Explaining the MENA paradox with returns to education
Empirical evidence from Egypt
2013-‐2014
Master’s thesis economics – development economics Supervisor: Noemi Peter
Student: Amber Leverne Schothorst Studentnumber: 5883695
Disclaimer: The Economic Research Forum and the Palestinian Bureau of Statistics granted Amber Leverne Schothorst access to relevant data, after subjecting data to processing aiming to preserve the confidentiality of individual data. Amber Leverne Schothorst is solely responsible for the conclusions and inferences drawn upon
Table of contents
1. Introduction 3
2. Literature review 4
2.1 Human capital theory and returns to education 4
2.2 Estimating the returns to education 5
2.3 Hypothesis 6
3. Methodology 7
3.1 Heckman two-‐step method 7
3.2 Heckman’s two-‐step model applied 8
4. Data and descriptive statistics 9
4.1 Data and sample 10
4.2 Representativeness of Egypt as a MENA country 10
4.3 Descriptive statistics 11
5. Empirical model 12
6. Results 13
7. Discussion and limitations 16
8. Conclusion 18
Acknowledgements 19
References 19
Appendices 22
Appendix 1 – Extended information on Egypt 22
Appendix 2 – Data 24
Appendix 3 – Limitations Heckman procedure 25
1. Introduction
According to a report by the World Bank (2012) on gender equality in MENA countries12, the MENA countries have showed a lot of progress in human development, especially in gender equality within education and health. However, the same report shows that only 25.5% of the women in MENA countries join the work force and this participation rate grows slowly; only 0.17 percentage points annually. In comparison with other developing countries with an average female labor force participation rate of 50%, a rate of 25.5% is low. Moreover, the labor force gender gap in the MENA region has doubled over the last 25 years and women face high rates of unemployment. The typical MENA country experiences the low labor force participation,
although the gender gap within education and health care has been reduced and the number of educated women is higher than ever. In most MENA countries this has not translated (yet) into equal roles for men and women in the labor markets (World Bank, 2013, p.4). So why has the increased human capital not yet contributed to an increase in female labor force participation like in the rest of the world?3
The essence of the MENA paradox is the question why the great achievements in human capital do not seem to have contributed to increased female labor force participation in the MENA region. Why is it important to study the MENA paradox? Gender equality in labor force participation is important, for both intrinsic and instrumental reasons. The intrinsic reason is that gender equality – equal rights – is one of the basic human rights. The instrumental reason is that gender equality is a contributor to economic development (World Bank, 2013, p. 3). Ideally, solving the MENA paradox -‐ and increasing gender equality -‐ would address basic human rights and stimulate growth in the MENA Region. Hopefully explaining the MENA paradox will help identify and eliminate possible barriers that currently prevent women from participating the labor market to the same extent that men do. Moreover, explaining the MENA paradox will provide a better understanding of dynamics around gender equality and female labor market participation.
This thesis investigates if, and to what extent the MENA paradox can be explained with the returns to education.
In the empirical research data will be used from the Egypt Labor Market Panel Survey (ELMPS, 2006) conducted by the Economic Research Forum and the Palestinian Bureau of Statistics. This is the second round of the longitudinal survey, containing 37,140 observations.
1
‘MENA countries’ and ‘MENA region’ are interchangeably used.
2
The MENA region: Algeria, Djibouti, Egypt, Iran, Iraq, Jordan, Lebanon, Libya, Morocco, Syria, Tunisia, West Bank and Gaza and
YemenThe data contains questions on individual characteristics, education, employment and income. As the identification strategy the Heckman two-‐step approach for sample selection is used to estimate the returns to education for women and men.
To the best of my knowledge, this thesis is the first to address the specific link between female labor force participation and returns to education in Egypt in light of the MENA paradox4. This thesis contributes by excluding the returns to education as an explanation for the low female labor force participation in Egypt. This outcome opens the door for further research, since a clear consensus about the explanation of the MENA paradox has not been reached. This thesis is structured as follows. Section 2 reviews literature on returns to education and provides previous empirical evidence on the relationship between schooling and education. Section 3 covers the methodology and presents the Heckman two-‐step method as the
identification strategy. Section 4 discusses the data, Egypt’s representativeness as a MENA country and descriptive statistics. In section 5, presents the empirical method and in section 6 the results are presented. Section 7 discusses the results and limitations while section 8 will conclude.
