2. Theory and previous research

Religious and ethnic identity can impact FLFP through several mechanisms. One key channel is the influence of social norms, which are shared expectations and informal rules that govern behavior within a group. These norms can shape attitudes toward women’s work and influence both individual choices and structural constraints in the labor market. Religious doctrines, for instance, may prescribe specific roles for women, reinforcing expectations that prioritize domestic responsibilities over workforce participation. In particular, norms related to gender segregation or the belief that women should occupy a subordinate role in the home or public sphere may create significant barriers to FLFP. Because norms are socially constructed and transmitted across generations, they often vary across religious and ethnic groups.

Empirical research has shown that countries with Muslim majorities tend to have significantly lower FLFP compared to the global average (Tzannatos, Reference Tzannatos1999). Several studies have examined the role of religiosity in shaping these patterns (Dildar, Reference Dildar2015; Ugur, Reference Ugur2018). Differences in social norms across religious sects can also contribute to variations in FLFP. For example, Akyol and Ökten (Reference Akyol and Ökten2024) argue that more gender-equal views among Alevi Muslims in Turkey led to an FLFP premium of about twelve percentage points in this group compared to their Sunni Muslim counterparts, even after controlling for observed characteristics. Likewise, ethnic cultures may carry deeply ingrained beliefs about gender roles. In the MENA region, for example, social norms specific to Arab culture have been theorized as a major factor in explaining low FLFP rates (Korotayev et al., Reference Korotayev, Issaev and Shishkina2015). These findings highlight how religious and ethnic norms can act as both direct constraints on FLFP and as broader structural forces that shape societal expectations and labor market dynamics.

On the other hand, group inequalities in socio-economic status or in access to education may also influence FLFP. Different ethnic and religious groups may face different economic opportunities, particularly if these groups are geographically segregated, or when institutional barriers or discrimination affects certain groups. Differences in economic development across areas can also shape FLFP rates in line with the U-feminization hypothesis. Other variables that influence FLFP in a general sense may also contribute to differences in FLFP across groups, such as the age distribution of the population, when differences in FLFP are present across age groups concurrently with differences in the age distributions of different religious/ethnic groups.

Early research into ethnic and religious differences in FLFP employed data from developed countries such as the US. Reimers (Reference Reimers1985) explores the ethnic gaps in FLFP in the US and finds that these can be mostly explained by observed characteristics such as age, education level, and family size. Similarly, Lehrer (Reference Lehrer1995) studies FLFP gaps between Christian denominations in the US and finds that the differences between denominations are partly but not fully accounted for by observed characteristics. More recent studies include cross-country analyses, as well as analyses that focus on developing country contexts. Ross (Reference Ross2008) theorizes that low FLFP in Muslim countries is due to the presence of oil rents, which reduces FLFP through the growth of non-tradable sectors (which include construction and retail services) and through high household incomes which increase women’s reservation wages. While cross-country regressions do not find a strong correlation between fuel exports and FLFP (Bayanpour-Tehrani & Sylwester, Reference Bayanpour-Tehrani and Sylwester2013), the nature of Middle Eastern economies can contribute to the observed low FLFP. Demand-side factors, such as the shrinking public sector (Assaad et al., Reference Assaad, Hendy, Lassassi and Yassin2020), and dualism in the labor market where public sector jobs are highly sought after while outside options are limited to low-quality informal jobs, are additional constraints on FLFP (Assaad, Reference Assaad2014). Gender segregation within the workplace may also shape the demand for female workers as Chatterjee and Vanneman (Reference Chatterjee and Vanneman2022) show in the case of India. In a similar vein, İlkkaracan (Reference İlkkaracan2012) suggests that the lack of a demand-side challenge to breadwinner model resulted in the institutionalization of the gendered labor division in Turkey.

