Introduction

1Sex ratio imbalances in favour of men are very pronounced in several Asian countries such China, Korea or India. Apart from the impact of sex-specific migrations or adult mortality, these imbalances originate from higher sex ratio at birth and better survival rates among boys. Because sex differentials in migration rates and adult age mortality rates play little or no role in the composition of the child population, the analysis of the child sex ratio (CSR) is a very useful tool in assessing the extent and the nature of sex discrimination in countries where men predominate.

2While the extent of the sex ratio disequilibrium in Asian countries is well known, the variations observed within regions are less often described1. This is unfortunate, as national or regional sex ratio averages often tend to blur the intraregional disparities that are observed. Some areas are indeed often characterized by much higher sex ratio values than reflected by regional averages. For this reason, it is essential to disaggregate sex ratio data whenever possible and to use the lowest administrative units to detect geographical differentials across administrative subunits.

3In this paper, we will examine spatial variations and their regional contexts simultaneously in China and India using a similar methodology. While sex discrimination in Asia shares a common historical origin related to both traditional patriarchal institutions and recent fertility decline2, our comparative analysis will highlight some of main similarities and discrepancies between the two nations. It may be important to stress at this point that this paper will not attempt to explore the causes of sex ratio imbalances or ponder over the respective weight of sex-selective abortion, of excess mortality or of other factors; such a task would definitely require examining in detail the available evidence on the sex ratio at birth vs. the child sex ratio, which is beyond the scope of this paper3.

4We will first discuss the comparability of the data and then proceed to map recent figures for the lowest geographical units. This will help to compare the extent and the geography of interregional disparities across China and India. We will then examine the major correlates (or presumed determinants) of variations in the CSR. While several social or economic factors play a similar role in both countries, it appears that some social and economic factors have actually an opposite impact on the sex ratio in each nation. Further spatial analysis based on geo-statistical modelling will show the CSR to follow rather different spatial patterns in China and India at both macro-and micro-levels. The paper will end with a preliminary conclusion based on our findings.

What are the data to explore variations in child sex ratio?

5Sex discrimination among children has a direct impact on the demographic composition of the population. Two main phenomena are at play: lower sex ratio at birth and higher mortality rates among female children. Female foeticide (sex-selective abortion), infanticide and neglect of girls are the main factors explaining the skewed sex distribution of the children. The sex ratio at birth tends to fluctuate around 1.05 boys per one girl and this value is almost a biological invariant observed in all human populations with only minor variations. The impact of sex-selective abortions is therefore readily discernible on the level of the sex ratio at birth observed in a given area4. However, in countries lacking a proper civil registration system, data on sex ratios at birth are incomplete, unreliable or sometimes simply missing. While available for Chinese regions, detailed statistics on sex ratios at birth are not available in India below the state level for want of reliable civil registration statistics or detailed sub-regional surveys. The sex ratio at birth is available only for the few places in India where under-registration is minimal. In other areas, the best estimates come from retrospective surveys such as the National Family and Health Survey: these statistics are available only at the state level. The only source on the district scale comes from indirect estimation based on both the composition of the population aged less than one year in a given census and simultaneous infant mortality estimates at the same administrative level (Sudha and Rajan 1999). Similarly, the quality of data on mortality among children often does not allow of the estimation of sex differentials in infant or child mortality, except for large administrative units. As a result, the adequate statistics required for an analytical description of the various factors responsible for skewed sex ratios among children are usually not available. In our case, the diversity of available demographic data in China and India makes the task more difficult, as both countries have different statistical traditions and seldom provide statistics that are readily comparable.

6In this paper, we will use the child sex ratio derived from censuses as an index for the intensity of sex discrimination. When data are missing or incomplete, the census turns out to be a crucial source for studying the impact of sex discrimination on the child population. The child sex ratio is readily available from the statistics published by the census and is of reasonably good quality because the child population is little affected by sex differentials in migration, as long as children are below schooling age. Another advantage of using the census figures is that they allow us to disaggregate the age and sex data to a micro-scale for each country, rather than limit the analysis to provincial average values or selected areas.

7This index is not perfect as sex differentials in age misreporting (especially for around age 5) may also have an additional impact, especially in a country such as India, where age data are often of poor quality. Fortunately, errors in single age groups tend to cancel out when we use quinquennial categories. What may be more worrying is the possible effect of sex-specific under-reporting of children in China during the 1990 census5. Because of the one-child population policy, families may find it beneficial to conceal the presence of a child, especially when it is a girl. This hypothesis appears very difficult to establish with certainty and findings from available studies are often ambiguous. Some have stressed that sources less likely to be affected by sex-selective under-reporting, such as hospital records, indicate an unequivocal rise in the sex ratio at birth in the 1980s (Coale and Banister 1994). Similarly, the evidence for prenatal sex selection has strengthened (Chu Junhong 2003). But at the same time, in-depth studies demonstrate also the possibility of first female births being under-reported and replaced with second births (Merli and Raftery 2000). The CSR remains, however, the only variable readily available in both countries that captures the effects of both pre-birth and post-birth sex discrimination on the child population.

