Education, employment probabilities, and rural - urban migration in Tanzania

World Bank Reprint Series: Number Forty-nine H. N. Barnum and R. H. Sabot Education, Enployment Probabilities, and Rural-Urban Migration in Tanzania Reprinted from Oxford Bulletin of Economics and Statistics 39 (May 1977)  The most recent editions of Catalog of Publications, describing the full rarige of World Bank publications, and World Bank Research Progranm, describing each of the continuing research programs of the Bank, are available without charge from: The World Bank, Publications Unit, 1818 H Street, N.W., Washington, D.C. 20433 U.S. A. WORLD BANK BOOKS ABOUT DEVELOPMENT Research Publications Inter iationa! Comtparisons of Real Product and Purchasing Powerby Irving B. Kravis, Alan Heston, and Robert Summers, published by The Johns Hopkins University Press, 1978 Experiments in Fanily Planning: Lessons from the Developing World by Roberto Cuca and Catherine S. Pierce, published by The Johns Hopkins University Press, 1978 Income Distribuitioni Policy in the Developing Countries: A Case Study of Korea by Irma Adelman and Sherman Robinson, published by Stanford University Press (in the Commonwealth, Oxford University' Press), 1978 Interdependence in Planning: Multilevel Programing Studies of the Ivory Coastby Louis M. Goreux, published by The Johns Hopkins University Press, 1977 The Mining Industry and the Developing Countries by Rex Bosson and Bension Varon, published by Oxford University Press, 1977 Patterns in Household Demand and Sazing by Constantino Lluch, Alan Powell, and Ross Williams, published by Oxford University Press, 1977 Unskilled Laborfor Development: Its Econiomzic Costby Orville McDiarmid, published by The Johns Hopkins University Press, 1977 Electricity Economics: Essays and Case Studies by Ralph Turvey and Dennis Anderson, published by The Johns Hopkins University Press, 1977 Housing for Low-income Urban Families; Economics and Policy in the Developinig World by Orville F. Grimes, Jr., published by The Johns Hopkins University Press, 1976 Village Water Supply: Economics and Policy in the Developing World by Robert Saunders and Jeremy Warford, published by The Johns Hopkins University Press, 1976 Economic A7nalysis of Projects by Lyn Squire and Herman G. van der Tak, published by The Johns Hopkins University Press, 1975 Th6 D-sign of Rural Development: Lessons fromn Africa by Uma Lele, published by The Johns Hopkins University Press, 1975 Economny-Wide Models and Development Planning edited by Charles R. Blitzer, Peter B. Clark, and Lance Taylor, published by Oxford Uni.ersity Press, 1975 Patterns of Development, 1950-1970by Hollis Cheriery and Moises Syrquin with Hazel Elkington, published by Oxford University Press, 1975 A System of International Comparisons of Gross Produiict and Puorrhasibg PoWer by Irving B. Kravis, Zoltan Kenessey, Alan Heston, and Robert Summers, published by TheJohns Hopkins University, Press, 1975 Country Econ6mic Reports Commonwealth Caribbcan: The Integralion Exp.erienice by Sidney E. Chernick and others, published by The Johns Hopkins University Press, 1978 7vory Coast; The Challenge of Siuccess by BasL ian den Tuinder and others, published by The Johns Hopkins University Press, 1978 Kenya: Into the Second Decade by John Burrows and others, published by The Johns Hopkins University Press, 1975 (contiinued on inside back cover)  Volume 39 May 1977 No. 2 Reprintedfrom OXFORD BULLETIN of ECONOMICS and STATISTICS CONTENTS PAGE Earning Dispersion in Local Labour Markets: Implications for Search Behaviour K. Mayhew 93 Education, Employment Probabilities and Rural-Urban Migration in Tanzania H. N. Barnum and R. H. Sabot 109 Discriminating between MNC Subsidiaries and Indigenous Companies: A Comparative Analysis of the British Mechanical Engineenng Industry Robert F. Solomon and Keith P. D. Ingham 127 Expectations and the Term Structure of Interest Rates Elias Karakitsos 139 The Principle of Transfers and the Variance of Logarithms Johnt Creedy 152 Marketed Surplus in India: Fact and Fallacy Ashwani Saith 159 Marketed Surplus in India: Fact and Fallacy: A Reply Subrata Ghatak 169 ( Basil Blackwell, 1977 ISSN 0305-9049 EDITORIAL BOARD A. D. Ilaz7ewood, J. B. Knight, K. Vayhent BASIL BLACK WELL - OXFORD  EDUCATION, EMPLOYMENT PROBABILITIES AND RURAL-URBAN MIGRATION IN TANZANIA* by H. N. BARNUM and R. H. SABOT In most LDC's annual increments to the urban labour force have exceeded increases in wage employment despite a secular rie in the proportion of investment allocated to urban areas and consequent high rates of growth of industrial output. Unemployment rates in excess of the level considered tolerable in industrialized countries are a chronic problem and in absolute terms, if not as a proportion of the urban labour force, the numbers of urban unemployed are steadily increasing.1 The remedial policies adopted by govemments do not reflect an understanding of the underlying causal mechanism, and in some instances their effects appear to have been perverse.2 Recently a class of models has been developed to explain the coexistence of high levels of urban unemployment and rural-urban mirrration. These models are characterized