Religious School Enrollment in Pakistan A Look at the Data Tahir Andrabi1 Pomona College Jishnu Das The World Bank Asim Ijaz Khwaja Harvard University Tristan Zajonc Harvard University Abstract Bold assertions have been made in policy reports and popular articles on the high and increasing enrollment in Pakistani religious schools, commonly known as madrassas. Given the importance placed on the subject by policy makers in Pakistan and those internationally, it is troubling that none of the reports and articles reviewed based their analysis on publicly available data or established statistical methodologies. This paper uses published data sources and a census of schooling choice to show that existing estimates are inflated by an order of magnitude. Madrassas account for less than 1 percent of all enrollment in the country and there is no evidence of a dramatic increase in recent years. The educational landscape in Pakistan has changed substantially in the last decade, but this is due to an explosion of private schools, an important fact that has been left out of the debate on Pakistani education. Moreover, when we look at school choice, we find that no one explanation fits the data. While most existing theories of madrassa enrollment are based on household attributes (for instance, a preference for religious schooling or the household's access to other schooling options) the data show that among households with at least one child enrolled in a madrassa, 75 percent send their second (and/or third) child to a public or private school or both. Widely promoted theories simply do not explain this substantial variation within households. 1Corresponding Author: Tahir Andrabi (tandrabi@pomona.edu). This study would not have been possible without the enthusiasm and continuous support we received from Tara Vishwanath. Charles Griffin first encouraged us to look at the data. We thank Veena Das, Shehla Andrabi, Sehr Jalal, Ritva Reinikka and Carolina Sánchez for their encouragement and to Hedy Sladovich for her excellent editorial suggestions. The paper has also benefited from comments by Ismail Radwan, Naveeda Khan, Shahzad Sharjeel and Shanta Devarajan. The research department of the World Bank provided funding for this study through the Knowledge for Change trust fund. The findings, interpretations and conclusions expressed in this paper are those of the authors and do not necessarily represent the views of the World Bank, its Executive Directors, or the governments they represent. Working papers describe research in progress by the authors and are published to elicit comments and to further debate. Contents 1. Introduction............................................................................................................. 3 2. What do Reports and the Popular Press Say? ......................................................... 6 3. The Data About Madrassa Enrollment ....................................................................... 8 3.1 Data Sources ....................................................................................................... 9 3.2 How Many?....................................................................................................... 10 3.3 Explaining Differences ..................................................................................... 11 4. Variation across Districts and Time...................................................................... 13 4.1 Variation across Districts.................................................................................. 14 4.2 Variation across Time....................................................................................... 15 5. Variation across Households................................................................................. 16 6. Taking Stock: What Can we Say about established hypothesis?.......................... 18 7. Conclusion and Caveats........................................................................................ 21 References......................................................................................................................... 24 Appendix........................................................................................................................... 26 Figures and Tables ............................................................................................................ 28 Appendix Tables............................................................................................................... 36 ii 1. Introduction Bold assertions, based on reports and articles, are being made about high and increasing enrollment in Pakistani religious schools in recent years. This "rise" is attributed to an increasing preference for religious schooling among families or the lack of other viable schooling options for the household (Burki 2001, Singer 2001). Given the importance placed on the subject by policy makers in Pakistan and those internationally, it is troubling that none of the reports and articles reviewed base their analysis on publicly available data sources or established statistical methodologies.2 This paper uses established data sources as well as data collected by the authors for a broader study on education enrollment in Pakistan to examine the size and importance of the religious education sector in Pakistan. Our findings differ by an order of magnitude from those reported by and in the media. The madrassa sector is small compared to educational options such as public and private schooling, and accounts for less than 1 percent of overall enrollment in the country. Even in the districts that border Afghanistan where madrassa enrollment is the highest in the country, it is less than 7.5 percent of all enrolled children. Furthermore, we find no evidence of a dramatic increase in madrassa enrollment in recent years. The share of madrassas in total enrollment declined before 1975 and has increased slowly since then. Since 2001 total enrollment in madrassas has remained constant in some districts and increased in others. When we look at school choice, explanations for madrassa enrollment based on household attributes such as religiosity appear inadequate. Even among the less than 1 percent of families who have children enrolled in madrassas, more than 75 percent send their other children to private and public schools. Madrassa enrollment figures cited in the popular press and institutional reports, none of which are substantiated using publicly verifiable data, are sometimes highly inflated. For example, between March 2002 and July 2002, figures for madrassa enrollment cited in The Washington Post tripled from 500,000 to 1.5 million. A report about madrassas by the International Crisis Group (ICG) in 2002 puts madrassa enrollment between 1 and 1.7 million; the report claimed that this number represents 33 percent of all Pakistani children enrolled in schools, when in fact it is less than 1 percent. Similar numbers were reported in other major newspapers and influential publications like the 9-11 Commission Report. 2See for example President Bush's remarks on June 24, 2003, President Musharraf's remarks on November 20th 2003, Colin Powell's on March 11, 2004, Hillary Clinton on February 24, 2004 and the 9-11 Commission Report. The only publicly available sources of data to document patterns of enrollment and available educational options for Pakistani families are household-based surveys. These are the official 1998 Census of Population (Government of Pakistan)3, the 1991, 1998, and 2001 rounds of the Pakistan Integrated Household Survey4, and a 2003 census of schooling choice conducted by our research team. The fact that three sources use different definitions of madrassa enrollment, and were collected at different times by individuals with very different institutional affiliations provides independent verification of enrollment estimates and allows us to determine the sensitivity of our results. The household data tell us whether a child is enrolled full-time in a madrassa, but not whether a child goes for an hour on any given day to study the Quran. Therefore this data does not confound full-time with part-time attendees--a child who attends a public school during the day and a madrassa in the evening is recorded as enrolled in a public school. This is an important distinction since parents might use a modicum of madrassa or mosque based education to teach their children about religion. Consequently, if we contrast these household-based numbers with numbers from establishment-based reports, discrepancies can arise. From virtually any policy perspective, including evening quran classes in enrollment figures seems misguided. Regrettably, until now almost all enrollment numbers cited have been based on establishment surveys which do just that. These data sources show that around 200,000 children were enrolled full-time in madrassas before 2001. Since 2001, our school census suggests that these numbers may have increased somewhat, although the experience varies across districts. To put this number in context, total primary enrollment (grades 1-5) in public and private schools stood at 17.4 million in 2003 (Government of Pakistan, Ministry of Finance, 2003). The choice of madrassa schooling viewed as either the percentage of eligible children or the percentage of enrolled children, is statistically insignificant for the average Pakistani household. Enrollment in madrassas accounts for approximately 0.3 percent of all children between the ages of 5 and 19. Given that the overall enrollment rate for this age group is roughly 42 percent, this represents less than 0.7 percent of all enrolled children, an order of magnitude less than the 33 percent cited by the International Crisis Group report (2002). 3This is complemented with the census of private schools carried out by the Federal Bureau of Statistics in 2000 to provide statistics for private versus public enrollment. 