Project on India

Determinants of Urban Growth during Demographic and Mobility Transitions: Evidence from India, Mexico, and the US

LEIWEN JIANG BRIAN C. O’NEILL

UNDERSTANDING HOW and why urban populations grow over time is fundamental for a variety of scholarly and policy arenas including mi- gration, economic development, and environmental change. Improved understanding would also allow for better projections of the future course of urbanization and its consequences for society and the environment. In demographic terms, aggregate urban growth at the national level is deter- mined by a combination of natural increase (births minus deaths) within urban areas, migration between rural and urban areas or to/from other countries, and the reclassification of land (and population living on it) from rural to urban. Reclassification typically occurs through the expansion of urban boundaries, resulting in larger contiguous areas of urban settlement, or through the addition of distinct new urban areas as rural areas grow in population and economic size and density, contributing positively to urban growth (UN 2001). In some instances, however, reclassification can affect urban growth negatively because of contraction of existing urban boundaries or subtraction of existing urban settlements. Conceptually, re- classification is not entirely distinct from the other determinants of growth, because it typically occurs under conditions of increasing population and economic densities. Therefore both natural population growth and migra- tion may indirectly contribute to reclassification by increasing population density in areas that later are reclassified as urban.

A number of scholars have argued that the relative roles of these de- terminants of urban growth vary in a regular manner as countries undergo demographic and mobility transitions (Becker 2007; Skeldon 2008; Kelley and Williamson 1984; Ledent 1982). While demographic transition refers to the changes from high to low death and birth rates as a country becomes industrialized (Thompson 1929; Caldwell et al. 2006), mobility transition indicates that internal migrations display distinct patterns at different stages of demographic transition and urbanization (Zelinsky 1971).1 We review

POPULATION AND DEVELOPMENT REVIEW 44(2): 363–389 (JUNE 2018) 363

C© 2018 The Authors. Population and Development Review published by Wiley Periodicals, Inc This is an open access article under the terms of the Creative Commons Attribution-NonCommercial-NoDerivs License, which permits use and distribution in any medium, provided the original work is properly cited, the use is non-commercial and no modifications or adaptations are made.

364 DE T E R M I N A N T S O F UR B A N GR O W T H

these arguments and empirical findings and synthesize them into an illus- trative framework (Figure 1). Our focus is on explaining the growth of ur- ban population, not urbanization levels; explaining trends in urbanization (i.e., the percent of the population defined as urban) would lead to a differ- ent (but related) set of conclusions, given the dependence of urbanization on both urban and rural growth.

In the early or pre–demographic transition period, migration is thought to dominate natural increase in driving urban growth because the death rate in urban areas is very high—usually higher than the birth rate (Figure 1a, Stage I), leading to a negative natural increase rate (Dyson 2011; Bocquier and Costa 2015). Rural-urban migration, even though it is low because of the low level of economic development (Figure 1b), is the only significant factor driving urban growth (Dyson 2011; Figure 1c). During Stage II of the demographic transition, mortality decline with moderniza- tion leads to positive natural population growth. In the meantime, rural population growth generates socioeconomic and environmental pressures that increase migration to urban areas, making an important contribution to urban growth. Natural increase is generally expected to outpace the in- crease in migration during Stage II. During Stage III, rates of both natu- ral increase and rural–urban migration peak and start to decline. Natural growth is still expected to outweigh migration in its influence on urban growth (Figure 1c), partly because the changing relative sizes of rural and urban populations lessen net migration to urban areas in absolute terms. Moreover, urban–rural migration rates increase while rural–urban migra- tion rates decline (Figure 1b) because of narrowing rural/urban disparities, urban congestion, and counter-urbanization in many parts of the world. In Stage IV of the demographic and mobility transitions, both natural increase and rural-urban migration continue to decline, so that their relative contri- butions to urban growth can be expected to eventually stabilize (Figure 1c).

While the roles of natural and migratory changes in urban growth dur- ing demographic and mobility transitions (particularly in earlier stages) are well documented, reclassification may assume varying and more immedi- ate effects on urban and rural growth as a result of changes in urban def- initions or industrial policies and other political reasons (e.g. Kulcsar and Brown 2011). The role of reclassification, separate from the roles of migra- tion and natural increase, has not previously been conceptualized as occur- ring in a regular manner over the course of the demographic and mobility transitions, largely due to lack of proper data (Bocquier and Costa 2015). Most authors present the combined effects of reclassification and migration when studying the relative contribution of reclassification to urban growth (Preston 1979; UN 2001). We therefore review available empirical studies to determine the extent to which we can add (and then test) expectations about the role of reclassification in the conceptual model of determinants of urban growth.

FIGURE 1 Effects of natural increase, migration, and reclassification on urban growth

NOTE: CDR = crude death rate; CBR = crude birth rate; NRI = natural rate of increase; effects of natural increase and migration on urban growth by stages of the demographic and mobility transitions are based on Dyson (2009) and Zelinsky (1971). The shaded area of Figure 1c is the range of uncertainty of contri- bution from natural growth, the short dashes represent the range of uncertainty of contribution from net migration, the long dashes represent the range of uncertainty from reclassification.

