ISSN (Print) - 0012-9976 | ISSN (Online) - 2349-8846

A+| A| A-

Primary Completion Rates across Socio-Religious Communities in West Bengal

Primary completion rates of Muslims in West Bengal are substantially lower than that of upper caste communities as well as backward castes, scheduled castes and tribes. Further, analysis of age-specific pcr indicates that differences in pcr between Muslims and other communities may have actually increased in recent years. An econometric analysis reveals that age, gender, household size and expenditure levels, education and gender of decision-maker, etc, are important determinants of these differences in pcr. But use of Census data and District Information System for Education statistics indicates that deficiencies in infrastructural facilities in Muslim-concentrated districts also have a significant role in the low pcrs of Muslim children.


Primary Completion Rates across Socio-Religious Communities in West Bengal

Zakir Husain, Amrita Chatterjee

Primary completion rates of Muslims in West Bengal are substantially lower than that of upper caste communities as well as backward castes, scheduled castes and tribes. Further, analysis of age-specific PCR indicates that differences in PCR between Muslims and other communities may have actually increased in recent years. An econometric analysis reveals that age, gender, household size and expenditure levels, education and gender of decision-maker, etc, are important determinants of these differences in PCR. But use of Census data and District Information System for Education statistics indicates that deficiencies in infrastructural facilities in Muslim-concentrated districts also have a significant role in the low PCRs of Muslim children.

The authors would like to express their thanks to Debdas Banerjee and Achin Chakraborty for their comments and suggestions. Needless to say all deficiencies are the sole responsibility of the authors.

Zakir Husain ( and Amrita Chatterjee ( are with the Institute of Development Studies, Kolkata.

he importance of education as an important development goal has been widely recognised. It has been pointed out that education has important consequences for growth. By improving technical and cognitive skills it increases the productivity of the workforce and facilitates diffusion of technology and new ideas (Bruns et al 2003). Education also generates substantial externalities, like better natural resource management ( Godoy and Contreras 2001) and lower incidence of HIV/AIDS (World Bank 2002). In particular, education of girls improves health outcomes for infants and children, family nutrition levels and (in subsequent generations) educational outcomes (Hadden and London 1996). Empirical studies have also noted the high correlation between education and lower levels of poverty and inequality (Birdsall and Londoño 1998). Not only does education facilitate sustained growth and development, but as pointed out by Sen (1985, 1999), it is important as an independent goal, by improving human capabilities.1

1 Introduction

Recognising the importance of education, policymakers initially emphasised increasing access to schooling. Over time, however, it was realised that non-attendance, grade repetition, dropouts and poor retention rates were reducing the value of gross enrolment rates as an indicator of progress on the education front. The 1990 World Conference on Education for All in Jomtien, Thailand, therefore, emphasised on providing universal primary education to ensure the reaping of the substantial benefits associated with education.2

Despite the attempt to increase primary completion rates (PCR) after the Jontien Conference, the progress made on this front has been modest – at the global level, PCR has increased from 72% from 1990 to 77% in 2000. South Asia has the lowest regional PCR (70%) after Africa (Bruns et al 2003), while PCR in I ndia is even lower at 66%, according to the 2001 Census. While a concerted attempt is needed to improve PCR, intervention strategies must be based on an understanding of the factors responsible for low PCRs. In particular, sections of communities with low PCR should be identified and an attempt made to address the causes underlying their relative deprivation.

There is a substantial body of literature on factors influencing probability of primary completion. Based on limited dependent regression models,3 studies have identified factors like family i ncome or wealth, parental education, empowerment and education of mother, credit constraints, age and gender of the child, family size or presence of siblings, caste affiliations, place of

Economic & Political Weekly

april 11, 2009 vol xliv no 15

r esidence and educational infrastructure as determinants of PCR (Akhtar 1996; Deolalikar 1997; Tansel 1998; Brown and Park 2002; Connelly and Zheng 2003; Boissiere 2004; Desai and Kulkarni 2005; SIS/DPP 2005, Das and Mukherjee 2007, 2008; Okumu et al 2008). One limitation of these studies is that they have failed to explore the importance of religious identity in educational decision-making. In developing countries, particularly those in south Asia, religion play an important role in determining educational attainment. In India, for instance, a recent report has shown that PCRs of Muslims is substantially below that of other communities (GOI 2006).

The objective of this paper is to probe deeper into this issue by examining the extent and causes of differences in probability of

Table 1: Share of Socio-Religious Communities in Population in West Bengal (2004)

Analytical Categories used Socio-Religious Communities Rural Urban

Hindu-upper caste (H-UC) Hindu – upper caste 27.35 12.64
Hindu-backward castes (H-BC) Hindu – scheduled castes 27.17 19.32
Hindu – scheduled tribes 6.37 1.21
Hindu – other backward castes 6.08 5.95
Muslims Muslims 31.14 14.00
Others Other minorities 0.99 1.01
Religion/caste not stated 0.91 0.36
Total population 100% 100%

completing primary level of education between Muslims and other socio-religious communities (SRCs). We consider only one state, West Bengal, as there are considerable differences in the historical and socio-cultural environment across regions. The f ocus on West Bengal may be justified on the following grounds:

  • (a) The absolute size of the minority population is higher in West Bengal than in other Indian states (except Uttar Pradesh);
  • (b) The proportion of minorities in total population is higher in West Bengal than in any other major state. Only some small states like Assam, Jammu and Kashmir, north-eastern states and some Union Territories have a higher proportion of minorities;
  • (c) Ideologically, the Left Front coalition government is committed to secular principles, and its electoral manifesto emphasises upliftment of marginalised sections of the population; and,
  • (d) Political stability has given the state government the opportunity to implement its objectives of social justice and progress through long-term plans.
  • The scheme of this paper is as follows: Section 2 states the d atabase and methodology used in the paper. Section 3 presents estimates of PCR among children aged 12-15 years for different analytical categories of the population; this is followed by a discussion of age-specific PCRs. In Section 4 we use standard econometric techniques to identify determinants of probability of PCRs. The next section “decomposes” differences in probability of PCR between “explained” and “unexplained” components.4 The latter may be due to lack of infrastructure in Muslim-dominated areas or due to lack of demand. Section 6 examines both these issues.

    2 Database, Methodology and Hypothesis

    This paper is based on unit level data from the National Sample Survey (NSS) 61st round (2004-05). This database has information on the religion and caste of the respondents. These two variables have been combined to create socio-religious communities

    60 for analytical purposes. The socio-religious communities are u pper caste Hindus, Hindu scheduled caste, scheduled tribes and other backward castes, Muslims, other minorities and a residual group who had not stated their religion. These categories are r eclassified into four socio-religious communities, given in Table 1.5

    The state-level figures are decomposed by gender and place of residence to get four analytical categories – urban males, urban females, rural males and rural females. Our analysis is undertaken for each of these categories as the process of educational decisions may vary across these categories.

