In very large countries, demographic behaviours are often geographically diverse, so analyses must be carried out at sub-national level to capture this heterogeneity. The finer the level of observation, the greater the degree of accuracy in identifying the geographical patterns of similar or divergent behaviours. India, a country of considerable economic, cultural and demographic diversity, is very interesting in this respect. Abhishek Singh, Kaushalendra Kumar, Praveen Kumar Pathak, Rajesh Kumar Chauhan and Adrita Banerjee study the spatial variations in fertility across the 640 districts of India to map its levels and determinants. The maps obtained provide demographers with a useful tool for analysing the spatial concentration of demographic behaviours.
1India, the second most populous country after China, is currently undergoing fertility transition. Fertility levels in the country have declined significantly over the last 50 years, from as high as 5.4 children per woman in 1961-1966 to 2.5 in 2010 (Registrar General of India, 2012; Rele, 1987). Although all the states of India are concerned, the pace of decline is variable. Whereas some states – mostly in the south – have achieved below-replacement fertility, others (mostly in the north) still have levels above 3 children per woman (Registrar General of India, 2012). Fertility in India varies not only by state, but also by urban-rural residence and by population subgroup (IIPS and Macro International, 2007).
2A number of studies have identified the key determinants of fertility. While mother’s schooling, economic development and contraceptive use are negatively associated with fertility, child mortality and son preference are positively associated (Aksan, 2014; Amin, Casterline and Spess, 2007; Basu, 2002; Bhat, 1996; Bhattacharya, 2006; Caldwell and MacDonald, 1982; Cleland, 2001; Dreze and Murthi, 2001; Guilmoto, 2005; Malhotra et al., 1995; Merrick, 2002).
3Although numerous studies have examined the determinants of fertility in India and its major states, we have found only nine studies that explain the variations in fertility at district level (Bhat, 1996; Bhattacharya, 2006; Das and Mohanty, 2012; Dreze and Murthi, 2001; Guilmoto, 2005; Malhotra et al., 1995; Mohanty et al. 2016; Murthi et al., 1995; Registrar General of India, 1997). Female literacy is an important determinant (Bhat, 1996; Das and Mohanty, 2012; Dreze and Murthi, 2001; Mohanty et al., 2016; Murthi et al., 1995; Registrar General of India, 1997). While a couple of studies have reported a significant negative relationship between economic development and fertility (Bhattacharya, 2006; Malhotra et al., 1995), others have found this relationship to be weak or non-existent (Guilmoto, 2005; Mohanty et al., 2016). Several studies have reported child mortality as an important determinant of fertility (Bhat 1996; Bhattacharya, 2006; Das and Mohanty, 2012; Dreze and Murthi, 2001; Malhotra et al., 1995; Mohanty et al., 2016; Registrar General of India, 1997). Another important determinant noted in the Indian literature is son preference (Bhattacharya, 2006; Dreze and Murthi, 2001). Dimensions of patriarchy and women’s autonomy are also associated with fertility (Bhattacharya, 2006; Malhotra et al., 1995; Murthi et al., 1995). Last, some studies have reported a positive association between fertility and share of Scheduled Castes/Tribes [1] population and of Muslim population (Bhat, 1996; Bhattacharya, 2006; Das and Mohanty, 2012; Murthi et al., 1995).
