The economic determinants of urbanization : the regional pattern of urbanization in India explained

目 THE ECONOMIC DETERMINANTS OF URBANISATION- THE REGIONAL PATTERN OF URBANISATION IN INDIA EXPLAINED by Rakesh Mohan Development Research Department World Bank Washington D.C. February 1984 I am arateful to M. K. Nanda for programming assistance. ABSTRACT This paper attempts to link various theories related to urbanisation in order to explain the regional differences in urbanisation as well as.the size distribution of cities in India simultaneously. It is found that the tight relationship between urbanisation and economic development linked with economic base theory and central place theory explains well the regional pattern of urbanisation in India. This is contrary to a common belief that - urbanisation in India does not have well understood economic determinants. A logistic relationship between level of per capita state domestic product and levels of urbanisation in states is found to fit the data well. Manufacturing employment is found to be a key determinant of population of large cities while employment in agriculture largely generates the demand for urban population in small towns. Thus, the growth of small and medium-sized towns is likely to be brought about by agricultural growth in the backward regions rather than indust:ial dispersal, while the policy on industrial dispersal is likely to succeed if there is a concentration of disp' rsal in large cities rather than a dispersal of concentration to small to . j. Central place theory would. then explain well the distribution of small towns mainly deriving demand from agriculture, while economic base theory is more relevant to the generation of larger cities, with large manufacturing employment constituting key base employment. Alternatively, different kinds of basic employment can be seen to generate succeeding "layers" of urbanisation in small and large towns as income increases with overall economic development. Introduction ,: Some Theoretical Background Interest in the regional aspects of development has a long and respectable history. Disputes over the causes of, as well as solutions to, the existence of regional disparities have beea of at least equally long- standing. Economic development tends to be concentrated in certain geographic regions as development proceeds from low levels of income. Moreover, the most obvious manifestations of concentration appear in Lhose cities which tend to stand out in sharp contrast to their surrounding countryside. Agriculture also is often seen to prosper in certain regions more than in others. The combined effect of such emerging patterns is that disparities are observed, on the one hand, between urbav and rural areas and, on the other hand, becreen regions. If agricultural prosperity is itself accompanied by higher urbanisation, the emerging regional differentials in income and welfare are thus even more clearly related to the differential rates of urbanisation. This paper is then an effort to relate the existing differentials in the levels and rates of urbanisation becween States to their levels of economic development. It is in teresting that although concern with inter-re-iortal disparities has been expressed in Indian plan documents for a long time, and that one of the a,rowed aims of planning is to reduce these disparities, there has been little attempt to relate their existence to differential rates of urbanisation. Most of- the concern with urbanisation has to do with the relative rates of growth in large and small cities. The reasons for this neglect are, perhaps, not hard to find. While there are respectable theories -2- linking urbanisation with economic development, others explaining the growth of individual cities (e.g. economic base theory), and yet others explaining the distribution of towns in agricultural regions (e.g., central place theory), there is little available combining these different strands simultaneously. A recent advance in this direction has been made by Koichi Mera (1974) by positing what he terms the "layer" concept of urbanisation. The tight link between levels of economic development and urbanisation is now well-established and accepted. A logistic equation relating levels of per capita GNP to levels of urbanisation, both measured traditionally, is found to fit international data quite well. Indeed, it is felt that the fit would be even better if there were not the inevitable measurement problems related to the definition of urban areas, which varies between countries as well as the well known problems in the measurement of GNP. It is found that as income increases, urbanisation proceeds slowly in the initial stages, accelerates (in the stage of countries which are now known as middle income countries) and again slows down at very high levels of income. This is found to be largely true for a cross section of countries at any given time and for a tiz- series of a single country over a period of time. The explanation for such a phenomenon can be stated in somewhat 6,ifferent ways. What is required is a framework for the whole economy and the place of urbanisation within it. Consider an economy with the three tradi':ional eccaomic sectors: agriculture, industiy and services. The production structure of these sectors can be modelled through the use of standard production functions. The returns to he factors of production (labor, capital and land), which interact to produce the output:: through these production functions, can then be attribuced -3- to each. Once this is done, the demand arising for eacr of the three products, given chese incomes, can be determined. Such a model can be simulated and the economy made to grow by adding rules of investment and saving. The results of such a model illustrate th typical path followed by an economy undergoing economic development and urbanisation. Agriculture is essentially located in rural areas, with industry and services in urban areas. Initially, at low levels of income, the 'large majority of the labor force is employed in agriculture producing at subsistence levels. Trade between the urban and rural areas is determined essentially, on the one hand, by the demand for food by tae small urban population and, on the other hand, by the small demand for urban goods by the rural population. At low income levels, large proportions of income are spent on food, while only the residual small amounts left are spent on urban gPods. As incomes rise, perhaps as a result of technological changes in agriculture, the proportion of income spent on food declines and, consequently, that on other goods increases. This is because, after nutritional needs are satisfied, only small increases can take place in the income spent on food; more technically, this is the familiar Engel effect resulting from the income elasticity of food being less than one. In the initial stages, when the marginal increases in income are still going towards satisfaction of nutritional needs, the increase in. demand for nonfood goods is slow. Consequently, the increase in demand for labor to produce the nonfood goods is also slow and hence the rate of urbanisation is slow. As incomes increase, and the proportion of income speat on nonfood goods continues to increase, the absolute demand for nonfood goods accelerates, the demand for labor in urban areas increases, and the -rare of urbanisation accelerates as well. At very high income levels, the rate of -4- urbanisation slows down again since there is a logical limit of 100 percent. The proportion of the labor force necessary to produce the food required Eor the urban areas approaches an irreducible minimum. This process can only happen when there are continual technological changes occurring in agriculture permits continual increases in labor productivity there. Some caveats and explanations are needed to.qualify the above tale. This somewhat elaborate, though perhaps commonsensical, explanation has been given since it is often said that the relationship between urbanisation and economic development is only an observed regularity and not well-based in a theoretical sense. The main question which remains unanswered in such a formulation relates to the reasons why non-agricultural activities are located in urban areas in the first place. This also raises the question, not addressed until now, on what constitutes an urban area. Conceptually, disregarding details and practical problems characteristically encountered in the classification of urban areas by census authorities, the essence of urban areas lies in (i) the size of settlements, (ii) the density of these settlements, and (iii) the predominance of non-agricultural activities. An urban area is essentially a settlement of some size with considerable density. The predominance of non-agricultural activities is actually a consequence of factors auch as size and density of a settlement. Since agricultural activities are land-intensive it is difficult for a large number of people engaged in agricultural activities to concentrate in one settlement. This is another way of saying that the elasticity of substitution between land and non-land inputs in agriculture is limited, while in industry and services it is high. As a result, industry and service activities tend to be capital-intensive and are therefore concentrated in space which leads to -5- concentration of human population as well. The other reason, though it is associated with the above, is that industries and services exhibit scale economies, as well as agglomeration economies much more than agriculture. The existence of industries exhibiting scale economies implies plants or certain optimal sizes, which induces coacentration. The labor employed by such a plant then requires associated services for consumption purposes. Similarly, the functioning of the plant requires other complementary services such as finance and banking. Moreover, complementary manufacturing facilities which provide inputs to the plant and which use as their inputs the plants' products find it profitable to locate near such a plant because of the existence of transport costs. Hence, it is the combined operation of scale economies, agglomeration economies, the existence of transport costs, and the comparatively higher nonland-land substitution elasticities in industry and services which make their location desirable in urban-areas. Koichi Mera's "layer" theory is similar to such a formulation attempting to explain the distribution of urbanisation. The agricultural sector's demand for non-food goods implies some urbanisation which results from agricultural