Gaps in economic development across regions may contribute to FLFP gaps between ethnic and religious groups, especially when these groups are geographically segregated. A substantial body of literature explores the impact of economic development on FLFP, frequently employing the feminization U hypothesis (Boserup, Reference Boserup1970; Goldin, Reference Goldin1995). The hypothesis suggests a pattern in FLFP across different stages of economic development. In agricultural economies, FLFP is generally high as women often participate in farming and related activities. However, as these economies transition to industrialization, FLFP tends to decline. This decline is attributed to the shift toward factory jobs and other industrial work, which historically have been less accessible or less culturally acceptable for women. As economies further evolve into post-industrial stages, focusing more on services and knowledge-based industries, FLFP begins to increase again. These sectors often offer more opportunities and greater flexibility for women, leading to higher FLFP. Research focused on developed countries shows the presence of long-term secular increase in female labor force participation over the course of the 20^th century (Stanfors & Goldscheider, Reference Stanfors and Goldscheider2017; Mancini, Reference Mancini2018). Changing cultural norms and the family division of labor are part of this story, while structural transformation may have also played an important role (Olivetti & Petrongolo, Reference Olivetti and Petrongolo2016). While this theory has been used to account for differences in FLFP across countries and across time, it has seldom been used to address FLFP gaps within countries or across ethnic and religion groups.

Syria’s complex social fabric and its historical progression from an agriculture-based economy to a more diverse economyFootnote ¹ have resulted in differential economic opportunities across groups, possibly impacting FLFP. The different FLFP rates across groups could be shaped by differentials in labor market conditions and in economic development as suggested by the feminization U hypothesis. Economic development can shape FLFP gaps, not only through regional differences in the nature of the economy and available opportunities for women, but also through differences in education attainment across groups, and the demographic composition of different communities, as well as through changing norms that evolve in the process of economic development.

This study contributes to the literature on ethnic and religious disparities in FLFP by focusing on Syria, a country that exhibits both the broader Middle Eastern pattern of low FLFP and significant heterogeneity in labor market outcomes across religious and ethnic groups. While much of the existing research attributes low FLFP in the Middle East and North Africa (MENA) region to cultural and religious norms (Algan & Cahuc, Reference Algan and Cahuc2006; Bayanpour-Tehrani & Sylwester, Reference Bayanpour-Tehrani and Sylwester2013; Dildar, Reference Dildar2015; Kızılca, Reference Kızılca2016; Atasoy, Reference Atasoy2017; Ugur, Reference Ugur2018; Akyol & Ökten, Reference Akyol and Ökten2024), this paper shifts the focus to structural economic and demographic determinants of FLFP disparities within a single Muslim-majority country. The analysis will examine the contributions of various factors to explaining ethnic and religious gaps in FLFP, including economic and demographic structure, and cultural norms. This paper takes an exploratory approach to assessing the relative importance of these factors. Rather than assuming a dominant explanatory mechanism, the analysis tests the strength of association between these factors and FLFP gaps, allowing for an empirical evaluation of whether structural economic and demographic conditions outweigh the influence of cultural and social norms.

3. Ethnic and religious differences in Syria

Syria stands out for its extraordinary ethnic and religious diversity, often described as a “mosaic society” to emphasize the coexistence of various cultural groups (Hinnebusch, Reference Hinnebusch2004). This diversity is deeply rooted in the country’s geography, which has long facilitated migratory movements and reinforced internal fragmentation. Historically, the region that is now Syria was part of the Ottoman Empire until the First World War and later came under French rule before gaining independence in 1945. Since the early 20th century, Syria has undergone significant economic and social transformations. These include state-led industrialization starting in the mid-20th century, followed by economic liberalization in the late 20th century, alongside rising regional economic inequality in the decades leading up to the recent civil war (Hassine, Reference Hassine2015).

Syria provides a uniquely valuable case study for analyzing FLFP in relation to religious, ethnic, and economic factors. While Syria’s FLFP rates are low, in line with other Middle Eastern countries, its diversity and pronounced regional economic inequalities offer a rich spectrum of opportunities and constraints for women. This combination allows for a deeper exploration of how cultural norms and economic conditions shape women’s participation in the labor market. As a Middle Eastern country, Syria has a Muslim and an Arabic-speaking majority, and its low levels of FLFP make it a microcosm for examining mechanisms in the broader region. Using data from the pre-civil war period allows for comparisons across groups in the same institutional context prior to the cantonization brought on by the war which starting in 2011.

