Back to school in Afghanistan: Determinants of school enrollment - Stephane Guimbert

ried out by the World Bank in collabora-2.2closer to a better school). A final note is that we use probit, logit, and linear obability model (LPM) specifications. Probit dels are our standard specification, while logit d LPM are used as alternatives—and we found nsistent and similar results. We also use robust ndard errors as data are likely to present teroskedasticity (as often with cross-sectional ta). . Datalower chance of being enrolled (hence it is unlikely that they have migrated just to movemroz and Panjsher. whIn Afghanistan, primary education consists of ades 1–6 starting at age 6, lower secondary It also added a complexity as two new provinces were created,tion on schools, teachers, non-teaching staff as well as some information on students.  National Risk and Vulnerability Assessment (NRVA): the 2003 NRVA survey was carried out by the Ministry of Rural Rehabilitation and Development in association with the Ministry of Agriculture and Animal Husbandry and with support from the World Food Program, the Food and Agriculture Organization, and the World Bank. It covered all 32 provinces which existed at the time and districts within provinces, excluding 11 districts for security reasons. House- hold-level data were collected from 85,577 individuals (43,377 female and 42,200 male members from 11,200 households) in 1850 villages. Most of our analysis is based on a database that combines the last two databases, by assigning district-level data from the SC to individual data in the NRVA. It should be noted that the SC was done in 2004 (1 year later than the NRVA), hence opening up some time discrepancy.5 In addition, we use the PETS database for some of our assessments of descriptive statistics and for simple regressions analysis in the next section. For all three databases, a number of caveats are in order. First, in the absence of a census, a household survey cannot be considered nationally representative. In the case of the NRVA, the sample size somewhat makes up for this issue. For the PETS, the sampling is much better (a stratified sampling based on the SC), but the size of the sample is very small, thus limiting the possibility of segmenting it to refine the analysis. Second, in all three cases, security concerns imposed some restric- tions on data collection. Finally, all three surveys were subject to difficulties in terms of training the enumerators, cleaning up the data, and ensuring comparability of data. 3. School system and drivers of enrollment: descriptive analysisich we had to combine. education consists of Grades 7–9, and higher secondary for Grades 10–12. Education is provided for free at public institutions from Grade 1 until the undergraduate level. Enrollment is heavily skewed toward the lower grades and toward boys, reflecting the disruptions of the 1990s (Fig. 2). The dynamics of enrollment will also imply an increase in secondary education in the coming 2–3 years (World Bank, 2006). Enrollment by age shows that, not surprisingly, in most cases, children are late in the curriculum, i.e. above the normal age level for their grade. In addition, significant disparities, by province and by gender, exist in enrollment (Fig. 3). While universal primary education enrollment is almost achieved in the three main cities of Kabul, Herat, and Mazaar, enrollment remains marginal (below 20%) in the south of Afghanistan, namely, provinces of Uruzgan, Helmand, and Badghes. ARTICLE IN PRESS - 100 200 300 600 1 2 3 4 5 6 7 8 9 10 11 12 13 14 Boys 700 600 500 400 300 200 100 - Girls 400 500 700 Fig. 2. Enrollment by grade and gender in 2004 (‘000 students). Source: School Census (GoA). 20 30 40 50 60 70 80 90 100 F A R Y A B K A N D A H A R C IT Y Boys Girls Average S. Guimbert et al. / International Journal of Educational Development 28 (2008) 419–434 423- 10 K A B U L C IT Y H E R A T C IT Y M A Z A R C IT Y K A B U L T O T A L H E R A T T O T A L B A D A K H S H A N K U N D U Z C IT Y J A L A L A B A D C IT Y B A L K H T O T A L B A G L A N B A L K H R U R A L K A P IS A H E R A T R U R A L K O N A R L A G M A N S A M A N G A N P A R W A N N A N G A H A R T O T A L T A K H A R W A R D A KFig. 3. Disparities in enrollmenK A B U L R U R A L J A W Z J A N P A K T IY A N A N G A H A R R U R A L L O G A R K U N D U Z T O T A L B A M Y A N S A R I P O L G H O R K H O S T G H A Z N I N IM R O Z K A N D A H A R T O T A L P A K T IK A K U N D U Z R U R A L F A R A H N O O R IS T A N Z A B U L K A N D A H A R R U R A L B A D G H E S H E L M A N D O R U Z G A Nt (% children 7–13 years). ARTICLE IN PRESS Table 2 Teaching practices Mean Who teaches double shifts What grade are they teaching 0.014 0.021 0.022 0.062 0.123 2.25* 3 0.137 bust s S. Guimbert et al. / International Journal of Educational Development 28 (2008) 419–434424There are also striking gender disparities; net primary enrollment is only 40% for girls and 67% for boys (Government of Afghanistan (Central Statistics Office) and UNICEF, 2003). 3.1. The supply of education The Government’s vision of the education sector, set forth in a number of policy documents, is to provide good quality education for all regardless of gender, ethnicity, language, religion and geographi- cal location, and to provide opportunities for secondary and higher education at international standard to build skilled human resources which are able to meet private-sector-driven national develop- Age (years) 36.53 Sex (1 ¼ man) 0.72  Years as teacher 8.26 Years as teacher squared Years in this school 4.53  1 if Teach 2 shifts 0.14 Grade taught 5.18 Constant  Number of obs. 21 Pseudo R2 Note: Probit regression on the dummy ‘‘teaches two shifts’’ and ro mean significance at 1%, 5% and 10%, respectively. Source: Authors’ calculations, based on PETS data.ment and reconstruction objectives. At the moment, international assistance provides the financial means to develop quality education in Afghanistan: education is one of the main budget items after security, and the sector benefits from substantial external funding as the Afghanistan Reconstruction Trust Fund (ARTF) largely finances teachers’ salaries. 