated with numerous skills. In order to provide a manageable list of skills to job seekers on the Harambee platform 0.1, the team decided to reduce this temptative list of skills. reucing this also allowed to addressed the transferability and signalling issues inherent to the unseen economy. To do so, an approach based on a Panel of professionals involved in the South-African labour market and in intermediation instances (among which Harambee) was chosen. The panel approach is meant to gain context-specific insights from a wide range of stakeholders. This would make it possible to overcomestake holders' biases pertaining to the transferatility and credibility of skills. In order to test this approach, we first implemented a panel at the scale of Harambee.

For the panel to happen efficiently (in a time limit of 5 hours for the pilot panel), the Tabiya team proceded to a first round of skill selection in order to delete skills that were deemed obviously not consistent with the definitiona nd description of ICATUS activities. To do so, the team followed 4 easily applicaable rules.

Isolate Knowdleges

The ESCO classification distinguishes between "skills" - that all describe an action - and "knowledges". For instance, the occupation "cook" is asociated with the skill "use cooking techniques" and the knowledge "cooking technique". As a first step, we chose to isolate knowledges and to mainly focus on skills. However, we chose to let the panel decide to keep certain knowledges when they brought important new informations. For instance, "common children's diseases" may be deemed an essentio

Delete irrelevant skills:

In ICATUS, each activity is associated with a definition, and explicit list of tasks included in the activty, a list of tasks that are not included, and one or more examples. ESCO is built similarly. However, IACTUS activities are typically not only broader, but conceptually different from ESCO occupations. For instance, "Budgeting, planning, organizing duties and activities in the household" is not a formal sector activity that is included in ESCO. Therefore, to assign ESCO skills to this ICATUS activity, it was matched to "Office clerk", "Accountant", "Bed and breakfast operator", three ESCO occupations that encompass all tasks involved in "Budgeting, planning, organizing duties and activities in the household". The issue is that they encompass more. For instance, because of "bed and breakfast operator", the skill "serve beverages" endded up being associated to "Budgeting, planning, organizing duties and activities in the household", even though serving atsks are not included in the description of the ICATUS activity. When we came across such cases, we decided to exclude the skills that had been unduly associated to the ICATUS activities.

• Delete skills that are deemed too formal: some of the skills associated with ESCO occupations directly refer to situations or tasks only imaginable in the seen economy, whether it is formal or informal. For instance, "maintain customer service" cannot be appropriately associated with "serving meals and snacks", as it describes serving meals and snacks to one's own children/family members.

• Delete redundant skills: ESCO occupations are typically associated with numerous skills, and each ICATUS activity was associated with mutltiple ESCO occupations by Tabiya's team. The lists of skills from each ESCO activity was added to a list of skills for each ICATUS actity, with deletion of skills that appear multiple time (where we only kept one appearance). Therefore, the temptative lists of skills associated to each ICATUS activty contain skills tht may be deemed redundant. For instance, "Outdoor cleaning" was associated with both "prune plants" and "prune hedges and trees". We considered that asociating both skills to "outdoor cleaning" did not bring new information.

Although it is constently improved to include more occupations and skills, the ESCO classification does not yet - and will probably never - cover the full universe of skills and occupations in Europe and, a fortiori, in the world. Working with Harambee and working to build an inclusive taxonomy for the South African youth thus called for solutions to adapt and supplement ESCO. These challenges provided us with valuable insights on reproducing our methodology in other countries.

Internal Inconsistencies in ESCO

One of the main takeaways of our work for Harambee is already that ESCO presents some internal inconsistencies. The ESCO classification results from the work of experts, as described above. This human based methodology has advantages:

• It allows to avert a data-mining (CVs, job offers) exercise based on data of unequal quality. Job offers and CVs also notoriously overestimate the skills associated with a given occupation.

• Relying on data-mining would have resulted in a taxonomy highlighting the skills that are frequently sought for by employers and displayed by current workers. Therefore, it would tend to invisibilize rare associations of skills and occupations.

