Migration and the Productivity of LDC Mega-cities: Evidence from the Cairo Conurbation

Migration and the Productivity of LDC Mega-cities: Evidence from the Cairo Conurbation Barry McCormick & Jackline Wahba University of Southampton, UK Address: Dept. of Economics University of Southampton Southampton, SO17 1BJ United Kingdom Fax: +44 23 80 593858 E-mail: bmc@soton.ac.uk jew3@soton.ac.uk April 2001

Abstract The growth of LDC mega-cities is conventionally high, viewed as the consequence of, locally unassimilable, rural population growth. A more recent view is that cities have expanded in order that the productivity gains from close proximity particularly for educated and young workers are extended to a greater share of the population. This is either because changes in technology or the terms of trade, have made workers in cities relatively more productive, or because the shares of the young and educated in these countries have increased over time. This paper uses individual data to explore migration flows into and out of, one of the most prominent LDC mega-cities the Cairo conurbation in order to explain the extent to which migration alters the labour force structure of a dense urban area in the direction predicted by the productivity hypothesis. This complements the LDC migration literature that has shown how migration contributes about 50-70% of the overall growth of LDC cities, but is mainly focussed on rural-urban and inter-regional migration Lucas (1997). Four hypotheses are explored explaining the structure of migration into and out of large LDCs cities and urban labour productivity: 1) Large cities drain both other urban areas, and rural areas, of educated workers; 2) Large cities have a labour force of below average age and partly achieve this by attracting net in-migration of young persons and prompting net out-migration of older individuals; 3) Distance has a less systematic role in explaining migration to a large city to that of migration generally, with large centres developing their own, perhaps distant, catchment areas; 4) Migration from rural areas to large cities is more sensitive to local labour market conditions than migration from urban areas. Evidence concerning in and out migration to the dense urban area in North Egypt, comprising Cairo, Giza and Alexandria are used to test these hypotheses. Keywords: urbanization, internal migration, developing countries, city growth. 2

1. INTRODUCTION It is estimated that, by the year 2015, over 10% of the world's population will live in 26 cities with 10 million or more inhabitants, and that 22 of these cities will be in the developing world. 1 Despite the importance of large cities in LDCs there has been relatively little econometric analysis that contributes to explaining their rapid growth. The conventional view is that LDC mega-cities are a result of high rural population growth that could not be locally absorbed, and of migration high distorted urban wages that have encouraged rural-urban migration. 2 However, there are other explanations of city growth that stem from the increasing recognition that cities have a distinctive productive role that arises from spatially concentrating workers who most benefit from close proximity. For example, changes in the terms of trade and tariffs, which affect the relative price of goods efficiently produced in densely populated areas, or in the supply of educated and skilled workers, which influence the relative costs of such goods, might all alter the equilibrium share of employment in LDC cities - Krugman and Livas-Elizondo (1996). It has also been hypothesized that educated workers secure greater benefits from (i) spatial learning spillovers, and/or (ii) the greater variety of consumption goods in cities. In these explanations, cities provide efficiency gains by concentrating certain types of workers primarily the young and the educated/skilled and migration not only influences city size, but also changes the composition of the labor-force. This paper is concerned to explore for one LDC mega-city whether the structure of migration into and out of the city is consistent with the view that cities 1 The evidence is summarized in the United Nations Report World Urbanization Prospects (1999). 2 See, for example, Lucas (1997), which surveys the evidence of that artificially high urban wages may prompt city growth. In this framework city growth is inefficient if workers leave rural jobs with positive productivity to become unemployed in cities. 3

have grown, at least in part, for efficiency rather than rural overspill reasons. In particular, we examine whether migration flows concentrate the young and educated into a mega-city, and whether the structure of migration identifies a separate influence for the variant of the efficiency approach to city growth which emphasizes that the learning and consumption benefits of city location are greater for educated workers. There exists a variety of arguments and evidence that analyses how the spatial concentration of workers within cities may improve the co-ordination of activities, accelerate the flow of ideas, and create spill-over benefits between skilled workers that reduce the costs of skill-acquisition and raise the returns to skill Lucas (1988), Rauch (1993), Glaeser et.al. (1992), and Glaeser (1999). Cities may also assist the matching of workers to jobs, enable industries that require a high quality or diversified labor force to benefit from physical proximity to each other for example, Glaeser and Mare (1994) and allow establishments to exploit economies of scale while benefiting from an elastic, and hence flexible, labor supply. Cities also spatially concentrate demand sufficiently that a wider diversity of consumption goods can be supplied in the city than in a less populated area. If productivity gains are to be provided by LDC cities, their labor force structure may need to considerably evolve, primarily to achieve large shares of the groups hypothesized to benefit from close proximity- the young and educated. This may occur either as a consequence of internal mechanisms such as training by employers and high quality city schools, spatially uneven fertility rates, or migration flows both into and from the city. However, while it is known that rural to urban migration is dominated by young adults 3, very little is known about the other features of migration to cities, or from cities, and the extent to which selective migration alters 4