2. Literature review
This section discusses theories concerning returns to education, human capital theory and the Mincerian wage equation. Moreover, this section provides empirical evidence on the
relationship between wages and schooling and the returns to education.
2.1 Human capital theory and returns to education
Human capital is the complete set of abilities and skills, a combination of endowments such as IQ and the capital acquired through investments. Human capital is often described as the economic value of a person, where human capital is viewed as a production factor that creates an output with a certain value. Human capital theory seeks to explain how investments in human capital have effects on productivity -‐ compared to investments in conventional capital – and what the magnitude of these effects is (Schultz, 1961). Blaug (1976) discusses that key to human capital theory is that people base their investment decision on their expected earnings and wealth that will follow from the decision, not on personal short-‐term satisfactions. Hereby foregone
earnings should be taken into account as investment costs. Also, human capital theory assumes that human capital can deteriorate when aging or being unemployed (Berndt, 1991, p. 156).
4 Galal (2001) addresses the paradox of education and employment from the angle of the supply side of education and it’s quantity
Berndt (1991, p. 154) speaks about schooling as an investment and makes the comparison between demand and supply on the labor market, whereas a worker will educate himself when the expectation of future earnings is compensating. The earlier one starts to invest in its human capital, the longer it can benefit from the investment, since earnings tend to grow with work experience (Berndt, 1991, p. 154). Weiss (1995) has a more nuanced view and argues that human capital theory is often associated with the idea that increasing years of schooling directly affects productivity and therefore directly increases wages. However, it is likely that endogenous factors driving the schooling decision affect productivity rather than the years of schooling itself. Card (1999, p. 1802) extends this theory by arguing that although literature often indicates a positive relationship between schooling and wages, one must be careful with causal inferences on the effect of schooling on wages. Moreover, it is the question whether individuals’ earnings are higher because of higher education or whether they are better educated due to greater abilities (Card (1999, p. 1802).
The endogenous factors driving the education decision can cause biases, whereas ability bias is a common example (Harmon et al., 2003, p. 119). Ability theory describes that people with higher initial endowments of human capital tend to be higher educated and receive higher wages, since initially their productivity is higher. However, when omitting ability variables, schooling will be endogenous and performing an OLS regression will cause the results to be biased (Harmon et al., 2003, p. 119). Moreover, an educational degree sends out a signal about a person’s productivity. For high-‐ability people it is easier to obtain a higher educational degree, therefore returns to education for high-‐ability people tends to be upward biased5. It is often argued that studying the returns to education for identical twins rules out ability bias and only captures the effect of education on wages (Harmon et al., 2003, p. 119). However, this falls outside the scope of this thesis6.
2.2 Estimating the returns to education
Mincer (1974, p. 2) argues that people differ in the amount of capital they accumulate and the returns they receive for their investment. He developed the core model for estimating the returns to investment in human capital. Investments, rate of returns and initial endowments in human capital differ among humans and initial endowments are not easily captured (Mincer, 1974, p. 3). The Mincer equation is also known as the human capital earnings function. This
5
See for example Arrow (1973), Spence (1973A), Spence (1973B), Becker (1975), Weiss (1971) and Weiss (1995).
6
See for example Angrist and Krueger (1990), Ashenfelter and Krueger (1994),
model tries to capture the effects of investments in human capital on earnings. Lemieux (2006) explains that the most commonly form of the Mincer equation is the following:
Log γ = log γ0 + rS + β1X + β2X2 (1)
Where log γ is the natural logarithm of earnings, S the total years of education and X the total years of potential market experience. Adding work experience as an exogenous variable is relevant since wages tend to grow with age (Berndt, 1991, p.154). The quadratic term of the potential market experience indicates the diminishing returns to scales of work experience. The Mincer equation forms the base of most empirical models that try to capture the effects of schooling and education on wages. Psacharopoulos and Patrinos (2004, p. 116) discuss that the Mincerian equation estimates the wage-‐effects rather than the returns to education. Moreover, Psacharopoulos and Patrinos (2004, p. 112) find that the average returns to education are the lowest for non-‐OECD countries and the MENA region and this rate is approximately 7%, below the world average of 10%. According to Harmon et al. (2004, p. 150) the world average returns to education are 6% (OLS) and 9% (IV).