4The PIHS is the equivalent of the widely used Living Standard Measurement Surveys (LSMS) implemented in various countries. See http://www.worldbank.org/lsms for extensive notes on the 1991 PIHS. See also www.statpak.gov.pk for information on the census and the Federal Bureau of Statistics data. 4 The significant differences across regions and trends over time help us understand average madrassa enrollment. In terms of proportions of the enrolled population, the top 10 districts out of a total of 101 lie in the so-called "Pashtun" belt on the western border of Afghanistan. Outside the Pashtun belt, madrassa enrollment is thinly, but evenly, spread across the rest of the country. Madrassa enrollment declined from 1940 to 1980 but increased during the religion-based resistance to the invasion of Afghanistan by the Soviets in 1979. The largest jump in madrassa enrollment is for the cohort aged 10 in the period 1989-93--coinciding with the withdrawal of the Soviet Union and the rise of the Taliban. Outside the "Pashtun" belt, the distribution of madrassa enrollment at the level of the village is spread evenly with most variation within rather than between villages. Among the households covered by our 2003 census, we find the same pattern. The prevailing hypothesis that households do not have other schooling options and thus send their children to religious schools, or that households are religiously minded and thus choose madrassas over private and public schools is not supported by the data. We looked at households with at least two enrolled children where one child attends a madrassa. Such households are small in number, accounting for less than 1 percent of the sample. Among these households, less than 25 percent send all their children to madrassas; in contrast, 50 percent send their children to both madrassas and public schools and another 27 percent use the private school option. This "stylized fact" requires a theory of variation between children rather than households to explain such enrollment.5 Widely promoted theories of madrassa enrollment based on household-level attributes such as income or religiosity and village-level characteristics, such as alternative schooling options, simply do not explain the substantial variation within households. Our data are not extensive enough to answer the important question of what explains enrollment in madrassas either. Instead we offer suggestive evidence for the various hypotheses forwarded in the literature on factors that may lead to enrollment in a madrassa. Although we find that the Afghan war during the 1980s had an impact on madrassa enrollment, our ability to predict madrassa enrollment using multivariate regressions is poor. There is weak evidence to support the hypothesis that poorer and less-educated families are more likely to send a child to a madrassa, and somewhat stronger findings that poor children in settlements without a school use 5 One such theory, familiar to historians of religious schooling in European countries, is primogeniture (inheritance of the first-born of the parent's wealth), which often led to secular education for the first-born and careers in the church for younger siblings. Other explanations could be differences between siblings in terms of their abilities, gender, or health (for instance, disabled children may be more likely to be enrolled in madrassas). 5 madrassas more often. There is no evidence for explanations based on household preferences such as religiosity. The remainder of the paper is structured as follows. Section 2 summarizes recent literature (including the popular press) on the extent and penetration of madrassa enrollment. Section 3 uses existing data sources as well as our own survey to provide a comprehensive picture of madrassa enrollment. Section 4 looks at geographical variation in madrassa enrollment and trends over time. Sections 5 and 6 examine variation across households and comment on the consistency of the data with some hypotheses that have been advanced by other authors. Section 7 concludes. 2. What do Reports and the Popular Press Say? We looked at three different types of articles and reports: articles in mainstream American and international newspapers; reports and articles by American and international scholars affiliated with international think tanks, institutes, and the government (including the 9- 11 Commission Report); and studies by Pakistani scholars working in Pakistan and abroad. The sources for all these reports are either newspaper accounts of police estimates or interviews with policymakers. We have yet to find a single article that tries to validate these numbers using established data sources. We searched 10 major newspapers using FACTIVA in June 2004 for articles published in the past three years that mention "madrassas." Table A1 in the appendix shows the range of figures and their underlying sources for the 44 articles found.6 Three facts stand out. First, enrollment figures vary widely, ranging from 500,000 children to 1.5 million children. Worse (as discussed above), was a lack of consistency in estimates from the same newspaper. Second, only two articles benchmarked these numbers with the total number of children enrolled in schools (Hussain and McGory in The Times, August 2002 and Kraul in The Los Angeles Times, April 2003). There is a sea of difference between the proportions they report--according to The Los Angeles Times, 10 percent of all Pakistani students are being educated in madrassas; The Times, however, citing an International Crisis Group Report (discussed later) suggests that this proportion is closer to 33 percent. The use of numbers rather than percentages affects perceptions. As we see later, our understanding of madrassas and their prevalence changes dramatically when 6 Newspapers searched include the Boston Globe, Chicago Tribune, Financial Times, The Guardian (U.K.), ,The Independent (London), Los Angeles Times, The New York Times, The Philadelphia Inquirer, The Times (London) and The Washington Post for at least 3 occurrences of the word "madrassa" (spelled in different ways) from January 2001 to June 2004. 6 we present enrollment in madrassas as percentages, either of all school-aged children or of all enrolled students, rather than numbers. Third, all the articles rely completely on secondary sources. While we cannot reasonably expect newspapers to use primary data sources, the articles failed to acknowledge the accuracy of enrollment estimates using an appropriate disclaimer. The 9-11 Commission Report adopted the same methodology of interviews and reporting numbers based on secondary sources. We quote in full the passage relating to madrassas: "Pakistan's endemic poverty, widespread corruption, and often ineffective government create opportunities for Islamist recruitment. Poor education is a particular concern. Millions of families, especially those with little money, send their children to religious schools, or madrassahs. Many of these schools are the only opportunity available for an education, but some have been used as incubators for violent extremism. According to a Karachi's police commander, there are 859 madrassahs teaching more than 200,000 youngsters in his city alone." (Section 12.2). This report provides only a footnote quoting an interview with a police commander and does not attempt to validate the numbers provided. Striking, yet unsubstantiated claims such as "Millions of families...send their children to religious schools" are of particular concern given the emphasis on identifying and curbing potential sources of extremism. Moreover, even the secondary sources referred to in the report do not base their estimates on published data sources. The public perception of madrassas and their role in Pakistani education have been strongly influenced by three published pieces (both in journals and policy reports): an article written by Jessica Stern (2000) in Foreign Affairs, a piece by Jonathan Singer (2001) for the Brookings Institution, and a report by the International Crisis Group (2002), an independent, non- profit organization. Stern (2000) refers to "...the estimated 40,000 to 50,000 madrassas in Pakistan..." without identifying any verifiable source. Singer (2001) claims that, "...there are as many as 45,000 such schools within Pakistan..." although he admits that "...the exact number has never been determined." The ICG report, published in July 2002, is the only one that estimates the number of students enrolled. Consequently, it has been a definitive source for estimates of madrassa enrollment numbers--6 out of 11 articles written after July 2002 in our FACTIVA search referenced the report. The influence of this report was undoubtedly bolstered by its claim that, "about a third of all children in Pakistan in education attend madrassas." This 33 percent estimate of the fraction of enrolled children in madrassas is much higher than the only other 7 available figure, reported in The Los Angeles Times (2003), which puts the number at 10 percent. We discuss these in some detail below. Pakistani scholars also tend to use secondary or unverifiable sources. For instance, in the highly original and detailed work on Islamic religious scholarship by Zaman (2002) and a series of articles by Tariq Rahman (2004), the number of madrassa students and establishments is sourced from newspaper articles in Pakistan (The Dawn, 22 May 2002 and 16 January 2003). The newspaper articles in turn quote a police press release, where again the raw data is not available for verification. Similarly, Ahmed Rashid (2000) writes in his best-selling book on the Taliban: "...in 1988 there were 8,000 madrassa and 25,000 unregistered ones, educating over half a million students." The source cited in Footnote 13 of Chapter 6 of the book is "Intelligence report presented to the cabinet of Prime Minister Nawaz Sharif in 1992." Finally, the purpose of many of these articles and reports was to try and explain the popularity of madrassas, not estimate the numbers of enrolled children. The final column in Table A1 describes the main reasons advanced in newspaper articles. Of these, one of the most popular cause and effect arguments is the "failed-state" advanced by Singer (2001). According to this argument, the rise of madrassas is linked to the poor Pakistani public school system: "The reason for the madrassas new centrality stems from the weakening of the Pakistani state...the madrassas became immensely popular by targeting the lower class and refugee populations, whom the Pakistani state has failed to provide proper access to education." (Singer, 2001) This kind of coverage has fostered two conclusions: madrassa enrollment rates are high and increasing, and the popularity of madrassas should be understood as a response by the poor to the government's inability to provide public education and social welfare. Where is the evidence? 3. The Data About Madrassa Enrollment We examine three important questions related to madrassa enrollment: the number and fraction of children (both school-aged and enrolled) enrolled in madrassas; the geographical variation in madrassa enrollment across Pakistan; and enrollment trends over time with particular attention to two benchmarks--the Soviet invasion of Afghanistan (and the rise of the Jihadi movement against them in 1979) and September 11, 2001. 