the composition of a city s labor force. Thus while it is widely recognized that migration explains in the order of 50% to 70% of LDC city growth - for example, Mazumdar (1987) and Eastwood and Lipton (2000)- little is known about the structure of flows into and out of cities, and thus whether selected migration alters the composition of a city's labor force, as well as the scale. Important work by Henderson (1986), (1988) shows how the share of skilled workers in cities in US and Brazil increases with city size but does not address empirically the role of migration in the evolution of the cities. The enormous literature on developing countries migration as the valuable survey by Lucas (1997) notes, mostly focuses on rural-urban and inter-regional migration. Mazumdar (1987) points out that migration to and from large cities, as opposed to other urban areas, is an important type of movement, and likely to have distinctive characteristics. There are several studies of the structure of migration into large cities in developing countries - for example, Fuller (1985) and Mazumdar (1981) study Bangkok and Kuala Lumpur, respectively - and at least one study, of migration from a large city Pessino (1991), who studies Lima. However there do not appear to be any studies using individual data to test hypotheses concerning the comparative structure of flows into and out of an LDC city. This paper provides this analysis for Cairo/Giza/Alexandria. Section 2 draws together and extends in a minor way, the ideas explaining the structure of migration into and out of large cities. We then examine Sections 3 and 4 these ideas using Egyptian data. Since we are concerned with large cities and most studies of migration concern rural-urban flows we explore a three-way distinction between: rural areas, other urban areas and large cities. This enables us to examine 3 See Todaro (1976), for example. 5

separately hypotheses that distinguish between a) rural, and b) urban migration into and out of large cities. Consequently, our econometric work explores separately flows between the sprawling urban conurbation in Northern Egypt, comprising Cairo, Giza, and Alexandria, a) rural areas, and b) less dense urban areas. Section 5 concludes with a brief summary of findings. 6

2. THE STRUCTURE OF MIGRATION AND LARGE CITIES Hypotheses concerning the structure of migration to and from large cities are grouped under four headings. Hypothesis One: Large cities drain other urban and rural areas of educated workers. Beginning no later than Marshall (1890), economists have argued that large cities attract educated labor out of both rural areas and smaller cities. 4 At least three reasons have been given to explain why this may occur. First, on the demand side, it is hypothesized that firms with high-skill requirements benefit most from the efficiency gains resulting from close proximity. These benefits for high-skill firms from high density locations may arise in terms of lower costs of matching heterogeneous skilled workers to the tasks to be done, and the more rapid flow of information about competitor products and production methods. High skill firms are also frequently in the service sector where face-to-face customer contact is of particular value. Large cities generally possess a broader skill range and so can offer a more sophisticated service infrastructure than other urban areas - Henderson (1988)- and this may be of particular value to high-skill technology industries. On the supply side, high skill people are conjectured to be attracted to large cities because of the more extensive consumption amenities and services. This is sometimes called the bright lights hypothesis. One such amenity may be a high ratio of skilled to unskilled workers. Differences in skill composition across cities have various further implications. For example, if large cities have a relatively educated voting population, this may enhance the quality of local public decision making and the local quality of life. Glaeser (1999) discusses a second supply side reason why 7

skilled individuals are attracted to large cities: dense urban agglomerations provide learning opportunities. Thus if skilled workers have the highest returns to these learning opportunities, skilled workers are more likely to move to large cities as a form of human capital investment. 5 These theories suggest that in equilibrium the employment shares of high-skill workers are greater in large cities than towns or rural areas. Using US Census Data for 1990, Glaeser (1999) shows that larger cities have a greater share of high skilled persons relative to smaller cities. Henderson ((1986), (1988)) discusses evidence for US and shows that in the case of Brazil, the correlation between the ratio of high to low skill population and city size is 0.69 across all cities. (Pessino (1991) shows that for Peru, the highly educated are not more likely to migrate from Lima, and the same result applies to other urban areas. However, migration from rural areas, albeit to any destination is more likely for the highly educated. ) What are the implications for migration? First, it is worth noting that the preceding models do not imply that we should necessarily expect to observe that there will be net migration of educated workers in cities: in equilibrium the employment share of high skill workers, relative wages or the cost of living in cities may adjust to eliminate net migration flows. However, within each skill group the amenity value of cities must be offset by a compensating differential. Thus, the supply-side arguments imply that in competitive markets, if within-skill wage rates are equal over space that skilled workers will be more likely than unskilled workers to migrate into a city in order to benefit from either their greater utility from consumer amenities, or the 4 Alfred Marshall s (1890) observations are cited in Glaeser (1999) p2-3. 5 Marshall (1890) argues that skilled workers benefit from being near to one another in dense areas because they learn new skills, or as he calls it, learn about the mysteries of trade. 8