2.3 Hypothesis
The hypothesis in this thesis is ‘women have low labor force participation due to low returns to education’. If the returns to education are much lower for women than for men in Egypt, it could explain the low female labor force participation. When returns to education are low for women, it could go two ways. Either the costs of education are extremely high or the payoffs in terms of wages are low or a combination of both.. This thesis will focus on the latter and will investigate whether the returns to education are low enough to explain the low labor force participation. If the returns to education turn out to be low, this could explain the low female labor force participation rate. Low returns to education could imply that schooling does not increase human capital, in that case a diploma will be a piece of paper rather than a reflection of
productivity. In that case, the MENA paradox could be explained with the returns to education and policymakers and researchers should focus on the quality of education and the value of degrees. Highly simplified, if low returns to education for women are found it increases the solvability of the MENA paradox. However, the question why women do not use their acquired human capital remains in case of high returns to education. Possibly women invest in education to promote themselves on the marriage market. Also, exogenous factors could cause the low female labor force participation, such as discrimination by employers. This result would also be contributive since excluding a possible explanation of the MENA paradox is also informative.
3. Methodology
This section will elaborate on the methodology and the empirical model applied in this thesis. Sample selection bias occurs when those employed differ extremely from those who are not participating in the labor market. This is to be solved with a Heckman two-‐step correction model. Since the major challenge in this thesis is to estimate the returns to education for women who do not work (counterfactual), sample selection bias will be accounted for. However,
selection bias is not the only possible threat to estimating the returns to education, ability bias and signaling are possible threats as well. People with higher initial endowments of human capital earn more on average (ability bias). For people with higher ability it is easier to obtain a higher degree, by doing so they send out a signal of their high productivity (signaling). Applying an instrumental variables approach could solve these possible biases if a credible instrument would be available. Due to the lack of a good instrument and the scope of this thesis, these biases will not be accounted for.
3.1 Heckman two-‐step method
Heckman’s (1979) two-‐step model for sampling selection is applied. This model acknowledges possible sample selection, data truncation and provides a method to overcome the problems through Ordinary Least Squares Method (OLS). The Heckman two-‐step method was originally designed to overcome the problem of sample selection, specifically when data is truncated. The method was the answer to sample selection problems and Heckman (1979) introduced the case of estimating the wages of women who are not employed.
If one were to estimate the wages of women directly then it would only estimate the wages of women who are actually in the labor force. However, one cannot automatically say that these wages are representative for those who do not work. Since being employed could be the result of an endogenous, not observed decision. One example is self-‐selection: women who have a higher reservation wage are less willing to work for a certain wage than women with a lower reservation wage. Also it is possible that women not participating in the labor market have a lower productivity that would logically correspond to a lower wage. Solely performing an OLS regression would lead to inconsistent estimates of the coefficient on wages (Greene, 2003, p. 783).
Heckman acknowledges the possible problem of sample selection and treats the
endogenous driver of the unemployment decision as omitted variables bias. In short, in the first step the probability of a woman being employed is estimated. In the second step the wages of women are estimated. Since the wages are only estimated for those women who are in the first
step considered to be in the labor force, the Heckman correction term is added in the second step. Basically the Heckman correction term is no more than an ‘instrument’ for the omitted variable(s): the endogenous drivers of the decision to be unemployed. Adding the Heckman correction term overcomes the problem of sample selection and allows the wage equation to be representative for all women, employed and unemployed. Moreover, treating the sample
selection as omitted variables bias is that it enables to use Least Squares Method in step 1 and 2. Another benefit of using Heckman two-‐step model over other sample selection models (e.g. ML) is that the Heckman two-‐step model estimates the standard errors correctly.
3.2 Heckman’s two-‐step model applied
Heckman (1979) introduced the case where women’s endogenous employment decision is depending on their reservation wage.7 When the actual wage exceeds the reservation wage, people will enter the labor market. But when the actual wage is lower than the reservation wage, a person chooses not to be employed leading to unemployment and a wage of zero. However, the reservation wage cannot be observed, so another in the model exogenous variable is used to estimate the probability of working.