8 3.1 Data Sources We use three different types of data to verify our estimates and determine how sensitive they are to changes in definition and the year of the survey. Two sources are nationally representative, but date from 2001 or before, the third is data from a census of households carried out by the authors in 2003 as part of a project on educational choice. The first source is the "long" form of the population census in 1998, which is a large sample-based survey with information on enrollment. This survey is representative at the level of the district and region (rural or urban) and provides comprehensive coverage of the entire country.7 We use this data to examine enrollment patterns across districts. The second type of data, based on household surveys, are different rounds of the Pakistan Integrated Household Survey (PIHS) carried out in 1991, 1998 and 2001. While the data is not as extensive as the census, it contains detailed household information on schooling and income, and has been used extensively by researchers both in Pakistan and the United States. Finally, we use the census of schooling choice among households that our research team conducted in August 2003 (referred to as the project on "Learning and Educational Achievement in Punjab Schools", or LEAPS). These data are only for three districts in the province of Punjab, but is very recent, was conducted by an independent team of academics, and is a complete census of all households in the selected villages. Consequently, it yields sufficient madrassa enrollment to examine correlations with household attributes in a meaningful manner (this data source provides information on four times as many children as the PIHS). Table A2 in the appendix shows how these different data sources are used in the paper. Each source asks about madrassa enrollment in a slightly different but comparable way. The population census (1998) asks about the field-of-education ("What is name's field of education?") with options that include (for instance) engineering, medicine, or religious education. This question is also asked of all literate adults irrespective of their current enrollment status, allowing for comparisons in the stock of religious education over time. The PIHS rounds ask, "What type of school is name currently attending?" with options that include government school, private school, or deeni-madrassa (religious schooling). Finally, the LEAPS census directly asks, "Is the child enrolled in a madrassa or an Islamic education school?" Fortunately these different questions all give rise to similar numbers. This is reassuring since it suggests that any one particular result is not driven by the specific question or definition that was used. 7This is complemented with the census of private schools carried out by the Federal Bureau of Statistics in 2000 (www.statpak.gov.pk). 9 3.2 How Many? According to the 1998 census of population, 159,225 students (all ages) were enrolled in madrassas; this represents 0.31 percent of all children between the ages of 5 and 19 (Table 1, Row 1). Since the total gross enrollment rate (the gross enrollment rate or GER is defined as total enrollment divided by the number of "eligible" children--in this case, children between the ages of 5-19) is 45 percent, madrassa enrollment as a fraction of total enrollment increases to 0.7 percent. The numbers from the PIHS are very similar--with between 151,000 and 178,000 children enrolled in madrassas across the 1991, 1998 and 2001 rounds, this accounts for less than 1 percent of all enrollment (around 0.7 percent of all enrollments in the 1991, 1998 and 2001 rounds). Despite the different definitions used and the problem of accurately estimating a low probability event in the PIHS (these surveys typically identify less than 100 children enrolled in madrassas), the numbers are well within the standard error bounds, and within 0.1 percentage points of each other--that is, there is less than 1/1000th of a difference between the percentages of enrolled children going to madrassas depending on the different sources. These numbers can be benchmarked to enrollment in regular schools. Public schools run by the government enrolled between 16 and 17 million children in 1998 (Census of Pakistan, 1998); private schools enrolled almost one-third as many, at 6 million in 2000 (Federal Bureau of Statistics, Survey of Private Educational Institutes in Pakistan, 2000). As a percentage of children between the ages of 5 and 19, government schools accounted for 33 percent and private schools for another 12 percent. Again, since roughly one-half of all children between the ages of 5 and 19 are enrolled in school, as a percentage of enrollments these numbers approximately double to 73 percent and 26 percent.8 This comparison suggests that there are 38 times as many children in private and 104 times as many in government schools compared to madrassas.9 8 Boys are more likely to be enrolled in madrassas compared to girls. The census reports that there are only 43 females enrolled in madrassas for every 100 males. This imbalance is significantly greater than the 68 enrolled females for every 100 enrolled males in overall education. For both boys and girls, madrassa enrollment starts off at the same age cohort between 5 and 9 years, but girls' enrollment drops off sharply while boys enrollment jumps for children between 10 and 14 and then tapers off for the cohort in their mid-twenties. That there are fewer females in madrassas is not surprising, but the similar number of girls as boys in the youngest age category is somewhat unexpected. Interestingly, for children between 5 and 9, the enrollments sex ratio is slightly higher in madrassa than in general education (76.9 vs. 75.6 percent) but drops off at a much faster rate in the madrassa as children become older. The usual disclaimers about the numbers being too small to draw finer comparisons still apply. 9 We could also compare these numbers to enrollment across countries. In their work on religious enrollment, Berman and Stepanyan (2003) compare a number of countries, including Pakistan (albeit based only on the PIHS). The comparison is fraught with difficulties, since they sometimes uses stocks and sometimes flows and the data is at times at the household level and others at the individual level. Nevertheless, using their numbers, as a percentage of total enrollment, madrassa enrollment in Pakistan is roughly equivalent to that in Bangladesh and Côte d'Ivoire and much less than in India (two states only) or Indonesia. Interestingly, madrassa enrollment in Pakistan corresponds closely to 10 There may be grounds for skepticism about these estimates for madrassa enrollment. Since the data were collected prior to 2001, geopolitical changes after September 11 could have led to greater madrassa enrollment. In addition, the household-based survey faces the usual problems of accurately estimating a low-probability event--although enrollment is less than 1 percent in these surveys, the sampling error is large (see Bauman, 2001, for a description of similar problems in estimating home-schooling in the United States). Finally, while the census of populations does not face the problem of small samples, it is not that recent (1998) and some may have reservations regarding the quality of government data.10 The LEAPS census of schooling choice conducted in 2003 provides a rough check on these numbers (see appendix for details). This census was conducted in three districts of Punjab and villages were chosen randomly based on the criterion that each village must have at least one private school. Typically, this means that the villages lies somewhere between fully urban and fully rural populations and are not representative of the districts that they are in. Estimates from the LEAPS census show that as a percentage of enrolled children, the numbers in two of the three districts are slightly higher than those of the population census. In the third (Rahim Yar Khan) there is a large difference, with the census reporting that 1 percent of all school-going children attended madrassas, and the LEAPS showing that the fraction is closer to 3.7 percent (Table II). There are three potential explanations for this difference. First, the LEAPS data is not representative of the district and could be off the mark for districts with wide variation in madrassa enrollment across rural and urban samples. Second, the experience of the last five years could have varied dramatically across districts--in some, the enrollment fractions did not change and in others it increased substantially. Third, the data could point to systematic problems with the census estimates from certain districts, or the statistical problems that arise when we try to estimate low-probability events. 3.3 Explaining the Differences A number of reasons could account for differences between the estimates presented here and those in the popular press. 1. Differences in the sampling unit. Our estimates are all based on household surveys--an interviewer goes to a household and asks about the enrollment status of every child. Some census estimates of home rather than religious schooling in the United States--the former ranges from 1 to 2 percent (Bauman 2001) while the latter is closer to 8 percent (National Center for Education Statistics, 2001). 10 In our own analysis, we find the quality of the data generated by the Federal Bureau of Statistics in Pakistan to be consistently high. We have used the FBS Census of Private Schools (PEIP, 2000) to guide our fieldwork and feasibility study for LEAPS and found it tallied with the situation on the ground quite well, even in remote villages. 11 estimates in the literature are based on establishment surveys, where interviewers go to madrassas and ask about total enrollment in the institution. Our numbers could be lower than the actual enrollment if a significant proportion of those enrolled in madrassas are not linked to any household. This could happen if the share of orphans in madrassas is higher than their share in regular schools. While there are no countrywide estimates, a study by the Institute of Policy Studies in Islamabad (2002) found that 15 percent of all children enrolled in a sample of madrassas were orphans. In the worst-case scenario (where all orphans are enrolled only in madrassas), our numbers would have to be inflated by 15 percent to arrive at actual enrollment. 2. Differences in definition. We define a child as enrolled in a madrassa if he or she is enrolled full-time in such an institution. Alternative definitions also include those who are enrolled full-time in regular schools, but also attend madrassas (usually for evening classes) for lessons on the Quran. Thus, the police estimate for madrassa enrollment in Punjab is 235,000, which is substantially higher than the census estimate of 96,000. However, the police estimate includes children who are studying part-time; in all our data sources, these are excluded. If we take for granted that the existing data sources are dated, then reality is better represented by the more recent LEAPS census data. In this case, total enrollment in madrassas, using a population-weighted average across the three districts would be 1.7 times that in the 1998 government census. Adding in a (generous) 5 percent enrollment growth rate for every year, this puts the total number of children enrolled in madrassas at 410,000. This estimate is still conservative if a substantial fraction of children in madrassas are orphans not be picked up in household-based surveys. Thus, to arrive at a liberal estimate, we can add in a further 15 percent for orphans to take the total up to (approximately) 475,000, which is still below the lowest estimate in the existing literature. 