Attrition and Follow-Up Rules in Panel Surveys: Insights from a Tracking Experience in Madagascar

Texte intégral

(1)DOCUMENT DE TRAVAIL. DT/2010-13. The methodological challenge of. monitoring living conditions. Insights from a tracking experience in Madagascar.. Julia VAILLANT. UMR DIAL 225. Place du Maréchal de Lattre de Tassigny 75775 • Paris Cedex 16 •Tél. (33) 01 44 05 45 42 • Fax (33) 01 44 05 45 45 • 4, rue d’Enghien • 75010 Paris • Tél. (33) 01 53 24 14 50 • Fax (33) 01 53 24 14 51 E-mail : dial@dial.prd.fr • Site : www.dial.prd.fr.

(2) THE METHODOLOGICAL CHALLENGE OF MONITORING LIVING CONDITIONS. INSIGHTS FROM A TRACKING EXPERIENCE IN MADAGASCAR.1 Julia Vaillant Université Paris Dauphine, LEDa UMR 225 DIAL, IRD vaillant@dial.prd.fr Document de travail UMR DIAL Décembre 2010 Abstract Most longitudinal surveys recontact households only if they are still living in the same dwelling, producing very high attrition rates, especially in developing countries where rural-urban migration is prevalent. In this paper, we discuss the implications of the various follow-up rules used in longitudinal surveys in the light of an original tracking survey from Madagascar. This survey attempted in 2005 to search and interview all individuals who were living in the village of Bepako in 1995, the baseline year of a yearly survey, the Rural Observatories. The tracking survey yielded an individual recontact rate of 78.8%, more than halving attrition compared to a standard dwelling-based follow-up rule. The tracking reveals a very high rate of out-migration (38.8%) and household break-ups, as three quarters of recontacted households had divided between 1995 and 2005. The average income growth of the sample over the period increases by 28 percentage points when follow-up is extended to those who moved out of their household or village, suggesting that dwelling-based panels give a partial view of the welfare dynamics of the baseline sample. A higher baseline income per capita is associated with a higher probability of staying in Bepako and of being found in the tracking if one moved out. The hardest people to find are the poorest and most isolated. Special attention should be paid to collecting data that enable the identification and follow-up of individuals without which attrition is likely to remain a source of bias even after a tracking procedure is carried out. Key words : Panel data, tracking surveys, attrition, mobility. Résumé La plupart des enquêtes en panel ne recontactent les ménages enquêtés que s'ils vivent toujours dans le même logement, ce qui créé des taux d'attrition très élevés, en particulier dans les pays en développement où la migration vers les villes est importante. Dans cet article, nous discutons les implications des différentes règles de suivi utilisées dans les enquêtes longitudinales à la lumière d'une enquête tracking originale réalisées à Madagascar. Cet enquête a tenté, en 2005, de chercher et enquêter tous les individus originaires de Bepako, où une enquête annuelle est réalisée depuis 1995 (Observatoires Ruraux). Ce dispositif a permis de recontacter 78.8% des individus, réduisant ainsi de plus de moitié l'attrition par rapport à un suivi des individus basé sur le logement. Le tracking révèle un taux de migration très élevé (38.8%) et d'importantes recompositions et divisions de ménages, puisque les trois quarts des ménages recontactés s'étaient divisés entre 1995 et 2005. La croissance du revenu moyenne dans l'échantillon sur la période augmente de 28 points de pourcentage lorsque le suivi est étendu à ceux ayant changé de ménage ou de lieu de résidence, suggérant que les panels basés sur le lieu de résidence génèrent une vue partielle de la dynamique des revenus de l'échantillon initial. Un revenu par tête initial plus élevé est associé à une probabilité plus forte de rester à Bepako ou d'être retrouvé lors du tracking. Les personnes les plus difficiles à retrouver sont les plus pauvres et isolées. Une attention particulière doit être portée à la collecte d'information permettant d'identifier et de réenquêter les individus, sans laquelle il est probable que l'attrition restera une source de biais, même après avoir réalisé une enquête tracking. Mots clés : Données de panel, enquêtes tracking, attrition, mobilité. JEL Codes : C81, I32, O12, O15.. 1. The tracking survey used in this study was carried out as part of the research project “Dynamics of Rural Poverty” which has received financial support from the French Ministry of Research under the ACI initiative, “Sociétés et cultures dans le développement durable”. I would like to extend warm thanks to Flore Gubert and Anne-Sophie Robilliard for making the data available and for their very helpful comments during the preparation of this paper.. 2.

(3) Contents 1. Introduction. 5. 2. Rural observatories: an original longitudinal survey design. 7. 3. Monitoring changes in living conditions: measurement issues 3.1. 3.2. 4. 6. 10. 3.1.1. Consumption. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 11. 3.1.2. Income . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 12. Choice and construction of a price index . . . . . . . . . . . . . . . . . . . . . . .. 15. On attrition and experimenting a tracking survey to reduce it. 18. 4.1. Attrition and follow-up rules in longitudinal surveys. . . . . . . . . . . . . . . . .. 18. 4.2. Follow-up methodology and attrition in the ROR . . . . . . . . . . . . . . . . . .. 21. 4.3. Review of existing tracking surveys . . . . . . . . . . . . . . . . . . . . . . . . . .. 24. 4.4. Tracking survey design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 26. 4.4.1. Census of Bepako . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 27. 4.4.2. Individual Trajectories survey . . . . . . . . . . . . . . . . . . . . . . . . .. 27. 4.4.3. Tracking and interview of migrants . . . . . . . . . . . . . . . . . . . . . .. 28. 4.4.4. ROR survey in Bepako . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 28. Tracking results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 29. 4.5. 5. 9. Choice and construction of welfare measures . . . . . . . . . . . . . . . . . . . . .. What can be learned on attrition in the ROR from the tracking? 5.1. Theoretical and empirical aspects of the attrition bias in the literature. 5.2. Does the tracking bring more representativity to the ROR sample?. 33 . . . . . .. 34. . . . . . . . .. 36. 5.2.1. Descriptive statistics: comparison of sample characteristics . . . . . . . . .. 36. 5.2.2. Multivariate analysis: attrition probability model . . . . . . . . . . . . . .. 39. 5.2.3. Impact of attrition on initial behavioral relationships: BGLW test . . . . .. 42. 5.3. Dierential diagnosis on economic mobility with alternative follow-up methodologies 42. 5.4. How serious is post-tracking attrition?. . . . . . . . . . . . . . . . . . . . . . . . .. 46. 5.4.1. Descriptive statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 46. 5.4.2. Multivariate analysis of post-tracking attrition. 49. Conclusion. . . . . . . . . . . . . . . .. 51. References. 54. Appendix A. 56. Appendix B. 58. 3.