366 DE T E R M I N A N T S O F UR B A N GR O W T H

A study of Sweden in the early stages of the demographic and mo- bility transitions reveals that reclassification was virtually non-existent in 1750–1820 and increased only slightly and contributed marginally to urban growth from 1820 to 1910 (Bocquier and Costa 2015). The same study finds a larger effect of reclassification on urban growth in Belgium in the later stage of the two transitions (1860–1960). Studies of effects of reclassifica- tion in more recent decades in industrialized countries do not exist. Studies of experience over the past few decades in some developing countries find generally increasing but widely varying trends in the effects of reclassifica- tion over the course of the transitions. They range from the positive effects that are much stronger than those due to migration in China in the mid- 1980s (Goldstein 1990) and in the Philippines in 1960–70 (Permia 1976), to a negative contribution in Vietnam in the late 1970s (Banister 1993).

A few studies attempt to measure the relative contributions of natural increase, rural-urban migration, and reclassification to urban growth. Most of these compare the urban populations from two censuses, account for the effects of observed mortality and fertility, and assign the remainder of the population change to the effect of migration and reclassification. A major limitation of this approach is that, given the lack of data on migration or reclassification, it is impossible to differentiate between their effects (Preston 1979; Todaro 1984; Chen et al. 1998; National Research Council 2003). In one of the most cited studies, Chen et al. (1998) decompose the changes in urban population in developing countries in the 1960s–1980s, making the simplifying assumptions that urban mortality rates are 25 per- cent lower than those in rural areas and that urban and rural populations have the same mortality age profiles. They find that among 55 countries with data (17 in Africa, 22 in Latin America, and 16 in Asia), the median contribution of natural increase to urban growth is about 60 percent, with the remaining 40 percent attributable to the combination of migration and reclassification. According to Chen et al., the rural out-migration rate combined with reclassification increased from 0.61 percent of the rural population per year in the 1960s to 1.14 percent in the 1980s in developing countries. This trend differed significantly among the major developing re- gions. Based on data for the same 55 countries, during the same period the rural out-migration rate combined with reclassification declined steadily in Africa, but gradually increased in Asia, while in Latin America and the Caribbean it increased first between the 1960s and 1970s but then declined. Such trends are consistent with what is known about the urbanization levels/growth rates and changing economic situations in these regions.

Some progress has occurred in isolating the separate contributions of rural-urban migration and reclassification to urban growth in countries. Analysis for India using data on natural increase and migration reveals that net urban migration alone accounted for 19 percent of observed ur- ban growth in the 1960s and 1970s (Kundu 2011). Indirect estimates using

LE I W E N JI A N G / BR I A N C. O’NE I L L 367

the census survival ratio method also show that newly designated towns (reclassification) accounted for around 20 percent of observed urban popu- lation growth during the 1960s–1980s and the proportion slightly increased over time (Economic Commission for Asia and the Pacific 1995). Using more recent data, Bhagat and Mohanty (2009) find that migration accounted for around 20 percent of observed urban growth, although rates of migration and natural increase rates declined in recent decades. Studies in China show that while more than 70 percent of observed urban growth in the 1980s and 1990s was driven by the combination of rural-urban migration and reclas- sification (Lucas 1998), close to 50 percent of the urban growth was the result of reclassification alone (Chan 1994; OECD 2009). These studies sug- gest that relative contributions to urban growth of migration versus reclassi- fication may differ significantly across countries and stages of urbanization, and that when urbanization is rapid (as in China in the 1980s and 1990s), reclassification can play an especially significant role.

While studies have increased our understanding of forces driving ur- ban growth, they usually examine one or two demographic determinants but do not simultaneously consider all three components: natural increase, migration, and reclassification. These studies usually do not consider the differences in age and sex profiles of fertility, mortality, and migration rates (Bocquier and Costa 2015) and assume uniform distribution of the rates over the intercensal period (UN 2001). Most studies focus on the experi- ences of either a single country or a group of developing countries and do not systematically investigate the changes in countries across a spectrum of development and urbanization levels. Further, a variety of methods is employed across studies, making it difficult to compare their results.

Here we use multistate demographic methods combined with census data to derive and compare all three demographic determinants of urban growth in a recent decade for India (1991–2001), Mexico (2000–2010), and the United States (2000–2010). We select these three countries not only be- cause they are large countries in three continents at different economic de- velopment levels and in different stages of urbanization and demographic transition, but also because they have adequate data for the analysis. This research aims to identify recent driving forces of urban growth in differ- ent parts of the world and contribute to understanding how and why the spatial distribution of the human population changes with socioeconomic development and demographic dynamics.

