Changes in educational attainment in Bangladesh, *

Transcription

1 Changes in educational attainment in Bangladesh, * Samer Al-Samarrai alsamarrai@msbx.net April 2007 A background paper for the Bangladesh Poverty Assessment (2007) South Asia Region World Bank * The author is very grateful to Syed Rashed Al-Zayed of the World Bank for providing data on the number of schools across Bangladesh. Barry Reilly of the University of Sussex provided valuable comments and advice on the econometric modelling used in the paper. However the author alone is responsible for all errors and deficiencies.

2 Abstract The paper analyses changes in educational attainment during the first half of the 2000s. Primary gross enrolment rates have stagnated since 2000, although improvements in net rates mean more children of primary school age were attending primary school in Overall, the paper shows that there have been improvements in primary school completion rates since 2000 that were driven largely by increases in enrolment during the 1990s. Enrolment rates amongst the poorest boys have not kept pace with rates for boys in less poor households, however, and even when these boys enter school their educational attainment is poor. A contrasting story emerges at secondary, where there has been substantial growth in enrolment but declining completion rates. This has resulted in rapid increases in the proportion of the population that have started secondary education but failed to complete, particularly among women. Household factors such as income, levels of education, religion and household structure all play important roles in determining final attainment. The paper also finds that secondary school access for boys is constrained by inadequate school places; increasing school capacity is likely to be needed if further expansion at this level is to occur. However, the very low completion rates at this level suggest that more focus is needed on improving grade progression in already existing schools. 2

3 1. Introduction Recent studies have shown that education is strongly associated with poverty reduction in Bangladesh (see for example, Asadullah 2005; Asadullah and Rahman 2005; Al-Samarrai 2007a). For these reasons, education is a key component of the government s attempts to accelerate poverty reduction (GOB 2005). Bangladesh made impressive gains in education during the 1990s when enrolments expanded rapidly and gender differences narrowed. However, the quantitative gains gave rise to concerns about the proportion of children successfully completing primary and secondary education and more generally the quality of education being delivered. What has been happening to the education system since the end of the 1990s and has the impressive educational expansion continued at the same pace? This paper attempts to provide an answer to this question by exploring changes in student access and completion since The paper shows that since 2000 the education system has expanded very little. However, individuals that began their education in the 1990s, when enrolments were increasing, were leaving education in the first half of the 2000s and have therefore brought about increases in the average level of completed years of schooling. Completion of primary and secondary schooling remain key constraints to further education expansion as well as the likelihood that Bangladesh will achieve the education Millennium Development Goals (MDGs). For this reason the paper identifies factors that are important in determining initial access as well as the level of education students finally obtain. Changes in these determinants between 2000 and 2005 are identified to explain the changes in attainment seen over this period. The paper also explores gender differences in these determinants and finds that educational gaps between boys and girls appear to be growing. The paper focuses on the first 10 years of the education system, in line with international definitions of basic education. This is partly to focus on that part of the education system where many of the MDGs, concerning education, are centred. However, it is also the case that owing to the very small size of the post-secondary education sector in Bangladesh the household survey data used does not lend itself to a detailed examination of these higher levels. The next section looks at how the education system in Bangladesh changed between 2000 and 2005 using data from the Household Income and Expenditure Surveys (HIES) from these years. This section provides a first look at some of the major changes in educational attainment during this period. Section 3 outlines the methodology used to analyse the determinants of educational attainment which is followed by a discussion of the results. The results explore the determinants of attainment and how these differ between men and women and also what have been the main drivers of the changes in attainment described in Section 2. The final section offers some conclusions. 1

4 2. Trends in access and completion Before looking at how access to education has changed since 2000 it is important to understand how the education system had expanded prior to this period. The 1980s and 1990s were associated with steady growth in primary school enrolments. According to official statistics primary enrolments increased by 4% annually over this period, increasing from 8.2 million students in 1980 to 16.8 million in 1998 (DPE 2007). Primary gross enrolment rates increased more rapidly in the 1990s owing to faster growth in primary enrolment coupled with a faster decline in the growth of the school aged population during this period (NIPORT et al 2005). At the secondary and higher secondary school levels, enrolment rates remained relatively stagnant until the 1990s when enrolment rose sharply; between 1990 and 2000 secondary school enrolments almost tripled from 3 to 11 million students (BANBEIS 2007). Female enrolment at this level rose at a faster rate than male enrolment, in part due to the introduction of a country wide female secondary education stipends scheme (Khandker, Pitt et al. 2003). In 1990, girls represented only 27% of total enrolment compared to 51% in At the primary level the gender gap also narrowed but more modestly; female enrolment represented 40% of the total in 1980 compared to 45% in Trends in national access to education since 2000 How has access to basic education changed since the late 1990s? At the primary level, gross enrolment rates have remained relatively stable between 2000 and 2005, although net enrolment rates have increased slightly (see Figure 1). Taken together these findings imply that the proportion of over-age children in primary declined over the period. However, net enrolment rates for 2005 imply that over 30% of children of primary school going age are still not attending primary school. 1 Figure 1: Gross and net enrolment rates in Bangladesh, primary secondary higher secondary primary secondary higher secondary gross enrolment rates net enrolment rates Source: HIES (2000 and 2005)

