Declining female labour force participation: A review of literature

Table 1 outlines labour force participation rates (LFPRs) – in terms of the usual status including principal and subsidiary forms of work – for rural and urban sectors from the 43rd Round of the Employment and Unemployment Surveys of the NSSO (covering the period 1993–94) to the PLFS of 2018–19 for women aged 15 and above. Barring a rise in the mid-2000s, women’s LFPRs have been declining since the early 1990s in both rural and urban areas. While LFPRs are higher in rural areas than urban (Reference Klasen and PietersKlasen and Pieters, 2015), the fall has been much more significant in the former, and largely seen amongst agricultural workers (Reference RustagiRustagi, 2013). While rural LFPRs for women have been continuously falling since 2004–05, urban women’s LFPRs has remained roughly constant since 2011–12, at around the same level as that seen in 1999–2000.

Table 1. Women’s LFPRs (age 15 and above).

Source: Various NSS Rounds. The figures show the percentage of women aged 15 and above participating in the labour force.

The simultaneous occurrence of a fall in participation rates alongside significantly high rates of economic growth – particularly in the period between 2004–05 and 2011–12 – posed a significant paradox. This led to the hypothesis of an ‘income effect’, whereby increases in household incomes would lead to women withdrawing from the labour force due to income security being assured within the household (Reference Mehrotra and SinhaMehrotra and Sinha, 2017). This view implicitly sees women’s labour as a reserve, being called upon to address income shortfalls. At a macro level, this view assumes the existence of a ‘U-shaped’ curve between economic development and women’s labour force participation, where growth and structural change in a developing economy leads to an initial withdrawal of women’s labour and then a subsequent increase (Reference Goldin and SchultzGoldin, 1995). Reference Lahoti and SwaminathanLahoti and Swaminathan (2013), however, find no real link between economic growth and labour force participation at the state-wise level, while Reference Gaddis and KlasenGaddis and Klasen (2014) cast doubt on the very existence of a U-shaped curve itself.

Given that the most significant reductions in LFPRs have been seen amongst rural women, much of the literature focuses largely on rural women and conditions of employment in rural India (Reference Sanghi, Srija and VijaySanghi et al., 2015). A multitude of factors are responsible for this reduction (Reference Kapsos, Bourmpoula and SilbermanKapsos et al., 2014), such as increasing educational attainment (Reference Afridi, Dinkleman and MahajanAfridi et al., 2018; Reference BhallaBhalla, 2019; Reference Das and DesaiDas and Desai, 2003; Reference GhaiGhai, 2018) and increasing household incomes and rising rural real wages (Reference Mehrotra and ParidaMehrotra and Parida, 2017; Reference Neff, Sen and KlingNeff et al., 2012). Reference JayachandranJayachandran (2021) argues that cultural norms play a greater role in influencing women’s LFPRs amongst economies at similar levels of development. In this vein, Reference Dhanaraj and MahambareDhanaraj and Mahambare (2019) examine the Indian institution of a joint family; they find that married women residing in a joint family are less likely to participate in non-farm employment. Reference Deshpande and KabeerDeshpande and Kabeer (2021) complicate the question of norms, finding that religion and other markers of norms – such as veiling – are not as important as determinants of labour force participation as compared to the norm necessitating that the burden of domestic work be shouldered by women alone (see also Reference Afridi, Bishnu and MahajanAfridi et al., 2022b). Another strand of work focuses on measurement errors and the construction of these surveys itself, and how the nature of the survey and our limited conceptualisation of women’s work affects our measurements in rural and urban India (Reference Desai and JoshiDesai and Joshi, 2019; Reference HirwayHirway, 2012; Reference Kapur, Vaishnav and VerleyKapur et al., 2021; Reference SircarSircar, 2010).

