This is an interim intervention report. We have spent limited time to form an initial view of this program and, at this point, our views are preliminary. We plan to consider undertaking additional work on this program in the future.
- What is the program? The program consists of providing workers with certifications for non-cognitive skills, including traits such as trustworthiness and willingness to help others. It aims to increase workers’ earnings by improving their ability to signal soft skills to potential employers.
- What is its evidence of effectiveness? We have not found strong evidence that soft skill certification increases earnings. A randomized controlled trial (RCT) conducted in Uganda reports an increase in earnings of about 8% that is imprecisely estimated (our preferred specification of the key result is not statistically significant). Our best guess is that at least some of the benefits enjoyed by people who receive the certification are explained by losses incurred by others (e.g. jobseekers without certification on their application who seem lower-skilled despite having similar skills), but this effect was not tested in the study.
- How cost-effective is it? The evidence for the effectiveness of this program does not yet seem robust and we expect that it would have smaller effects at scale, so our best guess is that it is not in the range of cost-effectiveness of our priority programs.
- Does it have room for more funding? Very roughly, we would guess there are a large number of young job seekers and firms that may be able to benefit from this kind of program across Uganda if the program were effective. We are unsure as to whether there are organizations interested in, or able to, deliver the program at scale.
- Bottom line: We deprioritized this program because it does not stand out in terms of evidence of effectiveness (and therefore cost-effectiveness). Major reasons underlying our best guess that its cost-effectiveness is likely to be lower than our priority programs are: 1) our best guess is that a meaningful portion of the benefits of the program would be offset by losses incurred by people who do not receive the program (negative "general equilibrium/spillover" effects); 2) the key effects of this program seem relatively small and imprecisely estimated, and we would guess that there is a reasonable chance that they would not replicate; 3) we expect that the program is likely context-dependent and may have smaller effects in settings and scales outside of where it has been previously studied.
Published: May 2020
- What is the problem?
- What is the program?
- What is the program’s evidence of effectiveness?
- Is the program cost-effective?
- Room for more funding
- Should the investigation move further? If so, what are the next steps?
What is the problem?
Evidence suggests that jobs in developing and emerging countries are often poorly paid and unstable, and wages increase more slowly than in high-income countries.1 The International Labour Organization (ILO) estimates that, in 2017, around 730 million working people living in these countries were living on an income of less than $3.10 purchasing power parity (PPP) per day.2
One of the factors that might constrain workers’ ability to find better quality jobs is information friction.3 In particular, it can be hard for workers to signal their skills.4 These constraints might be especially relevant for young people, who have less experience and credentials.5
Caria and Lessing 2019, which discusses recent interventions tackling information friction, highlights the program discussed in Bassi & Nansamba 2020 as a potentially cost-effective way of addressing this problem.6 This motivated us to investigate the intervention.
What is the program?
The program consists of providing job seekers with soft skills certification. Soft skills are non-cognitive skills, such as trustworthiness, creativity, and willingness to help others.7 Assessment of most of these skills is based on teachers’ observations, while creativity was measured through a psychological test and trustworthiness through a trust game involving real money.8
This report focuses on the intervention described in Bassi & Nansamba 2020. The intervention took place in Uganda and included three components: (i) testing job-seekers’ soft skills, (ii) scheduling meetings between job-seekers and firms interested in recruiting, and (iii) providing job seekers with soft skills certificates. If workers attended meetings with firms, the certificates were presented to firms and workers during that meeting. Otherwise, certificates were disbursed to the worker after the first follow up survey.9
The intervention targeted young workers who had completed vocational training, and matched them with small and medium enterprises (SMEs).10 It focused on the following sectors: carpentry, catering, hairdressing, motor-mechanics, tailoring, welding.11 The program was delivered by BRAC, a large international NGO.12
What is the program’s evidence of effectiveness?
We judge the evidence for the program to be overall weak. The main source for the effectiveness of the intervention is a randomized controlled trial (RCT) (Bassi & Nansamba 2020) that reports an increase in earnings of about 8% that is imprecisely estimated (our preferred specification of the key result is not statistically significant). We are aware of other studies relevant to the program studied in Bassi & Nansamba 2020, but the interventions tested are sufficiently different that we do not consider them to be strongly informative of the effectiveness of this specific program.
We expect that a large part of the benefits of the program for treated job-seekers may be offset by negative effects on others. This is because a meaningful fraction of improvements in earnings of treated job-seekers may be due to a transfer between treated job-seekers and untreated jobseekers or firm owners, rather than being the result of an overall increase in productivity.