Step 1 – the selection equation:
W*i = ziγ + ui [Wi=1 if W*I > 0 and Wi=0 otherwise] (1)
• W*I = [yI -‐ y*I]
• Prob (Wi = 1|zi) = Φ( ziγ) • Prob (Wi = 0|zi) = 1 -‐ Φ( ziγ)
The underlying assumption is that employment (Wi) is determined by the difference (W*I) between the actual wage (yi) and the reservation wage (y*i). Wi is a binary variable whereas value 1 indicates person I is employed and 0 otherwise. Ziγ is a vector of exogenous variables determining W*I and the exogenous variables from the regression equation (2). Ziγ indicates whether the actual wage exceeds the reservation wage and thus whether someone will be employed (Wi=1). The selection equation estimates the difference between the actual wage (Wi) and reservation wage (y*i), whereas a female is considered to be employed (Wi=1) if the
reservation wage W*i > 0.
The selection equation basically provides an estimate of the probability that a person will be employed (Wi=1).
Step 2 -‐ regression equation:
yi = xiβ + [Heckman correction term] + εI [Only observed if Wi=1] (2)
Xi is a vector of individual endogenous variables that determine the wage (yi). The regression
equation is only estimating the wages for those who are in the sample, i.e. those who are employed (Wi=1). Key to the Heckman procedure is that the latent endogenous variable W*I is not included in the regression equation. Instead the sample selection that occurs in step 1 is considered as omitted variables bias is modeled by the Inverse Mill’s Ratio. It is possible to do so, since the Mill’s ratio follows the property of the truncated normal distribution. The omitted variables problem is modeled by the Heckman correction, whereas the Heckman correction is ρεu • σε • λI •(-‐ ziγ). An implication of the model is that coefficients are consistent and
asymptotically normal. By applying the Heckman two-‐step method, the coefficients on xi in the regression equation can be interpreted for both employed (Wi=1)) and unemployed (Wi=0) women. The Heckman two-‐step model has its limitations and threats, e.g. the estimator can be inconsistent (when working with small samples), model is dependent on its assumption and high rate of censoring causes inefficiency. See appendix 3 for more information.
The use of a selection variable is key to the Heckman two-‐step procedure. The selection variable allows selecting the counterfactual observations and therewith taking care of the sample selection problem. Without the selection variable, one would estimate the functional form in the selection equation, which is basically an OLS regression. Only estimating the functional form and omitting the Heckman correction would lead to inconsistent estimates of the coefficient on wages (Greene, 2003, p. 783) and does not allow to take care of the sample selection problem. However, one must be aware that in the end the representativeness and credibility of the estimates highly depends on the credibility of the selection variable.
4. Data and descriptive statistics
This section discusses the data and sample, the representativeness of Egypt for the whole MENA region and the descriptive statistics.
4.1 Data and sample
The Egypt Labor Market Panel Survey (ELMPS) 20068 contains almost 40,000 cases and is the second round of the ELMPS longitudinal surveys and is conducted by the by the Economic Research Forum and the Palestinian Bureau of Statistics. This is a follow-‐up survey on the households that were interviewed in the first round (1998). This longitudinal survey is
considered to be representative for the Egyptian population, which is of great importance to the credibility of the estimates on the returns to education (Psacharopoulos and Patrinos, 2004). To ensure the representativeness of the second round, a refresher sample of 2,500 households is added9. A total number of 8,349 households is reached in the 2006 sample. The dataset contains a large set of questions on individual characteristics, education, income and employment, which allows estimating the returns to education. The questionnaire includes household, individual and community level questions. All household and community level observations are linked to the individual level through ID-‐coding. All household members above the age of 6 are
interviewed, however, the question on marital status is asked only to females and males above respectively 16 and 18 years old – the legal marriage ages. Therefore, the ages range between 16-‐82 (males) and 18-‐90 (females) for the observations in the empirical research.