(4) List of Tables 1. Baseline sample size breakdown by observatory and village . . . . . . . . . . . . .. 2. Summary of tracking surveys in developing countries . . . . . . . . . . . . . . . .. 25. 3. Recontact rate at the household level . . . . . . . . . . . . . . . . . . . . . . . . .. 29. 4. Recontact rate at the individual level . . . . . . . . . . . . . . . . . . . . . . . . .. 31. 5. Household splits in 2005 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 33. 6. Baseline characteristics of households using dierent follow-up rules . . . . . . . .. 37. 7. Baseline characteristics of individuals using dierent follow-up rules . . . . . . . .. 38. 8. Probit model of remaining in the sample before and after tracking, household level. 40. 9. Probit model of remaining in the sample before and after tracking, individual level. 41. 10. BGLW test of dierence in initial behavioral relationships. 43. 11. Income growth using alternative follow-up rules at the individual level. . . . . . .. 12. Income growth using alternative follow-up rules at the household level. 13. Baseline characteristics of post-tracking attritors, individual level. . . . . . . . . . . . . .. 22. 44. . . . . . .. 45. . . . . . . . . .. 47. 14. Multinomial Logit Regression. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 50. A.1. Baseline caracteristics of households by tracking status . . . . . . . . . . . . . . .. 56. A.2. Baseline caracteristics of individuals by tracking status . . . . . . . . . . . . . . .. 57. List of Figures 1. Recontact rate of baseline households by year and observatory . . . . . . . . . . .. 23. 2. Household tracking results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 30. 3. Individual tracking results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 32. 4.

(5) 1. Introduction. This paper is a methodological contribution to the literature on economic mobility and poverty dynamics in developing countries. Economic mobility refers to the increase or decrease of wellbeing over long time periods, while poverty dynamics are movements in and out of poverty (Baulch & Hoddinott, 2000). This rather recent eld of development economics has recognized the dynamic aspect of poverty and the existence of an important fraction of the population that can be poor at some time but is not always poor. While characteristics of core, permanent poverty are relatively well known, factors explaining the changes of well-being are not. Policy implications are crucial however, because targeting the right groups for poverty reduction programs, designing safety net policies and equitable growth policies depends on the knowledge of which groups benet or lose from economic development, policy and shocks. Furthermore, there are empirical regularities found in the literature that give ground to a systematic study of that phenomenon in developing countries.. For example, the chapters brought together in the volume edited by. Baulch and Hoddinott (2000) nd that the group of sometimes poor is very large and that movements in and out of poverty are frequent. The essential input to carry out quantitative and rigorous studies of economic mobility or poverty dynamics is longitudinal data. Repeated interviews of the same units, households for example, using the same questionnaire, at more or less distant time intervals, give the opportunity to look at the evolution of individual well-being indicators, study the persistence of poverty, and the reasons of entrance into and exit from poverty.. Although panel data are practically. indispensable to study these phenomena, they carry a certain number of disadvantages and limitations such as measurement and sampling issues. In certain cases, measurement error can be increased by panel data, and economic mobility overstated. Besides, household longitudinal surveys require a rigorous denition of who is the successor of a given household, to ensure that the same unit is eectively re-interviewed, which is not straightforward. In addition, given that a particular denition of the longitudinal household is chosen, attrition and panel aging make panel data less representative than cross-sections over time (Glewwe & Jacoby, 2000). Considering these limitations, Ashenfelter, Deaton, and Solon wondered, in 1986, whether collecting panel data in developing countries [...] made sense (Ashenfelter et al., 1986). The increasing use since then of panel data econometrics has showed that researchers in development microeconomics consider them to be, at the very least, reliable enough to make useful inferences.. 5.

(6) However, they have often overlooked these limitations and in particular, not enough attention has been paid to the issue of attrition (Dercon & Shapiro, 2007). This problem is further aggravated in developing countries where rural-urban migration is prevalent (Alderman, Behrman, Kohler, Maluccio, & Watkins, 2001). High spatial mobility causes attrition which in turn causes neglect of these particular economic trajectories in panel data. In this paper, we attempt to discuss these issues in the light of an original dataset from. 1. Madagascar. This dataset is the combination of the Rural Observatories (ROR ) and a tracking survey, designed to complement the former. The ROR is a unique survey that has been conducted every year since 1995 in several selected Malagasy villages.. A simple follow-up rule was used. to create the panel, which was based on the head of the household. If the household head had left the village between two rounds but his spouse had not, she was re-interviewed in his place. This widespread follow-up rule and other factors created large attrition, and, in particular, made the study of migration impossible. One of the eight villages of the ROR, Bepako, was chosen in 2005 as the eld for an innovative survey that attempted to nd all individuals that belonged. 2. to the baseline sample in 1995, even if they had left the village.. This tracking survey cut. attrition by half and thus should improve the study of the dynamics of poverty by reducing the attrition bias and including those who moved out of the village. Because the data comes from one village only and is not statistically representative of a larger area, we believe it has little external validity.. What it can tell us, however, is how much we can learn from some specic. methodological choices made in the data collection process. Taking the data as two cross-sections from 1995 and 2005 yields an increase in the average village per capita income of 5.3%, while using the panel dimension (without tracking) results in an average per capita income growth of households of 40.5%. As the rst methodology does not take into account individual dynamics it is clearly not adapted to the context which requires longitudinal data. Besides, the addition of migrant households found in the course of the tracking to the standard panel of households who stayed in the village yields an average per capita income growth of 68.5%, which compared to the panel without tracking increases the average income growth over the period by 28 percentage points. The choice to collect longitudinal data and the extent to which sample units are followed are critical as they can determine to a large extent the results found in the studies of income. 1. The Rural Observatories were subsequently renamed the Rural Observatories Network (Réseau des Observa-. toires Ruraux), hence the ROR acronym used in this work. 2. A few other tracking surveys existed at the time, but they are rare.. 6.

(7) mobility and poverty dynamics. This motivates this paper which discusses methodological issues in the monitoring of living conditions in developing countries. The Rural Observatories are described in the next section. In section 3, we discuss measurement issues that arise in panel data. We detail the methodological choices made to construct a comparable welfare aggregate over time, dealing with inconsistent questionnaires and the lack of proper price indices.. Attrition in panel data, in particular the ROR data, is the focus of. section 4. The design and results of the tracking survey, designed to reduce attrition, are also be presented. Section 5 explores the extent of the attrition bias in the ROR, with and without the tracking. We attempt to draw lessons on the desirability of such tracking devices, their benets and limitations, and give recommendations to improve them in the conclusion.. 2. Rural observatories: an original longitudinal survey design. The Rural Observatories were created in 1995 as part of the MADIO project. In 1994, the Malagasy National Statistical Institute (INSTAT) and the French Research Institute for Development (IRD), and in particular DIAL, a research center which belongs to the IRD, put together a 4-year research program in Madagascar, called MADIO. MADIO had several goals: promote economic research in Madagascar, produce analyses of the transition that was taking place in the 1990s, and renovate the statistical system by creating several dierent specic surveys among which. 3. were the Rural Observatories (Droy, Ratovoarinony, & Roubaud, 2000).. A socioeconomic observatory is a statistical tool whose aim is to follow and monitor the population of a particular area in order to identify the dynamics of improvement or worsening of the living conditions of that population. The Rural Observatories were intended to illustrate particular key issues of Malagasy agriculture and were selected according to several criteria: the agro-climatic zone, the dominant production system, demographic characteristics, remoteness, etc. (Droy et al., 2000) In 1995, four observatories were set up:. •. The Mahafaly coastal plain observatory in the south-east, to study the livelihoods of the shermen and farmers of this very arid and remote region;. 3. Other surveys created by the MADIO project included a labor and employment survey, the 1-2-3 survey,. designed to study the informal sector and has since then been extended to other developing countries. For more information on MADIO, the reader can refer to the article by Droy et al. (2000) in a special issue of dedicated to the MADIO project. This section is largely inspired by this article.. 7. Stateco.