by an inflexible wage differential between the rural and urban sectors and by intersectoral labour transfers that continue until there is equality between the expected urban wage-defined as the product of the probability of being employed and the prevailing urban wage-and the rural wage.3 Far-reaching implications for policy can be deduced. In particular, attention has been directed at the critical importance of wage policies and rural development policies, and it has been shown that accelerated employment creation in urban areas may increase levels of unemployment. At the core of these models is a migration function with urban employment probability as well as rural and urban wages as determining variables. However, the enmpirical validity of job probability as a determinant of migration in a developing country has yet to be established. In this paper we estimate an urban migration function for Tanzania, where there is a significant level of urban unemployment (approximately 10 per cent) and an urban sector in which wages are downwardly inflexible. While migration contributed little more thain I per cent per year to the rate of growth of urban areas in the period 1900-48, in the period 1948-71 its contribution was roughly 75 per cent of total growth, or 4.5 per cent per year. In 1971 over two-thirds of the population in the seven towns covered by this study were migrants from rural areas.4 Tests are conducted by means of an analysis of variance in a multiple regression * We are grateful for financial sup2ort from the Development Centre of the OECD and the Rockefeller Foundation and for advice and criticism from participants in seminars at the third World Congress of the Economic Society, Toronto, and the Annual Meetings of the Populations Association of North America, Montreal where we presented earlier versions of this paper. I Turnham; Sabolo. 2 Harris and Todaro (1968). a The basic ideas regarding the relationship between migration and urban unemployment were first formalized in Todaro. Also see Frank, Green and Wellisz. The model has been ex- ten(led to allow for the analysis of welfare implications (Harris and Todaro, 1970), for dynamic considerations (Lal, Stiglitz), and for intersectoral capital flows (Corden and Findlay). 4 Sabot.  110 BULLETIN framework using data from a 1971 household survey of seven urban areas especially designed for the purpose. They lend confirmation to the hypotheses that urban job probability and rural-urban income differentials are significant explanatory variables. In addition we examine the influence of education on the pro)pensit) of rural residents to migrate. Education selectivity is a freq&iently observed characteristic of urban - :.'ant streams and, at least in Africa whiere the annual number of school leavei. . as only recently exceeded the increase in urban em- ployment opportunities, education has been linked with the eniergence of clhronic urban unemployment." However, whether education's influence on the nmobilitY decisions of rural residents is due to wider differentials in economic opportunities or to greater responsiveness has remained an open question. Of course these hypotheses are not mutually exclusive and our findings lend confirmation to both. Section I briefly sets forth a conceptual framework in which migrants respond to differences in expected incomes in labour markets where jobs are ratioiied by means other than the price mechanism. In Section II estimates of the migration function are presented together with the tests of the hypotheses. Section III summarizes our findings and briefly discusses their implications for policy. Tk-e Appendix discusses the sources of data for this study which permit the analysis to focus specifically on rural-urban migration (as opposed to inter-regional migra- tion) and allow a more disaggregated view of urban migration than has preN iou.ly been possible. I. CONCEPTUAL BACKGROUND Our analysis is conducted in the spirit of the human capital paradigm.6 We assume that an individual locaces himself spatially so as to maximize the present value of the stream of returns he enjoys, the economic component of which is the difference in the discounted present values of expected incomes between source and receiving areas, net of the direct costs of moving. It follows that the size, demo- graphic characteristics and directions of movement of the migrant stream are the result of rational decision making and not the result of random selection or deci- sions not taken by the individual migrant. A consideration of this assumption has led us to exclude from the analysis women and children, a high proportion of whom are not, in the Tanzanian context, autonomous decision makers. The basic 'investment in human capital' view of spatial mobility is easily extended to explain the coexistence of rural-urban migration and significant levels of open unemployment and marginal employment in urban areas.7 If institutional factors or strong wage-productivity relationships maintain urban wages at a level above the equilibrium supply price of labour, then urban surplus labour becomes the equilibrating factor in intersectoral labour allocation. For soine rural residents, the expected net economic returns to migration will be positive even if there is a period immediately a .er entering the urban labour force when no income is received C Caldwell, Byerlee. 