11 11Police surveys in Punjab and Sindh quoted in the literature estimate that there are 253,125 students enrolled in madrassas in Punjab in 2002 and 264,169 in Sindh in 2003. Our equivalent number from 1998 for the Punjab is 96,125. For Sindh the number is ten times greater than that in the census, with Karachi accounting for all the difference. While we try to investigate further, there are two observations. First, there is no information available on the methodology of the police surveys. Our interviews suggest a lack of documentation, certainly when compared to those of established statistical surveys. We have no way of knowing whether these numbers are from enrollment records in the madrassas or children attending on a given day or whether they are based on a census of establishments or a sample. Second, the police surveys are establishment-based surveys that include all students who attend madrassas (full-time students and children attending part-time for Quranic literacy classes), whereas the census and PIHS question include only those who attend madrassas full-time. Anecdotal evidence suggests that small stints in a madrassa, either for Quranic literacy or for a short time after primary school is fairly common. 12 Particularly surprising is the vast discrepancy in madrassa enrollment as a fraction of enrolled students. Our estimates suggest close to 1 percent, while the only other estimate in the literature (International Crisis Group, 2002) puts the number at 33 percent. Even if the number of children enrolled in madrassas were 1.5 million, the highest estimate in reports and articles, this would still imply that 3 percent of all children in the educational system are enrolled in madrassas. Are we totally off the mark? The International Crisis Group report based its estimate on two different statistics--the total number of children enrolled in madrassas, and the total number of children enrolled in all schools. It cites the total number of children enrolled in madrassas at between 1 and 1.7 million, and this figure is based on an interview with the Pakistani minister for religious affairs, Dr. Mahmood Ahmed Ghazi.12 For the total number of enrolled children, the report references the Economic Survey (Ministry of Finance, Government of Pakistan, 2002) and claims that enrollment in primary schools in 2001 was 1,992,132. Taken together, these two figures suggest that one-third of all children going to school were in madrassas. What the Economic Survey (2002) actually says is "Enrollment at primary, middle, and at higher levels were 19.92 million, 4.28 million, and 1.79 million respectively." This line in the text is corroborated by a table showing information on educational statistics, in which primary enrollment is stated as 19,921,232. The report omits the second `2' in the Economic Survey number, leading to an estimate off by a factor of ten. It is unfortunate that the report, which is informative in other respects, is marred by this fundamental mistake. Equally unfortunate is the immediate validation and propagation of this number without any attempt to "fact-check" the total enrollment numbers. 4. Variation across Districts and Time Using the published data sources we can also look at how madrassa enrollment varies across different districts--the difference between Attock and Rahim Yar Khan in the LEAPS data suggests that this may be substantial--and we can look at how such enrollment has changed over time. Both these exercises yield similar conclusions, pointing to the importance of the border region with Afghanistan and the impact of the Afghan war years on time trends in madrassa enrollment. 12 In a separate interview with one of the authors of this note (July 28th 2004) Dr. Ghazi confirmed these estimates although he did not identify a verifiable source, citing security reasons. 13 4.1 Variation across Districts Pakistan is divided administratively into four provinces with 102 districts--Punjab, Balochistan, North-West Frontier Provinces (NWFP), and Sindh--plus the federal capital Islamabad, the Federally Administered Tribal Areas (FATA), the federally administered Northern Areas and Azad Jammu and Kashmir (AJK). The four provinces or Punjab, Balochistan, Sindh and NWFP, together with Islamabad, account for more than 97 percent of the population. Geographically, parts of Balochistan, the NWFP and FATA border Afghanistan. Sindh and Balochistan are sparsely populated provinces, with the exception of Karachi in Sindh, which is the single biggest metropolis in the country with a population approaching 10 million. We use data from the population census, 1998, as well as the census of private schooling, 2000, to provide estimates of madrassa, private, and government school enrollment in each district except for those in the province of FATA. The geographical dispersion of madrassa enrollment depends on how we define madrassa prevalence. There are three alternatives. We could present a geographical breakdown of the total number of children enrolled in madrassas. This number is related to the total population of the district, and may thus reflect only the size of the district relative to others. Another option is to use the equivalent of the gross enrollment ratio (GER), defined as the total enrollment divided by the number of "eligible" children--in this case, children between the ages of 5-19. This statistic provides an estimate of the "penetration" of madrassas, but it does not take into account the overall enrollment decision of the family. Thus, a district with two children enrolled in madrassas, and 20 children enrolled in private or public schools out of a total of 100 children will have exactly the same gross enrollment ratio (GER) as a district with two children enrolled in madrassas and 98 children enrolled in regular schools. To the extent that we want to distinguish between these two districts, a third statistic, the ratio of children enrolled in madrassas to total enrollment (the madrassa fraction of enrollment or MFOE), can also be used. The picture changes dramatically depending on whether we use the raw numbers or the ratio of children enrolled in madrassas to total enrollment. However, since enrollment in madrassas is highly correlated with total enrollment, there is little difference in the pattern of madrassa enrollment whether we use the GER or the fraction of enrolled children in madrassas. Figure 1a shows the number of children enrolled in madrassas for every district in the country. As expected, numbers are closely linked to population size--the three most populated districts account for one-quarter of the enrollment, with the bulk of enrollment in large urban 14 metropolises. Madrassa enrollment is also higher in the Saraiki language belt districts of Southern Punjab (for instance, Rahim Yar Khan) compared to the rest of the country. The fact that the number of children enrolled in madrassas is higher in more populated districts indicates nothing more than the size of the district. Figure 1b normalizes this number with total enrollment. Districts in the dark solid colors have high madrassa enrollment compared to total enrollment (still less than 7.5 percent), and districts in diagonal stripes have the lowest enrollment fractions in madrassas. What stands out is the very high madrassa fraction along (and only along) the western border with Afghanistan. This is the Pashtun belt--the "Pashto" speaking population most directly influenced by events in Afghanistan. Pishin (the district bordering the Kandahar region of Afghanistan) is the only Pashto speaking district in the top 15 when we use the total number of children enrolled in madrassas, but when we deflate this number by total enrollment (or the number of eligible children) all the top ten districts are in the Pashto speaking belt. If we use 2 percent of total enrollment as a cutoff for "extreme" madrassa enrollment, all 14 districts that can be classified as such are either in Balochistan or the North-West Frontier Province. Leaving aside the Pashtun belt and Karachi, madrassa enrollment in the rest of the country is spread very thinly across all districts. That is, rather than districts with high enrollment and districts with low enrollment, a very small number of children in every district are enrolled in madrassas. Seventeen districts fall in the 1 to 2 percent MFOE group with the remainder reporting madrassa FOE below 1 percent. For the bulk of the districts in the country, madrassa enrollment falls between 0.02 and 1 percent of total enrollment. Thus, there are two distinct geographical patterns of madrassa enrollment in the country. Enrollment is systematically higher among districts that border Afghanistan (still with just over 4 percent of enrolled children in madrassas in all of them except Pishin). Apart from this group, enrollment is thinly spread among all districts in the country. Perhaps this pattern also reflects variation in the types of madrassas in these different areas--clearly, differences (if any) between madrassas in the Pashto belt and the rest of the country is a potential area for research. 4.2 Variation across Time The PIHS data show that the growth in madrassa enrollment during the 1990s was the same as that in enrollment in all schools. In fact, madrassa enrollment as a fraction of total enrollment decreased marginally, although the differences are well within the margin of error. Since the absolute number of madrassa-going students increased by 16 percent over the 10 years, 15 the decline in the proportion was due to slightly higher growths in overall enrollment during this period. Over a longer time horizon, we also compare the stock of individuals who stated a religious "field-of-education" on the census long-form. This cohort-based analysis assumes that changes in madrassa enrollment will be picked up 10 to 14 years after the birth of the relevant cohort. That is, a sudden increase in 1980 would correspond to an increase in the stock of individuals reporting religious education for the cohort born between 1966 and 1970. The stock of individuals reporting religious education as their "field" displays a distinct U-shaped pattern (Figure 2). People born between 1944 and 1948 were more likely to have religious education than those born between 1949 and 1953. This pattern continues until the cohort born between 1959 and 1963, then stabilizes for the next two cohorts and finally starts increasing. Those born between 1974 and 1978 were more likely to report religious education as their field than those before. This increase continues until the last cohort we have completed educational history data for, those born between 1979 and 1983. Interestingly, the downward trend halts for the age-cohort that comes of school-going age in the years surrounding the rise of the Zia-ul-Haq military government and the ensuing religion- based resistance to the Soviet invasion of Afghanistan starting in 1979. Further, the largest jump is in the cohort born in 1979-83. This cohort would be 10 years of age in the period 1989-93-- coinciding with the withdrawal of the Soviet Union and the rise of the Taliban. 