Introduction

1According to the last census, out of India’s total population of 846 million, 218 million, or only 26 per cent, were classified as living in urban areas in 1991. Thus India’s level or urbanization is quite low compared with many developing countries, and the level has hardly gone up at all in the decade 1981-1991. This reflects the changing rate of growth of the urban population. In the decade 1961-71 the growth of the urban population was 38 per cent, to be followed by the highest figure so far, 46 per cent, in the next decade, 1971-81. But in the most recent decade 1981-1991 the growth of the urban population has declined to 36 per cent (Pathak and Metha, 1995a). The general slowing down of urbanization has generated intense debate among scholars. Some scholars attribute it to under-enumeration of urban population while others present a wide range of plausible explanations. The most significant of these explanations are : a decline in the rural-urban migration, identification of relatively fewer new towns, and an increasing concentration of population in the rural areas adjacent to large urban centres (Premi, 1991 ; Krishnan, 1993). However, there are wide regional variations in the level of urbanization and the growth rate of urban populations, as well as in the development of regional urban conglomerations—as distinct from conurbations (Pathak and Mehta 1995b ; Jain, Ghosh and Kim, 1993).

2The Census of India has routinely since its inception placed towns and cities in 6 size classes, labelled I to VI. The class internals used have retained the same absolute population figures, and currently, as in the past, those cities with more than 100,000 population (1 lakh) are designated Class I. Census publications have usually tabled more extensive data in convenient form for these cities on an all-India basis, whereas data for smaller urban places is locked up in individual state volumes. Previous analyses of Class I cities in the 1961, 1971 and 1981 Census reports have revealed some striking patterns (Misra and Chapman, 1991 ; Chapman, 1983 ; Chapman and Wanmali, 1981). Perhaps principal among these has been the distinction between large, fast-growing industrial urban centres with adverse sex ratios on the one hand, and more stagnant, smaller, less industrial cities with more balanced sex ratios, and a great predominance of trade and domestic industry.

3This distinction in crude and general terms mirrors the distinction between generative and parasitic cities established in literature. The spatial pattern of urbanization has given credence to the correlations too : the more generative cities occur in clusters in the north-west, west and south of India, the more parasitic ones in the east, particularly in Bihar. A feature that has marked Indian urbanization as different from that in the west has been the positive correlation between size and specialization in employment. At what might perhaps be called a more mature stage of urbanization in the west, the largest cities have not shown the highest rates of growth in the last decades, and empirical data have supported the prediction that large places have less, rather than more, specialized employment characteristics.

4In this paper, we report on some of the findings of an analysis of data from the 1991 Census. For the first time the results of this census have been available on computer discs, thus enabling researchers to use more of the data in their analyses for any given investment of time, although of course requiring them to aggregate consistently from the different tables. Four principal data sets were available for the analysis reported on here (see list of variables in Table 6.1). These are :

1. For each state and union territory, a tabulation of the total for the rural areas of each district of persons in each of 36 categories.
2. For each state and union territory, a tabulation of the total for the urban areas of each district of persons in each of 36 categories.
3. For each state and union territory, a tabulation of the same 36 categories for each individual city, city component or town
4. For all Class I cities, the male and female populations in the 1991 Census, together with the same figures for the same cities in the 1971 and 1981 Censuses, even if they were then below the 100,000 threshold. This data set was not presented in this form in the 1991 Census, and was collated in the National Institute for Urban Affairs in New Delhi.

5From individual state tables in c) we have abstracted the set of variables for each of the cities listed in d), so that as well as the growth of the cities, we have the data for their employment characteristics and other social indicators such as sex ratios and literacy. There are 296 cities in d), contrasting with the 5,600 urban units for which data is provided in the total of state tables c). This does not mean that there are 5,600 urban places in India, since for larger urban places the tables provide breakdowns by internal subdivision of large urban areas. It is clear that c) is a very large data set warranting much more detailed study than we have managed so far, but that is beyond the scope of the current paper.

