1While women traditionally marry early in India, age at marriage has increased over the last two decades, following a pattern that varies between regions and states. Over the same period, women’s education has also improved substantially, although large spatial inequalities still exist. Using data from the censuses of 1981, 1991 and 2001, Premchand Dommaraju analyses the effect at aggregate (district) level of female schooling with respect to other socioeconomic factors (percentage of farmers, percentage urban, etc.), cultural factors (proportion of Muslims), demographic factors (marriage squeeze) and regional factors (east, west, south, north) on the timing of marriage. After controlling for all variables, the changes in schooling levels have little effect on trends in age at marriage in the districts. A very limited effect is observed between 1981 and 1991, and a significant, but modest, effect between 1991 and 2001. The more general social changes over this period are a more likely explanation for the decline in early marriages. Nonetheless, these findings at aggregate level by no means challenge the individual-level influence of schooling on age at marriage.

2Schooling has been considered as a catalyst for changes in marriage age in developing countries. A report by the United Nations Commission on Population and Development (2002) emphasizes the importance of schooling in bringing about changes in nuptiality, but this thesis has been questioned by Mensch et al. (2005). There are two key theoretical reasons why the influence of schooling could be weaker than commonly thought. First, there may be common factors that influence both schooling and marriage age. In such a case, the association between schooling and marriage age would weaken when these factors are accounted for in the models. In most cross-sectional studies, it is nearly impossible to control for such common factors (e.g. modernization or westernization) since they are not easily measured or observed. Second, increases in schooling could have been a response to changes in marriage age (rather than the reverse). If this were so, a simple model estimating the relationship between schooling and marriage age would be biased. Though the presence of such biases is acknowledged, few studies have addressed these issues. The temporal and spatial variations in marriage and schooling levels in India provide an ideal opportunity to untangle the complex relationship between schooling and marriage. The present paper analyses this relationship using Indian census data from 1981, 1991 and 2001.

I – The Indian context

3Marriage is universal and early in much of South Asia. In 2000, singulate mean age at marriage (SMAM) for women was 19 years in Bangladesh (Bangladesh Demographic and Health Survey 1999-2000 [1]), 22.7 years in Pakistan (Pakistan Reproductive Health and Family Planning Survey, 2000-2001 [2]) and 23.6 years in Sri Lanka (computed from the 2001 census data). By contrast, marriage age varied widely in East and Southeast Asia. In 2000, SMAM for women was 28.6 in Japan, 27.1 in South Korea, 24.1 in Thailand, 23.3 in China and 22.7 in Indonesia (Jones and Gubhaju, 2009).

4In India, SMAM for women was 20.2 years, and only about 2% of women age 30-34 were unmarried at the time of the 2001 census. However, age at marriage has been increasing gradually since the 1930s. As seen in Table 1, SMAM increased by about one year per decade between 1981 and 2001, and much of the increase has been due to decline in early marriage. As with other demographic indicators, especially fertility and mortality, nuptiality also shows strong regional patterns. In 1981, SMAM for women varied from a high of 21.8 in Kerala to a low of 16.1 in Rajasthan. By 2001, although marriage age had increased in both the states, they still retained their extreme positions at the top and bottom of the scale. Note that in the southern states, with the exception of Andhra Pradesh, women married at a comparatively late age. There is no major difference, however, in the timing of marriage between Hindus and Muslims, the two main religious groups in India. The SMAM was 20.0 for Hindus and 20.3 for Muslims in 2001. Marital dissolution through divorce or separation remains rare. As per the 2001 census, less than 1% of women were divorced or separated and this proportion has been remarkably stable over the last three decades. Other aspects of marriage, including spousal choice, have also remained relatively unchanged. In 2005, for instance, caste endogamy was still prevalent, with nearly 95% of women married to men of the same caste as their natal family (IHDS, 2005). In addition, arranged marriages remain the norm, with only around 5% of women selecting their spouse on their own (ibid.).