activities themselves. This is Hera's first layer and also the basis of Christaller's "Central Place Theory". This largely explains the distribution of lower order settlements which serve as market centers for agricultural inputs and for produce. Since agricultural activities are sparsely distributed,-we would expect such centers to be small and well- distributed. The existence of bigger centers is explained by the existence of large industries and services such as ports and transport nodes, along with governmental centers. If these activities (exogenously determined) are regarded as "basic" or "export-based", their distribution will determine the -6- distribution of the urban population which must locate near these activities to service them. These are the second and the third layers. Thus, each set of basic activities - primary, secondary and tertiary, ha.ve multipliers which determine the resulting urban population. According to this formulation, one has to isolate which of the secondary and tertiary activities are "basic". This is not easy, but, consistent with the idea of eocnomies of scale, large industry can be regarded as basic in the secondary sector. The location of much of large industry is also determined by proximity to raw materials - particularly iron and steel. This is determined by the principle of comparative advantage and may clearly be regarded as basic or exogenous. Similarly, the location of government activity can be regarded as exogenous. The latter is more troublesome, for presumably, the distribution of government activities is itself determined by the location of other activities. To the extent, however, that it is discretionary, it may be regarded as basic or exogenous. The task then in this paper is to investigate how the regional pattern of urbanisation and economic development in India conforms to this conception of an economy. Specifically, it is sought to relate the variation in urbanisation between states to the variation in income, and subsequently, to explain the variations in urban population to different measures of the employment in basic activities. This is done at this time in particular to understand the pattern of urbanisation that has become apparent from the 1981 Indian Census. It is important to make such an investigation since there is a widespread school of thought which believes that the pattern of urbanisation that exists in the country is highly concentrated and "dysfunctional" ./ Others allege that the urbanisation that has taken place in the country is in some sense "artificial" and not warranted by the existing structure of the economy. Vhile this paper does not offer normative views on the desirable pattern of urbanisation in India its aim is to document the existing pattern, disaggregated by states, to describe how it has evolved over the past 30 years, and to provide some pointers concerning the determinants of urbanisation. If 'he pattern is found to be consistent with what might be expected and with what is generally found in other countries, today and historically, it can at least be stated that the pattern observed is not surprising, is quite understandable and is at least warranted. It can still be regarded as undesirable from a normative viewpoint, but it would then also be clear that a change in the pattern would require much stronger policy measures. The next section will document the record of urbanisation disaggregated by states in India since Independence. It will be the aim of the following section to explain the record documented in this manner, given the theoretical background outlined above. II The Regional Pattern of urban Growth and Economic Development in India: the Record since Indecendence The release of the provisional population totals of the 198 Census reveals a significant acceleration in the speed of urbanisation in the country. This is true whether comparison is made with the historical record since the beginning of the century or with what was expected as recently as 1/ This is well argued by Kundu and Raza (1982) for example. -8- 1979. 1. The Sixth Five-Year Plan projected the level of urban population to be about 148 million in 1981 and the level of urbanisation to be 22.04 percent. In fact, the 1981 Census showed that the level was about 156 million (but this number excludes Jammu and Kashmir and Assam, not enumerated at the time of publication of Census results). In another recent paper (Mohan and Pant, 1982), we have documented the fact that India has had a relatively slow but stable rate of growth in its urban population since about 1921 during which the level of urbanisation has inched up slowly from about 11.3 percent of total population to about 23.7. percent now. Despite the recent acceleration, it must also be recognised that the rate of growth of urbanisation in India is still one of the slower in the world. A key feature of the Indian settlement pattern which emerged from the analysis was the very stable structure of settlements. While total urban population increased sixfold between 1901 and 1982 the number of settlements increased by only 80 percent. Thus most of the growth was because of the enlargement of existing towns at every level and not mainly because of the addition of new towns. This implies that the majority of settlements now classified as towns have exhibited urban characteristics for a very long time. The lack of addition of towns at the boccom end has remained something of a puzzle often remarked on by analysts. Indeed, one result of this growth pattern is that the Indian settlement structure has become increasingly top heavy. The proportion of urban population residing in towns above any cut-off point has risen continuously since the turn of the century. This phenomenon has led many 1/ Census of India, 1971, India. Series 1, Report of the Expert Committee on Population Projections Paper 1 of 1979. -9- observers to note that larger towns have grown faster than smaller ones. This is simply not true, as has been documented in detail in the aforementioned paper. In reality, the increasing proportion of urban population in larger cities has merely been the result of progressive accretion to existing- settlements of all sizes which are relatively well spread out spatially. The lack of emergence of new towns or of existing villages acquiring urban characteristics is indeed a matter for concern. To the extent that urbanisation is related to economic development and the distribution of lower settlements is well explained by Central Place Theory, the dearth of new urban settlements must be related to the relatively slow growth of agriculture in many regions of the country, coupled with the fact that settled cultivation has existed in India for a very long time. The latter would suggest a well- distributed pattern of settlements which would get denser only if there were a relatively fast rate of growth in agriculture and consequently in agricultural demand for urban goods and services. This is an eminently researchable issue but is not addressed directly here. Having stated that the Indian settlement structure has been quite stable, it is necessary to disaggregate the trends in urbanisacion, up to at least the state level. The variation in experience between states is surprisingly large and one that seems to be not readily explicable at first sight. Table I gives the level of urbanisation in the major states of India (i.e., those with population greater than 10 million in 1971 except Assam which is yet to be enumerated for the 1982 Census and Kerala). Kerala has not been included here because it poses particular problems in the definition of urban areas. Its overall population density was about 550 people per sq. km. in 1971, the rural density itself being about 480. Apart from those - 10 - settlements which have an organised local body such as a municipality, corporation, cantonment board or notified town area committee which are classified as urban, the Indian census classified settlements as urban if they have - (i) a minimum population of 5000, and (ii) a density of at least 400 people per sq.km., and (iii) at least 75 percent of the male labour force is in non- agricultural activities. It turns out that in Kerala, in 1971, about 80 percent of the rural population lives in villages of over 10,000 population and another 15 percent in villages in the range of 5000 to 10,000 population. Hence, almost all the settlements would qualify for classification as urban settlements were it not for the additional requirement of 75 percent of the male labour force being in non- agricultural activities. With such high overall densities it is also not easy to decide on the boundaries of settlements. Kerala has therefore been omitted in this descriptive part of the paper; however, it is included, in the more systematic analysis offered in the next section. Table 1 gives the level of urbanisation in States according to two definitions. First is the population of all towns as a percent of the total population and second is the same, but including the population in towns over 20,000 only. The definitional problems referred to above are essentially at the lowest levels. All settlements of 20,000 and over are easily classified as urban. Thus, this avoids kinks in trends due to definitional problems. The most industrialised states of Maharashtra, Gujarat and Tamil Nadu are now all over 30 percent urbanised (according to the Census definition). The industrial stagnation of Calcutta is reflected in the level of urbanisation in -11 - Table 1: LEVEL1 OF URBANISATION IN STATES2 1951-1981 Population of All Towns Population of towns above State as % of Total 20,000 only as % of total 1951 1961 1971 1981 1951 1961 1971 1981 1 2 3 4 5 6 7 8 9 Andhra Pradesh 17.41 17.4 19.3 23.3 10.8 13.2 16.0 21.1 Bihar 6.8 8.4 10.0 12.5 4.9 6.5 8.0 10.9 Gujarat 27.2 25.8 28.1 31.1 18.3 20.4 22.5 26.7 Haryana 17.0 17.2 17.7 22.0 10.9 12.5 13.9 18.1 Karnataka 22.9 22.3 24.3 28.9 13.4 15.6 18.2 23.9 Madhya Pradesh 12.0 14.3 16.3 20.3 7.7 9.7 12.2 15.7 Maharashtra 28.8 28.2 31.2 35.0 20.7 23.8 27.7 32.2 Orissa 4.1 6.3 8.4 11.3 2.1 3.5 6.0 9.0 Punjab 21.7 23.1 23.7 27.7 14.5 17.5 18.5 22.4 Rajasthan 24.4 26.7 30.3 33.0 13.0 21.0 26.3 29.9 Tamil Nadu 24.4 26.7 30.3 33.0 13.0 21.0 26.3 29.9 Uttar Pradesh 13.6 12.9 14.0 18.0 9.4 10.6 11.9 13.8 West Bengal 23.8 24.5 24.7 26.5 21.4 22.3 22.9 25.3 India3 17.6 18.3 20.2 23.7 12.3 14.5 16.9 20.5 Notes: 1. Urban population as proportion of tctal population. 2. Including all states with total population greater than 10 million in 1971 but excluding Kerala and Assam. 