8Recent censuses have been conducted in both China (2000) and India (2001). However, only limited sub-regional figures were available at the time this paper was written. For China, age data from the last census were not available in a geo-referenced format for sub-regional units. We had therefore to use 1990 census data for the population aged 0 to 4 years. At that time, the CSR at a national level was already very high at 1.1, but it was to peak at 1.20 in the next census. Provisional figures are already available for the last 2001 Indian census, but the detailed age distribution has not yet been published. We therefore have to use the provisional data on the population below 7 years of age that were published in 2002. Figure 1 sums up the evolution of the CSR in both countries during the last 50 years.

9In order to map our figures, we used for India a GIS (geographical information system) that had been already set up to study fertility decline (see Guilmoto and Rajan 2002). For China, we used a geo-referenced dataset of sub-regional data from the 1990 census6.

Regional and sub-regional divisions in China and India

10The sex ratio has been computed as the ratio of males to females from the census figures. For comparative purposes, we had to forgo the use of the usual Indian sex ratio that is computed as the number of females per 1000 males. This index as computed in India is probably more convenient than the male to female ratio as it points directly to the deficit of women. The usual sex ratio reflects the “male surplus” and is more useful when examining the impact of male migration on the population composition of different areas.

11At the time of the 1990 census, the People's Republic of China was divided into 31 provinces, autonomous regions and municipalities. This includes territories such as Taiwan that are not under the control of the Chinese authorities. These provinces are further subdivided into more than 300 prefectures and into counties (xian) and urban wards (qu). We will use here the counties, as this is the lowest administrative level for which we can prepare maps. Our original census database included 2975 units. But many urban wards had to merged so that they can be mapped at a common scale with the counties. We also excluded counties in areas such as Taiwan, for which no data are provided by the Chinese census. The final dataset covers 2420 counties (rural and urban) for which we have both statistical and spatial information.

12India comprises 35 states and territories, the population of which varies in a ratio of one to 1000. They were divided into 595 districts in 2001, as against 466 in the previous census7. These districts are further subdivided into more than 5500 sub-district units, referred to as tahsils or taluks in most states. However, the size of these sub-district entities varies widely: while some contain several hundred thousand inhabitants, others are very small units with less than one thousand inhabitants, as in Arunachal Pradesh. Computing the CSR on small population units may be unadvisable. Moreover, this administrative grid has not yet been geo-referenced and the district level remains the lowest level for which GIS data are available in India8.

13For these reasons, we have retained the Chinese counties and the Indian districts for this study of sex ratio variations. For homogeneity's sake, these units will also be referred to in this paper as sub-regional units (or sub-regions) while provinces in China or states in India will be called regions. Interregional and intraregional disparities correspond to variations observed respectively between and within these regions. Figure 2 displays the maps of administrative regions and sub-regions in China and India. The geographical and demographic characteristics of sub-regional units used in this study are now briefly summarized in Table 1.

Table 1: Demographic characteristics of sub-regional administrative units, China (1990) and India (2001)

14As can be seen, the Chinese administrative division into 2420 counties is based on a finer geographical grid than that of the 593 Indian districts. While the area of Chinese counties is slightly inferior to that of their Indian counterpart, their average population is four times smaller than in India. Chinese counties have less than half a million population on average and some of them in Tibet even have a population below 10,000 inhabitants. Comparatively, India’s administrative grid is much more balanced: unusually large or under-populated districts are very few. While the average size in sq. km of our sub-regions is comparable across countries, the coefficient of variation is much higher for China (251%) than for India (89%). Provinces and counties in western China tend to be substantially larger than in the rest in the county. This gets reflected in the measurement of density, which is almost twice as high in India as in China.

15The data in Table 2 show the CSR level to be marginally greater in China than in India. This only reflects the Chinese situation in 1990, as the deterioration of the sex ratio among the children has continued unabated during the 1990s. Interestingly enough, the CSR variability across sub-regions in both countries is almost identical.

Table 2: Child sex ratio in sub-regional administrative units, China (1990) and India (2001)

16When plotted in Figure 2, the distribution of CSR values appears roughly comparable in both countries. The slightly more regular distribution for China is due to its larger number of administrative units. The distribution of the sex ratio in both nations is slightly skewed rightwards because of the presence of sub-regions with extremely high sex ratios, while areas with unusually low sex ratios are almost non-existent. Obviously, sexual discrimination works in both countries mostly one way, i.e. in favour of boys, even when we can find several regions where girls outnumber boys, as in Xinjiang or in tribal India.

