Measuring Women and Men’s Work Main Findings from a Joint ILO and World Bank Study in Sri Lanka Measuring Women and Men’s Work Main Findings from a Joint ILO and World Bank Study in Sri Lanka Prepared by: The International Labor Organization and the World Bank Main authors: Antonio Rinaldo Discenza, International Labour Organization Isis Gaddis, World Bank Amparo Palacios-Lopez, World Bank Kieran Walsh, International Labour Organization Copyright © 2021 The World Bank. Rights and Permissions This work is available under the Creative Commons Attribution 3.0 IGO license (CC BY 3.0 IGO) http://creativecommons.org/licenses/ by/3.0/igo. Under the Creative Commons Attribution license, you are free to copy, distribute, transmit, and adapt this work, including for commercial purposes, under the following condition: Attribution—Please cite the work as follows: Discenza, A., Gaddis, I., Palacios-Lopez, A., Walsh, K. (2021). Measuring Women and Men’s Work: Main Findings from a Joint ILO and World Bank Study in Sri Lanka. Washington DC: World Bank. Disclaimer The findings, interpretations, and conclusions expressed in this report are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent or the International Labour Organization. Cover image: ©Lakshman Nadaraja / World Bank Table of Contents Acknowledgements.................................................................................................. v 1. Background and Summary.................................................................................. 1 1.1 Background.........................................................................................................................1 1.2 Measuring Women and Men’s Work: the 19th ICLS............................................ 3 1.2.1 Summary of findings: identification of employment................................................................................... 4 1.2.2 Summary of findings: the identification of other unpaid activities............................................... 5 2. Main Findings........................................................................................................ 8 2.1 Achieving the Comprehensive Measurement of Employment.................. 12 2.2 The Measurement Of Unpaid Working Activities.............................................. 17 2.2.1 Own-use production of goods......................................................................................................................................... 18 2.2.2 Own-use provision of services......................................................................................................................................... 22 2.3 Concurrent Work Activities and the Total Burden of Work..........................27 2.4 Work in Agriculture and Fishing...............................................................................32 2.5 Labour Underutilization.............................................................................................. 33 2.6 Other Issues of Note..................................................................................................... 34 3. Summary Conclusions...................................................................................... 36 References................................................................................................................ 40 Annex 1. 19th ICLS Statistical Standards.............................................................................. 42 Annex 2. Methodology of the Pilot Study............................................................................. 45 Annex 3. Identifying Employment in the LFS and MLSS Questionnaires......................... 51 Measuring Women and Men’s Work | Main Findings from a Joint ILO and World Bank Study in Sri Lanka iv Measuring Women and Men’s Work | Main Findings from a Joint ILO and World Bank Study in Sri Lanka Acknowledgements This report is a co-publication of the International Labour Organization and the World Bank based on a methodological study that was conducted in partnership with the Sri Lanka Department of Census and Statistics (DCS). This work was made possible by generous funding and support from the William and Flora Hewlett Foundation and Data2X under the Women’s Work and Employment Partnership. Akuffo Amankwah, Theophiline Bose-Duker and Peter Buwembo were members of the core team conducting the methodological study. The authors wish to thank Kathleen Beegle, Peter Betts, Michael Frosch, Yeon Soo Kim, Gayatri Koolwal, Michael Weber, and, Alberto Zezza for comments and guidance over the course of the study. Special thanks go to Indu Bandara, Gero Carletto, Rafael Diez de Medina, Caren Grown, K.A. Sajeewa Kodikara, Idah Pswarayi-Riddihough, Ritash Sarna, Asitha Seneviratne, and Simrin Singh for supporting the study and to all the dedicated staff at the DCS, especially Dhanushka Nanayakkara and Chandani Wijebandara, without whom this work would not have been possible. This report was proofread by Robert Zimmermann, and layout design was by studio Pietro Bartoleschi. v 1 Background and Summary 1. Background and Summary 1.1 Background supplement the unemployment rate. Enabling more meaningful gender analysis was a key objective of these various updates but this Between 2017 and 2019, the International can only be achieved when the standards and Labour Organization (ILO) and the World good measurement practices are applied Bank, in collaboration with the Department through household surveys. of Census and Statistics (DCS) of Sri Lanka, completed a pilot study in Sri Lanka with the It is important to highlight from the outset goal of developing guidance on good practice that the two household survey types that in the measurement of women and men’s are the focus of this study fulfil different work through household surveys. The study primary objectives. LFSs are the primary was designed to enable a comparison of the data source for the computation of labour outcomes of two types of household surveys, market indicators, while MLSSs are designed namely, the labour force survey (LFS) and the to allow broader measurement and analysis multitopic living standards survey (MLSS). It of living standards and poverty. While the was completed under the Women’s Work and absolute comparability of the results of the Employment Partnership hosted by Data2X two types of surveys should not be expected with the support of the William and Flora given the different primary objectives Hewlett Foundation. The motivation for the and methodologies, the classification of study was the 19th International Conference respondents, their working activities and of Labour Statisticians (ICLS) in October their engagement with the labour market 2013, which introduced major changes to the should be as consistent as possible. This is framework of definitions used to produce all the more important in developing country statistics on work and the labour market contexts, where surveys are often conducted (ILO 2013, see also Annex 1). Relative to the infrequently and many analytic studies (for standards of 1982, it reduced the scope of the example, to understand drivers of changes in statistical definition of employment to work poverty and living standards) draw on various done for pay or profit and applied a wider types of surveys on the assumption that they definition of work, along with the forms of are each generating coherent and consistent work framework, to support the analysis of information. participation in paid and unpaid productive activities. This new framework recognizes The sensitivity of statistics to survey design, that people may be engaged in multiple particularly statistics on labour, is well working activities within the same period, documented, often with a focus on the impacts thereby enabling a complete accounting of observed if the content of a survey is altered all work performed. An additional important (for example, see Bardasi et al. 2011). Studies development was the adoption of an extended have also been undertaken on the effects of set of labour underutilization indicators to different survey types on measurement that 1 Measuring Women and Men’s Work | Main Findings from a Joint ILO and World Bank Study in Sri Lanka focus on the ex post comparison of results, in estimates of work and the labour market such as a study in Egypt showing a substantial between the MLSS and the LFS. impact of the survey type on estimates of women’s work (Langsten and Salen 2008). To isolate the effect of survey type and Similarly, Floro and Komatsu (2011) show that differences in survey content on measures household surveys can easily miss temporary of work and employment, this study was or casual forms of employment. Among the conducted as a split-sample randomized concerns is that, especially in countries with experiment whereby the only differences strong social norms and/or culturally assigned between the two groups of households gender roles, women working in family randomly assigned to one of two treatment businesses may not consider the activity as arms were the questionnaire content and employment (or work) and therefore not report implementation. This study design permits the activity in response to standard questions conclusions to be drawn on the scale of about labour market engagement (Müller and differences, if any, and the possible cause of Sousa 2020). While these studies demonstrate these differences. This allows guidance to be the sensitivity of measurement to survey developed on good measurement practices. design, they do not provide specific solutions for any given survey beyond those they cover. This pilot study builds on previous rounds This requires more direct investigation of studies completed by the ILO (Benes and specific to the surveys under review, namely, Walsh 2018b) and the World Bank (Gaddis the LFS and MLSS. et al. 2020b), as well as a range of related research papers (Desiere and Costa 2019; To address these issues, the ILO and World Koolwal 2019). In addition to extending Bank conducted a joint pilot study in Sri Lanka, the scope of the available guidance, in collaboration with the Sri Lanka Department the experiences will be used to update of Census and Statistics (DCS). The study had published ILO model LFS questionnaires four broad objectives, i.e. to (i) support the and a World Bank model labour module for operationalization of the 19th ICLS standards MLSS questionnaires, have informed the in LFS and MLSS type surveys, (ii) assess and, new Living Standards Measurement Study if identified, reduce the undermeasurement (LSMS) Guidebook for measuring labor in of women’s employment and work (as MLSS-type surveys (Durazo et al. 2021). documented by the previous academic literature mentioned above and earlier ILO This report presents a first summary set of pilot studies) in these surveys, (iii) gain a better the findings of the pilot study. The findings understanding of the comparability of labour are being used to generate guidance on the market indicators obtained from LFS vs MLSS measurement of labour across different type surveys, and (iv) pilot changes in either types of household surveys. While the questionnaire that could narrow differences primary target audience of the guidance 2 1. Background and Summary will be those individuals tasked with the needed to meet user needs more effectively. completion of household surveys that These updates took the form of Resolution measure labour, the findings should also I of the 19th ICLS: Resolution concerning attract a wider audience, including data statistics of work, employment, and labour users who are interested in the measurement underutilization. practices behind the statistics or, more generally, in the improvement of the data The 19th ICLS standards revised available on women and men’s work. While the definitions of employment and highlighting issues of measurement, the unemployment and also established a much report also emphasizes the valuable data wider framework for statistics on paid and that can be generated if the guidelines and unpaid work and on labour underutilization. standards are implemented, such as the This has created a basis for a much wider more comprehensive measurement of all the range of analyses of the working lives of working contributions of men and women. individuals. A key motivation of the changes was a desire to explain differences in the working contributions and experiences 1.2 Measuring Women of women and men and to achieve a related understanding of labour market and Men’s Work: engagement. The objective is to achieve the the 19th ICLS mainstreaming of the measurement of all working activities in order to enable deeper The background of the study is related to the insights into the relationship between the international statistical standards adopted performance of work and interactions with by the international community at the 19th the labour market. ICLS in October 2013. The revised standards represent a framework for work and labour The survey questionnaires covered a mix market statistics and replace the standards of paid and unpaid working activities, adopted at the 13th ICLS in 1982. The latter namely, employment, the production of standards had been in use in many countries goods for own-use and the provision of for decades and had become synonymous services for own-use, as defined in the with labour statistics on a worldwide standards. The LFS questionnaire used basis, providing, for instance, definitions for the study was developed by the ILO of key concepts, such as employment, by building on the published model LFS unemployment and labour force participation. questionnaires. The MLSS questionnaire was developed by the World Bank using the The 1982 standards have been vital, but there multitopic household surveys with a focus had been a growing realization – as occurs in on poverty measurements, such as the ones many statistical domains – that updates were supported by the World Bank through the 3 Measuring Women and Men’s Work | Main Findings from a Joint ILO and World Bank Study in Sri Lanka Living Standards Measurement Study, as a The study in Sri Lanka sits within the context reference. The questionnaires both included 1 of ongoing efforts to provide support to similar numbers of questions to identify the countries in the implementation of the 19th labour force status of individuals, but the ICLS standards through household surveys LFS questionnaire contained more detailed that measure labour. The data have been questions on supplementary labour–related analysed following the completion of the factors, such as detailed characteristics of first and second waves of data collection, jobs, while the MLSS questionnaire contained which took place in March to April 2019 and questions on a range of other topics related September to October 2019, respectively. to living standards. The main findings are summarized below and detailed in the main body of the report. A message of this report is that the measurement of diverse forms of work adds immense value and provides a clearer 1.2.1 Summary of findings: perspective on gender differences than identification of statistics on employment alone. For employment example, three quarters of the total working time among employed male respondents The measurement of employment, across the three forms of work activities particularly employment among women, is – employment, the own-use production of sensitive to survey design and content. This goods and the own-use provision of services finding is consistent with the conclusions – was accounted for by employment. Among of many earlier studies and repeated across employed women, the corresponding share many settings (see Anker and Anker 1989; was less than half, and women spent more Boserup 1970; Comblon and Robilliard than half their average reported working time 2017; Mahmud and Tasneem 2011). While in unpaid household services, regardless of the contexts of the studies referenced their status as employed. As a result, a gap varied substantially and even though these of ten hours working time per week in favour studies generally pre-date the adoption of of men if only employment is considered the 19th ICLS standards, a similar pattern of becomes a gap of over ten hours in favour of undercounting women’s work was identified. women if the three forms of work activities are considered together, irrespective of the The results of the Sri Lanka study survey used to measure work. demonstrate that a clear risk continues to exist of undercounting various types of working activities, or of misclassification between paid and unpaid activities when the 1 The MLSS questionnaire is not based on the Sri Lanka Household Income and Expenditure Survey, because the latter does not 19th ICLS standards are applied. In the first include a dedicated module on household members’ labour market engagement. wave of data collection, the LFS identified 4 1. Background and Summary 22 percent more employed women than women than among men. These conclusions the MLSS (equivalent to an 8.1 percentage support the development of guidance on good point difference in measured employment measurement practices to avoid the risks, to population ratios). It also identified such as the need for recovery questions, approximately 3 percent more employed men careful wording and translations into local (a 2.4-percentage point difference in the language, to ensure that people with “small” employment-to-population ratios), leading jobs or helping in family businesses or farms to a gap of 10 percent overall between the are identified in the survey. These revisions to surveys (a 5.5 percentage point difference in the MLSS instrument, while important for the employment-to-population ratios). In-depth measurement of employment, also improve analysis of the data led to a conclusion that the measurement of own-use production the gap emanated from the fact that the work in agriculture (described below).3 MLSS, which, unlike the LFS, initially did not include any recovery questions, identified fewer people engaged in employment in three 1.2.2 Summary of findings: particular groups, namely (1) those with more the identification of casual, low-hours jobs, (2) helpers in family other unpaid activities businesses and farms and (3) others involved in informal working activities, with all of The Sri Lanka study also included questions these groups being primarily women. 