3. Including all states except Assam and Jammu and Kashmir. Source: Tables A.1 and A.2. - 12 - theat stable since 1951. These states conform more to middle-income countries with about $500 per capita income in terms of che level of urbanisation. At the other end of the scale are Orissa (11.3%) Bihar (12.5%), and Uttar Pradesh (18.0%). There are only about 10 countries in the world at the lowest per capita levels pressing levels of urbanisation below 12 percent. Thus, in terms of urbanisation levels, India's regions exhibit patterns spanning 'the whole range of about 50 countries with annual per capita incomes from about $100 to $400. The variation in levels of urbanisation 1 has, however, declined as measured by the coefficient of variation, from about 0.45 in 1951, to 0.42 in 1961, 0.50 in 1971 to 0.34 in 1981. The acceleration in the least urbanised states, along with the deceleration of the most urbanised ones, has caused this major change in the last decade. Table 2 gives the growth race of urban population in States from 1951 to 1981. As is evident from column 3, all of the relatively poorer states (Andhra Pradesh, Bihar, Madhya 1radesh, Orissa and Uttar Pradesh)Z have experienced rapid urban growth during 1971 tc 1981, while only Haryana among the richer states has experienced comparable growth. In fact, the old industrially developed states of West Bengal, Tamil Nadu and Maharashtra have the lowest growth rates. Naturally, absolute increases in urban population in these states continue to be large due to the higher initial levels. As in Table 1, columns 5 and 7 of Table 2 give comparable growth rates for urban 1/ Tncluding Bhutan, Nepal and Bangldesh. 2/ The coefficient of variation in the levels of urbanisation is taken for levels according to urban population in towns above 20,000. 3/ This is documented below. - 13 - Table 2: STATEWISE GROWTH OF URBAN POPULATION 1951-1981 (Percent per year) State All towns Towns above 20,000 only 1951-61 1961-71 1971-81 1951-61 1961-71 1971-81 1 2 3 4 5 6 7 Andhra Pradesh 1.5 2.9 4.0 3.5 3.9 5.0 Bihar 4.1 3.7 4.4 4.8 4.1 5.4 Gujarat 1.8 3.5 3.5 3.5 3.6 4.2 Haryana 3.1 3.1 4.8 4.4 3.9 5.2 Karnataka 1.7 3.1 4.2 3.5 3.8 5.3 Madhya Pradesh 4.0 3.9 4.6 4.6 5.0 4.8 Maharashtra 2.0 3.5 3.4 3.6 4.0 3.8 Orissa 6.5 5.2 5.3 7.4 7.9 6.0 Punjab 2.6 2.3 3.7 3.9 2.6 4.1 Rajasthan 1.1 3.3 4.6 2.8 4.3 5.4 Tamil Nadu 2.1 3.3 2.5 6.1 4.3 2.9 Uttar Pradesh 0.9 2.7 4.9 2.9 2.9 3.9 West Bengal 3.1 2.5 2.8 3.3 2.7 3.1 India3 2.33 3.26 3.85 3.59 3.85 4.18 Notes 1. Including all states with total population greater than 10 million in 1971 but excluding Kerala and Assam. 2. Including all States except Assam, Jammu and Kashmir. VRA/T-002U - 14 - population in towns above 20,000 only. While most of the observations made above are still valid, there are some changes that appear significant. The changes between decades are much less with this definition, presumably because the definition is more consistent between the States. The acceleration between decades is also less pronounced--although the rate of growth for each State is much higher. This result extends the idea that the Indian settlement structure is of longstanding and is also stable at the state level. Thus, the fact that the urbanisation rate for (class III) -L towns and above is higher than for all towns taken together again points to the fact that most urbanisation is by accretion to existing towns of all sizes, and there are only small additions of new towns at the low end of the range. This now appears to be true at the state le.vel as well. It may, however, be expected that in States such as Orissa, U.P. (particularly Eastern U.P.) and Madhya Pradesh (particularly Eastern Madhya Pradesh), where intial urbanisation levels are low and towns are located sparsely, there would be a greater potential for the emergence of- new towns. Indeed, one finds that with the second definition, an acceleration in urbanisation is not apparent in these states, indicating that much of the observed acceleration is because of the emergence of new towns. The problems caused by variation across states in the classification of towns at the low end has already been alluded to. Table 3 indicates the differences caused by the addition of new towns in the 1981 Census. The states which have added a significant number of new towns are essentially Haryana, Madhya Pradesh, Rajasthan, Orissa and Uttar Pradesh. Many of these 1/ Towns over 20,000 population. Lable 3: STATWISE GROWTH OF IURAN PPULUTION EXClUDING NEW T0WNS~ ADOED IN 1981 1,9 71-19 81 Tocal urban population Annual gowth race No. of excluding new Uncorrect:ed Corrected State No. of towns new towns c0'1ns (percen per year 1981 1981 L981 ('000) 1971-81) 2 3 4 5 6 Andhra Prade,sh 234 29(2) 12,160 4.0 3.8 Bihar 179 30(4) 8,374 4,4 4.1 Gujaraz . 220 29(2) 10,289 3.5 3.4 Haryana 79 17(6) 2 64 4.8 4.1 Karnataka 250 34(4) , 10, 3 9 3.3 3.4 Madhya Pad,esh 303 72(6) 9, 9,6 4.6 3.9 '-aharashra 276 31(2) 21,616 3.4 3.2 Orissa 103 27(9) 2,836 5.3 4.4 ?unjab 134 28(4) 4,419 3.7 3.2 Rajaschan 195 43(7) 6,612: 4.6· 3.8 Tamil Nadu 245 18(1) 15,7'74. 2.5 2.5 Uctar Pradesh 659 368(16) 16,829 4.9 3.1 Wes c Bengal 130 119(1) 14,236 2.8 2.7 India4 3,007 770(5) 136,169 3.81 3.35 NoIes: 1. All scates i 1971 populacion of 10 mill.ion and above. 2. No. of towns added in 1.981, i.e., towns not classified as such in 1971 (including chose classified as towns in an earlier census buc noc in 1971). 3. Figures in brackets give che populacion of new cowns as a percentage of tota.L urban populacion. 4. Including only che 13 scares above. 5. Race of growth of total urban populacion in States. 6. Race of growth of urban populatioo in state but excluding new cowns in 1981 and oluding 1971 eowns not found in L981 census as deca<led in Appendix 2.1 Source: 1. Census of India 1981 Series I, India Paper 2 of 1981. 2. Census of India 1971 Series I, India Part II-A(1) General opulacion Tables Scac,menc 5, page 135. 'TRA/T-0 03U - 16 towns in U.P. should iave been classified as such in 1971 (Premi and others, 1977). Thus, the aDpparent large acceleration of urbanisation in U.?. is at least partly illusory, though not entirely so. Taking the same towns as in 1971 (columna 6, Table 3), the growth rate did increase from 2.7 percent per year in 1961-71 to 3.1 percent in 1971-81. Similarly, if paly towns over 20,000 are con7)idered (columns 6, 7, Table 2), the rate increased from 2.9 to 3.9 percent a year. Similarly, for Haryana, thd corrected growth rate of 4..1. percent is still significantly higher than the 3.1 percent for 1961-71, The conclusion, on the basis of Table 3, then, is that the definitional problems of towns in the 1981 Census do make a difference to the recorded growth rates of 4 or 5 particular states--showing a much larger increase--but do not alter the overall pattern described from the growth rate of urban population, whether it is according to the usual definition or to that for towns above 20,000 population. The population of the added towns, as a proportion of state urban population, was 6 percent for Haryana, 16 percent for U.P. and 9 percent for Orissa. At the all India levels then, the difference made by these definitional issues is small. The total population of the towns newly classified as such, is only about 5 percent of the total. For some of these which should have been classified as towns for the first time in this Census, this should not be regarded as an error. The "error" is caused by those towns, mostly in U.P., which should have been classified as towns in the 1971 Census itself. Tabile 4 which gives the rates of growth of rural population, is of further interest. There have been significant declines in the rate of growth of rural population in the high agricultural productivity states of Haryana and Punjab, but small increases have taken place in the low productivity - 17 - Table 4: STATEWISE' GROWTH OF RURAL POPULATION DI DIDIA 1951-1981 Annual Rate of Growth of Rural Population State (percent per year) 1951-61 1961-71 1971-81 Andhra Pradesh 1.46 1.68 1.60 Bihar 1.64 1.78 1.89 Gujarat 2.61 2.29 2.00 Haryana 2.93 2.78 1.96 Karnataka 2.06 1.93 1.73 Madhya Pradesh 1.92 2.31 1.77 Maharashtra 2.2 2.03 1.43 Orissa 1.58 2.03 1.43 Punjab 1.79 1.89 1.55 Rajasthan 2.63 2.32 2.42 Tamil Nadu 0.81 1.53 1.20 Uttar Pradesh 1.65 1.68 1.81 West Bengal 2.80 2.37 1.85 India4 1.89 2.00 1.75 Notes: 1. Including all states with a population of 10 million or more in 1971 excluding Kerala and Assam. 2. Including all states except Assam, Jammu and Kashmir. VRA/T-004U/1.34 - 18 - states of Bihar, Rajasthan and U.P. There have, however, been declines in other poor states such as Orissa and Madhya Pradesh. Preliminary census results indicate that for the first time since the turn of the century, there may have been a perceptible decline in the proportion of the labour force engaged in agriculture - from 69.8 percent in 1971 to 66.7 percent in 1981 (J.N. Sinha, 1982). This is at least consistent with the decline in the overall rate of rural population growth. Increases in agricultural production are now coming mainly from productivity changes-only small increases in cropped area can now be detected. There are therefore indications that the absorptive capacity of agriculture for continued increases in population and labor force is now declining. Table 5 tabulates the urban-rural growth differential (U.R.G.D.)--yet another index of urbanisation. This is merely the difference between the urban and rural population growth rates. The acceleration in urbanisation is even more evident from this table--since urban population growth rates have increased in general, while rural growth rates have declined. Again, it is only Tamil Nadu which shows a decline; there have been significant increases have occurred in all other states. The progress over the three decades is quite remarkable. There were as many as 5 States with negative URGD, only one with less than 1.0, and as many as 9 with over 2.0. The rural urban transformation and its acceleration in all the States therefore stands out with much.greater clarity by considering URGD. It is interesting to find that the distribution of population growth rates for rural as well as urban areas has tended to become more uniform between states over the same period, The coefficient of variation between states of rural population growth rates has declined from 0.36 for the 1951-61 - 19 - Table 5: URBAN RURAL GROWTH DIFFERENTIAL (URGD) BY STATES1 1951-1981 State URGD3 1951-61 1961-71 1971-81 Andhra Pradesh 0.01 1.28 2.42 Bihar 2.43 1.93 2.55 Gujarat -0.76 1.21 1.48 Haryana 0.12 0.31 2.80 Karnataka -0.37 1.13 2.44 Madhya Pradesh 2.06 1.59 2.78 Maharashtra -0.27 1.45' 1.79 Orissa - 4.87 3.19 3.91 Punjab 0.79 0.39 2.14 Rajasthan -1.58 0.99 2.20 Tamil Nadu, 1.25 1.79 1.28 Uttar Pradesh -0.70 1.03 3.08 West Bengal 0.32 0.16 0.93 India2 0.48 1.29 2.11 Notes: 1. Including all states with a population of 10 million or more in 1971 but excluding Kerala and A<sam. 2. Including all states except Assam, Jammu and Kashmir. 