6 See Sjaastad and Bowles. 7 See Green; Harris and Todaro (1970); Todaro,  liDL'( ATION, ENIPLOYMENT PROBABILITIES AND MIGRATION 111 or an income below what they would have received in the rural areas. The hypothesis is that in situations in which there is a positive rural-urban income (lifferential net of the direct and psychic costs of migration, and in which urban inccmes are inflexible in a downward direction, an additional factor in migration decisions will lie the probability of finding a job, UTsing subscripts r and u to designate rural region of origin and urban destina- tion area respectively, and to indicate the time period, we express the migration relationship as the following function, continuously increasing withl WU and UP and decreasing with WR and DIST between the boundaries of zero and one, MFP,WU~ WRr M,=F (P., N'- 'usE(1 +yI, DISTru, UPu, Eru) (1) wliere M is the propensity to migrate, P is the probability of finding employment, NVU is urban income, WR is rural income and i is the applicable discount rate.8 The manner in which job probability enters the migration function is deliberately left unspecified in equation (1). In most of the regressions in the empirical section wve have entered probability and wages jointly as expected wages, thereby assuming that migrants are risk-neutral. We have also entered probability as a separate variable. Though no increase in the explanatory power of the model results from this change in specification, no firm conclusion regarding the mannler in which considerations of risk influence migration decisions can be drawn. Precise tests of hypotheses on the relationship between risk and the propensity to migrate would require additional information. DIST is the geograplhic distance between source and receiving areas; it serves as a proxy measure of the cost of transport, the efficiency of communication net- works, and the level of psychic costs of migration. The greater the distance the greater the psychic costs are likely to be as differences between source and receiving areas of language, dress, food and social practices tend to be accentuated the further the migra,rt ..av-Cks from his home area. Time lags and inaccuracies in information regarding the urban labour market, and consequently the level of uncertainty involved in the decision to move, are also likelv to increase with distance. A study by Schwartz concludes that in the United States distance is primarily a surrogate for information effects rather than psychic costs. The relatively greater variation of social characteristics among regions in Tanzania than in the United States suggests that the opposite may be the case in Tanzania. UP is the population of the urban receiving area. It serves as a proxy measure of labour market y9e, the strength of rural-urban contacts and the level of urban amenities. Stronger contacts with family and friends between rural regions and large relativ,e to small towns would be expected to reduce psychic costs, to reduce the risk component of migration costs as a consequence of better communication, and to reduce the job search component of costs as a consequence of greater availability of short-term support on arrival in town. Higher levels of urban anmenities per capita in large cities would be expected to increase real and psychic 8 See appendix for definitions of variables.  112 BLILLETIN returns of migration. E is an error term inicluded to capture effects on the nligra- tion rate arising from imprecision in specification anid measurement and is as ulned to be orthogonal to the independent variables and to have zero mean arnd uniform variance. The discounted value of the rural and urban ,incomv streams will be highly sensitive to small differences in the choice of discount rate (i) or time liorizon (T) when the level of the former is low and that of the latter is high. Precic iilforma- tion on the level of these components of the model is not available. Thus the choice of discounting parameters may introduce an arbitrary influence on the results. An alternative is to reduce the discounted income variable to a single, u.,discounted, average value in source and receiving areas. But this entails assuln- ing either that the time horizon is unlimited and that income differentials and dis- count rates are constant over time, or that the appropriate time horizon is a period that is sufficiently short that within it significant changes in income differentials and discount rates are not likely to occur. Either the discounted or the undiscounted income variable is likely to entail a degree of misspecification. In Tanzania the rate of increase of urban wages due to secular increases in the wage level and increases associated with seniority on the job, is likely to exceed rural inconoe increases over the course of an individual's working life. Thus, the former assumption is not strictly valid. In regard to the latter, although it would be convenient analytically to assume, as Todaro dlues, that potential migrants have a short time horizon, the