(8) •. The vanilla observatory in the north-east, to analyze the impact of trade liberalization on vanilla producers;. •. The Vakinankaratra observatory, to observe family smallholder agriculture in the Malagasy highlands;. •. The Marovoay observatory, to collect data on paddy farmers in an irrigated perimeter going through a heavy reorganization process.. Two villages were chosen in each observatory, across which 500 households were randomly drawn and interviewed. These households have been interviewed every year since then. At the start of each round, the survey team takes a census of the population of the village. If a household surveyed the previous year is not found, it is randomly replaced by another one in the village. As a large part of this paper is devoted to follow-up methodologies in panel surveys, the statistical implications of that survey design will be discussed in more depth in section 4.. In 1999, as. the MADIO project was coming to an end, the survey device was extended to several other observatories that became part of the Rural Observatories Network (ROR), ran by the local administration. An important drawback of the data collected is that they are not statistically representative at the regional or national level because the villages were chosen rather than randomly drawn. Each village survey is statistically representative only of the village surveyed itself.. It is not. possible to extrapolate gures at the country level for use in the national accounts for example, nor can one aggregate the data across the observatories. This particular survey design was chosen because, through the geographical concentration of the survey zones, better quality data could be collected at lower cost. It is also the only way to obtain sucient data on certain specic issues of the Malagasy agriculture such as vanilla production, which represents a fraction of the population too small to be studied using national survey data.. The Rural Observatories. were never intended to create nationally representative data or to substitute themselves to the agricultural statistical system of the country. We believe the non-representativeness of the data is however compensated by the longitudinal aspect of the survey, which is the only way to study poverty dynamics and economic mobility at the individual level. In addition, the ROR is still running and some of the historic villages are still surveyed every year as of today. This means that there are several 15-year long panels. 8.

(9) available in Madagascar, which is exceptional enough to be noted. The sampling unit of the survey is the household. The questionnaire is around 10 to 15 pages long and contains modules on the demographic and social characteristics of the household, its dwelling and level of comfort, expenditures on food, non-food and durables, and comprehensive and detailed modules on the farm inputs and outputs of the households.. There are specic. sections for paddy production, consumption and sales, and for other crops, livestock and livestock products. The main advantage of the Rural Observatories survey is that it is a yearly household panel. This makes the study of poverty dynamics, poverty persistence and economic mobility possible. Although longitudinal data enable us to go beyond what repeated cross-sections can do, there are nevertheless two main methodological issues that threaten the quality of the data and of the panel studies that use these data. Firstly, issues in the measurement of welfare, that exist in cross-sections as well, are rendered more complex by problems of comparability across time and price variations. Secondly, attrition, which is inherent to panels, is likely to cause biases in the analyses, if the reason for being lost to follow-up is in any way correlated with the outcome studied.. Besides, attrition caused by very local follow-up rules in the presence of geographic. mobility makes the study of migration impossible and restricts the sample to the particular part of the population which does not migrate. The remainder of this paper is devoted to discussing these issues and how they occur in the ROR data. We discuss measurement issues rst, then sampling problems.. 3. Monitoring changes in living conditions: measurement issues. In theory, the great advantage of panel data is that they make it possible to compare one's welfare over time. In practice, the problem is rather complex as there are several measurement issues that arise. Firstly, issues in the welfare measure itself exist. The choice of a variable such as consumption or income to measure welfare is not straightforward, and depends on assumptions, beliefs and the quality of the data. Once the welfare measure is chosen, it needs to be harmoniously measured across years.. Changes made to a questionnaire can threaten comparability. of variables across survey rounds.. In addition, price variations need to be corrected using an. appropriate price index to deate variables.. 9.

(10) 3.1 Choice and construction of welfare measures Analyses of poverty dynamics and economic mobility require that welfare be consistently measured over the period under study. This is not a minor issue in the context of the ROR data because questionnaires have changed between the rst and the last round. The very rst round, in particular, used quite a dierent questionnaire from the following ones. It is, however, our baseline and we have to make sure that the variables constructed from this dataset are comparable to those obtained in the subsequent years.. The choice of the indicator itself (income. or consumption) will be strongly determined by the availability of the data and the modules included in the questionnaires as we will explain below. This question of the measure of welfare has been discussed at length by Ravallion (1992) and Deaton (1997). A major issue in the measurement of welfare is the choice of the well-being indicator itself. Several measures exist, that can be either welfarist or non-welfarist. The former approach refers to measures that rely on individual utility levels, whereas the latter does not use the information on utilities. In that sense, non-welfarist measures have to include some kind of normative judgment, for example, setting a nutritional attainment level that is considered minimal to live a normal life. The most widely used concept is the "standard of living". It is a welfarist and non-welfarist approach that attaches importance to the individual consumption of privately supplied goods, usually measured by current consumption or income (Ravallion, 1992). Total consumption is usually considered a superior measure of well-being to total income, for several reasons. First, consumption smoothing possibilities, such as saving, borrowing and insurance mechanisms, make it a better indicator of long-term well-being, whereas income is usually prone to high volatility, especially in developing, rural areas, where farm income is very vulnerable to climate shocks. Second, data collection diculties and measurement errors are important in household surveys, and it seems that they are greater for income than for consumption. A large part of income in farming economies comes from self-employed agriculture and a substantial part of it is often consumed by the producing household rather than sold, which makes the reporting and assessment of its value dicult. However, that problem is also true of the measurement of consumption as it should include consumption of self-produced crops and produce. As stated by Herrera and Roubaud (2003), one cannot denitely establish that measurement diculties and errors should be larger when measuring income than consumption. Furthermore, the consumption smoothing assumption is not always veried, as shown for example. 10.

(11) by Dercon and Krishnan (2000) in Ethiopia, where consumption is volatile and uctuates largely over seasons and with shocks. The case against income as a measure of well-being is thus not strong enough to reject it completely, especially in the case of the ROR data, where comparability issues argue in favor of using income rather than consumption. We review these two measures in turn.. 3.1.1. Consumption. Guidelines for constructing a consumption aggregate are given by Deaton and Zaidi (2002). They recommend that it should be the sum of the value of food consumption (bought, home-produced, received), non-food (basic items, clothing, housing equipment, education expenses), housing rent (or its annual rental equivalent if the dwelling is owned) and the value of services received from the durable goods owned (measured by its annual rental equivalent). The ROR survey was not designed to construct a sophisticated consumption aggregate. The expenditures module of the 1995 questionnaire is practically non-existent. It only includes questions on durable goods such as housing equipment and clothing, and schooling, health, transportation, administration and ceremony expenses. This is very incomplete, as it does not include expenditures on any food or non-food products. More detail was added in 1996 and, since 1997 the module has included expenditures on basic foodstu and non-foodstu: oil, sugar, salt, oil, sh, legume, rice, cassava, corn, soap, petrol, tobacco. Respondents are asked to report usual amounts spent and quantities bought for each of these items at the frequency of their choice (daily, weekly, monthly, etc.). It is not possible to derive a satisfactory consumption aggregate from the 1995 questionnaire and we do not try to. The lack of detail in the expenditures module is the results of a trade-o made with farm related modules. The ROR survey is intended to monitor agriculture and farmers' livelihoods. As such, it was designed to include detailed farm related modules, but not consumption, to keep interviews reasonably short and gain quality on the other parts of the questionnaire.. 4. 4. It is possible to calculate a consumption aggregate using the 2005 data. However, this measure is not used in. this paper as the analyses presented below either look at baseline characteristics or make intertemporal welfare comparisons, for which both baseline and nal values are needed.. 11.