Methods and data

Method

Previous studies of the determinants of urban growth mostly use aggregate- level accounting equations that do not consider the effects of differing rural

368 DE T E R M I N A N T S O F UR B A N GR O W T H

and urban age and sex compositions. Most of the studies are not able to consider births to migrants in the destination region and therefore overesti- mate the effects of growth from migration and understate the contribution of natural increase (Visaria 1997). This limitation also affects estimates of the effects of reclassification (UN 2001). According to Rogers (1982), assess- ment using an aggregate-level projection model with assumed uniformly distributed rates over age “is of limited value for answering the question on whether it is natural increase or net migration that is the principal source of urban population growth”; therefore, “a more realistic model that allows the natural increase rate to change over time along with rate of net urban in-migration” is needed. This model should be able to disaggregate the ru- ral and urban population by age and sex and to specify the age and sex profiles of demographic events in rural and urban populations. Our study uses the National Center for Atmospheric Research (NCAR) multiregional population/urbanization model, which is based on multistate demographic methods that simultaneously project population change by age, sex, and ru- ral/urban residence (Rogers 1975, 1995). While a few demographers have used multistate models to calculate the relative contributions to urbaniza- tion of natural increase and migration, they have generally ignored the con- tribution of reclassification of areas from rural to urban (UN 2001).

A detailed description of the NCAR multiregional population/ urbanization model is contained in Jiang and O’Neill (2009); the basic ac- counting strategy of the model is given in the Appendix.

We use the model to project changes in the populations by age, sex, and rural/urban residence between the two censuses, given the known de- mographic rates and estimating the effects of unknown demographic events with reference to the census population numbers. Data on fertility and mor- tality differentiated by urban and rural residence allow us to identify the contribution of natural increase, and data on rural-urban migration allow us to estimate the migration contribution using the multiregional model. We calculate reclassification as the residual. Our multiregional model represents gross rural-urban and urban-rural migration rates, rather than a single net migration rate, allowing for more detailed analysis of the contribution of migration to urban growth.

Data

The US. We use the PUMS (Public Use Microdata Sample) data from the US censuses for 2000 (5 percent) and 2010 (10 percent) to derive populations by age, sex, and rural/urban status over the two time periods. We derive the rural/urban status of households (and therefore of the individuals living in them) for the two years according to their “Metropolitan status.” We define households as urban if they are in a metro area and as rural otherwise. We exclude households whose metropolitan status is “Not identifiable,” which

LE I W E N JI A N G / BR I A N C. O’NE I L L 369

results in a higher percentage of urban population than national urban- ization levels reported by the US Census Bureau. We use the age and sex structure of rural and urban populations derived from the census data for the two time points and scale them to match the total rural and urban pop- ulation according to the published US Census Bureau statistics.

We also use the census micro-samples to derive age-specific fertility rates for the rural and urban population. The 2010 census contains data on births over the last year, which we use to calculate the fertility rate. The derived national total fertility rate (TFR) of 2.0 births per woman is the same as reported by the American Community Survey (Figure 2), which suggests that births in the last year are a reliable source for representing rural and urban fertility. Unfortunately, the 2000 census has only data on the ages of the youngest and oldest own child. Using the age of own child and assuming that children aged less than 1 year represent new births in the last year, the calculated fertility rates are much too low, for instance a TFR of 1.62 (rather than 2.0) for the year 2010. However, the two approaches produce very similar age patterns of fertility (Figure 3). Figure 4 shows that the standardized cumulative fertility by age derived from the two methods

FIGURE 2 US TFR according to various sources, 1990–2013

NOTE: ACS = American Community Survey; CDC = Centers for Disease Control and Prevention; CPS = Current Population Survey.

370 DE T E R M I N A N T S O F UR B A N GR O W T H

FIGURE 3 US urban and rural age-specific fertility rates (ASFR), calculated from the 2010 census, using birth in last year and own child aged less than 1 year old

FIGURE 4 US standardized cumulative urban and rural age-specific fertility rates (ASFR), calculated from the 2010 census, using birth in last year and own child aged less than 1 year old

are almost identical. We conclude that the age-specific fertility rates of the rural and urban population derived by using age of own child from the 2000 census can be used to represent the age patterns of fertility in that year.

We use the patterns of age-specific fertility derived from the age of own child in the 2000 census and scale them up so that the national TFRs match the value reported by the US Centers for Disease Control and Prevention (CDC) for every year in the period 2000–2010 (Figure 5). We use TFRs

LE I W E N JI A N G / BR I A N C. O’NE I L L 371

from the CDC because the ACS only started from 2001, and the reported TFR from the Current Population Survey (CPS) is only calculated based on fertility of women aged up to 45 and is therefore systematically lower than other sources, which include fertility of women aged up to 49 (Figure 2). Our method results in TFRs for rural women of 2.42 and 2.12 for 2000 and 2010 respectively, compared to urban TFRs of 1.95 and 1.98. In general, the fertility level is higher in rural than in urban areas and the rural/urban difference fluctuates over time.

The single-year age- and sex-specific mortality rates for the US popu- lation in 2000 and 2010 are derived from the Human Mortality Database, which has rates only for the country as a whole, not by rural/urban res- idence. Therefore, we apply the national mortality rates to the rural and urban population for both time points. Considering the relatively small differences in mortality across rural and urban areas in the US and the less significant role of mortality in population growth in the US, we believe this treatment does not introduce a large bias in the natural increase of the urban population.