However, relying on focus groups also creates consistency issues. For instance, the ESCO research group has highlighed the existence of . Because there are multiple reference groups, some conceptually similar skills were created for different occupations under different names. More broadly, we faced two issues. First, the decisions rules used to assign skills to occupations are unclear. Second, using a taxonomy like ESCO entails decisions about a tradeoff between the representativity and the inclusivity of the taxonomy.

Do we need to extend ESCO?

The ESCO taxonomy under its current appears to cover most of the "seen" occupations found in the South African labour market, and the same can be said of skills. The data collected among microentrepreneurs suggests that the latter rarely claim to have occupations or skills that are not already covered by ESCO. Moreover, the Harambee platform aims at offering an "other occupation" and "other skill" option. Therefore, collecting data directly from a large number of job-seekers will allow to evaluate the overall representativity of ESCO in the South African labour market.

However, our work with Harambee in South-Africa already convinced us that extending ESCO to include the unseen economy should be at the core of our work, as it allows to build a more inclusive taxonomy. Collecting data about users usage of this part of the taxonomy will allow us to measure the importance of this tool to cover the full range of human capital in the South African labour market.

Adequacy of ESCO for non-European Economies

When it comes seen activities, the microentrepreneurship survey suggests that ESCO satisfactorily covers skills and occupations found in the South African labour market.

This page summarises a proposal for how to implement A/B testing of how to elicit skills for SAyouth.mobi users’ inclusive CV. This research is motivated by concerns that users may struggle to select between different skills, and an interest in comparing different potential metrics for skill priority. The proposed research is an on-platform randomised evaluation of how users are asked to capture the skills relating to their work experiences.

Background

In the search to highlight and surface the skills South African youth have as a result of the activities they have performed and are currently involved in, we conducted a survey that asked microentrepreneurs to rate the importance of different skills associated with their occupation. Specifically, we asked about 1500 microentrepreneurs “On the 1 to 5 scale where 1 is ‘Not important at all’ and 5 is ‘Very important’, how important is [skill tag] to your work?”. The results showed a tendency for participants to assign high importance to all skills presented to them.

This poses a potential challenge for the Harambee platform. If job seekers prioritize all skills equally, they might be overly reliant on the initial set of skills they encounter during their platform search. This could lead them to select skills that aren't necessarily the most relevant for their desired careers, hindering their job search effectiveness. To address this concern, our research aims to investigate how the order of skill presentation on the Harambee platform influences job seeker choices.

Thus we seek to answer the following questions:

1. What is the impact of presenting skills to platform users in different orders on the skills selected?

2. Is there an order that offers users the best support in their job search journey?

3. Do different skill priorities have downstream impacts on job search patterns?

4. Do different skill priorities have an impact on employer interest in the candidate?

5. How often should we ask users to update their skills?

Proposal

We propose implementing on-platform A/B testing of how to surface skills to users. This testing should help us to evaluate how users respond to different presentations of skills and consequently refine how best to elicit skill reporting from platform users.

The ideal method for testing skills surfacing is evaluating actual user skill selections on platform. This would be achieved by comparing how similar (statistically comparable) users select skills when presented with different list orders and lists of varying numbers of skills. Thus, we wish to randomly assign users to be presented with different lists when they are asked to select the skills they wish to include on their CV.

We propose a number of potential skill priority metrics:

1. Random This serves as a neutral benchmark. It is also the most theoretically “hands off”.

2. Transferability score Transferability score = number of appearances of this skill in ESCO / number of occupations in ESCO . This may nudge users towards listing skills that are commonly used in many other opportunities, or in jobs commonly listed on the platform. If users are encouraged to select skills that are more transferable, we may be able to improve the visibility of skills that firms are interested in knowing about.

3. Inverse scarcity of skills Inverse scarcity of skills = inverse of number of other users with that score, conditional on the number being above 10 (to rule out idiosyncratic skills). More rare skills may also help job seekers to stand out more from the rest of the field of applicants, and help firms differentiate between job seekers more easily.