evidence suggests that this is not the case in Tanzania. During the colonial period migration was circular, and rural males left their families behind while they participated in the wage sector of the economy. In these circumstances the migratory journey was short term as was, it could be assumed, the time horizon of the migrants. Today, with higler urban wages and the stabilization of the labour force, migrants who are successful in finding urban employment tend to bring their families to town and remain in town for the remainder of their working lives. A long time horizon implies both that the second assumption is not entirely accurate and that small errors in the estima- tion of T will have a significant impact on the discounted value of income streams. Although only the undiscounted version is presented below, we have also estimated the migration function in the discounted form and found that the results are highlY consistent. A second specification problem concerns the probability variable which should be delined as the ratio of the number of non-marginal job openings in the job search period to total urban surplus labour.9 Data on uner'oploymcnt but not on em- ployed surplus labour, i.e., employed urban workers who earn less than they would in rural areas, are available for years prior to 1971.10 The omission of einploxyed 9 The appropriate definition of probability can vary depending on assumptions regazrling changes over tini in the probability of an individual getting a job once he has arrived in towvn. We assume that the probability remains constant over the job search period. For a discussion of alternative definitions of employment probability see Todaro, Stiglitz, and Zarembka. 10 The allecative efficiency justification for rural marginal produtict as the income criterion for distinguishing between urban workers who are fully cmployed1 and the nideremployed is in Sabot (1975).  EDUCATION, EMPLOYMENT PROBABILITIES AND MIGRATION 113 surplus labour from the migration function results in an upward bias in the estima- ted probabilities. The significance of the coefficient on the probability variable will be affected only if the ratio of employed surplus labour to unemployment varies markedly over time, among towns and educational sub-groups, i.e. the dimensions by which the model is disaggregated. However, an examination of the data for 1971 revealed that the ratio of employed surplus labour to unemployment does not vary significantly among urban sub-groups. In urban areas education level, occupation level and the wage rate are all posi- tively correlated. When moving from high to low skilled occupations there is a shift from labour scarcity to labour surplus, which s -..,gests a positive relationlship between education anid urban employment probabilities. In addition, differences among education groups in psychic costs and returns as well as differences in information may influence migrant behaviour. The educational systeem may be selective of people with utility functions consistently different from those of the rest of the population. It is also conceivable that there is a systematic decrease over the course of an individual's formal education in place attachments and an increased preference for urban occupations and for those urban consumer items not found, or considerably more expensiv e, in the rural basket of goods and ser- vices. If so, this would explain higher levels of inohilit among the educated even in a situation in which expected income differentials are the same for all educational groups. Disaggregating the model by educational sub-groups allows us to assess the relative strength of alternative economic and non-economic explanations of the clear association between the education level and the propensity to migrate of rural residents." To define cross-section migration rates as ratios of total current urban stocks of migrants to current total storks of rural population in the relevant sub-groups is likely to introduce a bias along the time series dimension implicit in the model. The current stock of migrants arrived in different years. If rural-urban economic opportunity differentials changed over time, as they have in Tanzania, then the estimated model will be biased as net economic returns to migration vary among current age groups. When only contemrporary data on the economic variables are available, only recent migrants should be included in the numerator of the migration rate.12 A preferable procedure, adopted in this study, for minimizing 11 While Levy and Wadycki, and Bowles have incorporated education into their standard regional models of migrant behavior in this preferable manner, several existing econometric studies, e.g. Beals et al, Sahota, Schultz, introduce the average level of education of the source area population as an independent variable. A significant asssociation between regional average educational levels and migration rates is not sufficient to confirm that the educated have a higher propensity to move to town than the uneducated. The relationship may be due to a relatively higher level of mobility among all members of the regional population, educated and uneducated alike, suggestinig that education is serv,ing as a proxv for an unspecified independ- ent variable with which it is highly correlated. Even if it is established indepenlentlv that the educated have a higher miigration propensity there is noi way of dcetermining whether it is due predominantly to a higher level of responsiveness to a given ruiral-urban expected income differential or to wider income differentials for the educated than the unedlucated. While in four of the studies the sign of the education variable was as anticipated, Beals, et. al. obtained a negative coefficient. After a consideration of alternative explanations lKnight concludes that no helpful implications can be drawn from this study concerning the effects of education on migration. 