Page 41 - IJPS-3-2

Do young children prohibit mothers from working in Ethiopia?

reversal in the positive effect of the number of children, although it is not statistically significant.
4 Discussion

Some key points emerge for discussion from the results section. The first is the relative importance of coefficients’
magnitudes for estimates from exogenous probit and ivprobit models. That is, in some cases coefficients from exogenous
estimates are larger than those from endogenous estimates, and in some other cases, the reverse is the case. The lack of
consistency in coefficient size from the ivprobit estimator compared to the exogenous probit estimator in the present
study, however, is in line with the available research evidence for several other countries. Most previous research shows
larger coefficients from exogenous models exaggerating the effect compared to those from endogenous models (e.g.,
Angrist and Evans, 1998 and references therein). However, there is also evidence documenting larger coefficients for
estimates from endogenous rather than exogenous models. For example, Rosenzweig and Wolpin (1980a) note that
instrumenting endogeneity increases the coefficients compared to the exogenous model. For Korea, Chun and Oh (2002)
found larger coefficients using endogenous estimates compared to exogenous estimators when using households with at
least one child, but smaller endogenous estimates when using households with at least two children.
Researchers (e.g., Aassve and Arpino, 2007) attribute this inconsistency in exogenous and endogenous coefficients to
the instrumental variable used. Whereas the use of sibling sex composition provides a natural experiment whereby
households with same sex siblings are treatment groups and those with mixed sex siblings are control groups, a lack of
consistency is expected since the two models estimate essentially different things owing to their reference to different
samples (Aassve and Arpino, 2007). That is, the exogenous estimator coefficient represents the average effect of the
number of children over the entire population in the sample, whereas the endogenous estimator coefficient represents the
average effect of the number of children for those households whose first and second siblings have same sex. In such a
case, results from the exogenous estimator may be due to variables other than the number of children such as biases from
omitted variables, hence making causal inferences problematic.
The second point is regarding the heterogeneity in coefficient signs between the rural and the urban sub-samples. The
negative ivprobit coefficient on the probability of maternal work participation effect of young children for the urban
mother in the second panel (despite its being statistically insignificant) and the positive and statistically significant
ivprobit coefficient in the third panel is consistent with most previous evidence, although most such research is based on
rural-urban dummy instead of running separate analysis for rural and urban mothers (see e.g., Angrist and Evans, 1998;
Cáceres-Delpiano, 2008; Chun and Oh, 2002; Cruces and Galiani, 2007; Dupta and Dubey, 2003; Kim and Aassve,
2006; Orbeta, 2005; and references therein). This result is also consistent with previous lifecycle evidences. For
example, Hotz and Miller (1988) found that children tended to have negative effects during their early ages but not
during their adult ages, and that the intensity of time the mother spent tending her children markedly declined as children
grew up. Similarly, Assaad and Zouari (2003) for urban Morocco found that the presence of school-age children
significantly reduced the participation of women from all types of paid work.
The positive and statistically significant ivprobit coefficient for the rural sub-sample in the second panel, and the
negative though statistically insignificant ivprobit coefficient in the third panel, nevertheless, are inconsistent with the
theoretical prediction that holds that, other factors held constant, the mother’s probability of work participant decreases
with an increase in the number of young children and increases when children become more adult. However, consistent
with this result, using data from the 2000 Ethiopian DHS and instrumenting the number of children with the husband’s
desire for children, Solomon and Kimmel (2009) found positive (but statistically insignificant) labor supply effect of
children. In this connection, Angrist and Evans (1998:463) also cite a review that found that fertility either has no effect
on maternal labor supply, or it has a positive effect when endogeneity is considered.
The question now is why is this so? In the present study, it is argued that, despite the lack of statistical significance for
many of the ivprobit coefficients, the quantitative results’ inconsistency with theory and most previous research for the
rural households is rather due to the rural-urban difference in the employment structure and the effect of the household’s
lifecycle. Context-specific literature review and qualitative data seem to be revealing in this particular case.
First, the prevalence of household enterprises and traditional nature of farming in rural areas of poor economies
including Ethiopia means that more rural women have to work longer compared to urban women (see e.g., Arbache,
Kolev and Filipiak, 2010). In Ethiopia, farm plots are fragmented, farming is done manually, and productivity is low.
Households have to invest a lot of manual labor per unit area, and, as such, it would be likely for women to work in the
farms especially when there are other children to look after very young children at home. In such circumstances, children
may not be considered that much prohibitive to the mother’s work given the nature of the economy and the mother’s
need to work for the family, despite the adverse health implications that this is likely to have on the young children.

International Journal of Population Studies | 2017, Volume 3, Issue 2 35