5 As is the case in other countries, enrolment in primary school is correlated with socioeconomic status in Bangladesh. The net enrolment rate for poor students was 61% in 2005, compared to 75% for non-poor students (see Appendix Table 1). While there has been some improvement in the overall net enrolment rate between 2000 and 2005, differences between poor and non-poor students persisted. Furthermore, the gender gap amongst the poorest 20% of households widened considerably over the 5 year period. In 2005, the female net enrolment rate (53%) was only one percentage point higher than the male rate. By 2005, the female rate (62%) exceeded the male rate by 7 percentage points (see Appendix Table 1). This suggests that the poorest boys are increasingly being left behind. At the secondary level there has been a significant rise in enrolment rates over the last five years. By 2005 approximately half of all children of secondary school age were enrolled in secondary school (see Figure 1). The chances of secondary school attendance are seen to be even more strongly correlated with socio-economic status than at the primary level: 80% of children of secondary school age who completed primary in the richest quintile are currently enrolled in secondary, compared to only 36% in the poorest quintile. Enrolment rates at higher secondary are low and appear to have declined since 2000; in 2000 the higher secondary gross enrolment rate was 54% compared to 41% in Interestingly, these declines have been concentrated amongst the non-poor with the largest drops occurring in the richest 40% of households. It is unclear why these drops have occurred, but it is possible that declines in secondary school completion (see below) have affected enrolment in higher secondary. Despite these declines amongst wealthier households, differences in enrolment rates between poor and non-poor households are widest at this level; the net enrolment rate for poor households was 3% in 2005 compared to 12% for non-poor households (see Appendix Table 1). Divisional differences in access Recent estimates of poverty have shown that poverty rates vary widely across regions (BBS 2006). In particular, poverty tends to be lowest in Chittagong, Dhaka, and Sylhet divisions compared to Barisal, Khulna and Rajshahi. It is also the case that between 2000 and 2005 divisions with the lowest poverty rates also exhibited the largest reductions in poverty. Are there similar patterns in enrolment rates? Interestingly, primary enrolment rates tend to be higher in divisions with higher poverty rates. For example, Dhaka has the lowest incidence of poverty as well as the lowest primary enrolment rate. This may be partly picking up the greater labour market opportunities in the more vibrant economies of divisions with lower poverty rates and a greater demand for child labour. Divisional differences in secondary and higher secondary enrolment rates tend to follow a similar pattern, although they do not appear to be correlated with poverty. There are large differences in enrolment rates at these levels. For example, Khulna division has the highest secondary enrolment rate, or 20 percentage points higher than that of Sylhet, which has the lowest. The patterns in secondary enrolment may be 3

6 partly explained by the level of school supply in these divisions: the number of secondary school age children per secondary school in Khulna is 717 compared to 1,352 in Sylhet. 3 Figure 2: Divisional gross enrolment rates, Barisal Chittagong Dhaka Khulna Rajshahi Sylhet Source: HIES (2005) primary ger secondary ger higher secondary ger As Figure 1 showed, average primary enrolment rates have increased marginally since However, in Chittagong, primary enrolment rates actually fell between 2000 and 2005 from 96% to 90% and resulted from a decline in both male and female enrolment rates (see Appendix Table 3). In other divisions while the trend in female primary enrolment rates has been upwards, trends in male rates have been more mixed. In three of the six divisions (Chittagong, Khulna and Sylhet), male primary enrolment rates fell. Therefore, gender gaps in primary and secondary education have generally narrowed in all divisions. 4 In Sylhet, the absolute gender gap narrowed substantially at the primary level. In 2000, male enrolment rates were 15 percentage points higher than female rates, but by 2005 this gap had narrowed to 4 percentage points, and indeed reversed, so that female enrolment rates were higher than male rates (see Appendix Table 3). The narrowing of the gender gap was the result of increases in female enrolment as well as a 7 percentage point decline in male enrolment rates. Schooling status of primary and secondary school aged students What are school aged children doing if they are not attending school? Table 1 shows the schooling status for children of primary school age (6-10). Nearly 20% of this group are currently not going to school, and over two-thirds of these non-attenders reside in poor households. Unfortunately, it is not possible to accurately distinguish individuals that have dropped-out of primary school using the HIES data. 5 However, 4