17A closer look at this figure would indicate that CSRs in China seem to be more evenly distributed around its modal value of 1.05. In India, there is a large number of districts for which the ratio is clearly below the usual 1.05 benchmark. But at the same time, there is a small group of districts recording very high values, including CSRs above 1.2. In India, it seems therefore that the "standard" distribution of CSRs has been recently disturbed in many districts, while the rest of the country appears less affected by the trend towards an increasing girl deficit. Conversely, changes in CSRs were much more widespread in the 1990s in China and the number of unaffected counties is significantly lower.

Mapping sex ratio variations

18We reclassified child sex ratio values into six categories, using regular value classes from 1.00 to 1.20. The breakpoint for the CSR is here assumed to be 1.05, even if the standard value is usually lower in most countries of the world: according to United Nations figures, the sex ratio below 5 is of 1.038 in the world, after excluding China and India. While 1.05 reflects the average value of the sex ratio among children born, it tends to decrease regularly as age increases and is therefore lower for the composite 0-4 age group.

19From the original maps based on sub-regional administrative divisions, we prepared a new set of contoured maps that are easier to read. These maps are no longer dependent on the administrative grid. To draw these maps, we first converted each sub-region's polygon into a single point, located at the centre of the original polygon, and we then interpolated these point values using the kriging technique to estimate continuous values for all locations on the map (except for extreme locations). The first results consist in raster maps of interpolated CSR values. These maps were finally contoured using the same classes as for administrative regions9.

20From these maps, the shape of larger homogeneous regions that cut across administrative boundaries appears now more clearly. The rest of this section is devoted to a brief overview of the maps shown in Figure 3.

China

21China’s micro-regional map of juvenile sex ratios proves much harder to read than the province-level maps that have been so far commented upon by demographers10. The distribution of the sex ratio in large regions tends indeed to follow a global profile, moving from normal values in the west to higher values as one proceeds eastwards, with the peak of masculinity found in provinces located in the south-east, such as Hainan and Guangdong.

22The micro-regional picture indeed proves to be more complex: sex ratio values are never uniformly distributed within the provincial boundaries and sub-regional irregularities are now noticeable. Let us first consider the western provinces, which exhibit normal values that are mostly below 1.05. The detailed mapping now shows that there appears to be large, contiguous pockets of distorted CSRs that are found in the north-western tip of Xinjiang (Ürümqi), in Gansu and in the western part of Inner Mongolia. Regarding the eastern provinces, our map shows that the CSR is also far from homogeneous. It reaches in some areas extremely high values (such as 1.25), which are much beyond the regional average.

23In fact, the map serves chiefly to highlight the sharp internal disparities of sex ratio values within provinces. Sichuan province11 offers a one such example, where we see normal values to the west contrasting with extremely high values in the eastern counties. This gap corresponds broadly to the distribution of the Han population, which is concentrated in the eastern half of the region. Yunan to the south is another province where one may distinguish small tracts characterized by high sex ratios that are located east of its capital city Kunming. In this case, the association between the distribution of minority populations and low sex ratios is, however, less clear-cut.

24The situation is yet more complex in other provinces like Hubei or Zhejiang, where the Han population predominates: sex ratio levels below 1.05 are found in several pockets of both provinces, for example, in central Hubei and in areas adjacent to Shanghai province, respectively. At the same time, the rest of these provinces exhibit a pronounced deficit of girls. Sex ratios can even exceed 1.2 boys per girl around Wuhan or in coastal Zhejiang. In-depth studies on small areas such as Hainan, where our map shows variations in the sex ratio to be sizeable, confirm that high levels of internal heterogeneity with respect to son preference may be frequent (Lavely et al. 2000).

25With such a level of internal heterogeneity, no global spatial pattern seems to emerge in eastern China. Even the urban-rural distinction is of little help: while cities often exhibit a lower CSR, as is the case with Beijing, Shanghai or Tianjin, this is no more true of the prosperous south-eastern littoral (Guangdong, Fujian), where the urban CSR is usually higher than that of the rural hinterland.

India

26The Indian scene is comparatively much easier to describe that than of China. The main reason for this difference lies in the existence of a few geographical patterns accounting for most of the variations in the CSR observed across districts in India. The high CSR regions are indeed concentrated in North-West India. Both Punjab and Haryana states figure predominantly in this picture: the sex ratio in this corner of the country reached the most dramatic values in 2002, significantly above 1.20 boys per girl. Some sub-district units have even recorded child sex ratios as high as 1.4, such as Bhulath tahsil in Punjab

27The sex ratio in surrounding areas tends to decline regularly as one moves farther away from these core areas. The decrease can be extremely fast towards the Himalayas, as exemplified by the normal values recorded in neighbouring Himachal Pradesh, or, on the contrary, very gradual towards western Uttar Pradesh or Gwalior in Madhya Pradesh.