A growing literature has begun to look at demand-side factors such as technical change, employment generation and the demand for women’s labour in rural India. Reference Afridi, Bishnu and MahajanAfridi et al. (2022a) examine how mechanisation in agriculture reduced the need for women’s work in labour-intensive operations like weeding. Reductions in women’s agricultural employment did not lead to absorption in the non-farm sector owing to the inability of the rural non-farm sector to generate enough livelihood opportunities (Reference Chatterjee, Murgai and RamaChatterjee et al., 2015). The unavailability of suitable alternate livelihood opportunities in rural India affects both men and women; men, however, have the option of migrating to cities, while opportunities for women are limited (Reference Choithani, van Duijne and NijmanChoithani et al., 2021). Where alternate employment does exist, the ability for women to access such employment safely plays a major role in whether they take up these opportunities. Reference Lei, Desai and VannemanLei et al. (2019) find a positive effect of road transport infrastructure on non-farm rural employment, with a greater impact for women than men, while Reference Chatterjee and SircarChatterjee and Sircar (2021) who undertook surveys across four urban clusters of North India found a reduced willingness to engage in work for longer hours of commute.

The focus on demand-side factors is an important one, expanding the field beyond a focus on demographic and supply-side characteristics like education and household income. The use of longitudinal data can better illuminate these questions by precisely estimating important measures of labour demand such as job-finding rates. Reference BremanBreman’s (2012) work on ‘footloose labour’ studied the circulation of labourers amongst informal sectors of the economy through fieldwork and an anthropological look at the lives of migrant labour; large-scale statistical studies remained sparse. Some of the important studies that examine transitions utilise two successive rounds of the Indian Human Development Survey (IHDS), carried out in 2004–05 and 2011–12. Reference KesarKesar (2020) studied informality in the context of the Indian economy, examining transitions between informal and formal sources of work for households, while Reference Raj, Schotte and SenRaj et al. (2020) looked at the ability of individuals to transition from informal to formal jobs. Reference Neog and SahooNeog and Sahoo (2020) use the methodology of labour market transitions to examine intergenerational occupational mobility. Reference Sarkar, Sahoo and KlasenSarkar et al. (2019) examine women’s labour force transitions over this period, finding existence of a significant income effect. Though they provide valuable insights into the dynamics of the labour market, the large time period involved – 7 years separate the two successive rounds of the IHDS – makes it difficult to understand the nature of short-term transitions. Furthermore, these studies do not disaggregate movements within the labour force to understand the nature of transitions through employment and unemployment.

Reference BhattacharyaBhattacharya (2021) utilises PLFS data to study gross and net flows of labour. In contrast, we examine conditional transition probabilities, which allows for a deeper examination of labour dynamics. Reference Deshpande and SinghDeshpande and Singh (2021) cover similar grounds using data from the Consumer Pyramid Household Survey (CPHS) of the Centre for Monitoring Indian Economy, analysing the factors that influence entries and exits into the labour force. They find that the probability of labour force entry (exit) is negatively (positively) related to the increase in income of other household members and positively (negatively) related to the presence of unemployed male members. They further stress the importance of focusing on demand-side explanations of falling women’s LFPRs, pointing to the inability of demographic characteristics to explain much of the drop in women’s LFPRs over the study period (from 2016 to 2019). Despite some similarities, there exist significant differences regarding our analysis; the CPHS suffers from a relative under-estimation of women’s labour force participation owing to its restrictive definition of employment (Reference Abraham and ShrivastavaAbraham and Shrivastava, 2022) and an over-estimation of richer households (Reference Dreze and SomanchiDreze and Somanchi, 2021). Secondly, we distinguish between categories within the labour force – that is, employment and unemployment – so as to better examine demand-side constraints by calculating job-finding rates and examining the relative differences across gender and education categories.

The study of labour market transitions

An individual exists in any one of three states in the labour market: either employed (E), unemployed (U) or being out of the labour force (OLF) (non-participation). Assume an individual in state A_t at time t, transitioning to state B_t+1 in period (t + 1). The transition probability p(AB) is measured by dividing the number of people who have made this transition by the total number of people in state A_t. With nine such transitions possible, nine transition probabilities can be calculated (Reference TheeuwesTheeuwes, 1986).