3.1.1. Teachers There were around 110,000 teachers, according to the 2004 SC.6 This number has increased signifi- cantly over the last 2 years. Only 21% of teachers are female, with an even lower proportion in 6The PETS also include some data on head teachers. They are somewhat older than teachers (42 against 37 years on average, using PETS data) and have a similar proportion of women (around 20%).primary school, and this proportion is as low as 12% in rural areas. Teachers for higher grades seem to be more experienced, though not necessarily older (Table 2). Surprisingly, controlling for other factors, women seem to have a better chance to teach higher grades. The higher share of women in teacher-training schools is an encouraging sign for the future. In all, 53% of primary teachers and 27% of secondary teachers are contract staff.7 More than nine out of 10 primary school teachers have completed only primary education (Table 3). The situation is slightly better in middle schools, notably for permanent teachers. But, in total, less than 50 teachers of primary and middle schools (less than 0.1%) have higher (university) education. The 0.032* 0.125* * 0.5*** 0.012*** ** 0.001 *** ** 4.3*** 213 4 0.168 tandard error OLS on ‘‘highest grade taught’’. (***), (**), and (*)student-per-teacher ratio is around 39:1, which is in line with prescriptions from the Education For All (EFA) framework (Bruns et al., 2003). However this hides significant disparities across the country, ranging from 28:1 to 65:1. Although there are differences among provinces, the PETS shows that most teachers are now being paid on time and in full, a sign of improvements in the public finance management which hopefully has a positive impact on teachers’ motivation: 77% of teachers are paid monthly, 67% of them are paid on time, and 69% did not report any fee paid for receiving salary payments. Further progress is likely to be challenging as recruitment of teachers seems to 7or ‘‘contract’’ staff are government staff hired on fixed-term contracts. ‘‘Karmand’’ are permanent, tenured government staff. See Evans et al. (2004). ARTICLE IN PRESS Table 3 Teachers in primary and secondary schools Primary school Permanent anspo S. Guimbert et al. / International Journal of Educational Development 28 (2008) 419–434 425Total (teachers) 21,469 Total (%) 47 Of which % women 15 Of which % primary level 90 Of which % middle level 5 Of which % high level 4 Of which % higher level 0 Source: based on school census (GoA). Table 4 Access to markets and facilities (rural areas) Food market Nearest tr In the community 5.3 34.9 Less than 1/4 day 46.8 33.5 1/4–1/2 day 27.1 13.4 1/2–1 day 13.8 5.4 More than 1 day 6.2 3.1 Not available 0.8 9.8have outpaced the headcount ceilings and budgetary authorizations available. 3.1.2. Schools There are approximately 6500 government-registered schools in the country. Two-thirds are primary schools. Islamic schools account for a very small proportion of enrollment, which might reflect two factors: (i) a number of these schools are not registered with the Government and hence are not in these data; and (ii) in many cases, mullahs teach religious studies to children either before or after public school, or during the summer or winter break (Hunte, 2005). It is also known that a number of informal education mechanisms exist in Afghani- stan (World Bank, 1999; Hunte, 2005). There are also private school as well as some NGO schools (Hunte, 2005), but their number remains small.8 Total 100.0 100.0 Source: NRVA (2003). 8An educated guess would suggest that private and NGO schools represent 3–5% of the total number of schools, probably even less in terms of the total number of students. A survey by UNICEF and MoE suggested that in 2002, of all ‘‘learning spaces’’, 31% were informal (NGO; community managed; Mosque managed; home-based).Middle school (*) Contract Permanent Contract 23,757 19,447 7296 53 73 27 14 34 19 93 55 81 2 44 18 4 1 1 – 0 – rtation Primary school Secondary school 46.4 11.6 31.6 32.6 6.3 9.1 1.6 3.1 0.6 1.1 13.7 42.5In rural areas, almost a half of households (46%) have a primary school in their community (Table 4). Overall three-quarters of households have a primary school accessible within a few hours away. This is better than access to food market or even transpor- tation. Nevertheless, for young children, this can represent a significant barrier if they have to travel more than 1 4 day twice a day. On the other hand, access to secondary schools is much lower, with less than 12% of households having a school in their community and 43% stating that they have no access to a secondary school at all. There is unfortunately very little data available on the physical conditions of these schools. Some of them are in fact little more than a tent for the teacher and the pupils to gather in. The SC has some data on reconstruction, but there is significant underreporting, with around a quarter of the schools not reporting. Setting aside these measure- ment errors, the Census suggests that some 29% of schools are or have been under reconstruction (reconstruction includes construction in the case of schools upgraded from a tent to a building). The proportion is greater for higher levels of schools 100.0 100.0 (55% for higher schools against 20% for primary schools). A correlation analysis at the district level (Table A1 in annex) suggests that reconstruction activities have been more prevalent in Pashtu- speaking areas, but they are not correlated with are approved. The PETS shows that most non- salary expenditures are covered by non-government ARTICLE IN PRESS S. Guimbert et al. / International Journal of Educational Development 28 (2008) 419–434426higher enrollment (see the next section on regression analysis). 