7 government statistics show that drop-out is high at the primary level; it would therefore be expected that a significant proportion of the group shown in Table 1 as not attending are primary school drop-outs. One in ten primary school aged children are attending pre-primary schools, which is consistent with the common finding that Bangladeshi households typically send their children to primary school when they are older than the official starting age of 6: In 2005 the average age of children in Class 1 was 7. It is also interesting to note that a similar proportion of poor and rich households send their children to pre-primary school. Table 1: Schooling status of primary school aged children in 2005 (%) Not attending Attending pre-school Completed primary and/or attending other schooling Attending primary (NER) Male Female Poorest quintile Richest quintile Poor Non-poor Total Source: HIES (2005) Only 57% of children of secondary school age (11-15) have completed primary school and are therefore eligible to attend secondary (see Table 2). This is a relatively low proportion and reflects the fact that children start primary late: 24% of this age group were still attending primary school. It is also the case that approximately 20% of this age group are not eligible to attend secondary school because of failure to complete primary education. A greater proportion of females in this age group are eligible to attend secondary school than males, reflecting the higher completion rates of girls in primary school (see below). Of those that are eligible to attend secondary school, 11% had not started their secondary education in Therefore, the majority of children who complete their primary schooling continue on to secondary; an important implication of this is that any further expansion of secondary schooling is likely to be constrained by the lack of primary school graduates. Over a half of all non-starters come from poor households, even though poor households make up around 40% of the total population of households. It is also the case that even when poor children begin their secondary education they are more likely to drop-out compared to their richer counterparts (see Table 2). These findings are likely to reflect the much greater costs of secondary 5

8 compared to primary schooling in Bangladesh, and the greater use of private tuition by richer families to facilitate secondary school progress (see Al-Samarrai 2007b). Table 2: Schooling status of secondary school aged children in 2005 (%) Eligible to attend secondary school Of those eligible to attend secondary school Completed secondary Dropped and/or out of attending secondary other school schooling Not started secondary schooling Attending secondary Male Female Poorest quintile Richest quintile Poor Non-poor Total Source: HIES(2005) Using HIES figures it is estimated that there were 8.5 million children of primary and secondary school going age out of school in 2005 (see Figure 3). 6 Approximately 40% of these are of primary school going age. The total number of children of primary age that are out of school fell by approximately 1 million between 2000 and This has been driven primarily by the improvements in net enrolment rates discussed earlier. However, in Chittagong division, the drop in primary enrolment rates already discussed actually led to an estimated rise in the number of out of school children in this group. Figure 3 also shows that 60% of children (6-15) out of school are in poor households. 7 The out of school population is very unevenly distributed across Bangladesh and largely reflects differences in overall population levels and enrolment rates across divisions. Similar to the findings on enrolment rates, divisions with the highest numbers of out of school children tend to have the lowest poverty rates. For example, Dhaka division has approximately 2.5 million children out of school, or almost 30% of all out of school children, even though it has the lowest poverty rate. 6

9 Figure 3: Estimated number of children aged 6-15 out of school, ,000,000 5,000,000 4,000,000 3,000,000 2,000,000 1,000,000 - Barisal Chittagong Dhaka Khulna Rajshahi Sylhet poor non-poor male female age 6-10 age Source: HIES (2005) Notes: Out of school children are defined as children in the 6-15 year age group that are not attending any form of schooling. Educational attainment The previous sections have outlined changes in the patterns of enrolment amongst school aged children since How have these changes impacted on overall education levels of the population? Have these changes translated into higher completion rates? Table 3 shows that there have been some marked improvements in the overall education level of the population under the age of 40. As a whole the average years of education of individuals aged between 16 and 40 has increased by nearly a whole year between 2000 and It is also clear from the table that there is a relatively steady progression of increasing attainment over time; younger cohorts in the table have higher levels of education than the older cohorts. 8 These increases in educational attainment are likely to be rooted in the increases in enrolment seen in the 1990s, when most of the cohorts shown would have been entering primary school. With the stagnation in enrolment rates since 2000, it is reasonable to suppose that future increases in the average years of education of the population will be small, unless improvements in retention are also seen. 7