The US census reports the migration status of households in the past five years, and the 2010 census also reports migration status in the past year. The census indicates the metropolitan status of migrants’ previous residence, from which we derive the rural and urban status of the migrants before moving. Because the projection model used in the backcasting analysis takes a single-year step, we calculate the average annual age- and sex-specific rural-urban and urban-rural migration rates for all three time periods. Figure 6 shows that annual rural-urban migration rates in the US are higher than in many developing countries, even though the country is already highly urbanized.

FIGURE 5 US urban and rural age-specific fertility rates (ASFR), 2000

SOURCE: Authors’ calculations.

372 DE T E R M I N A N T S O F UR B A N GR O W T H

FIGURE 6 US annual rural-urban migration rates (percent/year), by age group, derived from the 2000 and 2010 census

SOURCE: Authors’ calculations.

LE I W E N JI A N G / BR I A N C. O’NE I L L 373

To produce our estimate of annual migration rates over the decade 2000–2010, we derive a linear trend in the migration rate for 2005–2010 such that its average over 2005–2010 and its final value in 2010 are consis- tent with the census data for 2000 and 2010. We then linearly interpolate from the year 2000 (assumed to have the mean 1995–2000 value) to the value for 2005 from our derived 2005–2010 trend. We end up with two linked linear trends for the two five-year periods over the decade of interest.

The age- and sex-specific international migration rates are obtained from the NCAR-CDM (Community Demographic Model) International Mi- gration Dataset (Nawrotzki and Jiang 2015). This dataset contains interna- tional migration rates by the age and sex of migrants calculated from data on international migrants of all countries from the UN Global Migration Database. Figure 7 displays international net immigration rates for the US by age and sex for the year 2000. We allocate international migrants to rural and urban areas according to current urbanization levels. In other words, we assume no direct contribution to urbanization from international migra- tion. Given that the effects on urban growth from international migration are small compared to the effects of internal migration, this approach does not have a significant effect on our conclusions.

India. Data for India are from the detailed tables of the Indian cen- suses of 1991 and 2001. While the census provides population by age, sex, and rural/urban status, it has serious age-heaping problems (Figure 8). We

FIGURE 7 Annual US international migration rate (percent/year), by age, 2000

SOURCE: Authors’ calculations.

FIGURE 8 India 1991 population by age, sex, and rural/urban before and after adjusting for age-heaping

LE I W E N JI A N G / BR I A N C. O’NE I L L 375

FIGURE 9 India urban and rural age-specific fertility rates (ASFR), 1991 and 2001

SOURCE: Authors’ calculations from Indian 2001 and 1991 census reports.

use the Whipple’s Index approach to smooth the age structure and adjust the rural and urban population to match national statistics on urbanization levels. Figure 9 shows the age-specific fertility rate of the rural and urban population derived from the two censuses. From the India statistics reports, we also obtained TFRs for both years. The rates reflect the substantial fertil- ity decline over the decade for all-India (3.64 in 1991, 3.10 in 2001), with the fertility reduction being more significant in urban areas (3.02 in 1991, 2.20 in 2001) than in rural areas (3.84 and 3.48).

The India censuses reports of 1991 and 2001 include tables of migra- tion for rural and urban populations over the past ten years, five years, and one year. We used the five-year data to calculate annual average

376 DE T E R M I N A N T S O F UR B A N GR O W T H

FIGURE 10 India annual rural-urban migration rates (percent/year), by age group, 1991–2001

SOURCE: Authors’ calculations.

rates over the 1996–2001 period, and the difference between the five- and ten-year data to calculate the annual average rate over the period 1991– 1996 (both shown in Figure 10). By assuming that these rates were con- stant over each five-year period, we specified migration for the decade. Al- though the census-reported rural-urban migration rate in India is generally

LE I W E N JI A N G / BR I A N C. O’NE I L L 377

low (much lower than in neighboring countries such as China and Vietnam and in the US), it increased considerably in the period 1991–2001. While rural-urban migration in India is expected continue, the migration tables from the 2011 census have not been released.

Indian age- and sex-specific mortality rates across rural and urban ar- eas are obtained from Sample Registration data.

Mexico. We use unharmonized micro-samples from the 2000 and 2010 censuses of Mexico from the IPUMS database to obtain population by age, sex, and rural/urban status. The Mexican census does not indicate the ru- ral/urban status of the households. The most detailed administrative unit included in the data is the municipality. We obtain from the Mexican National Population Council (CONAPO) separate data on the population locality, a sub-unit of the municipality, which can be used to define ru- ral/urban status according to the Mexican definition of a locality with 2,500 or more people as urban. The proportion of a municipality’s urban popu- lation is calculated based on the rural and urban population of its locali- ties. We identify a threshold of percent urban for defining a municipality as urban, such that the derived national urbanization level matches national statistics. Based on the rural/urban status of the municipalities thus derived, the rural and urban population by age and sex are obtained for 2000 and 2010. The directly obtained populations also have age-heaping problems, although less serious than for India. Again we use Whipple’s Index to re- solve the problem (Figure 11).