4. Some weighted average of these goals Its likely some mixture of these priorities is appropriate. It may also be that different jobs or different fields may require different weightings of the set of priorities. Developing a preferred weighted average may require some desk work to get input from field experts and literature about relative importance of different measures for particular industries or worker experience profiles.

5. “Other” field to search for skills Offering users the chance to describe their own skills offers another extreme hands off benchmark. This arm can also be used to validate the accessibility of the language used by the taxonomy to describe the skills surfaced to users.

6. More skill options For each of the above options we propose surfacing five skills. In this option we propose extending the list to surface more skill options for users to choose from.

We will follow a stratified random sampling strategy based on gender and the type of work the individual has performed (i.e. currently unemployed, microentrepreneur, and employed). The sample will be randomly assigned into treatment groups and each group will be presented with a pre-ranked set of skills. The testing could run for one to three weeks depending on logistical constraints for Harambee and the target sample size for each treatment cell. A/B testing in principle does not require detailed power calculations because patterns are informative even if they do not cross the threshold for academic statistically certainty. Samples as small as 20 users per arm could already offer indicative patterns, although on-platform may make it easy to reach far more users. A larger sample would naturally offer the advantages of more precise estimates of effect and the opportunity to account for the role demographic characteristics may play in mediating responses to different skills list orderings. We propose two levels of randomisation.

Across Users Effect

Randomise at the day/week level. While we prefer a day level randomisation, we are happy to work with logistical constraints of bringing various lists live on the platform. We could do every two days, or every week if one of these is easier. Each new randomisation block we will apply a different skill priority ranking. Cycle through each of the treatment arms repeatedly over 3 weeks. This will generate a treatment arm for each list priority order and therefore enable the central comparison of different potential skill priorities.

Within Users Effect

Once a month prompt an opportunity to review skills reported previously and surface skills in a randomly chosen priority (may be the same or different priority users have previously been asked). This will enable us to examine whether the same user reacts differently to receiving skills in different orders, and for users who are shown the same list twice, how stable elicitation is. A measure of skill reporting stability is useful because it can be used to decide how often the platform should prompt users to update their skills or CV. There are a number of reasons to believe that this will be less useful information than the primary randomisation, however, it will be important that the platform has the technical capacity to encourage users to modify which skills they are capturing in their inclusive CV so that we can implement the findings of the A/B testing for platform users in the control group.

Measurement

Many of the indicators of which list to prefer can be collected from administrative data already on the platform. The primary measure of impact is the simple comparison of skills chosen under different treatment conditions. Statistical properties of the skills chosen offer information a number of dimensions of choice quality:

1. Correlation between skills chosen under different list orders can tell us the degree to which the different orders make a difference to the skills ultimately chosen.

2. Variation in skills chosen within a single priority type can tell us whether all individuals are behaving the same way, indicative of mechanical clicking/less engagement with individual matches with skills or thoughtful engagement.

3. Measures of how often the first three skills, or the first skill, or the last skill, or the skill most centrally displayed on the screen can capture indicators or mechanical clicking, or reduced engagement.

4. We could manually review a subset of respondents to evaluate skill choice against their profile.

5. UX measures of time to complete fields, number who start and abandon part way through.

Demographics which may mediate how users interact with skill lists can also be drawn from user profiles. We would likely focus on:

• Gender

• Date of birth (Month and Year)

• Occupation/activity

• Education level

• Geographic location

• Experience history (previous years employed)

• Disability status

All of this information is available for users on the platform, so it would simply be a question of pulling the relevant administrative data.

If there is interest in using downstream outcomes to adjudicate between list priorities, we could also draw data on which jobs individuals who have been exposed to different lists click on and apply for. We could also ask firms to select skills they would like to see prioritized among platform applicants from similarly ordered lists to obtain a measure of firm preferences.

We could also collect additional data either using on-platform prompts, off-platform follow up surveys, or some small focus groups. Additional data would enable us to measure user satisfaction with the different lists, elicit user criteria for how they select between skills and evaluate how confident or committed users are about the skills they selected.