Results of Bassi & Nansamba 2020, the key study for this report
Our assessment of soft skills certification mainly focuses on Bassi & Nansamba 2020, an RCT conducted in Uganda. The trial took place between 2014 and 2017.13 The sample included 787 job-seekers and 422 SMEs.14
The intervention included three components: (i) testing job-seekers’ soft skills, (ii) scheduling meetings between job-seekers and firms interested in recruiting, and (iii) providing job seekers with soft skills certificates. If workers attended meetings with firms, the certificates were presented to firms and workers during that meeting. Otherwise, certificates were disbursed to the worker after the first follow up survey. Control areas received the first and second components, so the study tested the effectiveness of the third component (providing certificates) only. Data was collected in two surveys, which took place 12 and 26 months after the intervention.15
The intervention seeks to improve job seekers’ earnings by making it easier for workers to signal soft skills, and for managers to verify them; this aims to improve the quality of the match between job-seekers and firms, thus increasing firms’ productivity and workers’ wages.16
Pooling results from the two surveys, the program found a roughly 8% (95% confidence interval -5.3% to 21.4%) increase in total earnings for all workers.17 This effect is imprecisely estimated and the 95% confidence interval substantially overlaps zero. Our best guess is that if this study were replicated in a larger sample that it would find substantially smaller effects. We do not think this result provides strong evidence in favor of the effectiveness of the intervention, since our estimate suggests a p-value of 0.24, which is considerably higher than levels conventionally taken to indicate statistical significance.18
We focus our analysis on the estimated effect for all workers (above) rather than the effect highlighted in the study's abstract (which is restricted to the sample of workers with positive earnings) because (a) we see the former as most relevant for analyzing the effects of the program on the full population targeted by the intervention, which seems most relevant from a policy perspective (the author of the paper agrees with us on this point), and (b) we see the latter as a case of subgroup analysis that risks overstating the true effect. More detail in the following footnote.19
Are results from other studies informative about the effectiveness of this intervention?
We are aware of other studies related to the soft skills certification program studied in Bassi & Nansamba 2020, but the interventions tested are sufficiently different that we do not consider them to be strongly informative of the effectiveness of this specific program.
Aside from Bassi & Nansamba 2020, we are aware of the following evidence related to this program:
- Carranza et al. 2019, an RCT assessing the effectiveness of certification of both hard and soft skills in South Africa.
- Abebe et al. 2018, an RCT assessing the treatment and displacement effects of an intervention including a one-day orientation (focusing on resume writing, application letters and job interviews) and certification of hard skills in Ethiopia.
- Groh et al. 2015, an RCT testing the effectiveness of a job-matching service based on educational backgrounds and psychometric assessment (including both hard and soft skills) in Jordan.
- Abel et al. 2016, an RCT assessing the effect of providing reference letters for job-seekers in South Africa.
- Pallais 2014, an RCT testing the effect of providing job and performance feedback in an online marketplace.
We have not looked at these papers in detail as we believe they are sufficiently different that they'd only be weakly informative of the effectiveness of the program analysed in Bassi & Nansamba 2020, due to differences in context and in the nature of the intervention implemented.20
Are there effects of this intervention on untreated individuals?
We expect that the benefits of the program for treated job-seekers may be substantially offset by negative effects on untreated individuals. Considering both theory and evidence on mechanisms, we do not see strong reason to believe that the income effects in Bassi & Nansamba 2020 were due to productivity gains. Rather, we would guess that at least some of the benefits enjoyed by people who receive the certification are explained by losses incurred by others (e.g. because jobseekers without certification on their application seem lower-skilled despite having similar skills). Additionally, three RCTs testing labor market programs find negative spillover effects (though a fourth RCT does not). That said, our analysis so far is fairly cursory; we have not yet attempted to engage with complex models of the benefits of improved job matching.
A major source of uncertainty about the effectiveness of this program concerns whether the intervention has negative effects on untreated individuals.21 This might occur if increased earnings for treated workers were not the result of an overall increase in productivity, but rather of a transfer between treated job-seekers and untreated jobseekers, or firm owners. Negative spillover effects on untreated workers were not tested in Bassi & Nansamba 2020. Within the study, we do not find strong evidence supporting the thesis that increased earnings are explained by improved matching and increased productivity.22 Furthermore, it seems theoretically highly plausible that the intervention would transfer, rather than increase, value. In general, it seems likely that signalling interventions that reach only a subset of eligible individuals would benefit participants and disadvantage non-participants. Moreover, we have not conducted an in depth assessment of the reliability of testing employed to measure soft skills, and we are unsure of the extent to which they would track skills that increase workers’ productivity.