    The analysis is undertaken based on the following hypotheses:

  • (1) There is a significant disparity in PCRs between Muslims and other communities;
  • (2) Within the Muslim community, girls constitute a particularly vulnerable section;
  • (3) The low current PCR of Muslims is in line with a history of backwardness; and,
  • (4) Educational backwardness of Muslims may be explained only partially by household and personal characteristics (like household size, per capita income, gender and education level of head of family, gender of the child, his/her age).
  • Two sets of methodologies are used – descriptive analysis and econometric. Firstly, the NSS data is used to estimate the proportion of children completing primary education out of children aged 12-15 years.6 Following the Bruns et al (2003) definition, the PCR is defined as the ratio of number of children completing primary level of education in the appropriate age group to number of children in the age group.7 This is followed by estimation of age-specific primary completion rates as a proxy for time trends in primary completion rates. The methodology is explained in Section 3.2.1.

    Secondly, we attempt to identify the factors influencing enrolment, completion of primary education and choice of schools based on econometric methods. The analysis is undertaken using logit models. Subsequently, the results are decomposed by the extension of the Oaxaca method (Oaxaca 1973) applicable to logit models suggested by Bauer and Sinning (2008) and Fairlie (1999, 2003, 2005).

    3 Main Findings

    This section discusses the main findings of the analysis.

    3.1 Status of Primary Education Completion

    The PCR among children aged 12 to 15 years is 70% in West B engal. However, minority communities have a substantially lower PCR – it is 37% for Muslim children and 35% for children from others. In contrast, PCR is higher among not only H-UC (51%), but also H-BC (45%).

    Table 2: Primary Completion Rates among Children Aged 12-15 Years (2004)

    Socio-Religious Urban Communities Total Boy Girls Muslim 65.8 61.2 70.5 H-UC 86.8 88.4 85.1 H-BC 72.8 71.4 74.2 Others 100.0 100.0 100.0 Total 78.8 78.3 79.3 april 11, 2009 vol xliv no 15 Rural Total Boy Girls 65.0 61.8 68.1 80.5 80.9 80.0 69.4 73.9 64.6 51.4 58.8 44.4 70.0 70.8 69.3 Economic & Political Weekly

    Disaggregation of state-level figures by place of residence and gender reveal a similar picture of educational backwardness of Muslims (Table 2, p 60). Muslims have a lower PCR than both H-UC and H-BC in both urban and rural areas. The gap is relatively wider in urban areas. This may be because of greater p overty among urban Muslims (GOI 2006) and the presence of a substantial non-Bengali population, who are believed to attach lesser value to education.

    The observed differences in PCR across gender go against the well-known belief that Muslim parents are only interested in e ducation of their boy child – primary completion rates are higher among Muslim girls than among Muslim boys. In rural areas, in particular, PCR of Muslim girls is higher than that of H-BC and other minorities. Overall, however, gender differences are not marked.

    3.2 Trends in Primary Education Completion Rates

    First the methodology and then the findings of the analysis:

    3.2.1 Methodology

    Although this paper focuses on the educational status of SRCs in 2004, it is necessary to place the findings in a temporal context. This requires us to examine changes in status of education of SRCs over the years. Unfortunately, NSS neither provides time series data nor does it have sufficient large number of rounds containing religion-wise data.8 Therefore to create a time series on educational attainment we have followed the methodology suggested by Shariff (1999).

    As NSS provides data on age and education level of respondents, we may estimate the percentage of persons completing primary education in different age groups. Such age-specific primary education completion rates may be used to estimate the primary

    Among urban females, H-UC historically has the highest PCRs. Among Muslims, PCR has increased steadily from the late 1980s, but has decreased thereafter. Currently, it is about 71%, which is lower than even the PCR of H-BCs. The rate of increase in PCR among H-BCs is sharply pronounced from the 1980s; as a result their PCR has converged with that of H-UCs.

    Figure 1a: Trends in PCR in West Bengal – Urban Males

    (% population completing primary level)

    1950 1960 1970 1980 1990 2000 2010 100 80 60 40 20 0 Muslim H-UC H-BC

    Figure 1b: Trends in PCR in West Bengal – Urban Females

    (% population completing primary level)

    1950 1960 1970 1980 1990 2000 2010 100 80 60 40 20 0 Muslim H-UC H-BC

    A similar picture of elative deprivation is observed for Muslim males in rural areas. While PCR has increas ed for all the three SRCs, it has remained highest for H-UC and lowest for Muslims. While H-BCs were educationally backward in the 1950s, they have almost managed to converge with H-UC in recent years. Again, the PCR of Muslims is significantly lower than for other communities. In fact, after 2000, PCR has actually declined.9

    education completion rates (PCR) at dif- Table 3: Conversion of Age Groups into Years The picture for rural females is inter
    ferent points of time as follows. We as- Age Group (in 2004) Year Age Group (in 2004) Year esting. While PCR has increased for all
    sume that a person will complete his/her 12-15 2004 41-45 1975 SRCs, Muslim females have performed
    primary level within 12 to 15 years. Given 16-20 2000 46-50 1970 quite well. By and large, Muslim females
    that a person aged 16 to 20 years in 2004, 21-25 1995 51-55 1965 have performed better than H-BC. Although
    when the NSS 61st round survey was conducted, it is an easy task to compute the 26-30 31-35 36-40 1990 1985 1980 56-60 61 and above 1960 1958 their PCR is lower than that of H-UC, the difference seems to have narrowed some

    year (2000) when that person would have aged 12 to 15 years, and should have completed his primary education. Proceeding in this way, the corresponding year in which each age group had attained 12 to 15 years, can be calculated. This is illustrated in Table 3.

    Since the sample size of others may be small in some age groups, results for this community may be misleading. So the analysis is undertaken only for H-UC, H-BCs and Muslims.

    3.2.2 Findings

    Figure 1a presents the trend among urban males. The PCR for H-UC has remained steady at above 80%. On the whole, PCR for H-BC exhibits an upward rising trend, although there are sharp fluctuations in some years. In the case of Muslims, PCR had i ncreased from the 1970s, but has decreased from the mid-1990s. It should also be noted that PCR of Muslims has generally r emained lower than even that of H-BCs.

    Economic & Political Weekly

    april 11, 2009 vol xliv no 15

    what over the last two decades.

    Thus, age-specific analysis of PCRs indicates that low PCR among Muslims is not a recent phenomenon. Statistical tests (based on the standard normal statistic, τ) clearly show that such disparities have been a persistent feature of West Bengal society since Independence. After Partition the more educated among the Muslims migrated to East Pakistan so that the education level of the entire community fell below that of even H-BCs. Unfortunately, while PCR has risen for all communities, it has risen at a slower rate among Muslims. The consequent divergence between PCR of Muslims and PCR of other communities constitutes an important area for policy intervention.