4Whereas there is consensus on the link between female education or child mortality and fertility, findings differ when it comes to the relationship of other factors with district-level variations in fertility. For example, while Bhat (1996) found a significant positive association between unmet need for contraception and fertility, Das and Mohanty (2012) found no relationship with contraceptive use in Uttar Pradesh and Bihar when controlling for age at marriage. While urbanization was negatively associated with fertility in the study conducted by Das and Mohanty (2012), no association was found by Dreze and Murthi (2001). We also find mixed results with regard to the relationship of proportion of Scheduled Castes/Tribes population and of Muslim population with fertility. For example, while they were found in some studies to be significantly related to district level fertility (Bhat, 1996; Das and Mohanty, 2012), Dreze and Murthi (2001) found no association. While the proportion of Muslim population was statistically associated with fertility in Bhattacharya (2006), the proportion of Scheduled Castes/Tribes population was not. Another finding that deserves mention is the relationship between poverty and fertility. Three of the nine studies included poverty as an independent variable in the statistical models. Interestingly, poverty was not associated with fertility in any of the three studies (Dreze and Murthi, 2001; Das and Mohanty, 2012; Mohanty et al. 2016). The infertility variable is very rarely studied. Various studies of sub-Saharan Africa have stressed the association between sterility and fertility in the past (Bongaarts et al., 1984; Caldwell and Caldwell, 1983; Frank, 1983), attributable mainly to the high prevalence of sexually transmitted diseases. Evidence suggests that childlessness varies significantly across the districts of India (Ram, 2006), and available literature shows that it is much higher in the south than in the north and central states (Agrawal et al., 2012; Ganguly and Unisa, Ram, 2006; 2010; Unisa, 1999). Trend analysis of Indian Census data also suggests similar findings (Ram 2006). It is worth mentioning here that fertility in the south is much lower than in the north and central states (Registrar General of India, 2012). Such distinct spatial patterns of fertility and childlessness in India prompt us to examine the association between childlessness and fertility in our study.
5Although a few attempts have been made to understand spatial patterns of fertility decline in India (Bhat and Zavier, 1999; Guilmoto, 2005, 2016; Guilmoto and Rajan, 2001), no studies to date have analysed the spatial patterns of fertility and its major determinants. In addition to socio-economic factors, this article explores the role of childlessness in variations in the totally fertility rate (TFR) in 640 districts of India spread across 29 states and 6 Union Territories. It also aims to test spatial correlations between variables, i.e. whether an inter-relation can be found between the spatial distribution of fertility and a range of other variables. Spatial effects are important, as they may invalidate certain standard methodological results (Anselin, 1988). Ignoring spatial effects gives rise to artificially low standard errors.
I – Data and Methods
Data
6We use data drawn primarily from the 2011 Indian Census conducted in all the 29 states and 6 union territories of India, divided administratively into 640 districts, and from the third round of the District Level Household Survey (DLHS 3) conducted in 2007-2008 to provide estimates of maternal and child health, family planning and other reproductive health indicators (IIPS, 2010). The data were collected from 720,320 households from different states and union territories. A total of 643,944 ever married women aged 15-49 and 166,260 unmarried women aged 15-24 years were interviewed in the survey. The details about DLHS 3 can be obtained elsewhere (IIPS, 2010).
Dependent variable
7The dependent variable used in the analysis is the total fertility rate (TFR). We used 2011 Census data to derive the 640 district-level estimates using the total number of births in the preceding year and the total number of women in the 5-year age groups [2] (ORGI, 2015). The values obtained in this way may potentially be underestimated (Registrar General of India, 1989). We compared the state-level unadjusted estimates of TFRs with TFRs obtained from the Sample Registration System (SRS). The two sets of estimates were close, except for the states of Uttar Pradesh, Bihar and Jammu and Kashmir. Their correlation coefficient was 0.77. To compensate for potential underestimation, we adjusted the district-level estimated TFRs by adopting the method suggested by Bhat et al., (1984) and outlined by Vosti and Lipton (1991). This involves computing an inflation factor that takes into account the age structure of the childbearing population and child mortality. [3] After adjustment, the estimates of TFR for Uttar Pradesh and Bihar came close to that of the SRS. The correlation coefficient between the adjusted TFRs and the state level TFRs obtained from the SRS was as high as 0.86; between our adjusted estimates and that given by Guilmoto and Rajan (2013) it was above 0.90. These consistency checks show that our district-level adjusted estimates of fertility are generally trustworthy and are broadly consistent with the estimates of the SRS, as far as interregional variations are concerned. A number of previous studies have used the information on births in the preceding year to estimate TFRs for Indian districts (Dreze and Murthi, 2001; Malhotra et al, 1995; Murthi et al., 1995). We note, however, that the estimates of TFR for Jammu and Kashmir were rather high due to a possible exaggeration of the child population in the 2011 Census (Guilmoto and Rajan, 2013). Hence, the results for Jammu and Kashmir must be interpreted with caution.