6No census was held in Jammu and Kashmir in 1991. So the data in a) and b) covers 449 districts (we have omitted the offshore Island territories). In the earlier works using the 1971 and 1981 Censuses, cited above, equivalent district data was not incorporated, hence in this paper we have for the first time the opportunity not only to analyse the characteristics of the Class I cities in 1991, but also to relate these if possible to the regional context revealed by district level data.

7The aim of the paper is to analyse the spatial variation of urbanization in India, to some extent to contrast urban and rural variables, and to examine in more detail the characteristics of the large cities. Correlations within and between the data sets are explored to see what further light these throw on the processes of urbanization an on urban structure. The paper is exploratory and descriptive in nature.

8Some of this data was not used because : 1) the data on area was inconsistent between data sets and quite often absent ; 2) the definition of occupied residential houses was thought to be open to variable interpretation ; 3) and similarly we were not certain of the consistency of the definition of household ; 4) we noticed that there were very few persons counted as marginal workers, and that the definition was again questionable ; 5) rather than use Non-workers, we preferred to use Employment Participation Rates based on the number of people actually defined in work categories.

9Most of the data was consistent between the different sets, although there were two minor discrepancies between the Class I city population figures, and figures derived from the state lists of urban areas. There were also inconsistencies in the spelling of some town and district names, which required careful attention and adjustment.

Spatial averaging

10Although the geographical surface of India is continuous, the data reflects in part the arbitrary way that it is aggegated into dichotomous spatial unit suc as districts and urban areas. This means that, for example, two adjacent districts might have a very high and a very low measure of percent urban, although they share a single urban area spreading from one to the other, so that the overall figure for that region ought to be somewhere between the two. To overcome these problems the district data has been subjected to spatial averaging. A more sophisticated method would be to use population potentials, as used by Chapman and Wanmali (1981), but on this occasion the resources were not available to make an equivalent analysis. Instead, the averaging has proceeded in a very simple way. All the districts have had their geographical centre recorded as X and y co-ordinates. Then for each district, the original value of any variable is replaced by an average value of the district itself and its six nearest neighbours. There is nothing special or magic about six, except that it is the average of the number of contact neighbours in a random point pattern. The transformed data produces a smoother surface, and is the data used for most of the maps of district variation shown here.

11For a correlation analysis of the district data, the raw data is used because by definition the averaging process introduces spatial auto correlation.

12From both the averaged data and the original data, new variables have been derived as listed in Table 6.2, all of which are available for the urban and rural values of each district separately, except of course a variable such as “% urban.”

13In order to examine the broad spatial patterns, the averaged data is further manipulated. Firstly, highly skewed data is transformed by taking either log or power functions as appropriate, and secondly this data is re-expressed in standardized scores (often known as “z” scores or sometimes “t” scores) with means of 0 and standard deviation of 1. As an example, Figure 6.1 shows the Urban Population as a percentage of the total. The symbols range from values (expressed in standard deviations) below -2 to values above 2. The absolute values are not depicted (but are shown in Table 6.3), but the result shows clearly the variation within India. In a sense, since relative values are being used, the procedure cannot fail to show variation : what is significant is whether the spatial pattern reveals new understanding. In this case it clearly does. The more lowly urbanized areas of Arunachal Pradesh, Assam, Bihar, eastern Uttar Pradesh and Orissa are expected. The higher levels in Tamil Nadu are expected : what was not expected to be so clear is the way in which two of the four metropolitan areas are so clearly much more intensively urbanized regional systems (Delhi and Mumbai) than the other two (Calcutta and Chennai). Though Calcutta is a huge metropolis, it is more singular massive urban area within a more rural hinterland. In the case of Mumbai, the urban system clearly stretches north into Gujarat—a point which will be picked up again below in a slightly different context.