    Another finding with important policy implications is that, from the late 1990s onwards, there has actually been a decline in the PCR of Muslim males. This decline is remarkable in urban areas, where the decline is about 17%! Unless there has been a sharp change in attitude of Muslims,10 the explanation has to be sought in supply side factors.

    4 Econometric Analysis

    This section carries out an econometric analysis on the factors determining the probability of completing primary education.

    4.1 Variables

    We next undertake an econometric analysis to identify the factors determining probability of completing primary education. Based on the literature referred to in Section 1, we regress probability of completing primary education on the following independent v ariables:

    (1) Monthly family expenditure (MFE): Higher levels of MFE mean that the household has the resources to invest in education of children. This is important, given the high costs of the supposedly “free” primary education (PROBE 1999; NSS 1998). Thus, we

    Figure 1c: Trends in PCR in West Bengal – Rural Males

    (% population completing primary level)






    0 1950 1960 1970 1980 1990 2000 2010

    H-UC H-BC Muslim

    Figure 1d: Trends in PCR in West Bengal – Rural Females

    (% population completing primary level)

    100 80 60 40 20 0 Muslim H-UC H-BC

    1950 1960 1970 1980 1990 2000 2010

    expect that MFE will have a positive effect on enrolment rate and completion rate of primary education. Correspondingly, dropout should be lower with high levels of MFE. Given the high values of MFE with respect to the values of dependent variables, we have taken the logarithmic transformation of MFE (LMFE).

  • (2) Household size (HHSIZE): The household size is likely to affect educational status adversely, unless the dependency ratio is low.11
  • (3) Gender (FML): There is a vast body of literature (PROBE 1999) arguing that families tend to invest less in the education of the girl child. To verify this hypothesis, we have constructed a gender dummy, FML equal to unity if the child is female and 0 otherwise.
  • (4) Age (AGE): As a child becomes older it becomes more difficult for him/her to join school. This is partly because of incompatibility with other children of the same class, or because of loss of i nterest in schooling, or because parents tend to engage him/her in productive or household tasks, rather than keep him/her idle (increasing opportunity costs of schooling). After getting e nrolled, the probability of dropout increases as the child grows older. This may be because parents have a target level of education that is typically low, or because they want to divert the
  • 62 financial resources for schooling to younger siblings (Husain 2005). However, the probability of his/her completing primary level should be higher than younger children.

    (5) Educational level of head of household (EDU _ DM): Literature shows that if the decision-maker is him(her)-self educated then he(she) is more likely to educate his(her) children. While studies generally focus on literacy we have taken different levels of edu


    cation as defined in NSS.

  • (6) Gender of the head of the household (SEX _ DM): Gender of the head of the household also affects education-related decisions. If a female is the decision-maker, it may be expected that she is more responsive to the needs of educating a child, particularly the girl child.13 On the other hand, if she is also the main income earner then she may have to remove girls (to undertake household chores) and even boys (if gender discrimination in the l abour market results in her getting relatively lower wages, then family income may have to be supplemented).
  • (7) Socio-Religious Community: Recent literature has shown the presence of inequality in educational attainments at the all-India level between socio-religious communities (Desai and Kulkarni 2005; Sengupta and Guha 2006; GOI 2006; Alam and Raju 2007). This is also supported in our findings for West Bengal. We have therefore constructed dummies to capture differences in educational attainments between SRCs, taking Muslims as the reference category: H-UC = 1 if respondent belongs to Hindu upper caste, = 0 otherwise H-BC = 1 if respondent belongs to Hindu backward caste,14 = 0 otherwise Others = 1 if respondent belongs to non-Muslim minorities and others, = 0 otherwise
  • 4.2 Econometric Model

    The dependent variable – whether the child has completed primary level or not – is a binary variable, assuming values of 1 (if the child has completed primary level) or 0 (if the child has failed to complete primary education). In this situation, logit or probit models are commonly used, estimated using the maximum likelihood method. These models differ with respect to specification of the error term – if the error term follows a logistic distribution logit model is used, while probit model is used if the error term follows a normal distribution. Since the cumulative normal and logistic distributions are very close to each other except at the tails, we are not likely to get very different results for the logit and probit models. However, while the results of the nonlinear decomposition carried out subsequently has been theoretically verified for the logit model, they have been only observed as an empirical regularity for probit models (Fairlie 2005). We have therefore used the logit model in our analysis. The results of this econometric exercise are given in the appendix, while a discussion of the results is undertaken in the subsequent sub-section.

    4.3 Results

    Three models are estimated – for the entire state (incorporating a dummy for urban areas), the rural population and the urban p opulation.

    april 11, 2009 vol xliv no 15

    The results of the econometric exercise are, by and large, as h ypothesised (Table 4). It can be seen that family expenditure levels, household size, age of child and educational level of d ecisionmaker are statistically significant and affect PCR as h ypothesised. PCR is not statistically lower for girls,15 or if the d ecision-maker is a

    Table 4: Determinants of PCR

    Variables Hypothesised Model 1: West Bengal Model 2: Rural Model 3: Urban
    Family income (log) + + + +
    Household size - - - -
    Female child - Insig Insig Insig
    Age of child + + + +
    Education of decision-maker + + + +
    Female decision-maker + Insig Insig Insig
    Urban + Insig NA NA
    H-UC + + + +
    H-BC + Insig Insig
    Others - - - No variation

    female. The former finding is consistent with the low difference in PCR of boys and girls in both rural and urban areas, observed in Table 1 earlier. The place of residence is not significant in Model 1, despite sharp differences in rural and urban PCRs (Table 1). The reason for differences observed in Table 1 is mainly due to differences in household expenditure level. Controlling for this difference, the rural-urban difference in PCR is eliminated.

    The results also show that socio-religious identity is important in determining PCR. Upper caste Hindu children have a higher probability of completing primary level than Muslim children. While H-BC have a higher probability of completing primary level than Muslims in model 1, the difference is statistically insignificant when regression models are run for rural and urban areas separately. Children from non-Muslim minority communities, however, have an even lower probability of completing primary education in rural areas than Muslim children.

    5 Decomposing Differences in PCRs across Communities

    Now the differences in PCRs between communities observed in this analysis may be explained in terms of household and individual characteristics. It may also be attributed to “unexplained” factors – lack of demand for education or inadequate educational infrastructure in areas with high concentration of Muslims. This section estimates the contribution of household and individual characteristics in explaining differences in PCR between Muslims and H-UC.