Independent variables
8Existing studies have listed a number of variables associated with fertility variations in India (Bhat, 1996; Das and Mohanty, 2012; Dreze and Murthi, 2001; Malhotra et al., 1995; Mohanty et al, 2016; Murthi et al., 1995; Registrar General of India, 1997). Accordingly, a number of socioeconomic variables (percentage of women literate, [4] female workforce participation, percentage of poorest households and percentage urban), cultural variables (percentage of Scheduled Castes/Tribes and Muslim) and fertility-related variables (percentage childless, percentage not using modern contraception, percentage of women aged 20-29 years married before age 18 years) were included in the analysis. We also included child mortality and gender preference ratio (an indicator of son preference) in the descriptive and multivariate analysis. The detailed definition of each independent variable is given in Table 1. All the independent variables were estimated using the 2011 Indian Census, excepting gender preference ratio, percentage of women not using modern contraception and percentage of women aged 20-29 married before age 18 years, which were estimated using DLHS 3. Note that DLHS 3 was conducted in 601 districts of India. Hence, for the districts for which DLHS 3 data are not available the state averages of the variables were considered in the analysis. The child mortality estimates for the 640 districts of India were borrowed from Guilmoto and Rajan (2013).
9Childlessness can be defined in a number of ways (Mascarenhas et al., 2012; Rutstein and Shah, 2004; WHO, 2006; WHO-ICMART, 2009). As we cannot distinguish between involuntary (biological) and voluntary infertility in this study, it is defined here simply as the proportion of women aged 45-49 years who have never had children.
10For the first time, the 2011 Indian Census provided micro level housing data on a sample basis (ORGI, 2011). These data include information on household ownership, household type, electricity availability, types of wall, types of floor, roof type, water availability, type of bathroom and latrine use, drainage and sewerage, type of fuel use, availability of different modes of communication and different modes of entertainment. We used these to estimate a wealth index at the national level by means of principal components analysis (PCA). The index was subsequently coded into five quintiles – poorest (bottom 20% of households), poorer, middle, richer and richest (top 20% of households). Past studies have shown that wealth index is a good proxy of household economic status (Rutstein, 1999; Rutstein et al., 2000). In this study, we use the percentage of households at district level in the poorest national wealth quintile as a proxy of wealth status.
Statistical analysis
11We first used ArcGIS to generate descriptive maps. The shape files from ArcGIS were later exported to the Geoda environment to perform the spatial analysis. [5] We used first-order contiguity matrix as weight for conducting the spatial analysis.
12We estimated Moran’s I, and univariate and bivariate Local Indicators of Spatial Association (LISA). Moran’s I is the Pearson coefficient measure of spatial autocorrelation (Moran, 1950) which measures the degree to which data points are similar or dissimilar to their spatial neighbours. It is given by
14where Zi is the standardized variable of interest; Wij is the weight matrix; C is the ratio of total spatial units and the sum of all spatial weights.
15Negative (positive) values indicate negative (positive) spatial autocorrelation. Positive autocorrelation indicates that points with similar attribute values are closely distributed in space, whereas negative spatial autocorrelation indicates that closely associated points are more dissimilar. Values of I range from −1 (indicating perfect dispersion) to +1 (perfect correlation). A zero value indicates a random spatial pattern.
16Univariate LISA measures the correlation of neighbourhood values around a specific spatial location. It determines the extent of spatial clustering present in the data. The LISA functionality in Geoda offers two important options: cluster maps and significance maps (Figure 1).
17Bivariate LISA measures the local correlation between a variable and the weighted average of another variable in the neighbourhood. It indicates whether the spatial distributions of the dependent variable (in our case TFR) and the independent variables are inter-related (Figure 2). The bivariate LISA involves the cross product of the standardized values of one variable at location i (e.g. TFR) with those of the average neighbouring values of another variable (e.g. women not using modern contraception).