28Secondary core areas are found in Gujarat (Ahmedabad) and in central Maharashtra, and their impact on adjacent areas is also clearly visible. Two smaller tracts where sex discrimination is prevalent are also discernible on the map. The first one is the Salem area in Tamil Nadu, which is discussed in Stéphanie Vella’s paper. The second one is located in coastal Orissa and happens to be the most recent addition to the map of the deteriorating child sex ratio in India. These districts remain more or less isolated pockets that seem to have little influence on the neighbouring areas.

29The rest of India is by and large exempt from any abnormal sex ratio level. These areas altogether account for almost half of the country, including important states such as Andhra Pradesh, Kerala or West Bengal. The sex ratio is rarely above 1.05 boys per girl. However, the situation is probably changing very rapidly, as the deterioration of the sex ratio might expand to new areas in the near future.

30Unlike in China, therefore, the areas characterized by intense sex discrimination in India form a compact group that is clearly distinguished from the rest of country. The Indian map shows changes across adjacent districts to be moderate and gradual. A basic statistical analysis of variance (anova) leads to a similar conclusion. Provinces and states play different roles in China and India in explaining sub-regional differences: Chinese provinces explain less than 24% of the total variance of the CSR across counties, whereas Indian states explain almost 70% of the CSR variance across districts.

A comparative statistical analysis of sub-regional variations

31The statistical analysis of Indian data has been performed by different authors who have previously attempted to model sub-regional variations in the CSR using other social, economic or demographic characteristics12. Such analyses aimed mostly at explaining these disparities and offering an explanatory framework for sexual discrimination, but will not be repeated here. As a matter of fact, variables used for one country may not be found for the other country or may prove simply inapplicable. For instance, the question of joint family is irrelevant for the Chinese context. Similarly, no equivalent for county-wise estimates of the gross value of agricultural output found for China can be found for Indian districts.

32What will be attempted here is the use of a common set of variables to be applied to both countries simultaneously. Our aim is to draw a more general comparison of factors associated with distorted child sex ratios. To do this, we will mainly resort to four broad domains, i.e. to social, cultural, economic and demographic factors. Analysts usually relate these characteristics to sex ratio deterioration in Asia. Thus, factors related to economic or social development (such as female literacy, living standards, urbanization, etc.) are often supposed to impact favourably on the child sex ratio. For its part, group membership (to ethnic, linguistic or religious entities) is assumed to have a great importance in distinguishing behaviours within society, with specific groups displaying much higher CSRs than other groups. The prevalence of patriarchal norms, which is deeply entrenched in many groups, fuels son preference and can be seen on the whole as unfavourable to girls. The demographic variables have an ambiguous role, but fertility decline is often supposed to adversely affect the sex ratio. Nevertheless, the relationship between the sex ratio and these various factors (and other social, political, economic or historical characteristics omitted here) is still hotly debated among scholars, as some elements of the debate on the sex ratio in India may illustrate13.

33CSR data for counties and districts have been correlated with a set of available factors for both India and China14. The results given here are the correlation coefficients (r). These variables will only be briefly described in the tables that follow (see the unit column). Table 3 presents the results of a first-order correlation analysis for a first set of factors.

34Density (measured in logarithm) is an unexpectedly strong positive correlate to the child sex ratio in both countries, while urbanization provides opposite results: the girl deficit is more acute in the Chinese countryside and in Indian towns. Similarly, the share of agriculture in the local economy plays an opposite role in the two nations: the low CSR is associated with the Chinese peasant economy15 while in India, it is in the non-agricultural sector that the girl deficit tends to be observed. When combined, these indicators point to different hot spots of sex discrimination: the dense, agricultural countryside in China, on one hand, and non-agricultural urban and peri-urban localities in India, on the other hand.

Table 3: Correlation of child sex ratio with census variables, China (1990) and India (2001)

35Literacy indicators display a complete agreement across countries and indicate that education seems to have a significant positive impact on distorted CSRs. This is a most worrying dimension of sex discrimination, as progress in literacy in both countries, especially in India which lags behind China in this respect, could be associated with a worsening sex ratio scenario. The demographic factors have somewhat similar profiles in both countries: two indicators of fertility levels (birth rates and fertility measurements) confirm that the highest CSR is found in areas where fertility decline is more advanced. Low birth rates and higher education levels are two central dimensions of social development that are prima facie correlated with girl deficits in the two Asian giants.