12 Examples of one of the problems that arise when this potential bias is ignored are found  114 131tL.ETIN the bias is to interpret age-specific migration rates from a time series perspective and to disaggregate the independent variables by time periods. The specification adjustments yield equation (2) where subscript e indlicates education and a indicates time period. Mruea =F(P,ea, WUuea, WRrea, DIST,u, UP, R)ue,). (2) II. EMPIRICAL RESULTS The migration function is estimated using the data described in the Appendix. Tables lA, IB, and IC present the rural-urban migration rate, urban wvage rate and urban employment probability disaggregated by three times periods and four education levels. The data set used for estimation is also disaggregated into three rural regions and three urban areas; there are in all 108 observations. Comparing the da"a aggregated over time periods, there is a clear positive association between educational level and migration rates, on the one hand, and lbetween educational level and urban wage rates and employment probabilities, on the other. Given the assumption that rural regional incomes and the direct costs of migration are in- variant with regard to education, the implication is that at least part of the ex- planation of the cross-section differential in the propensity to migrate of rural residents is that the economic returns to migration are anl increasing function of education. A comparison of the data aggregated over educational groups also reveals a strong relationshli) between the migration rate, urban wages and emloy- ment probabilities over time. The ratio of the monthly urban wage to the mnthily rural per capita income varies over the 'ample from 3 for the lowest educational group in the 1955-60 period to 12 for the highest educational group in the 1966-70 period."3 We use linear regressions to assess the evidence bearing on the hypotheses regarding cross-section and inter-temporal differentials in income anid emplo nwment probabilities as factors explaining differentials in rates of urban migration. Sec- tion A examines the importance of income differential and employment probability variables in the migration function with the data pooled over all educational groups. Also, the role of an independent education variable as proxy -ir the economic variables is assessed. Section B presents the results obtained when the migration function is fitted for separate education groups, and assesses the non- economic influence of education on the propensity to migrate. Ordinary least squares techniques and simple linear forms of the equations were in Levy and 'Wadvcki (pp. 379-380), Sahota and Schultz. Sahota and Scultz attribute the relatively higher coefficients on the economic variables for young niig.rants to factors, such as a longer remain,ig time in remunerative activities which constitute explanations for a higher degree of responsiveness armong young rural residents. liowever, as a conseqliienc t of age selectivity, the young age groups of the urban stock of nigrants are comprised predlonm'iiantly of more recent arrivals. Since in these stu(dies age specific mi gration rates are all related to economic variables from the same year, it is possilble that a significant portinn of the (lifferentials in the coefficient is explained by a wider gap betwveen gross and net migration rates among relatively early arrivals and by a -widening of income dlifferentials over time. 13 See Barnum and Sabot for a more detailed discussion.  EDUCATION, EMPLOYMENT PROBABILITIES AND MIGRATION 115 TABLE 1 A. Rates of Rural- Urban Migration for Four Education Groups in Three Time Periods: Tanzanian Males Education Aggregate Arrival Standard Standard Form 1 over Current age period' None 1-4 5-8 and up education 20-24 1967-70 0.014 0.047 0.277 0.318 0.085 25-34 1962-66 0.013 0,038 0.144 0,399 0.047 35 and above 1955-61 0.012 0.030 0,106 0.180 0,022 Aggregated over age groups (time periods): 1955-70 0.011 0.032 0.123 0,249 0.035 'Period in which the majority of migrants in the age sub-group arrived in town. B. Average Urban Wages (Shillings per Month) for Four Education Groups itn Three Time Periods: Tanzanian Males Education Aggregated Standard Standard Form 1 over 72ime period None 1-4 5-9 antd up education 1966-70 230 246 322 448 288 1960-66 179 194 247 336 213 1955-60 106 121 150 200 125 Aggregated over time periods: 1955-70 181 198 253 347 221 C. Average Probabilities of Finding an Urban Job (Within Four