Figure 6.1 : Proportion of district population in urban areas

Figure 6.2 : Proportion of the district rural population below the age of 7 years

Figure 6.3 : Variation in rural sex ratio, by district

Figure 6.4 : Variation in male / female urban literacy rates, by district

Figure 6.5 : Variation in male/femal rate)/ rural literacy rates, by district)

Figure 6.6 : (Male urban participation (female urban participation rate), by district

Figure 6.7 : (Proportion of tribal population in rural areas) / (proportion of tribal population in urban areas), by district

Figure 6.8 : Q-Analysis structure of Table 6.4 district correlations

14Figure 6.2 shows the variation in the percentage of the rural population below the age of 7 years. The south-north distinction is extremely clear, with the more youthful population stretching down nearly all of the Ganges areas of Punjab and Valley, and into Assam. Exceptions in the north are the Himachal Pradesh, which are consistently different on map after map, the Calcutta hinterland, and the southern areas of the north-eastern states. Again, in map after map, Nagaland and Manipur, and sometimes Mizoram and Tripura too, are distinct from Assam and Arunachal Pradesh. In the south of India, the two states of Kerala and Tamil Nadu form a complete contrast to the northern states. The map showing the percentage of urban population under 7 years (not shown) is almost identical.

15Figure 6.3 shows the variations in the rural sex ratio (see also AtKin et al. in tnis volume). The north-south gradient is known and expected, as also the peaking of the highest values in Kerala. What is perhaps less expected is the pattern within the north, where the region of very low values stands out so clearly within western Uttar Pradesh and close to the Delhi conurbation. The urban sex ratio (not shown) is similar, though not picking out western Uttar Pradesh quite so emphatically.

16Literacy is examined here by comparing the male and female literacy rates for both rural and urban areas. The two Figures 6.4 and 6.5, are essentially maps of discrimination by sex. The discrimination in favour of males in rural areas reveals a concentration in the “cow belt” of Hindu orthodoxy from Rajasthan to Bengal. Apart from the areas of the north-eastern states, this map shows a close proximity to the percentage of rural population under the age of 7. It is a fairly graphic association of the implicit link between low levels of female education and high birth rates.

17The next set of four maps (not shown) concern Employment Participation Rates —the percentage of male or female population recorded as employed. On the whole the rural maps show a lower rate of employment in the Gangetic north, with the exception of high male rates in Punjab and Haryana ; a higher rate in the whole of the Deccan with the exception that the participation rate in Kerala is low — which is something that is quite well known. The urban maps display more marked regional contrasts —in Nagaland, Manipur, Tripura and Mizoram the female urban participation rates show strongly, as indeed they do in Kerala. By contrast, in Punjab, Himachal Pradesh and the capital region, male urban participation rates are high, while female ones are low. Figure 6.6 shows explicitly the ratio of urban participation rates for males to the same rates for females. The greatest “imbalance” occurs in the south in Kerala and southern Tamil Nadu.

Figure 6.9 : Figurative diagram of correlations of Table 6.4

18Other kinds of “imbalance” can be shown for the difference for any variable between urban and rural areas. Figure 6.7 shows the rural/urban ratio of the tribal population. The “provocation” for producing this map was the recently announced intention of the company P and P to build a new port at Vardhavan north of Mumbai, in Thane district and close to the Gujarati border, in an area inhabited by the Warli tribe. Opponents of the plan highlight the impact that such developments would have on Warli tribal culture. In south Bihar in the 1960s and 1970s, the influx of Biharis and Bengalis into new industrial towns in tribal areas caused political tension and the reinvigoration of the demand for a tribal state, Jharkand. Figure 6.7 highlights several areas in India where such tensions may be more probable. In the case of the area north of Mumbai, it is also worth again considering Figure 6.1.

Correlation analysis at the district level

19The spatially-averaged data cannot be used for a correlation analysis, since by definition it has had autocorrelation built into it. Here we perform an analysis using data, which has not been averaged. The question remains open however, as to whether this data is still spatially auto-correlated, since neighbouring districts may be more likely to reflect each other than distant ones. However, if such autocorrelation is present, this does not mean that we cannot find patterns that are interesting, but rather that it is not possible to apply the usual tests of significance. In fact it is very easy to show that auto correlation is present, since several variables correlate with the south-north coordinate (and some, to a lesser degree, with the east-west).