3. Taken as the difference between the growth rate of total urban population (census definition) and rural population growth rates. VRA/T-005U/1.84 - 20 - period to 0.17 for 1961-71 to 0.16 for 1971-81. The corresponding- coefficients for urban population growth are 0.31, 0.31 and 0.20, respec- tively. 1" Furthermore, the pattern is confirmed if taken back to the 1941-51 decade. The coefficient of variation for rural population growth rates for . that decade was about 0.49, and for urban population growth rates was about 0.31 again. What is of great interest is that the variation in urban population growth rates has declined as well in the past decade. It may be somewhat premature to draw'strong conclusions from the decline in one decade of this one inequality measure--the coefficient of variation--after it had been stable for 3 previous decades. Firstly, this decline is presumably a result of the lower variation across states in total population growth rates. But it may also be a result of the dispersal of industrialisation that has taken place between states--although this is still of small magnitude. Sekhar (1983) has documented the decline in the index (the Theil Index) of inequality between states in organised industry over the last 2 decades--whether measured by value added'or employment. It, must be emphasised that the changes have been small; as late as 1976, Haharashtra, West Bengal, Gujarat and Tamil 'Nadu together accounted for about 5 percent of .all value added and 52 percent of all employment in the manufacturing (factory) sector. In 1961, in comparison, these four states accounted for as much as 66 percent of valued added and 58 percent of employment. Given this small change, not much can be made of the decline in variation of the rate of urbanisation--only that two results are at least consistent with each other. 1/ These coefficients of variation for urban population growth are calculated for the growth of towns above 20,000 only (Columns 5,6,7 in Table 2) to avoid the variations due to low and definitional problems. - 21 - Table 6: URBAN GROWTH IN STATES DURING 1951-1981 Towns of population Category of states All towns above 20,000 only 1 2 3 I. States where the rate Andhra, Karnataka Andhra, Gujarat of growth of urban Rajasthan and Uttar Karnataka and population has increased Pradesh Rajasthan continuously since 1951-61. II. States where the rate Gujarat, Maharashtra Madhya Pradesh, increased between and Tamil Nadu Manarashtra and 1951-61 and 1961-71 Orissa but declined or remained constant thereafter. III. States where the rate Bihar, Haryana, Bihar, Haryana, declined or remained Madhya Pradesh, Punjab, Uttar constant between 1951- Orissa, Punjab Pradesh and 61 and 1961-71 but and West Bengal West Bengal increased thereafter. IV. States where the rate None Tamil Nadu of growth of urban population has increas- ed continuously since 1951. Source: Table 2. VRA/T-006U/1.84 - 22 - How can all these data be summarised? Table 6 groups all of the states into four categories. The variation in the experience of the different groups of states is immediately apparent. Thus, even in the case of richer states, two patterns that are diametrically opposite are revealed. Thus, in the old industrially and commercially developed states of Maharashtra and Gujarat, the urban growth rate increased initially, but tapered off thereafter, while in the agriculturally progressive developed states of Punjab and Haryana, the opposite was experienced. Clearly different forces are at work in these States, and more detailed analysis is required before further generalisation can be made. Having examined the record of urbanisation, it is now necessary to look at different indices of economic development over the same period. Consistent data sources are, however,.difficult to obtain for 1951 because of the reorganisation of States in 1956. Hence, only the record since 1961 is presented. Cons.iderable controversy surrounds the magnitudes of per capita state product. Given a countrywide common market it is difficult to compute state domestic product.. However, State Statistical, Bureaus as well a, the Central Statistical Office (CSO) make annual estimates of the state domestic product (SDP). There are usually discrepancies between the two sets of estimates, and the CSO issues a set of consistent estimates with some lag. These are currently available for the period 1970-71 to 1975-76. More difficult still is a constant price series of SDP, and no official series has been issued. The data used here is from a compilation of data by the Center for Monitoring the Indian Economy, an independent organisation. While recognising that individual errors might exist in this data, it is considered adequate for the purpose at hand: it does give a good picture of the magnitude - 23 - and pattern of disparity in incomes between states and how this pattern has evolved over the past 20 years. Table 7(1) gives estimates of per capita SDP at constant 1970-71 prices for 1961,1971 and 1978-79 the latest data available for a "normal" year. The coefficient of variation (CV) for per capita income has moved from 0123 in 1961 to 0.26 in 1971 to 0.33 in 1979. This is in the opposite direction from the CV for level of urbanisation which are 0.37, 0.34 and 0.29 for the same years respectively. The ratio of the per capita SDP of the richest to the poorest state was about 2 in 1961, 2.5 in 1971 and 3 in 1981. By international standards these are not high numbers but obviously a matter of some concern that spatial disparities are getting worse in the country. It is useful to get an idea of regional disparities in other countries to get a better perspective on our own. For example, the CV for Argentina was 0.59, Brazil 0.58, Philippines 0.75 and Thailand ).50, in the early to late sixties. -/ The average (unweighted) for 15 less developed countries tabulated was 0.48. the ratio of the poorest to richest regions was about 12 in Indonesia, 9 in Brazil and 5 in Colombia. By these standards, the income disparity between states is very low in India but what may be a matter of deep concern is the worsening of the disparity over time. this has been well appreciated in Indian planning and preferential allocation of funds to the poorer states has been incorporated for some time. Furthermore, there is an elaborate system of incentives for industries to locate i the poorer regions of the country. I/ From Gugler and Gilbert (1982). -24 Table 7(1) also presents data on the per capita value added in the factory sector. Note that these data are in current prices. The CV has declined from 0.92 in 1961 to 0.67 in 1971, and finally to 0.62 in 1978. The ratio from the highest to the lowest was about 14 in 1961, and 7 in 1971 and 1978 (there being a small decline between the latter two years). It is clear then that the worsening disparity in state incomes has not been caused by the manufacturing factory sector--indeed, there has"been a strong opposite trend toward equalisation in this sector. The poorest 3 States in 1961 were Bihar, Orissa and Uttar Pradesh, in terms of income. They also had, along with Andhra Pradesh, Madhya Pradesh and Rajasthan, the lowest value added per capita in the factory sector. The position has not changed much in 1981 and these three states remain among the bottom five or six according to both of the indicators.. The rates of increase in the factory sector's value added have, however, been higher in the poorer states. It is quite noticeable from the factory sector valued added table that from a situation when the only substantial manufacturing was in Gujarat, Maharashtra, West Bengal and Tamil Nadu, and located around their capital cities, there has been a considerable dispersal of industry by now. The first panel in Table 7(2) gives the data on agricultural productivity. This has been defined as the total production of foodgrains per male agricultural laborer. Only male labor has been used because of the well known definitional changes in the 1971 Census which has made the female agricultural labor data noncomparable with ocher years. The data for males is believed to cause fewer problems. The increase in disparity--CV increasing from 0.30 in 1961, to 0.59 in 1971 and 0.73 in 1981--is quite striking, and consistent with the increasing income disparity. The stagnation in-agricul - 25 - Table 7(1): SOME SELECTED INDICATORS OF ECONOMIC DEVELOPMENT Per capita state NDP Value added in factory State (constant 1970-71 prices)- sector (current Rs. per capita) 1961 1971 1981 1961 1971 1981 1 2 3 4 5 6 7 Andhra Pradesh 518 586 678 9 29 74 Bihar 390 418 438 14 31 57 Gujarat 697 845 884 52 108 245 Haryana 627 932 1,029 - 70 169 Karnataka 559 675 723 14 62 107 Madhya Pradesh 472 489 463 a 27 68 Maharashtia 769 811 1,008 69 167 324 Orissa 392 541 514 6 27 62 Punjab 760 1,067 1,308 4 52 .136 Rajasthan 500 629 591 5 26 56 Tamil Nadu 571 616 582 24 75 166 Uttar Pradesh 457 493 527 9 24 47 West Bengal 758 729 465 50 97 173 Mean 558 676 727 23 60 127 4 Coefficient of variation 0.23 0.26 0.33 0.92 0.67 0.62 Notes: 1. 1978-79 data 2. 1977-78 data VRA/T-007U/1.84 -26- Table 7(2): SOME SELECTED INDICATORS OF ECONOMIC DEVELOPMENT Agricultural5 Power-consumption Productivity per capita State (tons per person) (KWH) 19611 19712 19813' 1961 1971 19814 1 2 3 4 5 6 7 Andhra Pradesh 0.88 0.83 1.06 20 56 93 Bihar 0.71 0.69 0.67 23 65 87 Gujarat 0.59 0.94 0.85 55 138 231 Haryana - 2.77 2.91 - .98 211 Karnataka 0.82 1.06 1.13 41 104 149 Madhya Pradesh 1.24 1.26 0.98 16 54 96 Maharashtra 0.92 0.65 1.17 78 158 230 Orissa 0.98 0.96 0.98 32 95 115 Punjab 1.736 3.07 4.36 706 159 303 Rajasthan 1.10 1.44 1.05 3 50 93 Tamil Nadu 0.90 0.98 1.07 58 130 184 Uttar Pradesh 0.85 0.99 0.99 14 60 88 West Bengal 0.92 1.11 1.04 75 118 118 Mean 0.94 1.23 1.35 40 97 150 Coefficient of variation 0.30 0.59 0.73 0.61 0.38 0.45 Notes 1. Average of 159-60 to 1961-62. '2. Average of 1970-71 and 1971-72. 3. Average of 1978-79 to 1980-81. 4. 1978-79 data. 5. Defined as total foodgrains output in the State divided by total male agricultural labor. 