2 on unpaid working activities. Specifically, work done to produce goods for own- Changes to address these issues were consumption (called the own-use production successful in partially closing the gap in the of goods in the standards), which covers, second wave of data collection (6 percent but is not limited to subsistence farming, gap for both men and women, equivalent and unpaid work to provide services to the to a 3.5-percentage point difference in household (called the own-use provision employment to population ratios). of services in the standards), such as housework, childcare and other activities This finding that risks of misclassification of predominantly carried out by women. In work are most concentrated among certain combination, the standards refer to these types or groups of workers corroborates two types of activity as own-use production earlier findings of the ILO (Benes and Walsh 2018b), that these risks are greater among 3 In the MLSS, a common set of questions is used to identify employment in agriculture (that is, agricultural work for pay or profit) and own-use production work in agriculture (that is, for own or family consumption). The distinction between these two concepts is fleshed out in subsequent questions, which seek information on the intended 2 Recovery questions are here defined as questions whose purpose is use of the agricultural outputs (for pay or profit or for own or family to “recover” persons who were not classified as employed during the consumption). Any revisions that improve the ability of the MLSS to core questioning designed to capture employment, even though they capture employment in agriculture will thus also enhance the ability were engaged in activities that count as employment. of the survey to measure own-use production work in agriculture. 5 Measuring Women and Men’s Work | Main Findings from a Joint ILO and World Bank Study in Sri Lanka work. The other forms of work covered by were successful in narrowing the gap the standards, namely, unpaid trainee work between the MLSS and the LFS. and volunteer work, were not examined in the Sri Lanka pilot study. Even more notable was the sensitivity of the data on hours worked in own-use provision In the first wave of data collection, relative of services. While the MLSS identified fewer to the MLSS, the LFS recorded a greater people engaged in these activities in wave prevalence of both forms of unpaid work. 1, it showed a substantially higher average The difference was concentrated in crop- number of hours worked (34.2 versus 24.8 in farming, while there was relatively less the LFS). Analysis narrowed this down to care difference across other types of activities. work (care of children or dependent adults), This reflects the fact that – as described and a review of practices identified the source above – the MLSS identified fewer family as a difference in implementation between helpers and other marginal workers in the two surveys. While the two surveys used farming. The updates undertaken after similar questions to identify individuals wave 1 caused a reduction in the recorded engaged in care work for adults and children, gap. The difference observed in wave 2 the LFS emphasized active caregiving (and was relatively minimal, suggesting that the included a descriptive text to be read by additional questions and updates in wording LFS interviewers). In contrast, there was no 6 1. Background and Summary explicit emphasis on active caregiving in the to the single question. However, while the MLSS. As a consequence, the MLSS estimate direction and scale of the impact is quite for caring activities in wave 1 was nearly three consistent, which of the two sets of results is times the LFS estimate (43.8 versus 16.1). more valid is not certain. During the wave 2 training, both sets of The study covered many other issues, interviewers were instructed to read the the analysis of which enhances the additional text. The impact on results was understanding of good practices in the clear. The LFS result was relatively consistent measurement of work, employment and with wave 1, while the MLSS estimate fell by labour underutilization, as framed by the half, leaving a much smaller gap and resulting 19th ICLS standards. Perhaps a general in a minimal gap in the overall estimate of summary should highlight, as above, that the time spent in the own-use provision of the measurement of work can be sensitive services in wave 2 (26.1 hours per week in the to questionnaire design, implementation MLSS and 25.3 hours in the LFS). and context, and the study has allowed an identification of the areas in which the The study also shows that measured weekly misclassification risks appear greatest. hours spent on the own-use provision of services are significantly lower if the Another general point is the need for good survey relies on only one question (seeking questionnaire development and testing information on the hours worked during the practices to establish a solid survey footing. previous week) rather than two questions This is true at the international level in the (on the days worked during the previous activities of international agencies and at week and the average hours worked per the national level among national statistical day). In wave 2, both the LFS and the MLSS compilers. In the absence of appropriate administered to half the samples the one- testing, the degree of sensitivity of question approach and to the other half the measurement may never truly become visible, two-question approach. The results in both leaving open the possibility that the statistics surveys were highly consistent. The two- generated may not capture reality in the way question approach yielded weekly hours desired, for example the differences between spent on own-use production of services that women and men’s working lives. Activities were approximately 30 percent higher than at the international level can provide a major weekly hours based on only one question. support to countries, but not entirely replace This pattern was repeated among both men the need for sound translation and the and women albeit with slightly different gaps. adaptation of questionnaires to the national A possible explanation is that the rounding context, a process that needs to be supported of the daily averages in the two-question by testing at the national level. approach leads to an overestimation relative 7 2 Main Findings 2. Main Findings The measurement of employment and those adopted at the 19th ICLS. A primary other working activities is sensitive to objective of the revised standards was to survey design; this is particularly true in address gender biases in the basic concepts the case of women. A clear risk exists of used to measure employment and economic undercounting the various types of working activity, as well as to promote a much wider activities or misclassifying paid and unpaid range of statistics on paid and unpaid work activities. This risk can be reduced by and engagement with the labour market, careful survey design, testing and training. relative to previous standards. (See Annex 1 Misclassifications, if they occur, can for a description of the 19th ICLS standards.) seriously limit the analysis of the variations across the experiences and contributions of The implementation of the revised women and men to productive activities, as standards needs to be accompanied by well as the barriers and constraints they face good measurement practices to achieve to changing their situation. This hampers the an improvement in the data on women and identification or evaluation of appropriate men’s engagement in employment and other policies, including those seeking to promote forms of work. The Sri Lanka study is part women’s economic empowerment. of a longer-term series of studies designed to provide comprehensive guidance to This is one of a number of key findings of countries on the implementation of the a pilot study completed in Sri Lanka in a standards. In the ILO case, this builds on an cooperative effort of the DCS of Sri Lanka, earlier round of pilot studies that focused on the ILO, and the World Bank. The pilot study the implementation of key elements of the was completed through the Women’s Work standards through the LFS (Benes and Walsh and Employment Partnership hosted by 2018a). This work had been used to develop Data2X with the support of the William and model LFS questionnaires that were the Flora Hewlett Foundation. starting point for the LFS questionnaire used in the Sri Lanka study.4 For the World Bank, The findings of the pilot study will advance the study builds on previous methodological the cause of the proper measurement and studies conducted under the umbrella of reporting of paid and unpaid work across the Living Standards Measurement Study household surveys (particularly the LFS and Program to improve labour measurement in the MLSS) focused on measuring welfare by household surveys. While this study reiterated identifying measurement difficulties in the some of the findings of the earlier rounds of domain of work and the related solutions and good practices. This endeavour has been carried out in the context of the need 4 See Labour Force Survey (LFS) Resources (dashboard), ILOSTAT, for support in implementing the latest International Labour Organization, Geneva, https://ilostat.ilo.org/ resources/lfs-resources/.ttps://ilostat.ilo.org/resources/lfs- international statistical standards, especially resources/. 9 Measuring Women and Men’s Work | Main Findings from a Joint ILO and World Bank Study in Sri Lanka studies, it is unique because it was explicitly operandi of the quantitative test was to designed to allow a comparison of the labour administer a “typical” LFS questionnaire and indicators generated by two different survey a “typical” MLSS questionnaire to a similar instruments (the LFS and the MLSS). In sample of households through a split- addition, the study added substantially to the sample randomized design. Within each understanding of some topics, such as the PSU, 10 households were randomly assigned measurement of agricultural work and of the to the LFS, and 10 to the MLSS treatment time spent on unpaid household service work. arms.5 (See Annex 2 for a description of the It also highlighted areas where more study methodology of the pilot study.) 6 would be beneficial. As proposed by Presser et al. (2004), such The lessons learned will inform more rounds a split-sample approach can be used if of questionnaire development and testing on the objective is to compare the outcomes key related issues, such as the use of time- of different survey questionnaires and if use approaches to improve the measurement all aspects of the sampling, methodology of unpaid household service work. The and implementation, other than the ultimate objective will be a comprehensive questionnaires, are the same. In line with guidance covering the full range of issues approaches proposed by Fowler (2004) touched on by the 19th ICLS standards, statistics are generated and compared namely, the performance of paid and unpaid for the concepts covered by both work and labour market engagement. questionnaires. If differences were observed, for example, in the proportion of working- The Sri Lanka pilot study involved multiple age respondents identified as employed, a rounds of data collection, allowing more in-depth analysis was undertaken to comparisons across the outcomes at try to isolate the source of the differences. different times. The first round of testing This type of experimental approach is being involved cognitive interviews among increasingly used and has been found to 20 respondents for each questionnaire. be valuable in generating improvements This was followed by a quantitative test in questionnaire design (for instance, see based on a representative sample of Beaman and Dillon 2012; Beegle et al. 2012; households in three districts of Sri Lanka, Benes and Walsh 2018b; Gaddis et al. 2020a; namely, Anuradhapura, Galle and Kurunegala. Heath et al. 2020; Kilic and Sohnesen 2017). The quantitative test was based on a total sample of 980 households per survey type and per wave across 98 primary sampling 5 This implies that, within each household, all individuals were units (PSUs). The households were selected administered the same questionnaire. from the census blocks of the continuous 6 All estimates of labour market indicators reported in this document use post-stratification weights to benchmark the MLSS and LFS LFS in the selected districts. The modus samples to a common reference population. 10 2. Main Findings For example, when a difference was the impact of the solutions identified. The identified in the proportion of working- order of the report broadly follows the study age respondents in employment in wave 1, design. Thus, the findings of wave 1 are a detailed analysis took place of the generally described initially for any given characteristics of employment and working issue. This is followed by a description of the time of respondents to each questionnaire, changes made to the survey instrument in as well as the contribution of the various wave 2 and the results of wave 2, along with questions to the total measured level of the conclusions drawn. employment. This analysis then supported a conclusion that the difference emanated Despite the above, achieving absolute from a greater emphasis in the LFS consistency between the LFS and MLSS, questionnaire on the recovery of small jobs or any other household survey, in the and helpers in family businesses and farms, measurement of work and labour is not a as revealed by differences in working time, realistic goal. Absolute consistency will be industry, occupation, and so on. An analysis unlikely because of differences in the primary across the three districts showed that a objectives and many aspects of the design similar scale of variation was observed in of various household surveys. For instance, each district, further supporting a conclusion the LFS will typically be administered to a that the difference could be related to larger sample of households and be focused questionnaire content given that it appeared primarily on the labour market and work- to be systematic. related issues to generate a wide range of indicators on these topics. The MLSS may In the absence of a split-sample randomized involve smaller samples and will cover a study design, it would have been difficult wide range of topics relevant to the analysis to rigorously isolate the effects of the of poverty and living standards. While questionnaire used on the outcomes of information on the engagement of each interest, detect the sources of differences household member in different forms of with any degree of specificity, and identify paid and unpaid work is key to the analysis of ways to close measurement gaps. The poverty and living standards, MLSSs inevitably multiple wave approach also performed include fewer questions on labour and capture an important function, insofar as it gave less detail on the topic than a dedicated LFS. the study team the opportunity to make The outputs of the two surveys will therefore changes to the questionnaires before a vary substantially in scope, focus, the type of second wave of field data collection with the disaggregations, and so on. Nonetheless, the same households, and to assess the improving the consistency in measurement, impact of these changes on labour indicators to the extent possible, will be valuable. generated by both survey types during the Regardless of the survey, it is desirable that second wave. This enabled an analysis of a person who is employed (as defined by the 11 Measuring