36The following Table 3 presents the results of the correlation analysis performed with data on the composition of the population as well as three Indian macro-regions. The first result refers to what we term the “majority group”, i.e. the Han in China and the Hindu population of India. In China, there is a major distinction between the Han and the minorities: Han-populated counties are characterized by significantly higher CSRs. In India, Hindus as a religious group do not display any significant feature with respect to the CSR.

37The available sub-regional statistics allow us to decompose in greater detail the ethnic or religious factors in both countries. As the correlation coefficients in Table 4 show, the CSR in most minority groups in China is lower than among the Han. This is especially true of Muslim minorities such as the Hui, the Uygur or the Kazakh. But this is also the case for other minority populations such as the Tibetans, who live in underdeveloped areas in the south-west, or the more prosperous Koreans established in the north. Having said that, the association between a high level of discrimination towards girls and the Han population is not unqualified. Some areas inhabited by the Han display relatively normal sex ratio values, as was observed from the map of China. Moreover, some minorities are also affected by sex imbalances. Worth noting is the apparently higher CSR associated with the presence of some ethnic groups, as can be seen for the Zhuang, the Dong and the Yao: these are populations chiefly found in the Guangxi province and who have widely adopted the Chinese language and many other of the behavioural traits from the majority community.

Table 4: Correlation of child sex ratio with census variables, China (1990) and India (2001)

38The analysis for India examines first the difference among major religious groups. The Christian and the Muslim minorities that are spread all over India display rather low sex ratios compared to the Hindu majority. On the contrary, the Sikhs, who predominantly inhabit North-West India and Punjab as well Haryana, Chandigarh and Delhi, are characterized by extreme sex ratio levels. This association between specific regions and religious groups is, however, neither exclusive nor absolute: we can note for instance that the other pockets of high sex ratio in Gujarat or in Maharashtra that were identified on the map of India are almost devoid of any concentration of Sikh populations. While the presence of Sikhs is correlated to a higher CSR, the reverse is not necessarily true.

39An additional category in India is the proportion of Dalits (ex-Untouchables) and Tribals in the district population. Both populations have different relationships to high CSRs. It is well known that girls and women enjoy a better status in the tribal areas of central India than elsewhere and this is clearly reflected in the findings16. The positive association between the CSR and Dalits is less documented; the multiple linear regression below will clarify this.

40We have also tested the regional impact, using some of the common regional groupings employed in Indian social geography. The South and the North-East have much lower sex ratios as shown on the previous map. However, the Bimaru states (Bihar, Madhya Pradesh, Rajasthan and Uttar Pradesh), which are often associated with some of the ills of Indian demography such as high child mortality or high fertility, fail to display any noticeable feature. As the map showed us, the Bimaru states fall across both sides of the sex ratio dividing line: States like Bihar or Jharkhand (previously part of Bihar) record normal or low CSRs, while the north of Rajasthan or Madhya Pradesh are in a situation akin to that of Punjab or Haryana with a pronounced female deficit.

A statistical synthesis

41The analysis has so far examined the pair-wise association between sex ratio and a host of social characteristics of administrative sub-regions. Many of these variables are nevertheless redundant and a multiple regression model will clarify the exact contribution of the different factors in accounting for the sizeable CSR variations that are observed in both China and India. To give an example, let us consider the low sex ratio observed in Tibet (Xizang). Are these rather low values related to the Tibetan culture, to the underdevelopment of Tibet characterized by high rates of illiteracy among men and women, to its extremely low population density or to its high fertility? As we see, possible explanations for Tibet’s peculiar situation may refer equally to demographic, economic or cultural factors. To sort out such issues, a multiple regression is in order.

Table 5: Multiple linear regression of child sex ratio, China (1990) and India (2001)

42Table 5 summarizes the results of a synthetic analysis of sex ratio disparities in the two countries. The technique used here is an ordinary multiple linear regression model that has been applied once again to the sub-regional data. Coefficients displayed are the standardized beta coefficients so that the strength of the association can be compared across variables. As the number of variables is rather large, the table shows variables that turned out to be statistically significant. Most other variables have been omitted from the table.

43The findings indicate that the CSR peaks in China are more prevalent in high-density counties combining both urban characteristics and a large number of agricultural households. This somewhat odd combination probably designates the dynamic semi-urban counties of eastern China, where intensive agriculture blends with industrial activities. While higher than average literacy rates are still positively correlated with a high sex ratio, it now emerges that the fertility effect per se is nil. Several other correlates that were brought to light in the previous correlation analysis, such as the share of agricultural sector in the workforce analysis, have also disappeared from the general model.