20Skewed variables were transformed using either log or power functions, which eliminated most but not all of the skew from most of them. The data was then analysed using a Pearson product moment correlation coefficient. The exercise was repeated using a rank correlation coefficient, to eliminated the effects of whatever skew remained. The two correlation tables agreed with each other fairly closely, varying little in some cases in the strength of an indicated correlation. The results of the more robust (assumption free) rank correlation exercise are shown in Table 6.4. Given the number or observations (449) any correlation with an absolute value in excess of 15 is theoretically statistically significant at the 0.1 per cent level. However, autocorrelation means this value cannot be used with authority, and in any event the variance explained remains low.

Figure 6.10 : Q-analysis structure from Table 6.4 city correlations

21The variables clearly mesh with each other over a network of direct and indirect correlation. The next step in orthodox procedure would be use factor analysis, to impose orthogonal axes in place of the original variables. This procedure is not very satisfactory, because in a network of connections orthogonality seems an odd imposition, and because some variables may load partially on more than one factor. Here two different steps are followed. Firstly, and totally arbitrarily, correlations of less than [.5] are dropped from the analysis ; and secondly the remaining correlations are analysed in terms of Q-connectivity —a device which considers multi-dimensional and indirect connectivity. Q-analysis was developed by Atkin (1974), and is further exemplified by Chapman (1981 and 1984). The structure vector is shown in Figure 6.8. The vector shows how one variable —the Female Participation Rate— is connected to many others, but stands in a group somewhat on its own. On the other hand, both male and female literacy, both urban and rural, form a high dimensional cluster with the percentage of children under 7 years in both rural and urban areas. This information enables us to plot quite quickly the correlation diagram shown in Figure 6.9, where the direction of the correlations can also be shown. The correlations show both urban and rural characteristics vary together. Hence both urban and rural literacy rates inversely connect with both urban and rural populations under the age of 7. This does not mean that there are no differences between urban and rural areas —in absolute terms the levels of literacy and the percentage of children under 7 may differ quite widely in many districts. But it does say that variation across India in the absolute levels or urban areas will tend to mirror variations in rural areas, and vice versa. Hence variation within India will be found to be regional, rather than at the urban/rural interface.

22The diagram also highlights the pivotal nature of the Female Urban Participation Rate. It is connected both to urban and rural sex ratios, and is higher wherever these are more “favourable” (more females per male). These three together are all negatively correlated with the north-south coordinate “y” The value of y decreases with decreasing latitude, so again what is displayed is not urban-rural differences, but regional differences, again highlighting the higher southern sex ratios. The Female participation rate also correlates with Percent Scheduled Tribe in urban areas and the Female Rural Participation Rates. Inspection of the maps shows that this again is a regional result, reflecting in particular the combination of values in the north-eastern states.

23In the diagram the variable % Urban is strongly associated with % Scheduled Caste urban. This in a sense implies that there may be more work opportunities for the lower strata of society where urbanization rates are higher, but it may also reflect some of the regional impacts evident in the Punjab and north-west. Two variables remain disconnected — the urban and the rural Male Participation Rates. Inspection of the correlation Table shows that the rural rate has no particular high correlations— suggesting that it will be the result of local circumstance rather than any pan-national associations.

24The urban rate has quite strong negative association with the Percent of Rural Population under 7, and with urban Female Literacy Rates. One can only speculate in very general terms why this should be so ; that the Male Participation Rate is like the Female Participation Rate a sign of “modernity”, but less markedly so.

Table 6.4 : Rank correlations between 15 variables, 449 districts, 1991.

Table 6.5. Correlation tables, class 1 cities, 296 cases, 35 variables

Analysis of class 1 cities

25This is conducted in two stages. First, there is an analysis of the cities and their variables on their own in this section. Then an attempt is made to analyse the cities in relation to their regional settings.