3.1.3. Budget and management Education (excluding higher education) accounts for almost 20% of total operating expenditures in the core budget in Afghanistan.9 However, alloca- tion for non-salary expenditures is low and uneven: in 2004/05, 81% of salaries were spent in provinces but only 67% of non-salary expenditures, with even larger disparities in terms of budget allocations (World Bank, 2006). Often funds do not reach the schools. At the provincial level, half of the provincial education departments indicated that the Mustoufiat (the provincial office of the Ministry of Finance) did not release funds, and a quarter indicated that the central MoE did not provide for an allotment (the quarterly release of funds). The second point is ironic because the central MoE does not manage to spend its allotment in full, while the provincial departments usually spend it in full and would need more. In fact, provincial departments have little say in the planning process: almost half of the surveyed provincial education departments did not even prepare an annual budget for non-salary expendi- tures that would have indicated their needs to Kabul (the situation was much better for the planning of teachers). The fact that the Ministry and its provincial departments hardly demand inputs for the budget process and hardly allocate funds to the service delivery units is also evident at the school level. According to the PETS, only half of the schools had submitted a budget for teachers and a low 6% had submitted a budget for non-salary expenditures. The same happens during the year, with schools barely asking for non-salary expenditures. This is not surprising given that, even with this low level of requests, the estimated time for approval ranges from 1 to 12 months and only 33% of the requests 9The ‘‘core budget’’ includes all revenues and expenditures flowing through the Government’s systems (its bank accounts; using its procurement requirements; etc.). Part of this budget is financed by donors. This is opposed to the ‘‘external budget’’,which is entirely controlled by external donors.organizations or when they are covered by the government they come in kind from the district or provincial department. This seems to be the main external contribution, as most schools reported that they collected no fee (only contributions through PTA, see below).10 Most schools (94%) reported being visited for supervision, a practice to build on to improve the quality of education (PETS). These are reasonably frequent visits (weekly for 15%, monthly for 25%, and quarterly for 37%). The supervision is often administrative (36% to assess teachers’ attendance, 17% to review financial records, 11% to review students’ attendance, 8% to review the school facility) but can also be pedagogical (17% to assess teaching practice). Finally, around 75% of schools reported having a Parent Teacher Association (PTA) and a School Management Committee (SMC). Most of these bodies were reported as contributing to school activities. In 9% of the schools, the SMCs were also reported as assisting in raising funds for schools. 3.1.4. Quality Very little is known about the quality of education. In NRVA (2003), parents were asked about the main problem at school: lack of school books and supplies was the main issue raised (38%), together with poor quality of facilities (25%), lack of teachers (15%), and poor teaching (11%). In Human Rights Research and Advocacy Consortium (2004), a rough estimate suggests that ‘‘74% of girls and 56% of boys drop out of school by the time they reach Grade 5’’. Indeed, input indicators (such as teacher training, availability of non-salary budgets, textbooks, etc.) suggest that quality is poor. There are also concerns about the pedagogical approach, including excessive reliance on memorization and rote-learning, and about the outdated curriculum, including the text- books (World Bank, 1999). 3.2. Demand side 3.2.1. Children We now turn to the demand side of our analytical framework and first start with the characteristics of 10It should be noted, however, that these data come from a school and teacher survey, not from a parent survey, and hencethere is a potential underreporting bias. the children themselves. First, as discussed above, a key issue is that most children are already late in the curriculum for the intended age/grade. Second, disability is a constraint that affects 2.5–3.0% of 7–13-year-old children (Government of Afghanistan (Central Statistics Office) and UNICEF, 2003). Child labor is also a key constraint, affecting 11% of children aged 6–18 years in rural areas, with significant disparities across households.11 A min- ority (about a third) of working children still go to ments are increasing with income, but primary enrolment is consistently higher and more evenly distributed than secondary enrollment. In both cases, the enrollment differential between the richest and poorest is largest for boys than girls. The primary enrollment rate is different only for house- holds on the extreme of the expenditure distribu- Pashu-speaking schools tend to have: (i) lower ARTICLE IN PRESS S. Guimbert et al. / International Journal of Educational Development 28 (2008) 419–434 427school. The issue is more acute for older children. According to the Government of Afghanistan (Central Statistics Office) and UNICEF (2003), this number is slightly lower in urban areas (7%) and higher for boys than for girls (11% against 5%, for a national average of 8% of children 7–13 years that have worked at least half a day for income). In addition, 17% of children have domestic chores, as high as 23% of girls (11% of boys) and 19% in rural areas (children 7–13 years that did domestic chores worked at least half a day, Government of Afghanistan (Central Statistics Office) and UNICEF, 2003). 