We briefly reviewed the literature on the negative spillover effects of active labor programs. We focused on RCTs and found four papers attempting to measure negative effects on untreated job-seekers, listed below. We have not vetted this evidence.
- Crepón et al. 2012, an RCT assessing the treatment and displacement effects of a job placement assistance program in France.
- Cheung et al. 2017, an RCT assessing the treatment and displacement effects of three interventions for unemployed in Sweden.
- Alfonsi et al. 2017, an RCT assessing the treatment and displacement effects of vocational training and wage subsidies to train workers on-the-job in Uganda.
- Abebe et al. 2018, described above.
Three of these studies (Crepón et al. 2012, Cheung et al. 2017, and Alfonsi et al. 2017) find that a large proportion of positive treatment effects were due to displacement of untreated individuals. Abebe et al. 2018 does not find evidence of negative spillover effects.23
Based on our current understanding of the literature, and lack of strong evidence for a mechanism based on increased productivity, our best guess is that the overall effectiveness of the program is likely to substantially decrease once negative spillover effects are accounted for.
Overall, our analysis so far is fairly cursory and we have not yet attempted to engage with complex models of the benefits of improved job matching.
How generalizable is the evidence of effectiveness for this program?
Overall, we would guess the evidence on this program is not highly generalizable. This is because there is only one RCT measuring the effect of the program and external validity concerns are likely to be especially pressing for this type of intervention. The effectiveness of the intervention is likely to be affected by factors such as: employment rate, the extent to which firms value certain skills, and the extent to which there are information constraints. We would guess these factors vary substantially across different contexts.
Is the program cost-effective?
The evidence for the effectiveness of this program does not yet seem robust and we expect that it would have smaller effects at scale, so our best guess is that it is not in the range of cost-effectiveness of our priority programs.
We have completed a rough, preliminary cost-effectiveness analysis here. This analysis indicates that, given what we currently know about the program, reasonable adjustments to the observed effects of these programs would imply cost-effectiveness lower than our priority programs. Consequently, we have not put substantial effort into the precise figures and adjustments in the analysis.
Note that our cost-effectiveness analyses are simplified models that do not take into account a number of factors. There are limitations to this kind of cost-effectiveness analysis, and we believe that cost-effectiveness estimates such as these should not be taken literally due to the significant uncertainty around them. We provide these estimates (a) for comparative purposes and (b) because working on them helps us ensure that we are thinking through as many of the relevant issues as possible.
Major uncertainties in our cost-effectiveness model include:
- To what extent should we adjust the reported effect on earnings for all workers, to account for the confidence interval highly overlapping with zero? The current model provides a subjective adjustment for internal validity that accounts for the low level of statistical significance, but we are not confident about whether the size of the adjustment is appropriate.
- How would we expect the effectiveness of the program to change in a larger implementation? The effectiveness of the intervention is likely to be affected by factors such as: employment rate, the extent to which firms value certain skills, and the extent to which there are information constraints. We would guess these factors vary substantially across different contexts, which decreases our confidence in the results replicating in a different context. The current model provides a subjective adjustment for external validity that accounts for those factors, but we are not confident about whether the size of the adjustment is appropriate.
- Relationship between income and consumption. We use monthly earnings as a proxy for income, and income as a proxy for consumption. However, this is likely to be an underestimate of beneficiaries’ consumption, since this might also be affected by other sources of household income. Accordingly, our current assumption is an optimistic estimate of the intervention’s effect on beneficiaries’ welfare.
- Number of years benefits last. Our current cost-effectiveness estimate assumes benefit last three years, which is roughly a year longer than the time-span tested by the intervention.24 However, it is possible benefits would last longer.
- Does the intervention have a negative welfare effect on other jobseekers? If so, how large? As we discussed above, we are uncertain as to the effect the intervention would have on untreated job-seekers. Specifically, we do not know to what extent improvements in earnings of treated job-seekers are due to gains in productivity, or rather to the displacement of untreated job-seekers. The current model provides an estimate based on related literature, but we are uncertain of the extent to which those findings apply to the program considered.25
- What are the total costs of the intervention? The paper only reports costs for the certification component, which is the component that was tested by the trial. However, the overall intervention also included a matching component, which was administered to both treatment and control. This means that the reported cost of the intervention might underestimate the total cost of scaling up the program.