4.2 Representativeness of Egypt as a MENA country
Since this research is using data from Egypt only, this paragraph will briefly discuss whether Egypt is a representative MENA country10. Comparing Egypt to MENA shows that female labor force participation, life expectancy at birth and literacy rates are similar. However, the GNI of Egypt is approximately 15% lower than the average of MENA. School completion rates are higher in Egypt than in the whole MENA region, but this is probably due to the fact that primary school is compulsory in Egypt and attending a governmental school is for free (Herrera and Badr, 2011). So there is a large incentive to attend primary education in Egypt. The percentage of females holding a seat in parliament, which is often used as a proxy for gender equality, is 8 times smaller in Egypt than in MENA. However, compared to previous years, this rate has decreased and could be a result of the Arab Spring tensions.
For the MENA region, it is not an exception when both higher and less educated people compete for the same job, which makes it harder for those with middle or lower education to find a job. According to Bardak et al. (2006) the number of higher educated women in the MENA
8 A third round from 2012 is available but not used because of the Arab Spring events.
9
The difference comes from households that have split between 1998 and 2006,
10
See appendix 2 for further information, statistics and tables
region that are willing to wait for a challenging job is increasing. This happens mostly in the public sector, since women in the MENA region hardly enter the private sector (Bardak et al. 2006, p. 16).
In conclusion, the facts indicate that Egypt shows the pattern that is typical for a MENA country; the female labor force participation is low, while educational and health measures indicate great progress over the last decade.
4.3 Descriptive statistics
Table 1 presents the descriptive statistics. The average age in this sample is approximately 26.5 years old, which is quite low. However it is not uncommon for a MENA country since 30% of the population in the MENA region is below 14 years old, only 5% is older than 6511. On average 42% of the women are married which is slightly higher than 42% of men. A possible explanation for the difference is that polygamous marriages for men are legal. In 1981 Egypt introduced a law stating that education is compulsory up to preparatory education, equal to 9 years of
schooling. The mean years of schooling in the sample concentrates around 10-‐11 years, which is -‐ as expected – higher than 9 years. However, regardless of the compulsory schooling law, only 70% in the sample has at least attained preparatory education12. This implies that – assuming one starts school at age 6 – schooling is compulsory until the age of 15. The rate of scholars dropping out early (before age 15) has not decreased since 1981, this suggests that the introduction of the law has not prevented early dropouts ever since.
Variable Female Male
Basic variables N=18,553 N=18,587 Mean age 26.97 (19.61) 26.31 (19.22) Married (fraction) 0.42 (0.49) 0.40 (0.49) Education
Mean years of schooling N=7,477 10.43 (4.46) N=9,222 10.88 (4.76)
Highest education level attained13
Lower education Intermediate education Higher education Cumulative 0.373 0.771 1.000 Cumulative 0.375 0.747 0.999* 11
Source: http://wdi.worldbank.org/table/2.1
12See appendix 3.
13
Lower: nothing, primary, preparatory, Intermediate: general secondary, technical secondary 3-‐years, technical secondary 5-‐
years, Higher: Above intermediate, university, post graduate.Labor LFP (fraction) N=3,544 0.57 (0.50) N=10,465 0.81 (0.40) Income
Mean hourly wage in Egyptian pound (EGP) N=1,501 3.36 (8.94) N=5,171 3.76 (9.61) Table 1
* Rounding difference ** see appendix 3
The average hourly wage for women in Egypt is 3.36 EGP; the hourly wage is slightly higher for men (3.76 EGP). This could also be a result of the low female labor force participation, where females that do participate on the labor market are often higher educated and receive a relatively higher wage (IMF, 2013, p. 9).
The labor force participation for women in this sample is 57%, which is conflicting with the rate of 24% reported by the World Bank over 201214. However, the gender gap in labor force participation is large (57% against 81%) while the gender gap in total years of schooling is small15.
5. Empirical model
This thesis follows the two-‐step model for sampling selection by Heckman (1979) and consists of the following two equations that will be estimated for men and women separately.
LFP*i = γ0 +γ1Si +γ2Agei + γ3Age2i + γ4 Marri + u1 (3)
• LFPi=1 if
γ
0+γ
1S
i+γ
2Age
i+ γ
3Age2
i+ γ
4Marr
i+ u
1 > 0 and LFPi=0 otherwiseEquation (3) is the selection equation and estimates the probability of a person being employed (LFPi=1). Si represents the total years of schooling of individual i and Expi indicates the total
potential work experience in years.16 Marri is a binary variable whereas value 1 indicates that
person i is married and 0 otherwise. Marri is the selection variable, all other variables come from
the regression equation (equation (4)).