18We first use ordinary least squares (OLS) to examine the relationship between TFR and the independent variables, with the exception of child mortality for which endogeneity is an important issue. While high child mortality rates tend to raise fertility, high fertility rates might raise child mortality for various reasons, so the regression coefficients in an OLS regression are generally biased (Greene 2012; Kennedy 2003). To examine the association between child mortality and fertility, and to correct for this bias, we use two-stage least squares regression, for which it is important to select appropriate instrument(s). In our case, this must be a variable that is correlated strongly with child mortality but not with fertility. Dreze and Murthi (2001) identified access to safe drinking water, and we in turn used access to treated tap water. A Hausman test [6] was used to examine whether TFR and child mortality were endogenous.
19We also used a spatial error model available in Geoda to examine the relationship between dependent and the independent variables (Anselin, 2005). This model evaluates clustering of an outcome variable that is not explained by the independent variables. Spatial clustering is explained with reference to the clustering of the error terms. The spatial error model can be mathematically expressed as
21where u is the model prediction error; e are the residues (spatially uncorrelated); m is the spatial auto regressive parameter; and W is the spatial weights matrix.
22We used a two-stage spatial regression model to examine the association between child mortality and fertility after adjusting for spatial clustering. Note that if spatial autocorrelation is ignored, more variables becoming significant as the standard errors appear smaller than they really are. In addition, the presence of spatial autocorrelation results in exaggerated precision.
II – Results
Spatial autocorrelation
23Table 1 presents the means of the dependent and independent variables used in the analysis. The TFR ranged between a minimum of 1.6 in Kolkata (West Bengal) and a maximum of 7.3 in Kupwara (Jammu and Kashmir), with an average value of 3.3.
24Table 2 presents the Moran’s I values for the dependent and independent variables included in the analysis. The Moran’s I for TFR was 0.68. It indicates high spatial autocorrelation in TFR over the districts of India. Spatial autocorrelation is positive when similar values occur near one another in space. Results indicate that the districts with similar TFRs are near one another. Moran’s I ranges between 0.40 (for gender preference ratio) and 0.83 (for percentage of women aged 20-29 married below age 18 years). It is also very high (0.80) for percentage of households in the poorest wealth quintile, female workforce participation (0.75), child mortality, non-use of modern contraception (0.74), and percentage of Muslims (0.72). The high Moran’s I statistics clearly indicate substantial spatial autocorrelation in the variables included in the analysis.
LISA maps
25Figure 1 presents univariate LISA maps for the dependent and the independent variables. The cluster map shows locations with a significant local Moran’s statistic classified by the type of spatial correlation: dark green for high-high associations, dark grey for low-low, pale grey for low-high and light green for high-low. For example, high-high TFR means that districts with above-average TFRs also share boundaries with neighbouring districts that have above-average TFRs (Figure 1A). On the other hand, high-low means that districts with above-average TFRs are surrounded by districts with below average values. High-high are also referred to as hot spots and low-low as cold spots (Iwasawa et al., 2009; Sridharan et al., 2011; Weinreb et al., 2008). Striking geographic clustering of high TFR is observed in several districts of northern, central, eastern and north-eastern India (Figure 1A), while districts with low TFRs are clustered primarily in southern India, but also in certain northern and eastern districts. High childlessness is clustered in southern districts and on the eastern coast (Figure 1B). Several districts in north-eastern India also show clustering of high childlessness, while low childlessness is clustered in northern and central districts. Clustering of high child mortality is found in central, eastern and northern districts (Figure 1C). The poorest households are clustered primarily in the eastern, central and north-eastern districts (Figure 1D). High gender preference ratios are clustered in northern, central and eastern districts (Figure 1F) and low ratios in the south.
