Determinants of Urban Labour Earnings in Tanzania,

Transcription

1 Determinants of Urban Labour Earnings in Tanzania, by Vincent Leyaro, Priscilla Twumasi Baffour, Oliver Morrissey and Trudy Owens University of Dar es Salaam (Leyaro) and University of Nottingham Abstract The working age population in Dar-es-Salaam (DSM) has grown much faster than the rate of formal job creation so the informal sector accounts for an increasing share of employment but with low and irregular earnings compared to the formal sector. Making use of the Tanzania Integrated Labour Force Survey (ILFS) for 2000/01 and 2006, complemented by the Urban Household Worker Survey (UHWS) for 2004, 2005 and 2006, the main aim of this paper is to identify core features of urban labour markets, in particular the determinants of earnings and selection into sector of employment (public, private and informal). In a later stage this analysis will be extended to draw inferences for labour market dynamics and current policy. Preliminary investigation of the UHWS show that public sector workers (who tend to be more educated with longer tenure) earn more on average than workers in other sectors; among private wage employees, those in large firms earn considerably more than those in small firms; average earnings are lowest in the informal sector, but with significant variation (a small group of entrepreneurs have high earnings). There has been a general decline in real earnings from 2004 to 2006, but this is largely driven by a 25 per cent decrease in average real informal sector earnings. These results to a certain extent are corroborated using the larger but less detailed ILFS data. Although paid workers (central government and parastatal organisation) earns on average more than most of those in other economic activities, preliminary findings from ILFS revealed that self employed with employee (indicating a medium to large size enterprise or firm) earns the highest relatively to the rest. Engaging in farming activities in suburb of Dar es Salaam does not pay, as those working on own or family farm have low income with insignificant negative coefficients with a very low returns. The same is the case for unpaid work in family business or unpaid family helper (agricultural and non-agricultural). JEL Classifications: J6, J62, J69 Keywords: Labour Earnings, Urban Labour, Tanzania Preliminary paper for Labour Market Dynamics in Times of Crisis in Africa, Project Workshop, University of Oxford, 17 March

2 1 Introduction Rapid urban population growth in Dar-es-Salaam (DSM) provides a useful example of the challenge of creating employment for the increasing labour force in urban centres in Africa, where the informal sector accounts for an increasing share of employment with lower earnings compared to the formal sector. This has occurred against a background of high national economic growth rates which has core features of a growth without jobs problem: labour is attracted to urban areas because of increasing economic activity (suggesting a vibrant informal sector) rather than because of growing employment demand from the formal sector. The result is greater stress on the informal sector effective unemployment rises and/or average earnings fall to accommodate the growing labour supply. The aim of the paper is to analyse labour market dynamics (earnings and employment) in Dar-es-Salaam during the period from 2000/01 to In developing countries, urban labour markets are generally recognised as having two distinct sectors, a regulated/protected formal sector and an unregulated /unprotected informal sector (Pradhan and van Soest, 1995). The informal sector seems to absorb many job seekers who are unable to secure employment in the stagnant formal sector of employment. In Tanzania, the self-employed outnumber wage employees by almost twice in the urban labour market and it is the fastest growing segment of the labour force across rural and urban areas, typical of most developing nations particularly in Africa. Among the enormous challenges that face governments in developing countries including Tanzania is the need to identify development strategies that can generate new employment and income opportunities to reduce unemployment and under-employment. An understanding of earnings determination in the informal sector is as a consequence vital to understanding the labour market and income determination/distribution in the country. Studies which analyse the determinants of earnings typically use the Mincerian earnings model to estimate the mean effect of schooling and other individual characteristic variables on earnings. In Tanzania, these studies include Soderbom, Teal, Wambugu and Kahyarara (2005), Quinn and Teal (2008), Rankin, Sandefur and Teal (2010). Evidence from developed countries (Bushnisky 1994; Fitzenberger and Kurz 1998, Machado and Mata 2000) indicates that returns to skills and experience in the Mincerian model can differ across the earnings distribution, implying the usefulness of estimating determinants of earnings across quantiles of the conditional earnings distribution. We examine whether real earnings and private returns to education for urban workers in Tanzania are identical for low and high earners. Unlike other studies that applied quantile methods in Africa, such as Mwabu and 2