44The role played by the ethnic composition of the population has not been reversed, except for the fact that most ethnic variables exhibit much smaller correlation coefficients. Some of them have even lost any type of significance, as is seen for the Tibetans who figured prominently in the previous analysis. Apparently, the contribution of other factors related to the characteristics of the Tibet plateau, such as low population density or low literacy, accounts for this change.

45The synthetic model for India tends also to streamline the findings from the previous analysis. Several associations with variables such agricultural workforce or Dalit population observed above have now disappeared. The result remains very ambiguous, as factors associated with a high sex ratio include both progressive factors, such as literacy, urbanization and late age at marriage among women, and a rather patriarchal trait, such as high fertility. This analysis probably suffers from introducing a potentially endogenous variable such as fertility in our regression without proper econometric specification17.

46Among the other variables, the proportions of Muslims and Tribals in the population continue to reduce the CSR, while the Sikh population is still a major enhancing factor. An equation reduced to these three latter variables would in fact account for no less than 44% of the total variance of the CSR.

47Reading simultaneously the Chinese and Indian regression models leads to a few observations. The only factors that remain common in both countries are urbanization and literacy, which are positively correlated to girl deficit. This shows the evolution of the CSR to be linked to processes of social and economic development in China and India, which is a somewhat counter-intuitive observation. However, the explanatory power of these factors remains modest and counter-examples abound: South India and the Shanghai region are characterized by higher than average levels of urbanization and literacy without displaying a dramatic increase in the CSR. We may also observe the role played by ethnical or religious variables in both settings: differences closely correspond to the composition of the population and it is easy to identify groups that are systematically associated with lower or higher CSRs in China and India. While the risk of ecological fallacy (attributing regularities observed for administrative units to actual individual behaviour) remains a possibility, the statistical characterization of hot and cold spots of the CSR follows very closely the geographical boundaries of social and cultural groups, as is especially visible for minorities. This tends to lessen the strength of the development argument referred to above: cultural and social conditioning appear to mediate the impact of development variables such as education and urbanization on gender discrimination.

48Another important observation derived from a comparative examination of the regression results is that the Chinese model does a rather poor job in explaining the observed regional differentials: the overall correlation is less than 20% of the sample variance of the CSR. The Indian model performs significantly better, even though almost half of the observed variations still remain unaccounted for. This is partly due to the fact that some variables have not been introduced in our analysis for lack of statistical or social equivalents in the other country. For instance, detailed modelling of sex distribution in India includes variables such as female exogamy or medical infrastructures. Similarly, the impact of economic development on the CSR in India was not tested using the available variable for lack of Indian equivalents for the 2001 figures. However, studies published so far indicate that the modelling of CSR differences within India or China is often incomplete because of unobserved characteristics. Studies related to China are restricted to specific provinces or to province-level data and therefore are not directly comparable to our findings. In India, where district-level studies using 1981 and 1991 census data were more common, the models inevitably introduce several regional dummies or a spatial dependence factor to tackle spatial autocorrelation (Kishor 1993, Murthi et al. 1995). The proportion of the total variations in the CSR explained without these geographical dimensions is not reported in these studies, but close inspection of the results suggest that they would be hardly better than findings from this analysis.

49All this suggests that the strong geographical patterning observed in India remains a central dimension of variations in gender discrimination. In China’s case, the spatial distribution is more difficult to interpret for lack of large-scale, pan-Chinese patterns. In the next section, we will endeavour to examine more closely the geographical patterns using geo-statistical tools.

The spatial autocorrelation of child sex ratio

50This section presents the result of a geo-statistical analysis of the CSR distribution in China and India. A recurrent dimension in our analysis so far has been the strong geographical patterning observed in the two countries. A more systematic analysis of spatial autocorrelation makes it possible to test whether the correlation between sub-regional CSR estimates is a function of the distances separating these sub-regional units. In the presence of spatial autocorrelation, the nearest counties (or districts) should be characterized by similar CSR levels.

51For a more precise comparison, we computed a spatial autocorrelation index of the CSR in both China and India. This index (Moran’s I) measures the correlation between pairs of observations at a given distance (Haining 2003). Moran’s index is equal to the ratio of the covariance between different observations to the general sample variance:

where csr i is the (normalized) child sex ratio for unit i, n is the number of pairs of locations i and j with distance(i, j) = h, and m is the number of observations.

52As this formula shows, Moran’s index is analogous to the ordinary correlation coefficient (r) for two different variables, except that it is now computed for the same variable csr (child sex ratio) over pairs of observations classified by distance (d). In the absence of spatial autocorrelation, this indicator is close to 0. However, if regional units are close by and the CSR levels are similar, I (d) tends to be higher, perhaps even close to 1 when spatial dependence is maximum.