Women and Men’s Work | Main Findings from a Joint ILO and World Bank Study in Sri Lanka standards) be classified as employed, likewise Deeper analysis of this result suggested for unemployment or other key concepts. that the greatest gap centred on people Differences in classification have implications helping in family businesses or farms, for coherence across surveys. This is people with more casual jobs or jobs with especially important in developing countries, lower average working hours. These findings where surveys may be conducted infrequently are consistent with the results of several and labour market information systems may previous studies. Müller and Sousa (2020) have to draw on various types of surveys on note, in particular, the tendency of women the assumption they are each generating working in family businesses to self-identify coherent and consistent information. as housewives, which was often seen by the respondents as mutually exclusive with employment. Consequently, these 2.1 Achieving the respondents did not report their activities when they were asked about their jobs or Comprehensive businesses. Benes and Walsh (2018b) find Measurement of that dedicated recovery questions were Employment required to target more casual jobs or the work of those helping in family businesses. A similar conclusion was reached by In the first wave of field data collection, Sudarshan and Bhattacharya (2008), who the LFS questionnaire identified one show that these types of undercounts can tenth more employed respondents than be addressed by intensive probing. the MLSS questionnaire. The two surveys generated employment to population The types of working activities at greatest ratios of 57.0 percent and 51.5 percent, risk of undercount are predominantly respectively (see Figure 1). This difference performed by women. In the case of the was particularly acute and statistically Sri Lanka study, this was confirmed by an significant among female respondents. The assessment of the differences between the LFS identified 22.5 percent more employed surveys in the distribution of jobs by status women (a ratio of employment to population in employment, sector and average hours of 44.1 percent versus 36.0 percent), while a worked. More specifically, the LFS identified small difference was also recorded among larger numbers of contributing family men (72.4 percent versus 70.0 percent).7 workers, own-account workers and persons with low-hours jobs. Changes were made to the MLSS questionnaire used during wave 2 of the field data collection to reflect these 7 The indicators of work and the labour market shown in this report refer to the working-age population (WAP). In line with para. 65 of the conclusions. In particular, the wordings 19th ICLS resolution (ILO 2013) this includes all persons aged 15 years and above. of some questions were changed, and 12 2. Main Findings Figure 1 Employment to population ratio (% of working-age population (WAP)), by sex, wave of data collection and survey 57.0 Wave 1 TOTAL 51.5 57.4 Wave 2 53.9 72.4 Wave 1 70.0 MALES 73.5 Wave 2 68.9 44.1 Wave 1 36.0 FEMALES 43.8 Wave 2 41.3 0 10 20 30 40 50 60 70 80 LFS MLSS Source: Joint DCS, ILO, and World Bank pilot study in Sri Lanka, Wave 1 and Wave 2, March–October 2019. recovery questions were added to target This meant that, in wave 2, the LFS was people engaged in the types of activities identifying 6.5 percent more employed than apparently missed by wave 1 (see Annex 3 the MLSS (or an employment to population for details). Further in-depth analysis of the ratio that was 3.5 percentage points higher). MLSS wave 2 data, presented in Annex 3, This suggests the changes made were at Figure 3.1, shows that without the recovery least partially successful and were especially questions, 9 percent of employed women important for women, reducing the gap would not have been captured as employed. from 22.5 percent to 6.0 percent. It is worth For men, all four recovery questions noting that, while the difference in estimates combined identified only slightly more than of total employment remained statistically 2 percent of total employment. significant the difference for women was no longer statistically significant in wave 2. In wave 2, the gap between surveys was The remaining gaps observed in wave 2, as reduced among women (43.8 percent in wave 1, were repeated across the three in the LFS versus 41.3 percent in the districts covered by the pilot study and nearly MLSS), while it slightly increased among all age groups, supporting the conclusion men (73.5 percent versus 68.9 percent). that the difference was relatively systematic. 13 Measuring Women and Men’s Work | Main Findings from a Joint ILO and World Bank Study in Sri Lanka The fact that the LFS identified more Annex 3.) This approach is consistent with employed respondents in this context than guidance provided by Grosh and Glewwe the MLSS may be attributed to the fact that (2000). It also reflects long-standing practice the LFS design is centred on a comprehensive in MLSSs and helps maintain a degree of identification and description of employment comparability over time. and labour market engagement, while the MLSS has a primary focus on poverty, thus These variations in approach reflect the dedicating fewer questions to the overall important differences in the objectives topic of labour. This may be seen in the of the surveys, and it is unsurprising that questionnaires presented in Annex 3. the results are not completely consistent. Nonetheless, the finding of the pilot The LFS questionnaire used for the study that relatively minor adjustments to study dedicated early questions to the questionnaires can reduce, if not eliminate, comprehensive identification of employment, gaps is useful for any household survey without seeking to categorize employment covering labour-related issues.8 by industry, occupation, and so on. This additional detail was captured through the The differences observed in wave 2, while later sections of the questionnaire. The smaller in magnitude, were concentrated, answer to a single question might identify a as follows: respondent as employed, or several might be needed. Benes and Walsh (2018b) show that, if  The differences in wave 2 were well designed and implemented, this approach concentrated among the self-employed can be efficient in minimizing the survey (17.5 percent higher in the LFS), largely, but burden on most employed respondents, while not exclusively in the agriculture sector. capturing more difficult cases (for instance, casual jobs) through additional questions.  The LFS also recorded a larger number of employees in wave 2, but the difference By contrast, the MLSS questionnaire was less substantial than the gap for the combined the objective of identification self-employed. and a certain level of classification of the employment though the initial questions on 8 Because the LFS is dedicated to the measurement of employment labour, reflecting the fact that it dedicates and work and has been extensively tested in previous rounds of methodological investigation, the analysis generally considers the fewer questions to the topic overall. Also, LFS estimates as a benchmark against which the MLSS is evaluated. respondents to the MLSS tended to answer Of course, there is still also the possibility of some degree of under- or overcounting or of mismeasurement more broadly in the LFS all the initial questions and thus provided that the study was not designed to assess comprehensively, even if some misclassification issues could have been observed. Moreover, a categorization of all the employment and we assume that seasonality, as captured by differences in labour market indicators between waves 1 and 2, would affect the two own-use production of foodstuffs undertaken survey instruments proportionately (and this is one of the reasons this report emphasizes relative, rather than absolute gaps, between by the respondent. (See the questionnaire in the two instruments). 14 2. Main Findings  In wave 2, the MLSS identified a slightly because they cover the dimensions that higher number of respondents who, as typically distinguish women and men’s a main job, were working without pay in experiences in the labour market. family businesses and farms (that is, they were contributing family workers). Another implication of the differences in the identification of employment is evident in The implications of these differences the analysis of the data on working time. The are important because they have an MLSS picked up fewer jobs with low working obvious and direct impact on many of the hours than the LFS in wave 1 (see Figure 2), indicators describing the prevalence and leading to higher average working time (41.4 characteristics of paid work and on any versus 38.3). By wave 2, this gap had narrowed analysis that builds upon such indicators. because of the improved recovery of people This includes analysis of economic sectors, with casual or low-hours jobs, but a gap still status in employment, occupation, remained (39.8 versus 37.9), supporting the informality, working time, and so on. These conclusion that the changes made in the aspects exhibit high gender relevance surveys may not have fully closed the gap. Figure 2 Average hours actually worked per week in employment (in all jobs) and the gender gap, by sex, wave of data collection and survey 38.3 Wave 1 TOTAL 41.4 37.9 Wave 2 39.8 42.5 Wave 1 45.5 MALES 42.2 Wave 2 44.2 10.1 32.4 Wave 1 10.8 34.7 FEMALES 10.4 31.8 Wave 2 10.4 33.8 0 10 20 30 40 50 LFS MLSS GENDER GAP Source: Joint DCS, ILO, and World Bank pilot study in Sri Lanka, Wave 1 and Wave 2, March–October 2019. Note: The red diamond indicates the gender gap in working time in the activities covered. The diamond is included on the bar of the gender with less working time. If it is included on the bar for women, this thus shows the amount by which the average working time of women in the activity was less than among males and vice versa if it shown on the bar for men. 15 Measuring Women and Men’s Work | Main Findings from a Joint ILO and World Bank Study in Sri Lanka Figure 3 illustrates this additionally, wave 1, the LFS identified 23 percent more showing how respondents in each survey respondents who had worked between were distributed by hours worked in the two 10 and 30 hours in the reference week waves. In both wave 1 and wave 2, almost (314 compared with 256 in the MLSS). By identical numbers of respondents had wave 2, there was still a gap, but it had actual working time in all jobs of 40 hours or decreased to 13 percent (334 compared more in the reference week. This suggests 9 with 296). that both surveys were able to capture full- time employment. The differences were One area of consistency between the two observed in the lower bands of working time surveys was the gap in average actual in both wave 1 and wave 2. However, the working time between male and female gaps between the two surveys were smaller respondents (see Figure 2). Across both in wave 2 than in wave 1. For example, in waves and in both surveys the average Figure 3 Distribution of employed respondents, by bands of hours actually worked per week (in all jobs) and by wave of data collection and survey 1600 1508 1518 1423 1400 1363 1200 1000 800 804 797 799 800 600 191 181 156 164 400 314 334 296 256 200 118 78 140 118 74 51 64 52 0 LFS MLSS LFS MLSS Wave 1 Wave 1 Wave 2 Wave 2 Missing/Don’t know Temporarily absent 1 to < 10 10 to < 30 30 to < 40 40 or more Source: Joint DCS, ILO, and World Bank pilot study in Sri Lanka, Wave 1 and Wave 2, March–October 2019. Note: The LFS had 10 and the MLSS 11 missing/don’t know values for the hours worked in wave 1. 9 The actual number of respondents to each questionnaire was slightly different; so, the results were reweighted to impose a common total number of respondents for each survey and wave. This allows a direct comparison of the number of respondents in different groups across the two surveys and the two waves. 16 2. Main Findings working time in employment was be related to paid work and labour market approximately 10 hours greater among men engagement. Another advancement is the than among women with an identical gap recognition of the reality that people can be of 10.4 hours in wave 2 in both surveys. One engaged in multiple forms of work in a single possible conclusion from this finding is that, reference period, for instance, employed, even if some difference in estimates existed but also engaged in the production of goods across the surveys, the difference was not for family consumption, and so on. This is a particularly sex differentiated, at least not contrast relative to the 1982 standards, which in the case of working time in employment. excluded unpaid services within households Put differently, it was as likely to influence from the concept of economic activity and, the reporting of working time in employment at the same time, assigned people to one among both men and women. category only (employed, unemployed, or not economically active). The gender gap in working time is shown by the red diamonds in Figure 2 (and other The new framework promotes the figures containing information on working measurement of the different forms of work time). The diamonds are presented on the to enable indicators to be generated on the bar of the gender with lower average working prevalence of participation and the time spent time in the activity. For example, in Figure 2, in each of them, as well as the interaction the diamond for wave 2 in both surveys is between the various forms of work, the total on the bar for women with the number 10.4, work burden and how these activities are indicating that the average working time distributed across household members. of female respondents in employment was 10.4 hours less than the average among men. The pilot study included different sets of questions and flows to identify people carrying out unpaid working activities 2.2 The Measurement and the time spent on these activities. As with employment, the intention is to draw Of Unpaid Working conclusions on good measurement practices Activities for household surveys. Specifically, the questionnaires both covered the own- An important development associated with use production of goods and the own-use the adoption of the 19th ICLS standards is the provision of services. (See Annex 1 for a creation of a coherent framework identifying description of the 19th ICLS standards.) different forms of unpaid work, alongside employment. One goal is to mainstream the measurement of unpaid working activities, and in a way that allows the activities to 17 Measuring Women and Men’s Work | Main Findings from a Joint ILO and World Bank Study in Sri Lanka 2.2.1 Own-use production family workers) also improved the survey’s of goods ability to measure own-use production work in agriculture. Own-use production of goods covers a diverse range of activities performed by people Both surveys identified a high proportion to produce goods for their own household of respondents engaged in the own-use or family consumption. This includes production of goods. The comparison subsistence farming or fishing activities, but between the two surveys was impacted by also activities such as gathering firewood, the same issues identified in the case of fetching water, hunting, gathering wild employment, namely, in the MLSS in wave 1, foodstuffs, manufacturing clothing or other a relative undercount of people engaged in household goods, construction and major family farming activities and a reduction of renovation, or the preservation of foodstuffs the gap by wave 2, as follows: for consumption later. Thus, it covers many activities that are especially prevalent in  In wave 1, the LFS revealed that developing countries and, in some cases, 45.0 percent of respondents had subject to important gender asymmetries, engaged in own-use production of goods including the fact that those activities in the reference week, compared with predominantly carried out by women are less 37.7 percent in the MLSS (see Figure 4). frequently captured in the statistics. This cross-survey gap was relatively similar among both male and female The LFS and MLSS questionnaires both respondents with the LFS recording 8 included questions on the various activities percentage points higher participation covered by own-use production, albeit with for men and 7 percentage points for different structures, flows, and wording. In the women. Both surveys indicated that the MLSS, a common set of questions was used to rate of participation was higher among distinguish employment in agriculture (that is, women than among men and by similar agricultural work for pay or profit) and own- margins. For example, in wave 1, the LFS use production work in agriculture (that is, for showed a gap between the participation own or family consumption). The distinction of men and women of 12.6 percentage between these two concepts was illuminated points, compared with 13.5 percentage in subsequent questions, which asked about points in the MLSS. By wave 2, these the intended use of the agricultural outputs gaps were 13.5 percentage points and (for pay or profit versus for own or family 15.0 percentage points, respectively. consumption). The revisions highlighted in the previous section that improved the  By wave 2, the gap between the surveys MLSS’s ability to capture employment in the had nearly disappeared (38.7 percent in agricultural sector (especially contributing the LFS, compared with 39.8 percent in 18 2. Main Findings the MLSS). In addition, the differences The average hours worked in own-use between survey instruments are for production of goods (see Figure 5) by those the most part no longer statistically engaged in that form of work were quite significant. The rate found by the LFS similar between the surveys in both waves, fell substantially between wave 1 and for example 6.3 hours per week in wave 2 wave 2, which can be linked to the timing of the LFS, compared with 6.2 hours in the of the surveys; wave 2 took place during a MLSS. This highlights that, while own- period of higher rainfall and thus greater use production of goods was a common restriction on movement and outdoor activity, it was a low intensity activity work. However, the participation levels relative to employment in this setting. reported in the MLSS rose moderately, illustrating the success of the updates made to the MLSS questionnaire.10 Figure 4 Participation rate (% of WAP) in own-use production of goods, by sex, wave of data collection and survey 45.0 Wave 1 TOTAL 37.7 38.7 Wave 2 39.8 38.2 Wave 1 30.4 MALES 33.1 Wave 2 31.6 50.8 Wave 1 43.9 FEMALES 43.5 Wave 2 46.6 0 10 20 30 40 50 60 LFS MLSS Source: Joint DCS, ILO, and World Bank pilot study in Sri Lanka, Wave 1 and Wave 2, March–October 2019. 