5. Variation across Households This section looks at variation in madrassa enrollment at the level of the household. This study was not designed, and is not capable of, presenting definitive statements regarding the choice of madrassas among households as a schooling choice. As such, we do not attempt to go beyond simple tabulations and associations. Nevertheless, the exercise adds some value to our understanding of madrassa enrollment.13 Previous studies on this important question (Berman and Stepanyan, 2004) have been hampered by small sample sizes on the one hand (there are 100 children who attend madrassas in 13Differences in madrassa enrollment could be driven by differences across villages rather than households--i.e., in some villages all the children go to madrassas and in others none of the children attend. Most of the variation in madrassa enrollment (in the set of enrolled children) is within rather then between villages. For example the difference in enrollment ratios in a village at the 90th percentile of madrassa enrollment and one at the 10th percentile is only 3.4 percent. This also holds at the more disaggregated settlement level (some villages have more than one settlement), although less strongly, with more variation stemming from within settlement differences than between. 16 the PIHS data) and lack of household-level data (the census only provides district-level aggregates). In contrast, the LEAPS census has schooling choice data on 150,000 children at the household level; even with 1 percent enrollment, this gives us 1,500 children enrolled in madrassas, a number that is sufficiently large to draw meaningful conclusions. We first ask a simple question. If we look at households who have a child enrolled in a madrassa, what do the schooling choices of other children in the household look like? We look at households that had at least two enrolled children, and classify them as (a) "all madrassa" if all the children attend a madrassa, (b) "madrassa/public" if at least one child goes to the madrassa and one to a public school, (c) "madrassa/private" if one attends the madrassa and the other attends private school and (d) "madrassa/public/private" if the households had 3 or more enrolled children using all three options simultaneously. Finally, we repeated this exercise in households with at least one child going to public school and with one child going to private school. The results are startling (Figure 3). Among households with at least one child enrolled in a madrassa (call them "madrassa households") only 23.5 percent can be classified as "all madrassa" households. The majority of "madrassa households" (just under 50 percent) use both madrassas and public schools, and another 28 percent use either madrassas and private schools or all three simultaneously. Among households with at least one child enrolled in private school, 48.5 percent enroll all their children in private schools and another 49.6 percent use the private/public option simultaneously. If the choice of a madrassa or a private school provides information about the ideology of the household, the data suggest that the choice of a private school is more ideologically driven than the choice of a madrassa. The fact that most variation in madrassa enrollment is within rather than between households implies that any predictions about who will send their child to a madrassa based on household attributes will be fairly poor. The comparison done by Berman and Stepanyan (2004) looked at differences between "madrassa households" and "non-madrassa" households. A similar comparison in the LEAPS data (Table III) shows an association between lower-income households and madrassa enrollment and households with less educated heads and madrassa enrollment, but the magnitude of these associations is small. The heads of "madrassa households" are illiterate in 7 percent more cases and slightly poorer (0.5 percent more likely to earn less than Rs.7500 per month).14 The largest difference between household types is their proximity to a 14 The significant difference between madrassa and non-madrassa households in terms of income arises only when we use Rs.7,500 as the monthly income cutoff to distinguish poor and non-poor households--alternative cutoffs 17 private school--among households with a child in a madrassa, 49 percent live in settlements with a private school; this number is 72 percent for households with no children in madrassas. 6. Taking Stock: What can we say about established hypotheses? Although the last finding could be seen as supporting the theory that madrassas have emerged as the only viable alternative to poor government schooling, a closer look shows that the prevalence of private schools substantially complicates matters. We look at the fraction of children enrolled and the market shares of schools offering public, private, and madrassa education under three different scenarios. One scenario is when both private and public schools are present in the settlement, the second is when either a private or a public school exists (non- exclusively), and the third when neither is available in the settlement that the household is located in.15 Three interesting findings emerge (Figure 4). 1. In settlements with both private and public schools, the share of private schools increases with income. Nevertheless, even among the poor more than 30 percent of all families send their children to private schools. Madrassas account for less than 1 percent of all enrollments, and this share is no higher among the poor compared to the rich. 2. In settlements with either a private or a public school the exact same patterns are observed with regard to income and again, madrassa shares are no different among the rich and the poor. 3. In settlements with no public or private schools the enrollment patterns are very different. The largest differences arise in the choice of enrollment, rather than schooling choice conditional on enrollment. In settlements without a public or a private school, families are more likely to exit from the educational system altogether rather than enroll their child in a madrassa. The drop in enrollment is dramatic, falling from 70 percent to 40 percent among the poor and from 87 percent to 68 percent among the rich. The market share of the different types of schools also changes. Government schools and madrassas increase their share while the private schools share drops. However, the increase in the market share of government schools is insufficient to overcome the drop in enrollment, so of Rs.2,500 or Rs.5,000 show no significant differences between poor and non-poor households. Importantly, 97% of all households earn below Rs.7,500. 15Recall that the LEAPS sample was taken from a list frame of villages that had at least one private school. To look at schooling choices under these three scenarios, we divided villages into settlements and plotted enrollment shares in each type of settlements. These settlements are often far from each other, and thus act more as self-contained units than the administrative definition of a village. In the LEAPS data, there are 112 villages, but 253 settlements, generating considerable variation for this exercise. These findings are robust in a multivariate regression context. 18 that total number of children in government schools still declines; the opposite is true for madrassas. These settlements are also the only ones where there is a clear relationship between income and madrassa enrollment--among the poorest, 4 percent of all enrolled children are in madrassas and among the rich this decreases to 2.5 percent. These numbers suggest that the schooling decision for an average Pakistani household in a rural region consists of an enrollment decision (should I send my child to school) followed by a private/public decision, with a madrassas possibility. When there are no nearby schools, households exit from the educational system altogether although there is evidence of an increase in the market share of madrassas among the poor in these settlements. When both private and public schools are available, richer households exit to the private system, but there is no difference in madrassa shares with household income. It's possible that when richer households use the private sector, resources are freed up for use in public schools leading to an overall increase in enrollment. The key issue then becomes the placement of public and private schools. Andrabi, Das, and Khwaja (2004) show that private schools overwhelmingly locate in villages where there are pre-existing public schools. This may not be an insurmountable barrier as 2,500 new private schools were set up between 1994 and 1995. Just three years later in 1998, 6,000 new schools were set up and in 1999 this increased to 8,000. Half of the growth in private schools occurred in rural villages. If current trends continue (and our data collected in 2004 suggests it will), the number of villages in Punjab province, with half the country's population, with a private school will approach 50 percent by the end of the decade. These schools are not particularly expensive to attend. The median annual tuition fee in rural Punjab is Rs.650 (just under $1 per month). Using household expenditure data from the PIHS, this represents 1.7 percent of average household expenditure so that a family with 4 children would spend 7 percent of total expenditure if all four children attended private schools (Andrabi, Das, and Khwaja, 2003). By comparison, the ratio of private school fees to household income in high-income countries is high--for instance, the average annual fees in a private school in the United States of $2,200 (National Center for Educational Statistics, 2001) corresponds to 9 percent of median per capita income. The analog in Punjab is closer to 2.5 percent suggesting a four-fold difference in the tuition to income ratio across the two economies. A number of explanations apart from the "failed-state" hypothesis have been proposed in the literature. Another, equally valid hypothesis is that the rise of madrassa enrollment was linked to the Afghan war. The data does show a reversal of the declining trend in madrassa enrollment 19 between 1948 and 1965, with increases in the percentage of adults with religious education in the cohorts born after this date. There is also wide geographical dispersion in the prevalence of madrassa education in Pakistan. Although all districts report that less than 2.5 percent of children in the relevant age group (children between the ages of 5 and 19) are going to madrassas, the Pashto speaking belt that borders Afghanistan stands out in terms of the popularity of madrassas as an educational choice. The notion that the madrassa movement coincided with resistance to the Soviet invasion of Afghanistan is supported by the 1998 data from the population census. The increase in the stock of religiously educated individuals starts with the cohort that came of age in 1979 (the year of the Soviet invasion of