The Skills Space in Informal Work: Insights from Bangalore Slums

Abstract

We develop a framework for mapping and analysing informal worker skills using microdata from nearly 1500 workers residing in the slums of Bangalore, India. Alongside econometric modelling, we employ machine learning techniques to explore relationships between skills crowdsourced from respondents. We find that informal workers rely on a host of task, language, personal and social skills. Further, we identify skill claims associated with both levels and stability of wage earnings. Our results include insights on gender disparities in skill claims, importance of English and computer literacy and the central role of personal and social skills in the Indian informal labour market.

Acknowledgements

Financial Support for this project was received from the National Science Foundation, University of Chicago Delhi Centre, and Azim Premji University. The authors are grateful to Indian Institute of Technology Bombay for administrative support. Amitabh Chaudhary at University of Chicago provided critical technical guidance and Vishal Vincent Joseph provided excellent research assistance. Prof Santosh Mehrotra and Rakesh Ranjan Kumar at Jawaharlal Nehru University provided valuable suggestions to strengthen the paper. Comments from participants at workshops and talks at the University Chicago Center in Delhi, Azim Premji University, Ashoka University, ISI Delhi, IIT Delhi and IIIT Delhi were very helpful.

Disclosure statement

Declaration of Interest Statement by Nandana Sengupta: Financial Support for this study came from the National Science Foundation, University of Chicago Delhi Centre, and Azim Premji University, Bangalore. Administrative support for conducting the study was provided by Indian Institute of Technology Bombay (Project Code RD/0117-TUC0000-001). The IRB approval was obtained from Azim Premji University, Bangalore. Reviewing rights were shared with co-authors of the study. Nandana Sengupta has nothing further to disclose including any conflicts of interest that could inappropriately influence the work.

Declaration of Interest Statement by Sarthak Gaurav: Financial Support for this study came from the National Science Foundation, University of Chicago Delhi Centre, and Azim Premji University, Bangalore. Administrative support for conducting the study was provided by Indian Institute of Technology Bombay (Project Code RD/0117-TUC0000-001). IRB approval was obtained from Azim Premji University, Bangalore. Reviewing rights were shared with co-authors of the study. Sarthak Gaurav has nothing further to disclose including any conflicts of interest that could inappropriately influence the work.

Declaration of Interest Statement by James Evans: Financial Support for this study came from the National Science Foundation, University of Chicago Delhi Centre, and Azim Premji University, Bangalore. Administrative support for conducting the study was provided by Indian Institute of Technology Bombay (Project Code RD/0117-TUC0000-001). The IRB approval was obtained from Azim Premji University, Bangalore. Reviewing rights were shared with co-authors of the study. James Evans has nothing further to disclose including any conflicts of interest that could inappropriately influence the work.

Supplementary material

Supplementary Materials are available for this article which can be accessed via the online version of this journal available at https://doi.org/10.1080/00220388.2021.1898593

Correction Statement

This article has been republished with minor changes. These changes do not impact the academic content of the article.

Notes

1. The Government of India “Report on Conditions of Work and Promotion of Livelihoods in the Unorganised Sector’ (2008) estimates that informal (or unorganised) workers comprise about $92\mathrm{\%}$ of the total workforce in India. We use the same definition for informal or unorganised workers as this report: ‘Unorganised workers consist of those working in the unorganised enterprises or households, excluding regular workers with social security benefits, and the workers in the formal sector without any employment/social security benefits provided by the employers.’

2. A discussion on the institutional setup of skilling initiatives in India is provided in the supplementary material for the paper.

3. Beyond the informal worker context, there are a few studies on cognitive and non-cognitive skills of children Helmers and Patnam (2011) as well as adults Krishnan and Krutikova (2013).

4. For instance in their paper Adhvaryu et al. (2018) mention that ‘ $\cdots$ [The training initiative] which is named Personal Advancement and Career Enhancement (P.A.C.E.), aims to empower female garment workers through training in a broad variety of life skills, including modules on communication, time management, financial literacy, successful task execution, and problem-solving. These skills are important inputs into production in the ready-made garments context $\cdots$ ’. The study finds that wages rise very modestly with this training (by $0.5$ percent).

5. Background on informal work in Bangalore is provided in the supplementary material for the paper.

6. Official agencies categorise slums as either ‘declared’ (including ‘notified’ or ‘recognised’ slums) or ‘undeclared’ (Krishna, 2013, p. 1013). As per the Census (2011) classification, notified slums comprise those specified areas in a city or town that are notified as ‘Slum’ by the state or local government or Union Territory (UT) under any Act including a ‘Slum Act’. Recognised slums consist of those specified areas that are notified as ‘Slum’ by the state or local government or UT, or Housing and Slum Boards,which may have not been notified earlier. The third Census category – ‘identified’ slums – comprises congested tenements with poor civic amenities and having a population or number of household criteria (Census, 2011).