Means of the dependent and independent variables
Means of the dependent and independent variables
Note: Standard deviations in parentheses.26Figure 2 presents bivariate LISA maps depicting the associations between selected independent variables and TFR at the local level. Low TFRs are statistically correlated with high childlessness in 71 districts, located mainly in southern India. A few eastern districts also exhibit such correlations (Figure 2A). In comparison, high TFRs are significantly correlated with low childlessness in 65 central, eastern and northern districts. There are a few outliers as well. High TFRs are correlated with high childlessness in a few central and eastern districts.
Moran’s I for the dependent and independent variables
Moran’s I for the dependent and independent variables
Interpretation: Values of I range from −1 (indicating perfect dispersion) to +1 (perfect correlation). A zero value indicates a random spatial pattern.27The map of spatial correlation between child mortality and fertility also reveals interesting spatial structures (Figure 2B), Overall, high TFRs are significantly correlated with high child mortality primarily in the districts of central, eastern and northern India and in a few north-eastern districts. In comparison, low TFRs are significantly correlated with low child mortality in southern districts. High child mortality is significantly correlated with low TFRs in a few districts of Andhra Pradesh, Karnataka, Jammu and Kashmir, Himachal Pradesh and Odisha. The same pattern is observed for gender preference ratio (Figure 2E). In Figure 2F, districts with a high percentage of women married below 18 years of age and high TFR are primarily concentrated in northern, central and eastern districts, while those where these two values are low are found mainly in the south. Interestingly, some districts (particularly in Andhra Pradesh and Telangana) have a high percentage of women married below age 18 and low TFRs.
28Overall, the bivariate LISA indicators suggest that spatial distribution of TFR, on the one hand, and of the selected variables, on the other, systematically contrasts the districts of central and eastern India (e.g. in Uttar Pradesh) with those of the extreme south (e.g. in Tamil Nadu, Kerala, etc.). They also suggest an association between childlessness, gender preference ratio, non-use of modern contraception, female literacy, and TFR. We further examined this association by estimating mean district-level TFRs according to these variables (Table 3). The findings confirm the associations obtained in bivariate LISA analysis. The mean TFR in districts having a gender preference ratio between 4.0 and 24.5 is 1.2 children higher than in districts with a ratio between 0.0 and 0.19. Indeed, fertility is lower in districts with higher levels of female literacy and high childlessness. In districts with female literacy below 50%, TFR is 1.7 children higher, on average, than in those where it is above 64%.
Univariate LISA maps showing the clustering of dependent and independent variables
Univariate LISA maps showing the clustering of dependent and independent variables
OLS regression models
29We ran an OLS regression model to examine the determinants of fertility in India. The results are presented in Table 4.
30A key purpose of the regression analysis is to examine the association between childlessness and fertility. As expected, childlessness is negatively associated with TFR, as is female literacy (Model 1). In contrast, the gender preference ratio is positively associated with TFR. This is also the case for percentage Muslim, non-use of modern contraception, percentage of Scheduled Castes/Tribes, and percentage of women married below 18 years of age. The variables not associated with TFR are female workforce participation, household poverty, and percentage urban.
Bivariate LISA maps showing spatial correlation between dependent and independent variables
Bivariate LISA maps showing spatial correlation between dependent and independent variables
31We estimated a two-stage least squares model (Model 2) to examine the association between child mortality and TFR. The Hausman test confirmed that endogeneity is a serious issue, but that treated tap water was a good instrument. [7] Interestingly, the effect of childlessness remains unchanged in the two-stage least squares. Gender preference ratio, no modern contraception, percentage of Scheduled Castes/Tribes, and percentage Muslim are also statistically associated with fertility. Female literacy and women married below 18 years of age become statistically insignificant.
Mean estimated district-level TFRs according to childlessness, gender preference ratio, and non-use of modern contraception
Mean estimated district-level TFRs according to childlessness, gender preference ratio, and non-use of modern contraception
32We also estimated spatial error (Model 3) and two-stage spatial regression models (Model 4) to account for the spatial clustering while examining the association between TFR and the independent variables. The results shown in Table 4 indicate statistically significant spatial autocorrelation (m). Lambda is statistically significant in both types of models. These results indicate the fitness of the spatial models presented here. [8] Childlessness is indeed negatively associated with TFR. Child mortality is not associated with TFR in the two-stage spatial model, while gender preference ratio, non-use of modern contraception, percentage of scheduled castes/tribes and percentage Muslim are all positively associated with TFR.