53The analysis is here performed with the Chinese counties and with the Indian districts. Results are plotted in Figure 5. For the sake of comparison, Moran’s index has here been computed for the same 50-km intervals, starting from distances less than 50 km. The results are eloquent. Both countries exhibit strong spatial autocorrelation at short distances. In China, almost half of the variation between neighbouring counties is explained by their proximity. For India, the correlation is so strong that it even exceeds 1 for districts that are distant by less than 100 km. This means that girl deficit is significantly less pronounced in China than in India, as was already observed in the previous maps.

54The curve for India points to a very broad pattern of spatial dependence that was already visible on the map. Districts that are 200 km apart still have a lot in common in terms of gender disparities, for the index is still as high as 0.75. As a matter of fact, the regional patterning of sex distribution in India is still detectable at distances of 400 km. Comparison with other census statistics (not shown here) indicates that the level of spatial autocorrelation for the CSR is possibly the highest recorded in 2001, stressing the rather unique geographical patterning of sex discrimination across Indian regions.

55The examination of Moran’s index allows us to go further in the comparison. We may indeed notice that spatial dependence tends to decline extremely fast in the case of Chinese counties: when observations are 200 km apart form each other, the measured autocorrelation index tends to be negligible, as if observed CSR values were almost completely unrelated. What is also surprising is the moderate intensity of spatial autocorrelation at short distance: it is only 0.45 for the closest counties. While this value is undoubtedly significant, it may seem relatively weak, as demographic behaviour is often spatially very clustered. Other demographic indicators (such as the proportion of married women in the 15-19 age group) exhibit stronger spatial dependence than the CSR. As Chinese counties were of distinctly smaller area18, we hypothesized that the scale factor might be to blame for this discrepancy. We therefore decided to re-aggregate Chinese data into larger regional units that are exactly of the average size of India’s. Once done, we computed again Moran’s index and found almost similar values (I = 0.53 for the new units below 50 km). This means that unit scale is not the major factor accounting for the observed spatial irregularities.

56At this point, it is difficult not to raise once again the possible effect of selective under-reporting of girls in 1990. If deliberate under-reporting of girls did indeed occur during the 1990 census, it must have resulted from a combination of active concealment from household members as well as data manipulation at a higher bureaucratic level. This latter aspect of underreporting refers to the active manipulation of census records in response to the population policy pressures from higher officials: in order to lower the fertility estimates, female children would be under reported. Such a phenomenon is likely to have had a somewhat erratic “spatial signature”, as local bureaucracies may have reacted very differently to the regional political context: while most officials might have kept the original figures, some isolated officials would have found it feasible and expedient to twist the records19. But these are only conjectures at this point and systematic comparison with the 2000 census figures is required to sort out the issues involved here.

57The message from this analysis is that the specific features of the CSR in India and China seen on the map can be also be modelled with geo-statistical tools and are therefore crucial characteristics of the geographical gender disparities in the two nations. Nonetheless, the patterns are somewhat different: clusters of high sex ratio are usually rather small in China and tend to fade away over a larger distance. In India, the influence of hot spots such as Punjab or cold spots such as tribal central India are perceptible over a very large distance and confer to the Indian map a global pattern.

58These results may now be related to the previous statistical analysis. The last section will briefly recapitulate our project and offer preliminary conclusions as to the specific features of gender inequalities in India and China.

Discussion

59Our aim was not to methodically examine the commonalities and differences of gender imbalances among children in China and India. This exercise is too vast to be fully conducted in such a preliminary analysis. What we have attempted to do here is more modest, as we have (temporarily) excluded the examination of causal processes of gender discrimination from our discussion. Whereas other scholars have looked for determinants of gender differentials, we rest content with correlates, well aware that a more systematic analysis in a comparative perspective would require a much larger array of explanatory variables and a more sophisticated statistical analysis in order to tackle the problems arising from endogenous variables. However, we have tried to provide a more systematic description of these differentials. For this, we have used disaggregated county-level data from the 1990 Chinese census that have hardly ever been examined and surprisingly have never before been mapped (to the best of our knowledge). Corresponding figures from India have been more often used to investigate the nature of the CSR in the country, but the 2001 figures have not yet been investigated thoroughly.