7. We used the Cochran (1963) formula for sample size calculation:

$S=\frac{n}{1+\frac{n}{N}};\mathrm{w}\mathrm{h}\mathrm{e}\mathrm{r}\mathrm{e}n=\frac{Z^2}{E^2}P(1-P).$

$Z=1.96$ is the Z-score for $5\mathrm{\%}$ level of significance. Based on the official statistics of declared slums in Bangalore, the population residing in declared slums, $N=306537$. $P$ is the estimated proportion of cases in the population of interest. In our study, we approximate $P$ as the proportion of population in the city in the $15-59$ years age group ($P=0.7$) based on the population pyramid of Bangalore detailed in the $2011$ Census of India. $E$ is the error margin. In larger sample surveys comprising notified and non-notified slums in the city, a margin of 3 per cent is commonly used (Roy et al., 2018). Our budget and resource constraints did not support this margin, but rather a feasible error margin we could achieve for our surveys. For the first round, the margin of error chosen was $E=3.5\mathrm{\%}$ leading to a sample size of $S_1=657$. For the second round, we fixed the margin of error at $E=3.25\mathrm{\%}$ leading to a sample size of $S_2=762$ (we were able to marginally reduce the value of $E$ from $3.5\mathrm{\%}$ to $3.25\mathrm{\%}$ due to the availability of additional funding for the project at the start of Round 2). We calculated the final sample sizes ($S_1=706$ and $S_2=819$) using a response rate of $93\mathrm{\%}$ which was found to be feasible from pilot surveys.

8. Our analysis of respondent characteristics corroborates this and is included in the supplementary material for the paper. We do not claim that our sample is representative of the remaining types of slums in the city that is ‘undeclared’, ‘non-notified’ or ‘de-notified’ slums. In Bengaluru, slums fall under the jurisdiction of three governing bodies namely the Karnataka Slum Development Board (KSDB), Bruhat Bengaluru Mahanagara Palike (BBMP) and the Directorate of Municipal Administration. For declared slums, KSDB, has the obligation for developing civic amenities as well as upgrading and slum rehabilitation. Slums that have undergone in-situ redevelopment for provision of housing for eligible slum dwellers are supposed to be de-notified. Although there are official estimates of the total number of declared and undeclared slums in the city, there is ambiguity about actual numbers (Krishna, Sriram, & Prakash, 2014). During fieldwork before our surveys commenced, we came across residents in several slums referring to the local administration’s viewing most slums, particularly ‘de-notified’ or ‘undeclared’ slums, in the city as illegal. Therefore, finding a sampling frame of all such slums in the city is challenging and informal workers residing there become a hard-to-reach population.

9. Detailed tables of our sampling methodology are presented in the supplementary material for the paper.

10. We did not use a sampling frame such as voter list as we expected to sample individuals in the 15-59 years age group. Voter IDs would only be available for those who 18 years or older. Furthermore, many migrants remain unlisted on the voter list of the constituency.

11. We also considered other coding schemes for the personality trait responses, including coding them as missing values to check the robustness of the analysis. Results remained qualitatively similar and have been included in the supplementary material for the paper.

12. The question asked is reproduced here verbatim: ‘What skills, knowledge or values do you have which you think help in getting or keeping a job? List as many as you can think of. (If no clear answers, surveyor should ask – if current, former or potential employers were considering both you and another person for the same job, why did they pick you?)’.

13. The analysis is carried out in R using the package corrplot to generate correlation plots.

14. Each skill is matched only to a single cluster.

15. We implement the algorithm in R using the command kmean() for clustering and the package factoextra for estimating optimal number of clusters.

16. Principal Component embeddings were also generated, which represented comparable insights, and are available from the authors upon request.

17. Available at https://github.com/facebookresearch/poincareembeddingshttps://github.com/facebookresearch/poincareembeddings.

18. This is calculated by subtracting the age at which the respondent started working from her current age.

19. In the absence of valid instrument in selection model, the Heckman model results provided in the supplementary material do not address potential endogeneity.

20. Calculated as the sum of dependent members under $15$ years of age and over $60$ years of age over the sum of wage earners in the household.

21. PAN stands for Permanent Account Number. It is a unique identity number issued by tax authorities in India. There is a mandatory requirement to link an individual’s PAN to her/his bank account.

22. We point interested readers to Taylor and Tibshirani (2015) for an introduction to the burgeoning literature on post-selection statistical inference.

23. We use the command cv.glmnet to estimate $\hat{\lambda }$ through crossvalidation and the command glmnet to fit the resulting LASSO model.

24. The full list of skills is available from the authors on request.

25. Emerging evidence on returns to schooling indicates that the opportunity cost of schooling is an important consideration surrounding investment in skilling for informal worker households (Datta & Mishra, 2019; Kanjilal-Bhaduri, 2018) as is the local demand for such skills (Adukia, Asher, & Novosad, 2020). With female labour force participation rates declining despite growing school enrolment, increased investment in skill training may be a crucial policy tool.

26. These results are robust to heteroskedasticity corrections and inclusion of dummies for the two major occupation categories as controls: Domestic Work and Construction Work.

27. The assumption here is that a weak form of ‘relationship with employers’ existed before the acquisition of the job. This assumption is supported by the finding that nearly two thirds of the jobs in Round $1$ were obtained by referrals from a relative or a friend who knows the employer. More details on respondent characteristics are available in the supplementary material for the paper.

28. Informal workers, particularly in small and medium enterprises are most likely to be excluded from apprenticeship training (Saxena & Gandhi, 2014). Our framework offers insights into developing platforms for mapping ‘skills gap’ by trade to address localised demand and supply mismatches.

Additional information

Funding

This work was supported by the NSF SciSIP (Science of Science and Innovation Policy) [grant number 1158803].