(1) $p\left(\mathrm{AB}\right)=\frac{A_t{B}_{t+1}}{A_t}$

All urban individuals in the PLFS sample are interviewed once every quarter for four quarters, thus allowing for the construction of a panel that tracks changing labour force status four times – one every quarter – over a year. Transition matrices – which present the nine possible transition probabilities – are estimated for a pooled sample over the periods 2017–18 and 2018–19, covering the transitions made between the first quarter – covering the months July to September – and the fourth quarter – the months April to June of the subsequent year – for all individuals aged 15–65. The method of panel construction is outlined in the Appendix. The individual panels are largely similar in construction each consisting of roughly 31,800 individuals (as shown in Table 2) for a total sample size of 63,633 individuals. The employment status of individuals is determined according to their Current Weekly Status (CWS), which classifies status based on an individual’s activities over the 7 days prior to the date on which the survey was conducted.^{Footnote 1} In this paper, we make no disaggregation regarding an individual’s nature of employment (whether self-employed, regular wage or casual), nor between formal, informal or precarious forms of work.

Table 2. Description of the panel.

Source: PLFS panel for 2017–18 and 2018–19 as constructed by the authors.

The methodology we follow differs from the literature in certain respects. We estimate transitions between the first and fourth quarters; this differs from much of the literature which estimates short-term transitions, either month-to-month or quarter-to-quarter (e.g. Reference LeevesLeeves, 1997). Secondly, our conditional transition matrices are the probabilities of transitions conditional on certain demographic characteristics, differing from work that attempts an estimation of transition probabilities conditional on the labour market status in a previous month (Reference GomesGomes, 2012; Reference Hall and KudlyakHall and Kudlyak, 2020; Reference Krueger, Cramer and ChoKrueger et al., 2014; Reference Kudlyak and LangeKudlyak and Lange, 2018). Our conditional transition matrices are predicted probabilities of transitions estimated through the calculation of marginal effects at sample values from the following logistic regressions:

(2) $\log \left[\mathrm{Tr}\left(\mathrm{AB}\right)\right]={\beta }_o+{\sum }_{i=1}^n{\beta }_i.X_i+u_i$

where log[Tr(AB)] is the log of the odds of transitioning from state A to B. The individual is coded 1 if she makes the transition, and 0 otherwise.

The demographic characteristics chosen are age, gender, caste – with the ‘Others’ or General category as the base (a proxy for the privileged Upper Castes), and Scheduled Tribes (ST), Scheduled Castes (SC) and the Other Backward Classes (OBC) as the relevant categories – and education – workers without schooling form the base, with the other categories representing workers with some schooling and those with graduate degrees and higher. We construct three categories of workers – the young (aged 15–25), prime-age (26–50) and older workers (51 and above) – to evaluate younger workers’ outcomes vis-à-vis their older counterparts. We do not include such variables that may influence transitions for women alone, such as the number of children within the household nor changes in marital status, a methodology more appropriate for a causal analysis. Our estimates are not estimates of the factors influencing labour supply but are an estimation of the relative strength of different labour flows for specified demographic categories, so as to provide directions and raise questions for future research.

A total of nine logistic regressions are run, one for each relevant transition. The sample for each regression consists only of those individuals who are in state A in the initial time period (first quarter of the year); the results of the logistic regressions are displayed in Appendix 1. Predicted probabilities for each demographic category are then calculated through the estimation of marginal effects, and the relevant conditional transition matrices are presented, along with confidence intervals of each estimate.