10 As mentioned earlier, we assume that seasonal changes affected the LFS and MLSS proportionately, and therefore did not have a strong influence on the gap (in relative terms) between the two surveys. However, it remains a possibility that seasonality affected one survey instrument more than the other and thus contributed to the narrowing of the gap between the two surveys. 19 Measuring Women and Men’s Work | Main Findings from a Joint ILO and World Bank Study in Sri Lanka Figure 5 Average hours actually worked per week in own-use production of goods, by sex, wave of data collection and survey 6.6 Wave 1 6.3 TOTAL 6.3 Wave 2 6.2 7.6 Wave 1 6.9 MALES 6.7 Wave 2 7.5 1.6 6.0 Wave 1 1.0 5.9 FEMALES 0.6 6.1 Wave 2 2.1 5.5 0 1.0 2.0 3.0 4.0 5.0 6.0 7.0 8.0 LFS MLSS GENDER GAP Source: Joint DCS, ILO, and World Bank pilot study in Sri Lanka, Wave 1 and Wave 2, March–October 2019. Note: The red diamond indicates the gender gap in working time in the activities covered. The diamond is included on the bar of the gender with lower working time. If it is included on the bar for women, this shows the amount by which the average working time of women in the activity was lower than among men and vice versa if it shown on the bar for men. In both the LFS and the MLSS, there were  The LFS identified more respondents who gender gaps in both waves. The average were engaged in crop farming to produce hours worked were higher among male foodstuffs for family or household use respondents, though the size of the gap compared with the MLSS (10.2 percent was somewhat different across the two versus 7.2 percent). This may be survey types. This suggests there is some linked to the structural differences in volatility or sensitivity in the reporting the questionnaires, particularly the on hours related to differences in the additional sets of questions in the LFS questionnaire content, but, on balance, this to ensure the complete coverage of was not substantial. this group. Evidently, this becomes important in the analysis of total labour In wave 2, splitting own-use production of input to agriculture, the identification goods into the various activities covered, one of agricultural households, or various may note interesting patterns (see Table 1). other analyses relying on measures of agricultural work (see below). 20 2. Main Findings  In some of the other activities covered, In both these activities, the number variations were observed despite the fact of men participating was essentially the surveys included identical questions. identical. No obvious explanation for these For example, the LFS identified more inconsistencies is available, indicating respondents engaged in the gathering of that the measurement of some own-use wild fruits. The difference was entirely production activities may be sensitive to among women (15.3 percent versus issues other than the wording of survey 11.4 percent). This situation was reversed questions, such as interviewer effects, in the engagement in the collection question placement and order, the context of firewood, a common activity in the effect, and so on. However, this is not survey areas. The MLSS identified more universal. There is a fairly high degree of respondents who were engaged in this consistency in the case of fetching water activity, all women (33.4 percent versus and other activities covered by own-use 27.2 percent). production of goods. Table 1 Shares of respondents of working age engaged in own-use production of goods in wave 2, by sex, type of activity and survey TOTAL MALES FEMALES MLSS (pps) MLSS (pps) MLSS (pps) Sign. Level Sign. Level Sign. Level Shares of Shares of Shares of Diff LFS- Diff LFS- Diff LFS- Coeff. of Coeff. of Coeff. of WAP (%) WAP (%) WAP (%) Std. Err. Std. Err. Std. Err. var. (%) var. (%) var. (%) Crop LFS 10.2 1.0 10.2 12.0 1.2 10.2 8.8 1.1 12.6 3.0 ** 4.2 *** 2.0 * farming MLSS 7.2 0.7 9.1 7.7 0.9 12.2 6.7 0.6 9.6 Rearing of LFS 1.1 0.3 25.9 1.1 0.4 31.3 1.1 0.4 32.3 -0.7 -0.5 -0.8 ** livestock MLSS 1.8 0.4 20.0 1.6 0.4 25.8 1.9 0.4 21.9 LFS 0.2 0.1 62.6 0.3 0.2 62.6 0.0 0.0 Fishing 0.2 0.3 0.0 MLSS 0.0 0.0 0.0 0.0 0.0 0.0 LFS 11.3 1.1 9.5 6.6 1.0 14.6 15.3 1.5 9.6 Hunting and Gathering 2.1 0.0 3.9 ** MLSS 9.2 0.8 8.9 6.6 0.8 11.7 11.4 1.1 9.7 Preserving LFS 1.7 0.3 16.1 1.1 0.3 29.9 2.2 0.4 18.1 0.4 0.5 0.3 food MLSS 1.3 0.2 18.7 0.6 0.2 36.6 1.9 0.4 20.6 Fetching LFS 10.3 1.0 9.7 9.9 1.1 11.1 10.5 1.2 11.6 -0.5 -1.4 0.2 water MLSS 10.8 0.9 7.9 11.3 1.1 9.6 10.3 1.0 9.3 Collecting LFS 21.6 1.3 5.9 14.9 1.3 8.6 27.2 1.7 6.2 -3.4 ** -0.2 -6.1 *** firewood MLSS 25.0 1.2 4.9 15.1 1.2 8.2 33.3 1.6 4.9 Manufacturing of LFS 2.8 0.4 15.6 1.2 0.5 41.7 4.1 0.6 15.4 other household 0.4 0.5 0.4 goods MLSS 2.3 0.3 12.8 0.7 0.3 41.3 3.7 0.5 13.5 Building and major LFS 2.1 0.4 17.2 2.6 0.4 17.3 1.6 0.4 25.2 -0.5 -0.3 0.7 renovations MLSS 2.6 0.5 19.5 2.8 0.6 22.4 2.3 0.5 20.3 Source: Joint DCS, ILO, and World Bank pilot study in Sri Lanka, Wave 1 and Wave 2, March–October 2019. Significance levels: * = 10 percent ** = 5 percent *** = 1 percent 21 Measuring Women and Men’s Work | Main Findings from a Joint ILO and World Bank Study in Sri Lanka 2.2.2 Own-use provision women contributed approximately four of services fifths of all the time spent in own-use provision of services. In the past, such gaps Own-use provision of services refers to often went unreported. the many services people provide to their own households or families without pay, The results of wave 1 of the Sri Lanka including housework, gardening, small pilot study highlighted how sensitive the repairs, arranging services, caring for measurement of time spent in unpaid children, the elderly or other dependent household and care work is to questionnaire family members, and so on. It aligns with design and survey implementation. Both the scope of Sustainable Development Goal surveys used the same approach and Indicator 5.4.1 – the proportion of time spent sequence of questions. However, there on unpaid domestic and care work, by sex, were some slight differences in wording age and location – meaning that statistics and implementation. These differences may compiled on own-use provision of services seem minor from a designer’s perspective, can form the basis for this indicator. In but they were evidently substantial from the addition, it forms part of the analysis of respondent’s perspective. the total burden of work and is critical in understanding the differences between The LFS showed a higher level of women and men in the contributions to participation in wave 1 (87 percent of the household well-being, often misrepresented working-age population (WAP) versus if only employment is counted. 81.1 percent in the MLSS). The difference was entirely associated with the male Before addressing the lessons learned on respondents; the levels among women the measurement of this form of work, it were essentially equivalent (see Figure 6). is useful to show the value these data can Closer analysis of the data showed that the generate, especially in highlighting gender difference among men was concentrated in gaps that are often not visible because only one of the three districts covered by the of infrequent measurement. Female survey, suggesting some type of local effect respondents in the study in Sri Lanka had that may be related to inconsistencies approximately three times as much working in interviewer practices or instructions, time per week on average in the provision of making it less likely that male respondents services for own use than male respondents to the MLSS would report these activities. engaged in the activity. This was true in both surveys, in both waves and in applying any During the preparations for wave 2, of the different measurement approaches interviewer training was used to emphasize tested. If the differences in participation the need to ask all respondents about the are combined with average working hours, full list of activities exhaustively, and this 22 2. Main Findings Figure 6 Participation rate (% of WAP) in own-use provision of services, by sex, wave of data collection and survey 87.0 Wave 1 81.1 TOTAL 87.6 Wave 2 84.6 81.5 Wave 1 69.6 MALES 82.3 Wave 2 75.8 91.5 Wave 1 90.7 FEMALES 92.0 Wave 2 91.9 0 20 40 60 80 100 LFS MLSS Source: Joint DCS, ILO, and World Bank pilot study in Sri Lanka, Wave 1 and Wave 2, March–October 2019. seems to have been successful in reducing relative to the LFS (24.8) (see Figure 7). The the gap. Notably, the estimates between average hours calculation excluded those wave 1 and wave 2 of the LFS were consistent respondents who did not report that they among both men and women and among had undertaken any own-use provision of women responding to the MLSS. However, services during the reference period (that is, the participation of men recorded in the zero hours reported). MLSS rose between the two waves of data collection, thus narrowing the overall gap Disaggregation by activity showed that between the surveys. the difference was concentrated in childcare and adult care (see Table 2). The The sensitivity of measurement was even number of hours spent in other activities, more obvious in the reporting on the including cleaning, cooking, and so on, were time spent in various unpaid household comparable between the two surveys. For service activities. In the first wave of data example, the MLSS showed average reported collection, the average hours captured by hours in childcare of 42.5 hours in wave 1, the MLSS (34.2) were 38.0 percent higher compared with 15.4 hours in the LFS. 23 Measuring Women and Men’s Work | Main Findings from a Joint ILO and World Bank Study in Sri Lanka Figure 7 Average hours actually worked in the own-use provision of services, by sex, wave of data collection and survey 24.8 Wave 1 34.2 TOTAL 25.3 Wave 2 26.1 12.8 20.8 Wave 1 16.4 29.4 MALES 12.0 23.4 Wave 2 11.9 24.1 33.6 Wave 1 45.7 FEMALES 35.4 Wave 2 36.0 0 10.0 20.0 30.0 40.0 50.0 LFS MLSS GENDER GAP Source: Joint DCS, ILO, and World Bank pilot study in Sri Lanka, Wave 1 and Wave 2, March–October 2019. Note: Averages were calculated only for those respondents who reported that they had carried out some own-use provision of services during the reference period. The red diamond indicates the gender gap in working time in the activities covered. The diamond is included on the bar of the gender with lower working time. If it is included on the bar for women, it shows the amount by which the average working time of women in the activity was less relative to men and vice versa if it is shown on the bar for men. A review of the two questionnaires and of gave examples of the types of activities interviewing practices identified multiple, involved. The purpose of this additional related possible causes of this disparity, text was to emphasize that only time including the following: spent in “active” care (for example, bathing children, taking children to school, tutoring,  The MLSS questionnaire asked respondents and so on) was to be reported. There was if they had performed any of a list of no such additional statement in the MLSS. activities. One was “Look after children (17 Likewise, the enumerator training for the years or younger)”. The LFS had a dedicated MLSS did not emphasize active caregiving. question and longer wording referring to: “looking after children living in this  The LFS asked only one question per household or the children of your relatives”. activity to capture working time in the previous week (that is, “Last week  LFS interviewers were instructed to read did you...”). The MLSS, following the an additional text below the question that recommendation of the DCS, split this 24 2. Main Findings into two questions: first, on the number of approach could be isolated from possible days, and, then, the average hours per day. implementation differences between the two As discussed further below, the number survey types. of measured hours worked in the own-use production of services tend to be greater Harmonizing the two questionnaires reduced in the two-question approach than in the the gap between the two survey types (see one-question approach. Table 2). The estimates produced by the LFS were relatively stable between wave 1 and wave The two questionnaires were harmonized in 2 (for instance, the average time spent in care wave 2, adopting the question wording and activities of 16.1 hours in wave 1 versus 15.1 approach of the LFS. In addition, both surveys hours in wave 2). The estimates of the MLSS fell included an experiment to assess the impact by half, from 43.8 hours on average on all care of the use of one question on the hours activities to 21.9 hours. While a statistically reported for work during the previous week significant gap between the surveys remained, versus the two-question approach (number there was a far higher degree of consistency, of days and hours per day). Both approaches and the knock-on effect was that the estimates were administered to half the sample for each of time spent on all own-use provision of of the questionnaires in wave 2 to ensure that services in wave 2 were similar between the the effect of the one- versus the two-question two surveys (see Figure 7). Table 2 Average hours actually worked during the reference week by respondents engaged in care activities, by sex, wave of data collection and survey AVERAGE HOURS ACTUALLY WORKED IN THE REFERENCE WEEK TOTAL MALES FEMALES Sign. Level Sign. Level Sign. Level LFS-MLSS LFS-MLSS LFS-MLSS Coeff. of Coeff. of Coeff. of Std. Err. Std. Err. Std. Err. var. (%) var. (%) var. (%) Hours Hours Hours Diff Diff Diff LFS 15.1 0.5 3.5 9.3 0.6 6.5 18.5 0.7 3.6 Care Activities -6.7 *** -2.6 *** -8.3 *** MLSS 21.9 0.7 3.2 11.8 0.7 5.6 26.8 1.0 3.6 Care for LFS 11.2 0.9 8.0 12.1 2.0 16.5 10.6 0.8 7.2 WAVE 2 adults -3.4 ** 2.1 -6.0 *** of MLSS 14.5 1.3 9.0 10.1 1.5 14.4 16.7 1.7 10.3 which LFS 14.6 0.5 3.5 8.0 0.4 5.6 18.3 0.7 3.8 Care of children -6.7 *** -3.5 *** -7.7 *** MLSS 21.3 0.7 3.2 11.4 0.7 6.3 26.0 0.9 3.6 LFS 16.1 0.6 3.5 10.3 0.5 5.3 19.6 0.8 3.9 Care Activities -27.6 *** -18.4 *** -31.6 *** MLSS 43.8 1.1 2.5 28.7 1.4 4.9 51.2 1.5 2.8 Care for LFS 11.6 0.9 8.1 9.5 1.6 16.4 12.8 1.0 7.7 WAVE 1 adults -21.6 *** -17.5 *** -23.5 *** of MLSS 33.2 2.7 8.2 27.0 4.0 14.9 36.3 3.0 8.3 which LFS 15.4 0.6 3.8 9.5 0.5 5.3 18.8 0.8 4.2 Care of children -27.1 *** -17.2 *** -31.6 *** MLSS 42.5 1.3 3.0 26.7 1.5 5.7 50.3 1.6 3.2 Source: Joint DCS, ILO, and World Bank pilot study in Sri Lanka, Wave 1 and Wave 2, March–October 2019. Significance levels: * = 10 percent ** = 5 percent *** = 1 percent 25 Measuring Women and Men’s Work | Main Findings from a Joint ILO and World Bank Study in Sri Lanka The additional experiment to compare the 30 percent more than the one-question use of one question versus two questions approach. This pattern was repeated among on the number of hours worked also yielded both men and women albeit with slightly interesting conclusions (see Figure 8). different gaps. A possible explanation is The results for both surveys were highly that the rounding of daily averages in the consistent. For example, the one-question two-question approach leads to a relative approach yielded averages of 22.0 hours overestimation relative to the one-question spent on the own-use provision of services approach. However, while the direction and in the LFS, compared with 22.7 hours in scale of the impact are quite consistent, the MLSS. The two-question approach which of the two sets of results may be yielded 28.8 and 29.7 hours, respectively, more valid is not certain. Figure 8 Average hours actually worked during the reference week by own-use providers of services, by sex, survey and type of questions used to capture working time One question 22.0 LFS Two questions 28.8 TOTAL One question 22.7 MLSS Two questions 29.7 One question 10.2 LFS Two questions 13.8 MALES One question 11.0 MLSS Two questions 12.9 One question 31.0 LFS Two questions 39.7 FEMALES One question 31.0 MLSS Two questions 41.0 0.0 5.0 10.0 15.0 20.0 25.0 30.0 35.0 40.0 LFS MLSS Source: Joint DCS, ILO, and World Bank pilot study in Sri Lanka, Wave 1 and Wave 2, March–October 2019. Note: Each of the two survey samples was divided into two random groups. The questions about number of hours worked were asked using only one question to one of the groups (that is, “How many hours did you spend doing this last week?”) and with two questions among the other group (that is, “Last week, on how many days did you do this work?” and “And, on average, how many hours per day did (you/NAME) spend doing this last week?”). 