It is assumed that the employment decision is based on the reservation wage, i.e. the minimum wage a woman is willing to work for, which cannot be observed directly. Marriage
14
See appendix 1; table A3.
15
Independent group t-‐test indicates difference in mean years of schooling is significantly different from zero.
16
See appendix 3 for definitions and computations of variables
does not affect productivity directly thus it does not directly affect the wage one receives.
However, marriage does affect the reservation wage of women, since married women are part of a household whereas the husband often provides the main income. When the total household income is higher due to the contribution by the husband, the reservation wage of a woman increases. Therefore Marri is included as the selection variable.
Ln Yi = β0 +β1Si +β2Agei + β3Age2i + [Inverse Mills Ratioi] + ε1 (4)
• [Only observed if LFP=1]
Equation (4) represents the regression equation and β1 captures the effect of schooling on
wages. Ln Yi represents the total income from work, including bonuses etc., excluding subsidies or benefits from the government and is calculated as wage on an hourly basis.
Concluding, the sample selection is treated as omitted variable bias, which is on its turn modeled by the Inverse Mill’s Ratioi. By adding the Heckman correction term Inverse Mill’s Ratio, the coefficient on schooling represents the returns to education for all women regardless of their labor force participation status.
6. Results
Table 2 shows the results of the Heckman two-‐step and OLS estimates. The Heckman two-‐step model strongly depends on the correct specification and requires the correlation between the selection and regression equation to be nonzero. A correlation of zero will cause the model to be biased (Guo and Fraser, 2010, p. 124). However, this is considered not to be a problem since both Rho’s are nonzero. Moreover, the significance of the Inverse Mill’s Ratio’s for females and males indicates that the problem should be modeled with the Heckman two-‐step method, since OLS would lead to biased estimates17.
The more counterfactual observations are sampled, the smaller lambda and sample selection becomes less of a problem. By ‘counterfactual observations’ the part of the population is meant that could have been in the labor force – selected by the selection variable. Sigma is by definition larger than zero, rho lies between [-‐1,1] and lambda is the product of sigma and rho (ρεu • σε =λ). A positive rho means that omitted factors are positively correlated with the selection variable and the dependent variable in the regression. A negative rho means that omitted factors are negatively correlated with the selection variable and positively correlated with the dependent variable – or vice versa. The lower the value of lambda, the higher the probability that an observation contains data for the dependent variable (Heckman, 1976, p.
479).
Regression equation Female Male
Heckman OLS Heckman OLS
(logmtotalreal) Schooling (years) Age Age^2 Constant 0.063** (0.007) 0.094** (0.013) -‐0.001** (0.000) -‐2.500** (0.304) 0.050** (0.005) 0.076** (0.011) -‐0.000** (0.000) -‐1.915** (0.194) 0.035** (0.003) -‐0.058** (0.017) 0.001** (0.000) 1.354** (0.351) 0.040** (0.002) 0.051** (0.006) -‐0.000** (0.000) -‐0.866** (0.101)
Selection equation Female Male
Heckman Heckman (LFP) Schooling (years) Age Age^2 Married Constant 0.133** (0.008) 0.234** (0.016) -‐0.003** (0.000) -‐0.908** (0.078) -‐4.704** (0.295) 0.021** (0.005) 0.232** (0.010) -‐0.003** (0.000) 0.526** (0.060) -‐3.392** (0.170) Mills Lambda 0.249* (0.098) -‐0.847** (0.124) Rho sigma 0.37810 0.65776736 -‐0.99934 0.84793435 N 2,294 1,441 6,167 4,709 Censored Uncensored 853 1,441 1,456 4,702 Table 2
** significant at the 1% level * Significant at the 5% level
+ Significant at the 10% level
For females the lambda is low (0.249) and significant and rho is small (0.378). This indicates that unobserved factors that are positively correlated with LFP tend to be positively correlated with the dependent variable, resulting in higher wages. Heckman two-‐step corrects
for this, therefore Heckman two-‐step reports a higher return to education than the OLS estimate. Increasing schooling with one year respectively leads to a 6.3% (Heckman two-‐step) and 5.0% (OLS) increase in the hourly wage. 18
For males the lambda is -‐0.847, which is significant, negative and larger than the lambda for females. This indicates that on average the probability that an observation i has data for the dependent variable is lower for males than for females. The negative rho indicates that
unobserved factors that are positively (negatively) correlated with LFP tend to be negatively (positively) correlated with wages. Since the Heckman procedure corrects for these unobserved factors, it explains why Heckman reports a lower returns to education than the OLS estimation. Increasing schooling with one year respectively leads to a 3.5% (Heckman two-‐step) and 4.0% (OLS) increase in the hourly wage.
