Employment’s Role in Enabling and Constraining Marriage in the Middle East and North Africa

We focus on marriage rather than engagement because stages of the engagement process happen at different spacings, making these intermediate stages less comparable (Gebel and Heyne 2014). Engagements are commonly broken and are not the same durable transition to adulthood as marriage (Hoodfar 1997). Additionally, the couples’ living conditions are negotiated during the engagement process, and thus marriage timing depends in part on employment during engagement. We also test lagged employment status, which could be interpreted as employment status when engaged.

Because the hazards are very low at young and old ages, we combine the dummy variables for ages 15–18 and those for ages 35+ but do not drop any observations.

We tested dropping individuals who moved from their place of birth from our sample, and results were substantively similar. The exception was for women in Tunisia, for whom the result lost significance (possibly because of reduced sample size) but had the same direction.

For more information see Assaad and Krafft (2013), Assaad (2014b), and Assaad et al. (2016). Data are publicly available from ERF at www.​erfdataportal.​com.

We restrict our data to the 30 years preceding each survey. The Jordanian data do not distinguish between urban/rural areas of birth in the residential history data.

Because employment histories are available only starting at age 15, we start our analysis of marriage timing from age 15, excluding those (very few) individuals married prior to that age. Missing data also limits the size of the sample, particularly in Tunisia.

In the labor market history data, we can capture only market work, which is likely to underestimate work, particularly for women (Donahoe 1999; Langsten and Salem 2008). However, comparisons of labor market history and contemporaneous data show reasonably good data consistency, particularly for more persistent statuses, such as public sector work (Assaad et al. 2018b).

In Tunisia, because of small sample sizes, a number of neighboring governorates are combined.

They tend to be negative because if after contemporaneous local public sector employment opportunities are accounted for, the last period’s local public sector employment was higher, then the location is shedding public sector employment and thus is unlikely to hire a young person.

The larger size of the effects in the 3SRI model may be due to the fact that local average treatment effects (LATE), the effects for compliers, could be larger than the average treatment effects (ATE) for the sample as a whole (Angrist et al. 1996). In this case, and consistent with the monotonicity assumption, we expect that there will be compliers (individuals who would not have worked in the public sector otherwise induced to do so by increases in local public sector employment opportunities) but no defiers (individuals who would have otherwise worked in the public sector who do not do so as a result of greater local opportunities).

The simulations are run for a secondary graduate who was in school until age 18; who has secondary-educated parents; whose father was a self-employed professional; whose mother did not work; and who was located in Cairo for Egypt, Amman for Jordan, and Tunis for Tunisia. She or he has two brothers and two sisters and was born 35 years before the survey round. These characteristics create a baseline hazard from our models, over which we vary employment statuses. Different characteristics would lead to different ages at marriage but the same general structure of tradeoffs. For example, if we simulate a man born in rural Lower Egypt, instead of Greater Cairo, because the hazard ratio of rural Lower Egypt is greater than 1 (see Table 2), marriages will be systematically a year earlier for all profiles than shown in Fig. 1. A similar result of shifted ages pertains with other characteristics; they will be systematically shifted, with the direction depending on their hazard ratios.