26 2. Main Findings On balance, the analysis of the participation Volunteers project from which the findings and time spent in own-use provision of have been published separately. services reinforces the message that the measurement of these activities is highly Given the findings already discussed, a sensitive to questionnaire content design few conclusions are evident, including the and implementation. Seemingly minor following: differences in implementation can yield substantial differences in results. This  The large majority of the WAP engaged suggests that more study of this topic is in some form of work (given that close needed, for example, to allow comparisons to 90 percent were engaged in the between the results generated by diary- own-use provision of services). In fact, based approaches and the results of the 92.8 percent of the LFS respondents and types of stylized questions used in the 91.2 percent of the MLSS respondents Sri Lanka pilot study. This might enable were engaged in at least one form of firmer conclusions to be drawn on the best work during the survey reference period approaches to balance respondent burden in wave 2 (see Figure 9). There was a gap and data quality. between the surveys that was driven by the factors discussed above, but the gap narrowed between wave 1 and wave 2 as a 2.3 Concurrent result of the changes made in the MLSS, such as those that led to the identification Work Activities of more employed respondents. and the Total Burden of Work  Women were more likely than men to be engaged in some work. The overall The recognition that people may be engaged female participation rate was similar in in multiple forms of work during a single both surveys, particularly in wave 2. The reference period is an important evolution LFS recorded higher male participation associated with the 19th ICLS standards. This rates, resulting in a smaller gender gap enables a look not only at participation rates than the MLSS. in different forms of work, but also the extent to which people mix these activities and, by While rates of participation and differences extension, their total burden of work. The across subgroups are clearly of interest, the Sri Lanka pilot study did not cover volunteer analytical possibilities become particularly work or unpaid trainee work, both of which rich using data captured on the number of are also defined within the standards. The hours worked. measurement of volunteer work is the subject of a dedicated ILO–United Nations 27 Measuring Women and Men’s Work | Main Findings from a Joint ILO and World Bank Study in Sri Lanka Figure 9 Shares (% of WAP) of respondents engaged in one of the forms of work (employment, own-use production of goods, own-use provision of services), by sex, wave of data collection and survey 92.4 Wave 1 89.8 TOTAL 92.8 Wave 2 91.2 91.1 Wave 1 87.0 MALES 92.4 Wave 2 89.2 93.4 Wave 1 92.2 FEMALES 93.2 Wave 2 92.8 80.0 82.0 84.0 86.0 88.0 90.0 92.0 94.0 96.0 LFS MLSS Source: Joint DCS, ILO, and World Bank pilot study in Sri Lanka, Wave 1 and Wave 2, March–October 2019. Figure 10 shows the average hours spent on exhibited a greater average number of all the forms of work by respondents who working hours than men. In wave 1, the gap had engaged in any of these forms of work was quite wide in the MLSS (61.4 hours (that is, 92.8 percent of all respondents to the among women versus 52 hours among men), LFS in wave 2 and 91.2 percent in the case but, by wave 2, both surveys showed an of the MLSS). The total work burden (across almost identical gap, with women reporting the different forms of work) was, on average, an average of approximately 6 hours more over 50 hours per week in both waves in both working time per week than men (53.3 hours surveys. In wave 1, there was a clear gap versus 46.8 hours in the MLSS). The between the LFS (50.5 hours) and the MLSS reduction in the gender gap in the MLSS (57.3 hours). Following the various changes between the two waves partly reflects the described above, the gap disappeared in emphasis on active caregiving in wave 2, wave 2. implying that at least part of the additional hours reported in wave 1 reflected the Another conclusion that may be drawn is reporting of passive caregiving. While this that, in both waves in both surveys, women is obviously important from a measurement 28 2. Main Findings Figure 10 Average hours actually worked by people engaged in one of the forms of work (employment, own-use production of goods, own-use provision of services), by sex, wave of data collection and survey 50.5 Wave 1 57.3 TOTAL 50.0 Wave 2 50.4 2.6 49.0 Wave 1 9.4 52.0 MALES 6.1 46.7 Wave 2 6.5 46.8 51.6 Wave 1 61.4 FEMALES 52.8 Wave 2 53.3 0 10.0 20.0 30.0 40.0 50.0 60.0 70.0 LFS MLSS GENDER GAP Source: Joint DCS, ILO, and World Bank pilot study in Sri Lanka, Wave 1 and Wave 2, March–October 2019. Note: The red diamond indicates the gender gap in working time in the activities covered. The diamond is included on the bar of the gender with lower working time. If it is included on the bar for women, it shows the amount by which the average working time of women in the activity was less than among men and vice versa if it shown on the bar for men. perspective, it is clearly also relevant to The report shows the average number the interpretation of the data. The time of hours worked in each form of work spent on passive caregiving was, by design, individually. Figure 10 illustrates the total excluded in wave 2 in both surveys in the work burden of all respondents who carried interests of comparability, but would show out any of the forms of work. A look at how an additional gender gap if it were also the work burden of individuals is distributed measured and reported. This does not across forms of work is also revealing. mean that passive caregiving should, by default, be excluded from consideration, Starting with employment, Figure 11 shows although, as has been acknowledged by the that employed respondents worked a UN Expert Group on Time-Use Statistics substantial number of additional hours in (UNSD 2019), more discussion would be other forms of work. Specifically, employed required to provide a clearer definition as a respondents to the LFS in wave 2 worked basis for the relevant measurement through an average of 2.7 additional hours in the household surveys. own-use production of goods and 20.5 hours 29 Measuring Women and Men’s Work | Main Findings from a Joint ILO and World Bank Study in Sri Lanka in the own-use provision of services. As emphasized throughout the report, the The results of the MLSS in wave 2 were analysis of working time is highly relevant similar. Overall, this meant that employed from a gender perspective and never more respondents to the LFS worked 61.1 hours so than in the analysis of total working per week overall across the three forms of time across forms of work. Taking the work, compared with 63.6 hours in the MLSS. wave 2 data of the MLSS to illustrate (see This gap was created by the relatively minor Figure 11), employed men worked 43.9 hours remaining differences in employment and in employment on average, an additional own-use provision of services. The gaps 2.6 hours in own-use production of goods between the surveys were much less than and 11.4 hours in the own-use provision those found in wave 1, demonstrating that of services. The results of the LFS were the various questionnaire changes were at relatively similar. Over three quarters of all least partially successful in improving the working time among employed men was thus consistency between the two surveys. in employment. Figure 11 Average hours worked by respondents in employment and additional hours worked in the own-use production of goods and services, by sex, wave of data collection and survey WAVE 1 LFS 38.4 3.2 19.7 61.4 MLSS 68.7 TOTAL 39.9 2.4 26.4 WAVE 2 LFS 37.9 2.7 20.5 61.1 MLSS 39.3 2.5 21.9 63.6 WAVE 1 LFS 10.9 42.7 3.0 11.0 56.7 MLSS 16.1 44.3 2.3 15.2 61.8 MALES WAVE 2 LFS 15.2 42.2 2.3 10.3 54.8 MLSS 12.8 43.9 2.6 11.4 57.8 WAVE 1 LFS 32.5 3.5 31.7 67.7 MLSS 34.1 2.5 41.3 77.9 FEMALES WAVE 2 LFS 31.8 3.3 34.8 70.0 MLSS 33.8 2.5 34.4 70.6 0.0 20.0 40.0 60.0 80.0 Hours worked in employment (all jobs) Additional hours worked in own-use production of goods Additional hours worked in own-use provision of services GENDER GAP Source: Joint DCS, ILO, and World Bank pilot study in Sri Lanka, Wave 1 and Wave 2, March–October 2019. Note: The averages for working time in employment are not identical to those in Figure 2 because some employed respondents did not provide information on time spent on the own-use production of goods or the own-use provision of services and are thus excluded from the analysis in Figure 12. The red diamond indicates the gender gap in working time in the activities covered. The diamond is included on the bar of the gender with lower working time. If it is included on the bar for women, it shows the amount by which the average working time of women in the activity was lower than among men and vice versa if it shown on the bar for men. 30 2. Main Findings Employed women showed quite a different burden faced by women who report high levels profile. While working time in employment of working time in unpaid household services, was less among women than among men even when employed. (33.8  hours in wave 2 in the MLSS), the additional working time in the own-use Another interesting pattern is underlined by a provision of services was high among women. look at the unpaid working time of those not in In fact, in both surveys in wave 2, the number employment. Figure 12 shows this information of hours spent in the own-use provision of for those who were engaged in the own-use services was even greater than the number production of goods but not employed. In of hours spent in employment. As a result, line with findings described above, there was women in employment showed around 25 a substantial gap between surveys in the percent more working time than men across reported hours in wave 1 (59.9 hours in the the three forms of work in wave 2 of both MLSS versus 40.3 hours in the LFS), which surveys. This highlights clearly the double significantly narrowed in wave 2. Figure 12 Average hours worked by own-use producers of goods who are not in employment and additional hours worked in the own-use provision of services, by sex, wave of data collection and survey WAVE 1 LFS 6.4 33.9 40.3 MLSS 59.9 TOTAL 7.5 52.5 WAVE 2 LFS 6.5 35.3 41.8 MLSS 7.2 38.2 45.4 WAVE 1 LFS 9.7 11.7 20.8 30.5 MLSS 13.6 18.4 30.8 44.3 MALES WAVE 2 LFS 8.0 18.0 20.1 26.0 MLSS 9.9 16.7 23.7 26.6 WAVE 1 LFS 5.8 36.4 42.2 MLSS 6.4 56.3 62.7 FEMALES WAVE 2 LFS 6.1 40.0 46.1 MLSS 6.6 43.7 50.3 0.0 10.0 20.0 30.0 40.0 50.0 60.0 70.0 80.0 Additional hours worked in own-use production of goods Additional hours worked in own-use provision of services GENDER GAP Source: Joint DCS, ILO, and World Bank pilot study in Sri Lanka, Wave 1 and Wave 2, March–October 2019. Note: The red diamond indicates the gender gap in working time in the activities covered. The diamond is included on the bar of the gender with lower working time. If it is included on the bar for women, it shows the amount by which the average working time of women in the activity was lower than among men and vice versa if it shown on the bar for men. 31 Measuring Women and Men’s Work | Main Findings from a Joint ILO and World Bank Study in Sri Lanka The gender disparity shown in Figure 12 The revised standards differentiate between is striking. A look at wave 2 data reveals employment and own-use production work in that, in the LFS, women in this situation agriculture and other sectors. This enables (not employed, but performing the own-use a wide range of analyses and may allow production of goods) worked 6.1 hours on for more meaningful policy interventions, average per week in the own-use production for example by recognizing that people of goods. Men, by comparison, worked 8 hours engaged in subsistence farming may be per week on average. However, women did an seeking or available for paid work, which additional 40 hours of work in the own-use was not reflected in the old standards. provision of services, compared with 18 hours Distinguishing between employment and among men. The pattern was relatively similar own-use production work in agriculture can in the MLSS, again showing that women, to also be challenging in contexts where mixed a greater extent than men, retained a high agriculture prevails (Gaddis et al. 2020b). number of working hours in unpaid forms of work, regardless of their employment status. One of the more straightforward indicators that can be generated within this new framework is the total number or proportion 2.4 Work in Agriculture of people engaged in agricultural or fishing activities. Figure 13 illustrates the difficulty and Fishing in capturing some of the agricultural work in wave 1 of the MLSS, and the improvement in Improving sectoral analysis is one important wave 2. While the proportion of agriculture objective of the 19th ICLS standards. and fishing workers identified by the Specifically, the labour input to any sector or LFS remained relatively stable (close to grouping of working activities is a combination 30 percent), the proportion identified by the of paid and unpaid work. This is particularly MLSS rose from 20.4 percent to 26.0 percent. important in certain sectors. Agriculture The increase in the MLSS was greater among and fishing are key examples, while care female than among male respondents. work is another example. In rural areas of developing countries, a large proportion of the There will be many other indicators of population is typically engaged in some type interest for those seeking to understand the of agricultural work. Some, possibly a large agricultural sector, such as the proportion of part, of this work will be performed with the agricultural households. Evidently, any issues intention of producing goods for consumption in the measurement of agricultural work by the household or family, while another part directly impact indicators on the number of involves producing goods for sale. Indeed, agricultural households if the classification households often engage in mixed production, of households is based on the presence of keeping some and selling the rest. agricultural workers in the household. 