10 Table 3: Changes in the average years of education, Age group male female total male female total 16 to to to to to Poor Non-poor Total Source: HIES (2000 and 2005) Notes: There is no information in the HIES on the number of years of education completed for individuals that are not literate. To generate years of education these individuals are assumed to have zero years of education. Reported statistics include individuals that were still in school at the time of the survey and therefore have not completed their education. It is also clear from Table 3 that gains in education have not been equally shared between men and women. On average, the increase in women s education levels over the period has been twice that of men. Looking at the earlier cohorts the differences are even more striking. For example, in 2000 the gender difference in the average years of education for 21 to 25 year olds was 2 years, compared to less than 1 year in For women in this age group average education levels rose by 35% or 1.5 years over that period. These are remarkable changes over such a short period of time. Differences in the level of education of the poor and non-poor are very large and have remained relatively stable since 2000 (see Table 3). 9 The smallest gains appear to have occurred for poor males and show a similar pattern as changes in male primary enrolment rates amongst the poorest (see Appendix Table 2). Table 3 does not give a sense of the proportion of children actually completing different levels of education. School completion is important as it demonstrates that an individual has mastered the knowledge and skills associated with a particular level of schooling. For example, successful completion of primary is generally associated with mastery of literacy and numeracy. 10 It is difficult to measure primary school completion rates because children start and finish primary school at different times. As the paper has shown, late starting is common in Bangladesh and 9% of 13 year olds (3 years older than the end of the official primary school age group) are still enrolled in primary school. Figure 4 shows the proportion of each age cohort that has completed primary education. The completion rates shown include children who went on from primary school to complete higher levels of education. Primary completion rates for children aged between 14 and 19 are between 60% and 75%. These are similar to rates reported in national statistics based on reconstructed cohort methods (see for example, DPE 2006). Primary completion rates tend to be higher for females compared to males; for 15 year olds, male completion rates in 2005 were 64% compared to 83% for girls. These gender differences in favour of girls in primary school completion are a relatively recent phenomenon as male completion rates are higher than female rates for individuals over the age of 20 (see Appendix Figure 1). 8

11 Figure 4: Primary school completion rates by age, 2000 and Percentage of cohort that completed primary education Age Source: HIES (2000 and 2005) How have completion rates changed since 2000? The figure also shows that completion rates have improved between 2000 and 2005 for all cohorts of children. For example, in 2000, 62% of 15 year olds had completed primary education compared to 73% in It should be noted that year olds in 2000 began their primary schooling in the mid 1990s, a time of increasing primary school enrolments. Therefore improvements in completion rates will have come about through a greater proportion of children going to school as well as better progression through the system. 11 However, the current stagnation in primary enrolment rates suggests that any future improvements in completion rates will need to come about through improvements in children s progression through the school system. A very different picture emerges at the secondary level where completion rates are low despite the large increases in enrolment rates reported in Figure 1. A major reason for this is the high drop-out rate at this level. Amongst year olds that have completed their education 33% started but did not complete their secondary education in In 2000, the equivalent percentage was only 23% suggesting that a much higher proportion of children are starting but failing to complete their secondary education (see Appendix Table 4). Alternatively, of all the students in this age group in 2005 that started secondary education only 28% completed (see Appendix Table 4). 12 9

12 Figure 5: Secondary school completion rates by age, 2000 and Percentage of cohort that completed secondary education Age Source: HIES (2000 and 2005) At the secondary level, completion rates amongst boys and girls of the same age appear to be changing fast. In 2000, completion rates for females above the age of 16 were much lower than completion rates for boys. For example, amongst year olds that have completed their education, female completion rates were 13% compared to 24% for boys. However, for the younger cohort of students completion rates are similar for both boys and girls. While there are signs, therefore, that the gender gap is closing, completion rates at secondary remain low for both sexes. As expected, completion rates amongst the poor are much lower than for the nonpoor, but the gap in primary completion rates has narrowed between 2000 and 2005 (see Table 4). However a large gap remains and clearly feeds into the very low completion rates of the poor at the secondary level. Only 6-7% of the poor between the ages of 16 and 25 have completed secondary education compared to 21-35% of the non-poor. The completion rate gap between poor and non-poor students has also narrowed at secondary although this has come about from a decline in completion rates amongst the non-poor, rather than a more rapid rise in completion rates amongst the poor. Table 4: Completion rates by poverty status, (%) Primary completion rates Secondary completion rates Age category Poor Nonpoor Poor Nonpoor Poor Nonpoor Poor Nonpoor Source: HIES (2000 and 2005) 10