6. Undivided Punjab. VRA/T-007U-2/1.84 - 27 - tural productivity in the whole Eastern Region is the main cause of the increasing inequality. Productivity has actually decreased or remained almost constant in Bihar, Orissa, West Bengal, Madhya Pradesh and Uttar Pradesh. The data for 1981 is somewhat on the low side because the extremely bad year of 1979-80 has been included in the computed average. The coorelative picture would, however, not change much. The next panel gives another useful indicator, the power consumption per capita. It is clear that there have been large overall increases in power consumption in all states. The CV declined from 0.61 in 1961 to 0.38 in 1971, but then increased again in 1981. Note the very high increases in Punjab and Haryana. Consistent with industrial as well as agricultural stagnation of West Bengal, it is the only state which does not record an increase in per capita consumption of power. Table 7(3) gives two more indicators--the male agricultural employment and total factory employment-as percentages of their respective totals. There has been a clear decline-in percentage male agricultural employment, in all states except Tamil Nadu and U.P., between 1961 and 1981. There was a general increase between 1961 and 1971, possibly because of definitional problems. On the other hand, except for West Bengal, and as might be expected, the proportion of people employed in the factory sector has increased between 1961 and 1981 in all the the states. All of these results are quite consistent with some of the conjectures offered for the observed pattern of urbanisation. The decrease -in disparity in the manufacturing factory sector--both in terms of value added as well as in employment--would lead to more uniform urbanisation, since almost all factories would be found in urban areas. To the extent that 'accory - 28 - Table 7(3): SOME SELECTED INDICATORS OF ECONOMIC DEVELOPMENT3 % Male agricultural % Total factory State employment1 employment2 1961 1971 1981 1961 1971 1981 1 2 3 4 5 6 7 Andhra Pradesh 67 69 6.3 1.2 1.5 2.3 Bihar 78 82 78 1.0 1.6 1.8 Gujarat 63 65 57 4.3 5.3 5.8 . Haryana - .67 50 - 3.5 4.5 Karnataka 69 59 52 1.6 2.8 3.6 . Madhya Pradesh 77 78 72 1.0 1.5 1.7 Maharashtra 62 60 52 4.4 5.7 4.9 Orissa 77 80 74 0.5 1.0 1.0 Punjab 613 65 60 1.93 3.0 3.9 Rajasthan 74 75 67 0.6 1.1 1.8 Tamil Nadu 59 61 64, 2.3 3.1 3.3 Uttar Pradesh 73 77 74 1.2 1.5 1.7 West Bengal 58 61 56 6.4 6.8 5.7 Mean 66.5 68.8 62.7 2.25 2.99 3.31 Coefficient of variation 0.16 0.12 0.16 0.76 0.59 0.47 Notes: 1. Total male agricultural employment as percentage of total male employment. 2. Total factory employment as percentage of total employment. 3. For undivided Punjab. V-RA/T-007U-3/1. 84 - 29 - employment his large multipliers for service, small manufacturing, and government, its effect on urbanization would be reinforced. The decrease in the proportion employed in agriculture (a very significant decrease by Indi,an historical standards), accompanied by stagnation in agricultural productivity, suggests the "push" factor operating in the poorer states. This is particularly so since, despite the decrease in agricultural employment (proportionally), it has been pointed out that foodgrain output per male agricultural laborer has declined or remained constant in the poorer states. Although there is a significant increase in the labor employed in factories in these states, they do not appear large enough to cause the decrease in agricultural employment. The composite picture that emerges then is as follows. There has been a perceptible increase in interstate inequality as measured by-per capita state domestic product. This has been caused largely by the stagnation in agricultural productivity in the whole Eastern, Central-Eastern and South Eastern Regions. There have been notable increases in factory production in all states and particularly so, in relative terms, in the hitherto backward states. This movement has somewhat mitigated the increasing interstate inequality 'Out the favorable changes have not been large enough to counteract the adverse changes in agriculture. The combination of increased manufacturing production and agricultural stagnation in the poorer states (during the last decade) is likely to have caused the significant acceleration in urbanisation over this period. The data suggests that the absorptive power of agriculture has probably been stretched to its limit in the sense that minimum per capita subsistence limits may have been reached. In the earlier periods, there was considerable scope for an exteojive increase in the area - 30 - under cultivation. I,t was, therefore, possible to accommodate the increase in population and, consequently, the resultant growth in agricultural labor, for cultivation of hitheVto uncultivated land. These possibilities havenow been exhausted. Increases in agricultural productions can now only take place by productivity changes, j.e., technological change. This seems very clear for the whole rice regior,: if there is labor using productivity change there will be an initial increase in labor absorption. The indications from Punjab and Haryana are that after the initial labor absorption occurs, after income increases have taken place, and agriculture becomes more technologically- intensive, it is likely that urbanisation would again accelerate. The indications therefore are that both agricultural stagnation, as well as growth, are likely to contribute toward further urbanisation in the country in the foreseeable future. III. The Determinants of Urbanisation: Some Explorations 3.1 The Level of Urbanisation As mentioned in the introduction, the relationship between urbanisation and per capita income has been a long-established one for international cross-section and time series data. The hypothesis is that the shap of the relationship is logistic as expressed by an equation of the type I+b -clog y exp where U is the percent of population urbanised and Y is per capita GNP. Good - 31 - fits are generally found for such an equation. The most recent estimation of this type has been by Mills and Becker (1982) who have estimated a variant of (1). They have added a variable measuring the level of agricultural activity--the share of the labor force in agriculture. An alter-native, to include a measure of manufacturing activity, was found to be a better predictor of urbanisation in such a formulation. They estimated a logistic form of the equation along with a quadratic one of the form: U = a +aY+aY + aA + a t(2) o 1 2 3 -4(2 Where A is the share in agricultural employment and t is a time trend. If a, > 0 and a. < 0 this equation predicts U to start declining after some maximum for a constant A. They found both these formulations to perform quite well with a large international data set spanning the 1960 to 1980 period. The level of explanation achieved (R2) was about 0.77, i.e., as simple a formulation as this was found to account for 77% of the variance in urbanisation levels in the world. I utilize a similar framework for attempting to explain the variance in urbanisation between states in India. Consistent data are now largely available, as shown in-the last section for 1961, 1971 and 1981. Except for population data, most of the other data is not available yet for 1981, e.g., for SDP, agricultural production, etc. In each case, the nearest available year has been used. Equations such as (1) and (2) are justified using the same kind of theoretical bases given in Section 1. They are neither - 32 - structural nor reduced form equations. Such variable as per capita income urbanisation levels, share of agricultural employment, are all endogenous, resulting from the kind of interactions sketched earlier. But the confirmation of the associations predicted do give some support to the theory. I estimated the following three types of equations: U = a0 + a1Y2 + a2 + a + at (3) 2 or U = b0 + bY + b2Y + b3M + b4t (4) or U = c0 + cY + c2 2 + c3A + cM + c5C (5) where U = level of. urbanisation in the State Y = per capita state domestic product A = an index of agricultural activity M = an index of manufacturing activity- and t = time trend. The exact definition of the variables used is given in Table 8. A measure of state domestic product per capita (SDP) ac constant prices is used for Y; the share of male agricultural employment in total male employment (%LAGRM) and agricultural laborer productivity (FGNS) expressed as total foodgrain production per agricultural labor are used as alternative measures for A. The share of factory employment, value added per capita in the factory sector and per capita power consumption are the alternative measures for M. - 33 - Table 8: DEFINITION OF VARIABLES Dependent variables 1. URB Level of urbanisation (total urban population as a proportion of total population). 2. URBPOP Total urban population (in thousands of persons) 3. CLASS I Total population of class I cities (in thousands of persons) i.e., cities ,under 100,000 population. 4. NONCLASS I Total population of nonclass I cities (in thousands of persons) i.e., cities under 100,000 population. IndeDendent variables A. Income variables 1. SDP State domestic product per capita (in 1970-71 constant prices). 1978-79 data used for 1981. 2. SDP2 (SDP)2 B. Employment variables (all measured in thousands of persons). 3. LFACT Total employment in factories (all factories using power and employing more than 10 people and others employing more than 20 people). 1979 data used for 1981 4. LAGRM Total male employment in agriculture. 5. LmNHm Total male employment in nonhousehold industries. (no 1981 data). 6. LFACMX Total employment in factories in "net export manufacturing activities". - LFACT. LFACT - minI -(total employment.) 1 total employment. VRA/T-008U/1. 84 Table 8 (contd) (Table 8 cont'd) - 33A - 7. LGOCT Total employment in Central government. (This does not include "productive" enterprises, i.e., public enterprises). 1980 data used for 1981. 8. .LGOST Total employment in State government. 1980 data used for 1981. 9. % LFACT LFACT as percentage of total male employment in the State. 10. % LAGRM LAGRM as percentage of total male employment in the State. C. Other Indicators of Economic Develooments 11. VAFAC Value added per capita in factory sector in the State (current Rs. per person). 1977-78 data used for 1981. 12. KWH Total power consumed per capita in the State (KWH/person). 1978-79 data used for 1981. 13. FGNS Total foodgrains output per male agricultural labor in the State (tons per person). total food grains output LAGRM D.Other 14. t Time variable. 1961=1, 1971=2, 1981=3. 1/ This is defined as the excess employment in manufacturing over that required for domestic services. This mrinimum is approximated by the lowest proportion found among the sample of states. Note: See Appendix for sources of data. VRA/T-908U/1. 