III – Discussion
33A number of studies have examined the determinants of fertility at the national and state levels in India, but analysis on these scales cannot detect the large variations that exist at district level. Districts are the smallest administrative units in India, and the level at which policies and programs are generally implemented, yet we came across only a few studies that have examined the determinants of fertility at this level. In addition, many of these studies failed to capture the spatial autocorrelation present in the data. Our primary objective in this article was to examine the spatial patterns of fertility and its determinants using data from the 2011 Indian Census. A key finding is the robust association between childlessness and fertility. At the district level, childlessness is significantly and negatively associated with TFR in all four statistical models. The spatial distributions of fertility and childlessness are indeed inter-related, and systematic contrasts are found between the districts of central and eastern India and those of the extreme south. Similar associations between sterility and fertility have been documented for sub-Saharan Africa (Bongaarts et al., 1984; Frank, 1983). The study by Frank (1983) revealed that sterility was associated with a considerable fertility shortfall in some large countries (Cameroon, Mozambique, Tanzania, Zaire) and in several small ones (Central African Republic, Congo, Gabon). A study by Caldwell and Caldwell (1983) also reported sterility as an important cause of abnormally low fertility in tropical Africa. The high levels of sterility in sub-Saharan African countries were primarily due to sexually transmitted diseases, but as access to antibiotics increased, sterility almost disappeared.
Results of regression models for TFR, Indian districts
Results of regression models for TFR, Indian districts
Significance levels: ** p ≤ 0.01; * p ≤ 0.05.Childlessness and fertility in India
34High childlessness is clustered in southern India and on the eastern coast, and low childlessness in northern and central districts. Analysis of childlessness among married women aged 40-44 and 45-49 years based on the 2001 Census also depicts high rates of childlessness along the southern and eastern coasts (Ram, 2006). The data from the 1981 Indian Census also suggest a similar pattern among married women aged 40-44 and 45-49 years. A study by Agrawal et al. (2012) suggests a high prevalence of childlessness in all the southern states, far above the national average: 13% in Andhra Pradesh, 12% in Kerala and 9% in Tamil Nadu. Unisa (1999) also finds high levels in Andhra Pradesh. Using data from the National Family Health Survey 2005-2006, Ganguly and Unisa (2010) found that prevalence is highest in the southern states. Unfortunately, the epidemiology of childlessness in India has rarely been explored. A few small-scale studies suggest pelvic inflammatory disease, herpes simplex virus type-2, tubal blockage, etc. as the main factors leading to infertility (Adamson et al., 2011; Mittal et al., 2015; Zargar et al, 1997). Recent data from DLHS 3 suggests high demand for infertility treatment in India; almost 80% of women aged 15-49 with primary infertility seek treatment (IIPS, 2010). This strong demand indicates that childlessness is an important issue in the country. Far more research is required on all aspects of this question, including childlessness as a determinant of demographic behaviour in India, and the possible role of voluntary childlessness in explaining Indian fertility.
Son preference
35Another interesting finding that emerges from our analysis is the statistically significant and positive association between gender preference ratio (an indicator of son preference) and fertility in India. Using data from the 1981 and 1991 censuses, Dreze and Murthi (2001) and Bhattacharya (2006) also showed a significant association. Our findings, based on the 2011 Indian Census, reveal that son preference is still an important determinant of fertility in India, and bivariate LISA shows that it is clearly a barrier to fertility reduction in central, eastern and northern districts. To our surprise, son preference was not correlated with TFR in the western districts with distorted sex ratios. Interestingly, the spatial distribution of fertility and son preference also systematically opposes the districts of central, eastern and northern India with those of the south.