32 2. Main Findings Figure 13 Participation rate in agriculture and fishing (% of WAP), by sex, wave of data collection and survey 29.8 Wave 1 20.4 TOTAL 29.1 Wave 2 26.0 33.4 Wave 1 24.7 MALES 32.5 Wave 2 29.1 26.9 Wave 1 16.8 FEMALES 26.2 Wave 2 23.5 0.0 5.0 10.0 15.0 20.0 25.0 30.0 35.0 LFS MLSS Source: Joint DCS, ILO, and World Bank pilot study in Sri Lanka, Wave 1 and Wave 2, March–October 2019. Note: Includes employed whose main or second jobs are in agriculture, forestry or fishing (International Standard Industrial Classification codes 01, 02, and 05) as well as own-use producers of goods who are engaged in crop farming, animal rearing or fishing. In summary, any comprehensive analysis subsequent questions to establish whether of work in agriculture or related indicators the agricultural outputs are mainly intended requires that surveys cover both employment for sale or for own use. and own-use production work in agriculture comprehensively in questionnaires. It is thus recommended that, in countries 2.5 Labour with substantial prevalence of agricultural activities, both forms of work should be Underutilization covered in any survey measuring labour input. The lessons of the Sri Lanka pilot A consistent set of questions was used in study should be borne in mind in designing the two questionnaires to capture job search the questionnaires. In particular, the and availability as a basis for estimates on questionnaires should be designed to capture unemployment and the degree of attachment both forms of work in a dedicated manner by to the labour force of those not in employment. including questions that seek to identify all Both questionnaires also included questions to agricultural work comprehensively, as well as identify time-related underemployment, which 33 Measuring Women and Men’s Work highlights people who report that they have an insufficient volume of paid work relative to their preferences. The LFS identified more respondents in time- related underemployment, and the MLSS identified more unemployed respondents. The most obvious explanation for this is the difference between the surveys in the number of employed respondents identified. Given a higher number of employed respondents, particularly among those with low hours of work, it is unsurprising that the LFS identified more people willing and available to work additional hours (the time- related underemployed). Similarly, because the MLSS identified fewer employed respondents than the LFS and, especially, seemed to undercount respondents with low working hours, it is suitable and can be recommended as a not surprising that it identified more people basis for the development of questionnaires seeking and available for work who were not covering these variables. identified as employed. The gaps noted above were lower in wave 2 2.6 Other Issues of Note than in wave 1, indicating that the primary driver of the gaps was the inconsistency in the measurement of employment. Also given In line with earlier pilot studies, the that the questions used to capture labour importance of translation and national underutilization were the same across the adaptation was evident in the Sri Lanka pilot two surveys, any additional survey-specific study. For example, some of the difficulties impacts cannot be identified. in the measurement of farming work in wave 1 of data collection in the MLSS can be linked Furthermore, the sets of questions used, to difficulties in identifying appropriate which were based on published ILO model everyday terminology in Sinhalese for questionnaires, operated well in the field. some of the farming-related questions. This suggests that the questions were These issues were addressed in wave 2. 34 2. Main Findings For example, based on the experiences in differences were noted between the wave 1, the MLSS questionnaire in wave 2 LFS and the MLSS, such as the number avoided more abstract terminology (e.g. own of respondents with multiple jobs and account crop farming) and instead opted variations in the identification of work in for simpler terms with examples (e.g. work agriculture as main or secondary jobs. on a family farm to prepare or maintain the Conclusions can be drawn on some of the land, or to plant, grow or harvest any crops sources of the differences, but, for others, vegetables or fruits). (See Annex 3, Table 3.2 the conclusions are not straightforward. for additional examples.) This presumably However, it is clear that the greatest contributed to the improvements in the measurement issues centre on people in consistency of the wave 2 results. informal employment, particularly those in own-account activities in both agriculture The analysis in this report focuses on and other sectors. This suggests a need indicators of participation in various to undertake studies on appropriate forms of work. A related, important topic questionnaire content to identify and is the measurement and analysis of the describe properly the full range of people’s characteristics of jobs, including the jobs and businesses, particularly in the case number of jobs held. Various important of informal employment. 35 3 Summary Conclusions 3. Summary Conclusions The first key conclusion is that the field differences in the measurement of key experiment undertaken in Sri Lanka labour-related variables, particularly generated a wealth of rich data that may participation in the various forms of be used to identify good practices in paid and unpaid work. The impact of the questionnaire design and apply the latest changes made to the MLSS before wave 2 standards in the domain of labour statistics. of data collection suggests that some The design of the study enables the of the differences can be reduced if not existing guidance to be extended to surveys removed entirely through relatively minor other than the LFS, although absolute changes in questionnaire content or survey consistency between the LFS and other implementation. household surveys remains unlikely given the differences in design and in objectives. The A variety of other sensitivities could be depth and breadth of conclusions generated identified, such as the sensitivity of the through the study would not realistically have measurement of working time in unpaid been possible through another mechanism. work to the measurement approach (for example, one or two questions). This was From a gender perspective, the value of the especially evident in unpaid care work. In study is difficult to overstate. As highlighted addition, all surveys should emphasize good in this report, a much larger part of women’s translation and national adaptation, as well (compared to men’s) work tends to be invisible as interviewer training and supervision, to or at risk of being underreported or simply not promote consistency in measurement. measured at all in official statistics. The value of the data is the subject of additional reports The risks of misclassification and (Discenza and Walsh 2020a, 2020b), but it is measurement difficulties were concentrated already clear that pilot studies such as this in the case of people engaged in certain one are extremely valuable in improving the types of activities. For instance, the higher measurement of paid and unpaid work. This risk of misclassification or undercounting is the subject of a long, ongoing process, was clear among people engaged in casual, which gained significant momentum through low-hours work or people helping on family the adoption of the 19th ICLS standards and farms or businesses. This is highly relevant has been the main focus of the ILO and World in the measurement of outcomes by sex Bank agenda to operationalize the 19th ICLS given that these types of activities were standards and improve survey methods more common among women than among on labour through the Women’s Work and male respondents, which is likely to be true in Employment Partnership. many settings. On the measurement side, the Sri Lanka The questionnaires were successful in study revealed important cross-survey capturing a range of paid and unpaid working 37 Measuring Women and Men’s Work | Main Findings from a Joint ILO and World Bank Study in Sri Lanka activities. This has unlocked great analytical Nonetheless, the model questionnaires, potential through, for example, the deeper guidance and tools now or soon to become understanding of gender gaps in working available are an excellent reference for those activities and labour market engagement. facing the task of designing a questionnaire An important future goal, urged forward by to capture work- and labour-related issues this study and the related work to develop through a household survey. The appropriate guidance on good measurement practices, choice of content for a particular survey is the mainstreaming of the measurement of involves a balance among the objectives unpaid working activities to enable this type of the survey, the desired outputs and the of analysis on a regular, wide-scale basis. appropriate level of respondent burden, ranging from minimal approaches typically The harmonization of questionnaire content included in population censuses to the most represents a way to improve the consistency detailed content one expects from the LFS. of measurement, but it cannot be assumed that absolute consistency can be achieved The findings presented in this report are a or that the need for a national process of subset of the many findings possible from adaptation and testing can be avoided. such studies. The findings will be used to Differences in surveys and across countries generate guidance and additional technical mean that questionnaires should be adapted notes to be published at a later date, as well to context and fully tested to enhance the as to facilitate updates to the guidance, tools quality of the statistics generated. Other and support provided by the ILO for LFSs differences in survey objectives or aspects of and the World Bank for the next round of methodology, such as sample size, mean that household surveys supported by the Living full comparability across surveys is unlikely. Standards Measurement Study team. 38 3. Summary Conclusions 39 Measuring Women and Men’s Work | Main Findings from a Joint ILO and World Bank Study in Sri Lanka References Anker, Richard, and Martha Anker. 1989. Comblon, Virginie, and Anne-Sophie Robilliard. “Measuring the Female Labour Force in Egypt.” 2017. “Are Female Employment Statistics More International Labour Review 128 (4): 511–20. Sensitive Than Male Ones to Questionnaire Design? Evidence from Three West African Bardasi, Elena, Kathleen Beegle, Andrew Dillon, Countries.” DIAL Working Paper DT/2015- and Pieter Serneels. 2011. “Do Labor Statistics 22 (version February 2017), Unité Mixte de Depend on How and to Whom the Questions Recherche DIAL (Développement, Institutions Are Asked? Results from a Survey Experiment et Mondialisation), Institut de Recherche pour in Tanzania.” World Bank Economic Review 25 le Développement, Université Paris-Dauphine, (3): 418–47. Paris. Beaman, Lori, and Andrew Dillon. 2012. “Do Desiere, Sam, and Valentina Costa. 2019. Household Definitions Matter in Survey “Employment Data in Household Surveys: Taking Design? Results from a Randomized Survey Stock, Looking Ahead.” Policy Research Working Experiment in Mali.” Journal of Development Paper 8882, World Bank, Washington, DC. Economics 98 (1): 124–35. Discenza, Antonio R., and Kieran Walsh. 2020a. Beegle, Kathleen, Joachim De Weerdt, Jed “Closing Gender Data Gaps in the World of Friedman, and John Gibson. 2012. “Methods Work: Role of the 19th ICLS Standards.” of Household Consumption Measurement Lessons from the COVID-19 Pandemic, through Surveys: Experimental Results form Statistical Methodology Series 8 (October), Tanzania.” Journal of Developing Economics 98 ILO Department of Statistics, International (1): 3–18. Labour Office, Geneva. https://ilo.org/ Benes, Elisa M., and Kieran Walsh. 2018a. “ILO wcmsp5/groups/public/---dgreports/---stat/ LFS Pilot Studies in Follow Up to the 19th ICLS: documents/publication/wcms_757964.pdf. Background, Objectives, and Methodology.” Discenza, Antonio R., and Kieran Walsh. 2020b. Statistical Methodology Series 1 (May), ILO “Lessons from the COVID-19 Pandemic: Gender Department of Statistics, International Relevance of the 19th ICLS Standards.” ILO Labour Office, Geneva. https://www.ilo.org/ Brief (October), International Labour Office, wcmsp5/groups/public/---dgreports/---stat/ Geneva. https://ilo.org/wcmsp5/groups/ documents/publication/wcms_627873.pdf. public/---dgreports/---stat/documents/ Benes, Elisa M., and Kieran Walsh. 2018b. publication/wcms_757969.pdf. “Measuring Employment in Labour Force Durazo, Josefine, Valentina Costa, Amparo Surveys: Main Findings from the ILO LFS Palacios-Lopez, Isis Gaddis. 2021. Pilot Studies.” Statistical Methodology “Employment and Own-Use Production in Series 4 (July), ILO Department of Statistics, Household Surveys : A Practical Guide for International Labour Office, Geneva. https:// Measuring Labor (English).” LSMS Guidebook www.ilo.org/wcmsp5/groups/public/--- Washington, D.C.: World Bank Group. dgreports/---stat/documents/publication/ http://documents.worldbank.org/curated/ wcms_635732.pdf. en/336141630489348107/Employment-and- Boserup, Ester. 1970. Women’s Role in Economic Own-Use-Production-in-Household-Surveys- Development. New York: St. Martin’s Press. A-Practical-Guide-for-Measuring-Labor Floro, Maria S., and Hitomi Komatsu. 2011. “Gender and Work in South Africa. What Can Time-Use Data Reveal?” Feminist Economics 17 (4): 33–66. 40 References Fowler, Floyd Jackson, Jr. 2004. “The Case for Koolwal, Gayatri. 2019. “Improving the More Split-Sample Experiments in Developing Measurement of Rural Women’s Employment: Survey Instruments.” In Methods for Testing Global Momentum and Survey Research and Evaluating Survey Questionnaires, edited Priorities.” Policy Research Working Paper by Stanley Presser, Jennifer M. Rothgeb, Mick 8840, World Bank, Washington, DC. P. Couper, Judith T. Lessler, Elizabeth Martin, Jean Martin, and Eleanor Singer, 173–88. New Langsten, Ray, and Rania Salen. 2008. “Two York: Wiley. Approaches to Measuring Women’s Work in Developing Countries: A Comparison of Survey Gaddis, Isis, Gbemisola Oseni, Amparo Palacios- Data from Egypt.” Population and Development Lopez, and Janneke Pieters. 2020a. Review 34 (2): 283–305. “Measuring Farm Labor: Survey Experimental Evidence from Ghana.” World Bank Economic Mahmud, Simeen, and Sakiba Tasneem. 2011. “The Review. Published ahead of print, May 21, 2020. Under Reporting of Women’s Economic Activity https://doi.org/10.1093/wber/lhaa012. in Bangladesh: An Examination of Official Statistics.” BDI Working Paper 01 (February), Gaddis, Isis, Gbemisola Oseni, Amparo Palacios- BRAC Development Institute, BRAC University, Lopez, and Janneke Pieters. 2020b. “Who Is Dhaka, Bangladesh. Employed? Evidence from Sub-Saharan Africa on Redefining Employment.” Policy Research Müller, Miriam, and Liliana Do Couto Sousa. 2020. Working Paper 9370, World Bank, Washington, “‘She Helps Me All the Time’: Underestimating DC. Women’s Economic Engagement in Rural Honduras.” Policy Research Working Paper Grosh, Margaret E., and Paul W. Glewwe, 9217, World Bank, Washington, DC. eds. 2000. Designing Household Survey Questionnaires for Developing Countries. Presser, Stanley, Jennifer M. Rothgeb, Mick P. Lessons from 15 Years of the Living Standards Couper, Judith T. Lessler, Elizabeth Martin, Measurement Study. 3 vols. Washington, DC: Jean Martin, and Eleanor Singer, eds. 2004. World Bank. Methods for Testing and Evaluating Survey Questionnaires. New York: Wiley. Heath, Rachel, Ghazala Mansuri, Bob Rijkers, William Seitz, and Dhiraj Sharma. 2020. “Measuring Sudarshan, Ratna M., and Shrayana Bhattacharya. Employment: Experimental Evidence from Urban 2008. “Through the Magnifying Glass: Women’s Ghana.” World Bank Economic Review. Published Work and Labour Force Participation in Urban ahead of print, June 12, 2020. https://doi. Delhi.” ILO Asia-Pacific Working Paper (May), org/10.1093/wber/lhaa014. Subregional Office for South Asia, International Labour Organization, New Delhi. https://www. ILO (International Labour Organization). 2013. ilo.org/newdelhi/whatwedo/publications/ “Resolution Concerning Statistics of Work, WCMS_123539/lang--en/index.htm. Employment, and Labour Underutilization, Adopted by the Nineteenth International UNSD (United Nations Statistics Division). 2019. Conference of Labour Statisticians (October “Conclusions and Recommendations: Third 2013).” International Labour Organization, Meeting of the Expert Group on Innovative and Geneva. https://www.ilo.org/global/ Effective Ways to Collect Time-Use Statistics statistics-and-databases/meetings-and- (EG-TUS), New York, USA, 4–6 November 2019.” events/international-conference-of-labour- Document ESA/STAT/AC.372/L3, Statistics statisticians/19/WCMS_230304/lang--en/ Division, Department of Economic and Social index.htm. Affairs, United Nations, New York. https:// unstats.un.org/unsd/demographic-social/ Kilic, Talip, and Thomas Pave Sohnesen. 2017. “Same meetings/2019/newyork-egm-tus-nov/3rd_ Question But Different Answer: Experimental EGM-Conclusions_and_recommendations- Evidence on Questionnaire Design’s Impact on FINAL%20posted.pdf. Poverty Measured by Proxies.” Review of Income and Wealth 65 (1): 144–65. 41 Measuring Women and Men’s Work | Main Findings from a Joint ILO and World Bank Study in Sri Lanka Annex 1. 19th ICLS Statistical Standards Resolution I of the 19th ICLS introduced a particular reference period, and (c) these major changes to the framework of different work activities should be measured definitions used to produce statistics on work and reported. and the labour market (ILO 2013). Relative to the 13th ICLS standards of 1982, it reduced This means that, unlike in the previous the scope of the statistical definition of standards, it is possible to capture fully the employment to work done for pay or profit and participation, contributions and working applied a wider definition of work, along with conditions of persons in, for example, the forms of work framework, to support the employment, volunteer work and own-use analysis of participation in paid and unpaid production work. This enables an analysis of productive activities. the total amount of hours spent by individuals “for the production of goods and services Among the various innovations within the for use by others or for own use” (across 19th ICLS standards is the recognition that: paid work, housework, work to produce (a) different forms of work can be defined foodstuffs, or other goods for own use, based on the intended destination of the volunteer work, and so on) and, if analysed at output and the motivation underlying the work the household level, can show the different (see Figure 1.1), (b) people can be engaged contributions of household members to in different forms of work simultaneously in overall household livelihoods and well-being. Figure 1.1 The forms of work framework Intended destination of for own for use production final use by others Forms of work own-use employment other* unpaid volunteer production work (work for pay trainee work or profit) work of of in market & in households services goods non-market producing units goods services Relation to within SNA production boundary 2008 SNA inside SNA General production boundary * Includes compulsory work performed without pay for others, not covered in the draft resolution. 