The compromised impact of ILO standards on laws and practices

There is substantial literature on the impact of the ILO’s global standard-setting, but this has yielded contrasting results depending on whether the focus is on the ILO’s impact on legislation or outcomes. Some studies find that states’ exposure to international labor standards through membership in the ILO, states’ participation in the International Labour Conference, and ratification of ILO conventions, increases the probability of the introduction of social policies regardless of the level of economic development or types of political regimes (Böger and Leisering, Reference Böger and Leisering2020; Kim, Reference Kim2010; Schmitt et al., Reference Schmitt, Lierse, Obinger and Seelkopf2015; Usui, Reference Usui, Janoski and Hicks1994). In a similar vein, shortly after the ILO specifies global blueprints of certain labor or social policies in the form of international standards, states have been shown to adopt the kind of policies identified in the conventions (Berkovitch, Reference Berkovitch1999). However, others fail to find evidence that states increase social expenditure or improve labor and social rights in practice after the ratification of ILO conventions (Peksen and Blanton, Reference Peksen and Blanton2017). This is especially the case for LMICs (Boockmann, Reference Boockmann2010; Strang and Chang, Reference Strang and Chang1993). For example, after finding that the ratification of ILO conventions did not contribute to the expansion of social expenditure in LMICs, Strang and Chang (Reference Strang and Chang1993) argue that their empirical results contradict the conclusion of numerous articles in the International Labour Review that ILO conventions substantially influence the adoption of national policies in ratifying countries.

One possible explanation for the discrepancy is that the impact of ILO standards is compromised at different stages of the policy process. A promising vein of research points out that workers in informal and less-recognized sectors are left out of the application of ILO standards due to the institutional features of welfare systems in LMICs and the limitation of the ILO. Reflecting their general tendency to provide social security only to privileged minorities (Haggard and Kaufman, Reference Haggard and Kaufman2008; Holland, Reference Holland2017), LMICs tend to ‘translate’ or ‘vernacularize’ international standards in such a way as to narrow the scope of the application by not acknowledging workers in the informal sector, such as peasants and the self-employed urban poor, as “workers”, or by using the loophole in international standards that allows states to apply social security systems for all only ‘in accordance with national circumstances’ (Boris, Reference Boris2019; Deacon, Reference Deacon2013; Seekings, Reference Seekings2019; Seekings, Reference Seekings2020). The ILO was also negligent in placing the informal sector under the social security system, as it did not realize the significance of the informal sector in LMICs until the 1990s. Although the 1944 Declaration of Philadelphia included a provision for social assistance, at the time this was seen as a limited measure to cover children, invalids, aged persons, widows, and the self-employed (Devereux, Reference Devereux2017), rather than as a means to include those in the large informal sector in LMICs. The first time the ILO coined the term ‘informal sector’ was at the ILO World Employment Mission to Kenya in 1972, before which the informal sector in LMICs was treated within the ILO as a ‘temporary problem’ (Bangasser, Reference Bangasser2000). Since the ILO expected that the informal sector would disappear along with economic development as it had in the high-income countries (van Ginneken, Reference van Ginneken1999: 3), it prioritized social insurance over a social assistance. Social assistance has rarely been mentioned in its technical assistance activities (Seekings, Reference Seekings2008) and has never been adopted as a resolution by the International Labour Conference from 1919 until 2019.Footnote ¹ It was only in the 1990s that the ILO began to adopt conventions that explicitly targeted informal sector workers such as the Home Work Convention of 1996 (C177), reflecting a change in the ILO’s strategy from eliminating informal employment to broadening the coverage of the social protection system to the informal sector (Boris, Reference Boris2019: 157–165).