    The issue of unravelling the relative contribution of explained (that is, dependent variables) and unexplained variables was first addressed by Blinder (1973) and Oaxaca (1973) in the context of discrimination in wages in the labour market. While a technical discussion is undertaken in Appendix B, an intuitive explanation of the issues is presented here. Starting with the proposition that differences in wages could be attributed to differences in endowments (given identical regression coefficients) and to differences in regression coefficients (given identical endowments), they a ttempted to estimate the relative contribution of each of these factors. An important problem that arises in this context is which coefficient vector to use in the decomposition process. While it was initially suggested (Oaxaca 1973) that the coefficient vector of either of the two groups be taken, there have been alternate

    Economic & Political Weekly

    april 11, 2009 vol xliv no 15

    suggestions. For instance, while Remiers (1983) proposes that the simple average of the two groups be used, Cotton (1988) suggests using the relative sample size of the majority group as weight. On the other hand, Neumark (1988) and Oaxaca and Ransom (1994) derive the coefficient vector based on the pooled sample.

    The Oaxaca decomposition is, however, not appropriate in the non-linear case. There has been several attempts to develop a ppropriate decomposition techniques in the non-linear case (Gomulka and Stern 1990; Yun 2004; Fairlie 1999; 2005; Bauer and Sinning 2008), and corresponding decomposition packages.

    Differences in probability of completing primary level of education between Muslims and H-UC are decomposed using the n ldecompose package (after minor modifications) in STATA.16 The explained proportion and the residual are given in Table 5.

    Table 5: Decomposition of Differences in Probability of Completing Primary Level

    Reference Category Components Rural Urban
    ‘Backward’ group Explained 69.94 69.59
    Residual 30.06 30.41
    ‘Advanced’ group Explained 71.71 61.41
    Residual 28.29 38.59
    Reimers (1983): Productivity* 71.26 66.24
    Simple average Advantaged** 14.86 15.02
    Disadvantaged*** 13.88 18.74
    Cotton (1988): Productivity 71.3 63.59
    Average, weighted by relative Advantaged 14.26 6.18
    sample size of groups Disadvantaged 14.44 30.24
    Neumark (1988): Productivity 73.84 66.81
    Pooled sample Advantaged 13.6 6.9
    Disadvantaged 12.56 26.28

    * Corresponding to variation explained by endowments. ** Represents the unexplained advantage of the superior group. *** Represents disadvantage of the inferior group.

    One of the disadvantages of the nldecompose package is that it fails to identify the contribution of individual variables to the e xplained difference. This deficiency is covered in the fairlie package, based on Fairlie (1999, 2005), developed by Ben Jann.17 We have used this package to identify contribution of individual variables for pooled sample – corresponding to the method suggested by Neumark (1988). The results are stated in Appendix A (p 67).

    The results show that factors like expenditure levels, household size, age of child and the educational level and gender of the decision-maker operate to create a gap in PCR between Muslims and H-UC. However, the contribution of gender of the child is interesting.18 In rural areas, it does not contribute to differences in PCR; in urban areas it actually reduces differences in PCR! In other words, after controlling for other factors, a girl child has a higher probability of completing the primary level if she is a Muslim, than if she belongs to H-UC community. While this does not support widely held notions about gender discrimination within the Muslim community, the finding is consistent with the earlier s ections and also with field observations during surveys of K olkata slums in 2003 and 2008.

    6 Alternative Explanations of the ‘Residual’ Component

    The results of Section 5 show that a significant proportion (about 30%) of the differences in probability of completing p rimary level remains unexplained by household and individual characteristics. This raises the question, why is PCR lower among Unfortunately, there is no reliable method by which we can Muslims, vis-à-vis other communities, even after controlling for measure demand for education, given a possible gap between factors like expenditure, household size, education and gender of a ctual demand for education and the revealed demand for educafamily head, and age and gender of the child. tion. A distinction between two possible situations becomes rele

    vant in this context:

    Table 6: Educational Infrastructure in Minority Concentrated and Other Districts of West Bengal (2001) (a) Muslims may not be availing of educational facilities exist-

    District Name % Minority Schools Population ('000

    ing in this area as they feel that education is not important. In

    Population Covered by Schools)

    such a situation, revealed demand is an acceptable measure of

    Govt Total Govt Total

    Schools Schools Schools Schools

    actual demand.

    Non-minority dominated districts

    (b) Muslims recognise the value of education, and demand

    Puruliya 7.5 2,985 3,313 0.8 0.85

    more facilities than existing currently. In this case, demand

    Bankura 7.7 3,914 3,914 0.8 0.82

    r evealed through enrolment figures will be much less than actual

    East Medinipur 11.6 3,490 3,878 2.5 2.75


    Hooghly 15.3 3,393 3,565 1.4 1.49

    In other words, we may have a truncated demand curve for

    Jalpaiguri 16.6 2,233 2,302 1.5 1.52 Bardhaman 20.4 4,138 4,800 1.4 education when we consider enrolment-based figures.

    1.67 Kolkata 22.1 1,532 2,007 2.3 2.98In this situation, figures relating to indicators like percentage Darjeeling 22.7 1,371 1,388 1.2 1.17of schools with more than 60 students per classroom, or with

    West Medinipur 23.7 4,831 5,392 -0.03 student-teacher ratio exceeding 100, etc. may provide a more Total 27,887 30,559 1.3 1.5

    r eliable sign of whether there is an infrastructural deficiency

    Minority dominated districts

    compared to demand for education. Estimates from district infor

    Cooch Bihar 24.4 1,872 2,070 1.2 1.32

    mation for school education (DISE) statistics for the year 2005-06

    North 24 Parganas 24.7 4,049 4,644 1.9 2.21

    (Table 7) reveal that Muslim-concentrated districts perform

    Howrah 24.9 2,148 2,627 1.6 1.99

    poorly relative to other districts for all these indicators.

    South Dinajpur 25.5 1,351 1,351 1.1 1.11 Nadia 26.1 2,787 3,063 1.5 1.65 This indicates that while population growth led to increase in South 24 Parganas 34.1 3,604 4,396 1.6 1.92demand for more educational facilities, the State government

    Birbhum 35.4 2,371 2,774 1.1 1.27has failed to respond adequately, particularly in Muslim North Dinajpur 48.0 1,438 1,632 1.5 1.70concentrated districts. Consequently, the gap between demand Maldah 50.0 1,909 2,221 1.5 1.72

    for educational facilities and infrastructure available is greater in

    Murshidabad 64.0 3,170 3,682 1.6 1.85

    Table 7: Indicators of Deficiency between Educational Infrastructure

    Total 24,699 28,460 1.5 1.7

    and Demand (2005-06)

    Source: Estimated from Census 2001 data: Table C1 (Population by Religious Community – Blockwise) and Table on Educational Amenities in West Bengal (