Figure 1 shows a map of Syrian cities, towns, and villages according to ethnic majority. The analysis in this paper follows the definition of ethnic identity which relates this identity to claims of shared ancestry and the cultural practices associated with those claims (Khaddour & Mazur, Reference Khaddour and Mazur2018). Ethnic identity is used here as an umbrella term for religious, sectarian, and national differences. The ethnic majority group in Syria is typically conceptualized as the Arab Sunni Muslim population. However, Khaddour and Mazur (Reference Khaddour and Mazur2018) suggest a more nuanced view, separating this group into tribal and non-tribal groups, employing the terms “Sunni tribes” and “Sunni families” – a terminology adopted in this paper. The Sunni tribes occupy the north-east of the country and have a distinct culture despite sharing the same language and religious affiliation with the Sunni families in the west. Apart from Sunni Muslims, major ethnic groups in Syria include Alawi Muslims, Druzes, Kurds, and Christians, as well as smaller groups such as Shia Muslims, Ismaili Muslims, and Circassians who are all aggregated under the category “Other minorities.” Most of the cities have mixed populations and are categorized as such, while smaller towns and villages are mostly homogenous with a clear majority group in each.

Figure 1. Ethnic composition and female labor force participation (FLFP).

Most Sunni Muslims in Syria follow the Hanafi school of Islamic jurisprudence, noted for its relative adaptability compared to other Islamic legal traditions. Still, religion holds a prominent place in daily life and may shape individuals’ viewpoints and preferences. Several other Muslim and non-Muslim groups are present alongside the Sunni Muslim majority. Antoun (Reference Antoun1991) emphasizes the distinctiveness of Syria’s religious sects, which typically occupy peripheral locations, cluster demographically in territorial “homelands,” and are exclusive in culture and social organization. Muslim sects like Alawis and Druzes, alongside Christians, do not follow the practices that may be common amongst Sunni Muslims, like gender segregation, such as wearing the veil which is common amongst Sunni Muslim women. Cultural and social norms of this kind may play a role in determining FLFP within each group. Sect groups in Syria not only have distinct informal norms, but they also maintain separate courts that deal with matters of personal law such as marriage and divorce, and inheritance (Van Eijk, Reference Van Eijk2016), a practice with deep historical roots in the MENA region.

Other groups like the Sunni tribal communities, Kurds, and other minorities like the Ismailis, Circassians, and Armenians all tend to have distinct social and cultural norms that distinguish each of them. Even though communities like Sunni tribes, Kurds, and Circassians, largely follow a similar form of Sunni Islam, the distinct norms in these communities may shape the lives of women and may affect their propensity to enter the labor market, based on views prevalent in each community regarding women’s proper role in society.

Ethnic and religious inequalities in socio-economic outcomes remain under-studied in this context. However, studies that address neighboring countries can be useful to inform our expectations as many ethnic communities cut across national borders in the region. An early study of religious gaps in fertility in Lebanon found lower fertility among Christian as compared to Muslims, as well as differences between Muslim and Christian sects, but these appeared to converge at higher levels of education (Chamie, Reference Chamie1977). Differentials in education levels between religious groups in the region are long-standing, with Christian school enrollment overtaking Muslim enrollment in the early 20^th century (Sassmannshausen, Reference Sassmannshausen2019). Under the French mandate of Syria and Lebanon (1923–1946), private and missionary school were mostly attended by Christian students, while public schools served a majority of Muslim students (Abi-Rached & Diwan, Reference Abi-Rached and Diwan2022), which suggests differences in access to education across groups. The post-colonial period saw the establishment of modern primary schools, which coincided with an increase in Muslim education levels in Egypt, and the beginning of a process of convergence in Muslim and Christian educational levels and labor market outcomes in that country (Saleh, Reference Saleh2016). Still, Christians in the MENA have overall higher incomes, higher educational attainment, and lower fertility than Muslims. Several studies draw on data from Turkey, where religiosity is found to have a negative association with FLFP (Dildar, Reference Dildar2015; Ugur, Reference Ugur2018). Additionally, Alevi Muslim women in Turkey have higher participation than their Sunni Muslim counterparts (Akyol & Ökten, Reference Akyol and Ökten2024). Some studies also assess religious differences in FLFP among Arab women in Israel (e.g., Khattab, Reference Khattab2002; Yonay et al., Reference Yonay, Yaish and Kraus2015) finding significant gaps between Christian, Druze, and Sunni Muslim women in that context.