Even when LMICs extended coverage to informal sectors, they often failed to enforce labor and social policies due to high regulation costs and low state capacity (Basu and Van, Reference Basu and Van1998; Boockmann, Reference Boockmann2010; Bosch et al., Reference Bosch, Melguizo and Pagés2013; Fombad, Reference Fombad2013; Seidman, Reference Seidman2007). For instance, in Latin America, employers in micro-enterprises are not motivated to register their employees in social security systems because states tend to regulate only middle- or large-sized firms in the formal sector due to the large cost of monitoring informal and illegal labor markets (Bosch et al., Reference Bosch, Melguizo and Pagés2013: 74–75). The ILO has not been successful in amending the implementation constraints in LMICs. Its monitoring involves comparing the text of conventions and national laws rather than using statistics on policy outcomes, and, therefore, it is less likely to notice non-compliance in the stage of policy implementation (Boockmann, Reference Boockmann2010: 682). Even if implementation failure is detected, ILO standards remain for the most part ‘soft’ laws without sanctioning power. The most powerful instrument at its disposal is the complaint procedure which enables trade unions to file complaints about persistent violations directly to the ILO. If states refuse to accept the Commission of Inquiry’s proposed steps, the ILO could take cases to the International Court of Justice. However, this procedure has rarely been employed: the court had seen only 26 cases by 2007 (Hartlapp, Reference Hartlapp2007), and by 2020, only one complaint related to the Maternity Protection Conventions had been submitted.

The selective application of international labor standards in the formal sector means that most labor forces in LMICs would not benefit from the spread of the standards. Furthermore, it may have perverse incentives, forcing workers into the informal sector to avoid burden of paying contributions as well as making the poor subsidize social insurance benefits for formal sector workers through taxes and higher prices (Levy, Reference Levy2008; Seekings, Reference Seekings2008). Yet, no study has systematically assessed the translation and implementation of ILO standards at the national level. This paper aims to fill the gap using paid maternity leave policy as an example. Historically, women used the ILO to pressure states into improving the maternity protection policies, compensating for their lacking political power within domestic political institutions (Berkovitch, Reference Berkovitch1999; Boris, Reference Boris2019), although studies that focus on the recent expansion of leave policies put more emphasis on women’s political representation at the national level (Htun and Weldon, Reference Htun and Weldon2018; Kittilson, Reference Kittilson2008; Morgan, Reference Morgan2013). At the same time, women in LMICs constitute a large proportion of those engaged in precarious labor, such as domestic workers, home-based workers, and family workers: 92.1 percent of employed women in low-income countries and 84.5 percent of employed women in lower-middle-income countries are engaged in informal employment (International Labour Organization, 2018). Since women’s economic rights are treated as secondary to their family responsibilities, women manage this tension by engaging in self-employment and other forms of informal work. If Deacon’s thesis is correct, the impact of the ILO Maternity Protection Conventions would not reach most women workers in LMICs due to the way that states translate and implement the ILO standards. Building on these insights, I suggest the following propositions:

1. 1. Although the ILO conventions triggered legislative change in maternity leave policies, LMICs tend to ignore international standards that concern the broadening of social security coverage to include the informal sector.

2. 2. There is a substantial discrepancy between law and practice in LMICs especially regarding those engaged in the informal sector.

The ILO Maternity Protection Conventions

The ILO had been the only standard-setter concerning maternity protection for women workers until the European Union implemented its own standards in 1992 (Pregnant Worker Directive 92/85/EEC). Other international organizations had long focused their attention on women as mothers rather than as income earners (Boris, Reference Boris2019: 92). For instance, the Food and Agriculture Organization emphasized mother and infant health while the World Bank did not refer to women in its economic productivity plans for LMICs until the 1970s. In contrast, the ILO contributed to the extension of the statutory support for pregnant women workers in both ratifying and non-ratifying countries (International Labour Organization, 1965).

The ILO adopted the first Maternity Protection Convention (C3) at its very first conference in 1919 and has since updated the standards for maternity protection every three decades in two consecutive conventions (C103, C183), in 1952 and 2000, respectively. The ILO Maternity Protection Conventions include the five main standards, as Table 1 shows: benefit amount, benefit duration, coverage, financing method, and introduction of social assistance scheme. The ILO included standards that addressed maternity protection measures for all after the 1944 Declaration of Philadelphia. The second Maternity Protection Convention required states to broaden the coverage of maternity benefits to all employed women, including domestic workers and farmworkers, and to introduce a social assistance scheme for women who do not qualify for maternity insurance. However, the ILO succumbed to LMICs’ complaints that it would be too costly to expand the coverage of maternity protection to all employed women, allowing states to adopt this standard in accordance with the level of national economic development (Boris, Reference Boris2019: 75). Given that the non-compliance of ILO standards with the compromise formulation is mostly taken for granted (Maupain, Reference Maupain2013), the provision of social assistance benefits was practically the only standard that pursued the universal coverage of maternity protection until 2000. The third convention consolidated its goal of broadening the coverage of maternity protection to all women by lifting the compromise formulation of the coverage standard and continuing to include the provision of a social assistance scheme.