Table 1

Female singulate mean age at marriage and literacy rates in India, major states, 1981-2001

State SMAM Literacy rate(c) (%) 1981(a) 1991(b) 2001(b) 1981 1991 2001 Andhra Pradesh 17.3 18.3 19.4 24.2 32.7 51.2 Assam – 21.1 21.7 – 43.0 56.0 Bihar 16.6 17.6 18.6 16.5 22.9 33.6 Gujarat 19.5 19.9 20.4 38.5 48.6 55.6 Haryana 17.8 18.9 19.7 26.9 40.5 56.3 Karnataka 19.2 20.1 20.9 33.2 44.3 57.5 Kerala 21.8 22.3 22.0 75.7 86.1 87.9 Madhya Pradesh 16.6 17.8 19.1 24.0 28.9 50.6 Maharashtra 18.8 19.7 20.6 41.0 52.3 67.5 Orissa 19.1 20.2 21.7 25.1 34.7 51.0 Punjab 21.1 21.0 21.6 39.7 50.4 63.6 Rajasthan 16.1 17.5 18.5 14.0 20.4 44.3 Tamil Nadu 20.2 20.9 21.4 40.4 51.3 64.6 Uttar Pradesh 16.7 18.1 19.6 17.2 25.3 43.0 West Bengal 19.2 19.7 20.0 36.1 46.6 60.2 All India 18.3 19.3 20.2 29.8 39.3 54.0 Note: All states refer to the pre-2000 state boundaries. The all India figures include smaller states that are not shown in the table. Sources: (a)1981: Registrar General, India, 1988; (b)1991 and 2001: calculated from the respective census data; (c) Government of India, 2002, Table 4.1.

Female singulate mean age at marriage and literacy rates in India, major states, 1981-2001

5While changes in nuptiality have been gradual, other family-related aspects have witnessed substantial changes. Fertility, for instance, declined from 4.9 children per woman in 1974-1980 to 2.7 in 2003-2005 (Bhat, 1996; IIPS and Macro International, 2007). This decline has occurred in the context of universal and relatively early marriage. Fertility decline has been achieved by widespread use of contraception, especially sterilization of women, rather than by increases in marriage age. In the southern state of Andhra Pradesh, for instance, fertility has declined to below-replacement level while age at marriage remains relatively low. It appears that changes in fertility rates have been far more rapid than changes in marriage patterns.

6There has been a large though spatially uneven increase in literacy levels in India in the last six decades since independence (1947). For the period under consideration, 1981-2001, literacy rates for females nearly doubled, with an especially pronounced increase between 1991 and 2001. While the reasons for this remain unclear, government policy in encouraging universal education and the economic changes beginning in the late 1980s may have provided opportunity and incentive for families to send their children to school. It should be noted that among literates, a significant proportion have not completed any schooling (the census defines literacy as merely the ability to read or write in any language as reported by the household head). In 2001, nearly one-third of literate women had not completed primary school (Census of India, 2001 [3]). But as with the case of marriage patterns, literacy rates vary from one state to another, ranging from a low of 34% in Bihar to a high of 88% in Kerala in 2001.

II – Schooling and marriage

7Four general scenarios may be anticipated.

8First scenario: Schooling lowers marriage rates and increases marriage age. A variety of pathways have been proposed to explain such an influence. It has been observed that marital and student roles are perceived to be incompatible, reflecting the commonly held expectation that individuals do certain activities at certain ages (Raymo, 1998; 2003). Thus with increasing levels of school enrolment, marriage rates for school-age girls may decline. Further, in contexts where female schooling translates into labour force participation, getting married when in school will entail significant opportunity costs. This higher opportunity cost in terms of barriers to achieving educational and career goals may encourage women to postpone marriage (Thornton, Axinn and Teachman, 1995). However, in settings where labour force participation is seen as incompatible with married life, women may choose to remain single or drop out of the labour force after marriage (Jones, 2005; Yu, 2005).