greater learning opportunities that city work offers the skilled. 6 By itself, however, this empirical prediction is not a very powerful test of the supply-side explanations of city growth; there are other reasons why the more educated have higher migration rates that are unrelated to spatial concentration for example, Schwartz (1976). However, these other arguments apply to migration to all destinations so that in Section 4 we explore the hypotheses given here by comparing the influence of education on migration from (i) urban, and (ii) rural areas into a large city, with that of education on migration from a large city to these areas. We shall reject the hypotheses that supply arguments have raised the share of educated employment in cities if education fails to have a significantly larger effect on the probability of migration to the city, all else equal, than it has on migration from the city to (i) urban, or (ii) rural areas. Hypothesis Two: Large cities achieve a large share of young workers by (i) attracting the net in-migration of young persons and (ii) encouraging the net out-migration of older individuals. It is widely conjectured that the young are a disproportionate share of those moving to cities for example, Todaro (1976) but only recently have economists discussed why migration might reduce the mean age of city workers below that elsewhere. Young persons have long been recognized to be more willing to invest in learning given that they can expect more years over which to realize the benefits. If cities offer learning spillover benefits from living near to large numbers of skilled workers, young people will become net migrants to cities, while older workers with fewer years to 6 Pessino (1991) shows that for Peru, the highly educated are not more likely to migrate from Lima, and the same result applies to other urban areas. However, migration from rural areas, albeit to any 9

enjoy the learning externalities from cities, and similar costs of living in dense areas, become net-migrants from cities. Moreover, if young workers seek to experience a range of jobs, then since cities spatially compress jobs, sampling need not require commuting or migration costs. Glaeser (1999) shows that younger persons have a greater propensity to live in US cities - especially large ones, using Census data on stocks of employment by US city size. 7 As with Hypothesis One, the force of this argument would appear stronger for those migrating from rural areas to cities than for rural migrants to cities, since rural areas are less likely than small urban centers to provide learning opportunities for young people. The higher price of land in cities may also explain the lower mean age in cities. Older workers, perhaps with a pension income or other financial wealth, may choose to work fewer hours. Since income from savings is the same at all household locations, but the cost of living is not, workers supplying fewer hours will have a lower benefit from, and thus lower demand for, city location. This mechanism would provide a second explanation of how migration enables the share of young to rise in cities, in order to allocate scarce city accommodation to those working more hours. As with education, there are several reasons why young workers are more likely to migrate that are unconnected with spatial density. Hence we adopt a similar econometric strategy to that outlined for education and test whether the influence of education on the probability of migration to a city exceeds that of the influence of education on migration from a city, ceteris paribus. Our next two hypotheses do not explain how migration alters the productive capacity of the labor force but do aim to explain the circumstances and spatial origins destination is more likely for the highly educated. 7 This evidence is presented in a Table analogous to our Table 1 - see Glaeser (1999). 10

of migrants to cities, and hence contribute to our choice of control variables in the econometric models. Hypothesis Three: Distance has a less systematic role in explaining migration to a very large city than that of inter-regional migration, with large centers developing, their own, perhaps distant, catchment areas. Distance between origin and destination is usually found to have a deterrent effect on migration for both developed and developing countries - for example, Greenwood (1997) and Lucas (1997). In contrast, as Mazumdar (1987) observes, Large cities typically develop their individual catchment areas from which migrants are drawn- and these areas are not necessarily concentrically distributed in terms of distance. These catchment areas could be determined by the historical accident of which family/ethnic groups first established communities in the city, and thereby acquired the widest network of contacts for potential migrants. (In the case of Egypt such an incident occurred when the construction of the High Dam in Aswan in 1972 caused Nubian communities to migrate to Cairo following loss of their land.) Thus the deterrent effect of distance on migration to large cities may differ to that for interregional migration. To develop this hypothesis we shall explore the extent to which certain distant rural areas in Egypt have become predominant suppliers of migrants to large cities. Work by Katz and Stark (1986) suggests a second reason why remote rural areas may be more likely to provide migrants to a large city, then rural areas nearer to the city. Rural migrants may be thought of as choosing whether to migrate to a nearby urban area or to a more distant city. If in regions far from large cities, the prosperity of urban areas is more highly correlated with the rural hunter land, then rural families 11

in such regions seeking to diversify their sources of income, will be more likely, ceteris paribus, to send workers to a distant city. Finally, we add a further new hypothesis to help to understand the circumstances in which certain areas supply migrants to the city. Hypothesis Four: In LDCs, migration from rural areas to large cities is more sensitive to local labor market conditions than migration from urban areas to large cities. In LDCs, public goods and subsidized services are often concentrated into urban areas, with rural areas receiving comparatively low levels of services. Thus in times of adverse labor market conditions, when families look to social provision for income/food/health support, the rural areas are less likely to be provided with social resources than urban areas. Hence for any given adverse movement in relative regional wages or unemployment, rural workers are more likely to migrate for better long-term opportunities than urban workers. We assume that some of these rural migrants seeking social protection go to the Cairo conurbation. 12