- Does our current model provide the best way of estimating the program’s benefit? As we discussed above, we do not see strong reason to believe that the income effects in Bassi & Nansamba 2020 were due to productivity gains. However, our analysis was only cursory and we have not yet attempted to engage with complex models of benefits of improved job-matching and long-term increase in productivity.
Room for more funding
Since our current analysis indicates this program is likely to be less cost-effective than our priority programs, we have not assessed room for more funding in depth.
Very roughly, we would guess that, across Uganda, there are 45,000 young job seekers and 100,000 firms that may be able to benefit from the program.26
We are unsure as to whether there are organizations interested in, or able to, deliver the program at scale. Our understanding is that BRAC is not currently implementing the intervention.27 We were unable to find other NGOs implementing the program.
Should the investigation move further? If so, what are the next steps?
In the absence of further evidence, we do not plan to investigate this intervention further.
We became aware of the program discussed in Bassi & Nansamba 2020 through Caria and Lessing 2019. We lightly scrutinized Bassi & Nansamba 2020 and used it as the basis for a cost-effectiveness analysis of this program. We looked for related literature by reviewing studies cited in Bassi & Nansamba 2020 and then reviewing studies cited by those studies; we do not expect this search to be comprehensive. We also spoke with one of the authors to discuss pending questions about the program.
“In developing countries, jobs are often poorly paid, informal, and unstable, trapping people in poverty and hindering economic growth.” Caria and Lessing 2019, Pg 1.
“In 2017, extreme working poverty remained widespread, with more than 300 million workers in emerging and developing countries having a per capita household income or consumption of less than US$1.90 (PPP) per day (...) Moderate working poverty, in which workers live on an income of between US$1.90 and US$3.10 per day in PPP, remains widespread, affecting 430 million workers in emerging and developing countries in 2017.” ILO, World Employment and Social Outlook: Trends 2018, Pg 1.
“Information is central to efficient market functioning, but in developing countries, crucial labour market information may not be widely available due to the limited diffusion of information technologies, fast urbanisation, and a disproportionately young labour force with little previous work experience. In these labour markets, jobseekers may not be able to easily access information about vacancies, and firms may not be able to accurately assess applicants’ skills.” Caria and Lessing 2019, Pg 1.
“Theoretical models of the labor market highlight how the eciency of the job matching process plays a key role in determining labor productivity and wages. Such labor market efficiency crucially depends on the information available to both workers and firms (Jovanovic,1979; Menzio and Shi, 2011; Chade and Eeckhout, 2017): difficulties in screening workers can prevent firms from selecting the right employees; at the same time, difficulties in signaling skills to employers can impact the ability of workers to match with the right jobs, or even their ex ante decision to acquire human capital (Spence, 1973).” Bassi & Nansamba 2020, Pg 1.
Our analysis is based on the September 2018 version of the draft, accessed on the SSRN website at https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3268523 on February 12th 2019, and shared in the link above. The draft is no longer available on the website.
“Certification is particularly relevant for young people, who may have limited formal work experience and credentials.” Caria and Lessing 2019, Pg 2.
“We focus the information revelation on non-cognitive or “soft" skills, such as trustworthiness, which have been shown to have high labor market returns in both high and low-income countries, but are hard to observe by nature.” Bassi & Nansamba 2020, Pg 2.
“We then selected the top four skills: creativity, trustworthiness, communication and willingness to help others, plus attendance, as a placebo skill.” Bassi & Nansamba 2020 p.10
“We used teacher surveys to measure those skills that are easier to assess for an external examiner, namely attendance, discipline, communication, pro-sociality and pro-activity. To measure creativity and trustworthiness we developed our own assessments: for creativity, we used a battery of questions; for trustworthiness, we made trainees play trust games with real money.” Bassi & Nansamba 2020, Pg 9.
“The intervention we implemented has three components: (i) a screening component, whereby information was collected on the soft skills of workers while they were still enrolled at the VTIs; (ii) a matching component, whereby job interviews were scheduled between workers and firms; (iii) a signaling component, by introducing experimental variation in whether information from the screening assessments was disclosed to both workers and firms during the job interview process, through the provision of skills certificates.” Bassi & Nansamba 2020, Pg 9.