The 2000 and 2010 Mexican censuses give the year and month of the last live-born child, from which we derive rural and urban age-specific fer- tility rates (Figure 12). Mexican total fertility declined from 2.60 in 2000 to 2.30 in 2010; the reduction in the TFR was larger in urban areas (from 2.50 to 2.20) than in rural areas (from 2.90 to 2.80). The Mexico National Popu- lation Council provides annual estimates of TFRs, which are used to derive annual rural and urban fertility rates based on the rural/urban fertility rates in the 2000 and 2010 censuses.

The two censuses also contain information on migration status in the past five years and on the municipality of migrants’ residence before mov- ing. We calculated annual average migration rates over the periods 1995– 2000 and 2005–2010 (Figure 13) and assumed a linear interpolation be- tween the two for the period 2000–2005. Rural-urban migration rates in Mexico (Figure 13) are slightly higher than in India (Figure 10) for the year 2001, but significantly lower than in the US (Figure 6). Moreover, the rural- urban migration rate actually declines during this period, while urban-rural rates remain nearly constant.

The Mexican census does not have information on mortality. We use the Mexican age-, sex-, and rural/urban-specific mortality rates from the UN’s 2006 Demographic Yearbook to differentiate rural/urban mortality

FIGURE 11 Mexico 2010 population by age, sex, and rural/urban before and after adjusting for age-heaping

LE I W E N JI A N G / BR I A N C. O’NE I L L 379

FIGURE 12 Mexico urban and rural age-specific fertility rates (ASFR), 2000 and 2010

SOURCE: Authors’ calculations from IPUMS Mexican 2000 and 2010 census micro sample data.

rates. The UN Model Life Table (UN 1980, 1982) is adopted to smooth irregular mortality rates of old-age groups (Figure 14). Assuming the dif- ferences between rural and urban areas in 2006 and using the sex-specific life expectancies from national statistics, we derive rural/urban mortality rates for 2000, 2005, and 2010 (UN 2014).

Decomposition

Using the multiregional model and the input data described above, we estimate the determinants of urban growth by backcasting urban popula- tion changes and urbanization levels between census time intervals. The

380 DE T E R M I N A N T S O F UR B A N GR O W T H

FIGURE 13 Mexico annual rural-urban migration rates (percent/year), by age group, 1995–2010

SOURCE: Authors’ calculations from IPUMS Mexican 2000 and 2010 census micro sample data.

LE I W E N JI A N G / BR I A N C. O’NE I L L 381

FIGURE 14 Mexico rural/urban mortality rates by age and sex

backcasting is conducted in single-year steps, and the fertility, mortality, and migration rates between the two censuses are assumed to change linearly if data for interim years are unavailable. We conduct the decom- position analysis in two parts. First, we project urban and rural population

382 DE T E R M I N A N T S O F UR B A N GR O W T H

TABLE 1 Urban population growth and urbanization in India, Mexico, and the US

India Mexico US

1991 2001 2000 2010 2000 2010

Percent urban, census 25.7 27.8 74.4 77.8 79.1 80.1 Urban population (million), census

217.7 286.6 77.4 91.7 224.2 247.9

Projected urban population (million), fertility + mortality

265.3 89.3 240.0

Contribution of natural increase (%)

69.1 83.4 66.7

Projected urban population (million), fertility + mortality + migration

279.8 89.8 241.3

Contribution of migration (%) 21.2 2.9 5.5 Contribution of reclassification (%)

9.6 13.6 27.7

NOTE: Census refers to the population size or percent of urban population reported by the censuses.

growth over the historical period driven solely by natural increase. Second, the projection considers fertility, mortality, and rural-urban migration rates and projects urban population growth driven by both natural increase and migration. The difference in urban population sizes between the number reported by the censuses and the number calculated from the second step of the multiregional population projection we regard as the effect of reclassification, although it also reflects uncertainties in the inputs derived from the census and other data sources.

Results

Table 1 shows the urban population sizes and urbanization levels reported by the censuses for each country, the urban population sizes projected by the multiregional population projections, and the implied contributions to urban growth of natural increase, migration, and reclassification, based on the decomposition method described above.

Natural increase is the main source of urban population changes in all three countries, accounting for 69.1 percent of total urban population growth for India in 1991–2001 and 83.4 percent for Mexico and 66.7 percent for the US, both in 2000–2010 (Figure 15). The contri- bution from net migration is the highest in India (21.2 percent) and rela- tively low in Mexico (2.9 percent) and the US (5.5 percent). The remaining change attributable to implied reclassification is the highest in the US (27.7 percent) and the lowest in India (9.6 percent), with Mexico in between (13.6 percent).

The results are consistent with the conceptual framework presented in Figure 1. In all three countries, natural increase is the primary determinant of urban growth, accounting for 67–83 percent of that growth. India and

LE I W E N JI A N G / BR I A N C. O’NE I L L 383

FIGURE 15 Sources of urban growth in India, Mexico, and the US

Mexico, respectively, are in Stage II and III of their demographic and mo- bility transitions and have high contributions of natural increase that are larger in the more advanced stage. The contribution from natural increase in the US, which is in Stage IV of the transition, is lower than in Mexico but still remains the primary determinant of urban growth. While this outcome is not inconsistent with our conceptual model, the model does not strongly

384 DE T E R M I N A N T S O F UR B A N GR O W T H

constrain the role of natural increase at later stages of demographic and mobility transitions.