3. THE EGYPTIAN CONTEXT AND DATA 3.1 Background to City Growth in Egypt With approximately 45% of the population living in towns and cities, mostly located near to the Nile, Egypt is a comparatively urbanized LDC. Although Cairo continues to dominate the urban population, small and medium-sized towns have grown strongly during the last three decades - Karoufi (1996). 8 These towns provide various service activities- administration, education and health- as well as manufacturing centers. In contrast, the 1986 and 1996 Census evidence shows no signs of population growth in both of the largest cities: Cairo and Alexandria. While the data studied here confirm the low net migration to Cairo/Alexandria in the 1980s and 1990s, we shall explore how far migration is changing the composition of the labor force in these cities. 3.2 The Data The data sets used are the labor mobility modules of the 1988 and 1998 Egyptian Labor Force Sample Surveys 9 which provide detailed information on approximately 20,000 individuals. The 1988 LFSS reports residential and work locations in October 1988 (the time of the survey) and retrospective information concerning October 1981, while the 1998 LFSS reports locations in November 1998 and retrospective information concerning August 1990. 10 We examine migration to 8 Between 1976 and 1986, the inter-census period, the urban population grew by 10% and the number of towns having more than 100,000 inhabitants rose from 20 to 24. Egypt certainly conforms to Lucas' conjecture (1997), p.727, that it is not "true that the largest cities are out-growing the medium size cities." 9 The 1988 and 1998 surveys were conducted by the Central Agency for Public Mobilisation and Statistics (CAPMAS) in Egypt. Both surveys are special rounds of the LFSS, which include additional information on mobility. 10 October 1981 was chosen to benchmark retrospective information since President Sadat was assassinated in that month. August 1990 was chosen because it was the time of Kuwait's invasion. 13

and from the biggest three cities in Egypt: Cairo, Giza and Alexandria. These three cities accounted for 52.5% of the total urban population in 1986 and are located within around 200 kilometers of each other in Northern Egypt. The analysis is confined to male workers in the labor force who are aged between 15-64. The total sample size used is 7783 individuals of whom 208 are migrants. Table 1(a) displays the distribution of the working population between large cities (Cairo, Giza & Alexandria), urban areas (small cities and towns) and rural areas in Egypt, by level of education (secondary and university degree holders), and by age category (15-34, 35-54 and 55-64). Consistent with other studies (for example, Henderson (1986) and (1988)), large cities have a higher share of educated workers than other urban areas: 23.2% of the large cities working male population have secondary or higher level of education, compared to 20.6% in other urban areas, and 6.1% in rural areas. Furthermore, 52% of the labor-force with secondary or higher education (at least twelve years of schooling) live in the large cities - Cairo, Alexandria, or Giza. However, it is striking that amongst those under 35, the proportion of educated workers in the large cities is only slightly greater (1%) than in all other urban areas, and workers under 35 in Egypt comprise a substantial fraction - about 41%- of the labor-force. Thus the education gap between large cities and other urban areas is very small for those under 35. The proportion of young (15-34) in big cities is 1% greater than in other urban areas and 9% less than in rural areas, where fertility is high. The proportion of older workers (55-64) in the cities (8%) is slightly less than in other urban areas, and Table 2 also shows that the mean age of the work force is marginally lower in the large cities than in other urban areas. Overall, the descriptive evidence provides only slight support to the view that large cities have higher shares of young workers, but stronger 14

evidence that the large cities have more educated labor forces than other urban or the rural areas. The flow of migrants, in our sample, between the rural, urban, and large city areas, is given in Table 2. There are 50 migrants from rural areas to large cities, and 27 in the opposite direction. These figures reflect a net flow of both illiterate and welleducated rural migrants into the large cities. There are 38 migrants into large cities from urban areas, and 55 migrants from large cities to other urban areas. However the flows of educated workers are about equal in each direction, and so it is a flow of illiterates from the large cities to other urban areas that creates the net flow from large cities. Thus, perhaps surprisingly, for flows between big cities and other urban areas, migration contributes to the higher educational levels in the large cities by draining less educated workers from the large cities, rather than by "draining" highly educated workers from other urban areas. The educational composition of migration from rural areas to large cities is very similar to that of migration from large cities to rural areas (Columns 2 and 9). However, since the scale of rural migration to big cities is about double that from big cities, migration flows raise educational levels in the cities only if migrants are more educated than mean city levels. Since the education levels of rural-city migrants are marginally lower than city inhabitants (stayers), we cannot conclude that rural migration is directly contributing to the comparatively high educational levels that we have found in the Cairo/Alexandria conurbation. Overall, the only mechanism whereby migration directly helps to raise educational levels in the conurbation is through net outflows of less educated workers to other urban areas. Tables 1(b) and 2 provide descriptive evidence concerning the contribution of migration selectivity to the lower mean age in large cities. Table 1(b) shows that 39% 15