Our paper suffers from three main limitations. The use of the CWS to define quarterly labour force status implies that individuals’ labour force status for the entire quarter is defined according to their activities in the given reference week within that quarter. They might undertake several transitions within the quarter outside the reference week which cannot be taken account of. Secondly, the PLFS records individuals over a relatively short period – once every quarter for four quarters – unlike the CPHS, which records individuals monthly over a couple of years. Our panel is thus restricted to analysing individuals’ labour market behaviour over a period of 1 year alone. And finally, since such a panel has been constructed in the NSSO for the urban sector alone, we cannot comment on the dynamics of the rural sector, a sector which has seen the largest falls in women’s labour force participation.

Descriptive analysis

We begin by outlining broad trends as shown in Table 3. A high proportion of individuals – nearly 89% – remain in the same status – either employed, unemployed or outside the labour force – in each quarter; roughly 11% of the total sample experience at least one transition within the year. Nearly 48% of individuals spend all four quarters OLF, while 44% spend all four quarters within it. About 38.4% of the total sample remain employed throughout all four quarters, while 2.28% remain unemployed.

Table 3. Broad movements (as % of entire sample).

Source: From panel constructed by authors.

Across both genders, the proportions of those who retain the same status in all four quarters is roughly the same (between 88% and 89%). The difference lies in the rates of labour force participation. While only 21% of men remain outside the labour force in all periods, three-quarters of women do not register their presence in the labour force for even one quarter. While 63.75% of men remain employed in all four quarters, only 13% of women remain so, testifying to the significantly low levels of labour force engagement of urban women in India.

Of those who do make transitions, the share of those who make multiple transitions, either through different statuses – that is, from employment to unemployment and then non-participation – or through the same statuses but at different periods – that is, from employment to unemployment and back again to employment – is relatively low, at around 4.65% of the overall population and 40.5% of those who make at least one transition. The share of men making more than one transition is higher than that of women.

Table 4 presents gross labour market flows between the first and fourth quarter disaggregated by gender. As mentioned above, a greater proportion of men tend to remain employed throughout the year; however, the share of men who remain unemployed in both periods is significantly higher than that of women.

Table 4. Gross flows (as % of entire sample).

Source: From panel constructed by authors.

The proportions of those moving from unemployment to employment is relatively low at only 1.14%, indicating low rates of employment generation; movement from employment to unemployment is higher at 1.15%. The difference in terms of gender is stark, with nearly 2% of men moving from unemployment to employment as compared to only 0.3% of women. A high proportion of women, in contrast, move from employment to non-participation. The multiple difficulties faced by women is displayed in the tables above; a reduced propensity to participate in the labour force, lower movements into employment and high movements out of employment to non-participation.

Transition probabilities

Table 5 presents the unconditional transition matrices at the aggregate level. Each cell represents the transition probability (as outlined in equation (1)) from employment (E), unemployment (U) or non-participation (Out of the Labour Force or OLF) in the initial quarter to any one of these three possible states in the fourth quarter; each row sums up to 100. These tables confirm the findings outlined when studying gross flows: both those employed and those OLF in the first quarter largely remain in the same state by the fourth quarter.

Table 5. Unconditional transition matrix: Aggregate.

The table above outlines transition probabilities calculated as shown in equation (1). Figures in brackets are sample sizes for that transition.

The consideration of gross flows alone wouldn’t provide an adequate understanding of transitions if there existed significant differences in LFPRs.^{Footnote 2} Consider, for example, the movements from employment and unemployment to non-participation. Gross flows from employment to non-participation is 2.17% while from unemployment, 0.91%. This would imply that the problem of moving from employment to non-participation is more acute than movements from unemployment. But this relative ranking is due to the share of employed in the population being much larger than that of the unemployed. When considering transition probabilities, we see that 5% of the employed move to non-participation as compared to 18.6% of the unemployed.

Transition probabilities provide insights into the dynamics of unemployment, a question relatively unaddressed in the field of Indian labour economics. Around 58.16% of those unemployed at the beginning of the year are still unemployed by the fourth quarter, indicating a significant problem of long-term unemployment. However, of those who do exit unemployment, the share of those who actually find employment is low; only 23.25% of the unemployed find work at the end of the year, while 18.6% move to non-participation. The problem of unemployment is heightened due to low employment generation; the problem would be even worse were it not for the movement to non-participation.