“All the treatment workers who showed up to the job interviews were given the certificates (corresponding to 49% of treatment workers). The remaining treatment certificates were disbursed to the workers shortly after the first worker follow-up survey. So by the second follow-up survey about 81% of treatment workers had received the certificate” Bassi & Nansamba 2020 p.13, footnote 32
“Our sample includes young workers fresh out of vocational training and looking for jobs, as well as Small and Medium Enterprises (SMEs) looking for workers.” Bassi & Nansamba 2020, Pg 2.
Bassi & Nansamba 2020, Table 1 and Table 2
“The project was implemented in partnership with a large and reputable NGO, BRAC Uganda.” Bassi & Nansamba 2020, Pg 5.
Bassi & Nansamba 2020, Figure 1: Timeline.
Bassi & Nansamba 2020, Figure 4: Compliance and attrition.
Bassi & Nansamba 2020, Figure 1: Timeline.
“We highlight the role of mismatch between workers and jobs as one important channel through which information frictions on skills can reduce output and earnings.” Bassi & Nansamba 2020, Pg 4.
“Column 2 shows that when all workers are included in the regression, so that individuals with no earnings are assigned a value of zero, we find an 8% increase in earnings, but this is not significant at conventional levels" Bassi & Nansamba 2020, Pg 27. The confidence interval was calculated using the standard error and number of observations provided in the study. Calculations can be found in the ‘Confidence interval and p-value’ tab in this document.
The p-value was calculated using the standard error provided in the study. Calculations can be found in the ‘Confidence interval and p-value’ tab in this document. We believe that programs can easily fail to have an impact. The social science replication crisis shows that even widely accepted results can later be shown not to replicate. Therefore when we assess the effects of a program we look for strong evidence and/or very plausible mechanistic explanations in favor of the effects of a program.
The study reported earnings increased by roughly 11% (95% confidence interval: about 0.3% to 22.4%) for the sample of workers with positive earnings ( “In column 7 of Table 8 we report impacts for the sample of workers that are employed at follow-up: we find that certificates lead to an increase of about $7 per month, a result significant at the 5% level, and corresponding to an 11% increase in monthly earnings, relative to the control mean” Bassi & Nansamba 2020, Pg 27.) The confidence interval was calculated using the standard error and number of observations provided in the study. Calculations can be found in the ‘Confidence interval and p-value’ tab in this document.
We judge Carranza et al. 2019, Abebe et al. 2018, and Groh et al. 2015 to be the most relevant interventions. Even in this case, however, we do not take the papers to be strongly informative of the intervention discussed in Bassi & Nansamba 2020. First, external validity concerns are likely to be especially pressing for this type of intervention. The effectiveness of the intervention is likely to be affected by factors such as: employment rate, the extent to which firms value a certain type of skill, and the extent to which there are information constraints. We would guess these factors vary substantially across the different contexts in which the interventions were tested. Secondly, there are substantive differences across the interventions. For instance, while Bassi & Nansamba 2020 focuses on the certification of soft skills, the intervention assessed in Groh et al. 2015 focuses on soft and hard skills as well as educational background, Carranza et al. 2019 focuses on soft and hard skills, and Abebe et al. 2018 focuses on hard skills; moreover, Groh et al. 2015 includes a job-matching service, Abebe et al. 2018 includes a workshop component, and Carranza et al. 2019 only includes skills certification, while Bassi & Nansamba 2020 includes a matching intervention that is also delivered to the control group.
“An important caveat is that we have little direct evidence on the effects of these interventions on other jobseekers. The cost-benefit case would be weaker if these interventions caused displacement of other jobseekers.” Caria and Lessing 2019, Pg 4.
“Understanding general equilibrium better: a key concern with many of these policies directed at particular job-seekers is that they merely change who gets the jobs firms are advertising, without increasing the total number of jobs available” McKenzie 2017, Pg 19.
The authors also stressed this point in the paper's conclusion “Looking ahead, a promising extension would be to understand the general equilibrium effects of scaling up this type of information interventions. This would require a randomization at the regional level, introducing certificates in some local labor markets but not others. While the challenges of implementing
such a design would be substantial due to the high spatial mobility of workers in developing countries, this type of study would generate important directives for labor market policy, and so is something worth attempting in future research.” Bassi & Nansamba 2020, Pg 31.
Evidence that may support the idea that earnings increases are due to improved matching and increased productivity includes:
- The study reports effects of the intervention on job seekers and managers’ beliefs: it finds that workers increase their labor market expectations and managers revise upwards their beliefs in workers’ skills (see below, sections (a) and (b) in this footnote).