9Besides role incompatibility and labour force participation, schooling can delay marriage by bringing about ideational changes. Such changes include a shift from traditional values to individual-oriented values, secularization, rationality, and greater individual autonomy (Caldwell, 2005). Schooling exposes girls to new ideas, attitudes and aspirations that give them power to question traditional values (Kritz and Gurak, 1989). These ideational changes tend to delay marriage as they increase the say of women in marriage decisions, encourage alternative institutions to marriage, such as cohabitation, and make it more acceptable to remain single. Studies in India at the individual level have generally found schooling to delay marriage, though there has been no consistent causal explanation for these delays (Bloom and Reddy, 1986; Das and Dey, 1998; Dharmalingam, 1994).

10Second scenario: Marriage age can influence schooling levels. The causal arrow between schooling and marriage age can point in both directions. In societies where early marriage is the norm, the opportunity cost of sending girls to school would be high. In such contexts, as Amin and Huq (2008) have noted, schooling for girls is often seen as “something they do while they are waiting to get married”. If this is true, schooling levels in such situations would be influenced by the prevailing normative marriage age for women. Also, in South Asia education is rarely continued after marriage and schooling is effectively ended at the time of marriage. Changes in marriage age for reasons unrelated to education may thus influence schooling levels. Studies that have properly modelled the endogenous relationship between schooling and marriage age are rare, a notable exception being Field and Ambrus (2008) who show that marriage postponement in Bangladesh would substantially increase schooling levels.

11Third scenario: Changes in both schooling and marriage could be driven by same unobserved factors. Hatti and Ohlsson (1985) have argued that the higher marriage age for educated women in India could be due to exogenous factors that influence both schooling and marriage age. Though the authors do not elaborate, these extraneous factors could include the spread of modern ideas and attitudes about women’s roles. Audinarayana (1993) reports that in India, irrespective of education level, it is now common to get married a year or two after menarche. This change in marriage age to past menarche reflects a shift in thinking about childhood. In interviews conducted by Caldwell, Reddy and Caldwell (1983) in Karnataka and more recently by Santhya, Haberland and Singh (2006) in Rajasthan, parents often articulate their objection to marrying their daughters before menarche by noting the loss of childhood and by expressing concern that a degree of physical and emotional maturity is needed to take on the responsibilities of married life. While there is not much literature on the influence of ideational factors in explaining changes in attitudes towards early marriage, it is quite possible that ideational changes have led to a secular shift in attitudes and norms regarding marriage age.

12Fourth scenario: Patterns seen at the individual level between education and marriage may be different from those observed at the aggregate level. Besides an individual’s own education, community schooling levels and other community factors could mediate the relationship between schooling and marriage. In communities where some girls are in school or participate in non-family work, their marital behaviour could bring about normative changes and legitimize late marriage even for those not studying or working. As Amin et al. (1998) have noted, marriage age for girls is higher in communities with higher female labour force participation, irrespective of individual work or educational status. Thus the influence of schooling can go beyond the individual’s own educational attainment and marriage age could be influenced by broader community factors.
Much of the recent research on schooling and marriage linkages has focused on individual agents and causes. While such a focus is well suited to identifying the nuances of the first scenario, it may be less so for investigating the other scenarios outlined above. A pitfall of using individual data is that the individual is removed from the surrounding environment and considered in isolation as a free agent (Voss, 2007), and the results of such analyses “are often generalized to country, regional, and even global levels and are used, without justification, to interpret and anticipate time trends at the macro level” (Ní Bhrolcháin and Dyson, 2007, p. 1). This study uses aggregate panel data to investigate the relationship between schooling and marriage timing. Such an approach complements the current understanding offered by individual-level studies.

III – Data, variables and estimation methods

Variables

14The main dependent variable is the percentage of never-married women aged 15-19 in the district. Over the study period, major changes in marriage behaviour at younger ages were observed. Between 1981 and 2001, at the national level, among women aged 15-19, the proportion of never-married increased from about 56% to 75%, compared with an increase from 3% to 6% among women aged 25-29. The dependent variable is selected to focus analysis on these changes at younger ages.