3.2.2. Households Household characteristics are also likely to impact upon enrollment. In rural areas, the average size of a household is 7.3 (NRVA). For 88% of households, the household head is a man; in 98% (s)he has attended school but only in 25% of the cases does (s)he know how to write and read. Even if, as pointed out in the 1990s by Nancy Dupree, ‘‘a great respect exists for the learned and for books’’ (World Bank, 1999), international experience sug- gests that literacy of the parents is an important factor. In many countries, the household level of income or expenditure has an impact on enrollment. To measure this, we use a notion of welfare based on consumption.12 Both primary and secondary enrol- 11In the NRVA, child labor was defined as having worked for a pay over a recall period of 7 days. 12A vector of food consumption for 11,227 households was obtained from the NRVA household survey. This was then multiplied by a fixed vector of prices (a national median) to obtain a measure of expenditure on food that is the closest approximation to money metric utility available. Non-food expenditure on education, medicine, clothing, taxes, fuel, and oil was then obtained from 5559 wealth group surveys within villages. This was an estimate of the typical amount a household in one of three wealth groups in that village would spend on these items and was estimated by a gathering of household membersfrom that wealth group (who for convenience were often laterprimary enrollment; (ii) higher student per teacher ratios, but lower student per permanent teacher ratios; (iii) a smaller proportion of female schools and female teachers; but (iv) a higher proportion of schools under reconstruction (see above). Access to school (cf. above and Table 4) also has a child, household, and community dimension. Depending on the community, the cost of going to (footnote continued) chosen as sample households). Therefore there were between two and three data points which had the same non-food expenditures in each village. A total expenditure measure was then calculated by summing food and non-food expenditure. We thank David Atkin and Philippe Auffret for providing us with these data. 13This refers to the primary teaching languages. The other 2,375,000 students are in schools with mixed languages or othertion. On the other hand, the more significant difference in secondary enrollment between the first and last deciles (4% and 10%, respectively) suggests that income might be a more important constraint at this level. This is not to say that the constraint is the direct cost of education, as the cost could also be about the opportunity cost of older children not working. 3.2.3. Communities Finally, there are various aspects of the commu- nities, including their history and traditions, which will affect enrollment. Most of these aspects are unobservable, in a statistical sense, not in an anthropological sense. The local language is one of the factors that can be measured and it can be linked in broad terms to specific cultures. It appears that the Pashtu-speaking schools (657,000 students), compared to the Dari-speaking schools (1,394,000 students),13 have: (i) lower enrollment overall and notably for girls; (ii) lower enrollment in middle schools, probably reflecting a more limited pool of student 4 years ago; (iii) higher enrollment in Islamic schools; and (iv) higher student/teacher ratio (but slightly lower student/school ratio) (based on the SC). The correlation analysis (Table A1 in the annex) suggests that districts with a majority ofprimary languages. sch of mi no by De dir of bin nit the (in tha ho 4. fir en ea eq  somewhat lower than those of age 7–11; after 12      ARTICLE IN PRESS S. Guimbert et al. / International Journal of Educational Development 28 (2008) 419–434428years of age the likelihood of going to school declines quickly. This reflects a mix of security- related and generational issues, that is, older children would have had to start school during the civil war or the Taleban period. It is also possibly linked to the issue of child labor which is difficult to assess for econometric reasons (see next point). Being a boy increases the likelihood of going to school by almost 40% compared to girls. Physically disabled children have a 14% lower probability to go to school, while mentally disabled children—who are much fewer—go to school 20% less often. As discussed below, these results on age and gender reflect a mix of issues related to tradition and cultural norms, as well as lack of availability of girls’ schools and female teachers.  Given the endogeneity problems, we analyzed the impact of child labor in a separate regression. Children who have been working and being paid are less likely to go to school. But the coefficient is quite small (13%), reflecting that many children who work also go to school though thispending on the household, these costs (the ect cost of transport and the foregone income having a child in school) will be more or less ding. Households will also look at their commu- y’s behavior as a norm (Hunte, 2006). Finally, se costs will also vary with gender and age Hunte, 2005, a barrier for older girls is t ‘‘they are too old girls to go out of the usehold’’). Empirical estimates in rural areas We now turn to our regression analysis. Let us st summarize the main correlates of primary rollment. (Table 5 shows the marginal effect of ch variable on enrollment.) Everything else being ual, this analysis suggests the following: Among child characteristics, age and gender are very important. The likelihood of going to school for the very young (6) is, everything being equal,isool might change, with different risks (including kidnapping) and costs (e.g. transport). There ght also be social or ethnic barriers as well, tably for girls walking alone and being