- There is evidence that the intervention lead the average manager to revise upwards their beliefs on workers with higher skills (Bassi & Nansamba 2020, Table 3, columns 3-4, and Figure A8, Panel A)
- There is some evidence that workers with higher skills gained more from the intervention in terms of earnings conditional on employment (Bassi & Nansamba 2020, Figure A8, Panel C)
- One of the authors has told us in conversation that the long term nature of the increase in earnings is suggestive of the increase in earning being explained by an increase in productivity. GiveWell's non-verbatim summary of a conversation with Vittorio Bassi, June 20, 2019
However, the absence of effects on firms’ profits, or other direct measures of firm productivity, significantly limits our confidence in the thesis that the increase in earnings is explained by an increase in productivity.
(a) “We show that workers also react to te certificates by revising upwards their labor market expectations: in the two years post intervention workers with a certificate report 7% higher expected earnings , 5% higher expected employment probability, a higher intention to bargain for wages and a larger size of their ideal employer” Bassi & Nansamba 2020, Pg 2.
(b) “In the previous section, we showed that, upon receiving the certificates: (i) managers revise upwards their beliefs on the skills of workers, with a stronger effect among higher ability managers.” Bassi & Nansamba 2020, Pg 20.
More details can be found in the ‘Displacement literature’ tab in this document. We followed the following process: first, we calculated what proportion of benefits to treated individuals reported in the study was due to reallocation from untreated individuals. For instance, if a study reported a 10 percentage points increase in employment rate for treated individuals and a 9 percentage points decrease in employment for untreated individuals, we report that 90% of treatment benefits were due to reallocation. Secondly, we estimated which proportion of treatment benefits was due to reallocation in Bassi & Nansamba 2020. To do so, we calculated a weighted average of the estimates in the four studies mentioned above, by weighting each result on the basis of how closely the intervention delivered and effect measured matched Bassi & Nansamba 2020.
This is a subjective estimate. The reason why we do not estimate benefits to extend further is that we do not judge there to be strong evidence of a mechanism for impact, which limits our confidence in benefits extending in time.
More details can be found in the ‘Displacement literature’ tab in this document.
There are 572 accredited vocational training institutes in Uganda. The list of accredited centers can be found here. The six sectors included in the study constitute 27% of the total of sectors covered. The list of sectors can be retrieved here. Data from the trial suggests the number of students enrolled in each centre is likely to be between 350 and 600, and roughly 80% of the eligible centres contacted in the trial and 80% of the eligible students in these centres were interested in participating to the program. (See section (a), (b) and (c) below).
Data from the study also suggests there were roughly 23,000 firms operating in the relevant sectors in the eleven districts where the intervention was conducted; there are 121 districts in Uganda (see section (d) below). Roughly 40% of eligible firms were interested in participating (see sections (e) and (f) below).
(a) Bassi & Nansamba 2020 Supplemental Online Appendix, Table S2: Basic VTI descriptives.
(b) “Only five of the 24 VTIs reported not being interested to participate” Bassi & Nansamba 2020 Supplemental Online Appendix, Pg 1.
(c) “About 20% of the eligible workers decided not to participate in the intervention, and this is consistent with them realizing they would not have benefited from it.” Bassi & Nansamba 2020, Pg 29.
(d) “Our census was conducted in 17 urban areas in 11 of the 121 districts of Uganda. In each urban area, the census took place within a 4km radius from the local BRAC branch. Our census covered about 1% of the total area of the 11 districts we worked in. The 2010 Census of Business Establishments further reveals that in 2010 there were 23,366 firms operating in the same sectors and districts targeted by our census, so that we covered less than 5% of the firms.” Bassi & Nansamba 2020, Footnote 4, Pg 5.
(e) “The table uses data from the initial census of 1,086 firms conducted for the job placement intervention”, Bassi & Nansamba 2020, Table 1: Firm descriptives from initial census.
(f) “The 422 firms which confirmed their interest in the program and completed the survey form our experimental sample.” Bassi & Nansamba 2020, Pg 6.
This is based on the activities listed on BRAC’s website for their work in Uganda.
Bassi, Vittorio, and Aisha Nansamba. "Screening and Signaling Non-Cognitive Skills: Experimental Evidence from Uganda, Supplemental Appendix for Online Publication." 20 Jan. 2020.