By assessing the model specific parameters rho and lambda, it is concluded that the selection equation is likely to be specified correctly for females. However rho for males is almost equal to [-‐1]. Since rho is the correlation between the error terms of the regression and selection equations, there is not much information in the selection variable marriage, that distinguishes between those two equations. Therefore one must be careful when interpreting the Heckman two-‐step estimate for males.
Variable Female Male
Age Schooling (years) Married (fraction) Uncensored N=1,441. 36.13 (10.75) 13.78 (3.55) 0.66 (0.47) Censored N=853 38.07 (15.90) 10.21 (4.86) 0.75 (0.43) Uncensored N= 4,702 35.46 (11.23) 11.77 (4.62) 0.71 (0.45) Censored N=1,465 41.91 (21.78) 10.07 (5.12) 0.51 (0.50) Table 3
Table 3 shows descriptive statistics on the observations in the Heckman two-‐step
estimates. For females, the fraction of people married is higher under the censored observations (75% against 66%). This is consistent with the hypothesis that marriage raises the reservation wage and therefore plays a role in the decision to be employed or not. The results suggest that it is likely that these women would have worked, if they had not been married.
The censored males in the Heckman two-‐step estimate have a lower percentage of marriage (51% censored against 71% uncensored). Possibly marriage plays less of a role in the decision to be employed. It is not expected that being married increases a man’s reservation wage. For both females and males the monthly household benefits are higher for the censored
observations, which raises the reservation wage for both (and is not gender specific). It should be noted that the monthly household benefits include illness support and pension benefits; it is possible that the ‘censored’ people are not unemployed because of reservation wage
considerations but because of the inability to work.
7. Discussion and limitations
The hypothesis in this thesis is ‘women have low labor force participation due to low returns to education’. The female labor force participation is 57% whereas the male labor force
participation is 81%. This large difference is noteworthy since the gender gap in education is very small, men have on average 0.41 more years of education compared to women. The
Heckman two-‐step estimates for the returns to education show that women have higher returns to education than men – 6.3% and 3.5% respectively. This rules out the possibility that a degree is a piece of paper rather than a reflection of productivity; according to the rate of return these women do not lack productivity according to the rate of return.
However, the question remains why women do not put their acquired human capital into practice. There is more than one possible explanation. One is that women invest in education to promote themselves on the marriage market. Pencavel (1998, p. 326) argues that it is well known that there is a positive correlation between the education of men and women. Moreover he argues that the importance of educational degrees in assortative mating has increased since the 1950’s. Assortative mating is the theory that describes how individuals like to marry spouses that are similar to themselves in many aspects, such as education, ethnic background, religion etc. (Lefgren and McIntyre, 2006, p. 789). Furthermore, Anderson and Hamori (2000, p. 230) argue that the potential of an individual translates to a social price that reflects the attraction and future prospects of the individual as a spouse. Lefgren and McIntyre (2006, p. 788) found that half or more of the correlation between women’s education and consumption runs through the marriage market. For women, education is thus an efficient way to increase future prospects. Furthermore, Lefgren and McIntyre (2006) discuss that men with higher income tend to marry women who have the potential of earning a good salary when entering the labor market. The data from ELMPS 2006 shows a significant negative effect of marriage on being in the labor force19. This suggests that being married reduces the probability of joining the labor market. This is in line with the hypothesis that women educate themselves to find an appropriate spouse on the marriage market rather than having the intention of being employed.