26Table 6.5 shows the complete correlation matrix for the selected variables. Again, the approach adopted is not to use factor analysis to force orthogonal vectors onto this material, but to investigate the connectivity through a Q-analysis and a descriptive diagram. The value chosen for slicing the correlation matrix is arbitrary : in this case it is for absolute r values greater than .35 — a value at which a clear structure emerges in the Q-analysis. At low values, e.g. r = .1, many variables are related to many others, and at high values many variables and many connections between variables are excluded.

Figure 6.11 : Figurative diagram of correlations of Table 6.5

27The analysis shown in Figures 6.10 and 6.11 immediately highlights one point by comparison with the analysis of the 1981 data set. In that set the growth of cities over the previous decade 1971-1981 was correlated quite strongly with several variables, including size, the presence of organized industry, employment rate, and specialization. This time growth hardly seems correlated with anything : in the diagram it appears to be correlated with Mining and Quarrying, but closer inspection of the data reveals this relationship to be overstated, because the data remains skewed even after a log transformation, and one city, Ramagundam in Andhra Pradesh, has had by far the highest rate of growth and has high levels of employment in mining. That growth which has no particularly strong correlates in the more recent decade is interesting : a more diverse set of causes of the growth of cities perhaps indicates a more mature stage of urbanization.

28Many of the variables have been calculated separately for the two sexes. It is apparent that there is a close correlation between the two sexes for nearly all of these —for example the percentage of female and male employment in Mining and Quarrying, or in Industrial Manufacturing. This does not mean that the employment levels of the two sexes are the same for any category of employment —but it does mean that given whatever absolute levels occur, then variations from one city to another in one variable is reflected by variation in the other. This in turn suggests that different types of city economy do not have a discriminating effect on male versus female employment which is additional to discrimination from other sources — e.g. cultural or demographic variation.

29Three variables stand out in terms of the number of other variables with which they are correlated, and in terms of their total “r” values. These are Sex Ratio, north-south Co-ordinate (y), and percentage Females in Other Services. The first is commonly referred to, as there appears to be such a regional difference between high Female/Male ratios in the south, and low ones in the north. The correlation of this with the north-south coordinate confirms this. Given such a crude way of measuring location in a single dimension, the relationship is extraordinarily strong. The Other Services category is interesting too —it conjures up an image of pressure to find work in the informal sector. For both Females and Males it is negatively correlated with the respective workforce Participation Rate, and both are negatively correlated with levels of Industrial Manufacturing employment rates, and less specialized centres with more diverse employment at lower participation rates.

30All of these variables have been plotted on maps. Five of these are reproduced here in Figures 6.12-6.16, two of which will receive specific comment. The variable Specialization is an entropy statistic derived from all the employment categories together, and is a measure of how much employment is concentrated in one or a few categories. It does not differentiate between concentrations in two different categories — i.e. one city could be specialized in Industry and one in Household manufacturing, but both be equally specialized. The variable percent Males in Construction emphatically and curiously does not correlate with urban growth. There does appear to be some geographical pattern worth explaining though, since the low values occur in eastern Uttar Pradesh, Bihar, West Bengal and Orissa. These states collectively make up the stagnant part of two-speed India (Chapman, 1992) — not in the sense that population is not growing, but in that incomes per head in this zone have not improved for at least 30 years. This reminds us that a key variable which one would like to include in further analysis would be average incomes per head in these cities.

Figure 6.12 : Variations in proportion of males in construction industry in cities

Figure 6.13 : Variations in proportion of males in agriculture labour in cities

Figure 6.14 : Variations in sex ratio of cities

Figure 6.15 : Variations in degree of employment specialization by city

Figure 6.16 : Variations in females in other services predicted values

Figure 6.17 : Residual variation of male participation rate in cities from local

Figure 6.18 : Residual variation of female participation rate in cities from local predicted values