84 - 34 - Table 9(a) gives the regression results estimated by ordinary least squares. Variables are added in a stepwise fashion to test for the stability of individual variables. The coefficients for SDP and (SDP)2 are highly significant and have the predicted signs. The time trend does not add to the level. of explanation. This suggests that there is no built-in trend towards urbanisation over time. It argues against those who claim that much of the urbanisation taking place is merely because of population pressure and is unrelated to economic growth. The consistently insignificant coefficient of t suggests that with stagnating income one would not expect increasing urbanisation over time. Regressions (3) and (4) suffer somewhat from collinearity problems: %LFACT has a correlation coefficient of about 0.66 with SDP and of about -0.7 with LAGRM. Nonetheless, the coefficient of %LFACT (1.06) in regression (3) is significant at the 5% level. Ceteris paribus, i.e., keeping income constant, a one percent increase in the share of factory employment would lead to a one percent increase in the level of urbanisa- tion. Me coefficient of LAGRM has the expected negative sign--an increase in the share of agricultural employme.nt would lead to a small decrease in urtanisation. The output variables for M perform somewhat better. The coefficient of VA.FAC (the value added per capita in the factory sector) is highly significant (regressions 5 and 7) and adds considerably to the magnitude of variance explained, despite its correlation with SDP (0.65). The other proxy for M, power consumption per capita (KWR), has coefficients which are also highly significant and stable (regressions 6,8,9). Finally, the last variable utilised, FGNS (agricultural labor productivity), has a negative sign, but is not found to be statistically significant. It does indicate that, ceteris - 35 - Table 9: RBANISATION AND ECONOMIC DEVELOPMENT IN INDIAN STATES 1961-1981(1) Dependent variable: URB (Level of urbanisation) Independent variables 1 2 3 4 5 Constant -24.30 -25.11 -17.23 -7.80 -13.92 (1.69) (1.72) (0.97) (0.41) (0.98) SDP 0.108 0.107 0.081 0.079 0.083 (5.43) (5.34) (3.29) (3.20) (4.25) SDP2 -0.000055 -0.000055 -0.000043 -0.000040 -0.000044 (4.29) (4.27) (2.86) (2.77) (3.73) %LFACT 1.06 0.672 , (1.71) (1.07) %LAGRM -0.112 (1.07) VAFAC 0.049 (3.16) WTH FGNS t 0.72 0.65 0.67 -1.14 (0.77) (0.72) (0.74) (1.11) R 2 0.631 0.637 0.664 0.675 0.716 22 0.612 0.608 0.627 0.629 0.684 (Adjusted R2 Notes: 1. All regressions run with 41 observations. 13 states in 1961 and 14 each in 1971 and 1981. (Punjab and Haryana combined in 1961). 2. t statistics in parenthesis. VRhA/T-009U-1/1.84 - 36 - Table 9(b): URBANISATION AND ECONOMIC DEVELOPMENT IN INDLA STATES 1961-1981(2) Dependent variable: URB (Level of Urbanisation) Independent variables 6 7 8 9 10 Constant -19.85 -4.28 -6.47 -5.06 -3.70 (1.49) (0.27) (0.42) (0.21) SDP 0.103 0.075 0.090 0.079 0.069 (5.80) (3.64) (4.73) (3.71) (3.16) SDP2 -0.000065 -0.000040 -0.000058 -0.000043 -0.000031 (5.61) (3.24) (4.68) (2.50) (1.85) %LFACT %LAGRM -0.098 -0.125 0.087 -0.075 (1.14) (1.55) (1.01) (0.83) VAFAC 0.044 0.031 (2.74) (1.45) KWH 0.093 0.087 0.071 (3.47) (3.28) (2.37) FONS -2.34 -2.17 (1.19) (0.89) t -2.72 -0.96 -2.52 -1.97 -0.54 (2.12) (0.93) (1.99) (1.47) (0.48) R2 0.728 0.726 0.746 0.756 0.732 R2 0.698 0.687 0.709 0.713 0.685 Notes: See Table 9(a). VRA/T-009U-2/1, 84 - 37 - paribus, higher agricultural labor productivity would tend to decrease urbanisation. It is of interest to analyse the implications of these regressions a bit further. The R2S obtained are gratifyingly high--in the range of 0.75. These equations succeed in explaining as much as 75% of the variance in urbanisation between states. Furthermore, if the definition of urban population is taken to include only towns above'20,000 population, and the dependent variable ERB is changed accordingly, it is found that the R2 increases to about 0.82. This is presumably because, as was suggested in the last section, the definitional problems at the lower end of the settlement scale are thereby eliminated. Consequently, the variation between states and over time due to inconsistent definition is eliminated. One interesting feature of these estimates is that the-maximum positive effect of income on urbanisation is reached at relatively low income levels and is in fact within the sample. The maximum effect is reached in the range of about Rs.750 to 950 per capita income (1970-71 prices) in different regressions. In international data, this turning point naturally comes at much higher incomes. The reason is essentially that this is a ceteris paribus result: if the share of agricultural employment or factory employment remains constant, this is actually what would happen. This can be illustrated with reference to Punjab. Punjab now has the highest SDP, but its urbanisation level of 27.7 percent is only the fifth among States. 'Increases in income without accompanying declines in agricultural employment or increases in manufacturing employment tend to keep the level of urbanisation down. The interconnected simultaneous processes sketched in Section 1 are therefore being captured well in these equations. It is admitted that it is somewhat tautological to say - 38 - that without a decline in agricultural employment urbanisation will not increase. But when does this happen? This occurs in two kinds of situations. First, when income increases only in the urban sector, that is, in a highly daulistic economy, without matching changes in the subsistence agricultural sector. Second, when income increases in the agricultural sector because of labor-using technological changes-perhaps as in Punjab.. But both scenarios have their limits. When urban/rural income disparities become too great, migration is induced, but the employment problem remains unless labor utilizing urban technologies are found. Similarly, as may be happening in the Punjab, as agricultural incomes increase further and the demand for urban goods increases, increased demand for urban labor would tend to take labor from rural areas which would in turn have the effect of quicker introduction of labor-saving technology in agriculture. The results from the equations estimated would be consistent with these kinds of changes in the economy. Further insight is gained by comnparing the fitted values with the actual levels of urbanisation. To the extent that the estimations reflect some "norm", the errors in prediction reflect "under" or "over"- urbanisation. Consider the case of the poorer states first, which have been found to urbanise relatively rapidly in the past decade. Bihar and Orissa are found to be consistently "under"-urbanised, i.e., they are expected to be more urbanised than they actually are. But this gap has been consistently reduced over the last 20 years, particularly in the last decade. This presumably reflects the highly capital-intensive nature of the industries in those states--given their income levels, albeit low, higher levels of urbanisation would have been expected. To the extent that the nature of industry has not changed over this period, and that this has been accompanied by agricultural - 39 - stagnation, it would be valid to infer that this catch-up phenomenon has been of the push migration variety. It was documented in Mohan and Pant (1982) that this increased urbanisation rate in these states is largely accounted or by very high growth in urban population in the regions which have received the most industry. These, however, are still quite concentrated--in the Southern Region of Bihar, for example--so that the expected level of urbanisation is still higher than the actual, even after accounting for the rapid'urbanisation of the past decade. Among the other poor states, U..P. is close to the predicted value in all years when small towns with less than 20,000 population are excluded. But if the total urban population is used, the 1981 actual value is much higher than predicted, implying "over"-urbanisation. This suggests that the acceleration in urbanisation in U.P. is essentially due to definitional changes, as was found in the last section. Otherwise, U.P. gives no cause for surprise. Madhya Pradesh was close to the predicted value in 1961 but has "over"-urbanised considerably over the past two decades. This is true even if small towns are excluded, and is therefore not a definitional problem. Given that industrial investment there has been concentrated in the Eastern region of the state, this result again suggests 1oush" from rural areas of Madhya Pradesh also. Among the other states, Kerala is consistently "lunder"-urbanised-- this may be because of the definitional problems alluded to earlier. Haryana is also "under"-urbanised, though less so in 1981--this is consistent with the productivity gains made there in agriculture and which have been labor using. Urbanisation is, however, catching up, because of continuing income increases. The rest of the states--Andhra Pradesh, Gujarat, Karnataka, - 40- Maharashtra, Rajasthan and West Bengal may be termed "normal" in that their predicted and actual values are quite close in all three years. 