3 Schultz (1996) and Nielsen and Rosholm (2001), this paper addresses endogeneity and measurement error bias. The Tanzania Integrated Labour Force Survey (ILFS), for 2000/01 and 2006, and the Urban Household Worker Survey (UHWS) for 2004, 2005 and 2006 provide information on earnings for the informal sector (self-employed) formal sectors (public and private) with worker characteristics (such as age, education, tenure). This allows for comparison across sectors to shed more light on the importance of heterogeneity in earnings determination. The UHWS has a modest sample size and a small panel element, and includes some useful information such as parent s education and job history. The ILFS has a larger sample, but no panel element, not all the same variables (e.g. no parent s education) and there are some issues with the reported incomes in Using both surveys permits extended analysis to check robustness of the main findings. The structure of the paper is as follows. Section 2 provides a brief literature review, concentrating on studies for Tanzania. Section 3 briefly outlines the specification and econometric methods. Section 4 provides data sources and descriptive statistics for both surveys (comparing ILFS with the UHWS). Section 5 presents results for ILFS comparing with the UHWS results. Section 6 concludes and outlines the next stage of the research. 2 Literature Review Soderbom, Teal, Wambugu and Kahyarara (2006) used repeated cross-section surveys for Tanzanian and Kenyan manufacturing sectors; the data for Kenya covered and 2000 and the data for Tanzania covered and The study used the control function approach to control for endogenous education by instruments. Results generally showed a pattern of upward biased OLS estimates contrary to the more recent studies reviewed for mainly developed countries. The conclusion from the study was contrary to the conventional view of concavity between earnings and education: the marginal return to education is found to increase with increased education in both Tanzania and Kenya, an indication of a convex earnings function with education. Quinn and Teal (2008) used three rounds (2004, 2005 and 2006) of the UHWS to examine determinants of earnings and earnings growth, pooling all three rounds for an OLS estimation of earnings equation. Results indicate a significant convex effect of education on earnings, with substantial heterogeneity between and within sectors. These results are shown 3

4 to be robust to control of endogenous education with instruments (parent s education and occupation) and a Raven s Progressive Matrices Score to control for unobserved ability. Rankin, Sandefur and Teal (2010) pool the 2004 and 2005 rounds of the UHWS to investigate how the role of formal education and time spent in the labour market explain labour market outcomes of urban workers (using a similar survey for Ghana to compare with Tanzania). The study adopted the standard Mincerian earnings function and controlled for endogenous education with instruments (including distance to the nearest primary school at age 6, distance to nearest secondary at age 16, mother s education, father s education and dummy variables to indicate whether mother and/or father had a formal sector employment) in four models (self employment, public sector and private sector which was further categorised by number of employees into small and large firms). After controlling for selection bias using the methods of Lee (1983) and Durbin and McFaddden (1984), the paper concludes there are convex returns to education in self employment but concave returns to education in large firms, no significant effect of education in the public sector. Studies that have utilised quantile regression method within the Mincerian framework include Buchinsky (1994), who analysed the U.S. wage structure by using Current Population Survey (CPS) data. Buchinsky proves the returns to education in the U.S increase considerably over the quantiles of the conditional distribution of wages. Fitzenberger and Kurz (1998), applied quantile regression in a study of earnings in Germany and Machado and Mata (2000) in a study of wage inequality in Portugal, all found varying returns across quantiles. Mwabu and Schultz (1996), in a similar manner used quantile regression on a sample of South African men and obtained varying returns across quantiles. Nielsen and Rosholm (2001), applied quantile method on a Zambian data set and obtained similar results. 3 Empirical Methods The basic assumption is that an individual s earnings reflect its labour productivity and that investment in human capital in the form of foregone earnings in the past pays off in the form of higher wages in the future (Card 1998). Mincer (1974) provides the theoretical model from which the following wage equation is derived: log w X S x x u i 2 0 i 1 i 2 i 3 i i... 1 Where w i is an earnings measure for individual i such as earnings per hour, week or month. S i represents a measure of schooling, this proxy s human capital acquired through formal education, x i is an experience measure (typically age the age an individual left 4

5 school), this captures human capital acquired on-the-job. X i is a set of other variables assumed to affect earnings and u i is the disturbance term which captures all factors other than schooling and labour market experience that affect individual wages. The error term is assumed to be normally distributed and uncorrelated with the human capital variables as well as between individuals and across time in panel data analysis. Experience is included as a quadratic term to capture the concavity of the earnings profile. The derivation of the empirical model by Mincer implies that, under the assumptions made (particularly of no tuition cost). β 1 can be considered the private financial return to schooling as well as the proportionate effect on wages of an increment in S. To control for unobserved ability in the earnings equation, we adopt a two-stage control function approach. Card (2001) discusses the control function and shows the control function is more robust than the 2SLS where slope parameters potentially co-vary with the unobserved factors in the model. Even with constant slope parameters 2SLS will result in relatively imprecise parameter estimates given that the model is non-linear in the endogenous variable. Consequently, at the first stage, a regression of education on a set of instruments) is run and the residuals are estimated. This residual which captures all unobserved factors that affect education are saved and used in a second stage earnings regression as a control variable for ability. This procedure produces consistent parameter estimates, when conditions for identification and independence of instrumental variables are met 1. This requires valid exclusion restrictions (variables correlated with schooling but uncorrelated with the earnings residual). The UHWS provides information on the distance in kilometres to the nearest primary school at age six and to secondary school at age sixteen of respondents in addition to parent s education, and occupations. Such supply side measures of education like distance to school can be reasonably argued to be correlated with education and not with ability (Card, 2001). Many studies on earning have used such supply side measures of education as instruments for education, while others have similarly used family background variables as instruments for education. While OLS captures the effect of education on an individual on the mean earnings, quantile regression looks at the returns at some other parts of the earnings distribution for example bottom or top quartile. In essence, the focus is the quantile treatment effects of education on earnings rather than on the average treatment effect, and this adds value to estimation results. 1 Blundell and Powell (2001) show zero covariance between z i and δ i is generally not sufficient for consistency given that the model is non-linear in the endogenous explanatory variable. 5