(12) 3.1.2. Income. Unlike expenditures, the ROR has fairly comprehensive modules concerning income components. It includes detailed information about farm output production and sales, input purchases, as well as in kind income, and income from other secondary activities, as required by Deaton (1997) to compute a satisfactory income measure. The paddy production, in particular, which accounts for the largest share of the household's income, was reported in very similar ways in 1995 and 2005. We argue that our measure of income can thus be a fairly good indicator of resource ows. We will dene two dierent measures of income, respectively for rural households and for urban households.. 5. We dene the rural income aggregate as the total gross farm and non-farm income, minus production costs. The components of the gross farm income are the production of paddy and other crops and the income generated by livestock and the produce of livestock.. These com-. ponents are often expressed in the data as physical quantities, which need to be converted into monetary values to be summed up in the income aggregate. Therefore, a price for each of them had to be computed. The general method used to obtain a price was to divide, for each observatory, the total sales of a crop or produce by the total physical quantity sold. This yields an average unit value per crop or produce and by observatory. When that price was not available, because no household had sold a particular type of produce in one particular observatory, but was needed all the same because some households had produced it (for home consumption), the average price computed over all households of all observatories was used. This is not entirely satisfactory, as there could exist systematic dierences between the price of a specic observatory and the average price, but we do not have another source of data, and must value the production of all crops anyhow. The production of the tracked households that had stayed rural, but had moved from Bepako, was, for all crops and livestock, valued at the Marovoay observatory price, because there were not enough observations to compute a specic price for each district. This assumption can be made as the households were only tracked if they stayed in the same region. To the crop output is added the livestock output, measured as the sales of animal produce (meat, dairy, eggs) plus 10% of the value of the household's herd, which corresponds to the appreciation rate of livestock.. 5. 6 We add the income generated by the rental of rice elds to other farmers. The inclusion of urban households in this discussion stems from the specicity of the tracking which recon-. tacted individuals that had moved to urban areas.. 6. We follow Jacoby (1993), who arbitrarily chooses an appreciation rate of 20% and assumes it to be constant. 12.

(13) and the non-agricultural income. Non-agricultural income sources are wage labor and secondary activities, such as shing or small businesses and trade. The net rural income is nally obtained by deducting the production costs from the gross income. There are three types of costs: land, labor and inputs. The cost of land is the amount paid in money or in kind for the rental of a cultivated area. The in kind payments are made in paddy, and therefore valued using the unit value of paddy imputed in the way we mentioned above. The cost of labor is the amount paid for the labor force hired on the farm, and the input costs are the amounts spent for the inputs necessary to the production, such as fertilizer, seeds, or food and medical care for the livestock. By denition, no farm modules were included in the questionnaire administered to households living in urban areas in 2005.. A household was administered the urban questionnaire if his. main activity was not farming or cattle-raising. This much shorter questionnaire asked direct questions on main and secondary activities and on the income generated by these activities. The income of urban households is dened as the sum of three components: total wages and prots (if the individual owns a business), secondary activity income and non-labor income (such as pensions and nancial income). For the most important part, rice production, the questionnaires are comparable between 1995 and 2005. However the 1995 questionnaire did leave out a few questions that were added later. When the necessary modules to compute the dierent income components were not included, the general methodology we chose was to impute a value for that component using the data available for the other years. We then used rank correlation tests on the 1995 sample to test if the ranking of households according to their income was robust to the alternative assumptions made. A rst problem was that the livestock produce module of the 1995 questionnaire only referred to milk production (and not other produce) and was not precise enough to calculate the value of the milk yearly production itself. We assumed that the value of livestock produce was a constant proportion of the value of the livestock owned over time and across households in a village. The average ratio (produce/livestock) over the 1996-2005 period was computed and multiplied by the value of the livestock in 1995. We did the same with the single year 1996 ratio, and thus obtained two proxies for that missing component. Using a Spearman rank correlation test, we found that over households. We assess robustness using a Spearman rank correlation test for other values (5%, 10%), and nd that the ranking of households is robust to the choice of the appreciation rate. We chose a rate of 10% for our baseline measure of income, which seems more reasonable for Madagascar.. 13.

(14) the ranking of households according to the income in 1995 is not modied by the choice of one assumption over another, namely, using the 1996-2005 average, the 1996 ratio, or simply deleting that component from the measure of gross income. We choose to use the 1996-2005 average for our baseline income measure because it uses more information than a single-year ratio. Another problem was that the module in the 1995 questionnaire concerning the income generated by the renting out of land was also somewhat dierent from the 2005 module, because where the 1996 to 2005 questionnaires asked about the rice elds rented out, the 1995 questionnaires asked about. all. elds rented. However, that problem was easily solved, as the amounts in. 1995 and the other years were very similar, and as the data collected on the costs of the rental of farming land showed that the rental of land other than for rice cultivation was practically null. We could therefore assume without much risk that the questions about the rice elds rented out and all elds rented out were actually the same. The non-farm component of the gross income was quite dicult to compute because of the dierences between the questionnaires.. That component is dened as the sum of wage. labor income and secondary activity income. The 2005 questionnaire design enables a separate calculation of wage labor income and secondary activity income, unlike the 1995 survey, for which only an aggregate measure of non-farm income can be computed. One important assumption had to be made about wage labor. The question asked in the 1995 questionnaire was how much did you earn the last month you worked?. which was not very clear.. We assumed that the. answer to that question was the monthly wage. To impute the annual income, we multiplied the assumed monthly wage by twelve, except where the work was declared irregular and seasonal, in which case it was then multiplied by seven months, which corresponds to the average period of rice growing and harvest, during which day labor is hired. Assumptions also had to be made about the costs of agricultural inputs, which are subtracted from the gross income. We had indeed no information at all on the use of these inputs in the 1995 questionnaire. produce.. The strategy adopted here was analogous to the one used for livestock. We assumed that the value of agricultural inputs was a constant proportion of the. value of the agricultural production over time and across households in a village, and multiplied the production of each household, rst, by the average 1996-2005 ratio, then by the 1996 ratio. It would have been optimal to use the technical coecient, that is, the ratio of the volume of inputs over the volume of production, but we do not have price data for inputs, so we can only carry out. 14.

(15) an approximation of that, by using the values. Once again, using a Spearman rank correlation test, we nd that the ranking of households is not modied by using the average 1996-2005 ratio, the 1996 ratio or by simply not including the cost of inputs in the income aggregate. We choose to use the average value over the 1996-2005 period, because it uses the most information.. 3.2 Choice and construction of a price index Another key issue when one wants to make welfare comparisons is the choice of an appropriate price index to deate nominal income. In this section, we discuss the possibilities and choices made to deate the welfare measures. It is particularly crucial here, because we are comparing real incomes in 1995 and 2005, which is a long period with regard to ination.. Furthermore,. in developing countries, spatial price dierences might arise because of imperfectly integrated markets, high transport and marketing costs, especially between rural and urban areas (Deaton, 1997).. In our sample, a number of individuals moved to another rural district or an urban. area, so an assessment of the evolution of their real income requires a spatially dierentiated price index.. The matter is further complicated by the fact that the consumption structure. between rural and urban households could be dierent, because of dierences in preferences and relative prices (Ravallion, 1998). Hence, to compare the purchasing power provided by dierent levels of income in our sample, we would ideally deate incomes by a price index that would account for dierences in cost-of-living in dierent years and locations and individual preference. Unfortunately, the necessary data are not available, and we must nd proxies for the price index.. 7 does not collect very disaggregated price data. It publishes a. As can be expected, INSTAT. Consumer Price Index with prices collected in seven major cities of Madagascar since 2000. Before 2000, it collected price data only in Antananarivo.. The ROR itself started to collect. market prices at the end of the year 1996 in every observatory, 3 times a month. Unfortunately these prices are neither available in 1995 nor in 2005, because this operation was given up before 2005. In addition, market prices were not collected in the various locations in which the tracked households lived in 2005. As none of the price indices needed are available, we choose to calculate an index using information from the expenditures module of the 2005 questionnaire. Following the recommendations of Deaton and Zaidi (2002), we compute a Paasche index, which is a weighted average of the. 7. National Statistical Institute of Madagascar.. 15.