Table 1. Five components of the standards of the ILO Maternity Protection Conventions

Empirical Analysis

Impact of the ILO conventions on legislative change

I use measures of the year when 160 independent states improved the five standards of maternity protection through the introduction or expansionary reforms of maternity protection policies from 1883 to 2018 as dependent variables. The expansionary reform in this article does not necessarily indicate the formal enactment of maternity leave legislation. States often change the substance of the legal conditions of maternity leave through administrative regulations instead of legislation, particularly in the case of the benefit amount or benefit duration. Conversely, the enactment of new maternity laws sometimes does not change any substance of the five standards of maternity leave. More specifically, I use five dichotomous measures of whether the five maternity protection standards, as stated in the ILO Maternity Protection Conventions, are extended at the national level. For instance, the amount variable indicates incidences when states extended the benefit amount of maternity leave. The amount variable, duration variable, and coverage variable include the first adoption of paid maternity leave since benefit amount, benefit duration, and coverage were extended in this incidence. In contrast, the financing variable and the social assistance variable do not include all adoptions of maternity leave but count only the incidences when specific types of social security systems, e.g. social insurance and social assistance, were introduced.

Figure 1 shows the share of countries per year that extended the provision of maternity protection in terms of the five standards of the ILO Maternity Protection Convention. The figure reveals a substantial variation of the five international standards in the degree that states incorporated them into domestic laws. While we can observe more than 250 events in which states extended the amount, duration, or coverage of maternity benefits, only 83 countries changed their financing method to social insurance, and just 50 countries introduced a social assistance program.

Figure 1. Share of countries per year that extended the provision of maternity protection by standards of paid maternity leave

To systematically analyze the impact of the ILO conventions on the timing of the extension of maternity protection policies, I estimate a Cox proportional hazards model with standard errors, adjusted for clustering in countries. I use two types of event history analyses, a recurrent and a non-recurrent, depending on the maximum number of legislative changes per standard. Namely, states extend benefit amount, benefit duration, and coverage more than once to conform to the ILO standards that are updated every three decades, while states transition the financing method of maternity leave benefit from employer liability to social insurance or introduce social assistance schemes only once. The countries in recurrent analyses are considered until 2018 while the countries in non-recurrent analyses are considered until they introduced social insurance or social assistance. The number of countries in non-recurrent analyses is also smaller than the one in recurrent analyses due to the cases that predecessor states (e.g. colonies, USSR, and Yugoslavia) had already institutionalized social insurance or social assistance, reflecting the exclusion of dependent states from the empirical analyses. Although the main research interest of this paper is the influence of ILO conventions on independent states’ behavior, I acknowledge that this exclusion would induce underestimation of colonial experience’s influence on social policy expansion in former colonies, many of which have introduced social policies during their colonial time (see Appendix 1 for details). The recurrent event history analysis is stratified by how many reforms each country has already adopted: since the logic behind the timing of the first and the fifth legislative change, for instance, would be different (Box-Steffensmeier and Jones, Reference Box-Steffensmeier and Jones2004: 160–161). To test the robustness of the results in terms of temporal dependency, I estimate logistic regressions with a cubic polynomial approximation (Carter and Signorino, Reference Carter and Signorino2010).

The key explanatory variable is the three ILO Maternity Protection Conventions. Based on prior research on the effectiveness of ILO conventions (Berkovitch, Reference Berkovitch1999; Helfer, Reference Helfer2006), I expect that states will extend provisions of maternity protection shortly after a large number of states have initial access to ILO conventions – namely, when member states submit a convention to the competent national authorities within eighteen months from the adoption of the convention. Thus, the ILO Convention variable identifies the years 1921, 1954, and 2002 as dichotomous variables. I set a five-year period during which the effects of ILO conventions can unfold because it is expected that the effects of the conventions would fade as the salience of the maternity protection issue, awakened by the submission of the conventions, fades within a reasonable period. I include several control variables to take alternative determinants of the initiation and extension of maternity leave policies into account. The regression models include ILO membership, logged GDP per capita, the change in total fertility rate, political regime type (democracy), initiation of other welfare programs, religion, and colonial heritage (see Appendix 2 for operationalization of variables).