6 Given a set of explanatory variables, quantile regression estimates the dependent variable conditional on the selected quantile. The estimation of the model at different quantiles enables us to trace the entire conditional distribution of earnings given a set of regressors. Afterwards, comparing the estimated returns across the whole earnings distribution, we can infer the extent to which education exacerbates or reduces underlying inequalities. Particularly, how schooling, individual characteristic, sector of employment and firm size affect earnings differently at different points of the conditional distribution of earnings. An additional advantage of employing this estimation method is that the regression coefficient vector is not sensitive to outlying values of the dependent variable, as the quantile regression objective function is a weighted sum of absolute deviations. Provided error terms are homoscedastic, Koenker and Bassett (1982) and Rogers (1992), this method would be adequate to calculate the variance covariance matrix. Rogers (1992), shows in the presence of heteroscedastic errors, this method understates the standard errors. We consequently use bootstrapped estimator of standard errors as suggested by Roger to cater for any such under estimated standard errors. This method however requires that, there is adequate dispersion of the independent variables over the earnings distribution to enable identification of coefficients for each quartile (decile). The Tanzanian surveys appear satisfactory in this regard. Quantile regression uses the whole sample available and allows us to estimate the return to education within different quantiles of the earnings distribution (Buchinsky, 1994), so we can trace the entire conditional distribution of earnings given a set of regressors. Then one can compare the estimated returns across the whole earnings distribution and infer the extent to which education exacerbates or reduces underlying inequalities. Quantile regression should also reveal if gender or tenure earnings differentials vary according to the position in the earnings distribution. This facilitates a richer exploration of the survey data. Given that labour force data are truncated on the basis of the wage/earnings variable due to self-selection into the various employment sectors, allocation into employment sectors and the resultant earnings is not entirely random (implying biased OLS estimates). To correct for selectivity in the earnings model we adopt the Heckman two-stage procedure, and first estimate a probit model for the probability of sorting into the various sectors relative to being unemployed to estimate the inverse Mills ratios (selection terms), which are then inserted in the earnings equations at the second stage. Covariates in the probit model include education, age, sex, marital status and a dummy variable for whether the individual has children or not to fulfil the exclusion restriction. 6

7 4 Data Sources and Description This study draws on two principal data sources that are to be complimented with two more data sets. The first is the Integrate Labour Force Survey (ILFS), available for 2000/01 and The second is the Dar es Salaam (DSM) Urban Household Worker Survey (UHWS) for 2004, 2005 and We use the two surveys on DSM to assess if they yield comparable inferences on determinants and evolution of sector income and employment share as an attempt to analyse labour markets dynamics and draw inference for current policy. As there are issues with measurement of income in ILFS, will use Household Budget Surveys (HBS) for 2001 and 2007 data for DSM sample to construct cohorts (i.e. households by district and main source of income) to match an expenditure-based measure of income to the ILFS. One problem with these data sets is that they predates the real financial and economic crisis as they end up to 2007, base on availability, we will extend the analysis by using the most recent survey, the National Panel Survey for 2008/09 and 2010/11 to draw inferences on what happened. 4.1 Integrated Labour Forces Survey The ILFS are nationally representative surveys conducted by the National Bureau of Statistics base on National Master Sample (NMS) for Mainland Tanzania, covering individuals in private households. They provide detail information on labour market characteristics including sectors of employment income (wages and earnings) and sectors composition of employment, sufficient to answer questions on determinants of workers earnings and workers mobility between sectors. The information provided range widely, including: demographics and gender; migration; education; training; health; economic activities (wage/paid employment, self employment, own family business, farm, unpaid family, etc); main occupation; main sector of employment; main industry; main business; status of employment; sources of income (wage/salaries, earnings, remittances, pension, rent, interest, dividend, etc); average household monthly cash income from all sources; paid monthly income; self employed income; amenities and community services (type of house, assets owned, sources of energy/fuel, distance to water, school, health centres, markets, type of transportation, land). The 2000/01 LFS is the third comprehensive survey of its kind since independence and covers the general labour force, child labour and informal sector. The data classify the labour force according to the four main domains rural areas, urban areas, Dar es Salaam 7