13 The section has shown that there have been improvements in primary school completion over the five year period. However, completion rates remain low and recent declines have been driven primarily by previous increases in enrolment rather than an improved likelihood of completion once enrolled. Given the slowdown in enrolment expansion, any further improvements in primary school completion will need to come about through improvements in these chances. The section has also shown that secondary schooling completion rates are very low and despite a closing of the gender gap have stagnated since The remaining sections of the paper attempt to analyse the determinants of educational attainment. 3. Data and methodology The aim of the paper is to look at how educational attainment has changed since 2000 and identify the factors that drive changes in attainment in Bangladesh. In particular, the paper is interested in understanding the factors that determine initial enrolment as well as primary and secondary school completion. To identify these factors the paper uses regression analysis. There are a number of issues particular to the modelling of grade attainment that need to be taken into account. Firstly, grade attainment is not a continuous variable and often populations are clustered around certain values of attainment. For example, in many countries a large proportion of the population dropout at the ends of each schooling level (e.g. primary, secondary etc.). Simple ordinary least squares (OLS) regression analysis does not accommodate discrete dependent variables particularly well. Secondly, there are often many observations in the sample that are still attending school. For these individuals, final grade attainment is not known and failing to account for this censoring is likely to lead to biased estimates. Finally, there are often many observations in the sample that have not begun their school careers and although their current attainment level is zero this is unlikely to be the case in the future. 13 Again, if this censoring is not taken into account estimates of the determinants of grade attainment are likely to be biased. In order to address the discreteness of grade attainment and the right-censoring of a significant proportion of the population of interest, a censored ordered probit model is estimated. Suppressing i subscripts for convenience, the desired level of schooling, S*, for an individual i is defined as: S* = X'β + ε where X is a vector of household and individual level characteristics taken to influence the desired level of schooling for individual i, β denotes a vector of unknown parameters to be estimated and ε is a normally distributed error term with an assumed unit variance and zero mean. S* is not observed for any individual. However, for individuals that have completed their schooling the level of schooling attained, S, is observed such that: S=j if θ j-1 < S* < θ j j = 1, 2,, 6 where j is the level of schooling. In this paper, six schooling levels are defined; j=1 refers to no schooling, S=2 incomplete primary education, S=3 complete primary 11

14 education, S=4 incomplete secondary education, S=5 complete secondary education and S=6 post secondary education. 14 The θ s represent threshold parameters and θ 0 = - and θ 6 =. Since S is observed it is possible to infer a range into which the latent dependent variable S* falls. For example, if S=1, S* falls between the threshold parameters θ 0 and θ 1. Under the assumption that ε is distributed normally it follows that: Prob(S = j) = Ф(θ j X'β) - Ф(θ j-1 X'β) These probabilities form the likelihood function for the uncensored observations. Individuals that are still enrolled in school are censored because the final level of schooling they obtain is not known. However, it is known that the desired level of schooling is greater than the observed level of completed education (j-1). 15 The probability that S* lies above the completed level of education is: Prob(S*>j-1)=Prob(S*> θ j-1 )=1- Ф(θ j-1 X'β) For censored observations these probabilities form the basis of the likelihood function to be maximised. The likelihood function for uncensored observations is: 6 [ ( θ - X' β ) - Φ( θ - X' β ] loglu = Φ j j-1 ) j= 1 and for censored observations: logl 6 = [ 1 Φ( θ - X' β ] c j-1 ) j= 1 Adding these two log likelihood functions together provides the log likelihood function to be maximised. 16 While this estimation procedure accommodates individuals that have not completed their education career (right censored observations) it does not account for delayed or late enrolment. Late starting is common in Bangladesh where children are, on average, 7 years of age when they begin primary school even though the official school starting age is 6. In order to ensure that non-enrolment effects are not confounded with the effects of late starting, children below the age of 8 are excluded from the models estimated. Over 85% of Class 1 students are 8 years of age or over which suggests that if children are going to go to school most will have begun by the age of 8. Differences in educational attainment over time and by gender are analysed using data from the 2000 and 2005 Household Income and Expenditure surveys. All the variables used in the analysis are drawn from comparable questions in both data sets and descriptions and summary statistics are provided in Appendix Table 8. The dependent variable in the model outlined in this section is a categorical variable 12

15 detailing the highest level of educational attainment an individual has achieved. Unfortunately, the HIES does not ask respondents who are illiterate and not attending school what their highest level of educational attainment was. Therefore, uncensored observations (i.e. individuals who are not currently attending school) in the first category (i.e. no schooling ) are defined as illiterate. It is recognised that this group will include individuals that attended some primary school and therefore a cautious interpretation of the results is warranted. 17 Of the 14,234 (17,604) respondents in 2000 (2005) contained in the HIES samples 13,979 (16,207) had information on all the variables in the models estimated. 18 Household expenditure per capita is used in the paper as a proxy for household income. Expenditure data is generally more accurately collected in household surveys and more adequately reflects a households permanent income since it smoothes transitory income fluctuations (see Behrman and Knowles 1999). It has also been argued that household schooling decisions are simultaneously determined with the overall level of income and therefore including household income/expenditure as an explanatory variable is likely to lead to biased results (Montgomery, Oliver et al. 1995; Lavy 1996). In order to ensure that the unobservables influencing educational attainment and household spending are independent, the expenditure variable is instrumented using household asset variables. 19 Tests of endogeneity, reported in Table 5, Appendix Table 5 and Appendix Table 6, show strong evidence of endogeneity and therefore justify the use of a predicted measure. 20 Sample weights are used in the regression analysis as the regression analysis is used as a device to summarise the characteristics of the population. Furthermore, the demand functions estimated are reduced form equations and hence are descriptive rather than structural. Therefore, the justification for using weights is no different to the justification for using weights to calculate means and other summary statistics from representative sample data. 4. Results This section attempts to report the determinants of educational attainment and how these differ by gender, as well as identifying the main factors that have led to the changes in attainment outlined in Section 3. To aid clarity, only the aggregated (male and female) estimates are reported in the text while the gender disaggregated estimates are detailed in Appendix Table 5 and Appendix Table Where there are large gender differences in the determinants of educational attainment these are highlighted in the text with reference to the appropriate tables in the appendix. Results from 2000 are reported in Appendix Table 7, and a discussion of the changes in the determinants of educational attainment between 2000 and 2005 is included at the end of the section. Table 5 reports the empirical results of estimating the educational attainment model for 2005 outlined in the previous section. To ease interpretation the reported coefficients are transformed into marginal and impact effects for the continuous and binary variables respectively. In general, these effects report the change in the probability for each education category (e.g. incomplete primary, complete primary etc.) of a unit change in the explanatory variable. For example, Table 5 shows that if 13