teased young boys (in Hunte, 2005, ‘‘we are Pashtun’’ presented as a reason for not going to school).could have an impact on learning achievements.higher in the Eastern and Southern regions. Kuchis (the nomadic people in Afghanistan) have a clear disadvantage (likelihood 41% low- er). Pasthu speakers are less likely to be enrolledimpact on enrollment. Migration status has little impact on enrollment. The only exception seems to be that households that have moved from one rural place in Afghanistan to another are less likely to send their children to school. This could reflect that households have moved to find work or because of security reasons, and are hence less likely to send their children to school. Contrary to a finding by Hunte (2006), our results suggest that, in rural areas, returned refugees do not have a higher likelihood of enrolling in school. Household environment also counts in the enrollment decision: children with parents who can read or write are more likely (5%) to school. Children in a household with a radio are also more likely to go to school (8%), while TV has no impact. The community environment counts as well. Even taking into account many other factors, enroll- ment depends on the region, with the probability of enrollment lower in the Western region andanalysis leads to this results—on the contrary, assuming that no answer means no enrollment would lead to a significant and positive impact of income (poorest people go less to school, see column 10 in the annex). There are other factors that are related to the economic environment and available opportunities of the households: or- phans as well as children in a household without regular salary, especially those paid in kind, are less likely to go to school. Participation of the household in some welfare program has noeducation; and (ii) cost might be important, but not a major driver. Nevertheless, this conclusion is subject to some data limits: a large number of the poorest people have not answered the question on enrollment. Our assumption to exclude the individuals with no answer from theAmong household characteristics, size is impor- tant: children in large households have a significantly lower probability of sending their children to school. Income differentials (measured by a notion of household expenditure) have no strong correla- tion to enrollment. This suggests that: (i) all segments of the population place similar hopes in(again, everything else being equal), with the ARTICLE IN PRESS Ta Sum enrollme Va Ch A A B Ho S H A O M rur Co K R L S D Sch D com O A G P N P Me All See S. Guimbert et al. / International Journal of Educational Development 28 (2008) 419–434 429ble 5 mary regression analysis Primarylikelihood of enrollment being around 10% lower than all other speakers. As we note below, this possibly reflects a lack of schools where chil- dren’s mother tongue is used (hence the con-  riables Mean All ild ge 9.70 0.36* ge squared 105.08 0.0 oys 0.50 0.38* usehold ize (number) 8.39 0.0 ead can read/write 0.34 0.05* ctivity other than regular salary 0.08 0.2 0.4 wn a radio 0.51 0.08* igrant: most of the last 5 years in other part of al Afghanistan 0.06 0.1 mmunity uchi 0.05 0.4 egion (base ¼ East-central) Eastern 0.28 0.17* North Eastern 0.03 0.0 North 0.20 0.0 Western 0.13 0.0 Central 0.01 0.05 Southern 0.12 0.12* anguage (base ¼ Dari) Pastho 0.52 0.1 ecurity issues 0.05 0.0 istrict level proportion of Dari speakers 33.9811 0.001 ools istance to primary school (base ¼ within munity) At less than 1/4 day 0.34 0.1 1/4–1/2 days 0.06 0.2 1/2–1 days 0.01 0.3 More than 1 day 0.00 0.3 Not available 0.11 0.5 perated by (base ¼ Government) Mosque 0.01 0.28* Others 0.00 0.28* vailability of teachers (at the district level) Higher teacher/school 0.21 0.02* Share of contract teachers 0.54 0.0 ender issues (at the district level) Share of male schools 0.42 0.0 Share of female teachers 0.06 0.17* ass rate 0.63 0.15* umber of obs. 13,05 seudo R2 27.51 an and regressions on primary enrollment on 4–15-year-old children; re with survey weights. Coefficients for are marginal effects. (***), (**), a details in annex (including variables not reported here). Source: Authont Secondary enrollmentcerned households just state that there is no school available for their child). This phenomenon is further strengthened by the impact of other members of the community on Girls All Girls ** 0.34*** 0.09*** 0.02* 2*** 0.02*** 0.00*** 0.00* ** 0.26*** 2*** 0.03*** 0.00** 0.00*** ** 0.04** 0.03*** 0.00 2 to 9*** 0.15** 0.03*** ** 0.04** 0.02*** 0.01** 5*** 0.06 0.01 0.00 1*** 0.31*** 0.05*** 0.00 ** 0.23*** 0.01 0.00 1 0.02 0.06*** 0.07*** 1 0.04 0.02** 0.01 7** 0.06 0.02 0.00 0.02 0.11* 0.06 ** 0.04 0.05* 0.03 1*** 0.15*** 0.02** 0.01 9** 0.08* 0.03* 0.01 4*** 0.0027*** 0.0001 0.0001 0*** 0.09*** 5*** 0.17*** 5*** 0.24** 9*** 0.28** 2*** 0.34*** ** 0.44*** 0.03 0.00 * 0.13*** 0.03 0.00 0.01 0.01* 0.00 8* 0.15*** 0.08*** 0.03 4 0.21*** 0.01 0.01 * 0.30*** 0.01 0.01 ** 0.28*** 0.00 0.00 2 6024 4921 2103 23.98 49.06 26.43 gression on secondary enrollment on 10–18-year-old children. nd (*) mean significance at 1%, 5%, and 10%, respectively. rs’ calculation, see Annex.      