8 city and other urban. The rural component of the sample consists of 100 villages, while for the urban component, 122 Enumeration Areas (EAs) were used based on the 1988 Population Census. The data was collected from 3,660 households (or 30 households per enumeration area) in urban areas and 8,000 households (or 80 households in each village) in rural areas, thus making a total of 11,660 households; of these the DSM sample is 1,497 households. In the 2006 ILFS (the fourth comprehensive survey) covered 16,445 households, with 6,107 households from urban areas (244 enumeration areas with 30 households selected in each); of these DSM sample is 1,836 households. The selected households were distributed more or less equally across the four quarters of the year for the purpose of measuring seasonality, and pooled 2000/01 and 2006 for DSM sample make a total of 3,333 households. As the main objective of this study is to identify core features of the labour market in DSM in the 2000s, the focus of this section is to get feel of which sectors appear to be growing in terms of size (share of employment or as sources of income); how have incomes (average and dispersion) evolved in those sectors; and what determines worker earnings and mobility across sectors. Thus understanding how the variables are measured and their trend between 2001 and 2006 is important. Appendix Table A1 gives the summary statistics of the variables used in the estimation and one such key variable is the measure of income. In ILFS income is measured in four ways: as monthly paid employment income; as monthly or weekly self employment gross income; as monthly or weekly self employment net income, and average household monthly cash income; the sum of paid and self employment income makes total household income. From the summary statistics Table the mean self employment income is twice the paid employment income (might be because the self employed are more than paid employment). Age that captures work experience is measured as number of years of individuals 15 years and above, and its square. Education is measured in two ways, as a continuous variable the highest number of education years attained and its square, and as categorical variable, that is: those with no or pre-school education, primary, secondary, post secondary and tertiary education. As there is no a tenure variable in ILFS data, we proxy that by using two variables: migration and type of training. Migration is measured as duration of residence: since birth equal to 1, less than one years equal 2, one to three years equal to 3, three to five years equal to 4 and five years and above equal to 5. Training is measured as type of training, which are in three categories: none (1); 8

9 on the job training (2) and other trainings (3). 2 Most people, around 70% have no on job training. Table 1: Demographic, Education and Training Variables, 2000/ Percent 2000/ Earnings in TShs (%) Percent Earnings in TShs (%) Income Ratio Sex Male ,990 (79%) ,516 (73%) 1.58 Female ,661 (21%) ,806(27%) 2.15 Age Youth (13%) (25%) 2.65 Adult (69%) (65%) 1.34 Aged (18%) (10%) 0.82 Marital status Single (28%) (30%) 1.14 Married (30%) (70%) 2.32 Widowed (24%) Divorce (19%) Education No or Pre-school (0%) (6%) Primary (6%) (10% 2.51 Secondary (20%) (14%) 1.01 Advanced ndary (34%) (34%) 1.49 Tertiary (40%) (37) 1.36 Training None (13%) (18%) 2.13 On job training (31%) (30%) 1.45 Other training (56%) (51%) 1.38 Source: Authors own calculations from the Tanzania Integrated Labour Force Surveys for 2000/01 and 2006 years Table 1 looks at education, training and demographic variables that capture labour market characteristics in terms of share and income size. There is almost the same number of male as female, 50 percent each, both in 2000/01 and 2006, but a huge difference in terms of income share, with male accounting around 75 percent of the income. For both male and female the income has increased, more for female than male. Youth (age years) make the big share of population in Dar es Salam, 64% in 2000/01 and 62% in 2006, followed by Adult (35-64 years) around 32% and aged (65 years and above) around 4%. However, it is the Youth who earn the least, accounting only for 13% of income in 2000/ None, 02-On the job, 03-Certificate 1 or less than 2 years, 04-Certificate 2 or more years, 05-Formal Apprenticeship, 06-Informal Apprenticeship, 07-Diploma 2 or more years, 08-University,09-Other Courses after University, 10-Other Courses, 99 Not stated. However after cleaning, what is left is 1, 2 and 3. 9

10 compared to 69% for Adult, indicating a huge inequality. While the income of both Youth and Adult has been rising between 2000/01 and 2006, those of Aged people have been falling. People with primary education are the majority, accounting for 69% in 2000/01 and 71% in 2006, followed by those with secondary education around 23%, advanced secondary education 5% and those with tertiary account for only 1%. However, the opposite is the case when it comes to income share, as shown, the higher the level of the education the higher the earnings, with tertiary accounting for the largest share. We decompose the data further to try to get the feel of which sectors appear to be growing both in share of employment and size of income, between 2000/01 and Table 2 starts by looking at the main economic activities. The main economic activity for majority of people in Dar es Salaam is in private sector, both formal and informal activities, as it accounted for 70% in 2000/01 and 56% in 2006, with self employed in a business without employee being many, followed by those employed in private sector and then self employed in a business with employee. While the share of those working on own or family farm has remained constant, that s 7.6% in 2000/01 and 7.9% in 2006 (indicating there are few farmers in suburb of Dar es Salaam), the unpaid work in family business has increased tremendously, from 5% in 2000/01 in 2000/01 to 28.7% in 2006 (that might be the indication of migration of youth from rural to urban areas who willing to stay in extended family unemployed and unpaid). Table 2: Employment and Earnings Shares by Main Economic Activity, 2000/ Main economic activity Employment (%) 2000/ Earnings (TShs) % Change 2000/ Income Ratio Working on own or family farm ,355 19, Employee central government , , Employee parastatal organisation , , Employee political party , , Employee cooperative ,500 78, Employee NGO/religious organisation , , Employee private sector , , Self employed in a business with employee , , Self employed in a business without employee , , Unpaid work in family business ,225 5, Other ,108 36, Source: As for Table 1 10