42 Annex 1. 19th ICLS Statistical Standards Furthermore, it will be possible to evaluate but is supplemented by time-related how participation in one form of work underemployment and the newly introduced impacts participation in another form of concept and measure of the potential labour work. This is a major departure from the force (see Figure 1.2). Together, these three previous standards under which each measures are recommended for the broader individual had only one status in one monitoring of insufficient labour absorption reference period (employed, unemployed, or, from a social perspective, the unmet not economically active), and the many need for employment. For dissemination unpaid working activities people undertook purposes, a range of labour underutilization were either conceptually included under indicators, LU1–LU4, based on different employment or not defined at all. combinations of the three measures has also been recommended. An additional important development within the 19th ICLS standards was Table 1.1 attempts to summarize the the establishment of a set of labour developments from the 1982 standards to underutilization indicators to supplement the 2013 standards. This illustrates that the the unemployment rate, which has, new forms of work framework, combined for decades, been a key labour market with the new labour underutilization indicator. The new indicators focus on indicators, offers the potential for far richer issues of insufficient labour absorption as insights into the productive activities in shown by an inadequate quantity of work. which people are engaged, how people Unemployment remains a key part of a range interact with the labour market, and how of measures of labour underutilization, these are interrelated. Figure 1.2 Components of labour underutilization to monitor the unmet need for employment WORKING AGE POPULATION Labour force Outside the labour force Employed Others outside the labour force Potential Labour Force Time-related • seeking, not available Unemployed underemployd • not seeking, but wanting and available Labour underutilization (unmet need for employment) 43 Measuring Women and Men’s Work | Main Findings from a Joint ILO and World Bank Study in Sri Lanka With all the above said, it should be decades. The hope and expectation is that the recognized that the 13th ICLS standards adoption of the 19th ICLS will be a precursor played an important role in providing a clear to a similar expansion in the availability of basis for the development of labour market more comprehensive data on paid and unpaid statistics. This has supported the major work, and labour market engagement over the expansion in availability of labour market coming years and decades. related data across all regions in recent Table 1.1 Comparison of the scope of statistical standards 1982 standards (13th ICLS) 2013 standards (19th ICLS) Employment definition All activities within the production Work done in exchange for pay or profit boundary in the system of national accounts, including some unpaid activities such as subsistence farming Recognition of forms of work No Yes beyond employment Ability to measure the total Not completely: the framework only Yes, along with the fact that multiple forms of work burden of work (paid and identified one status during a reference could be performed in the same reference period (such unpaid) period, but did apply a wide definition as housework, farming for own-use, employment, and of employment, which included some so on) but not all unpaid work Relationship to the system of Conceptually a one-to-one relationship The different forms of work can be combined to align national accounts between employment and productive with both the system of national accounts production activities within the production boundary and the general production boundary boundary in the system of national accounts (not necessarily applied in practice) Labour underutilization Limited to unemployment indicators Recognition of unemployment, time-related and subsequently time-related underemployment and the “potential labour force”, underemployment which combine into four labour underutilization indicators 44 Annex 2. Methodology of the Pilot Study The pilot study in Sri Lanka was organized The aim of the study was to assess any as a comparative test of an LFS and a MLSS differences in results and to yield evidence questionnaire. While the two surveys have to allow the development of guidance on the different primary objectives (the generation implementation of the 19th ICLS standards of labour indicators versus the measurement in different types of household surveys. This of living standards and poverty), both overlap guidance will supplement and be used to in their coverage of labour and work. In update supporting materials already available the case of the LFS, the primary objective from the ILO for the LFS and from the World is to generate labour market and work- Bank for the MLSS. related indicators, while the primary focus of the MLSS is on the use of work-related To allow different types of assessment information in the broader analysis of of the questionnaires, the study included poverty and living standards. The objectives both qualitative and quantitative stages. differ, but it nonetheless remains important The qualitative stage took the form of to ensure that, to the extent possible, cognitive interviews, while the quantitative respondents are consistently classified stage involved the field testing of the across the two surveys in line with labour- questionnaires with samples of households. related statistical standards, for example to These stages are described below. Each ensure that a respondent who is employed, stage was heavily supported through training, as defined in the standards, is classified as supervision and remote technical assistance employed regardless of the survey. by ILO and World Bank experts with the full support of staff from the DCS. DCS staff were In the case of the pilot study in Sri Lanka, engaged as interviewers and supervisors and the focus was on the implementation of to provide all other support and management the standards adopted at the 19th ICLS. required during the process. The cognitive (See Annex 1 for more details.) Specifically, interviews were further supported by both questionnaires were designed to an expert of the UK’s Office for National cover employment, labour underutilization, Statistics (ONS). the own-use production of goods, and the own-use provision of services as defined in Resolution I of the 19th ICLS (ILO 2013). 45 Measuring Women and Men’s Work | Main Findings from a Joint ILO and World Bank Study in Sri Lanka Cognitive interviewing the World Bank, be translated into Sinhalese. This proved useful and prompted updates to Cognitive interviews were carried out in both the English and Sinhalese versions of the rural areas of Kalutara and Negombo on the questionnaires. 22–27 October 2018. They aimed at verifying how respondents understood the questions from the two different questionnaires, Field testing the LFS-type questionnaire developed by the ILO and the MLSS-type labour module Given the general measurement objectives developed by the World Bank. It was a of the study and the need to assess the small-scale test carried out with about outcomes under different conditions of 20 respondents for each questionnaire, agricultural work intensity, the study protocol involving an in-depth probe of a small envisaged two waves of data collection. The number of crucial questions. two waves were intended to take place at two different points of an agricultural or fishing The cognitive interviewing stage was useful season, that is, a peak period in agricultural in identifying possible areas of difficulty and or fishing work (planting or harvesting staple enabled both questionnaires to be updated in food, and so on) and a period with much lower various ways. An important challenge faced intensity of agricultural or fishing activity was the fact that the cognitive interviewing (e.g. either when crops are growing or after evidently required that the questionnaires, the harvest when farmers are waiting for the originally prepared in English by the ILO and new season of planting). 46 Annex 2. Methodology of the Pilot Study of the fieldwork and the concurrent work schedule of the pilot and other surveys, it was decided to undertake the pilot study in three administrative districts outside the capital, Colombo, namely, Anuradhapura (the centre- north of the country), Galle (the southwest), and Kurunegala (centre). The sample included all the PSUs – corresponding to census enumeration areas – used for the current LFS in the fourth quarter of 2018, giving a total of 98 PSUs. From each PSU, ten new households (not interviewed for the regular LFS) were selected for the LFS and ten for the MLSS using a randomized approach, generating a total sample size of However, identifying the most suitable approximately 980 households for each of the districts and periods for data collection two questionnaires. while taking into account the requirements of the study and the time constraints was Interviews were administered through challenging because of the different crop computer assisted personal interviews calendars and seasons across the districts (CAPI) on tablets. The CAPI questionnaires of Sri Lanka. The most important aspect was were built using the World Bank’s Survey that, in Sri Lanka, there are three agricultural Solutions software. systems (dry, wet and intermediate zones); this means that: (a) different products are Fieldwork in wave 1 of the pilot study took produced in different districts (for instance, place in the three districts selected for the rice, rubber, maize, groundnuts, tea, study from 18 March to 7 April 2019, a period vegetables, and so on) and (b) similar products with high intensity of agricultural work at the have different seasonality in different end of the main agricultural season. In-depth districts (that is, planting and harvesting can training was delivered by ILO and World start earlier in some districts and later in Bank officials during the previous week. others). In addition, recent climate change ILO and World Bank officers supervised is making the start and end of the various the three initial days of fieldwork and agricultural seasons less predictable. provided feedback. Across the two surveys, 1,937 household interviews were completed: After an evaluation of several alternatives 964 for the LFS and 973 for the MLSS. This and taking into account also the organization led to data being captured for 2,588 and 47 Measuring Women and Men’s Work | Main Findings from a Joint ILO and World Bank Study in Sri Lanka 2,701 individuals of working age, respectively Between wave 1 and wave 2, some updates (see Table 2.1, which also shows detailed were made to both questionnaires based breakdown by sex, age group and education). on observations of the operation of the questionnaires in the field, feedback from Additional balance tests show that interviewers and analysis of the data. households were well balanced across treatment arms for most individual- The fieldwork for wave 2 took place from level characteristics, with no significant 8  September to 7 October 2019, a period of low differences in the share of males, average intensity of work in agriculture between the age and the share of household members two high agricultural seasons. The households aged 15+ without any schooling (Table 2.2). interviewed in wave 1 were reinterviewed. The However, the share of household members number of household interviews completed aged 15+ who had passed at least grade 10 is for the LFS in wave 2 were 956, and 960 signficantly higher in the MLSS (67.8 percent) for the MLSS; giving a total of 2,604 and than in the LFS (50 percent), which suggests 2,643 respondents of working age, respectively. that – despite the randomization approach – some differences in educational attainment The teams of interviewers from wave 1 across the two samples remained. In terms were retained, with few changes. The week of household characteristics, there are no before the fieldwork, the ILO and World significant differences in household size, Bank officers delivered a five-day in-depth the number of children in the household, face-to-face refresher training course to the the share of households headed by a male, respective teams, which also familiarized the head’s marital status and ethno-religious interviewers with the changes made to the group affiliation, and access to electricity. questionnaires between waves 1 and 2. However, households in the LFS sample have, on average, slightly more bedrooms The use of CAPI allowed data to be transmitted than households in the MLSS sample (2.63 vs. to the DCS each day. Data were processed by 2.54) and are less likely to use an improved the DCS and subsequently shared with the ILO source of drinking water (4.6 percent vs. and the World Bank for analysis. To facilitate a 6.5 percent). Both differences are marginally direct comparison of the figures and estimates statistically significant (e.g. at 10 percent). In from the two samples, it was decided to use the view of the authors, these differences are grossing weights to benchmark the sample unlikely to have substantially impacted the results to a common reference population. analysis presented in this report, given the Poststratification weights were calculated for re-weighting procedures used (as described the two sets of microdata using the distribution below) and the fact that the differences by district, sex and age group obtained as across the two samples are not substantial. averages of the two sample distributions. 48 Annex 2. Methodology of the Pilot Study Table 2.1 Basic characteristics of the two samples in wave 1 and Pearson Chi Square test SEX BY AGE-GROUP LFS MLSS TOTAL MALES 0-14 447 431 878 MALES 15-24 208 267 475 MALES 25-34 164 171 335 MALES 35-44 205 207 412 MALES 45-54 206 219 425 MALES 55-64 204 196 400 MALES 65 + 186 181 367 FEMALES 0-14 417 413 830 FEMALES 15-24 233 254 487 FEMALES 25-34 222 216 438 FEMALES 35-44 280 273 553 FEMALES 45-54 233 252 485 FEMALES 55-64 216 221 437 FEMALES 65 + 231 245 476 TOTAL 3452 3546 6998 Person chi2 test (13) = 9.450 p-value = 0.738 SEX BY RELATIONSHIP TO HEAD LFS MLSS TOTAL MALES - REFERENCE PERSON/HEAD 710 707 1417 MALES - SPOUSE/PARTNER 18 23 41 MALES - SON/DAUGHTER 657 707 1364 MALES - MOTHER/FATHER 23 22 45 MALES - OTHER 212 213 425 FEMALES - REFERENCE PERSON/HEAD 254 266 520 FEMALES - SPOUSE/PARTNER 631 631 1262 FEMALES - SON/DAUGHTER 609 591 1200 FEMALES - MOTHER/FATHER 68 69 137 FEMALES - OTHER 270 317 587 TOTAL 3452 3546 6998 Person chi2 test (9) = 5.529 p-value = 0.786 EDUCATIONAL LEVEL LFS MLSS TOTAL STUDYING IN GRADE 1 65 73 138 PASSED GRADE 1 77 78 155 PASSED GRADE 2 120 123 243 PASSED GRADE 3 107 117 224 PASSED GRADE 4 173 152 325 PASSED GRADE 5 192 173 365 PASSED GRADE 6 137 135 272 PASSED GRADE 7 179 163 342 PASSED GRADE 8 212 222 434 PASSED GRADE 9 146 143 289 49 Measuring Women and Men’s Work | Main Findings from a Joint ILO and World Bank Study in Sri Lanka EDUCATIONAL LEVEL LFS MLSS TOTAL PASSED GRADE 10 732 749 1481 PASSED G.C.E.(O/L) OR EQUIVALENT 351 389 740 PASSED GRADE 12 194 242 436 PASSED G.C.E.