3.2 Urban Population: Employment Multipliers Having established that the pattern of interstate variation in overall urbanisation found in India is quite consistent with a simple formulaton of the relationship between urbanisaEion and economic development, it is now iv-essary to elaborate this further by attempting to find the determinants of the absolute levels of urban population found. The methodology adopted is essentially that of Mera (1975), who has suggested a "layer" theory of urbanisation, as outlined in Section I. The idea is to relate the magnitude of urban population to categories of basic employment, i.e.., U dP + d2S + d3T (6) where U1 is urban population, and P, S and T represent basic employment in the primary, secondary and tertiary sectors respectively. In this formulation,dj, d2 and d3 are the population multipliers resulting from basic employment. This formulation brings into focus a neglected aspect of urbanisation until now: the role of the tertiary or service sector in the generation of urban employment and population. The justification for this neglect is essentially the idea that all tertiary sector employment is derived from the demands of agricultural and manufacturing activities. This assumption is now relaxed and the tertiary sector can now have at least some component of "basic" employment. -41 Equation (6) merely says that each type of basic employment adds a layer to urban population, which are additive. Interactions between P, S and T are ignored, by definition, since they constitute "'basic" or "export-based" or "exogenous" employment. In the primary sector all male agricultural employment is regarded as basic employment. Two variables are used for each of the secondary aid tertiary sectors. LFACT is total employment in the factory sector. This may be regarded as exogenous, since this is employment in only the larger factories, defined as those employing more than 10 employees, if using power, and more than 20, if not using power. These cut- off points are, in reality, quite low: it would be more correct to use higher cut-off points, on the argument that it is the larger factories which exhibit economies of scale and are therefore "more exogenous". Hera suggests construction of another variable to get around this problem. The objective is to purge LFACT of that factory employment which is not basic, i.e., that which is derived. If it is assumed that the State which has the lowest proportion of LFACT has only derived factory employment, this proportion would be necessary in all the other States for local consumption purposes. The excess over this proportion then comprises "basic" or "export" employment (LFACMX) in all other States. For the tertiary sector, government employment is regarded as basic and the two measures used are State Government employees and Central Government employees. Three different measures of urban population are used as the dependent variable in order to investigate the functions of different sizes of cities. First is URBPOP, the total urban population; second is CLASS-I which is the population of cities over 100,000 only; and the third is NONCLASS-I, which is merely (URBPOP - CLASS-I), i.e., the total population of small towns. - 42 - As was suggested in Section I, it is expected that the size and distribution of urban population in small towns is likely to be determined by agricultural demand (or agricultural employment), while that of large cities is expected to be uncorrelated with agriculture. Table 10(a) gives the regressions for URBPOP. Once again the different variables are first introduced separately to test for collinearity effects that may arise when they are combined together in one regression. Agricultural employment is common to all the regressions reported. The coefficient of LAGRM is quite unstable over the 7 regressions, suggesting high correlation (0.72) with State Government employment (LGOST). Its correlation with Central Government employment (LGOCT) is also high (0.66). This suggests the derived nature of government employment. Indeed, a comparison of regressions (1) and (4) shows that the addition of LGOCT adds nothing to the level of explanation provided by just agricultural and factory employment. Indeed, the correlation of government employment, both central and state, is even more highly correlated with LFACT. Note that in equations where there is no LGOST where the coefficient of LAGRM is statistically significant, it is also quite stable at about 0.4. This means that about 5 agricultural jobs generate about 2 heads of urban population. Using a participation rate of about 35-36%, this means that it takes about 7 agricultural jobs to generate I urban job. The coefficient of LFACT varies between about 8 and 12 and that for LFACMX between 6 and 10. Accounting for the correlation with government employment, one can say that every basic manufacturing job induces an increase in urban population of about 8 to 10 or that it induces 3 to 4 other jobs. It is interesting to see that, contrary to expectation, cen-tral government employment appears to be almost totally derived: its coefficient is -43 Table 10(a): THE DETERMINANTS OF URBAN POPULATION IN INDIAN STATES Dependent variable: URBPOP Independent variables 1 2 3 4 5 6 7 Constant -1687 -924 1253 -1701 -409 -2812 -935 (1.72) (0.66) (0.84) (0.90) (0.22) (1.42) (0.52) LAGRM 0.414 0.046 -0.166 0.425 0.124 0.440 0.127 (6.26) (0.42) (1.61) (3.68) (1.15) (3.03) (1.11) LFACT 12.16 12.40 7.50 (11.43) (4.90) (4.33) LFACM%X 6.49 9.63 (0.90) (4.32) LGOCT 29.70 -0.74 (7.97) (0.11) .LGOST 16.55 7.78 (10.24) (3.21) t 1084 1698 -567 1073 228 1852 452 (3.07) (3.81) (1.26) (2.88) (0.55) (4.74) (0.98) R2 0.881 0.801 0.859 0.881 0.907 0.854 0.902 R2 0.871 0.785 0.848 0.868 0.897 0.838 0.891 (Adjusted R2 Notes: 1. No. of observations in-all regressions. 41 13 for 1961 (Punjab and Haryana taken as one State in 1961) 14 for 1971; 14 for 1981 2. t statistics in parenthesis. - 44- unstable. Government employment continues to be significant when included in an equation with agricultural, as well as factory, employment. The population multiplier for state government employment may be as high as that for LFACT, that is, between about 8 and 10. Note in Table 10a that regression (1) is not improved on significantly by the addition of government employment. It would then be reasonable to conclude that it is essentially agricultural and factory employment that is basic, while government employment is derived from these activities. Table 10b gives the same results for CLASS-1 cities only, i.e., the determinants of population in large cities. The first thing to note is that the significance of the coefficient of agricultural employment declines, as does its magnitude. It is now only about 0.15. This means that it takes about 6-7 agricultural jobs to induce an increase in large city population of 1, or that it takes about 15-20 agricultural jobs to create the demand for 1 job in large cities. The coefficients for LFACT and LAFCIX are quite stable, large and significant, as expected. Every factory job *creates 2 to 3 jobs in large cities. Note also that the coefficient of central government employment (LGOCT) is significant at the 5% level in regression (2), and of sizeable magnitude, quite similar to that of LFACT. This suggests that central government jobs are largely located in large cities and they do indeed have a multiplier effect. The coefficients of LGOST are of about the same magnitude as they were in the URBPOP regressions. Regressions (5) and (6) are for 1961 and 1971 data only and use yet another measure of manufacturing employment. This is LMNHM, male employment in the non-household industry. This includes all of manufacturing except the household industry, and is therefore not an adequate representation of basic manufacturing employment--ics coefficients - 45 - Table 10(b): THE DETERMINANTS OF URBAN POPULATION INDIAN STATES Dependent variable: CLASS I Independent variables 1 2 3 4 5 6 Constant -2613 -3439 -1569 -2088 -1135 -648 (2.12) (2.72) (1.43) (0.52) (1.17) (0.61) LAGRM 0.118 0.146 -0.079 -0.60 0.075 -0.072 (1.57)'- (1.57) (1.24) (0.92) (1.32) (1.14) LFACT 8.31 5.64 .(5.02) (5.54) LFACMX 7.14 5.34 (3.99) (5.62) LMNHM1 4.40 3.36 (5.17) (3.72) LGOCT' 4.20 8.05 2.95 (0.94) (1.78) (0.82) LGOST 6.82 9.63 5.75 (4.79) (4.32) (3.20) t 1030 968 217 443 315 -314 (4.22) (4.13) (0.89) (1.69) (1.10) (1.06) R2 0.900 0.898 0.938 0.-939 0.924 0.948 R2 0.889 0.890 0.931 0.932 0.910 0.938 No. of observations 41 41 41 41 27 27 Notes: 1. Data not available yet from 1981 Census. 2. See Table 10(a). VRA/T-0010U-2/1. 84 - 46 - being smaller than those for factory employment-are consistent with this idea. Table 10(c) reports similar estimates for small towns. The first feature to note is the consistently significant and stable coefficient of male agricultural employment (LAGRI). About 9 to 10 agricultural jobs give rise to 1 job in a small town. This suggests that in the absence of any factory employment, the level of urbanisation in an exclusively 'agricultural economy should be in the region of about 10 percent. This is not very far from the lowest levels that are observed internationally, as well as in India's most backward regions. The coefficients of LFACT and LFACMX are seen to decline in size and in significance level. It is noteworthy, however, that the most *inc'lusive measure of manufacturing employment, LMHNHMT, essentiaily retains its magnitude and significance. It would seem that.central government employment is not important in small and medium towns. The negative sign implies that most central government employment goes to large cities and perhaps attracts population out of small and medium towns. However, this argument cannot be taken too far, since the coefficient is not statistically significant. State government employment continues to be significant but highly correlated with agricultural employment, suggesting that state government employment is essential servicing agriculture. What emerges from these results? First, the level of explanation is surprisingly high. About 90 percent: of the variance in urban population, as well as that of large cities, is accounted for by these formulations. The variance of small town population is not as well explained but still quite respectable, with R2 of about 0.65. That large cities are not intimately connected with their hinterlands is supported by these results. Factory - 47 - Table 10(c): THE DETERMINANTS OF URBAN POPULATION IN INDIAN STATES Dependent variable: NONCLASS I Independent All years 1961, 1971, 1981 1961 and 1971 onl variables 1 2 3 4 5 6 Constant 912 627 1160 1153 859 1672 (0.81) (0.56) (0.90) (0.94) (0.80) (1.15) LAGRM 0.306 0.294 0.203 0.187 0.255 -0.011 (4.45) (3.58) (2.73) (2.41) (4.05) (0.13) LFACT 4.09 1.87 (2.71) (1.57) LF ACMX 2.87 1.26 (1.82) (1.12) .MNHM1 4.35 0.92 (4.65) (1.13) LGOCT -4.95 -1.56 -9.66 (1.20) (0.38) (2.45) LGOST 0.96 1.18 4.67 (0.58) (1.20) (1.95) t 42.4 293 10.8 8.96 -49.3 -427 (0.19) (1.32) (0.04) (0.03) (0.17 (1.04) R2 0.672 0.638 0.661 0.650 0.763 0.743 R2 0.635 0.598 0.624 0.612 0.720 0.696 No. of observations 41 41 41 41 27 27 Notes: 1. Data not available yet from 1981 Census. - 2. See Table 10(a). - 48 - employment has a multiplier of about 3 to 4. Government employment is largely derived from the demands of the primary and secondary sectors: it does not appear to autonomous or exogenous. In trying .to explain government employment, it was regressed against agricultural and manufacturing employment (regressions not reported here). About 90 percent of the variance is explained. About 75 agricultural jobs create I central government job, while only about 25 agricultural jobs create I state government job. Similarly, 3 factory jobs create 1 central governmient job, but create 2 state government ones. Small towns are more organically connected with agriculture, and the employment within them is essentially for the service of agriculture. Government employment has been found to be almost totally derived: it is clearly not