(16) ratio of prices faced by a household relative to a set of reference prices. In the Paasche index, the weights are the budget shares devoted by a household to each good concerned. The following formula is an approximation of the Paasche index:. Ph = exp where. wih. i=1 X. ph wih ln i0 pi N. is the budget share devoted by household. faced by household. h, p0i. is the reference price of good. !. h. (1). to good. i, phi. is the price of good. i. i.. Alternatively, one could use a Laspeyres price index, in which weights are the same for all individuals. This would give an indication of how much better or worse o a household is than a reference household, however it does not represent welfare correctly. This is why the Paasche index is preferred, as it yields a Money Metric Utility, that is, the amount of money required to sustain a level of living (see Deaton and Zaidi (2002), for a more detailed discussion on the choice of a price index). The methodology for calculating the. phi , p0i and wih is as follows.. Unit values are used to proxy. for market prices, because we do not have market-collected prices. For each item of expenditure, respondents are asked to report both the quantity and value of purchases over the period of their choice, which is then annualized by the interviewer. Unit values are the amounts divided by quantity for each item. These items are sugar, salt, oil, legume, rice, cassava and corn. As discussed by Deaton (1997), unit values are not prices, as they are aected by the choice of quality, and thus are partly chosen by consumers, whereas they have no control of real market prices. Higher quality will yield higher unit values, which are consequently positively correlated with income. Deaton (1997) nds that, by using several dierent datasets, this eect exists but its size is modest, and unit values are thus a useful indicator of spatial price variations.. We. must however be aware that the quality eect might underestimate an income increase between two years or locations, as, in the measure of real income, both the numerator (nominal income) and the denominator (ination index) will rise simultaneously, leaving real income constant. Another drawback is that recalling diculties by the respondent or reporting mistakes by the surveyor might cause important measurement errors.. To minimize sensitivity to outliers and. unit inconsistencies, we take the median of unit values of each good in each district and sector, instead of the household unit values. We thus replace. 16. phi by pdi in the formula, where d denotes the.

(17) district and sector. Weights in the consumption basket are then calculated for each household, by taking the ratio of the value of the consumption of each item over total consumption (of those basic food items only). For rural households, the denominator also includes self-produced goods.With this method, each household has its own Paasche index. As the number of items is extremely limited and our sample size small, we also compute a more aggregate index, calculated at the district and sector level:. i=1 X. Pd = exp. N. pd wid ln i0 pi. ! (2). Weights are averaged over all households in each sector and district, which provides democratic rather than plutocratic indices. This means that, unlike CPIs computed by statistical oces, in which wealthier households are over-represented, each consumption structure in the. 8 The choice of a reference price system p0 depends i. sample has the same weight in the price index.. on the specic survey design and calendar, and will be discussed in the next section. Using a Paasche index allows for dierences in the structure of the consumption basket. However, our ination index only includes a very small number of items.. This is its main. weakness. Firstly, it assumes that the basic foodstus purchasing power is a good proxy for the global purchasing power.. Secondly, it is probably not very appropriate for urban households,. who live in areas where the supply of goods is much wider and varied than in rural areas of Madagascar. Given the strong data constraints, we argue that this price index is the best that can be done to adjust for dierent cost-of-living between urban and rural areas, and across districts. Finally, we need a price index to deate income between 1995 and 2005. Unfortunately, we cannot use the same methodology as for the 2005 spatial price index because there is no basic foodstus expenditures module in 1995 from which to derive unit values and weights. We choose to use the national CPI instead. This index itself has drawbacks because it uses prices collected in Antananarivo before 2000, and in several cities after 2000.. It thus uses only urban prices,. whereas our sample is mostly rural. Having discussed methodological choices made to calculate consistent welfare aggregates, we now turn to the other main issue of panel data: sample attrition.. 8. In one district there were not enough households to disaggregate by sector, so only one index was calculated. for the whole district. 17.

(18) 4. On attrition and experimenting a tracking survey to reduce it. 4.1 Attrition and follow-up rules in longitudinal surveys Attrition has been called the Achilles heel of longitudinal household surveys (Thomas, Frankenberg, & Smith, 2001). It is the process by which a panel loses its sampling units over time. This can occur if the surveyed unit dies, refuses to participate in the survey, moves to another dwelling (if the identication of the surveyed household is dwelling-based) or village or region. The breakup of a household can create attrition as well, depending on how the follow-up rule is dened. If attrition is not random with respect to the outcome studied, that is, if there is selection of attritors, there will be an attrition bias in the estimates because the balanced sample will not be representative of the population. A related but slightly dierent issue is panel aging, which also reduces representativeness as the individuals of the sample become older than the population every year. In a basic panel design, sampling is done once, at the onset of the survey, which has the advantage of reducing costs, compared to repeated cross-sections.. But as panel members grow older, as individuals. leave their households to join or create new ones, the sample will no longer be a random draw of the population. This is not because the sample itself is selected, as in attrition, but because the population out of which it was drawn has changed while the sample has not. There is thus a trade-o between following individual income dynamics, which can only be done with longitudinal data, and representativeness (Ashenfelter et al., 1986; Ravallion, 1992; Deaton, 1997). A particular case of attrition occurs when respondents are lost because they move out of their initial location, where the survey is run. If migrants are not followed and interviewed, an attrition bias can be potentially generated because it is likely that migrants are not a random sample of the population. Besides, it makes the study of the welfare dynamics of migrants impossible. This creates an important limitation in the study of poverty dynamics and economic mobility using panel data because spatial mobility has been shown to be one of the strategies that individuals implement to improve their welfare and cope with risk (Dercon & Shapiro, 2007; Ashenfelter et al., 1986). While part of attrition is completely independent of the surveyor's will (deaths for example), the survey methodology and in particular the choice of a follow-up rule can crucially determine the extent of attrition in panel data. The World Bank's Living Standard Measurement Surveys. 18.