    District % Schools % Schools Student Percentage Percentage % DataTables/08/FrameTables-e_1.htm). with More with Teacher of Enrolled of Enrolled Minority Than 60 Student Ratio (in Childern Childern Population Two alternative explanations may be examined. Traditionally, Students in a Teacher Govt Studying Studying Class Room Ratio > 101 Schools) in Single in Schools

    it is argued that Muslims do not recognise the value of education,

    Teacher without particularly mainstream education (Hunter 1869; Baig 1974; Building


    Murshidabad 73.6 13.15 57.2 0.1 0.3 63.92

    Sharma 1978; Jehangir 1991; Salamatullah 1994; Sengupta and

    Maldah 75.3 25.56 69.3 0.5 1.4 49.99

    Guha 2006). This is manifested in low rates of enrolment in

    North Dinajpur 85.1 22.11 65.2 1.0 0.0 47.93

    mainstream schools and preference for Madrasahs (Jehangir

    Birbhum 47.8 2.74 39.9 1.2 0.2 35.35

    1991; Ruhela 1998; Salamatullah 1994; Bandhopadhyay 2002).19

    South 24 Parganas 67.9 31.91 74.1 9.3 0.4 34.06

    This view has been questioned in recent years (Alam and Raju

    Zone A 69.9 19.1 61.1 2.4 0.5 46.2

    2007).20 The High Level Committee Report (GOI 2006) has also

    Nadia 58.1 3.44 47.9 0.5 0.3 26.09 identified the lack of an adequate number of schools and infra-South Dinajpur 45.7 6.14 50.9 1.6 0.0 25.51 structural facilities in Muslim-dominated areas as a major cause Howrah 48.7 4.33 40.4 0.4 0.0 24.70 for educational backwardness of this community. North 24 Parganas 51.9 9.21 54.3 1.1 0.4 24.63 Cooch Bihar 50.6 8.01 49.3 0.1 0.2 24.36

    Analysis of census data reveals that population coverage of

    Darjeeling 12.2 5.97 34.3 10.9 1.3 22.63

    schools (population in thousands, divided by number of schools)

    Bardhaman 29.7 1.23 41.9 0.5 0.3 20.36

    is lower in Muslim-dominated districts (Table 6). Moreover, in

    Zone B 42.4 5.5 45.6 2.2 0.4 24.0

    Muslim-dominated areas, the availability of basic infrastructure

    Jalpaiguri 57.0 6.74 55.8 1.0 0.3 16.53

    (like benches, boards, chalk, duster, toilets) and personnel is

    Hooghly 28.5 3.89 48.8 1.2 0.1 15.30

    o ften poor (GOI 2006). West Medinipur 24.4 3.71 41.7 2.9 0.4 13.41 Population-school ratio by schools, however, is not enough to East Medinipur 20.7 18.94 62.3 0.5 0.2 10.07 refute the value theory. If Muslims really do not acknowledge the Bankura 28.3 1.97 40.5 3.6 0.3 7.61

    value of education, they will not enrol their children in schools. Puruliya 29.3 5.39 45.1 7.8 0.9 7.43 Zone C 31.4 6.8 49.0 2.8 0.3 11.7

    Given scarce resources, a geographical distribution of schools

    Kolkata 11.3 2.81 36.9 1.1 0.0 21.63

    b iased against Muslim-concentrated areas may be optimal from

    Source: Estimated from District Information for School Education, District Report Cards, 2005-06. the utilisation point of view.21 Accessed at on 1 October 2007.

    64 april 11, 2009 vol xliv no 15

    Muslim-concentrated districts. This may have resulted in a lower-The relative importance of demand for education and infraing of primary completion rates of Muslims below levels warranted structural deficiencies are examined in section 6. Since the by their economic status and demographic characteristics. d emand revealed through enrolment is a truncated version of a ctual demand for education, we look at indicators reflecting

    7 Conclusions

    infra structural gap vis-à-vis revealed demand. The proportion of The analysis undertaken in this paper provides useful insights schools in Muslim dominated districts with high student-classinto the relative status of Muslims and non-Muslim minorities in room ratio and student-teacher ratio, or proportion of children West Bengal with regard to primary educational attainments. In enrolled in single classroom schools or schools without buildings Section 3 we had seen that PCR of Muslims is lower than that of in such areas indicate government lacunae in addressing the not only H-UC, but also of H-BCs in both rural and urban West e ducational needs of Muslims. While the Central and State gov-Bengal. Statistical tests clearly show that these rates have ernments have emphasised on schemes like the merit-cum-means r emained significantly lower than that within H-UC and H-BC scholarship and on expanding madrasa education, such meascommunities over time in West Bengal.22 While Section 4 shows ures are inadequate.23 As a result, Muslims are slipping behind that such differences are due to household and personal charac-other communities in education. The lack of human capital is getteristics like lower income levels, larger household size, lower ting reflected in their inability to shift to the formal sector and to educational level of household head, etc, such differences a ccount more remunerative jobs, and is a major factor underlying the high for only about 70% of variations in probability of primary com-levels of poverty (GOI 2006; Husain 2008) and overall economic pletion between Muslims and upper caste Hindus. backwardness found in both rural and urban West Bengal.