Months) for Four Education Groups in Three Time Periods: Tanzanian Males Education Aggregated Standard Standard Form 1 over Time period None 1-4 5-8 and up education 1966-70 0.10 0.19 0.47 1,0 0,33 1960-66 0,16 0.21 0.71 1.0 0.34 1955-60 0,12 0.22 0.78 1.0 0.25 Aggregated over time periods: 1955-70 0.14 0.20 0.56 1.0 0.30 employed in all regressions.14 Except where specifically noted, the first figure below the coefficient estimate in the tables is the standard error and the second figure is the elasticity calculated at mean values (see Table 2), of the variables. An asterisk indicates that the coefficient is significant at the five percent level using a one-tailed t test. TABLE 2 Mean Values" Variable M WU WR P P.WU-WR DIST UP Mean 0.437 231.1 34.6 0.419 81.7 67.6 11.4 ' The migration rate has been rescaled by a factor of ten (actual, .0437), WU and WR are in shillings per month, observations on P lie in the interval between zero and one, DIST is an index based on a map scale, UP is in tens of thousands. 14 An analysis of the residuals ranked by source area population, Rr,a, did not reveal marked heteroskedasticity. Regressions in which the observations of all variables were divided by V Rra, and /Rr., was also included as an additional variable, are fully consistent with the results reported below.  116 BULLETIN A. THE MIGRATION FUNCTIO:; USING POOLED DATA The wage variables are introduced into the migration function in four different forms. Rural and urban wages, WU and WR, are entered separately or as an absolute differential, WU-WR.15 The results are presented in Table 3. The estimated coefficients of all the wage variables have the correct sign and, with the exception of WR in regression 1, the coefficients are significant at a 5 per cent level. TABLE 3 Regression Explaining M:gra'ion Rates -Including Wages as Explanatory Variables (108 Observations-3 x 3 x 4 x 3) TJNDISCO1JNTED WAGES Regression No. Constant VIti WR (WIT-WR) DIST UP R2 SSR' 1 0.15 0.0040' -0.0106 -0.0085* 0.027* 0.48 22.621 (0.40) (0.0005) (0.0067 (0,0039) (0.004) (2.09) (0.83) (-1.31) (0.71) 2 -0.14 (0.26) 0.0039* -0,0073* 0.027* 0.49 22.836 (0.0005) (0.0037) (0.004) (1.76) (-1.13) (0,70) aSSR is the sum of the squares of the estimated residuals. The total sumn of the squares for all regressions in tables 2 through 5 is 45,06. The urban probability variable is first added to the migration function as a component of the expected wage, calculated as the product of job probability and the urban wage. This form of the equation, in which the elasticities of the two variables are constrained to be the same, would be appropriate if potential migrants are risk-neutral and evaluate their income possibilities in terms of a simple actuar- ial value. Migrants may respond differently to a wage differential when there is a risk of unemployment than to the same differential when there is assurance of finding a job. Therefore the wage and probability variables are also entered separately. Table 4 presents the resulting regressiGns. In both forms of the T1ABLE 4 Migration Regressions Including Employment Probabilily UND3SCOUNTED WAGES (1) EXPECTED WAGES Regression coefficiets Regression No. Constant D(WL').WR DIST UP R2 SSR 3 0.43 0.0030' -0.0074 0.0220 0.55 19.820 (0.23) (0.0003) (0.0035) (0.004) (0.57) (-i.14) (0,58) (2) PROBABILITY ENTERED SEPARATELY Regression No. Constant P Wl7 WR WU-WR DIST UP1l R2 Sr,R 4 0.11 0.666* 0.0024' -0.0077 -0.0077* 0.023' 0.55 19.429 (0.3 7) (0.163) (0.0006) (0.0063) (0.0036) (0.004) (0.64) (1.26) (-0.56) (-1.20) (0.60) 5 -0.09 0.676* 0.00230 -0.0069* 0,023' 0.55 19.534 (0.24) (0.162) (0).0(0)06) (0.0035) (0.004) (0.65) (1,05) (-1.07) (0.59) 15 A relative wage ratio, WUIWRXVR was also tried and gave regressions consistent with those reported below in which an absolute wage differential was employed.  P.1)(UATION, EMPLOYMENT PRO)BABI.LITIES ANI) MI;RA1ION 117 regression where the probability variable is entered separately (regressions 4 and 5), its coefficient is significant at above a 5 per cent level. Similarly the coefficient of the expected wage variable (equation 3) is significant at a 5 per cent level, Comparing equations 1 and 4 or 2 and 5, we find that the addition to the explained sum of squares in moving from the specification without probability to the speci- fication including probability as a separate variable is significant at a 99 per cent confidence level. We suggested that disaggregating both the dopendent azn(l itndependIent variablcs by education will increase the explanatory power of the ecoitomic variables relative to regressions in whiclh the educational level is entered as an independent variable. If the significance of the independent education variables derives at least in part from the fact that they serve as surrogates for rural-urban wage (lifferentials by education and educational differentials in urban employment prohahility, then the significance of dummy shift variables representing each of the four educational groups, would be expected to decrease with the introduction of the wage or probability variables. Table 5 reveals that this is what occurs, The F statistic TABIE 5 Migration Regrcsion. uith Fducailon ntgered as S .