(19) program recommends a dwelling-based follow-up rule, which means that the sampling unit is a dwelling that is randomly drawn in the population of enumerated dwellings, and whoever lives there will be the interviewed household. If the household leaves the house between survey rounds, then it will be lost to follow-up, which will result in attrition in the data (Glewwe & Jacoby, 2000). The sample is updated at each wave by adding a random sample of newly-built dwellings. A major inconvenient of these widely used residence-based panels is that they cannot take into account geographical mobility: migrants are automatically excluded from the panel after they move. Several papers have brought arguments against this type of follow-up. Rosenzweig (2003) or Thomas et al. (2001) for example nd that it creates signicant biases in the study of economic mobility, and recommend following movers if they leave the baseline location. This is the purpose of tracking surveys, which can be local, national or even international.. These surveys try to. reduce attrition due to spatial mobility by searching and interviewing those who have moved away from their initial location and would be lost to follow-up if no search was attempted. A trade-o has to be made between a rigorous and comprehensive follow-up of respondents, through tracking surveys, and the cost of the survey. This will depend on how much mobility there is in the sample. If an extremely small fraction of the sample has moved far away, it might not be benecial to try to nd them, given the poor representativeness of these units compared to the cost of follow-up. A very mobile sample would gain from a local or national tracking as migration is an important characteristic of the population studied (Glewwe & Jacoby, 2000). In theory, all attrition due to spatial mobility could vanish with the use of extensive tracking surveys (regardless of their cost). But the problem is more complex than that as the choice of a sampling unit (between individuals and households for example) is not straightforward and results. 9 Most living standard surveys are household panels, and although. in dierent attrition patterns.. they usually collect information on members of the households, they do not include exactly the same individuals each year as they move in and out of the household. Panel of individuals, on the other hand, attempt to interview the same individuals in each round even if the household they belong to changes or breaks-up between waves. The choice between individual- and householdbased panels will have dierent consequences in terms of attrition and panel aging.. If one. attempts to interview all the individuals of a household, attrition is likely to be higher than if. 9. In this paper we consider only household surveys. Firms, for example, can be sampling units in other types. of surveys.. 19.

(20) the household is followed as a global entity. However, a household-based panel can create more panel aging as it does not account for household break-ups and formations. The central issue here is the denition of the household and how to take into account its dynamics nature, especially in developing countries. Taking the apparently simple example of a nuclear family in which the couple divorces in the subsequent round and the children live with the mother in another location, we see how the problem is far from obvious. Which of the two new households is the successor of the initial one? Should it be the one where the initial household head lives or where the majority of the initial members reside?. It could also be none of the. above, according to a denition based on household type (single individual, couple with children, single parent family). Previous works by demographers on developing countries have proposed longitudinal household denitions, which set clear rules to decide which households are the same across time and which ones are dierent (McMillen & Herriot, 1985). This methodology is subject to criticism because studies using longitudinal households as the unit of analysis ignore individuals who do not belong to longitudinal households, that is, those who live in splitos that have been classied as dieren from the antecedent household.. This will create a. selection bias as those excluded from the panel have disproportionately experienced events such as migration or divorce (Duncan & Hill, 1985).. Besides, if household division is non-random. and depends on individual endowments, a sampling rule that drops split-o households will produce substantial biases in estimates of economic mobility (Rosenzweig, 2003). Duncan and Hill (1985)recommend the alternative approach of dening the individual as the unit of analysis rather than the household and attributing to the individual the characteristics of the household in which he lives. More recently, Dercon and Shapiro (2007) make a convincing case for basing sampling and tracking strategies on individuals rather than households in developing countries, based on ndings in works on poverty mobility reviewed in their paper. Surveys based on individuals lead to an increasing number of households in the sample, as, in theory, every household in which a member of a baseline household lives should be interviewed. Unlike classic household- or residence-based panels, they give the opportunity to analyze household formations and break-ups over time. Until fairly recently, this issue was largely ignored in the development microeconomics literature, as the household structure was taken as exogenous (Foster & Rosenzweig, 2002). Changes in household size and structure, through marriage and divorce, births and deaths, child fostering, are being increasingly viewed by researchers, not only. 20.

(21) as factors of welfare changes, but also as strategies to move out of poverty and cope with risk. This is particularly true in rural economies, where the household is both the production and consumption locus (Fafchamps & Quisumbing, 2007; DeVreyer, Lambert, Sar, & Sylla, 2008). In the next section we discuss these issues in the particular context of the ROR: what follow-up methodology is used and how much attrition it has generated over the years.. 4.2 Follow-up methodology and attrition in the ROR There were four observatories in 1995, in each of which around 500 households were interviewed (with two villages per observatory).. 10. The sampling unit in the ROR is the household.. A. household is dened in the interviewer's guide as. a group of persons related or not, who live under the same roof or on the same compound, take their meals together or in small groups, share part or all of their income for the good functioning of the group, and whose expenditures depend on a common authority called the `household head'.. Households are interviewed every year.. 11. At the start of each round, the survey team takes a. census of the village to count the population, see who moved out and whether there are new households. The follow-up methodology is local, but not as stringent as the dwelling-based rule, as a household that changed houses but still lives in the same village is still included in the panel. If the head of the household dies or moves out, the remaining spouse is surveyed instead. Households are lost to follow-up if both the head and its spouse leave the village: all migrants are thus excluded from the panel. If a household surveyed the previous year is not found, refuses to answer, or died, it is replaced by another one randomly drawn from the village census. In 1995, there were 2016 households and 10749 individuals in the ROR, as shown in Table 1. The follow-up methodology used in the ROR, although superior to a dwelling-based panel, creates attrition if only by excluding migrants. Besides, the number of villages surveyed in each observatory was progressively extended, while keeping sample sizes constant (500 households per observatory). This created additional attrition. As households were excluded randomly from the. 10. This is in contrast with traditional agricultural surveys that take the farm as the sampling unit. The aim is. to include all economic activities of household members, not only agricultural, and to make a distinction between the rural sphere and the agricultural sphere (Droy et al., 2000).. 11. Ensemble de personnes avec ou sans lien de parenté, vivant sous le même toit ou dans la même concession,. prenant leur repas ensemble ou par petits groupes, mettant une partie ou la totalité de leurs revenus en commun pour la bonne marche du groupe, et dépendant du point de vue des dépenses d'une même autorité appelée `chef de ménage'. (ROR, 1995). 21.

(22) Table 1: Baseline sample size breakdown by observatory and village. Observatory Antalaha Antsirabe Marovoay Tuléar. Village Maromandia Ampohibe Soanindrariny Vinany Bepako Madiromiongana Beheloka Itampolo. Total. Households 262 252 252 251 307 188 210 294 2016. Individuals 1223 1346 1491 1458 1490 919 1338 1484 10749. Source: ROR (1995).. panel to reduce the sample size in each village, this should not create an attrition bias. However it does reduce substantially the size of the samples. Figure 1 shows yearly recontact rates of baseline households by observatory. The recontact rate in a given year is calculated as the proportion of households included in the panel every year since baseline on the total baseline sample. By this denition, re-entries are not allowed, so any household who is not interviewed (because he refuses or is absent for example) in a round is excluded from the panel permanently. The recontact rate is therefore necessarily decreasing. As shown by gure 1, attrition so dened is very high. The exogenous reduction in sample size in 1999 (mentioned above) is particularly visible in the Marovoay Observatory. The Tuléar and Antalaha observatories were dropped in 2002 and 2003 respectively. The sharp drop in recontact rates in Antsirabe in 2003 can be explained by the exclusion of one village from the sample. The two villages of the Marovoay observatory (in which we are particularly interested because of the tracking survey that took place there), Bepako and Madiromiongana, were surveyed every year without interruption until 2005. The recontact rate at the end of the 10-year period is 24.6%: a quarter only of baseline households were interviewed every single year since 1995. The ROR actually does not exclude re-entries in the panel, but it attempts to follow-up those interviewed the previous year only.. Households who re-enter the panel are replacement. households, drawn randomly from the village census. They are not identied by the survey team as previous sample households, and are attributed a new identier. Thus a merging of rounds based on the household identier only will yield these very low recontact rates. Only a careful examination of the households in the sample using a dedicated survey, such as the one we will detail below, can enable the construction of a more complete panel, in which households that. 22.