15Schooling is measured using two variables: percentage of women in the district with primary school as the highest education level, and percentage of women in the district who have completed any level of schooling (primary, middle, secondary or higher [4]). The first variable is used to test whether a low level of schooling has the potential to bring about changes in marriage patterns. The second variable represents all levels of schooling and captures the general educational level in the district.
The models also control for percentage agricultural workers, percentage Muslim, percentage scheduled castes/tribes, [5] percentage urban in the district, geographical region and marriage squeeze (numerical imbalance between “marriageable” men and women). Since there is no direct measure for the economic status of the district in the censuses, percentage agricultural workers is used as a proxy. As Bhattacharya (2006, p. 270) has noted, “districts with higher percentages of agricultural workers, ceteris paribus, would have experienced less change in their economic structure”. Bhat and Halli (1999) have documented the influence of marriage squeeze on marriage patterns in India. To account for this influence, a simple measure of marriage squeeze is introduced in the models. The definition of all the variables is presented in Table 2.

Table 2

Variable definitions and sample means (standard deviations in brackets)

Variable Definitions 1981 1991 2001 Unmarried women, 15-19 Percentage of female population in district aged 15-19 who are unmarried 50.8 (21.5) 60.6(19.6) 73.7 (13.4) Primary school, women Percentage of women in district with primary school as the highest educational level 7.3 (4.9) 9.5 (5.6) 12.4 (4.2) Female schooling Percentage of women in district with any level of completed schooling (primary, middle, matriculation or higher) 14.2 (10.6) 21.1(13.3) 28.5(12.5) Teachers Number of teachers in district per 1,000 population aged 5-19 (instrumental variable) 14.5 (5.1) 18.1 (6.4) – Marriage squeeze Ratio of unmarried men aged 20-24 to unmarried women aged 15-19 in district (%) 100.40 (27.5) 95.2(22.5) 82.6(17.1) Agricultural workers Percentage of workers in district aged 15+ categorized as agricultural workers 68.9 (16.3) 67.7(17.4) 62.4(18.9) Urbanization Percentage of district’s population living in urban areas 20.5 (15.1) 22.4(16.1) 23.1(16.9) Scheduled castes Percentage of district’s population belonging to a scheduled caste 15.9 (7.0) 16.7 (7.1) 16.8(7.4) Scheduled tribes Percentage of district’s population belonging to a scheduled tribe 8.9 (15.2) 8.9(15.5) 10.1(16.8) Muslim Percentage of Muslim population in district 9.8 (9.1) 10.7(10.1) 10.4 (9.9) North Districts in the states of Haryana, Punjab, Madhya Pradesh, Rajasthan, and Uttar Pradesh. 46 45 45 South Districts in the states of Andhra Pradesh, Karnataka, Kerala, and Tamil Nadu 21 22 20 East Districts in the states of Bihar, Orissa, and West Bengal 18 20 22 West Districts in the states of Gujarat and Maharashtra 14 14 13 Number of districts 326 362 469 Period indicator 0 at start of period and 1 at end of period for fixed-effect models Source: Censuses of 1981, 1991 and 2001.

Variable definitions and sample means (standard deviations in brackets)

Estimation methods

16Three types of model are used: OLS regression, generalized least square fixed-effect, and two-stage least square model. The increase in the proportion of never-married women aged 15-19 between 1981 and 1991 (or between 1991 and 2001) could be due either to inter-district variation (i.e. variation in marriage patterns from one district to the next) or to intra-district variation (i.e. variation within each district over time). A simple OLS regression would model the inter-district variation. The fixed-effect approach models the intra-district variation over time, but the influence of variables that are not time-sensitive (e.g. geographic location of a district) cannot be estimated. Panel data could also be analysed using the random effect approach that models variations both across and within districts, and allows estimation of the influence of variables that do not change over time. A key assumption of the random effect model is that the regressors and the random individual effects are exogenous. A Hausman test (1978) indicated that the models from the Indian data violate this assumption, so estimates from fixed-effect models only are presented.