Reference Donovan, Lu and SchoellmanDonovan et al. (2020) construct and analyse cross-country measures of labour market flows, ranging from relatively low-income countries like Palestine – with a gross domestic product (GDP) per capita in the range of 2800–4600 in purchasing power parity terms – to South Africa – GDP per capita 11,800–12,400 – to richer countries like Denmark, the United Kingdom and the United States (For comparison, India’s GDP per capita lies in the range 6500–6900). They find an inverse relation between flows and GDP, with job-finding and job-loss rates higher in poorer economies. However, India’s quarterly job-finding rates are much lower than even richer countries in the sample. Reference Ward-Warmedinger and MacchiarelliWard-Warmedinger and Macchiarelli (2014) report an average annual job-finding rate (transition from unemployment to employment) of Nordic economies – for those aged 16–64 – of 39.7% (over the period 2004–2008), higher than that of India at 23.25%, even though the Nordic economies are all significantly richer. A developing economy like Vietnam also exhibits higher job-finding rates, with the average quarterly measure (across the period 2011–19) of the transition from unemployment to employment calculated at 39.9% (Reference Samaniego and ViegelahnSamaniego and Viegelahn, 2021).

The above table presents unconditional transitions and does not control for the impact of different demographical variables – such as age, caste and educational status – on the probability of transitions. We calculate conditional transition probabilities for each variable through the methodology discussed above. The conditional transition probability matrix for gender is displayed in Table 6. The transition dynamics for men are similar to that seen at the all-India level, with significant retention of employment, low job-loss rates (the transition probability from employment to unemployment) and relatively low job-finding rates (the transition probability from unemployment to employment). While 27% of unemployed men find some form of employment over the year, only 12.45% of unemployed women are able to secure employment. In comparison, the average annual job-finding rate for women in Nordic economies as estimated by Reference Ward-Warmedinger and MacchiarelliWard-Warmedinger and Macchiarelli (2014) is 39.47%. While the period of comparison might differ – their estimates are averages for 2004–2008 – it does uncover a significant problem. One would expect a relatively lower-income country like India to have higher job-finding rates, given the presence of a large informal economy. The fact that it does not outlines the onerous burdens faced by Indian women.

Table 6. Conditional transitions matrix: Gender.

Conditional transition probabilities derived from marginal effects calculated from regressions using equation (2).

In spite of low job-finding rates, the problem of long-term unemployment is higher for men than women, with 61.25% of unemployed men remaining unemployed in the fourth quarter compared to 50.5% of women. This is due to the significant movement to non-participation seen in the case of women; 36% of unemployed women and 17.8% of employed women move to non-participation by the fourth quarter. It is unclear whether this is due to women being fired from jobs at a faster rate and not spending as much time in unemployment as men, or because women voluntarily leave employment. Moreover, it is unclear whether transitions from unemployment to non-participation are because of low job-finding rates – women, discouraged from not finding suitable employment, cease the process of job search – or determine low job-finding rates, as women face social constraints and norms that cause them to cease job search early and hence do not take up what employment opportunities are available. These are important questions for future research; the main contribution of this paper is to highlight future avenues of research in what is already a well-studied phenomenon.

What is surprising is that even in an environment of low labour demand for women’s work, there exists a significant flow from employment to non-participation. As outlined above, a large literature has emerged pointing to several factors – childbirth, household incomes etc. – that influence women’s exits from the labour force. The fact that exits can occur from employment even when employment opportunities are so hard to come by is indicative of the significant pressures faced by Indian women.

We next provide estimates on the influence of age (Table 7) and education (Table 8) on transitions. Workers aged 15–25 faced higher rates of job-loss – 7.25% compared to prime-age workers (2.01%) and older workers (1.39%) – and lower job-finding rates – 17.13% compared to 29.25% for prime-age workers and 38.44% for older workers. Nearly 63.5% of young unemployed workers remained unemployed over four quarters, compared to 27.75% of older unemployed workers.