The second most important determinant of urban growth is migration in India (21 percent) and reclassification in Mexico and the US (14 percent and 28 percent, respectively), and in all cases these determinants are much larger contributors than the remaining (third) determinant. The outcome for India is consistent with the conceptual model for Stage II countries depicted in Figure 1, in that migration is a more important determinant than reclassification in this stage. The absolute volume of rural-urban migration is large in India (consistent with Stage II, see Figure 1b), despite its very low urban in-migration rates. This is because of India’s very low urbanization level and relatively large rural/urban difference in vital rates, which lead to a large flow of rural-urban migration generated from the large rural population.

The larger role of reclassification (relative to migration) in Mexico and the US is consistent with highly urbanized countries in Stage III and IV of the transitions, although our conceptual model allows for a range of rela- tionships between determinants at late stages of the transitions (Figure 1c). At these stages, inter- and intra-urban migration prevail over rural-urban migration, even though rural-urban migration rates could remain high (in the US, for instance). There is wide uncertainty in the contribution of re- classification, however, given the possibility of periodic changes in govern- ment policies and changes in urban definitions. For instance, the highest proportional contribution from reclassification in the US is mainly due to the recent changes in its definition of urban places.2

To test the robustness of the decomposition results given the uncer- tainty in the demographic rates, for the US and India we carried out an analysis of sensitivity to assumptions about migration, the rate with the largest uncertainty. For the US, we use different assumptions about migra- tion rates over the period 2000–2010, keeping fertility and mortality rates the same as in the original analysis, and examine the effect on the results of the decomposition analysis. We consider three different migration rate as- sumptions: (a) annual migration rates are constant at their 1995–2000 level; (b) migration rates are constant at the 2005 level; and (c) migration rates are constant at their 2009–10 level. We compare decomposition results in these cases to the original results using annual migration rates derived from data for the three time periods as described above. Although there are some variations in the decomposition results, the basic conclusion remains un- changed in all cases: the contribution from natural increase is the largest, in the range of 66.7 to 67.6 percent, reclassification contributes between 22.7 and 27.7 percent, and migration makes the smallest contribution at 5.4 to 10.1 percent. Thus, despite the uncertainty in annual migration rates over 2000–2010, we are confident in the results of our decomposition analysis for the US.

LE I W E N JI A N G / BR I A N C. O’NE I L L 385

Conducting a similar analysis for India, we again found results to be relatively insensitive to uncertainty in the migration data. We used the original annual migration rates for the periods 1991–1996 and 1996–2001 and made two alternative assumptions: (a) annual migration rates remain constant over the whole decade at the 1991–1996 average level; and (b) annual migration rates remain constant over the decade at the 1996–2001 average level, in both cases keeping fertility and mortality rates the same as in the original analysis. We compared results in these cases to the original results using migration rates from both 1991–1996 and 1996–2001. The overall conclusions remain the same in all cases. Natural increase is the largest contributor, ranging from 72.2 to 73.2 percent, the contribution of migration is 22.3–28.3 percent, and reclassification contributes between –1.5 and 5.5 percent.

The negative contribution from reclassification in case b is the result of applying an extreme (high) assumption for migration rates, in which mi- gration over the full ten-year period is at the level that is actually reached only in the second half of the decade. In addition, the age profile of female migrants is concentrated in the childbearing age groups, which also leads to large number of births in urban areas and thus to an increase in the con- tribution of natural increase. The result is that reclassification, calculated as the residual, takes on a negative value. While this did not actually oc- cur in India in this period, it is possible theoretically. In a detailed analysis of the spatial distribution of India’s population by rural/urban settlements for the years 1991 and 2001 (not reported here), we observed some urban settlements being reclassified as rural.

Conclusion and further research

This research aims to improve understanding of the relative importance of the determinants of urban population growth during the demographic and mobility transitions. We synthesize a conceptual model for the relative contributions of natural increase, migration, and reclassification to urban population growth as demographic and mobility transitions proceed. The contribution of migration is dominant at early stages but declines in impor- tance as the transitions proceed, while the contribution of natural increase grows at first and declines at later stages, and the contribution of reclassifi- cation rises over the course of the transitions. For the later stages of the transitions, we lack empirical evidence about the changes and therefore the model is more uncertain about the relative contributions of the three determinants.

Our analysis suggests that when adequate data are available, the multi- regional population model can be used to analyze the multiple driving forces of urban population growth at national or subnational levels for countries at different stages of demographic and mobility transition. The decomposition

386 DE T E R M I N A N T S O F UR B A N GR O W T H

of the relative contribution of natural increase, migration, and reclassifica- tion to urban growth in India, Mexico, and the US is consistent with the conceptual model we identify. Natural increase is the dominant determi- nant of urban growth, consistent with expectations for India and Mexico (which are at earlier transition stages) and less consistent with expectations for the US given the model’s larger uncertainty regarding countries at late stages of the transitions. Rural-urban migration makes a more important contribution in India (even though the country has relatively low migra- tion rates) and a less important one in Mexico and the US, consistent with the model’s expectation of the declining importance of migration over the two transitions.