Courbage and Todd (Reference Courbage and Todd2014) address the fertility gaps among ethnic groups in Syria, focusing on differences in fertility between Alawis, Sunni Muslims, and the tribal regions of the north-east. The authors propose that these divergences are a product of entrenched family structures. Specifically, they highlight that the Alawi community has preserved traditional matrilineal practices, while the Sunni tribes remain the most patrilineal among the nation’s ethnic cohorts. Balanche (Reference Balanche2015), on the other hand, links the diminished Alawi fertility with the relative advancement of the Alawi minority which commenced in the second half of the 20^th century. Balanche posits that the acceleration in the fertility transition was due to increased education levels and the induction of Alawi women into public sector employment. These contrasting theories on Alawi fertility can also be used to address women’s entry into the labor force. Elevated FLFP in Alawi regions might stem from beliefs or traditions that inform preferences within this group. Alternatively, higher FLFP could be attributed to more recent factors pertaining to the labor market structure in Alawi areas. The same argumentation can be applied to addressing the Christian advantage in FLFP. The FLFP advantage may be due to differences in deep rooted beliefs among Christians and Sunni Muslims, or they may be a product of other factors including the demographic and economic structure at the local level.

Previous research has highlighted some implications of structural change for gender dynamics in the MENA and how the resulting changes in patterns of life and work affect men and women differently. Abdelali-Martini et al. (Reference Abdelali-Martini, Goldey, Jones and Bailey2003) describes the feminization of agricultural work which took place in MENA countries including Syria (see also Abdelali-Martini & Dey de Pryck, Reference Abdelali-Martini and Dey de Pryck2015). The increased availability of work outside of the agricultural sector led to an exodus of men into more productive and higher paid occupations, increasing the proportion of women among agricultural workers. This points to the spread of the male breadwinner model as a result of structural change. Female labor in the agricultural sector in the MENA includes both waged work in agricultural gangs such as those investigated by Abdelali-Martini, as well as traditional forms of work within the family unit which endures in many areas.

Figure 2 shows the sectoral composition of the labor force across the geography of Syria, which clarifies the inequalities discussed above. The north-east of Syria has a predominantly agricultural economy, while the secondary sector is largest in the western parts of the country apart from the coastal region, and the latter has a mainly service-based economy coinciding with the highest rates of employment in the public sector.

Figure 2. Labor force composition by sector at the local level in Syria.

Based on the figure above and the U-feminization hypothesis, we would expect large differences to emerge in FLFP rates between groups in Syria. The first panel in Figure 3 shows FLFP in Syrian towns and villages as a function of income. The line of fit shows a clear U-shaped pattern, with FLFP decreasing as a function of income at low values of income, and then increasing at higher values of income. Areas with a majority of Sunni tribes have higher FLFP that is falling with income, while Sunni families appear to have reached the lowest FLFP rates in the mid-section of the U-curve. Meanwhile, Alawis and Christians have higher FLFP owing to their high average incomes and higher economic development and associated education levels in their areas.

Figure 3. (a) Female labor force participation as a function of income. (b) Female labor force participation as a function of income by ethnic majority. Note: The income measure is imputed based on the sectoral composition of the labor force. The line of fit is estimated using a non-parametric loess function with observations weighed according to population size.