    Notes 11 It may be better to take the dependency ratio, Interestingly, NSS data reveals that wages paid to
    1 Capability refers to the power to reflect, make rather than household size. However, dependency Muslims is substantially lower than that paid to
    choices, seek to participate in the civil society and ratio will have to be derived manually – making it upper caste and even backward caste com
    enjoy a better life. an extremely cumbersome, though not impossi munities for nearly all educational levels in urban
    2 3 4 5 6 7 Primary education corresponds to at least four or six years of education. In this paper, we have d efined primary education in terms of the first five years of education. Logistic models are generally used. Akhtar (1996) uses hazard rate analysis. By “explained” we mean the proportion of difference that may be attributed to differences in “endowments” – the determinants of PCR used in the regression method. The small proportion of other minorities (about 1% in both rural and urban areas in West Bengal) and others (which is even lower, see Table 1) means that any meaningful analysis of these categories separately will not be possible. We have therefore clubbed them together, along with those who did not reveal their socio-religious identity, as “others”. Similarly, Hindu scheduled castes, tribes, and other backward castes have been clubbed together to form a category called “Hindu backward castes” (H-BC). The age group taken in this case is 12 to 15 years, as children aged 6 or slightly more have no chance of completing primary education. Assuming enrolment at 6 to 10 years and 5 years of education, a child should have completed primary education by 15 years. Incidentally, there are no failures in Government schools in the West B engal. All s tudents are promoted irrespective of academic performance, thereby ruling out grade repetition. Alternatively, we could have taken only the number of enrolled children as the denominator. However, this implies overlooking the considerable number of children without access to schooling. 12 13 14 15 16 17 18 19 ble, task. Hence we have taken household size, rather than dependency ratio. The variable is therefore a continuous variable, representing illiterate (1), literate without formal schooling: EGS/NFEC/AEC (2), TLC (3), literate: below primary (5), primary (6), middle (7), secondary (8), higher secondary (10), diploma/certificate course (11), graduate (12), postgraduate and above (13). For instance, Das and Mukherjee (2007, 2008) a rgue, based on all-India NSS data, that empowerment of women will increase enrolment. Defined as SC, ST and OBC. Similar findings questioning the popular belief of gender discrimination in education within M uslim communities have been observed in the I slamic country of Pakistan (Akhtar 1996). Available in Mathias Sinning’s homepage: http:// html. Accessed on 17 November 2008. Fairlie: Stata module to generate nonlinear decomposition of binary outcome differentials. Available: /bocode/ s456727.html. Accessed on 17 November 2008. Gender of child is defined as =0 for the boy child, and = 1 for the girl child. Based on a study of slum-dwellers in Kolkata, Husain (2005) argued that Muslims do recognise the importance of education, but perceived discrimination in the labour market reduces expected employability and/or wages paid to Muslims relative to other communities. This lowers perceived r eturns to education, and hence effective demand. 20 21 22 23 West Bengal. Only among illiterates and matriculates were differences marginal, while among postgraduate Muslims earned higher wages ( Husain 2008: pp 42). Coupled with the fact that Muslims are underrepresented in the formal s ector (particularly in the government sector) there seems to be prima facie grounds for this p erception. Analysis of NSS data reveals that parents of only 2.6% children aged 6 to 12 years believed that e ducation was unimportant. In 70% of such cases, parents were illiterates. For instance, during deliberations on the Annual Plan for 2007-08 for West Bengal, officials of the school education board argued before the State Planning Board that there was no point in setting up schools in Muslim-dominated areas as Muslim parents were not interested in sending their children to government schools. This contrasts interestingly to the finding that more than 80% of rural Muslims are enrolled in government schools. The case of West Bengal is not isolated. Increasing disparities in primary completion rates between Muslims and other SRCs (and even completion rates for higher levels of education) have been observed at the all-India level and most of the major states (GOI 2006). Officials of the minority affairs department, government of West Bengal, r evealed, during the State Planning Board meetings to finalise allocations for 2008-09, that in 2008 about 7.5 lakh m inority candidates (nearly all of whom were Muslims) applied for the pre-matric merits-cum-means scholarship. Only about 44,460 (representing less than 6% of

    8 Religion-wise data is being collected in NSS surveys only since 1987-88.

    9 The τ-value (-2.0507) of the difference between proportion of Muslims completing primary level in 2000 and 2004 is statistically significant.

    10 This may have been caused by greater alienation from the mainstream following the 1992 riots, subsequent communalisation of Indian politics and increasing scepticism about the sincerity of the Left Front Government to improve the conditions of Muslims. During primary surveys of Muslim-dominated Kolkata slums (2003, 2007-08), a sense of frustration about the Left Front within the community could be perceived. These factors may have led to a reduction in perceived benefits to education (in terms of increased probability of securing employment).

    Review of Women’s Studies

    (Forthcoming – April 25, 2009)
    Women’s Citizenship Still Mired in the Private-Public Dichotomy Maithreyi Krishnaraj
    Maya Machhindra and Amarjyoti: Reaffirmation of the Normative Vaishali Diwakar
    Exploring Gender, Hindutva and Seva:
    A Case Study of Dnyana Prabodhini Swati Dyahadroy
    Gendered Citizenship and Women’s Movement Anurekha Chari
    Making of the ‘Bajarbasavi’: Reading Devadasi Practice through
    Popular Marathi Literature Anagha Tambe

    Economic & Political Weekly

    april 11, 2009 vol xliv no 15

    applicants) would be awarded scholarships, amounting to Rs 1,100 per annum. It should also be noted that the amounts will be r eleased only after about a year from the date of application. The dubious utility of expanding m adrasa education may be seen from the fact that only 4.6% of Muslim children aged 6-19 years are enrolled in such institutions (GOI 2006).


    Akhtar, S (1996): “Do Girls Have a Higher School Drop-out Rate Than Boys? A Hazard Rate Analysis of Evidence from a Third World City”, Urban Studies, 33(1): 49-62.

    Alam, M S and Saraswati Raju (2007): “Contextualising Inter-, Intra-Religious and Gendered Literacy and Educational Disparities in Rural Bihar”, Economic & Political Weekly, Vol 42 (18), pp 1613-22.

    Bandyopadhyay, D (2002): “Madrasa Education and the Condition of Indian Muslim”, Economic & Political Weekly, 37(16): 1481-84.

    Baig, M R A (1974): Muslim Dilemma in India (Delhi: Vikas Publication).

    Bauer, T K and M Sinning (2008): “An Extension of the Blinder-Oaxaca Decomposition to Non-Linear Models”, Advances in Statistical Analysis, 92: 197-206.

    Birdsall, N and J L Londoño (1998): “No Trade-off: E fficient Growth via More Equal Human Capital Accumulation in Latin America” in N Birdsall, C Graham and R Sabot (ed.), Beyond Trade-offs: Market Reform and Equitable Growth in Latin America (Washington DC: Brookings Institution and Inter-American Development Bank).

    Blinder, A S (1973): “Wage Discrimination: Reduced Form and Structural Estimates”, Journal of H uman Resources, 8: 436-55.

    Boissiere, M (2004): “Determinants of Primary Education Outcomes in Developing Countries”, Background paper for evaluation of the World Bank’s support to primary education, The World Bank, Washington.

    Brown, P H and A Park (2002): “Education and Poverty in Rural China”, Economics of Education Review, 21(6): 523-41.

    Bruns, B, A Mingat and R Rakotomalala (2003): “Achieving Universal Primary Education by 2015: A Chance for Every Child”, The World Bank, Washington.

    Connelly, R and Z Zheng (2003): “Determinants of School Enrolment and Completion of 10-18 Year Ilds in China”, Economics of Education Review, 22(4): 379-88.

    Cotton, J (1988): “On the Decomposition of Wage- Differentials”, The Review of Economics and S tatistics, 70: 236-43.

    Das, S and D Mukherjee (2007): “Role of Women in Schooling and Child Labour Decision: The Case of Urban Boys in India”, Social Indicators Research, 82(3): 463-86.

    – (2008): “Role of Parental Education in Schooling and Child Labour Decision: Urban India in the Last Decade”, Social Indicators Research, forthcoming.

    Deolalikar, A (1997): “The Determinants of Primary School Enrolment and Household School Expenditures in Kenya: Do They Vary by Income?” Mimeograph, Department of Economics, University of Washington.

    Desai, S and V Kulkarni (2005): “Changing Educational Inequalities in India in the Context of Affirmative Action”, Mimeograph, Deparmentt of Sociology, University of Maryland.