17The main advantage of using panel data is the ability to minimize (or eliminate) omitted variable bias (Baltagi, 2005, chapters 2 and 4). The equation of the panel model could be expressed as follows:

18

19where M_dt is the percentage of never-married women age 15-19 in district d at time t; x_dt is the vector of the explanatory variables; ? is the vector of the coefficients to be estimated; u_dt is the disturbance term; ?_d is the unobserved district-specific effect; ?_dt is the rest of the disturbance term. In the fixed-effect model ?_d is treated as a fixed term.

20Schooling and marriage may be endogenous. Endogeneity could be addressed using a two-stage least square regression model (Wooldridge, 2002, chapter 5). This method involves selecting an instrumental variable that is “highly correlated with that explanatory variable it is to replace, but which is uncorrelated with the error term” (Pearce, 1992, p. 209). The mark of a good instrumental variable according to Angrist and Krueger (2001, p. 73) is that it should be “correlated with the endogenous regressor for reasons the researcher can verify and explain, but uncorrelated with the outcome variable for reasons beyond its effect on the endogenous regressor”.

21The two-stage regression model could be expressed as follows. Consider the following regression model:

22

23where y₁ is the main dependent variable and y₂ is the endogenous variable, the z’s are other independent variables in the model and u is the error term.

24The first stage in the two-stage model is to regress the endogenous variable (y₂) using the instrument (I₁) as an independent variable. This can be expressed as follows:

25

In the second stage of the model, the fitted values of y₂ obtained from equation [2] are substituted for y₂ in equation [1].
In the case of schooling, it is reasonable to assume that availability of schools influences the decision to attend school. In the absence of a direct measure of availability of schools in the districts, number of teachers per 1,000 population, or better, per 1,000 school-age population (ages 5-19) can be used as a proxy. It makes intuitive sense to expect the number of teachers to be correlated with education; but at the same time, it is reasonable to assume that the number of teachers has limited or no direct influence on marriage age, i.e. the instrumental variable appears to be uncorrelated with the outcome of interest.

IV – Results

26Descriptive statistics for the variables used in the analyses are presented in Table 2. These measures indicate that women’s marriage age has been increasing. The proportion of never-married women at ages 15-19 increased by about 10 percentage points between 1981 and 1991, and by about 13 points between 1991 and 2001. Both the indicators of women’s schooling show improvement between 1981 and 2001. The proportion of women with primary schooling as their highest educational level increased from 7.3% in 1981 to 9.5% in 1991 and to 12.4% in 2001. The proportion of women with any level of completed schooling doubled between 1981 and 2001, reaching about 28% in 2001. The rest of the indicators indicate a slight increase in urbanization, a decline in the proportion of agricultural workers in the active population and an increase in scheduled tribal groups.
Table 3 presents the results from multivariate models that analyse the influence of primary schooling on marriage patterns of women aged 15-19. The first panel presents the results from OLS models for each year separately. The estimates indicate that primary schooling increases the percentage of never-married women aged 15-19, and its influence is strongest in 1981. A 1% increase in primary schooling at the district level increased the percentage of never-married women aged 15-19 by nearly 2.3% in 1981 and by about 2% in 2001. The cross-sectional models indicate that schooling is a significant factor in explaining the variations in marriage patterns.

Table 3

Primary schooling and marriage in India: regression estimates (dependent variable: percentage of never-married women aged 15-19)