Ethnicity-specific patterns can be seen more clearly in Figure 3b, which breaks down the same relationship according to ethnic group. The U-curve is most found clearly for Sunni families and Sunni tribes, where the U shape resembles that in the previous figure. The dashed line in Figure 3b shows a line of fit based on a similar model that adds a control variable for the share of the working population employed in the public sector, with the line of fit estimated at zero public sector shares for all groups. Controlling for public sector employment lowers the fitted values of FLFP in all groups, but its effect is seen most clearly in Alawi areas where a stark difference between the adjusted and unadjusted fit are found. Public sector employment may have shielded this group from the decline in FLFP that is found in Sunni majority areas. Meanwhile Kurdish FLFP appears to be falling with income, and Druze and Christian FLFP rates are rising with income. The limited data on these smaller groups and their geographic clustering may limit the extent to which the full U pattern can be seen. Similar results are found when using the alternative measure of income imputed using education data (See Appendix Figure 1). The finding that FLFP follows a U-shaped pattern as a function of income accords with the U-feminization hypothesis and justifies the use of second order polynomials in the regression equations in the next section.

4. Data and methods

The data employed in the analysis comes from the population census of 2004, which is the last census conducted in the country to date. The data is aggregated geographically at the level of cities, towns, and villages. It includes measures of female and male labor force participation, as well as data on demographics, economic structure, and education levels. Ethnicity data comes from previous work on the social structures in Syria (Khaddour & Mazur, Reference Khaddour and Mazur2018). A city, town, or village is assigned a majority based on an estimated 80% cutoff (i.e., 80% of the village is of the same ethnic group for the village to have an ethnic majority), with most towns and villages having clear ethnic majorities while cities are typically ethnically mixed. The data necessitates that the ethnic gaps are measured between ethnic majority areas but does not capture ethnic differences between populations within cities.

The use of aggregated data is justified by the nature of the research question, which seeks to explore gaps in FLFP across geographically distinct ethnic majority areas. Additionally, individual-level census data is unavailable in this context. It is important to note that this approach has several limitations: Aggregation omits within-area variation, making it impossible to capture individual-level differences within the same area, or those between ethnic minorities and majorities in mixed urban areas. Another limitation relates to the data on ethnic composition, which was collected based on expert interviews, rather than measured directly. Meaning that ethnic composition shares are likely to be highly imprecise. For this reason, a conservative 80% cutoff is used for assigning the ethnic majority for any city, town, or village, with areas at or below the cutoff treated as “mixed” areas.

To evaluate the relationship between FLFP and economic development, two alternative measures of income are imputed based on the sectoral composition of the labor force or alternatively based on the educational attainment of the labor force. Real wage data is used for the calculations which comes from Syrian Labour Force Survey of 2007. The imputation formula takes the following form for sector-based income:

$$Incom{e_i} = \;\;\mathop \sum \limits_{j = 1}^7 Sector\;wag{e_j}\;.\;\;Sector\;shar{e_{ij}}$$

Where i refers to the city, town, or village, j to the sector. Sector wage refers to the national-level average wage of sector j, and Sector share is the share of workers in the city/town/village employed in that sector. The set of sectors used to construct the sectoral income measure are agriculture, industry, construction, hotels and restaurants, transportation, finance and real estate, and other services. While wage imputation is standardly used in the study of income inequality (such as income scores, see Sobek Reference Sobek1995), its limitations should also be understood. In this case, the measure captures the sectoral composition of the labor force, with low values of sectoral income corresponding to areas where the labor force is predominantly employed in agriculture, while increasing values of the measure track with the transition of the labor force into secondary and tertiary sector occupations and eventually into service-sector and white-collar dominant economies. The sectoral income measure is standardized according to its z-score value, with the national average weighted by population size taking a value of 0 with a standard deviation of 1. The alternative specification of the income measure uses educational attainment instead of income using the following formula:

$$Incom{e_i} = \;\;\mathop \sum \limits_{j = 1}^7 Average\;wag{e_j}\;.\;\;Educational\;attainment\;shar{e_{ij}}$$

Where the educational attainment share refers to the share of the adult population in attainment categories: literate, elementary school, middle school, high school, middle academy, and university educated, with the illiterate group treated as the reference group. Average wage refers to the national-level average wage conditional on the educational attainment category.