60Moreover, we have followed a similar protocol to analyse the regional variations. This involved the following constraints: use of the finest geographical data, selection of a comparable index of gender discrimination among children, comparable mapping using the same procedure and the same class intervals, use of comparable independent variables for regression analysis, and identical geo-statistical analysis. Comparative studies are usually discouraged by the lack of a common framework to approach new demographic issues such as gender inequalities. This is sometimes due to the absence of common variables that would follow a standard DHS-like format. Another reason is also that the contextual approach tends to dominate in view of the role played by local institutions and path-dependent trajectories of demographic policies: gender outcomes may then be read as idiosyncratic products of specific social and historical developments in given cultural settings and, as a result, comparison across nations appears to be a risky enterprise. As a matter of fact, the role played by the contextual features of both Chinese and Indian histories is well established, but we believe that cross-national comparison may precisely help to underline the specific character of each evolution, as the work by Das Gupta et al. (2003) suggests. It is also obvious that India and China as well as other Asian nations (such as Taiwan, South Korea, etc.) followed an analogous trajectory in terms of gradual sex ratio deterioration among children in a somewhat parallel demographic environment characterized most notably by fertility transition. We should expect to be able to identify common features in the demographic evolution of these countries.

61Our analysis shows that the recent thrust of gender discrimination in China and India has not at all been felt uniformly in the two countries. In each country, spatial disparities are wide and the deterioration of sex ratio levels among children has impacted mostly on some specific regions or pockets, while the rest of the nation has been little affected. The emergence or the intensification of a female deficit follows partly some historically defined contours. In the Indian case, which is better documented because of the available data from the colonial censuses, the existence of sex discrimination in western India (from Gujarat to Punjab) is well attested, as well as the prevalence of female infanticide. Nevertheless, the recent deterioration in China and India also affect areas where sex ratio imbalances are new. Moreover, the phenomenon seems likely to be associated with fertility decline, as son preference gets now expressed in terms of “quality of children”.

62What appears rather puzzling is the spatial patterning of this increasing sex discrimination, with macro-regions in India vs. micro-regions in China. There is no major spatial correlate of a high CSR among the social, economic and even demographic variables, except for the positive effect of literacy and urbanizing, which accounts for but a minor part of the observed gender disparities. The most prominent factors that emerge, at least in the Indian case, relate to the composition of the population, with Sikhs and to a lesser extent Hindus on one side, and religious minorities and Tribals on the other side. But these dimensions usually display a very strong level of geographical clustering and the distinction between the regional dimension and the social (cultural, religious, ethnical) dimension appears rather thin. It is probable that a strictly geographical measure such as the distance from the hot spots observed in India would yield an even better predictor of sex ratio differentials than social variables used here.

63This means that the child sex ratio seems to maintain a high level of “spatial autonomy”, whereby basic principles of geographical organization dictate the spread of the phenomenon. Whether diffusion processes at work in India are an issue cannot be examined here, but it seems likely that the female deficit can be seen as spreading from core areas to peripheral zones as well as from dominant groups to the rest of the local society. There is indeed an important diffusion aspect to this process that can be addressed here.

64In China, regional differentials in the CSR cannot be interpreted at the national level, except for the critical gap between Han-dominated China and the rest of the country. However, western China is rather marginal in terms of demographic size and the minority factor accounts for a very limited share of the observed sex ratio heterogeneity. Within historical China, differences are still very important and most provinces where sex ratio values are above 1.1 still include pockets that appear to be discrimination-free. When compared to India, the Chinese situation is quite unexpected in view of its assumed homogeneity in terms of history, culture, language and political institutions. In India where regional and sub-regional diversity is expressed in language (including alphabets), religious traditions, caste structure and political history, one would expect on the contrary more clear-cut disparities in sex ratio.

65In addition, very few factors in China seem to statistically account for the bulk of differences within eastern China. The nature of the spatial clustering, which is described with greater accuracy by our geo-statistical analysis, tends to suggest that local factors have played a major role in shaping the map of the child sex ratio. Whether these local factors relate to political management of the fertility policy and the accommodations it provided, or a to different set of characteristics more akin to regional social institutions is difficult to fathom at this point. However, these factors may also have been shaped to some extent by the quality of census information: as suggested earlier, the frequency of the under-declaration of children, especially with respect to girls. Not only are results from our statistical modelling rather limited, but the level of spatial dependence is also moderate (as is also the case for Chinese fertility). One would rather expect demographic behaviour to display a very high degree of spatial autocorrelation, even if this autocorrelation does not extend very far. An alternate explanation for this spatial dispersion therefore comes down to the quality of the census enumeration and to local factors conducive to statistical manipulation. Available data do not allow us to confirm this hypothesis at this stage.

Figures. Figure 1: Child sex ratio in China and India, 1953-2001

Figure 2: Administrative map of China (1990) and India (2001)

Figure 3: Map of child sex ratio, China (1990) and India (2001)

Figure 4: Map of child sex ratio, China (1990) and India (2001)

Figure 5: Spatial autocorrelation (Moran) index of child sex ratio, China (1990) and India (2001)

Map abbreviations