Cross-section estimates Panel 1981-1991 Fixed effect IV (2SLS) 1981-1991 Panel 1991-2001 1981 1991 2001 I II III I II III Primary schooling, women 2.27 ** 1.57 ** 2.02 ** 0.98 ** 0.02 0.32 ** 1.10 1.33 ** 0.91 ** 0.88 ** (9.8) (7.5) (12.5) (7.7) (0.2) (3.1) (1.1) (11.6) (8.4) (8.6) Agricultural workers – 0.05 – 0.29 ** – 0.17 ** – 0.15 – 0.12 – 0.27 ** 0.00 – 0.35 ** 0.12 + 0.14 * (0.6) (3.2) (4.1) (1.0) (1.2) (2.7) (0.0) (6.3) (1.8) (2.2) Urbanization 0.24 ** 0.06 0.03 0.50 ** 0.05 0.13 – 0.11 – 0.16 – 0.09 0.05 (3.0) (0.8) (0.7) (3.9) (0.5) (1.5) – (0.6) (1.5) (1.0) (0.6) Scheduled castes 0.30 * 0.31 ** 0.13 + 1.25 ** 0.24 + 0.52 ** 0.34 + 0.46 – 0.03 0.24 (2.3) (2.7) (1.7) (6.3) (1.6) (3.6) (1.8) (1.1) – (0.1) (0.7) Scheduled tribes 0.18 ** 0.19 ** 0.21 ** 0.20 0.05 0.07 – 0.16 – 0.70 * – 0.47 + – 0.32 (3.2) (3.6) (5.8) (0.6) (0.2) (0.3) (0.4) (2.2) (1.7) (1.2) Muslim 0.05 0.05 0.09 + 1.95 ** – 0.35 – 0.22 0.22 0.26 0.18 0.25 + (0.6) (0.7) (1.7) (5.1) (1.2) (0.8) (0.4) (1.6) (1.2) (1.7) Marriage squeeze – 0.31 ** – 0.27 ** – 0.11 ** – 0.26 ** – 0.17 ** – 0.13 ** – 0.19 ** – 0.45 ** – 0.33 ** – 0.25 ** (11.0) (8.5) (3.2) (10.2) (9.2) (6.8) (6.7) (17.3) (12.6) (9.0) Time dummy(a) (1991/2001) 8.44 ** 10.91 ** 6.01 ** 7.21 ** 15.67 ** (18.4) (19.3) (2.7) (10.4) (10.9) Time* primary schooling – 0.38 ** – 0.60 ** (6.7) (6.3) South 8.81 ** 9.12 ** – 0.79 (4.1) (4.5) (0.6) East 7.81 ** 7.34 ** 0.43 (4.2) (4.0) (0.3) West 11.44 ** 8.15 ** 2.80 + (4.4) (3.1) (1.6) N 326 362 469 652 652 652 652 724 724 724 Significance level: ** p < 0.01, * p < 0.05, + p < 0.10. Notes: The numbers in brackets are t-ratios. (a) For the 1981-1991 panel, 1981 is the reference category; for the 1991-2001 panel, 1991 is the reference category. Sources: Censuses of 1981,1991 and 2001; author’s calculations.

Primary schooling and marriage in India: regression estimates (dependent variable: percentage of never-married women aged 15-19)

27The second panel presents estimates from the fixed-effect models for 1981-1991 period. There are three models in the panel: the first with all controls except time dummy, the second adds a control for time and the third has an interaction term for time and schooling. The time dummy captures the secular changes in marriage age between 1981 and 1991. The first fixed-effect model indicates a positive influence of schooling on marriage, but the magnitude of the estimates is substantially lower than the OLS estimates. In the next model with a control for time, the estimate for primary schooling is no longer significant. The coefficient of time indicates that the percentage of never-married women aged 15-19 was substantially higher (about 8.5%) in 1991 than in 1981, net of schooling and other controls. In the next model, an interaction term of time and education is introduced to test whether the influence of schooling is different in the two time periods. The estimate of the interaction term is negative and significant: the influence of primary schooling is 0.38 points lower in 1991 than in 1981, falling from 0.32 to –0.06. Finally, the next column in the table presents the estimates from the fixed-effect instrumental variable model. The estimate of schooling in this model is taken from the two-stage least square model in which, as described earlier, number of teachers per 1,000 inhabitants aged 5-19 in the district is used as an instrument. The estimate of primary schooling in this model is not significant, suggesting the absence of endogeneity.