of migrants to large cities are under 27 years, while less than 21% of migrants from large cities are under 27 years old. In other words, whilst migration is concentrated amongst the young, this is particularly so for migrants to Cairo, Alexandria and Giza, and less so for migrants out of these densely populated areas. The mean age of the migrant categories are given in row 1 of Table 2, and suggest that migration contributes to reducing the mean age of those living in the conurbation in Northern Egypt. However, this effect comes almost entirely from migration to and from rural areas, rather than less dense urban areas, with the mean age of rural migrants to the city over 7 years below that of migrants from the city to rural areas. It is noteworthy that rural migrants to the conurbation are disproportionately drawn from Upper Egypt (53%) the South- whereas urban migrants to the conurbation are from Lower Egypt and the Suez Canal area, (62 + 14)%. It is striking that migrants from the large cities travel only 44 kilometers on an average, whereas migrants to the large cities come from a much wider geographical "catchment", and travel on average 180 kilometers. Finally, the large cities appear to be exporting workers to manufacturing jobs rather than attracting them to such jobs from other urban areas, which is consistent with manufacturing growth sectors being based outside the Cairo/Alexandria conurbation. 16

4. HOW DOES MIGRATION ALTER LABOR SUPPLY IN CAIRO? ECONOMETRIC EVIDENCE The evidence described in Section 3 gives qualified support for the view that structure of migration between the Cairo/Giza/Alexandria (C.G.A.) conurbation and other areas (described in Table 2) helps enable LDC large cities to bring into close proximity those groups most likely to productively benefit - the young and educated and that net migration to these cities does not merely increase the scale of the city labor force. In this Section we explore this in more detail, allowing for various other influences on migration into and out of C.G.A., and investigate the hypotheses described in Section 2. Since our arguments distinguish C.G.A. from both other urban and rural areas, the econometric analysis first estimates models of the probability of individual migration into C.G.A. from both (i) rural, and (ii) other urban areas. We then estimate a model of an individual in C.G.A. choosing between migrating to a rural area, another segment of C.G.A., an urban area beyond C.G.A., or not migrating. By contrasting the parameter estimates of these models of migration into and out of C.G.A., we are able to test the implications for migration of the various hypotheses. To implement this, in each location choice model we assume that the indirect utility of location choice j for individual i can be written U ij = β X + ε j i ij where U ij is the utility from choice j by individual i, β j is a vector of parameters which may vary between location choices, and X i is a vector of attributes of individual i which includes characteristics of the region in which the individual works. The regional characteristics included are the relative wage and unemployment rates, and measures of the region s distance from Cairo. 17

If the errors ε ij are independently and identically distributed with the type I extreme value distribution, this gives the multinomial logit model. The probability of individual i choosing location j is given by P ij ' = exp( β X j i ) /(1 + Σ j ' exp( β X j i )) Thus we begin by estimating two binomial logit models of the probabilities of migrating from (i) rural areas to Cairo/Giza/Alexandria, and (ii) urban areas to Cairo/Giza/Alexandria. Next, we study the determinants of migration choice for individuals previously working in C.G.A., using a multinomial model (j =0, 1, 2 & 3). Let j be the regional location in the survey period (1988 or 1998). For C.G.A. workers in the reference year (1981 or 1990), j = 0 is staying in the origin part of C.G.A.; j = 1 migrating to another part of C.G.A.; j = 2 migrating to another urban area; and j=3 migrating to a rural area. To estimate the preceding models we combine Labor Force Surveys for 1988 and 1998 to study migration between 1981-1988 and 1990-1998, respectively. Since these two periods are 7 years and 8 years respectively, we include a dummy variable in the migration models to allow for this implied higher probability of migration in the second sample. The data used is described in Section 3. We now discuss the testing of the hypotheses in Section 2 and the various control variables used. Education. Evidence from both LDCs and DCs shows that both before and after controlling for other influences, the propensity to migrate is generally higher for the more educated for example, Mazumdar (1987) and Lucas (1997)- both before and after the inclusion of other explanatory variables. There are several possible explanations for this. First, formal qualifications may reduce the uncertainty associated with migration by contributing to a person s ability to collect and process 18

information. Also, the higher migration rate of the educated may partly reflect the responsiveness to larger spatial wage differentials than exist for the uneducated (see Fields (1982) and Schultz (1982)) as well as to the higher growth rate of jobs requiring educational qualifications in the urban economy. Support for Hypothesis One requires that the incremental influence of education on migration to large cities is not only positive but in excess of that found for migration from large cities. To examine this we distinguish between three educational groups: those with no education, those with less than secondary degree (less than primary, primary, and preparatory degrees) and those with secondary and higher (secondary, university and postgraduate degrees). Age. Younger persons are usually conjectured to be more mobile than older persons because the lifetime income gains to moving are larger for the young. In contrast, older individuals have higher moving costs from having longer locational family ties. In most developing countries migration is concentrated in the 15-30 age group, with a substantial portion in the 15-24 sub-group. Support for Hypothesis 2 requires not only that being young has a positive effect on migration to large cities, but also that the effect exceeds that estimated for migration from big cities. We shall use parameter estimates to construct profiles of the forecast probability of migration by age, holding constant other influences, for both in and out migration to large cities. We shall also differentiate by flows from rural and other urban locations. Three age groups dummies, 15-26, 27-34, 35-64 years old, are used to provide age differentiation. Distance and Regional Location. Most migration studies including those for LDCs - find that distance between origin and destination is an important deterrent to migration. As Yap (1977) points out, the distance coefficients in migration functions are 19