To understand the contribution of social norms to explaining ethnic gaps in FLFP, gender parity indices (GPIs) in education are used to proxy for bias against girls and women. This bias may take the form of lower investment in girls’ and women’s education as compared to their male counterparts, or as prescriptive norms that limit their freedoms and participation in activities outside the home by limiting their education. Three GPIs are calculated based on the secondary school enrollment ratio between girls and boys, the schooling attainment ratio between adult women and adult men at the secondary level and above, and at the tertiary level. The GPIs measure at 100 in places with equality between the genders, and measure below 100 if male outcomes are higher than female outcomes, and above 100 when female outcomes overtake male outcomes. The enrollment GPI is calculated based on the enrollment of girls and boys aged 15–17 during the stage of upper secondary schooling. In the Syrian context, schooling is mandatory until the age of 16, which allows for variation in the enrollment rate that is determined by parental decisions after the age of 16. The GPI for secondary education and above (termed secondary+) captures the gap in educational attainment at the secondary level and above. And another GPI is used for educational attainment at the tertiary level.

Educational attainment of both women and men is determined by parental investment during late teens and early 20s. Students in Syria typically rely on family support, and in particular women who choose to pursue education beyond the minimum required by law. Pursuing education at above the secondary level also requires postponing marriage, and parental decisions typically play an important role in the timing of marriage. The GPIs are used in the analysis to proxy for the kinds of social norms that inhibit women’s participation in wider society, including prescriptive norms that curb mixing between women and men, or those that favor the seclusion of women or their early marriage. While secondary level students in Syria are often segregated on gender, there is no such segregation above the secondary level. Furthermore, secondary schooling establishments for boys and girls are typically located in close proximity and attending school typically offers opportunity for students of different genders to interact. In effect, low values of the GPIs correspond to places with high differential in educational investment in favor of males, which proxy for both favoritism towards males and limitations on women’s freedom to pursue activities outside the home on par with their male counterparts. Since educational enrollment in Syria is free at all levels, costs of education are not expected to shape the decision to attend school at the secondary or tertiary levels. Though the opportunity cost of education may play a role, particularly in agricultural regions where the labor of young men and women is used on the farm. This possibility is accounted for by including secondary polynomial terms for all GPI measures in the regression analysis.

The Gelbach decomposition method (Gelbach, Reference Gelbach2016) is used to evaluate the role of various factors that determine the ethnic gaps in FLFP. This method allows for estimating the contribution of each individual factor to explaining gaps between groups, without introducing bias due to the order with which the different variables are included. The Gelbach decomposition starts with an unadjusted ordinary least-squares model of the form:

(1) $$FLF{P_i} = \;\alpha + \;\mathop \sum \limits_{d = 1}^7 {\mu _{d,\ unadjusted}}\;{D_{i}}+ {\varepsilon _i}$$

Where i refers to the city, town, or village. FLFP is the female labor force participation ratio, which measures the ratio of the female population in working age (18–65) who are engaged in the labor force as proportion of the total female population in the same age group. D includes a set of seven ethnic majority dummy variables, which separate the cities, towns, and villages into different ethnic majority groups with Sunni Families as the reference category.