16 an individual lives in Barisal they are 5% less likely to have no education compared to an individual living in Dhaka. It should be noted that the total change in the probability across all education categories of a change in an explanatory variable is zero. Table 5: The determinants of educational attainment, 2005 Marginal and impact effects Coefficients No education/ illiterate Incomplete primary Complete primary Incomplete secondary Complete secondary Post secondary Predicted household expenditure (log) 0.47** (0.18) Age -0.09** (0.02) Age squared (0.0006) Female 0.05* (0.02) Urban (0.03) Hindu -0.12** (0.04) Other religion (0.10) Head years of education 0.07** (0.01) Head salary wage earner (0.04) Head daily wage worker -0.30** (0.05) Head not in the labour force (0.04) Female household head 0.21** (0.07) 0.05** Education of spouse of head (0.01) Spouse of head not in -0.14* household (0.06) Birth order (0.02) Number of children -0.07** (0.02) Number of adults 0.10** (0.01) Total upazila primary schools (00s) Total upazila secondary schools (00s) (0.02) 0.18** (0.06) 14

17 Table 5 contd. Marginal and impact effects Coefficients No education/ illiterate Incomplete primary Complete primary Incomplete secondary Complete secondary Post secondary Barisal 0.20** (0.06) Chittagong 0.08* (0.03) Khulna 0.23** (0.04) Rajshahi 0.16** (0.04) Sylhet -0.18** (0.05) Observations 16,207 Endogeneity test 1.27 Instrument validity test 3.33 LR test stat pooled v nonpooled 322** Notes: 1. Robust standard errors in parentheses significant at 10%; * significant at 5%; ** significant at 1% 3. Variable is binary and therefore impact rather than marginal effects are calculated. 4. Endogeneity test is based on Smith and Blundell (1986). 5. Instrument validity test is based on a test of the joint significance of the instruments in a model with the original household expenditure per capita variable (see Section 3). 6. To conserve space, estimated thresholds are not reported but are available from the author on request. Table 5 shows that levels of household income proxied by predicted expenditure per capita impact positively on educational attainment although attainment appears to be relatively income inelastic. A 10% increase in income, at the mean level of income, leads to a 1% decline in the probability that an individual does not go to school and a 1% increase in the probability of post-secondary education. Interestingly the impact of income on attainment appears to be slightly stronger for girls than boys (see Appendix Table 5 and Appendix Table 6). Therefore increases in household income are likely to benefit girls more than boys. Age also appears to be a significant determinant of attainment. Since the model accommodates censoring this finding is likely to reflect a cohort effect; younger cohorts are likely to have higher educational attainment than the older cohorts because of the educational expansion during the 1990s outlined in Section 2. After controlling for the other explanatory factors in the estimated model gender is still a significant determinant of attainment; women tend to have higher attainment than men although the magnitude of this effect is small (see Table 5). It is interesting to note that a study using data from 1996 from one district in Bangladesh found similar results (Maitra 2003). This suggests that a significant gender difference in educational attainment, having controlled for other explanatory factors, has existed in Bangladesh for some time. 15