28The last panel presents estimates from the fixed-effect models for the 1991-2001 period. As was the case for 1981-1991 period, the fixed-effect estimate of primary education for the 1991-2001 is weaker than the corresponding cross-sectional estimate. But primary education has a statistically significant influence in predicting percentage never-married women even when a control for time is introduced (model II). The estimate of the time dummy suggests that the percentage of never-married women aged 15-19 was substantially higher (by about 7%) in 2001 than in 1991, net of schooling and other controls. However, the influence of time does not wipe out completely the influence of primary education. In the next model the interaction term for time and primary schooling indicates that the influence of primary schooling is about 0.6 points lower in 2001 than in 1991 (0.88 in 1991 and 0.28 in 2001).
Table 4 presents the estimates of the relationship between female schooling (percentage of women with any level of completed schooling: primary, middle, matriculation or higher) and the percentage of never-married women aged 15-19. The table is arranged in similar fashion to the previous one. Cross-sectional estimates suggest that female schooling is positively correlated with the percentage of never-married women aged 15-19 in the district. The influence is evident in all three time periods, and is especially strong in 1981. The panel model for the 1981-1991 period with no control for time (model I) also supports this finding, but the estimate of the fixed-effect model is smaller than the cross-sectional estimates. According to the fixed-effect estimate, a 1% increase in female schooling increases the percentage of never-married women aged 15-19 by about 0.8%. In the next model with time dummy included, schooling is no longer significant. The estimate of time dummy indicates a 9% increase in the percentage of never-married women aged 15-19, net of other factors. As was the case with primary education, the interaction term between schooling and time is significant. The estimate of the interaction term indicates that the influence of schooling was significantly lower in 1991 than in 1981.

Table 4. Any level of schooling and marriage in India: regression estimates (dependent variable: percentage of never-married women aged 15-19)

Cross-section estimates Panel 1981-1991 Fixed effect IV (2SLS) 1981-1991 Panel 1991-2001 1981 1991 2001 I II III I II III Female schooling 1.16 ** 0.88 ** 0.95 ** 0.77 ** –0.08 0.34 ** – 1.08 0.88 ** 0.60 ** 0.59 ** (9.8) (9.9) (14.9) (12.0) (1.1) (3.7) (1.1) (14.9) (7.3) (7.8) Agricultural workers 0.01 0.09 0.07 0.08 –0.16 – 0.18 + – 0.53 – 0.09 0.10 0.09 (0.1) (1.1) (1.5) (0.6) (1.4) (1.8) (1.3) (1.6) (1.5) (1.4) Urbanization 0.15 + 0.06 0.04 0.16 0.08 0.12 0.40 – 0.28 ** – 0.17 + – 0.07 (1.9) (0.9) (1.0) (1.3) (0.8) (1.4) (1.2) (2.9) (1.8) (0.8) Scheduled castes 0.43 ** 0.41 ** 0.17 * 0.58 ** 0.26 + 0.46 ** 0.55 0.10 0.06 0.08 (3.4) (3.6) (2.3) (3.0) (1.8) (3.2) (1.6) (0.3) (0.2) (0.2) Scheduled tribes 0.18 ** 0.20 ** 0.18 ** 0.26 0.06 0.02 0.06 – 0.55 + – 0.45 – 0.37 (3.2) (4.1) (5.4) (0.7) (0.2) (0.1) (0.2) (1.9) (1.6) (1.4) Muslim 0.08 0.07 0.10 * 1.13 ** – 0.40 – 0.08 – 0.91 0.17 0.17 0.23 + (1.0) (1.1) (2.0) (3.2) (1.4) (0.3) (1.4) (1.1) (1.1) (1.6) Marriage squeeze – 0.33 ** 0.30 ** – 0.15 ** – 0.27 ** – 0.16 ** – 0.14 ** – 0.08 – 0.45 ** – 0.39 ** – 0.26 ** (12.1) (9.9) (4.8) (11.8) (8.5) (7.6) (1.0) (19.2) (15.4) (9.4) Time dummy(a) (1991/2001) 8.99 ** 9.86 ** 15.07 * 4.43 ** 13.51 ** (14.8) (17.1) (2.4) (4.7) (10.1) Time* schooling – 0.20 ** – 0.26 ** (7.1) (8.8) South 10.14 ** 8.45 ** 0.16 (4.9) (4.5) (0.1) East 7.40 ** 5.58 ** 2.05 + (4.0) (3.2) (1.7) West 13.67 ** 9.24 ** 3.70 * (5.5) (4.0) (2.3) N 326 362 469 652 652 652 652 724 724 724 Significance level: ** p < 0.01, * p < 0.05, + p < 0.10. Notes: The numbers in brackets are t-ratios.(a) For the 1981-1991 panel, 1981 is the reference category; for the 1991-2001 panel, 1991 is the reference category. Sources: Censuses of 1981,1991 and 2001; author’s calculations.