11 For those whose main economic activity is not in the private sector, if we assume as well those working on own or family farm and the unpaid work in family business are private, make 15% in 2000/01 and 8% in Of these, the shares of employee of central government have remained nearly the same at around 6% in 2000/01 and 5% in Those employed in parastatal organisation follows, but have seen a huge fall in its share from 7% in 2000/01 to 1% in Others, except employee in NGO s and religious organisation that account for 1%, share insignificantly. This analysis is largely corroborated by what is in Table 3, where the focus is in term main sector of employment. As shown those in private sector, that is private sector self employed, private sector worked or others, and household own business account for 61% in 2000/01 and 89% in Table 4 does the same, with focus on the status of employment. Self employed with employee, self employed without employee, unpaid family helper (non-agric), unpaid family helper (agricultural), and on own farm or shamba which is considered as private account for 55% in 2000/01 and 63% in The same experience is reflected in Appendix Table A2 when the focus is on the main industry of the economy. Although most of the main industry can be taken as in private sector, those in wholesale, retail trade, restaurants and hotels, agriculture, hunting, forestry and fishing, and community, social and personal services reflect more of those in informal sector and accounted around 73% both in 2000/01 and Employment by main occupations in Table 5 reflects more of paid employment. Table 3: Employment and Earnings Shares by Main Sector, 2000/ Main sector Employment (%) 2000/ Earnings (TShs) % Change 2000/ Income Ratio central /local government , , parastatal , , political party ,167 11, partnership registered , , partnership unregistered , ngo/religious organisation , cooperative registered , , cooperative unregistered ,065 49, international organization ,000 - household ,253 12, private sector - self employed , , private sector (other) employed , , Source: As for Table 1 11

12 Table 4: Employment and Earnings Shares by Status Employment, 2000/ Status of employment Employment (%) Earnings (TShs) 2000/ % Change 2000/ Income Ratio Paid employee , , Self employed with employee , , Self employed with out employee , , Unpaid family helper (non-agric) unpaid family helper (agricultural) ,644 - On own farm or shamba ,458 15, Source: As for Table 1 Table 5: Employment and Earnings Shares by Main Occupation, 2000/ Main occupation Employment (%) 2000/ Earnings (TShs) % Change 2000/ Legislators and administrators , , Professionals , , Technicians and associates professionals , , Office clerks , , Service and shop sale workers , , Agriculture and fisheries workers ,008 27, Craft and related workers , , Plant, machine operators and assembles , , Elementary occupations ,111 74, Source: As for Table 1 Income Ratio Although share of employment between 2000/01 and 2006 have not increased significantly for most of main economic activities, main occupation, main industry or sectors and status of employment, income have increased, and some have increased significantly. As shown in Table 2, the income of individual whose main economic activity is in self employed in business without employee, employee in NGOs and religious organisation and employee of central government have seen their income increasing significantly. These coincide very much with what is in Table 4 that is the earnings share by the status of employment. If one look at sector of employment in Table 3, the income of central/local government, private sector employed and private sector self employed have almost doubled, followed by those parastatal and cooperative unregistered. The rest have seen fall in their earnings. Main occupation in Table 5 largely reflects paid employment earning and for most occupation the earning has increased significantly between 2000/01 and

13 In this section we have established that youth makes a huge share of population in Dar es Salaam just like those with only primary education, however, both of these groups earn less relatively to others. Men, who likely too to be paid employment, are by far well paid compared to women, indicating gender earnings disparities. The ILFS data have shown as well that the private sector, both formal and informal, makes the huge share of employment in Dar es Salaam, above 75%. And the private informal sector has seen huge increases in its share, relatively to other sectors. Most of other sectors, though slightly, have seen a decline in their share of employment. However the income for most sector or main economic activities has increased, particularly those of central government employee and self employed without employee. 4.2 Urban Household Workers Survey The Tanzanian Urban Household Worker Survey (UHWS) is conducted in six (6) urban areas (regions): Dar es Salaam, Arusha, Iringa, Morogoro, Mwanza and Tanga by the Centre for the Study of African Economies (CSAE) at the University of Oxford. The sample is based on a stratified random sample of urban households from the 2000/01 Tanzania Household Budget Survey (HBS). The survey has been conducted in 2004, 2005 and Survey questions include: levels of education and other training and qualifications; a history of the number of jobs since leaving school or in the last 20 years depending on which was the shortest and individual and household characteristics. Information on jobs include: type and length of each episode of employment; remuneration; and other sources of income. The unit of analysis in the data is the individual. The UHWS data has a feature which is important in answering the questions posed in this paper, since it provides comparable information including income data on both wage employees and the self employed (with a panel component to address endogeneity). Tables 6 to 9 give a description of the Tanzanian household worker surveys for 2004/5 and Individuals active in the labour market are categorised into formal (public or private) sector employed, informal sector employed and the unemployed (Table 6). As expected in a developing country like Tanzania, the informal sector is the largest sector in terms of employment (61% in 2004/5 and 45% in 2006). This is followed by the private and public sectors respectively. The pattern remains the same across the years, though informal sector employment share is shown to have decreased and picked up by the private sector. 13