The adjusted model takes a similar form and adds a set of controls:

(2) $$FLF{P_i} = \;\alpha + \;\mathop \sum \limits_{d = 1}^7 {\mu _{d,\;adjusted}}\;{D_{i\;}} + \;\sum \gamma \;{X_i} + \;{\varepsilon _i}$$

The set of controls $\;{X_i}$ include male labor force participation (MLFP), demographic controls, the imputed income variables, the shares of workers employed in the public sector and in the family sector, and gender parity indices. The demographic controls include the marriage fertility rate, and the proportion of the population aged 15–64. Second-order polynomial terms are used for the demographic controls, income measures, and the gender parity indices to account for the non-linearity due to the U-feminization pattern. An additional specification adds a broader set of controls that include education composition of the female population, a set of infrastructure variables that measure the proportion of homes connected to the electricity network, the sanitation network, and drinking water network, and a set of controls that capture information on informality in home ownership, as well as a broad set of geographic controls that include longitude and latitude and information on agricultural suitability. The observations are weighed by population size, which allows for generalizable analysis that is not skewed by the most numerous villages with the smallest populations.

The expanded set of variables is meant to examine mechanisms and additional competing hypotheses. Education, infrastructure, and housing informality capture various mechanisms for the effect of economic development on FLFP. The inclusion of geographic variables is motivated by the idea that social norms around FLFP may be shaped by environmental factors, particularly those related to agricultural suitability. Theories of gendered division of labor suggest that historical reliance on different forms of agriculture – such as plow-based versus shifting cultivation – has long-term effects on gender roles, including attitudes toward women’s work outside the household (Alesina et al., Reference Alesina, Giuliano and Nunn2013). By including controls for agricultural suitability, the analysis accounts for the possibility that historical economic conditions contributed to the development of group-specific labor market norms that persist today.

The FLFP measure is based on an expanded definition which includes both employment in the “modern” labor force, as well as work performed in both the informal sector and in the family sector. The family sector counts activities that are performed within the family unit – which mostly refers to agricultural work but also includes any production activities performed within the household, without necessarily receiving a wage in exchange for this activity. This definition of FLFP does not include any service activities that are not intended for marketization, such as household chores or care activities. This definition accords with the available data and is standardly used in census data collection in developing countries (Gaddis & Klasen, Reference Gaddis and Klasen2014).

Having run models (1) and (2), the explained ethnic gap in FLFP that is due to the added explanatory variables can be expressed as ${\mu _{d,\;adjusted}}$ - ${\mu _{d,\;unadjusted}}$ .” To estimate the contribution of each variable to this gap according to the Gelbach method, a set of auxiliary regressions were run for each added covariate and each ethnic group, with each variable used in equation (2) as the dependent variable. For each auxiliary regression, an indicator variable is included for the ethnic majority, alongside controls for longitude and latitude and a constant term. For example, an auxiliary regression for MLFP for Alawis takes the following form (with Sunni families as the base category, and restricting the sample to only Sunni Families and Alawis):

$$MLF{P_i} = \alpha +{\tau _{MLFP,\;Alawi}}+{\varepsilon _i}$$

Then the contribution of $MLFP$ to explaining the Alawi-Sunni gap in FLFP is calculated using the following formula:

$${\delta _{\;MLFP,\;Alawi}} ={\gamma _{MLFP}}{\tau _{MLFP,Alawi}}$$

Where ${\gamma _{MLFP}}$ is the coefficient for $MLFP$ obtained from equation (2). The same process is repeated for every ethnic group and every variable introduced in equation (2).

In addition to the Gelbach decomposition analysis, the contribution of each variable to explaining ethnic FLFP gaps is tested using models that include each variable as the sole explanatory variable (besides the ethnicity dummy variables and latitude and longitude). The findings from this analysis show similar results to the Gelbach decomposition and are therefore included in the appendix (see Appendix Tables A.1–A.7).

It is important to note that the Gelbach decomposition method explains changes in coefficients due to correlation, but does not imply causal mechanisms. The method has the same limitations as those of OLS, like sensitivity to omitted variable bias. The results presented in the forthcoming section should be read with this limitation in mind. While the available data allows for exploring the roles of a rich set of variables, it is not possible to rule out the presence of other factors outside of the included variables. For example, social norms that are unrelated to the included GPIs and geographic variables, such as those related to the male breadwinner status.