18 Non-Muslim individuals have significantly lower levels of attainment compared to their Muslim counterparts. For example, individuals from Hindu households are 3% less likely to attend school compared to individuals from Muslim households. Furthermore, these individuals are less likely to have continued their education beyond primary if they did enrol initially (see Table 5). It is difficult to unpick what might be driving this result. It may be that preferences for higher levels of education are more common in Muslim households compared to households following other religions. Equally, the result may be driven by differential access to schooling amongst these groups. For example, post-primary schooling opportunities may be lower for non-muslim groups. 22 Given that non-muslim households make up approximately 10% of the reported age group understanding more clearly the issues behind these findings is an important area for future research. The characteristics of the household head and his/her spouse are also included in the estimated model. In 2005, over 99% of the spouses, included in the sample, were women and the head of the household was the father or mother of over three-quarters of the sampled individuals. 23 Table 5 shows that the education of the household head and his/her partner has a positive and statistically significant impact on educational attainment. For example, the chances of an individual completing secondary school is 6 percentage points higher if their household head has secondary education compared to a household head with no education. This is a common finding in studies of the demand for education and is usually associated with the greater ability of educated parents to assist with their children s schooling as well as recognising its benefits more clearly than less educated parents (see for example, Colclough, Al-Samarrai et al. 2003). It is also often the case that preferences for better educated children may be higher in households with higher aggregate levels of education. Studies that have explored the determinants of initial educational access and attainment have often included mother s and father s education levels separately to allow for differences in their potential impact on attainment. Table 5 shows that the magnitude of the impact on attainment of the household head s education is slightly higher than for the spouse. This suggests therefore that the household head s education plays a slightly greater role in determining final attainment. It is interesting to note, however, that attainment is higher in households where the head is female. For example, the chances of secondary school completion are 2 percentage points higher in households headed by a female. Assuming that women have greater control of resources in female headed households this may reflect a greater preference amongst women for higher levels of educational attainment compared to men, regardless of their level of education. Does the education level of the household head and his/her spouse have a differential impact on male and female attainment? The separate male and female estimates presented in the Appendix show that the household head s education level has a much stronger impact on male attainment, whereas the spouse s level has a larger impact on female attainment (see Appendix Table 5 and Appendix Table 1). These relationships suggest different educational attainment preferences between the head and his spouse for male and female children; more educated spouses are associated with greater attainment of females while better educated household heads are associated with greater attainment of males. Similar findings have been found in other settings (see for example, Holmes 1999; Glick and Sahn 2000). However, the study in Bangladesh 16

19 conducted in 1996 mentioned previously found opposite effects on attainment of mothers and fathers education (Maitra 2003). What might account for these different results? It is possible that changes have occurred over the last 10 years in Bangladesh and girls in households where the mother is educated may have better schooling opportunities in It may also be the case that the relationship between parental education and attainment found in the 1996 study may be particular to the district of study and not representative of Bangladesh as a whole. In contrast to the effect of spouses education on attainment, preferences in female headed households strongly favour male members of the household. A boy living in a female headed household is 7 percentage points more likely to go beyond secondary schooling compared to a boy living in a male headed household. For girls, living in a female headed household increases the chances of continuing beyond secondary by only 3 percentage points. This implies that there is greater male preference in terms of educational attainment in female headed households. This pattern may arise out of a greater need for these households to have a well educated male to provide access to markets and social services. Educational attainment is poorer in households where the head is engaged in daily wage work compared to households where the head is engaged in self-employment. This effect also appears to be quite strong; children in these households are 7 percentage points more likely to have not attended school, and are 5 percentage points less likely to go beyond secondary schooling. Given that household income has been controlled for, this is not proxying for an income effect. It is possible that children in households where the head is a daily wage worker may have to contribute to the household economy in ways which are less flexible (longer hours and more often) than the children of self-employed heads. These demands may prevent them from going to school at all or on a regular basis which in turn reduces their final attainment. This potential explanation fits with the stronger negative impact this variable has on male compared to female attainment. It is also the case that daily wage work tends to be un-skilled work where the benefits to education are likely to be low. It may therefore be the case that preferences for children s education may be lower for household heads engaged in daily wage work than for head s in other occupations. The overall number of children in the household tends to have a negative impact on educational attainment. This is likely to be driven by a greater pressure in larger households to spread education investments (including both pecuniary and nonpecuniary benefits) amongst a larger number of children. This provides some evidence that there is a quantity-quality trade-off in terms of attainment in Bangladesh. 24 The estimates reported in Table 5 also show that earlier born children tend to have higher education attainment than younger members of the household. Looking at the gender disaggregated estimates, this effect is only statistically significant for girls and suggests that households tend to be biased in terms of educational investments towards later born girls (see Appendix Table 5). Combining these results imply that late-born girls in large families will have significantly lower attainment compared to early-born girls in small families. The total number of primary and secondary schools available in the upazila that a household resides in are also included to explore the impact of school supply on attainment. 25 While the supply of primary schools does not appear to have a 17