14 This could possibly be due to the formalisation of some informal sector businesses. Across the period under consideration (2004/5 2006), public sector and informal sector shares of employment decreased by 5% and 16% respectively. However under the same period, the share of private in employment increased by 18% and unemployment increased by 3%. A further categorisation of employments into manufacturing, non-manufacturing, informal sector and other in table 7 shows the informal sector is the largest sector in terms of employment, followed by non-manufacturing salaried wage, other sectors and the manufacturing sector. Between 2004/5 and 2006, the share of the informal sector in employment decreased by 16% whereas that of non-manufacturing and manufacturing increased by 13% and 4% respectively. Table 6: Percentage of Employment by Status, 2004/05 and 2006 Status 2004/ Change Public Private Self employed Unemployed Total Source: Calculations from UHWS 2004, 2005 and 2006 Table 7: Percentage of Employed Persons by Sector, 2004/05 Sector 2004/ Change Manufacturing Non-manufacturing(Salaried/wage) Informal Other Total Source: Calculations from UHWS 2004, 2005 and

15 Table 8 reports real monthly earnings for paid (formal) employment and self employment. The mean and median monthly earnings of paid employment are more than twice what is earned in self employment and is consistent across the two periods. A disaggregation of the data into male and female shows there are more females (55%) in the data than males across the years, however, men earn more on average than women. An additional analysis of monthly earnings based on levels of education indicates individuals with high levels of education earn more than those with lower or no education (table 9). 48% of individuals in the sample have primary education followed by 35% with secondary education and 12% with no education and 5% with tertiary education. Table 8: Real Monthly Mean and Median Income of Paid and Self-Employment 2004/05 and 2006 Mean (TShs) Median (TShs) Type of Employment 2004/ / Paid employment 79,292 71,050 52,419 53,957 Self employment 33,803 17,756 24,194 10,791 Total 47, ,226 17,986 Source: Calculations from UHWS 2004, 2005 and 2006 Table 9: Real Monthly Earnings by Sex and Education, 2004/05 and / Sex % Share (Tshs) Mean Median % Share (Tshs) Mean Median Male ,575 34, ,690 21,583 Female ,161 24, ,825 14,389 Education No or Pre-school ,277 15, ,095 7,194 Primary ,074 24, ,700 14,389 Secondary ,229 37, ,889 57,554 Tertiary ,445 82, , ,885 Source: Calculations from UHWS 2004, 2005 and

16 4.3 Household Budget Surveys Tanzania Household Budget Surveys, Dar es Sample for 2000/01 and 2006/07 The Household Budget Surveys (HBS) are nationally representative surveys conducted by the National Bureau of Statistics with data on household expenditure and characteristics. For the Dar es Salaam sample (DSM) we have a total of 1,225 households in 2000/01 and 3,435 households in 2007 make a pooled sample of 4660 households. 3 The HBS for 2001 and 2007 data for DSM sample is used to construct cohorts (i.e. households by district and main source of income) to match an expenditure-based measure of income to the ILFS, as an alternative measure of household income to the one obtained from ILFS. The surveys record everything that the interviewed households declare as consumed over one month, including purchased food and food grown by the households and consumed during the month. Respondents were asked to provide information on how much they spent on each item and on the quantity consumed. Total household expenditure per adult equivalent (per capita) is the measure of household income. Table 10: Real Household Total Expenditure by Main Source of Cash Income in TShs and Percentage Change for Dar es Salaam Sample, Main Source * % Change Sales of food crops 13, , Sales of livestock 20, , Sales of livestock ( products) 31, , Sales of cash crops 13, , Business income 19, , Wages or salaries in cash 23, , Other casual cash earning 18, , Cash remittances 15, , Fishing 9, , Other 15, , Average Income Change Note: Income is reported in Tanzanian Shillings (TShs) adjusted using survey weights. Percentages changes are based on Fisher index: 2007 prices while 2007* values are in 2001 prices. This is monthly real income. Source: Authors own calculations only for Dar es Salaam sample from Tanzania Household Budget Surveys for 2000/01 and To get the feel of what happens to the real income of nearly the same households as in ILFS, Table 10 reports changes in real total household expenditure between the 2000/01 and 3 The surveys provide data at the level of Dar es Salaam (the capital), other urban areas and rural areas. 16

17 2007, for the Dar es Salaam sample. 4 While there has been a substantial rise in real income for the main source of household income, on average by 30 percent between 2000/01 and 2007, there is huge difference between Dar es Salaam sample and the entire country (Leyaro et al, 2010). For instance, while the real income for sales of livestock and livestock products increased by 60% during this period for the country; it fell by 16.5% for Dar es Salaam sample. While the real income from sale of cash crops fell by 10.8 percent, wages and salaries fell by 4.37 percent and fishing by 11.4 percent for the country, they rose by 27.6 percent, 11.5 percent and 56.9 percent respectively for Dar es Salaam. This is easy to understand, most dwellers in Dar es Salaam are not livestock keepers, rather some of them are food and cash crops sales brokers, being along the Indian Ocean others are involved in fishing, and many more others engaged in other activities including business and informal sectors, and being the capital it has most of wage/salaries earners. As a result, others sources of real income have shown a very significant increase compared to the entire country. This includes real household income from sales of food crops increased by 45.7%, cash remittances by 30% and other sources by 109 percent. National Panel Survey: Dar es Sample (2008/09 and 2010/11) As said already, one problem with these data sets is that they predates the real financial and economic crisis as they end up to We will extend the analysis by using the most recent survey, the Tanzania National Panel Survey (TZNPS) for 2008/09 and 2010/11 (depending on availability) to draw inferences on what happened during and after the financial and economic crisis of 2008/09. TZNPS for 2008/09 is the first in a series of nationally representative household panel surveys that assembles information on a wide range of topics including agricultural production, non-farm income generating activities, consumption expenditures, labour market characteristics and a wealth of other socio-economic characteristics. The first year of the survey was conducted over twelve months from October 2008 to October It was implemented by the Tanzania National Bureau of Statistics (NBS). The second wave of the TZNPS was done for fall 2010/11 and the data are about to be released by now. Like ILFS and HBSA the sample was constructed based on the National Master Sample frame which is a list of all populated enumeration areas in the country developed from the 2002 Population and Housing Census. The sample includes a partial sub-sample of 4 Real total household expenditure from HBS is obtained by deflating the total household expenditure between 2000/01 and 2007 Fisher Ideal Index that based on Dar es Salaam basket. 17