Are the ILO Maternity Protection Conventions influential in the extension of maternity protection policies in the world? That seems to depend on the standard. Shortly after states have access to newly updated ILO standards, they tend to adopt the standards into domestic laws concerning benefit amount, benefit duration, coverage, and financing method as shown in Table 2. However, Model 5 shows that the ILO Maternity Protection Conventions do not influence the timing of the introduction of social assistance schemes. Particularly, LMICs tend to omit the standard that could benefit informal sector workers. Among 20 countries that adopted social assistance schemes before the 1990s, Chile was the only LMIC. Many LMICs began to introduce social assistance programs for pregnant women and their infants only after the 1990s, regardless of the adoption of the ILO conventions, as Figure 1 shows. This trend may reflect a change in the ILO’s strategy from eliminating informal employment to broadening social security coverage to cover informal sector workers in the 1990s, prior to which the ILO was also negligent in applying its standards in informal sectors in LMICs. For instance, the ILO’s lax enforcement of social assistance standards gave LMICs sufficient room to ignore this element. In its observations on non-compliance, the ILO did not bring up the issue of social assistance until 1982, when it suggested that the Bolivian government should try to introduce social assistance or lower the qualification period.Footnote ²

Table 2. Event history analyses: legislative changes of maternity protection policies

* p < 0.05, ** p < 0.01, *** p < 0.001

The results remain robust after changing estimation methods to logistic regressions with a cubic polynomial approximation (see Appendix 3). Taking a brief look at the control variables, states tend to adopt or extend maternity leave policies in a similar timeframe to the introduction of other welfare programs. While ILO members extend paid maternity leave policies more frequently than non-ILO members, Islamic countries improve the statutory support for working women to maintain paid work and maternal duty less frequently than majority-Christian countries. It is also notable that the benefit duration and coverage of maternity protection of former French colonies was stagnant after independence, as France transposed its high standards, 14 weeks of maternity leave for all employed women, onto its colonies during colonial times (Schmitt, Reference Schmitt2015; Son, Reference Son, Dobrotić, Blum and Koslowski2022).

The gap between laws and practices in LMICs

To examine whether there exists a large gap between laws and practices in LMICs in comparison to high-income countries, I tested a conditional effect of country income on the discrepancy between the level of generosity and the entitlement of laws and expenditure. I drew on the generosity score from the HDML that measures entitlement and generosity of maternity leave laws in each country. I used the equation of the generosity scores of the Comparative Welfare Entitlements Dataset (CWED) but with an adjustment. Originally, the generosity scores of the CWED standardize and aggregate various de jure and de facto conditions of welfare programs using (1) benefit amount, (2) benefit duration, (3) contribution period, and (4) de facto coverage (Scruggs, Reference Scruggs2014). Since the research interest of this paper lies in capturing the discrepancy between laws and practices, I substituted de facto coverage to de jure coverage that measures in which employment sectors workers are entitled to maternity leave benefits. Following the standards of the ILO Maternity Protection Conventions, coverage is categorized into four sectors: industrial, commercial, agricultural, and atypical. De jure coverage scores are generated through the addition of these four categories, yielding a variable ranging from 0 to 4, 0 indicating the absence of maternity leave policy, and 4 indicating full coverage of the four sectors. The adjusted equation follows:

Maternity generosity score = z(Benefit replacement rate) + z(Benefit duration weeks) - z(Benefit contribution weeks) + z(De jure coverage categories)