Effect of spatial distribution
36The spatial auto-correlation (m) came out to be statistically significant in the spatial error and two-stage spatial regression model, indicating that the relationship between fertility and independent variables at the macro-level (districts) may be misleading if spatial clustering is ignored. The evidence presented earlier suggests that there is indeed an inter-relationship between the spatial distribution of fertility and of several other variables. The AK test also confirmed the utility of spatial models in examining the determinants of fertility at the district level in India. This study is the first to identify spatial outliers. Another unique contribution is the utilization of sample-based housing micro data from the 2011 Indian Census for estimating wealth quintiles. The percentage of households in the poorest quintile was used in the statistical models. By doing so we avoided using arbitrary cut-offs, which are often controversial, to estimate the percentage of population below the poverty line.
Conclusion
37Univariate and bivariate LISA provide good insight on the spatial clustering of the variables included in the analysis. Univariate LISA maps also help to identify the districts/geographical regions where more focus is needed, for example, those where high proportions of the poorest households or of women married below age 18 years are concentrated. Bivariate LISA between women married below 18 years and TFR produced interesting patterns (i.e. in some districts of Telangana, Andhra Pradesh, West Bengal, etc., where a high proportion of women married below age 18 is associated with low TFRs). Notably, Andhra Pradesh and Telangana are known for low age at marriage, low age at first birth and low age at sterilization (Matthews et al., 2009; Padmadas et al., 2004; Singh et al., 2012). Finally, bivariate LISA also highlighted a number of outliers which cannot readily be explained by the present study, such as districts where high childlessness and high fertility coexist. It is difficult to conclude whether such relationships are attributable to specific behaviours, random variations or poor quality of data on births in the preceding year and on children ever born. Future studies must focus on outliers to determine the reasons for the observed relationship.
38Finally, our findings reveal a need for a greater investment in the social sector, particularly in the central and eastern states of India where fertility is still very high. They also suggest that reduction in son preference is an important goal to pursue. Fertility seems to be shaped primarily by social factors. The indicators of income/wealth and modernization, such as wealth status of households and percentage urban, do not exert a statistically significant influence on fertility.
Acknowledgements
The authors are grateful to Dr C. Z. Guilmoto for his helpful suggestions and comments and to Mr Shraban Sarkar for his help in preparing the maps. An earlier version of the paper was presented in the Asian Population Association (APA) Conference held in Kuala Lumpur on 27-30 July 2015. The comments and suggestions of the participants were helpful. The authors also gratefully acknowledge the three reviewers and the editors whose comments immensely improved the quality of the paper.Notes
-
[*]
International Institute for Population Sciences, Mumbai, India. Correspondence: Abhishek Singh, Associate Professor, International Institute for Population Sciences, Mumbai – 400 088, India, email: abhi_iips@yahoo.co.in
-
[**]
Department of Geography, Delhi School of Economics, New Delhi, India.
-
[***]
Population Research Center, Lucknow, India.
-
[1]
Scheduled Castes/Tribes are various officially designated groups of historically disadvantaged people, sometimes referred to as Dalits or untouchables.
-
[2]
Table F-9 of the 2011 Indian Census.
-
[3]
For a given district, the inflation factor is given by 0N5/∑((5Wa × 5fad × 5L0/5*l0) × 5) where 0N5 is the number of children age 0-5 years in the district, 5Wa is the number of women per five-year age group in the childbearing ages in the district, 5fad is the district-level age-specific fertility rate, and 5L0/5*l0 is the district-level childhood survival probability.
-
[4]
We included female literacy in the analysis rather than the years of schooling due to data limitations.
-
[5]
Geoda is a tool for spatial data analysis: data manipulation, mapping, and spatial regression analysis (Anselin, 2005).
-
[6]
The Hausman test checks for the endogeneity of a variable by comparing instrumental variable (IV) estimates with ordinary least squares (OLS) estimates.
-
[7]
That is, child mortality was statistically correlated with treated tap water, and the coefficient of treated tap water was significant in the first stage regression.
-
[8]
The Anselin-Kelejian (AK) test also rejected the null hypothesis that there is no spatial autocorrelation in the residuals of the estimated model and justified the use of the spatial error model.