Afghanistan) and the largest increase is for the cohort co-terminus with the rise of the Taliban. Combined with the fact that the largest enrollment percentage in Pakistan is in the Pashtun belt bordering Afghanistan, this suggests events in neighboring Afghanistan influence madrassa enrollment. Is there something intrinsic about Pashtun sensibility or tribal culture that leads to higher madrassa enrollment? The differentiation of the Pashtun and non-Pashtun districts does not extend to Pashtun and non-Pashtun households in the LEAPS data. We find no evidence that Pashtun households are more likely to send their children to Madrassas compared to the rest of the sample, suggesting that geopolitical factors and geographical proximity to Afghanistan matter more than cultural preferences.16 Similarly there is no evidence for religiosity or household preference-based models of madrassa enrollment. The radical religiosity argument suggests that children are more likely to be sent to madrassas when the family favors a radical brand of Islam. If true, what are we to make of the fact that more than 75 percent of all households with a child in a madrassa also send a child to a public or private school? In a multivariate context we checked whether households identified as "radically Islamic" were more likely to send their child to a madrassa.17 Again, we found no 16 The data from the LEAPS census asked about ethnic and caste identity, and households that classified themselves as "Pathan" or "Afghani" were used to represent Pashtun households. In line with the usual residential patterns of individuals with Pashtun backgrounds, most of these households are in district Attock in the North of Punjab. 17 In a largely Islamic country it is difficult to find good measures of religiosity. No data on religiosity was collected as part of the census and a more recent and detailed household survey that includes information on time-use elicits little variation--everyone reports high mosque attendance and regular prayers. An alternative, suggested by David Evans at Harvard University, which we pursue here, is to use recent developments in the use of "names." Research by Fryer and Leavitt (2004) demonstrates the increasing use of names to define race identity in the United States. We postulate that households who named (at least) one child "Osama" (also spelt Usamah, Usamma or Usama) are more likely to favor a radical brand of Islam. The use of the name Osama was minimal until 1998, and then peaks in 1998 and 2001, following disruptive events. Of course, the naming of the child may reflect name recognition rather 20 difference between these households and others--the probability of choosing a madrassa increased by 0.40 percent, but the increase was statistically insignificant and small. Many interviews in the press suggest that madrassas provide free food and clothing and stipends to the poor. This implies significantly higher madrassa enrollment among poorer segments of the population, a view also advocated in a number of interviews with policymakers in Pakistan. The "three-scenarios" figure (Figure 4) shows that this is true only in settlements where there are neither private nor public schools. In general, the effect of income on enrollment overwhelms the effect on the choice of education type making madrassa enrollment lower as a percentage of population in poorer families than in richer families 7. Conclusion and Caveats Analysis of published data sources and the LEAPS survey shows that existing estimates in the literature of madrassa enrollment are highly exaggerated. Our most liberal estimate, which doubles the census/PIHS numbers and adds in a further 15 percent for orphans, is still below the lowest estimate in newspaper articles and policy reports. This imbalance is accentuated when we look at the fraction of children enrolled in madrassas, either as a percentage of school-aged children or enrolled children. This fraction has been overstated by a factor of 10 in The Los Angeles Times (2003), and 33 in the report by the International Crisis Group (2002). Moreover, there is currently no evidence of a dramatic explosion of enrollment in madrassas in the 1990s. The geographical dispersion of madrassa enrollment helps us to understand the use of religious schooling. Madrassas are most popular in the Pashtun belt with the top ten districts in terms of the fraction of enrolled children in madrassas all bordering Afghanistan (where they still account for less than 2 percent of all school-aged children). Time trends also support a strong Afghan war influence. Madrassas declined in popularity for cohorts born between 1947 and 1974 and increased thereafter. The biggest jump is for the cohort born between 1979 and 1983; this corresponds to those children who would start attending school around the rise of the religion based resistance to the Soviet invasion of Afghanistan. It appears that this "Afghan" influence is related more to geographical proximity than to preferences for religious schooling among Afghan immigrants--the LEAPS data show no difference between Pashtun and non-Pashtun households in the use of madrassas. than ideology and this must be kept in mind. To the extent that naming a child "Osama" is a good indicator of radical religiosity, we find no evidence of this effect in the data. 21 At the level of the household, most variation in madrassas enrollment is within rather than between families. Among households with at least one child enrolled in a madrassa, 75 percent send their other child to a public or private school or both. Historians of religious education choices will find this reassuring--during the 18th and 19th centuries, most European countries followed a similar pattern with one child sent to the church and others sent for a secular education (or no schooling at all). Are poorer families more likely to send at least one child to a madrassa? At an aggregate level there is little difference between poor and rich households in the choice of religious schooling. However, this masks an important difference between two different types of settlements. In settlements where other schooling options exist, less than 1 percent of all enrolled children go to madrassas and this fraction is the same for all income groups. In settlements where there are no other schooling options, the fraction of children going to madrassas increases and is higher among the poor compared to the rich (although it stays below 4 percent for all income groups). Nevertheless, the biggest difference between these two types of settlements is not the increase in the use of madrassas but the dramatic decline in overall enrollment. Thus, though the share of madrassas increases, this is offset by a sharp reduction in the size of the overall enrollment pie. It is likely that the number of settlements without public or private schooling options will reduce considerably during the next decade, primarily due to an ongoing dramatic explosion in the growth of private schools. In 1983 there were approximately 3,300 private primary and secondary schools in the four biggest provinces (Jimenez and Tan 1987)). In 2000 the same four provinces had 32,000 private schools, an almost ten-fold increase in less than two decades (Andrabi, Das and Khwaja, 2003). The growth in low-cost rural private schools is particularly dramatic, a point left out of the current debate on education in Pakistan. For the average child (even a relatively poor one), the most popular alternative to government schooling is a private school, not a madrassa. This paper does not address a number of important questions. Both case studies and personal visits suggest that madrassas vary in their character and the education that they impart, ranging from neighborhood evening religious education schools to those incorporating a more extreme radical militant view. None of the data sources distinguish between different types of madrassas. All types of madrassas are included in our enrollment estimates. Furthermore, we are unable to provide an in-depth view of madrassa goers. For a case-based approach with detailed studies of select madrassas see Stern (2000) and various press reports. These case studies describe 22 the mindset of madrassa students, teachers, and religious leaders and provide psychological portraits of such individuals. The results presented here will speak differently to people with different concerns. One concern is to obtain a better understanding of how madrassas are incorporated into the educational decisions of households. This paper has talked about madrassa enrollment in a framework well known to empirical economists, which deals with issues of poverty and school quality in developing countries. Phrased in terms of household choice (should I send my child to public school, private school, or a madrassa?), the inclusion of madrassas as a schooling alternative has a negligible effect on household decisions. Consequently, for those interested in individual decisionmaking, our results suggest that madrassas do not form an important part of the decision-making of the average (or even the 98th percentile) of Pakistani households. A second concern relates to global security issues and, under this view, absolute numbers matter. While we do not have data on whether madrassas promote extremist views and recognize that this is likely to differ across different types of madrassas, we can conclude that current estimates of madrassa enrollment--both absolutely and in percentage terms--are significantly overstated. Moreover, existing theories fail to adequately explain madrassa enrollment and largely ignore intra-household considerations which appear important. If a pro-active policy toward madrassas is necessary despite the small numbers, more sophisticated theories as well as additional up-to-date, publicly available and verifiable data are needed. However, this is not an easy task. Given the spatial and temporal patterns of overall madrassa enrollment we found, and the very small percentage of children enrolled in madrassas, the only reliable way to capture such enrollment and the correlates of madrassa use is with very large-scale surveys, perhaps even censuses, in a representative sample of villages throughout the country. This would be a costly affair. 23 References The word "processed" describes informally reproduced works that may not be commonly available through library systems. Ahmad, Akbar S. 2002. "Ibn Khaldun's Understanding of Civilizations and the Dilemmas of Islam and the West Today." The Middle East Journal 56 (1): 20-45. Andrabi, Tahir, Jishnu Das and Asim I. Khwaja. 2003. "Private Schools in Pakistan: Catering to the Elite?" The World Bank, Washington D.C. Processed. Andrabi, Tahir, Jishnu Das, and Asim I. Khwaja. 2004. "Women as Teachers: Culture meets the Market in Pakistani Private Schools." The World Bank, Washington D.C. Processed. Bauman, Kurt. 2001. "Home Schooling in the United States: Trends and Characteristics." Working Paper Series 53. Population Division, U.S. Census Bureau, Washington, D.C. Berman Eli and Ara Stepanyan. 