(23) 0. 20. Recontact rate 40 60. 80. 100. Figure 1: Recontact rate of baseline households by year and observatory. 1995. 1996. 1997. 1998. 1999. 2000. Antalaha Marovoay. 2001. 2002. 2003. 2004. 2005. Antsirabe Tuléar. Source: ROR annual survey (1995−2005). left and re-entered the panel are identied as the same units. In itself, this operation will lead to an important reduction in the 1995-2005 household attrition rate in the village of Bepako. Although some data are collected on individuals, such as their age, sex, education or activity, the survey was not designed to create an individual panel: a person who leaves the surveyed household is not followed. In addition, individual identifying codes are not intended for matching across rounds, they are not reported on pre-lled questionnaires from one year to the next. Matching of individuals across waves is thus a laborious, and sometimes impossible task. In the Malagasy context, this is particularly complex, because last names are not family names: there is no way of identifying members of a family based on the name. Furthermore, nicknames are often used as substitutes to the ocial given name and dates of births are declared approximately. As a consequence, creating a panel of individuals. ex post,. by manually matching. across names and dates of birth is extremely dicult in this survey. Any kind of assessment of individual attrition will be thus imprecise. The design of the ROR survey, which follows households only locally, generates the loss of two categories of individuals, those of move out of their household and those who move out of the village. Firstly, as it attempts to re-interview households as a global entity, dened by its household head, it excludes all individuals who change households between survey rounds. Secondly, as the households are interviewed only if they still live in the village where they were initially surveyed, it loses all those who move to another area (village, region, country). While the data contained in the ROR panel are very rich, they cannot be used to study migrants' welfare. 23.

(24) dynamics, or to analyze household formations and break-ups in time. These lacunas in the ROR in particular and in panel data in general have created a need for tracking surveys. Tracking surveys aim to nd and interview households or individuals in a panel even if they have moved away from their initial location, which ultimately reduces attrition and, hopefully, the associated statistical bias. In 2005, ten years after the rst round, a tracking survey was implemented in one of the villages of the ROR, Bepako. This rather experimental enterprise reduced attrition by half and made the study of migration possible. The next sections are dedicated to tracking surveys. We start by reviewing existing tracking surveys, then move on to the presentation of the tracking survey in Bepako and its results.. 4.3 Review of existing tracking surveys In 1986, Ashenfelter et al. (1986) stated that if panel surveys in developing countries are intended to generate accurate information on income dynamics, they must attempt to follow movers or else devise methods to correct for the bias. Even though this idea is not recent and the potential selection bias created by not following movers demonstrated, surveys tracking movers were almost never carried out until recently. We have summarized in Table 2 the main surveys in which the follow-up implied some sort of tracking of households or individuals who moved out of their village.. 12 This table updates a previous, similar table in Dercon and Shapiro (2007). All these. surveys are based on a sample of households interviewed in the rst round. Tracking surveys were carried out 4 to 19 years after baseline. Using the LSMS follow-up rule as a benchmark, we dene the recontact rate without tracking as the proportion of the sample still living in the same dwelling at the time of follow-up.. Individual tracking yields lower recontact rates than. household tracking because the former attempts to nd all individuals targeted, whereas the latter considers a household recontacted if at least one member was found. This is not true for Bangladesh, where individuals were not actually physically searched and interviewed, but data were collected from key informants still residing in the initial village, which explains the 100% recontact rate.. Tracking protocols managed to increase recontact rate by 4 to 48 percentage. points depending on the survey (Table 2). For comparative purposes, a line summarizing the results of the survey described in this paper is added at the bottom of the table.. 12. We have found only ve tracking surveys, although there could be more that we do not know about.. 24.

(25) 25. 1995. Rural Observatories, Bepako (Madagascar). 307 households; 1490 individuals. 912 households; 6204 individuals. 1354 households. 1262 families including an evermarried woman. 7000 households. 308 households. Baseline sample. 2005. 2004. 1998. 1988-89. 1997. 2000. Follow-up. Individuals. Individuals. Households. Ever-married women. Households. Males aged <30 at baseline. Target. 49%. 52%. 80%. 46%. 84%. 79%. 82%. 84%. 72%. 94%. Recontact rate Without With tracking tracking 57% 100%. Author's calculation.. Beegle et al. (2010). Alderman et al. (2001); Maluccio (2000). Haaga et al. (1994). Thomas et al. (2001). Rosenzweig (2003). Source. Recontact rates without tracking represent the percentage of target living in the same dwelling as in baseline and. which explains the 100% recontact rate.. recontacted. In Bangladesh, individuals we not actually tracked, information on movers was obtained from interviews of key informants in the initial village,. one member of baseline household.. Notes: Target is the sample sought to be recontacted at follow-up. Individuals means all members of baseline households. Households means at least. 1991-94. 1976-77. Malaysian Family Life Survey. Kagera Health and Development Survey (Tanzania). 1993. Indonesia Family Life Survey. 1993. 1981-82. Bangladesh Nutrition Survey. KwaZulu-Natal Income Dynamics Study (South Africa). Baseline. Survey. Table 2: Summary of tracking surveys in developing countries.

(26) 4.4 Tracking survey design We now turn to the description of a tracking survey designed to minimize attrition between 1995 and 2005 in one of the villages of the ROR. The other motivations for running this survey were to include in the data individual trajectories that are naturally excluded from panels without followup, such as migration or household changes, and to account for household structure instability. 13. as death, marriages or divorce occur.. The tracking survey was implemented in Bepako, a village situated in Northwestern Madagascar, in the Observatory of Marovoay. This village was rst surveyed in 1995 and has been surveyed since then every year until 2009. Bepako was chosen for the tracking because of the size of the 1995 sample (307 households) and because it was in fact a census, as all households in the village were surveyed. Bepako is located about 80 km from the third most important city of the country, Mahajanga, and 5 km away from Marovoay, the nearest town. It is situated at the heart of one of the irrigated perimeters of Madagascar, a large, at, paddy growing area, where irrigation infrastructures such as canals and pumps were developed during the colonial era. This area grows a very signicant amount of the total rice produced in Madagascar, and most of the population in Bepako is involved in the paddy growing sector, as farmers, landowners or day laborers. The indigenous ethnic group, the Sakalava, is traditionally a tribe of nomadic cattle raisers rather than farmers. Most of the inhabitants of the region are migrants from the East and the Central Highlands of the country, who came to the Mahajanga region as farm workers when the irrigated perimeter was set up. When the infrastructure and land were not longer state owned and run, they had to buy the land and became smallholders. The irrigation scheme faced a serious crisis in the 1980s and the farmers no longer had an easy access to inputs and agricultural tools. The Marovoay Observatory attempts to collect data to analyze the strategies implemented by the farmers to deal with these issues. The eldwork in Bepako was carried out in four steps, described in detail in the following sections.. 13. The tracking survey was carried out by Flore Gubert et Anne-Sophie Robilliard (UMR 225 DIAL/IRD), in. the context of the project Dynamique de la pauvreté rurale en Afrique sur longue période : le cas de Madagascar.. 26.