The model does not have strong expectations for relative contribu- tions of determinants for countries at later transition stages because of a lack of empirical studies and evidence. The fact that Mexico and the US, both highly urbanized countries at later transition stages, show significantly higher contributions to urban growth from reclassification than from mi- gration suggests that the model might be refined in the future to formal- ize this expectation more broadly. Migration’s role can be expected to fall, given the diminishing size of rural populations as countries urbanize. On the other hand, the effects of reclassification can be highly volatile, depend- ing on government policies, regulations, and definitions related to settle- ment design and urban planning. Reclassification could either reflect gen- uine urban expansion as a consequence of socioeconomic and technological change and population growth, or simply be the result of sudden changes in the definitions of urban places. Moreover, different types of reclassification could lead to different urban land systems, depending on whether it oc- curs through expansion of existing urban boundaries or from the addition of separate and independent rural settlements. Further studies of histori- cal changes over longer time periods and additional countries would pro- vide important additional information for refining the model. In particu- lar, a study of accelerating urbanization in India in recent decades would shed light on whether migration is likely to play a more or less impor- tant role as the effects of natural increase decline due to declining fertil- ity and mortality rates across rural and urban areas. In addition, the anal- ysis of the effects of reclassification in 1990–2000 for the US could tell us whether recent reclassification exerted a similar or less significant role com- pared with previous decades. Further analysis of urban growth in earlier decades for Mexico could also examine whether there has been a change in the patterns of demographic determinants over time in that country. Analysis of additional countries at later stages of their demographic and mobility transitions could help reduce uncertainties in the model for such determinants.

LE I W E N JI A N G / BR I A N C. O’NE I L L 387

Notes

This work was funded, in part, by the US National Science Foundation award #1416860 to the City University of New York, the Population Council, the National Center for Atmospheric Research, and the Univer- sity of Colorado at Boulder.

1 According to Zelinsky, there is little or no migration and close to zero natural increase in the first stage of mobility tran- sition, before the onset of urbanization in a pre-modern society. During the second stage of this transition, the society observes a massive urban-ward migration accompa- nied by a rapid rate of natural increase due to demographic transition. This is followed by a third stage in which rural-urban mi- gration is surpassed by urban-urban migra- tion within an urban network in which cir- cular and non-economic migration starts to emerge. The trend of dominant inter- and intra-urban movements continues and is ac- celerated in Stages IV and V in advanced or super-advanced societies, when a large in- crease of migration flow from central city to suburban areas occurs, especially in Stage IV. In the late stages of mobility and demo- graphic transitions, natural increase becomes

moderate as fertility and mortality decline to very low levels. While Zelinsky’s model mainly addresses changes in internal migra- tion, it also mentions changes in interna- tional migration: emigration increases in the second stage but net immigration is observed in the third stage.

2 The US Census Bureau changed its definition of urban in 1880, 1890, 1900, 1950, 1960, and 1980 in response to changes in settlement patterns, data use needs, and technology available for use in defining ur- ban areas. In 2000, the Census Bureau adopted the concept of the urban cluster, de- fined as small densely populated geographic entities with 2,500 or more but fewer than 50,000 persons, so that urban areas include large urbanized areas plus small urban clus- ters. In addition, urban areas are delineated based solely on population density without reference to place boundaries (Bureau of the Census 2011). This change in urban defi- nition had a large effect on rural and ur- ban population estimates, which resulted in a large contribution of reclassification to ur- ban population growth in the US during the period 2000–2010.

References

Becker Charles, M. 2007. “Urbanization and rural-urban migration,” in Amitava Dutt and Jaime Ros (eds.). International Handbook of Development Economics. Cheltenham: Edward Elgar, pp. 516–532.

Bhagat, R.B. and Soumya Mohanty. 2009. “Emerging pattern of urbanization and the contribution of migration in urban growth in India,” Asian Population Studies 5(1): 5–20.

Bocquier, Philippe and Rafael Costa. 2015. “Which transition comes first? Urban and demographic transitions in Belgium and Sweden,” Demographic Research 33(48).

Brass, William. 1975. “Methods for estimating fertility and mortality from limited and defective data.” Chapel Hill: Carolina Population Center, University of North Carolina.

Bureau of the Census, 2011. Urban Definition of the 2010 Census. FR Doc. 2011-21647. Caldwell, John C., Bruce K. Caldwell, Pat Caldwell, Peter F McDonald, and Thomas Schindlmayr.

2006. Demographic Transition Theory. Dordrecht, The Netherlands: Springer. Chan, Kam Wing and Ying Hu. 2003. “Urbanization in China in the 1990s: New definition, different

series, and revised trends,” China Review 3(2): 48–71. Chen, Nancy, Paolo Valente, and Hania Zlotnik. 1998. “What do we know about recent trends

in urbanization?,” in Richard E. Bilsborrow (ed.), Migration, Urbanization, and Development: New Directions and Issues. New York: United National Population Fund and Kluwer Academic Publishers, pp. 59–88.

Dyson, Tim. 2009. “The role of the demographic transition in the process of urbanization,” Popula- tion and Development Review 37(Suppl.): 34–54.