typically negative and very significant when included with, for example, employment, wage, education and urbanization variables. The likelihood that with increased distance the psychic costs of migration rises, and availability of information about labor markets falls has provided the primary explanation of the deterrent effect of distance. In this paper the distance variable measures the distance between the area of origin rural or urban governorate and Cairo. In Table 2 we found above that Rural Upper Egypt (the South) provides a disproportionate share of rural migration to Cairo, Giza & Alexandria, and we will ask how far this occurs despite a significant distance variable effect, or whether it is facilitated by the unimportance of distance to destination. We also wish to account more generally for the factors behind extensive migration from the rural southern areas to the large cities in the North. In this analysis, location concerns the workplace (not residence) so that migrating into a city may reflect no change in the location of residence but a decision to commute. For households living within commuting distances to the city, migration is plainly easier and so we introduce a variable capturing whether a worker lived and worked within commuting distance seven/eight years before the interview. Our contiguous location dummy variable is intended to capture this and takes the value 1 if previous work location is less than or equal to 50 Kilometers from Cairo. Regional fixed effects are used to control for regional specific characteristics of region of origin that may influence migration. Two regional dummies are used; Upper is the much more remote south of Egypt; Lower is the north. Origin Local Labor Market Conditions. Various studies show that migrants respond to economic circumstances in local labor markets. However, relatively few of these studies allow for both earnings and employment probabilities at origin and destination. The available results together show that some combination of economic 20

incentives matter for migration, but there is little consensus emerging from LDCs' studies that one of these influences is especially important. 11 Lucas (1985) formulates a fairly general specification using micro data to study migration in Botswana and finds that both origin earnings and employment probabilities have an equal and opposite effect to the same variables at destination. We explore how the relative wage and unemployment rates in urban and rural areas in turn, explain the likelihood of migration to the mega-city. Since the latter is the sole destination, and we have a single cross-section, we are able to eliminate the destination condition from our models and express both urban and rural origin log wages and log unemployment relative to the national average. 12 To add specificity to these variables we (i) differentiate the hourly wage at origin by educational category, and express relative to the national average hourly wage, and (ii) differentiate the unemployment rate at origin by age category. Local Amenities: Urban bias in the allocation of public services and amenities is seen by many to be an important determinant of rural-urban migration. 13 To capture this effect, we use water supply, the proportion of the rural population with access to piped water, as a measure of the local provision of amenities in rural areas of origin. Employment Sector. Given the different opportunities for migration afforded by the public sector in Egypt, we allow for being a public sector employee. We also differentiate by sector of employment: agriculture, manufacturing, construction and 11 See the surveys by Lucas (1997) and Mazumdar (1987). 12 Hourly wage rates are calculated from the earnings module of the 1988 and 1998 LFSSs. Since using grouped (aggregated) data in individual level regressions, can potentially result in the standard errors being biased because of the correlation of the error term across individuals in a region or industry, we correct for the correlation of error terms across individuals in each province and report the robust estimates. Robust (Huber/White/Sandwich) estimator of the variance was used in place of the conventional Maximum Likelihood Estimation variance estimator and observations were allowed to be not independent within governorate/province. 13 See Eastwood and Lipton (2000) for a recent discussion. 21

services. We expect agricultural workers to be least mobile given that their sector specific skills are not easily used in urban areas for recent evidence see Tunali (1996). Manufacturing workers are often found to be less mobile than service and construction workers in developed countries studies, but this may not hold in countries where manufacturing skills are scare. Empirical Results The estimates of the models of the probability of migration to the C.G.A. conurbation are given in Table 4. Column 1 of Table 4, provides the odds ratio of urban migration to the. conurbation, while column 2 provides the odds ratio of rural migration to the conurbation. Estimates of the multinomial model of migration from the conurbation to (i) another area of the conurbation, (ii) another urban area, (iii) a rural area, are given in Table 5. More specifically, Table 5 gives the relative odds ratio of migration to another large city (column 1), migration to another urban area (column 2), or migration to a rural area (column 3) relative to not migrating from a large city. Consider first implications of the supply-side version of Hypothesis One, which requires that having a higher level of education, ceteris paribus, not only increases the migration probability of both rural and urban workers to large cities, but also that this effect is greater for migration to the city than from it. We find in Table 4 that for both rural and urban workers the estimates give a consistent picture of a positive effect of increasing levels of education on migration to large cities. Table 5 gives a more mixed picture of the effects of education on migration from the conurbation. It shows that highly educated workers are more likely to migrate from large cities, ceteris paribus, provided their destination is to other urban areas. However, this is not valid for literate workers without secondary education, who are less likely to 22