20 statistically significant impact on attainment, the number of secondary schools in the upazila does (see Table 5). Individuals residing in areas with more secondary schools tend to have a higher probability of continuing to secondary than individuals in upazilas with lower numbers of secondary schools. This mirrors the discussion in Section 2 that reported higher enrolment rates in divisions with higher secondary school provision. Therefore secondary school access does appear to be constrained to some degree by the lack of school places. It is also interesting to note that the supply of secondary school attainment appears only to be statistically significant for boys attainment. This more serious supply constraint for boys may arise from the strong incentives given to schools for enrolling girls. Schools participating in the nationwide female stipends programmes are given tuition payments, by the government, for eligible girls. These payments are often an important source of non-salary revenue for schools. Finally, a set of divisional dummies is included to control for unobservable differences in attainment across Bangladesh s six divisions. The coefficient estimates for these divisional dummies show a similar pattern to the enrolment rates reported in Figure 2. It is surprising to note that while Dhaka division has the lowest incidence of poverty in 2005 it does poorly in relation to other divisions in terms of educational attainment. The exception to this is Sylhet where, holding other things constant, attainment is lower than in Dhaka (see Table 5). This appears to be an entirely female phenomenon; girls are 8% less likely to go to school in Sylhet compared to girls in Dhaka division (see Appendix Table 5). Gender gaps in enrolment rates have also historically been high in Sylhet; in 2000 the male primary enrolment rate was 15 percentage points higher than the female rate. Sylhet has also had high levels of infant and child mortality compared to other divisions in Bangladesh (World Bank and ADB 2002; NIPORT et al. 2005). It is likely therefore, that these findings are reflecting a more general gender bias against women in Sylhet division. More recently, however, Sylhet has made some rapid gains in female enrolment, and gender gaps in education have narrowed significantly (see Appendix Table 3). It is likely therefore that if current trends continue, attainment in Sylhet will begin to converge with the other divisions in the future. Changes in attainment between 2000 and 2005 This paper has shown that average levels of attainment improved between 2000 and 2005 (see Table 3). As Section 2 suggested, this improvement has largely been in an increase in the proportion of the population with incomplete secondary education. For the age group used to estimate the attainment model, similar trends can be found. Using the model estimates from 2000 and 2005 it is possible to gain some further understanding of the reasons for these improvements. Improvements may have come about through improvements in household endowments (e.g. higher incomes, better levels of education) or through changes in preferences for educational attainment. Changes in household endowments can be identified by looking at how characteristics have changed over time, while changes in preferences can be identified by comparing the estimated coefficients over time. 26 Comparison is made between the estimates reported for 2005 (Table 5) and estimates reported for 2000 (Appendix Table 7). 27 The impact of household income on school attainment has declined between 2000 and In 2000, a 10% increase in household income per capita increased the chances 18

21 of going to school by 3% and going beyond secondary school by 2% (see Appendix Table 1. In 2005, these impacts had declined to 1% (see Table 5). Therefore, even though real household income increased over the period its impact on attainment fell. This is likely to explain in part the enrolment and completion patterns seen in Section 2. How do the estimated impacts of parental education compare between 2000 and 2005 in Bangladesh? The impact of the education of the household head appears to have remained relatively stable over the period. However, the level of education of the spouse is only statistically significant in the 2005 estimates (see Table 5 and Appendix Table 7). It is possible therefore that the relatively large impact on attainment of the education of the spouse in 2005, compared to its small and statistically insignificant impact in 2000, is a major driver of the improvements in educational attainment. This effect is further compounded by a small increase in the average level of education of the spouse over the period (see Appendix Table 8). It is possible that this result is being driven by the movement of educated women into higher paid wage employment activities over the same period (see Al-Samarrai 2007a). These opportunities may have increased women s decision-making power in the household, and led to a stronger impact of spouse s education on attainment. The results show a strong intergenerational effect of the education of heads of households and their children. It is tempting to conclude from this that greater investments in education today are likely to lead to even higher education investments tomorrow. However, this assumes that the relationship between parental education and children s attainment remains relatively steady over time. The estimates for the impact of spouses education on attainment detailed in this paper provide some cause for caution in interpreting the results in this way because of the big change in the impact of spouses education over a relatively short period of time. However, taking the results together intergenerational effects do appear to have got stronger since The preference given to earlier born children in the family appears to have lessened over time. This may in part be the reason for the growth in enrolment rates over the last years outlined in Section 2. The negative impact on the number of children in the household, has, however increased over time. For example, in 2000 an additional child in the household increase the chances of never attending school by about 1 percentage point. In 2005, an additional child would increase the probability of never attending school by 2 percentage points. This suggests that aggregate attainment in larger households fell between 2000 and However, over the same period the average number of children in a household also declined. Therefore, it is unlikely that these changes will have had a large impact on the average changes in educational attainment seen over the period. It is also interesting to note the large changes in the impact of religion on educational attainment between 2000 and In the estimated model for 2000, children in non- Muslim households tended to have higher educational attainment than their Muslim counterparts. However, by 2005 a child in a non-muslim household had significantly lower levels of educational attainment. The relative balance between demand and supply factors leading to this turnaround cannot be ascertained from the current data but this is clearly an area where further research is required. 19