    Fairlie, Robert W (1999): “The Absence of the African-American Owned Business: An Analysis of the Dynamics of Self-Employment”, Journal of Labour Economics, 17(1): 80-108.

  • (2003): An Extension of the Blinder-Oaxaca Decomposition Technique to Logit and P robit Models, Discussion Paper No 873, Economic Growth C entre, Yale University.
  • (2005): “An extension of the Blinder-Oaxaca D ecomposition Technique to Logit and Probit
  • Models”, Journal of Economic and Social Measure

    ment, 30: 305-16.

    Godoy, R and M Contreras (2001): “A Comparative Study of Education and Tropical Deforestation among Lowland Bolivian Amerindians: Forest Values, Environmental Externality, and School Subsidies”, Economic Development and Cultural Change, 49(3): 555-74.

    GOI (Government of India) (2006): Social, Economic and Educational Backwardness of Muslims in I ndia: A Report, Report of the Prime Minister’s High Level Committee, New Delhi.

    Gomulka, Joanna and Nicholas Stern (1990): “The Employment of Married Women in the United Kingdom 1970-83”, Economica, 57: 171-99.

    Hadden, K and B London (1996): “Educating Girls in the Third World: The Demographic, Basic Needs, and Economic Benefits”, International Journal of Comparative Sociology, 37:1-2.

    Hunter, W W (1869): Indian Musalmans (New Delhi: Indological Book).

    Husain, Z (2005): “Analysing Demand for Primary Education: Muslims Slum Dwellers of Kolkata”, Economic & Political Weekly, Vol XL (2), pp 137-47.

    – (2008): Employment and Socio-Economic Status of Socio-Religious Communities in West Bengal: Some Evidence from NSS Data, unpublished report, Dept of Minority Affairs and M adrasa Education, Government of West Bengal.

    Jehangir, K N (1991): Muslim Women in Bengal: Socio-Economic and Political Status (Calcutta: Minerva Publishers).

    National Sample Survey Organisation (1998): Attending Educational Institutions in India – Its Level, Nature and Costs: NSS 52nd Round (1997-98), R eport No 439, Department of Statistics, Government of India, New Delhi.

    Neumark, D (1988): “Employers’ Discriminatory B ehaviour and the Estimation of Wage Discrimination”, Journal of Human Resources, 23: 279-95.

    PROBE (1999): Public Report on Basic Education in I ndia(New Delhi: Oxford University Press).

    Oaxaca, R L (1973): “Male-Female Wage Differentials in Urban Labour Markets”, International Economic Review, 14: 693-709.

    Oaxaca, R L and M Ransom (1994): “On Discrimination and the Decomposition of Wage Differentials”, Journal of Econometrics, 61: 5-21.

    Okumu, Ibrahim M, Nakajjo, Alex and Isoke, Doreen (2008): Socioeconomic Determinants of Primary School Dropout: The Logistic Model Analysis. Munich Personal Repec Archive. Accessed at: http:// people/Okumu,_ Ibrahim_M=2E.html on 20 November 2008.

    Reimers, C (1983): “Labour Market Discrimination Against Hispanic and Black Men”, The Review of Economics and Statistics, 65: 570-79.

    Ruhela, S P (1998): “Religion, Social Class and Educational Opportunity: Case Studies of Eight Muslim Girls” in S P Ruhela (ed.), Empowerment of the I ndian Muslim Women (New Delhi: MD Publications), 1-21.

    Salamatullah (1994): Education of Muslims in Secular India (Chandigarh: Centre for Research in Rural and Industrial Development).

    Shariff, Abusaleh (1999): Indian Human Development Report (New Delhi: NCAER).

    Sharma, K D (1978): Education of a National Minority: A Reader in Social Class of Indian Muslim (New Delhi: Kalamkar Prakashan).

    Sen, A (1985): Commodities and Capabilities (Amsterdam: North Holland).

    – (1999): Development as Freedom (New York: Knopf).

    Sengupta, Piyali and Jaba Guha (2006): “Enrolment, Dropout and Grade Completion of Girl Children in West Bengal”, Economic & Political Weekly, XXXVII (17): 1621-37.

    Sinning M, M Hahn and T K Bauer (2008): “The Blinder-Oaxaca Decomposition for Non-Linear Regression Models”, The Stata Journal, forthcoming, Accessed at sinning/files/sj.pdf on 20 November 2008.

    SIS/DPP (Strategic Information Section and Division of Policy and Planning) (2005): “Levels, Trends and Determinants of Primary School Participation and Gender Parity” (New York: UNICEF).

    Tansel, A (1998): “Differences in School Attainments of Boys and Girls in Turkey”, Discussion Paper 789, Economic Growth Center, Yale University.

    World Bank (2002): “Education and HIV/AIDS: A Window of Hope” (Washington DC: The World Bank).

    Yun, Myeong-Su (2004): “Decomposing Differences in the First Moment”, Economics Letters, 82: 275-80.

    april 11, 2009

    Appendix A: Results of Econometric Analysis Appendix B: Oaxaca Decomposition and Extension to Non-Linear