(27) 4.4.1. Census of Bepako. The tracking survey started with a census of the population of Bepako, in July 2005. The list of households living in Bepako obtained through the census was then compared to the theoretical household roster, that is, all households ever included in the ROR since 1995 (for one or more years). Households were classied according to whether they were still living in the village, had moved away, were new in the village, or were deceased, if all the members of the household had died since the last interview.. 4.4.2. Individual Trajectories survey. The next step was to administer an Individual Trajectory questionnaire to households still residing in the village. The questionnaire used was designed to collect information on the composition of the household every year since 1995. The form was pre-lled using the information available in the ROR panel with the years of presence and absence of each member of the household. The table was completed during the interview with the household head. He was asked to conrm or correct the information, and complete with new members, their entry date, age, sex, link to head, etc. Reasons and years of exit, entrance and absence of each member were recorded. There were also questions on transfers from the members who had left the households, a detailed module on child fostering, and a module on the shocks undergone in the previous years by the household, such as bad crops or deaths in the family. A French template and a lled-out Malagasy version of the questionnaire can be found in Appendix B. If the household head could not be found, another member of the household answered the questionnaire. If the entire household was gone, it was lled in by a neighbor or the village chief. In this case, a Household Tracking Form was used to record the date the household had left the village and his new address, as well as the maximum information that could be given to help the survey team nd these households.. If. the household was found in Bepako, but some individual members had left the village, then an Individual Tracking Form was used to collect the same type of information. Even without tracking individuals who moved out of the village, this phase brings valuable information to the data as it identies the households that had left and re-entered the panel and thus were recorded as distinct households. Thanks to this, the household attrition rate between 1995 and 2005 could be greatly reduced, just by matching households across waves.. 27.

(28) 4.4.3. Tracking and interview of migrants. The tracking forms were the basis of the third step of the project, the search and interview of movers. The aim was to nd all individuals who were living in Bepako in 1995 (and were thus included in the baseline sample) and who had left the village since then. Once they were found, they were interviewed using a standard ROR type questionnaire if they had stayed in a rural area or if their main activity was farming or cattle-raising. If they had left the agricultural sector and lived in a town or city, they answered a dierent, specic urban questionnaire. A particular choice that was made in this tracking was to limit the search to individuals who had stayed in the region: short- and medium-distance movers, but not long-distance movers. A major reason for restricting the tracking to the region is to be found in the specicity of migrants in Madagascar, where it is a very widespread practice to be buried in one's region of origin, in the same vault as the rest of the family. Many inhabitants of Bepako are in fact migrants or children of migrants, and they originate from far away regions, often the East of the country or the Central Highlands. When they reach old age, if they can aord it, they return to their native region to die and be buried there. Chances of nding them still alive were thus very slim for a high cost of research, which motivated the decision not to try to nd them. We will see whether this particular choice was justied. 4.4.4. ex post. in the next sections.. ROR survey in Bepako. Finally, all households living in Bepako (and enumerated in the census mentioned above) were surveyed during the regular ROR round, in the following months of December and January. This means that in both 1995 and 2005, all households of Bepako were interviewed.. 14. This is. benecial from the point of view of attrition as it includes all those who were lost from the panel but still living in the village, either because they were excluded in the exogenous sample size. 14. The chronology of the survey had an impact on the choice of a reference location for the baseline price system. used in the Paasche index, mentioned in section 3. Ideally, prices in Bepako should be used as we are comparing welfare between individuals who either live or used to live in Bepako. reference.. It thus makes sense to consider it as a. However, this is not possible because the ROR survey was carried out in Bepako in December and. January, as it is the case every year, while the tracking survey was done in July. The end of the calendar year is the hunger season in Madagascar, where paddy stocks are low, the next harvest has not been reaped, and prices are consequently the highest.. Dierences in prices between Bepako and the rest of the sample will not reect. regional but seasonal dierences. Taking December-January prices to deate July values will cause real incomes of households living in Bepako to be underestimated compared to those living in other locations, but these would be spurious conclusions.. We deal with this issue by making the assumption that prices in rural parts of the. Marovoay district (where Bepako is situated) are homogenous. Hence, the reference price system is calculated using the data from households living in the Marovoay district excluding those living in Bepako and Marovoay town.. 28.

(29) reduction in 1999, or because of identiers mix-ups, or because they had left the panel once and re-entered but were not identied has the same households. This also adds to the wealth of data because it creates two pictures of a village, ten years apart, which can be used to study the demographic evolution of the village, identify newcomers' characteristics, etc.. 4.5 Tracking results Results of the tracking survey are considered in terms of recontact rates. A recontacted household is one in which at least one person who was a member of that household in 1995 is found and interviewed in 2005. A household can also be deceased, if all members have died since 1995, or not found, if none could be found in the tracking. Figure 2 recapitulates the denitions of the dierent categories of households with regard to the tracking. In the Individual Trajectory phase of the survey, split-os were identied, that is, individuals who moved out and created a new household or joined an existing one. A recontacted household will yield as many interviewed households in 2005 as there were split-os, to the extent that they could be found and were in the tracking zone.. Table 3: Recontact rate at the household level. Number 258. % 84.1. Always in. 70. 22.8. In Bepako. 139. 45.3. 49. 16.0. Recontacted Outside of Bepako. Not found All members deceased Total. 42 7 307. 13.7 2.3 100.0. Source: ROR annual survey (1995-2005), Tracking survey (2005).. There were 307 households interviewed in Bepako in 1995.. This represents the baseline. population. As shown in Table 3, the survey team was able to recontact 258 households, of which 70 were core panel (always in), 139 had at some point left the panel but were subsequently recontacted and 49 were found outside of Bepako. All members of seven households were deceased over the 10-year period and the remaining 42 households were either not found at all in 2005 or had moved too far away to be recontacted. Conditional on being alive, the household recontact rate was thus as high as 86%. 434 households were interviewed in 2005. These households are either baseline households or split-os: they all include at least one member of the baseline panel.. 29.

(30) Figure 2: Household tracking results. Followed-up every year in ROR “Always in” 70 (22.8%) At least one member stayed in Bepako “Stayers” 209 (68.1%). Baseline population 307. Entire household moved out of Bepako “Movers” 91 (29.7%). Dropped out of ROR and recontacted in 2005 “Recontacted in Bepako” 139 (45.3%). At least one member found “Recontacted outside of Bepako” 49 (16%) No members found “Not found” 42(13.7%). All members deceased 7 (2.3%) Legend Definition “Name of group” Number (% of baseline population). Recontacted sample in 2005. Definition “Name of group” Number (% of baseline population). Not recontacted in 2005. We now turn to the tracking results at the individual level, as the tracking was undertaken on an individual basis, trying to locate not only all baseline households, but also all baseline individuals wherever their location in 2005.. There were 1490 individuals in the 1995 sample. that can be classied in three broad categories in the tracking survey: deceased, stayers and movers.. We call stayer an individual who, in 2005, was still residing in Bepako and in the. same household as in 1995, while a mover is a person who moved out of his household, whether he stayed in Bepako or changed location. For the time being, we do not dierentiate between local movers (inside the village) and migrants, because with regard to attrition, they would all be lost to follow-up without the tracking.. As the panel is at the household level and not at. the individual's, a person who leaves the household is dropped from the panel.. Even though. the survey was not designed to create an individual panel, this does not create attrition at the household level but it does at the individual level.. 30. Besides, such a person might be living in.