parents, being in a household without regular salary, or being located in a community without a school or in a district with a low proportion of female teachers. Mentally disabled also have a markedly lower likelihood of going to school. Going to a non-government school also has a large positive impact, but it is rare as noted above. ARTICLE IN PRESS S. Guimbert et al. / International Journal of Educational Development 28 (2008) 419–434430enrollment. A higher pass rate in the district increases the probability of enrollment (although it also means that more children stay in school, hence introducing some reverse causality). In terms of gender, a high proportion of female teachers increases the likelihood of enrollment, while the proportion of male-only schools has little impact. In terms of magnitude of the correlates, gender is one of the most influential factors for enrollment. Among the large negative factors are the size ofhigh proportion of contract teachers (presumed less experienced than permanent teachers, even though we saw above that their formal qualifica- tions are similar) reduces the likelihood ofschools in a district (compared to the total number of school in the district) is positively correlated with enrollment. Reconstruction of schools seems to have no impact on enrollment—but, as noted above, the quality of this variable is poor. Enrollment is higher for schools not operated by the Government (likelihood higher by 0.28), even though there are too few of these schools in the sample (1%) to be definitive. Finally, there are several factors that underscore the importance of the quality of teaching. First, aor within 1 4 day. The availability of secondary schools also has some impact on primary enrollment. Similarly, a high proportion of ruralvery important (even if not in the community, it should be at least at less than 1 4 day from the community), although it should be noted that in 80% of cases there is a school in the communityenrollment rate in the district has a positive externality on individual decisions (given the correlation between this variable and other exogeneous variables, we tested this assumption in a separate regression). As noted by Hunte (2005), households take into account their neighbors’ enrollment decisions to make their own decisions. Security is also an important barrier to enroll- ment: enrollment is 9% less likely in for house- holds that have mentioned a ‘‘security incident’’ in the previous year. Finally, the availability and quality of schools is an important driver of enrollment. Proximity iseach household: everything being equal, enroll- ment is higher in districts with a higher propor- tion of Dari speakers. Similarly, a higherthe family, being a Kuchi, having lost one’sIn addition, we performed the same regression analysis on two sub-samples.14 For girls (Table 4), the results are very similar overall to the main regression. The main nuance is the even more central role of quality: the share of male schools in the district, the proportion of female teachers, the share of contract teachers, and the pass rate all have much stronger effect than in the regression with the whole sample. Besides, regional disparities are less pronounced, with only the Eastern regions signifi- cantly different from others (with higher likelihood of enrollment). Some of the environment factors, such as the proportion of Dari speaker in the district, play even more than for boys. Management by non-government seems to have, at the margin, a slightly stronger (positive) impact on girls’ enroll- ment (which comes somewhat as a surprise given that the result applies for schools managed by the mosque, while we have seen above that girls’ enrollment in these schools is low). On the other hand, the source of income (regular salary or not) was less correlated with girls enrollment. Also, the lower likelihood of Kuchis enrolling their children is slightly less marked for girls. We finally looked at the 6–7-year-old age cohort, as these are the children whose progress through the system might have been less impacted by the civil war (since they were not supposed to be in school before 2001). A sign of hope is that the lower likelihood of girls being enrolled is much less pronounced in this subsample (although this might just reflect the tradition factor that makes it more difficult for older girls to leave the household), and 14We also run some sensitivity analysis. First, the main regression was done through different statistical specifications. While probit is the preferred approach, tests with a LPM and a logit lead to largely consistent results. Second, changing some variables led to similar results. We tested the use of dummies for each age, instead of a continuous quadratic function. We also tested subjective wealth categories instead of consumption-based quintiles. We also performed the regressions on smaller samples, as indicated in the text. All these specifications led to similarresults, reducing the risk of a major systematic measurement bias. there are much lower regional disparities. Also, many of the household and community factors have less impact. And as for the previous two regressions, management by a NGO has a strong positive impact (a conclusion that needs to be nuanced by the small sample available for these schools). A complex empirical issue is related to the children, for which we do not have an answer to the question on enrollment. The standard approach is to exclude them from the sample, but this could int so en im to becomes a positive factor (higher income plays a so ru to Th ass ca co Sc ARTICLE IN PRESS S. Guimbert et al. / International Journal of Educational Development 28 (2008) 419–434 431positive role in the enrollment decision): this would be consistent with the assumption that the poorest do not answer the enrollment questions and is likely to introduce a bias. We conclude that income indeed plays a role in enrollment even if our main regression does not show that, but the role is not very large. Second, the Southern region now has a lower than average enrollment likelihood. We also analyzed the reasons given for non- enrollment (in the NRVA). For this part, we restricted our sample to all individuals that are not enrolled and we regressed the reasons that households gave for non-enrollment on the same variables than above (child, household, community, and school characteristics).15 Let us focus on the two regressions related to the two main reasons for non-enrollment:  In 42% of the cases, households claim that enrollment would be ‘‘contrary to a family commitment, the child’s marriage, or