(A/L)A OR EQUIVALENT 340 367 707 PASSED GAQ/GSQ/DEGREE 81 70 151 PASSED POST GRAD./DIPLOMA/PHD/SPECIAL EDUC. 32 13 45 NEVER ATTENDED SCHOOOL 75 82 157 UNDER 15 /MISSING 239 255 494 TOTAL 3452 3546 6998 Person chi2 test (17) = 21.182 p-value = 0.218 Note: The category “under 15/missing” combines household members below age 15 and those with missing values. Table 2.2 Additional household and individual sample characteristics by treatment arm LFS MLSS Difference   Standard Standard in means   Mean error Mean error test Individual Characteristics Males 0.469 0.008 0.472 0.008   Age 35.452 0.386 35.453 0.378   Has no schooling (age 15+) 0.021 0.002 0.027 0.003   Passed at least grade 10 (age 15+) 0.5 0.009 0.678 0.009 *** Total number of individuals 3,452 3,546   Household (Hh) Characteristics Household size 3.562 0.048 3.644 0.051   Number of children (0-14) in household 0.887 0.032 0.867 0.031   Male household head 0.737 0.014 0.727 0.014   Hh head single/divorced/widowed 0.192 0.013 0.203 0.013   Hh head belongs to Sinhala ethnic group 0.896 0.01 0.9 0.01   Hh head is Buddhist 0.877 0.011 0.879 0.01   Number of bedrooms in house 2.629 0.032 2.541 0.035 * Hh uses unimproved source of drinking water 0.046 0.007 0.065 0.008 * Hh has an electricity connection 0.944 0.007 0.933 0.008   Total number of households 964 973   Note: *** indicates statistical significance at 1%, ** at 5% and * at 10%. Unprotected source of drinking water includes the following sources: unprotected well, pond, river, canal, stream, lake, unprotected stream and others. 50 Annex 3. Identifying Employment in the LFS and MLSS Questionnaires This annex illustrates – in a simplified way The LFS questionnaire – the sequence of questions used in the two questionnaires to identify persons in In the LFS questionnaire, the only purpose employment in both wave 1 and wave 2. of the employment identification sequence While the flow and wording differ, the overall is to classify whether the respondent is intention of the sequences is the same, employed or not. Thus, once respondents namely, to identify comprehensively all are identified as employed, they will not be employed respondents to the survey. Any asked more questions from the sequence changes between wave 1 and wave 2 are and will continue with other parts of the highlighted in red (see Tables 3.1 and 3.2). questionnaire covering characteristics of their job, time worked, and so on. This differs Changes were not made to the LFS sequence from the approach in the MLSS questionnaire (Table 3.1) as no evident misclassification as outlined below. issues were identified in the wave 1 data. The changes made to the MLSS questionnaire, Table 3.1 illustrates the employment as discussed in the report, are shown identification questions used in the LFS in Table 3.2. In each table, the last two questionnaire for both wave 1 and wave 2. The rows show the total number of employed table shows four main profiles covering the respondents identified in each wave in each main situations encountered in the field. survey to show the relative importance of the different questions. In the interests  Profile 1 includes respondents who of comparability, these data only include reported “work for pay” in the first those respondents who were at work during question; in this case, respondents are the reference week and exclude those who immediately classified as employed, and were temporarily absent from work (about all the other screening questions are 4 percent of all employed). skipped. This accounted for 63 percent of the employed respondents in wave 1 and 58 percent in wave 2. 51 Measuring Women and Men’s Work | Main Findings from a Joint ILO and World Bank Study in Sri Lanka  Profile 2 includes respondents who this route and were identified as employed report “own-account work” in the second based on the criteria on the main intended question (having not reported “work destination of the output produced (18 for pay” in the first place). In a follow up percent in wave 2). question, it is determined that at least some of the work was in businesses As covered in the report, the number of other than “agricultural/fishing”: these employed respondents identified by the LFS respondents are classified as employed questionnaire was consistent between the and skip all the other questions in the two waves: 1,459 in wave 1 and 1,451 in wave sequence. In wave 1, 20 percent of all 2. The proportions of respondents identified employed respondents were covered by as fitting the various profiles varied this profile (22 percent in wave 2). somewhat, possibly reflecting the seasonal differences in activities at the time of the  Profile 3 is similar to profile 2 and includes two waves. respondents who report “helping” with the business or a job of a family member in the third question. This type of question has The MLSS questionnaire been shown to be important in avoiding an undercount of contributing family workers In the MLSS questionnaire, the screening who may not view themselves as working questions are used both to identify persons for pay or doing any kind of business. As in employment and to collect information an with profile 2, once it is confirmed that all the different jobs held by respondents, this involved work in a non-agricultural including the time worked in each of them business, the respondent is identified as during the reference week. They serve a dual employed and skips remaining questions purpose: to identify employed respondents in the sequence. This covered 4 percent of and to generate statistics on the engagement employed respondents in wave 1 and in different working activities (which, by 2 percent in wave 2. comparison, is captured later in the LFS questionnaire). Therefore, the questions in  Profile 4 includes respondents who report the sequence are asked of all respondents work activities that are subsequently of working age. This is common practice in identified as activities in agriculture or labour modules in MLSSs. fishing. Based on the 19th ICLS standards, it is necessary to determine if the output Table 3.2 illustrates the main questions from the work is mainly intended for sale used in the MLSS questionnaire to identify (employment) or for own use (the own-use employment. Based on the analysis of the production of goods). In wave 1, 14 percent results from wave 1 (as described in the of all employed respondents came through report), several changes were made to the 52 Annex 3. Identifying Employment in the LFS and MLSS Questionnaires Table 3.1 Sequence of screening questions used in the LFS to capture employment LFS WAVE 1 and WAVE 2 Questions for identification of persons in PROFILES WAVE 1 PROFILES WAVE 2 emplyment (at work) 1 2 3 4 TOTAL 1 2 3 4 TOTAL ATW_PAY Last week, that is from Monday (DATE) up to (last • • Sunday), did NAME do any work for someone else for pay, even if only for one hour? " ATW_PFT Last week, did NAME run or do any kind of business, farming or other activity to generate any income? READ ONLY IF NEEDED: For example: [making • • • • things for sale, buying and reselling things, provided paid services, growing products, raising animals or catching fish for sale, and [OTHER EXAMPLES RELEVANT IN NATIONAL CONTEXT] ATW_FAM Or, did NAME help with the business, farm or paid • • • • job of a household or family member?" AGF_ANY Last week, that is from Monday (DATE) up to (last • • Sunday), did NAME do any work in farming, rearing animals, [fishing or fish farming]?" AGF_CROP_CHK The work that you mentioned, was it farming of • • crops, vegetables or fruits? for example: rice, tea, rubber, flowers?" AGF_LIV_CHK • • Was it rearing or tending farm animals?" AGF_FISH_CHK Was it fishing, [FISH FARMING] or collecting • • shellfish?" AGF_OTHER_CHK • • • • Was it another type of job, business or activity?" AGF_MKT Thinking about all the (farming products, animals • • or fish) NAME worked on, are they intended…" AGF_MKT_MAIN Thinking about those (farming products, animals or • • fish), is it intended to sell...?" AGF_HIS In general, in the past have these products been • • mainly sold or mainly kept for family consumption?" AGF_HIR Was NAME hired or paid by someone else to do • • this work?" EMPLOYED identified by different profiles 914 292 52 201 1459 835 322 35 259 1451 EMPLOYED % from the different profiles 63% 20% 4% 14% 100% 58% 22% 2% 18% 100% 53 Measuring Women and Men’s Work | Main Findings from a Joint ILO and World Bank Study in Sri Lanka questions for wave 2; these are highlighted All the questions are asked of all in red in the table. Changes were made to the respondents. The questionnaire is thus wording of some questions (that is, S2Q7aa, structurally quite different from the LFS S2Q7ba, S2Q7ca); other questions, similar to questionnaire. Nonetheless, it was still those used in the LFS, were also added at the possible to build respondent profiles that end of the sequence (that is, S2Q8a, S2Q8aa) are broadly comparable with the LFS to verify whether these would “recover” other profiles highlighted above. In both waves, employed respondents not captured by the first four profiles shown in Table 3.2 are previous questions. Despite these additional logically identical to those identified in the recovery questions, the average interview LFS. A fifth profile was added in wave 2 to length for the MLSS labour module did not capture respondents who were identified as increase between waves 1 and 2. This shows employed by the new recovery questions that that the recovery questions, which were had been added to the end of the sequence. only asked of a small subset of household members who responded negatively to some Given the number of changes made to the of the previous questions, do not need to questionnaire, it is useful to look at the signficantly increase the interview burden of number of respondents identified under the survey. 11 the various profiles (the last two rows of the table). The overall number of employed respondents identified increased, as discussed in detail in the report (1,303 in wave 11 In wave 1, the average interview duration for the MLSS labour module (per eligible household member) was 17.7 minutes, compared 1 to 1,372 in wave 2) meaning that the gap to 16.5 minutes in wave 2. The slight decline in the length of the interview, despite the fact that additional questions were added, relative to the LFS was narrower in wave 2. probably reflects that the enumerators were more familiar with the survey instrument and CAPI application in wave 2. This suggests that the changes introduced 54 Annex 3. Identifying Employment in the LFS and MLSS Questionnaires were successful, even if they did not eliminate S2Q5a, S2Q6a, S2Q7aa, S2Q7ba, S2Q7ca). Two the gap entirely. The increases came in recovery questions on helping on a family profiles 2 to 5 of the second wave, which farm in terms of crop farming (S2Q7aaa) and supports the general conclusion of the report livestock production (S2Q7baa) identified that these questions were more effective an additional 4 and 2 percent of employed in identifying respondents engaged in own- women, while the final two recovery account farming for the market, those helping questions (S2Q8a and S2Q8aa combined) in family businesses or farms and those with identified only slightly less than 3 percent of small or casual jobs as outlined in the report. employed women. Thus, without using the recovery questions, 9 percent of employed The importance of the recovery questions for women would not have been captured capturing women’s employment in the MLSS as employed. For men, all four recovery instrument is further illustrated in Figure 3.1, questions combined identified only slightly which shows that 98 percent of employed more than 2 percent of total employment. men, but only 91 percent of employed This pattern of greater importance of women were identified by the core question recovery questions for women than men is sequence used to identify the employed consistent with earlier literature referenced population (these are the questions that in this report and findings from previous pilot were already included in wave 1, e.g. S2Q3a, studies by the ILO. Figure 3.1 Share of employed women and men identified by recovery questions, MLSS wave 2 0.7% 1.8% 4.4% FEMALE 91.1% 2.1% MALE 97.9% 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% Core questions to identify employed population S2Q7aaa: … did [NAME] help on a family farm to prepare or maintain the land or to plant, grow or harvest any crops, vegetables, or fruits …? S2Q7baa: … did [NAME] help on a family farm to raise or tend animals, such as heep, goats, pigs, chicken or cattle …? S2Q8a: … did [NAME] run or do any kind of business, farming, or other activity to generate an income …? S2Q8aa: … did [NAME] help with the business, farm or paid jobs of a household or family member …? 55 Measuring Women and Men’s Work | Main Findings from a Joint ILO and World Bank Study in Sri Lanka Table 3.2 Sequence of screening questions used in the MLSS to capture employment MLSS WAVE 1 Questions for identification of persons in emplyment (at work) PROFILES WAVE 1 1 2 3 4 TOTAL S2Q3a 'Last week, that is from Monday [DATE] up to Sunday [DATE], did NAME do any work for someone else for pay, even if only for one • hour? INCLUDES PAID APPRENTICESHIPS AND PAID INTERNSHIPS.” S2Q5a 'Last week, did [NAME] work in a non-farm family business that • [NAME] operates, even if only for one hour?" S2Q6a 'Last week, did [NAME] help in a non-farm family business that is • operated by another family member, even if only for one hour?" S2Q7aa 'Last week, did [NAME] work on own account crop farming, even • if only for one hour?" S2Q7ba 'Last week, did [NAME] work on raising animals, even if only for one hour? EXAMPLES: COWS, CHICKEN, GOATS” S2Q7ca 'Last week, did [NAME] work on own account fishing or • collecting shellfish, even if only for one hour?" S2Q7e Thinking about all the [farming products/animals/fish] [NAME] • worked on, are they intended... S2Q7f Thinking about those [products/animals/fish], is it intended to • sell...?" S2Q7g In general, in the past have these products been mainly sold or • mainly kept for family consumption?" EMPLOYED identified by different profiles 868 223 58 154 1303 EMPLOYED % from the different profiles 67% 17% 4% 12% 100% 56 Annex 3. Identifying Employment in the LFS and MLSS Questionnaires MLSS WAVE 2 Questions for identification of persons in emplyment (at work) PROFILES WAVE 2 1 2 3 4 5 TOTAL S2Q3a 'Last week, that is from Monday [DATE] up to Sunday [DATE], did [NAME] do any work for someone else for pay, even if only • for one hour? INCLUDES PAID APPRENTICESHIPS AND PAID INTERNSHIPS.” S2Q5a 'Last week, did [NAME] work in a non-farm family business that • [NAME] operates, even if only for one hour?" S2Q6a 'Last week, did [NAME] help in a non-farm family business that • is operated by another family member, even if only for one hour?" S2Q7aa 'Last week, did [NAME] do any work on a family farm [or in a kitchen garden] to prepare or maintain the land, or to plant, • grow or harvest any crops, vegetable or fruits, even if only for one hour? S2Q7aaa 'Last week, did [NAME] help on a family farm [or in a kitchen garden] preparing or maintaining the land, planting, growing • or harvesting any crops, vegetable, fruits or other agricultural products, even if only for one hour? S2Q7ba 'Last week, did [NAME] spend any time on a family farm raising • or tending animals such as sheep, goats, pigs, chickens or cattle, even if only for one hour? S2Q7baa 'Last week, did [NAME] spend any time helping on a family farm • raising or tending animals such as sheep, goats, pigs, chickens or cattle, even if only for one hour? S2Q7ca 'Last week, did [NAME] spend time in family fishing, pond • fishing or collecting shellfish, even if only for one hour? S2Q7e Thinking about all the family [farming products/animals/fish] • [NAME] worked on, are they intended…" S2Q7f Thinking about those [products/animals/fish], is it intended to • sell...?" S2Q7g In general, in the past have these products been mainly sold or • mainly kept for family consumption?" S2Q8a Last week, did [NAME] run or do any kind of business, farming or other activity to generate income? • READ ONLY IF NEEDED: For example: making things for sale, buying or reselling things, provided paid services, growing products, raising animals or catching fish for sale. S2Q8aa Or, did [NAME] help with the business, farm or paid job of a • household or family member? 57 Measuring Women and Men’s Work | Main Findings from a Joint ILO and World Bank Study in Sri Lanka MLSS WAVE 2 Questions for identification of persons in emplyment (at work) PROFILES WAVE 2 1 2 3 4 5 TOTAL S2Q8ad Was [NAME]'s work in family farming, tending/rearing animals, • or family fishing? S2Q8ae Thinking about all the [farming products/animals/fish] [NAME] • worked on, are they intended... S2Q8af Thinking about those [products/animals/fish], is it intended to • sell...? S2Q8ag In general, in the past have these products been mainly sold or • mainly kept for family consumption? EMPLOYED identified by different profiles 812 241 75 220 24 1372 EMPLOYED % from the different profiles 59% 18% 5% 16% 2% 100% Note: The recovery questions that were added in wave 2 were not administered to all household members but only to those who responded negatively to some of the previous questions: S2Q7aaa (S2Q7baa) was activated only for those household members who responded ‘no’ to S2Q7aa (S2Q7ba); S2Q8a was asked of those household members who responded ‘no’ to questions S2Q3a, S2Q5a, S2Q6a, S2Q7aa, S2Q7aaa, S2Q7ba, S2Q7baa, and S2Q7ca; S2Q8aa was asked of those who responded ‘no’ to S2Q8a. 58 CONTACTS Living Standards Measurement Study www.worlbank.org/lsms lsms@worldbank.org International Labor Organization www.ilo.org STATITSTICS@ilo.org