migrate from large cities than illiterates. We find no tendency for additional education to increase out migration from large cities to rural areas: the descriptive evidence in Table 2 appears to reflect the intervening effects of other variables. Thus the structure of rural/city flows support Hypothesis One. The effect of additional education on migration from large cities to other urban areas (Table 5, col.2) is weaker than that on migration to large cities from other urban areas. However, this difference is not sufficiently strong that we can reject the null hypothesis of identical effects. Pessino (1991), who provides the only comparable study of out migration, finds that more educated residents of Lima are less likely to migrate out. However she does not distinguish urban and rural destinations, and does not consider migration into Lima. Thus the Egyptian evidence supports the supply-side version of Hypothesis One for rural-large city flows, but is less clear for urban-conurbation migration. We infer that the higher relative risk of migration for educated workers from rural areas to the city helps to maintain a high share of educated workers in the city. The dim lights of other urban areas may be sufficiently strong for this effect to be of more marginal relevance for urban-city migration. Turning now to Hypothesis Two, and again using Tables 4 & 5, we find that the likelihood of migration to large cities is inversely related to age, regardless of the migrants' origin. Persons who are 15-26, inclusive, are the most likely to migrate. Thus, large cities attract younger persons from other areas. Table 5 indicates that young workers (15-26) are not more likely to migrate from large cities. Corrected for other effects, migration to large cities peaks at a younger age (15-26), while migration out of large cities, to rural and other urban areas, peaks in the 27-34 group. The implications for probabilities are given in Table 6. We conclude that there is evidence to support the hypothesis that migration patterns reduce the mean age of the large cities labor force by 23

increasing the probability of in-migration amongst younger individuals more than that same relative probability for out-migration. The influence of being above average age is more mixed. Thus the apparent influence of age in the summary statistics of Table 1b survives the extraction of the intervening influence of other variables, but the effect is strongest in migration to and from rural areas. The evidence concerning Hypothesis Three consistently supports the view that once nearby areas to C.G.A are treated separately, using the commuting distance dummy variable, the probability of migration to C.G.A is not significantly affected by the potential migrant's distance to the large cities. This supports our a priori discussion that distance has a less systematic role in explaining migration to a very large city than that of inter-regional migration. Various non-linear specifications for the role of distance yield similar conclusions. We also find that households living and working previously within commuting distances are more likely to migrate to work in the conurbation from both urban and rural areas, although this effect is stronger for urban workers. Turning to Hypothesis Four, migration from rural areas to large cities does not appear to be sensitive to local labor market conditions. The findings suggest that neither relative unemployment rates nor wage rates in rural areas affect rural migration to big cities. Although the economic factors, such as income and employment opportunities, do not seem to affect rural migration, access to local amenities does appear to have an important influence. However, improved access to water supply, our measure of local amenities, has a negative and significant impact on migration, indicating that the greater the provision of rural amenities, the less is the probability of rural migration to large cities. 24

The relative wage rate at origin appears to have the conventional effect in the case of urban migration to large cities: lower than average origin area wage rates induces urban migration to large cities. However, the impact of unemployment rates on urban migration to large cities is not significant. Overall, the evidence found using these data do not support Hypothesis Four. We now summarize the contribution of the control variables. The likelihood of migration by a public sector worker from rural areas is less than of a private sector worker. Being employed in the public sector has no significant impact on large city workers migrating to urban and rural areas. One explanation may be the fact that it is not uncommon to transfer public sector workers from large cities to smaller cities and rural areas. The industry to which the individual is attached plays a conventional role in determining migration in and out of large cities. Workers in the construction sector are more mobile than workers in the services sector, regardless of their origin and destination. Rural workers attached to the agriculture sector are less likely to migrate to large cities. This perhaps reflects their comparative lack of skills for work in urban areas or alternatively the existence of non- marketable rents, which are earned in agriculture sector work. 5. CONCLUSION The conventional explanation of the recent explosive growth of LDC megacities is that it provides a repository for high rural population growth that cannot be absorbed in the rural economy. Whereas various papers test how rapid LDC city growth may result from labor market distortions, relatively little attention has been given to exploring how far migration to LDC cities have expanded, at least partly, in order that the productivity gains from close proximity particularly for educated and 25

young workers are extended to a greater share of the population. The "productivity" theory implies that migration acts as a sorting device, raising the share of young and educated in cities. This paper explores whether that share of the young and educated is higher in cities and whether migration into and out of, one of the most prominent LDC "mega-cities" the Cairo conurbation alters the labor force composition of this densely populated urban area in the direction predicted by the bright lights and learning amenities version of the productivity hypothesis. Our main findings are that the share of educated workers is marginally higher in the large conurbation than in other urban areas. We also find that the pattern of rural-city migration raises the share of educated workers in the city, but evidence of the same for "other urban"-city migration is suggestive but statistically insignificant. Migration amongst the young enables the city to experience a significantly lower mean male age of worker, but migration amongst the old is not sufficiently differentiated to significantly reinforce this effect. These results suggest that migration patterns are broadly consistent with models in which LDC cities have grown partially in order to spatially concentrate young workers and educated workers. To what extent are these groups an increasing share of the labor force? The Labor Force Survey evidence shows that in 1988 28.9% of the labor force were under 29 years of age, but by 1998 this had risen to 35.4%. Evidence from the Census documents rising educational standards: between 1960 and 1986 the share of the population having University degrees rose from 1.7% to 7.5%; over the same period the percentage illiterate dropped from 63.8% to 45.6%. 26