According to the International Monetary Fund (2013, p.9) women in MENA countries that participate on the labor market, are often higher educated and receive a relatively higher wage than women all over the world. It is possible that the returns to education are, despite the Heckman correction, upwards biased since the females who are employed have a higher
educational attainment. The World Bank (2012, p.11) suggests that the low female labor force participation in the MENA region could be a result of culture in which marriage is an invisible hand that alters women’s opportunities to enter the labor market. To illustrate; the third round of the Egypt Labor Market Panel Survey in 2012 contains a question why women do not work – for those who do not work. 7.8% of the women that do not work indicate that it is refused by their husband/fiancé, 32.8% indicates that they would rather stay at home to take care of their children. Moreover, 13.6% of the women argue there are no suitable jobs and 17.5% argue that the wages of the available jobs are not suitable. Another 15.9% answered that they did not want to work.
It should be noted that this thesis has some limitations. The outcomes of this empirical research are dependent on the underlying assumptions. One could debate about the reflection of productivity in wages, the relevance and credibility of marriage as a selection variable and the definitions of the variables used in the specifications. Becker (1985) argues that the more time women dedicate to tasks within the household, the less energy they invest in their job compared to men, possibly resulting in lower productivity and salaries. The underlying assumption for using marriage as a selection variable is that marriage does only affect the reservation wage of a woman, not the productivity. So if Becker’s (1985) argument is true, this could be a threat to the use of marriage as the selection variable. However, the wage, that is assumed to reflect
productivity, is calculated on an hourly basis and is therefore not expected to cause biases. By applying the Heckman two-‐step method the estimator can be inconsistent (when working with small samples)20. The model is highly dependent on its assumptions and a high rate of censoring causes inefficiency. The dataset that is explored in this thesis is large, the assumptions are verified (nonzero correlation between the error terms of the two equations) and the rate of censoring is relatively low (between 23%-‐37%). These threats are therefore not considered to cause large problems.
In conclusion, the returns to education do not seem to provide an explanation for the MENA paradox. However, it is debatable how low the returns to education must be to conclude that the returns to education cause the low female labor force participation. Nonetheless, female labor force participation is much lower than male labor force participation even though the returns to education are much higher for females, which is somewhat contradictory. However, it is possible that there are unobserved (endogenous) factors driving the female labor force
participation decision that are not included in this research, but that does explain the
controversy in the proportions between men and women; the returns to education and labor force participation.
8. Conclusion
This thesis investigated whether the returns to education could explain the MENA paradox. The MENA paradox addresses the question why female labor force participation in the Middle East and North Africa is so low, while it should be higher given development indicators in
comparison to similar development regions. As indicated by economic literature, one can estimate the returns to education by applying the Heckman two-‐step model. This model estimates the returns to education as the effect of schooling on wages. This is estimated with data from the Egypt Labor Market Panel Survey 2006, which contains almost 40,000
observations and includes questions on individual characteristics, education and schooling. To account for the possible selection bias that could come from the fact that only wages of women in the labor force are observed, the Heckman two-‐step model corrects for this by estimating who are considered eligible for being in the labor force based on the selection variable marriage. The results show that the returns to education are higher for women than for men (6.3% and 3.5%), whereas the labor force participation rate is of women is lower compared to men (57% and 81%). In consideration of similar total years of schooling for both men and women (0.41 years difference) the result is puzzling. Under the assumption that wages reflect productivity, the theory that education is a piece of paper rather than a reflection of skills and abilities is excluded – which could be the case when the returns to education for women turns out to be low.
However, the returns to education for women are high, so why would women still invest in their education while most of them do not end up in the labor force? One possible explanation runs through the marriage channel; being educated gives females a higher probability of marrying an educated spouse. Also, it could be possible that women are discriminated on the labor market or that it is embedded in culture.
For further research I would recommend to investigate the marriage market theory and discrimination on the labor market. Also I would recommend investigating whether the women that do participate on the labor market are higher-‐educated and therefore earn more, as the IMF indicates However, it should not be ignored that the MENA paradox addresses a wide-‐spread multi-‐dimensional paradox, that can probably not be captured within the field of economics alone.