18 households interviewed during the 2006/2007 Household Budget Survey. The data is broken into for 4 different strata: Dar es Salaam, other urban areas on mainland Tanzania, rural mainland Tanzania, and Zanzibar. In total, the target sample was 3,280 households in 410 Enumeration Areas (2,064 households in rural areas and 1,216 urban areas). 5. Results and Discussion This section presents results obtained both from using Integrate Labour Force Survey (ILFS) and Dar es Salaam Urban Household Worker Survey (UHWS) to assess if they yield comparable inferences on determinants and evolution of sectors income. 5.1 Integrated Labour Forces Survey Results We begin in this section to look on ILFS results that are based on OLS estimator in Tables 11-13, where sectors of employment is defined as main economic activity an individual spent most of his/her time. The same estimations are then compared with the results in appendix B, where sectors of employment are defined as current status of employment. Table 11 uses the single measures of education - highest years of education attained, Table 12 include education as a squared term and Table 13 use education instead as underlying categorical variables. In each set of three regressions in each of the three Tables, the first regression pools all observation both in years (appending 2000/01 on 2006) and in earnings (sum of paid and self-employed earnings), the second regression is paid employment income (where those in self-employment are omitted) and the third regression is self-employment income (where those in paid-employment are omitted). Paid employment in 2001 accounted for 46% and 36% in 2006, while self employed accounted for 45% in 2000/01 and 40% in When we pooled 2001 and 2006 together, paid employment makes 41% while self employed makes 42% of the total sample. 5 5 While in the same household the head may be paid employee, the spouse may be self employee 18

19 Table 11: Determinant of Log Earnings Age, Tenure and Education: Pooled ILFS Dependent Variable: log earnings Age Age 2 /100 Male Tenure Pooled-Income (1) 0.072*** (9.964) *** (-7.947) 0.435*** (14.225) 0.143** (2.950) Tenure ** (-2.964) Education (years) Y *** (23.339) 0.782*** (26.405) Working on own or family farm a (-0.710) 0.449*** Employee central government (3.907) 0.905*** Employee parastatal organisation (7.565) 0.278* Employee NGO/religious organisation (1.703) 0.162* Employee private sector (1.656) 1.473*** Self employed in a business with employee (12.600) self employed in a business without employee Unpaid work in family business Constant 0.477*** (4.589) (-0.397) 7.095*** (37.378) Paid-EmpIncome (2) 0.074*** (10.408) *** (-6.717) 0.283*** (9.049) 0.232*** (5.282) *** (-4.848) 0.129*** (25.973) 0.313*** (11.004) 0.453*** (6.732) 0.806*** (11.192) 0.219** (2.088) 0.236*** (4.205) 6.767*** (43.764) Self-EmpIncome (3) 0.049*** (4.216) *** (-3.853) 0.531*** (10.899) (-0.523) (0.296) 0.140*** (13.875) 1.170*** (23.988) 0.399*** (3.375) 1.401*** (12.367) 0.350*** (3.716) 8.136*** (24.004) r N Source: Authors own calculations from the Tanzania Integrated Labour Force Surveys for 2000/01 and 2006 years Notes: -This s pooled OLS estimates. Figures in parentheses are t-ratios: *** denotes significant at 1 percent level, ** significant at 5 percent and * significant at 10 percent. -Tenure variable is measured as a migration variable a Omitted category are other economic activities, including employee of political party and cooperative Starting with Table 11, we consider age-earnings, gender-earnings 6 and tenureearnings 7 OLS level estimations. The basic results for age-earnings show a significant 6 Gender is measured by a Male variable, a dummy which equal 1if respondent is Male and zero otherwise 7 Tenure here is proxied by a migration variable as there is no a tenure variable in ILFS 19