Table 7. Conditional transition matrix: Age group.

See Table 5.

Table 8. Conditional transition matrix: Education.

See Table 5.

While the possession of a graduate degree does provide some protection against the risk of losing employment – job-loss rates were 2.14% for graduates as compared to 2.66% for workers with some level of schooling and 3.94% for illiterate workers – it confined its holders to longer spells in unemployment. The job-finding rate was only 16% for graduates as compared to 28% for workers with schooling and nearly 33% for illiterate workers, with the result that 67.7% of unemployed graduates remained unemployed over four quarters, compared to 37.17% of illiterate workers. The high rates of job-finding for illiterate workers could be due to working in the informal economy, where employment opportunities are (relatively) greater, but are of a precarious and ill-paid nature.

Gender, education and transitions

The above tables highlight the specific vulnerabilities faced by women, the young and the highly educated in the labour market. Women face significant pressures not only at the point of entry into the labour market, but also in the direction of exits. The young and graduates face different kinds of pressures – the inability to find employment and the threat of long-term unemployment. But do these vulnerabilities remain the same across genders? In the following tables, we calculate transition probabilities of gender and its intersection with age and education through the calculation of marginal effects for interacted variables.

Across both genders, the disparities seen with regards to age and education still hold, both the young and graduates face lower rates of job-finding and longer spells of unemployment in comparison to older and less educated workers, respectively. The disparity across genders within the same age and education categories, however, is stark. While 20.11% of young unemployed males find employment, only 8.5% of young female job-seekers are successful (Table 9). This disparity is seen for all age-groups, where the job-finding rate for prime-age and older women workers are nearly half that of prime-age and older male workers, respectively.

Table 9. Age and gender.

See Table 5.

As mentioned above, the low rates of women stuck in long-term unemployment in the face of low job-finding rates is due to exits to non-participation, exits which are worryingly high across all age groups. While 37.49% of young unemployed women leave the labour force by the fourth quarter, this proportion rises to 50.29% in the case of older women, rates that are much higher than for men. Prime-age unemployed women and men do face lower exit rates as compared to the young and the old, but it is still much higher for women (33.64%) as compared to men (10.4%).

We noted above that the rates of employment retention – the E-E transition – were lower for younger workers than for prime-age and older workers. Table 8 indicates just how skewed this in terms of gender. Only 66.74% of young women employed in the first quarter remain employed by the end of the fourth, in contrast to nearly 90% of young men. This is not because young women are moving into unemployment by the last quarter of the year – the E-U rate for young women is 5.5% compared to 7.68% for young men – but because they exit the labour force at higher rates. A staggering 23.14% of young women who are employed in the first quarter move to non-participation by the fourth, compared to only 2.74% of young men. This discrepancy is seen across all age-groups, with the E-E transition probabilities for prime-age and older women significantly less than that of men.

Similar differences can be seen with regards to education (Table 10). Across both genders, job-finding rates reduce as education levels increase; graduates find it harder to secure employment compared to illiterates. But within any given educational category, job-finding rates for women are much less than those of men. The job-finding rate for graduate women falls to only 8%, indicating a significant problem of employment generation for educated women in the urban Indian economy.

Table 10. Education and gender.

See Table 5.

The phenomenon of low job-finding rates and high labour force exists for women are seen across all educational categories. While only 9.5% of unemployed graduate men exit to non-participation by the end of four quarters, the proportion rises to 31.55% for women, resulting in the fact that roughly 71% of unemployed male graduates remain so after four quarters compared to roughly 60% of unemployed women graduates. At every education level, more men remain in long-term unemployment, but more unemployed women exit the labour force. Employment generation is failing urban women, but their large-scale exits from the labour force imply an inability to properly estimate the contours of the problem.