388 DE T E R M I N A N T S O F UR B A N GR O W T H

Economic Commission for Asia and the Pacific, 1995. “Trends, patterns and implications of rural- urban migration in India, Nepal and Thailand,” Asian Population Studies Series No. 138, United Nations, New York.

Goldstein, Sidney. 1990. “Urbanization in China, 1982–1987: Effects of migration and reclassifica- tion,” Population and Development Review 16: 673–701.

Jiang, Leiwen and Brian C. O’Neill. 2009. “Household projections for rural and urban areas of major regions of the world,” IIASA Interim Report IR-09-026, Laxenburg, Austria.

Kelley, Allen C. and Jeffery G. Williamson. 1984. “Population growth, industral revolutions, and the urban transition,” Population and Development Review 10: 419–441.

Kulscar, Laszlo J. and David L. Brown. 2011. The political economy of urban reclassification in post-socialist Hungary, Regional Studies, vol. 45 (4): 479–490.

Kundu, Amitabh. 2011. Trends and processes of Urbanisation in India, Human Settle Group and UNPD, Urbanization and Emerging Population Issues 6.

Ledent, Jacques. 1982. “Rural–urban migration, urbanization, and economic development,” Eco- nomic Development and Cultural Change 30: 507–538.

Lucas, Robert E. 1998: “Internal migration and urbanization: recent contribution and new evi- dence,” IED Discussion Paper Series No. 91, Boston University.

National Research Council. 2003. Cities Transformed—Demographic Change and its Implications in the Developing World. Washington DC: The National Academy Sciences.

Nawrotzki, Raphael and Leiwen Jiang. 2015. “Indirectly estimating international net migration flows by age and gender: The Community Demographic Model International Migration (CDM-IM) Dataset,” Historical Methods 48(3): 113–127.

OECD. 2009: Trends in Urbanisation and Urban Policies in OECD Countries: What Lessons for China. Pernia, Ernesto. 1976. “A method of decomposing urban pouplation growth and an application to

Philippine Data,” Papers of East-West Population Institute, No. 41, Honolulu, HI, East-West Center.

Preston, Samuel H. 1979. “Urban growth in developing countries: A demographic reappraisal,” Population and Development Review 5(2): 195–215.

Rogers, Andrei. 1975. Introduction to Multiregional Mathematical Demography. New York: Wiley. ———. 1982. “Sources of urban population growth and urbanization, 1950–2000: A demographic

accounting,” Economic Development and Cultural Change 30(3): 483–560. Skeldon, Ronald. 2008. “Demographic and urban transitions in a global system and policy re-

sponses,” in G. Martine, G. McGranahan, M. Montgomery, and R. Fernandez-Castilla (eds.), The New Global Frontier: Urbanization, Poverty and Environment in the 21th Century. London: Earthscan, pp. 55–71.

Thompson, Warren S. 1929. “Population,” American Journal of Sociology 34: 959–975. Todaro, Michael P. 1984. “Urbanization in developing nations: Trends, prospects, and policies,” in

P.K. Ghosh (ed.), Urban Development in the Third World. Westport: Greenwood Press. United Nations. 1980: Pattens of Urban and Rural Population Growth. New York. ———. 1982: Model Life Tables for Developing Countries. New York. ———. 2001. “The components of urban growth in developing countries,” ESA/WP. 169. ———. 2014. World Population Prospects 2010 Revision. New York. Visaria, Pravin. 1997. “Urbanization in India: An overview,” in Gavin Jones and Pravin Visaria

(eds.), Urbanization in Large Developing Countries: China, Indonesia, Brazil, and India. Oxford: Clarendon Press.

Zelinsky, Wilbur. 1971. “The hypothesis of the mobility transition,” Geographic Review 61: 219–249.

LE I W E N JI A N G / BR I A N C. O’NE I L L 389

Appendix: The multiregional population/urbanization model

Pt+1x+1,s,r = ( Ptx,s,r + Ptx,s,u × mx,s,u−r − Ptx,s,r × mx,s,r−u

) × Sx,s,r

Pt+1x+1,s,u = ( Ptx,s,u + Ptx,s,r × mx,s,r−u − Ptx,s,u × mx,s,u−r

) × Sx,s,u

where Pt+1x+1,s,r and P t+1 x+1,s,u are the population aged x + 1 with sex s, in rural

and urban area, at time t + 1; mx,s,u−r is the urban-to-rural migration rate of age x and sex s, and Sx,s,r is the survival rate for rural population of age x and sex s.

For the youngest age group in the rural area, the formula is expressed as

Pt+10−4,s,r = (

49∑ i=15

(Pti, f,r × ASFRi,r + Pti, f,r × Si, f,r × ASFRi+1,r )

÷ 2 +Pt0−4,s,r × 0.5 × m0,s,u−r − Pt0−4,s,u × 0.5 × m0,s,r−u) × S0,s,r

where ASFRi,r is the age-specific fertility rate of rural women, and P t i, f,r is

the rural female population of age i at time t.

Copyright of Population & Development Review is the property of Wiley-Blackwell and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.