The local context analysis

31Here we make an attempt to relate cities to their own regional context Each city is linked to the district within which it is located. For these districts we have the average values of variables from the six nearest surronding districts and the district itself. It is thus possible to calculate the local average value for any rural variable, in the proximity of any city. It is also possible to calculate the urban value from the surrounding six districts and the district itself, but in this case omitting the value of the Class I city included in the district (to avoid including this data on both sides of a correlation.) Table 6.6 shows the results for values above 0.3. The Table invites almost endless speculation, and we do not presume to be able to pronounce on all the associations revealed. Some of the results are simple and straight-forward and much as to be expected. For example, the Percent Scheduled Tribe in Class I cities both on average correlate with their respective regional trends, both urban and rural. It is for this reason that departures from the trend as revealed in Figure 6.7 are to be noted. Literacy rates in Class I cities correlate with their local regional trends in literacy, again both rural and urban. Male Class I Participation Rates correlate with local regional Male Participation Rates, and Female with local Female Participation Rates. But then there is a distinction between Males and Females. The Female Class I Participation Rate also correlates with regional Female Literacy Rates, but for the males this is not so. Presumably, this is a hint that the characteristics of the male population of large cities are less likely to reflect local trends, perhaps because males migrate further to large cities than females. However, the Sex Ratios of the Cities are correlates with the urban and rural local Sex Ratios. Class I Female construction workers are correlated with Rural Scheduled Tribes, and both Urban and Rural Female Participation Rates. This relationship depicts well the use of such labour in the arc of “new” industrial towns in south Bihar and West Bengal, in Jharkand (see above).

32A last example of the possibilities of manipulating this data is a consideration of the extent to which particular values of variables for Class I cities differ from those that would be predicted by a local regional trend. In this case, we are specifically interested in Participation Rates in Rural areas. The logic of this is that where there are large differentials, then we might expect local rural-urban migration rates to be higher too. The analytical procedure is as follows. The average values for the rural Male and Female Participation Rates for the district containing the Class I city and its six nearest neighbours have already been calculated. The actual values of the participation rates in the Cities are then regressed against these local rural averages. The correlation between the average rural values and the individual city values is high —in the order of 0.5 — so we are right in supposing that cities on average do show participation rates which reflect ocal circumstance. The last stage is to map the residuals between actual urban participation rates and the expected participation rate predicted by the regression. Two maps are shown here Figures 6.17 and 6.18 — for the Male and Female Participation Rate residuals — again scaled in standard deviations of the residuals. The map of Male Residuals emphasizes the Delhi, Calcutta and Mumbai metropolitan areas to some extent, but not Chennai. In the south, the industrial cities of inland Tamil Nadu make a very distinct pattern on the map. Rural male participation rates in this area are not particularly low by national standards. In one of the lower ranked districts of Tamil Nadu —Coimbatore— they are about national average. But the Male Participation Rates in town are clearly much higher than expected from these values.

33It would be exciting to say that this demonstrates a dynamic and growing urban area. However, the figures for urban growth do not show this area to be expanding fast, and indeed growth rates for the Class I cities remain the hardest to correlate with any other variables. We have yet to agree on the true significance of this pattern.

Conclusion

34In a previous paper (Misra and Chapman, 1991) as a result of an analysis of data on the Class I cities of the 1981 Census, the authors concluded by showing a regionalization of India’s major cities. The fact that they produced such a regionalization was indicative of the belief that urbanization in India has distinct regional characteristics, as well as also having some national trends. Prominent among the national trends was the ability to explain growth rates of Indian cities in terms of their size and their manufacturing base. One implication of the correlation was the unwelcome one that already very large cities could grow even larger.

35In this paper, we have not found it possible to explain the growth rates of cities in the last decade in any simple way. Specifically, it is no longer true that the largest cities are more likely to grow fastest. On the other hand, we have shown that regional systems of cities are important, and that in some of these the growth is dispersed throughout the city region. These regions —as with Mumbai, Pune and parts of Thane district— are too big and too dispersed to be called simple conurbations, but the reality of their regional performance is clear. We have also been able to go one step further, and to show how many of the characteristics of cities are strongly correlated with their regional hinterlands, even if the absolute values of many variables do vary between urban and rural areas. Lastly, we have shown that sensitive manipulation of some of the ratio measures between sexes or between urban and rural areas can reveal significant aspects of the spatial process of development.

Table 6.6. Correlation between class I city variables and local district variables