2003. "How Many Radical Islamists? Indirect Evidence from Five Countries." University of California. San Diego. Processed. Burki, Shahid Javed. "Basic Education in Pakistan and Afghanistan: The Current Crisis and Beyond." Brookings Forum on Universal Education. http://www.brook.edu/comm/transcripts/20011217.htm. Federal Bureau of Statistics. Various years. www.statpak.gov.pk Fryer, Ronald G. Jr., and Steven Levitt 2004. "The Causes and Consequences of Distinctively Black Names." Quarterly Journal of Economics 119 (3): 767-805. Goldberg, Jeffrey. 2000. "Inside Jihad University: The Education of a Holy Warrior." New York Times Magazine June 25. Hussain, Zahid and Daniel McGrory. 10th August 2002. "British Businessmen are Funding Schools that Breed Terror." The Times. London. International Crisis Group. 2002. "Pakistan: Madrassahs, Extremism and the Military." ICG Asia Report 36. Islamabad/Brussels. Jimenez, E. and J. P. Tan. 1987. "Decentralized and Private Education: the Case of Pakistan." Comparative Education 23: 173-90 IPS Task Force. 2002. Pakistan: Religious Education Institutions. An Overview. Institute of Policy Studies. Islamabad. Kraul, Chris. 14th April 2003. "The World Dollars to Help Pupils in Pakistan". Los Angeles Times. Los Angeles. Looney, Robert. 2002. "A U.S. Strategy for Achieving Stability in Pakistan: Expanding Educational Opportunities." Strategic Insights 1(7). http://www.ccc.nps.navy.mil/si Ministry of Education, Government of Pakistan. 2003. Directory of Deeni Madaris in Pakistan 2003. Curriculum Wing (Islamic Education Sector), Ministry of Education. Ministry of Finance, Government of Pakistan 2002. Economic Survey 2001-02. Government of Pakistan, Islamabad. Ministry of Finance, Government of Pakistan 2003. Economic Survey 2002-03. Government of Pakistan, Islamabad. 24 National Center for Education Statistics. 2001. National Center for Education Statistics Analysis Report. Rahman, Tariq. 2004. "The Madrassa and the State of Pakistan: Religion, Poverty and the Potential for Violence in Pakistan." Himal South Asian. February. Rashid, Ahmed. 2000. Taliban: Militant Islam, Oil and Fundamentalism in Central Asia. New Haven, Conn."Yale University Press. Singer, Peter W. 2001. "Pakistan's Madrassahs: Ensuring a System of Education not Jihad." Brookings Institutions Analysis Papers 41.Washington DC. Stern, Jessica. 2000. "Pakistan's Jihad Culture." Foreign Affairs 79(6): 115-126. The 9/11 Commission. 2004. The 9/11 Commission Report: Final Report of the National Commission on Terrorist Attacks Upon the United States. Zaman, Mohammad Qasim. 2002. The Ulama in Contemporary Islam: Custodians of Change. Princeton: Princeton University Press. 25 Appendix Our first data source is the 1998 population census where the question asked on the long form is the type of schooling attended by the respondents. One of the coded response options is "deeni taleem", exactly translated as religious education. The long form is not asked of every household, so the response is an estimate. However, the numbers are representative at the district rural/urban level. These census numbers are used for delimiting electoral constituencies and also as a sampling frame for most government and international surveys. We have used the census document for designing village level sampling and have found it to be generally consistent with our own fieldwork. We also exploit other district level data in the census document such as extent of public utilities like electrification and piped water as well as the quality of the housing stock (as a proxy for wealth) to correlate with religious enrollment. The second data source is the Pakistan Integrated Household Survey series where we use data from 199, 1998 and 2000. This is a widely used data set internationally for poverty and related matters. The data on religious schooling has been used by Berman and Stepanyan (2004) as well. In all three surveys the question is on type of school and Islamic/religious is a coded response option. Our third source is a specially conducted household educational census conducted by our research team in 125 villages in three districts of Punjab in 2003. This is an extremely rich data source that allows us to extensively look at school choice in rural Punjab at the household level as well as within the household. Here again, we classify a school as madrassah if it provides religious education and does not teach the state prescribed curriculum and a child as "enrolled in a madrassa" if he/she is attending a madrassa full-time The sampling for this data was based on an ongoing study of educational choices in Pakistan. We picked villages randomly from the three districts conditional on their having a private school. Typically, this meant that the villages in our sample are (a) bigger and (b) richer than the average village in the district. Problems of madrassa definition Finding the number of children who are enrolled in madrassas was not an easy task, even in a survey specially designed to understand schooling choice. Some unanticipated issues arose and future surveys might benefit from our experiences. First, children seemed to jump in and out of madrassas. That is, they would enroll in a regular school, then study in a madrassa for a year or two, and return to regular schooling after. 26 The numbers we present are the equivalent of a prevalence rate--that is, it counts the number of children who are currently studying in a madrassa. An alternative would be to capture the children who have ever studied in madrassas, which could yield different numbers. Second, the definition of a madrassa is unclear. We have defined madrassas as schools that teach a religious curriculum rather than that proposed by the ministry of education. There are schools in our sample that teach the government prescribed curriculum but add some elements of religious teaching and are not explicitly involved with any sectarian or religious group. This is in line with the approved principle of "additionality" of the ministry of education and most private schools that want to teach computer science or other additional subjects follow this regulation. We do not count these as madrassas (one type of school that could be mistakenly counted as a madrassa is a mosque school. These are schools housed in mosques that teach the normal state curriculum by regular education department teachers. These were started in the sixth five year plan (1983-1988) under the Junejo government as a cost effective way to use mosques but the program was not picked up later and is now gradually dying out. There are 8000 or so mosque schools in Punjab (EMIS)). Third, since our survey is household based, it will not pick up orphans who are not based in a household. One of the roles that madrassas play is to provide insurance and support. Consequently, case-studies suggest that enrollment in madrassas among orphans may be higher than among the general population. Table A3 shows what happens under alternate assumptions regarding the definition of madrassas. The table presents three different estimates, the lowest of which we call the "conservative" estimate and the highest the "liberal" estimate. Typically, allowing for more liberal estimates doubles the percentage of children enrolled in madrassas; for this paper we use the moderate estimate as the relevant number. This doubling though highlights an important problem. Since madrassa enrollment is very small as a percentage of the total, estimating the number precisely requires very large samples and a precise definition of what we mean. Despite interviews of over 150,000 children, we capture only 1,500 enrolled in madrassas. Small changes in the definition cause small absolute changes in the percentages, but could dramatically change the overall number. Likewise, for extremely low probability answers, errors in data entry even at a 1 percent rate can substantially affect the result. 27 Figures and Tables Figure 1a: Number of Children Enrolled in Religious Schools, Pakistan 1998 Source: The data for the districts is based on the "long-form" of the population census 1998. The long-form is administered to a sample of households in the census and is representative at the district/rural-urban level. We classify a child as enrolled in religious school if he/she reports her "main field of education" as "religious education". 28 Figure 1b: % Children Enrolled in Religious Schools, Pakistan 1998 Source: Population Census, 1998. The percent of all enrolled children in religious schools is defined as the total number of children enrolled in religious schools divided by the total number of children enrolled in any school. 29 FIGURE 2 Literacy Source by Year of Birth 800. noitacud 600. E suoigile 400. R noitcarF 200. 0 1988-1984 1983-1979 1978-1974 1973-1969 1968-1964 1963-1959 1958-1954 1953-1949 1948-1944 Source: Population Census, 1998 Note: Field of education is taken from the long form of the 1998 Population Census and is asked conditional on the interviewee being literate. Every bar shows the proportion of literature individuals in each birth cohort who report that they received a "religious education". The first birth-cohort is individuals born between 1984 and 1988, who were therefore between the ages of 10 and 14 at the time of the survey. The last bar is for individuals born between 1944 and 1948, who were between 56 and 60 at the time of the survey. We make no attempt to control for selection due to deaths over this period. 30 FIGURE 3 Household Enrollment Choices Households With Madrassa Enrollment Households With Public Enrollment 10.5% 0.4% 23.5% 24.0% 16.7% 1.7% 74.0% 49.3% All Madrassa Madrassa/Public All Public Public/Madrassa Madrassa/Private Madrassa/Public/Private Public/Private Public/Madrassa/Private Households With Private Enrollment Households Generally 0. 7% 0.6% 21.3% 18.8% 49.6% 48.5% 1.2% 59.2% All Private Private/Madrassa All Madrassa All Private Private/Public Private/Madrassa/Public All Public Mixed Source: LEAPS Population Census, 2003 Note: This graph shows enrollment choices among households with at least two enrolled children that have one or more children attending a madrassa (top-left), a government school (top-right), a private school (bottom-left) and any school (bottom-right). By construction, households with private, government and madrassa enrollment (Madrassa/Public/Private) must have at least three enrolled children. Thus, among households with at least one child in a madrassa, close to 50% have another child who is enrolled in a government school. The last graph (Households Generally) shows how prevalent every type of household is--for instance, only 0.6% of all households have all children enrolled in madrassas. 31 FIGURE 4 Enrollment by Presence of School and Expenditure 9. 40. tne 8. E OF 30. ml 7. orn 6. E 5. assarda 20. M 10. 4. 0 Rs.10000 Rs.10000 Expenditure Expenditure E 57. 5. OFtne 7. E 56. OF 4. mnrevo 6. 55. etavir 3. P G 5. 