20 concave age-earning relationship, so is the estimation for tenure except for the regression under self employed. These results are largely in line with other estimations for Dar es Salaam in section 5.2 and those by Quinn and Teal (2008), both of which used three rounds (2004, 2005 and 2006) of the UHWS to examine determinants of earnings. Though concavity of age-earnings relationship can be due to a number of reasons, given what we saw in data section that the Adult earns four times higher than the Youths, this might suggest a system of remuneration in Dar es Salaam labour market that reward more seniority than newcomer in the labour market or development of work specific skills. As we do not have actual tenure variable (i.e. number of years in current job), tenure is proxied by migration variable (i.e. years of residence in the urban area), which too could proxy network links and be associated with higher informal (self-employed) earnings given other determinants. Even though, tenure is not concave and is insignificant under self employment - difficult to interpret this. The dummy for male that is used to capture sex disparities is positive and statistically significant throughout, supporting what we saw under data section, that there is a huge earnings inequality between male and female in Dar es Salaam (male earns three times more than female). Education-earnings relationship is positive and statically significant in Table 11, and there is no a huge difference between a returns to education under paid and self employment earnings. On averaged returns to education is 13%. We allowed education to enter as a quadratic a quadratic term under the specifications in Table 12 to assess the educationearnings convexity. The education-earnings convexity is confirmed as the quadratic term is positive and statistically significant, implying that the more education one has the higher the earnings returns to him/her. These results are similar to what is in section in data section, section 5.2, Quinn and Teal (2008) and Rankin, Sandefur and Teal (2010). But unlike some of other results, these results are significant both under paid (public and private) and self employed earnings. We specify education according to its underlying categorical variables in Table 13, where excluded category is no or pre-school education. The results confirm the same trend observed in Tables 11 and 12, as all coefficients, both under paid and self-employed, are positive and statistically significant. The results support and enhance the education-earnings convexity observed, as people with more education either under paid or self employment tends to earns higher income relatively to those with low level of education (the results are consistent throughout). 20

21 Table 12: Determinant of Log Earnings Age, Tenure and Education 2 : Pooled ILFS Dependent Variable: log earnings Age Age 2 /100 Male Tenure Pooled-Income (1) 0.074*** (10.212) *** (-8.328) 0.438*** (14.316) 0.139** (2.868) Tenure ** (-2.890) Education (years) Education 2 /100 Y (0.523) 0.615*** (3.967) 0.776*** (26.226) Working on own or family farm a (-0.700) 0.437*** Employee central government (3.801) 0.889*** Employee parastatal organisation (7.439) 0.282* Employee NGO/religious organisation (1.731) 0.167* Employee private sector (1.609) Self employed in a business with 1.485*** employee (12.712) self employed in a business without employee Unpaid work in family business Constant 0.484*** (4.655) (-0.304) 7.577*** (33.668) Paid-EmpIncome (2) 0.077*** (10.798) *** (-7.262) 0.278*** (8.945) 0.233*** (5.327) *** (-4.896) (-1.120) 0.839*** (5.387) 0.306*** (10.801) 0.446*** (6.676) 0.796*** (11.117) 0.223** (2.135) 0.240*** (4.290) 7.468*** (37.080) Self-EmpIncome (3) 0.052*** (4.399) *** (-4.114) 0.533*** (10.948) (-0.601) (0.372) (0.819) 0.579** (2.202) 1.167*** (23.941) 0.393*** (3.323) 1.402*** (12.384) 0.357*** (3.790) 8.535*** (22.222) r N Source and Notes: As in Table 1 21

22 Table 13: Determinant of Log Earnings Age, Tenure and Education Categories: Pooled ILFS Dependent Variable: log earnings Pooled-Income (1) Paid-EmpIncome (2) Self-EmpIncome (3) Age 0.081*** (12.818) 0.078*** (11.713) 0.063*** (6.443) Age 2 / *** ( ) *** (-8.163) *** (-6.679) Male 0.472*** (15.652) 0.288*** (9.396) 0.571*** (12.060) Tenure 0.127** (2.670) 0.233*** (5.441) (-0.824) Tenure 2 (-2.670) (-5.000) (0.597) ** *** Primary a (6.086) (3.021) (4.887) 0.395*** 0.250** 0.435*** Secondary 1.020*** (14.134) 0.832*** (9.645) 1.150*** (10.751) Post Secondary 1.501*** (17.112) 1.373*** (14.689) 1.555*** (9.400) Tertiary 1.613*** (10.568) 1.485*** (11.591) 2.468*** (5.014) Y *** (26.042) 0.295*** (10.514) 1.122*** (23.717) Working on own or family farm a (-1.104) Employee central government 0.499*** (4.413) 0.508*** (7.748) Employee parastatal organisation 0.943*** (8.004) 0.853*** (12.136) Employee NGO/religious organisation 0.344** (2.137) 0.268** (2.624) Employee private sector 0.190* (1.875) 0.275*** (5.052) 0.385*** (3.391) Self employed in a business with employee 1.495*** (12.980) 1.403*** (12.824) self employed in a business without employee 0.479*** (4.711) 0.354*** (3.998) Unpaid work in family business (0.661) Constant 7.480*** (39.864) 7.331*** (44.975) 8.527*** (26.290) r N Source and Notes: As in Table 1 a Omitted category are no education or pre-school for education Somewhat these results are not in line with Quinn and Teal (2008), who found tertiary are not significantly different from zero and post secondary and tertiary education 22