Regarding dependent variables, the ILO Social Security Expenditure Database provides maternity expenditure as a percentage of GDP in mostly LMICs while the OECD expenditure database provides expenditure data for high-income countries. Although the OECD expenditure indicator provides the expenditure on cash benefits of both maternity and parental leave and the ILO expenditure indicator concerns only maternity leave, the Pearson correlation between the OECD and ILO indicators shows high and significant correlation in overlapped cases (r = .8, p = .0002). To operationalize the conditional hypothesis, I construct interaction terms between GDPs per capita and the generosity score of maternity leave policies. Although the HDML provides generosity scores for 160 countries with long time periods, the expenditure data only allows for the inclusion of 52 countries. The restricted samples include 27 high-income countries, 24 middle-income countries, and 1 low-income country, so a low proportion of low-income countries may cause the downward-biased estimation of a gap between laws and practices (see Appendix 4 for the full list of countries and their attributes). The period coverage of the dependent variable is also limited, covering only one or two years per country between 1989 and 1999. Thus, I use the average maternity expenditure from 1989 to 1999 as dependent variables and include independent and control variables in the standard of 1995 in OLS cross-section models.

As ‘dependent variable problem’ studies in the field of comparative welfare research argue, demand factors: such as the ratio of the elderly population and the unemployed, may blur the correlation between the generosity score and welfare expenditure (Clasen and Siegel, Reference Clasen and Siegel2007). Therefore, all models include demand factors: such as the total fertility rate and the ratio of female labor participation as control variables. Regression models also include many additional variables that are considered by comparative welfare state research as important variables of welfare expenditure. The Dublin-Watson test found no autocorrelation problems and the VIF scores also proved that no variables meet the conventional VIF threshold of 10. However, the Breusch-Pagan test revealed that one of the regression models has the heteroscedasticity problem. Since the presence of heteroscedasticity makes the OLS standard errors incorrect, I employed robust standard errors in the model.

The regression analyses capture a large discrepancy between law and practice in LMICs. Model 3 in Table 3 shows that generosity scores in LMICs do not correspond to maternity expenditure even after controlling for demand factors, while Model 2 shows a tight link between generosity and expenditure in high-income countries. Evidence from Model 4 further supports the conditional argument. The interaction effect variable between GDP per capita and maternity generosity score is positive and statistically significant (β = .040, p = .013), implying that the higher GDP per capita a country has, the more tightly linked the generosity of maternity leave policies will be with maternity expenditure. Figure 2 plots the predicted value of maternity expenditure on maternity generosity score at different levels of GDP per capita. The plot clearly shows that maternity generosity score is positively correlated with maternity expenditure with a 95 percent confidence interval in high-income countries (1 standard error above the mean of logged and scaled GDP per capita), whereas the relationship is unclear in low-income countries (1 standard error below the mean).

Table 3. OLS regression analysis: maternity expenditure

Note: Model 2 does not include polity as all high-income countries are democratized. *p<0.1; **p<0.05; ***p<0.01

Figure 2. Predicted maternity expenditure on maternity generosity score by GDP per capita

The gap between laws and practices in the informal sector

To capture the gap between coverage in laws and practices in the informal sector, I calculate expected coverage based on the HDML to estimate the proportion of women workers entitled to maternity leave benefits by laws with which I compare coverage in practice. I multiply the above-mentioned de jure coverage, the variable that measures in which sectors women workers are legally entitled to maternity leave benefits, with the ratio of female employment in each employment sector of all female employment to calculate the expected coverage of maternity leave. However, a special problem arises when calculating the expected coverage of the informal sector: while most legislation covers all employees in industrial, commercial, or agricultural sectors without disaggregating them into a detailed list of occupations, no maternity leave policies in the world cover all informal sectors. The expansionary reforms regarding informal sector workers are gradual, extending from one group (e.g. self-employed) to another group (e.g. domestic workers). Since no data exists regarding the proportion of female employees in disaggregated types of informal employment, I built two expected coverage indicators that embed two “extreme” assumptions: (1) all informal sector workers are covered if a maternity leave policy covers at least one informal sector or if states provide social assistance benefits; and (2) states would not enforce maternity leave laws in the informal sector. For example, if a maternity leave policy covers female employees in industrial, commercial, and agricultural sectors, as well as domestic workers, the first expected coverage indicator would be calculated as 100 percent. The second expected coverage indicator, by contrast, would be calculated as “(ratio of female employment in industrial, non-industrial, agricultural sector)*(1-ratio of female employment in informal sector).” Here, the ratio of informal sector workers is not simply deducted because the informal sector is a terminology based on types of employment rather than sectoral classification. The aggregated ratio of women workers in industrial, commercial, and agricultural sectors is 100 percent. The comparison of which expected coverage predicts coverage in practice would allow us to better detect the implementation constraints in the informal sector.