2. Rs.10000 Rs.10000 Expenditure Expenditure Either Public or Private No Public or Private Both Public and Private Source: LEAPS Population Census, 2003 Note: This figure shows the fraction of children enrolled (top-left) and the share of each sector in that enrollment. The figure on the top-right shows the share of madrassas, the bottom-left the share of government and the bottom-right the share of the private sector. In every figure, we show the appropriate fraction for three types of settlement--those with both a private and a public school, those with either a private or public school (including settlements with both) and those with neither a private nor a public school. These settlements were constructed through a mapping exercise in every village, and for the 125 villages in our sample, we have 253 settlements. We plot the shares against the self- reported monthly expenditures of the household. More than 95 percent of all households fall below the Rs.7,500 to Rs.10,000 cut-off. 32 TABLE I COUNTRY-WIDE MADRASSA ENROLLMENT ­ DIFFERENT SOURCES Data Source Madrassa Enrollment Madrassa as Fraction of Enrolled Census of Population, 1998 Total 159,225 0.70% Male 111,085 0.82% Female 48,140 0.53% 0.78% PIHS 1991 151,546 [0.16%] 0.74% PIHS 1998 178,436 [0.089%] 0.7% PIHS 2001 176,061 [0.093%] Note: Survey standard errors in [brackets] where applicable. The census of population covers all of Pakistan except the Federally Administered Tribal Area (FATA). Included are Punjab, Balochistan, North-West Frontier Provinces (NWFP) and Sindh, plus the federal capital Islamabad and the federally administered Northern Areas and Azad Jammu and Kashmir (AJK). The Census of Population, 1998 estimates are based on the census "long-form", which was administered on a sample basis to a large number of households. This data is representative at the district level for both rural and urban regions. The next three rows show estimates from the Pakistan Integrated Household Survey (PIHS) which is a household survey and is representative only at the provincial level for the four main provinces, which account for 97% of the country's population--Punjab, Sindh, Balochistan and NWFP. 33 TABLE II ENROLLED CHILDREN IN 3 DISTRICTS School Type Data Source Attock Faisalabad Rahim Yar Khan Government (%) LEAPS 67.73 71.96 71.38 Private (%) LEAPS 31.56 27.33 24.92 Madrassa (%) LEAPS 0.71 0.70 3.70 Population Census, 1998 0.50 0.49 1.03 Source: LEAPS, 2003. Population Census, 1998. Note: LEAPS reports school type for enrolled children ages 5 ­ 15. Population Census reports field of education for children 5-14. LEAPS sample villages were randomly drawn from a list-frame of rural villages with at least one private school and thus are not representative of the district as a whole. 34 TABLE III CHARACTERISTICS OF MADRASSA AND NON-MADRASSA HOUSEHOLDS Monthly Expenditure Household Under Rs. Settlement Has Household Type Head Illiterate 7500 No Land Private School Non-Madrassa Household 0.4469 0.9598 0.6171 0.7196 (0.0024) (0.0009) (0.0023) (0.0021) Madrassa Household 0.5159 0.9645 0.6047 0.4852 (0.0153) (0.0056) (0.0149) (0.0152) Source: LEAPS, 2003. Note: Standard error reported in parenthesis. Households are classified as "Madrassa Households" if one or more children are currently enrolled in a madrassa. "Non-Madrassa" households have at least one child enrolled in government or private school. Monthly expenditure under Rs. 7500 accounts for 97% of households. There is no difference in means between household types when a more even expenditure categorization is used. 35 Appendix Tables TABLE A1 ARTICLES ON MADRASSAS IN PAKISTAN MAJOR NEWSPAPERS, DECEMBER 2000 -JUNE 2004 Source Date Type of study Numbers Reasons for Madrassa Enrollment L.A. Times 28 Case-Study 8,000 Madrassas Studied Madrassa December was originally set up 2000 for Afghan refugees Financial Times 6 March Interview with 10,000 Madrassas, 1 Welfare service to the 2001 President Musharraf million students poor L.A. Times 12 August General article None Welfare service to the 2001 poor L.A. Times 19 Case-Study 18,000 in Peshawar Religious September indoctrination 2001 The Washington 20 Case-Study None Religious Post September indoctrination 2001 The Boston 4 October Case Study 11,000 Madrassas, 1 Boom during Afghan Globe 2001 million students war. Fills a gap due to failed government education. Financial Times 17 October Discussion of 10,000 Madrassas Religious Extremism 2001 meeting between Colin Powell and President Musharraf The Philadelphia 9 November General article 7,000 ­ 8,000 None Inquirer 2001 Madrassas 700,000 students Financial Times 17 General article and 4,000 Deobandi Religious teaching November interviews Madrassas 2001 The Philadelphia 25 General article and 8,000 registered + On tribal border Inquirer November interviews 25,000 unregistered areas, madrassas 2001 instill Islam and preach hatred for non-Muslims The Boston 29 General article Fill gap in public. Globe November education, provide 2001 religious indoctrination L.A. Times 10 General article and 10,000 madrassas Religious December interviews that dominate indoctrination 2001 education throughout rural Pakistan The Philadelphia 16 Interviews in Pishin None Religious Inquirer December district indoctrination in 2001 radical Islam 36 Source Date Type of study Numbers Reasons for Madrassa Enrollment Chicago Tribune 23 General article and None Fill gap in public December interviews education, private 2001 schools too expensive. Chicago Tribune 23 None Overwhelmingly December popular with 2001 Pakistan's poor, fill a gap in public education The Boston 25 General article and 10's of thousands, 1 Humanitarian aid for Globe December interviews million children poor 2001 L.A. Times 3 January General article and 5,000 madrassas Religious 2002 interviews indoctrination. Madrassas as training ground for Afghan was with Soviet Union L.A. Times 4 January General article and None 2002 interviews The Philadelphia 5 January News item on 6,000 madrassas None Inquirer 2002 government Chicago Tribune 13 January News item on None None 2002 government The Boston 14 January News item + None Religious Globe 2002 Interview indoctrination Chicago Tribune 18 January Case Study None Radical religious 2002 indoctrination L.A. Times 19 January News Item None See next 2002 The Philadelphia 23 January New Item Thousands of Explosion during the Inquirer 2002 madrassas 1980s due to Afghan refugees + Radical religious indoctrination Chicago Tribune 24 January Case Study None Radical Islam 2002 Washington Post 14 March General article 500,000 plus children Failed public 2002 educational system; Religious studies Poor children + orphans Boston Globe 18 March General article Thousands of 2002 madrassas 37 Source Date Type of study Numbers Reasons for Madrassa Enrollment L.A. Times 23 March Case Study 3,700 (NWFP only) Islamic studies based 2002 on a reactionary curriculum; service to the extremely poor; failed public educational system Washington Post 28 April General article + 7,000 madrassas Built for Afghan war 2002 Case Study against Soviet Union L.A. Times 29 June General article 1.5 million students Resistance to Soviet 2002 Union + Only schooling option for boys from poor families Chicago Tribune 30 June General article 8,000 to 10,000 Resistance to Soviet 2002 Madrassas Union + Only schooling option for boys from poor families Washington Post 14 July 2002 General article and 10,000 madrassas Resistance to Soviet interviews 1.5 million students Union; fighters for Rapid growth in Kashmir; failure of recent years public education system; social charity function The Times 10 August Report 1.5 million students Hard-line Islamic 2002 from poor rural schools families L.A. Times 12 October Report 8,000 to 10,000 2002 madrassas with 1.5 million students L.A. Times 2 February Interview with Madrassas reflect 2003 Foreign Minister poverty of the state. Mian Khursheed Cater to poor people. Mehmood Kasuri Public educational system has failed Financial Times 8 February General article and 40,000 to 50,000 None 2003 interviews madrassas L.A. Times 14 April Report 10,000 madrassas, Provide education for 2003 educate 10% of all the poor; failures in Pakistani students public education system Financial Times 19 August Expert comment Most madrassas on 2003 border with Afghanistan. Concentrates on Pashtun belt. Washington Post 2 September Report None Talks about 2003 madrassas in the Pashtun tribal belt 38 Source Date Type of study Numbers Reasons for Madrassa Enrollment L.A. Times 5 March Report on politics Madrassas are not the Madrassas are not the 2004 regarding Musharraf focus focus Washington Post 13 June General article and 10,000 madrassas Social safety net; 2004 interviews Radical religious indoctrination The Philadelphia 15 June General article and 8,000 madrassas Afghan refugees Inquirer 2004 interviews moving into Pakistan 39 TABLE A2 DATA SOURCE DOCUMENTATION Dataset Year Advantages Disadvantages Used for PIHS 1991 Representative for 4 Very low numbers for Representative 1998 provinces that account children enrolled in numbers for madrassa 2001 for 97% of the madrassas. The low enrollment in the 4 population numbers make it hard provinces over time. to study associations at the level of the household. PEIP 2000 Dataset on private No household level Size of private school schools, collected at the information and no enrollment. Often school level by the information on combined with 1998 Federal Bureau of madrassas. Age of Population Census in Statistics. Provides enrolled children not our analysis. wealth of information available. on private schools, their location and enrollment. Population 1998 Country wide coverage Information is Numbers on country Census, at the level of the aggregated at the level wide enrollment and Long Form district and region of the district. geographical (rural/urban) excluding dispersion of madrassa FATA. enrollment in the country. Associations at the level of the district. LEAPS 2003 Very recent data Very limited Presenting some Census collected by the authors. coverage--only 3 recent numbers on Large number of districts in Punjab. madrassa enrollment. children surveyed Limited household Associations at the allows for flexibility in level information. level of the household. studying association Villages not between household representative of attributes and schooling districts or country. choice. 40 TABLE A3 LEAPS MADRASSA ENROLLMENT ESTIMATES Estimate Definition Madrassa as Fraction of Enrolled ATTOCK Conservative 0.0033 Moderate 0.0070 Liberal 0.0108 FAISALABAD Conservative 0.0033 Moderate 0.0069 Liberal 0.0116 RAHIM YAR KHAN Conservative 0.0213 Moderate 0.0367 Liberal 0.0583 TOTAL Conservative 0.0081 Moderate 0.0148 Liberal 0.0236 Source: LEAPS, 2003. Note: Madrassa enrollment can be determined from several variables in the LEAPS dataset. Given data-entry and field errors, different estimates of madrassa enrollment can be obtained. Throughout this paper we use the moderate estimate. The conservative and liberal estimates should be viewed only as extreme lower and upper bounds. 41