REFERENCES Eastwood, R. and M. Lipton (2000). Rural-urban dimensions of inequality change.' United Nations University World Institute for Development Economics Research, Working Paper no. 200, September. Fields, G. (1982). Place-to-place migration in Colombia. Economic Development and Cultural Change, vol. 30, pp. 539-558. Fuller, T. D., et al. (1985). Rural-urban mobility in Thailand: a decision- making approach. Demography, vol. 22, pp. 565-579. Glaeser, E. (1999). 'Learning in cities', Journal of Urban Economics, vol.46, pp. 254-7. Glaeser, E. and D. Mare (1994). Cities and skills', National Bureau of Economic Research Working Paper: 4728, May. Glaeser, E. et. al. (1992) 'Growth in cities', Journal of Political Economy, vol.100, pp. 1126-52. Greenwood, M. J. (1997). Internal migration in developed countries. Ch.12, in Rosenzweig, M. and Stark, O. (eds.), Handbook of Population and Family Economics, vol. 1B, The Netherlands: Elsevier Science Publishers. Greenwood, M. J. (1969). The determinants of labor migration in Egypt. Journal of Regional Science, vol. 9, pp. 283-290. Henderson, J.V. (1986). Urbanization in a developing country: City size and population composition. Journal of Development Economics, vol. 22, pp. 269-293. Henderson, J.V. (1988). Urban Development: Theory, Fact and Illusion. Oxford: Oxford University Press. Karoufi, M. (1996). Urbanization and urban research in the Arab world, Management of Social Transformations, Urban Research in the Arab World, Discussion Paper Series, no.11. Lucas, R.E.B. (1985). Migration amongst the Botswana. The Economic Journal, vol. 95, pp. 358-382. Lucas, R.E.B. (1997). Internal migration in developing countries. Ch.13, in Rosenzweig, M. and Stark, O. (eds.), Handbook of Population and Family Economics, vol. 1B, The Netherlands: Elsevier Science Publishers. Katz, E. and Stark, O. (1986). Labor migration and risk aversion in less developed countries. Journal of Labor Economics, vol.4, pp.134-49. 27

Krugman, P. and Livas-Elizondo, R. (1996). Trade policy and the third world metropolis. Journal of Development Economics, vol. 49, pp.137-150. Maddala, G. (1983). Limited Dependent Variables and Qualitative Variables in Econometrics, Cambridge: Cambridge University Press. Mazumdar, D. (1981). The Urban Labor Market and Income Distribution: A study of Malaysia, USA: Oxford University Press. Mazumdar, D. (1987). Rural-urban migration in developing countries. in Mills, E.S. (ed.), Handbook of Regional and Urban Economics, vol. II, The Netherlands: Elsevier Science Publishers. Moomaw, R. and Shatler, A. (1996). 'Urbanization and economic development: A bias toward large cities. Journal of Urban Economics, vol. 40, pp.13-37. Pessino, C. (1991). Sequential migration theory and evidence from Peru. Journal of Development Economics, vol. 36, pp. 55-87. Rauch, J. (1993).'Productivity gains from geographic concentration of human capital: Evidence from the cities.' Journal of Urban Economics, vol. 34, pp. 380-400. Schultz, P. (1982). Lifetime migration within educational strata in Venezuela: Estimates of a logistic model. Economic Development and Cultural Change, vol. 30, pp. 559-593. Schwartz, A. (1976). Migration, age, and education. Journal of Political Economy, vol. 84, pp. 701-19. Speare, A. Jr. and Harris, J. (1986). Education, earnings and migration in Indonesia. Economic Development and Cultural Change, vol. 34, pp. 223-243 Todaro, M. (1976). Urban job expansion, induced migration and rising unemployment: a formulation and simplified empirical test for LDCs. Journal of Development Economics, vol. 3, pp.211-225. Tunali, I. (1996). Migration and remigration of male household heads in Turkey: 1963-1973. Economic Development and Cultural Change, vol. 45, pp. 31-67. United Nations, (1999). World Urbanization Prospects, New York: United Nations. Williamson, J. (1988). Migration and urbanization. in Chenery, H. and Srinivasan, T.N. (eds.), Handbook of Development Economics, vol. I,, The Netherlands: Elsevier Science Publishers. Yap, L. (1977). The attraction of cities: a review of migration literature. Journal of Development Economics, vol. 4, pp. 239-264. 28