    I. Determinants of Primary Completion
    Model 1: Logistic regression: West Bengal Number of obs = 3366
    LR chi2(10) = 560.07
    Prob > chi2 = 0.0000
    Log likelihood = -1694.2421 Pseudo R2 = 0.1418
    pcr Coef Std Err z P>|z| [95% Conf Interval]
    lmfe 1.008517 .1164784 8.66 0.000 .7802239 1.236811
    hhsize -.1134646 .0203882 -5.57 0.000 -.1534247 -.0735045
    fml -.0573438 .0846445 -0.68 0.498 -.2232438 .1085563
    age .1283985 .0368012 3.49 0.000 .0562695 .2005274
    edu_dm .2227545 .0166581 13.37 0.000 .1901053 .2554038
    sex_dm .1965441 .1542012 1.27 0.202 -.1056847 .4987728
    h-uc .4146125 .1166139 3.56 0.000 .1860535 .6431715
    h-bc .2053415 .0988943 2.08 0.038 .0115121 .3991708
    others -.7324167 .3595497 -2.04 0.042 -1.437121 -.0277124
    urban -.0273242 .1009594 -0.27 0.787 -.225201 .1705526
    _cons -9.297479 .9831879 -9.46 0.000 -11.22449 -7.370466
    Model 2: Logistic regression: Rural Number of obs = 2347
    LR chi2(9) = 400.47
    Prob > chi2 = 0.0000
    Log likelihood = -1233.379 Pseudo R2 = 0.1397
    pcr Coef Std Err z P>|z| [95% Conf Interval]
    lmfe 1.263685 .1485378 8.51 0.000 . 9725559 1.554813
    hhsize -.128999 .0239839 -5.38 0.000 -.1760066 -.0819913
    fml -.106916 .0988853 -1.08 0.280 -.3007276 .0868957
    age .1209443 .042918 2.82 0.005 .0368266 .205062
    edu_dm .2284959 .0198314 11.52 0.000 .189627 .2673648
    sex_dm .3182691 .1993255 1.60 0.110 -.0724016 .7089398
    h-uc .313782 .1402765 2.24 0.025 .0388452 .5887189
    h-bc .1589375 .1125664 1.41 0.158 -.0616885 .3795635
    others -.9587445 .3923839 -2.44 0.015 -1.727803 -.1896863
    _cons -11.18987 1.220716 -9.17 0.000 -13.58243 -8.797313
    Model 3: Logistic regression: Urban Number of obs = 1012
    LR chi2(8) = 145.42
    Prob > chi2 = 0.0000
    Log likelihood = -451.98989 Pseudo R2 = 0.1386
    pcr Coef Std Err z P>|z| [95% Conf Interval]
    lmfe .5692156 1852953 3.07 0.002 .2060435 .9323876
    hhsize -.0945059 .040426 -2.34 0.019 -.1737393 -.0152724
    fml .0710964 .167399 0.42 0.671 -.2569997 .3991925
    age .1366256 .0724641 1.89 0.059 -.0054013 .2786526
    edu_dm .2050891 .0309318 6.63 0.000 .1444639 .2657143
    sex_dm -.0641106 .2469988 -0.26 0.795 -.5482193 .4199982
    h-uc .7042587 .2183382 3.23 0.001 .2763237 1.132194
    h-bc .3642174 .2137738 1.70 0.088 -.0547715 .7832063
    _cons -5.92646 1.69557 -3.50 0.000 -9.249716 -2.603204
    II. Fairlie Decomposition Model A: Rural West Bengal Number of obs N of obs G=0 = 17399 = 8354
    N of obs G=0 = 9045
    Pr(Y!=0|G=0) Pr(Y!=0|G=1) Difference = .57672971= .36042012= .21630959
    Total explained = .15973376
    pcr Coef Std Err z P>|z| [95% Conf Interval]
    age .0058287 .0010711 5.44 0.000 .0037294 .007928
    fml .0000444 .0001587 0.28 0.780 -.0002667 .0003555
    hhsize .0041054 .0005378 7.63 0.000 .0030514 .0051594
    lmfe .0349091 .0016079 21.71 0.000 .0317577 .0380606
    sex_dm .0002753 .0000744 3.70 0.000 .0001295 .000421
    edu_dm .1145495 .0023178 49.42 0.000 .1100067 .1190923
    Model B: Urban West Bengal Number of obs N of obs G=0 = = 9575 7583
    N of obs G=0 = 1992
    Pr(Y!=0|G=0) Pr(Y!=0|G=1) Difference = = = .71040485.45180723.25859762
    Total explained = .17277084
    pcr Coef Std Err z P>|z| [95% Conf Interval]
    age .0260339 .00176 14.79 0.000 .0225844 .0294833
    fml -.0017305 .0002695 -6.42 0.000 -.0022586 -.0012023
    hhsize .0074341 .0021433 3.47 0.001 .0032334 .0116349
    lmfe .0148033 .0016871 8.77 0.000 .0114966 .01811
    sex_dm .0063252 .0008171 7.74 0.000 .0047236 .0079267
    edu_dm .1197523 .0035051 34.17 0.000 .1128825 .1266221

    Considering the linear regression model:

    Yig = Xig β + εig ...[1]


    for I = 1, 2, 3 …. N, and Σ N = N. Blinder (1973) and Oaxaca (1973)

    g gg

    proposes the following decomposition:

    –– – –^ –^ ^

    ΔOLS =YA - YB = (XA - XB) βA + XB (βA -βB) ...[2]

    where Y = N–1 ΣNg and X = N–1 ΣNg Xig. The first term on the right

    g gi=1 Yigggi=1

    hand side of [2] displays the difference in outcome variable between the two groups that is due to differences in observable characteristics, whereas the second term shows the differential that is due to differences in coefficient estimates, and is referred to as an estimate of discrimination by Oaxaca. In a subsequent paper, Oaxaca and Ransom (1994) generalised [2] as f ollows:

    – – – – –^ – ^

    YA - YB = (XA - XB) β*+ XA (βA -β*)+ XB (β*-βB) ...[3]

    In equation [3], β* is defined as the weighted average of the coefficient vectors βA and βB:

    β* = ΩβA+ (I - Ω) βB ...[4]

    where Ω is a weighting matrix, and I is an identity matrix. The above decomposition, however, is not appropriate in the non-linear (NL) case. The reason is that the conditional expectations E(Yig | Xig) may differ from Xβ. In the non-linear case, therefore, [2] has to be mod

    g g

    ified. There has been several works in this context (Gomulka and Stern 1990; Yun 2004; Fairlie 1999, 2005). In Bauer and Sinning (2008) the decomposition is assumed to take the form:


    = [EβA) - EβA )] + [EβA ) - EβB )] ...[5]

    A (YiA|XiA(YiB|XiB(YiB|XiB(YiB|XiB

    where Eβg (Yig|Xig) refers to the conditional expectation of Yig, while Eβh (Yih|Xih) refers to the conditional expectation of Yih evaluated at the p arameter vector β, with g, h = (A, B) and g ≠ h. If B is taken to be the


    reference group, then the decomposition expression becomes:


    = [EβB) - EβB )] + [EβA ) - EβA )] ...[6]

    B (YiA|XiA(YiB|XiB(YiA|XiA(YiB|XiB

    As earlier, the first term of the decomposition expression indicates the part of the differential in the outcome variable between the two groups that is due to differences in the covariates Xig, while the second term represents the differential in Iig due to coefficients.

    The generalised form of the non-linear decomposition (corresponding to [4] is:

    ΔNL = [Eβ*(YiA | XiA) - Eβ*(YiB | XiB)] + [EβA (YiA | XiA) - Eβ* (YiA | XiA)] + [Eβ* (YiB | XiB) - EβB (YiB | XiB)] ...[7]

    The two decomposition expressions, [5] and [6], represent special cases of the generalised equation [7] in which Ω is a null matrix or an identity matrix, respectively. It is also possible to interpret Ω in other ways. For instance, Remiers (1983) proposes that Ω = 0.5 * I, while Cotton (1988) suggests Ω = sI, when s denotes the relative sample size of the majority group. On the other hand, Neumark (1988) and Oaxaca and Ransom (1994) derives the counterfactual coefficient vector β* based on the pooled sample.

    * This section largely draws from Sinning et al (2008).

    Economic & Political Weekly

    april 11, 2009 vol xliv no 15

    Dear Reader,

    To continue reading, become a subscriber.

    Explore our attractive subscription offers.

    Click here

    Back to Top