their tradition’’. This reason is most often used for girls (age makes little difference, but this is more likely to apply to older children). Literate parents 15The use of these data raises two technical complexities. First, these answers cannot be used in our main specification since households answer them only in cases where the child was not going to school. Hence these data are only available for the sub- sample of non-enrolled children. Second, using them, we must keep in mind that households had to give one answer out of 11 choices, hence these variables are mutually exclusive. As a result, for instance, physically disabled children appear less likely to give the ‘‘family commitment/marriage/tradition’’ answer as an explanation for not going to school, which is likely to mainly reflect that they will give the ‘‘health/disability’’ answer. Similarlytoforroduce a sample bias if non-answer is related to me factors correlated with the decision not to roll. As a sensitivity analysis, we looked at the pact of assuming that non-response is equivalent non-enrollment. Many of our results are robust this change, except mainly two. First, incomemasecurity issues.me of the factors that correlate to enrollment in ral areas. In particular, lower security risks seem have increased the likelihood of going to school. e increase in the number of teachers was ociated with enrollment; however, it is signifi- ntly reduced by issues of quality; for example, the rrelation with contract teachers is less significant. hool construction seems to have had a verytheare less likely to give this answer. So are households close to a school run by a mosque or households in a district with a high proportion of male schools. Households returning from Pakistan are more likely than others to mention this reason. This reason is given more often in the Southern region, less in the North/North East- ern/Eastern regions.  In 30% of the cases, households indicate that there is no school available. This correlates well with our measure of distance to school. Every- thing else being equal, this is likely to be more of an issue for girls, younger children, mentally disabled, and children in richer households. Households that had recently migrated (notably from urban areas or from countries other than Pakistan) are also more likely to raise this issue, possibly due to their high expectations. Kuchis and non-Dari speakers are also more likely to raise this issue, everything else being equal: this probably reflects their expectation to have, in the community or close by, not only a school but also a school teaching in their language. Finally, we looked at secondary enrollment (Table 4). The drivers are very similar to primary enrollment. Surprisingly, everything else being equal, the impact of gender is somewhat smaller than for primary education. As in the case of primary education, income differentials have no significant impact; however, households growing poppy seem to be slightly more likely to send their children to secondary school. The regional dummies have a different profile, with smaller disparities across regions. Distance to schools seems to have a less powerful impact. 5. Conclusions In this article, we have first documented the dramatic increase in enrollment over the last 5 years. While we do not have data to fully ascertain drivers of this increase, this paper identifiesrginal relation to enrollment. But, overall, our ARTICLE IN PRESS T a b le A 1 C o rr el a ti o n m a tr ix a t th e d is tr ic t le v el P ri m a ry en ro lm en t ra te M id d le sc h o o l en ro lm en t ra te % D a ri sp ea k in g p eo p le % P a sh to sp ea k in g p eo p le A ll st u d en t o v er a ll te a ch er A ll st u d en t o v er p er m a n en t te a ch er F em a le sc h o o l ra te R u ra l sc h o o l ra te G ir l st u d en t ra te F em a le p er m a n en t te a ch er ra te F em a le te a ch er g ro ss ra te F em a le co n tr a ct te a ch er ra te C o n tr a ct te a ch er ra te H ig h er te a ch er / sc h o o l ra ti o R ec o n st ru ct io n sc h o o l ra te P ri m a ry en ro lm en t ra te 1 M id d le sc h o o l en ro lm en t ra te 0 .1 7 9 7 * 1 % D a ri sp ea k in g p eo p le 0 .2 9 6 7 * 0 .0 0 1 2 1 % P a sh to sp ea k in g p eo p le 0 .3 5 2 3 * 0 .0 2 4 1 0 .6 9 5 7 1 A ll st u d en t o v er a ll te a ch er 0 .0 4 3 5 0 .0 2 1 3 0 .0 9 9 3 0 .1 7 6 0 * 1 A ll st u d en t o v er p er m a n en t te a ch er 0 .0 4 6 3 0 .1 1 9 9 * 0 .1 2 1 0 * 0 .1 4 5 4 * 0 .1 9 8 4 1 F em a le sc h o o l ra te 0 .1 5 7 1 * 0 .0 3 3 1 0 .0 6 7 8 0 .1 7 1 9 * 0 .0 5 0 .0 0 5 1 1 R u ra l sc h o o l ra te 0 .1 1 7 9 * 0 .0 8 2 2 0 .0 5 1 1 0 .0 4 1 2 0 .0 4 4 1 0 .0 7 8 2 0 .1 5 2 2 * 1 G ir l st u d en t ra te 0 .4 1 9 1 * 0 .1 6 2 6 * 0 .2 6 7 1 * 0 .3 7 3 7 * 0 .0 7 1 2 0 .0 2 5 2 0 .6 1 7 1 * 0 .1 5 8 7 * 1 F em a le p er m a n en t te a ch er ra te 0 .1 9 5 5 * 0 .1 1 5 5 * 0 .1 3 1 2 * 0 .1 9 4 2 * 0 .0 5 2 5 0 .2 1 4 9 * 0 .2 8 5 8 * 0 .1 2 0 4 * 0 .2 9 0 1 * 1 F em a le te a ch er g ro ss ra te 0 .2 2 8 6 * 0 .0 4 9 9 0 .3 2 0 8 * 0 .3 3 9 9 * 0 .0 4 6 9 0 .0 6 0 7 0 .4 6 1 6 * 0 .1 0 0 7 * 0 .5 3 3 6 * 0 .3 9 0 7 * 1 F em a le co n tr a ct te a ch er ra te 0 .1 8 4 2 * 0 .0 6 0 6 0 .2 6 1 6 * 0 .2 1 8 3 * 0 .0 3 7 5 0 .0 0 0 6 0 .5 2 4 8 * 0 .0 4 5 0 .4 4 1 6 * 0 .3 3 0 6 * 0 .6 9 8 2 * 1 C o n tr a ct te a ch er ra te 0 .2 9 9 4 * 0 .3 7 3 9 * 0 .0 3 9 5 0 .0 9 1 9 0 .1 3 3 8 * 0 .3 7 9 8 * 0 .2 1 3 7 * 0 .3 0 2 9 * 0 .2 6 7 3 * 0 .2 0 9 1 * 0 .1 0 1 4 * 0 .1 7 8 1 * 1 H ig h er te a ch er /s ch o o l ra ti o 0 .0 4 3 2 0 .0 2 1 0 .0 6 4 7 0 .0 0 8 6 0 .0 3 3 1 0 .0 5 8 6 0 .0 0 6 4 0 .0 6 4 8 0 .0 3 9 1 0 .0 8 2 7 0 .1 9 5 1 * 0 .0 2 0 3 0 .2 2 0 4 * 1 R ec o n st ru ct io n sc h o o l ra te 0 .0 3 5 7 0 .0 2 9 8 0 .0 7 9 1 0 .1 5 4 4 * 0 .0 8 4 2 0 .1 0 5 1 0 .0 0 6 8 0 .0 1 3 6 0 .0 0 0 8 0 .0 2 1 0 .0 8 3 0 .0 1 3 4 0 .2 5 9 9 * 0 .2 5 4 2 * 1 S o u rc e: S C a n d N R V A , a u th o rs ’ ca lc u la ti o n s. S. Guimbert et al. / International Journal of Educational Development 28 (2008) 419–434432 Glewwe, P., Kremer, M., 2005. Schools, teachers, and education outcomes in developing countries. Bread Policy Paper, 9. ARTICLE IN PRESS S. Guimbert et al. / International Journal of Educational Development 28 (2008) 419–434 433regression analysis cannot account for the whole increase in enrollment. This obviously reflects a deeper structural change with the end of major conflict and an increase in the demand for education linked to the sudden change in potential returns to schooling, itself linked to security, large inflows of external assistance, and more broadly to economic growth. We have also described the massive gender gap. Our analysis emphasized the role of female teachers and girls’ schools, as well as the specific challenges of secondary education. Finally, our analysis stresses the importance of quality in education. Although we still measure this poorly, factors such as the share of contract teachers or the share of teachers with higher education tend to have higher correlation with enrollment. To make further progress toward universal enrollment, this analysis suggests a number of opportunities. First, demand for school still remains linked to the household context. There is therefore a role for promoting schooling, notably to reach illiterate heads of households. A possible approach is also to increase management of schools by non- government organizations which have good track record in providing social sector services, preferably education: although our data are too limited to test this adequately, they seem to indicate that in Afghanistan, as in other countries, there is scope for devolving management to local communities, an area where cautious piloting and evaluation could be initiated. Access to school is important, notably for primary education, but reconstruction of schools has a marginal correlation with enrollment. Similarly, it is unclear that more teachers would have a major direct impact on enrollment if teacher training does not expand. Of course, history—in Afghanistan, as well as the experience of other countries in their progress toward universal enrollment—calls for modesty. And beyond this most urgent challenge of universal primary enrollment are two other critical challenges, the quality of education and the provision of secondary education. We have noted some positive features for the secondary level (such as a higher proportion of female teachers and of better-educated teachers), but this is a sector that will need significant resources in the future. Beyond these conclusions, additional research will be needed to better understand the drivers of enrollment and how to target public interventions. In particular, the role and impact of management byGovernment of Afghanistan (Central Statistics Office) and UNICEF, 2003. Moving Beyond 2 Decades of War: Progress of Provinces: Multiple Indicator Cluster Survey 2003 Afgha- nistan. Kabul. Government of Afghanistan, 2006. Interim Afghanistan National Development Strategy, draft. Kabul. Human Rights Research and Advocacy Consortium, 2004. Report Card: Progress on Compulsory Education (Grades 1–9), March 2004. Hunte, P., 2005. Household decision-making and school enroll- ment in Afghanistan: case studies in Chahar Asyab District, Kabul Province; District 13 Pu-i-Khishk, Kabul City; Nesher Villages Belcheragh District, Faryab Province, District 2, Kandahar City. AREU Working Papers, December 2005. Hunte, P., 2006. Looking beyond the school walls: household decision-making and school enrollment in Afghanistan. AREU Briefing Paper, March 2006. Rashid, F., 2005. Education and gender disparity in Afghanistan. Master Thesis in Development Economics, Williams College, Massachusetts. The Guardian, 2006. Fears of lost generation of Afghan pupils asnon-government organizations needs to be further investigated, as this is potentially a key policy to scale up enrollment (and most likely quality). More careful monitoring also is of primary importance. At the household level, the 2005 NRVA will be a major step forward, with a more complete ques- tionnaire and better and expanded—to urban areas—sampling. At the administrative level, there is a need to monitor schools—human resources, performance, management—and to better track resource flows, possibly through a Public Expendi- ture Tracking Survey building on the one utilized in this paper. Appendix A1. Annex: detailed results See Table A1. References Bedi, A.S., Marshall, J.H., 2002. Primary school attendance in Honduras. Journal of Development Economics 69, 129–153. Bruns, B., Mingat, A., Rakotomalala, R., 2003. Achieving Universal Primary Education by 2015—A Chance for Every Child. The World Bank, Washington, DC. Buckland, P., 2005. Reshaping the Future: Education and Post Conflict Reconstruction. The World Bank, Washington, DC. Clemens, M.A., 2004. The long walk to school: international education goals in historical perspective. Center for Global Development Working Paper, 37. Evans, A., Manning, N., Osmani, Y., Wilder, A., 2004. A Guide to Government in Afghanistan. World Bank and AREU, Washington, DC.Taliban target schools. March 16, 2006. WFP, USAID/APEP, Save the Children USA, 2004. Results and discussion of education data collected in the Afghanistan National Risk and Vulnerability Assessment 2003. Mimeo. Wilder, A., 2005. A house divided? Analyzing the 2005 Afghan Elections. AREU Working Paper, December 2005. World Bank, 1999. Education for Afghans. World Bank Work- shops’ minutes. World Bank, 2005. Investing in Afghanistan’s future. Report No. 31563-AF, Washington, DC. World Bank, 2006. Afghanistan: managing public finances for development. Report No. 34582-AF, Washington, DC. ARTICLE IN PRESS S. Guimbert et al. / International Journal of Educational Development 28 (2008) 419–434434