autonomous. In comparing the actual urban, large city and small town populations with predicted ones, it is found that Orissa was seriously lacking in adequate small town population, as was Maharashtra. Bihar and Madhya Pradesh are found to lack adequate large city population. These results are suggestive of a methodology that could be used for regional investigations in order to determine whether the lack of different sized towns is warranted or not. IV. Urbanisation in Indian States: An Interpretation The statistical results found in the simple estimations attempted have been found to be surprisingly good and consistent with the theory offered in Section I. The pattern of urbanisation in India is clearly not chaotic: in most states, it is much as would be expected from theory. This is somewhat surprising since the regularity of the pattern is not obvious ac first glance. If the kind of simple equations used are regarded as good forecasting equations, such a system of equations could. even be used for forecasting and planning purposes. The prediction of inter-national or inter-regional growth rates of urbanisation has never been regarded as easy. An enlightened use of these estimation procedures could be quite useful. As predicted, there is a tight relationship betweeen levels of urbanisation and economic development. The surprising feature of the estimations is that the peak effect of income on urbanisation is found to be at quite low income levels and is in fact found within the sample of Indian States. This result is unexpected since the use of international data would suggest this turning point to occur at much higher income levels. The only interpretation that can be offered here is that the effect captured is a ceteris paribus effect which is the pattern of the effect of income on urbanisation when the structure of employment does not change appreciably. This happens when overall income increases because of industrial growth, unaccompanied by matching changes in the agricultural sector, i.e., the familiar dualistic pattern. This may be the case in the more industrialised States in the country, namely, Maharashtra, Tamil Nadu, West Bengal and Gujarat. All have experienced a significant deceleration in their rates of urbanisation, except for Gujarat, and all have experienced stagnating agricultural productivity. Their historically high industrialisation and urbanisatiot levels were clearly not due to intrinsic forces of urbanisation and economic development, but rather because of the demands of the colonial system. More light is shed on this issue by the effects of basic employment on urban population. It is of great interest to see that agricultural - 50 - employment essentially creates demand for services and industry in only small and medium-sized towns. The population of large cities has little relationship with agricultural activity, except in an indirect fashion through general income growth, which creates demand for urban goods. Hence, in the situation of countrywide agricultural stagnation, as was the Indian experience in the 50-100 years before independence, the small amount of urbanisation that took place had to be "exogenous", based on colonial trade patterns. The great policy relevance of these findings is that the growth of small and medium- sized towns is likely to be brought about by agricultural growth in the backward regions rather than by industrial dispersal. The highly significant effect of factory employment on the generation of population in large cities would suggest a policy of industrial dispersal to existing large cities, which may now be somewhat industrially backward. Or, alternatively, it suggests the creation of new centers which will be of substantial size. There should be a dispersal of concentrations rather than a dispersal of concentration, per se. The idea that economies of scale that exist in manufacturing cause the formation of concentrations is supported by the large population and employment multipiers found for factory employment. One of the key drawbacks in the work reported in this paper is the lack of a good indicator of "basic" tertiary sector employment. It is of great significance that government employment can largely be explained by basic agricultural and factory employment. The size of government is then not as* exogenous as is often thought: and is hence not easy to trim. The consistent story that emerges from the estimations in this paper is then described as follows. Increases in income do indeed lead to higher urbanisation levels, This is because of the diversification of activity that - 51 - usually accompanies growth in i"comes (because of Engel demand effects, etc). If this diversification does not take place, the urbanisation accompanying it would be slower than otherwise. Many of the non-agricultural activities exhibit economies of scale which determine the sizes of settlements that follow. The distribution between large, medium and small towns is therefore determined by the distribution of various econcmic activities. The number and distribution of small towns is particularly linked tightly with the level and distribution of agricultural activity. The hierarchy of economies of scale found in the manufacture of different products, along with the agglomeration economies as revealed by manufacturing employment multipliers, presumably, then determines the hierarchy of the size of settlements. Layers of urbanisation are cumulatively added on by the different basic activities. This interpretation attempts to link the overall theory of urbanisation and economic development, as captured in a structural model of the kind sketched in Sectio.a I, with the "central place" theories concerning the distribution of settlements and the "economic base" theories for the size of a settlement. These explorations suggest that spatial sub-models can be grafted on to such countrywide sectoral models. The total agricultural employment, for example, as determined by such a model, could be used, in a sub-model, to generate one layer of urbanisation--the small and medium-sized towns. Total manufacturing employment, similarly, could be disaggregated and then used to generate the larger order settlements--and so on. It is, perhaps, appropriate to conclude with a key policy implication of the findings in this paper. It is the growth of agriculture in the backward regions, and specifically the whole Eastern region, which needs the highest attention. This essentially means that technological innovations in - 52 - rice and dry area crops need the greatest research inputs. This agricultural stagnation in the Eastern Region is largely responsible for the increasing regional disparity in the country. The redressal of this disparity in agriculture is also, incidentally, the best policy for balanced urban growth. The focus on regional dispersal must be more on agricultural technology and less on manufacturing. - 53 - APPENDIX I Table A.1: URBAN POPULATION OF STATES 1/ IN INDIA 1951-1981 (In thousands) Population of towns above State Population of All Towns 20 OOO(Class I, II, III) 1951 1961 1971 . 1981 ,1951 1961 1971 1981 Andhra Pradesh 5419 6275 8403 12458 3358 4732 6942 11312 Bihar 2624 3914 5634 8699 1898 3023 4529 7634 Gujarat 4429 5317 7497 10556 2978 4212 6009 9060 Haryana 968 1308 1773 2822 617 950 1394 2321 Karnataka 4452 5466 7122 10711 2593 3669 5319 8870 Madhya Pradesh 3135 4627 6785 10589 2008 3141 5097 8170 Maharashtra 9201 11163' 15711 21966 6623 9398 13961 20225 Orissa 594 1110 1845 3.106 303 618 1317 2367 Punjab 1990 2568 3216 4620 1329 1940 2510 3742 Rajasthan 2955 3281 4544 7140 1646 2173 3321 5611 Tamil Nadu 7334 8991 12465 15928 3916 7090 10846 14446 Uttar Pradesh 8625 9480 12389 19973 5928 7853 10479 15288 West Bengel 6270 8541 10967 14433 5640 7793 10140 13761 India 2 61629 77566 106902 156192 43169 61418 89607 134898 Notes: 1. Including all states with total population greater thaa 10 million in 1971 but excluding Kerala and Assam. 2. Including all states except Assam and Jammu & Kashmir. Source: Census of India 1981-Provisionao Population Totals-Series I-Paper 2 of 1981. -54- APPENDIX II Sources of Data Variable name Source of Data 1. URBPOP ) 2. CLASSI ) 3. NONCLASSI ) Census of India 1961, 1971, 1981 4. LAGRM ) 5. LMNM4 ) 6. s.D.?. ) Centre for Monitoring the Indian Economy: Basic Statistics Relating to the Indian 7. VAFAC ) Economy, Vol II, 1981 8. FGNS) 9. LGOCT ) Government of India: Ministry of Labour: Directorate General of Employment and 10. LGOST ) Training Employment Review (Annual) Various Issues. 11. LFACT ) Government of India: Statistical Abstract ) Annual). Various Issues. 12. KWh ) - 55 - BIBLIOGRAPHY Bose, Ashish, D. B. Gupta and G. Raychaudhari Population SEatistics in India, New Delhi, Vikas Publications, 1977. Census of India 1971, India, Series, Report of the Expert Committee on Population Projections - Paper I of 1979, New Delhi, 1979. Census of India 1971, India, Series I, Part II A(i), General Population Tables, New Delhi, 1975. Census of India 1981, India, Series I Provisional Population Totals, Paper 2 of 1981, Paper 3 of 1981, New Delhi, 1981. Central Statistical Organisation, Statistical Abstract, New Delhi, (Annual) Various Issues. Centre for Monitoring the Indian Economy, Basic Statistics Relatihg to the Indian Economy, Volume II, States, Bombay, 1981. Gilbert, Alan and Josef'Guglar, Cities, Poverty and Development: Urbanisation in the Third World, New York: Oxford University Press, 1982. Kundu, Amitabh and Moonis Raza, Indian Economy: The Regional Dimension New Delhi: Spektrum Publications, 1982. Ministry of Labour: Directorate General of Employment and Training, Employment Review (Annual), New Delhi, Various Issues. Mera, Koichi, "A Multiple Layer Theory of National Urban Systems" in Swain and Mackinnon, 1975. Mills, Edwin S. and Charles Becker, "Urbanization and Economic Development" World. Bank, Mimeo. Mohan, Rakesh and Chandra Shekar Pant, "Morphology of Urbanization in India: Some Results from the 1981 Census" Economic and Political Weekly, Vol. 17, Nos. 38 and 39, September 18 and 25, 1982. Premi, M. K., D. B. Gupta and Amitabh Kundu, "The Concept of Urban Areas in the 1961-1971 Census" in Ashish Bose and Others, 1977. Sekhar, Udai, "Industrial Location Policy: The Indian Experience", World Bank Staff Working Papers No. 620, Washington D.C, 1984. Sinha, J. N., "1981 Census Economic Data: A Note" Economic and Political Weekly, Vol. 17, No. 6 pp. 195-203, February 6, 1982. Swain, H. and Ross D. Mackinnon (eds), Issues iH the Management of Urban System, Vienna, Austria, International Insticute of Applied Systems Analysis, 1973.