I use the number of contributors of maternity leave insurance and registered people within the social assistance scheme as coverage in practice (International Labour Office, 2010: 144–149). It measures the number of potential beneficiaries of maternity benefits and does not perfectly capture the number of actual beneficiaries. The number of actual beneficiaries is usually lower than the number of contributors as employers may refuse to pay maternity benefits to women workers who are legally eligible for maternity benefits due to lack of inspection and weak collective power (Boris, Reference Boris2019). Unfortunately, comparable data about the actual beneficiaries across countries do not yet exist. Hence, the empirical estimates on the link between coverage in law and practice may be downward-biased, as our alternative measures count only the cases in which women workers are not registered in maternity protection programs or give up contributing insurance. Nonetheless, case studies on implementation constraints in the informal sector argue that a large proportion of informal sector workers may not even be registered in the social insurance system due to additional costs for employers (Bosch et al., Reference Bosch, Melguizo and Pagés2013). The ILO provides data on the number of potential beneficiaries of 90 countries in 2010, thus the case number in the empirical analysis decreases (see Appendix 4 for the full list of countries and their attributes). As the indicator is composed of categorical values (0-9, 10-32, 33-65, 66-89, 90-100), I employ ordered logistic regressions but with a modification. The extremely low proportion of LMICs that reached near-universal coverage makes it impossible to test the proportional odds assumption. Therefore, I merge the two categories of 66-89 and 90-100. Brant tests revealed that the ordinary logistic regression models with the modified dependent variable do not violate the proportional odds assumption.

Figure 3 shows three bivariate plots of the expected coverage of maternity protection policies and coverage in practice with different sample sizes. While Plot 1 shows the patterns of the first expected coverage in all 90 countries, Plots 2 and 3 show correlations between two expected coverage variables and coverage in practice in 51 LMICs. The sample size decreases further in Plots 2 and 3 due to limited coverage of the informal employment data that is required to build the second expected coverage. While the limited samples of the earlier OLS regression analyses included disproportionately small numbers of low-income countries, the distribution of country income is relatively balanced in this sample that includes 19 upper-middle-income countries, 20 lower-middle-income countries, and 12 low-income-countries. The size of points shows the proportion of countries by country income group.

Figure 3. Bivariate scatterplots of two expected coverages and coverages in practice

Plot 1 shows that women workers in all sectors were legally entitled to maternity benefits in most countries – namely, 62 of 90 countries in 2010. All high-income countries covered female employees in the industrial, non-industrial, and agricultural sectors, as well as women workers in at least one of informal sector. Maternity protection policies in the majority of LMICs, 36 of 64 countries, provide the same level of legal coverage. Only six LMICs exclude female employees in the agricultural and informal sectors from their maternity insurance system, of which two countries provide social assistance benefits. However, in practice, high-income countries are separated from LMICs by a wide gulf in terms of coverage of maternity benefits. While only 9 of 26 high-income countries fail to reach 100 percent of coverage in practice, no LMICs, except Latvia and Serbia, achieve universal coverage. The median of coverage in practice in LMICs is between 10 to 32 percent of employed women, while the median for high-income countries reaches 90 to 100 percent. Less than 9 percent of women workers in one-third of LMICs would potentially receive maternity benefits.

Plots 2 and 3 demonstrate that the second expected coverage predicts the coverage in practice more precisely. The median of the second expected coverage (27 percent) matches better with the median of coverage in practice (10-32) than the median of the first expected coverage (63 percent). For instance, the second expected coverage predicts that 6 percent of working women would be eligible for maternity benefits in Cambodia even if the maternity protection laws cover all employment sectors including handicraft establishment and self-employment. Only 0 to 9 percent of working women would potentially receive maternity benefits. The results of ordinary logistic regression models that control for additional variables in Table 4 provide further evidence that women workers in informal sectors are not likely to receive maternity benefits, regardless of whether the statutory provision of maternity protection covers the informal sector.

Table 4. Ordered logistic regression analysis: coverage in practice

Note: *p<0.1; **p<0.05; ***p<0.01