Table 4. Any level of schooling and marriage in India: regression estimates (dependent variable: percentage of never-married women aged 15-19)

29The estimates from the instrumental variable model presented in the next column show that female schooling has no statistically significant influence on marriage timing. The coefficient of female schooling is negative and larger that the estimate in model II, though not statistically significant. This pattern is similar to what was observed for primary schooling in the previous table. The lack of significance of schooling in the IV models suggest that after accounting for possible endogeneity schooling still does not have a significant influence on marriage timing.

30Estimates for the 1991-2001 period presented in the last panel show that female schooling had a significant influence on entry into marriage for women aged 15-19. The influence of schooling is significant even when time dummy is introduced (model II), although the magnitude of the estimate is substantially reduced. This contrasts with the 1981-1991 panel where schooling does not have a significant influence after introduction of the time dummy. The estimate of the interaction term for time and schooling in the 1991-2001 panel shows that the influence of education declined between 1991 and 2001.

Conclusion

31This paper assessed the influence of schooling on marriage timing of women aged 15-19 in India. While schooling has a substantial influence on cross-sectional variations in marriage timing, the relationship is far from causal. When district-specific and time effects are taken into consideration, schooling, irrespective of the level, is no longer an important factor in explaining changes in marriage timing between 1981 and 1991. However, it seems to have had a small but statistically significant influence on changes in marriage timing between 1991 and 2001. The interaction estimates for both time periods indicate that the influence of schooling has declined over time. Further, the analysis suggests that increases in schooling are, for the most part, independent of changes in marriage age (i.e. the relationship is not endogenous) between 1981 and 1991. The indicator for time has a significant and large influence in explaining changes in marriage timing between 1981 and 1991, and between 1991 and 2001. As was noted earlier, the time dummy captures secular increases in marriage timing. When such secular changes are taken into account, schooling is no longer a significant factor in explaining changes in marriage timing between 1981 and 1991. However schooling does seem to have played a role between 1991 and 2001.

32The present analysis tests the relationship between schooling and marriage at the aggregate level. It would be erroneous to conclude from the analysis that, for individuals, schooling does not have a direct influence on marriage timing, as inferences from aggregate data are not necessarily true at the individual level (Schwartz, 1994). Thus women with higher levels of schooling may marry at a later age, even though schooling has no influence at the aggregate level. This discord at the individual and aggregate levels is to be expected. Schooling at the aggregate level, in contrast with individual-level measures, captures the broader context and social environment of the district. Further, and importantly, the influence of outside factors may not be the same at both these levels. Changes in schooling and marriage brought about by wider societal transformations can only be captured by the aggregate data. Similarly, secular changes in marriage age cannot be detected using a strictly individual-level analysis.

33This paper focused primarily on untangling the relationship between schooling and marriage using a quantitative approach. To understand fully the complex relationship between schooling and marriage age would require investigating the changing norms and attitudes towards both marriage and schooling in India. Further research is also necessary to examine the causal relationship between schooling and marriage age at the individual level and to reconcile the findings from the individual-level analysis with the aggregate-level findings.
Marriage age has been increasing irrespective of changes in schooling in India. It is clear from the present analysis that broader societal changes are an important factor in explaining marriage change before age 20. These changes could have been brought about by the spread of modern attitudes and values that promote greater say for women in spousal choice and other matters related to marriage. The nature of the social changes, including the role of ideational change in delaying entry into marriage in India, remains to be investigated.