The scope of the Migration Household Survey includes:
1. Household demographic characteristics
2. Housing conditions
3. Household assets and expenditures
4. Household use of financial services
5. Internal and International Migration and Remittances from Former Household Members
6. Remittances from Non-Former Household Members
7. Return migrants
18 of the 37 states in Nigeria were selected using procedures described in the methodology report
Producers and sponsors
Zibah Consults Limited
University of North Carolina
Provided oversight for the sampling and fieldwork
Provided technical support in data organization and consistency checks
Provided support in data entry, consistency checks and preliminary analysis
The World Bank
A. Sampling Frame
The sampling frame was the 2006 National Population Census. For administrative purposes, Nigeria has 36 states and the Federal Capital Territory. These states are grouped into six geopolitical zones - the North Central, North East, North West, South East, South South and South West. The states in turn are divided into 776 Local Governments. The demographic and political characteristics of the states vary considerably. For example, the number of component local government areas in the states ranges from 8 in Bayelsa State (in the South South) to 44 in Kano State (in the North West). Likewise state populations vary widely from 1.41 million in the Abuja Federal Capital Territory to 9.38 million in Kano State. The National Bureau of Statistics splits the country further into 23, 070 enumeration areas (EAs). While the enumeration areas are equally distributed across the local government areas, with each local government area having 30 enumeration areas, the differences in the number of local government areas across states implies that there are also huge differences in the number of enumeration areas across states. Appendix table 1 summarizes the population according to the 2006 population census (in absolute and proportionate numbers), number of local government areas, and number of enumeration areas in each state .
Given the above, a stratified random sampling technique was thought to be needed to select areas according to population and the expected prevalence of migrants. The National Bureau of Statistics (NBS) provided a randomly selected set of enumeration areas and households spread across all states in the Federation from the 2006 sampling frame. Every state in Nigeria has three senatorial zones (often referred to as North, Central and South or East, Central and West). The NBS sample enumeration areas were distributed such that within each state, local government areas from each senatorial zones were included in the sample, with Local Governments in each state nearly evenly distributed between rural and urban areas. In all, a total of 3188 enumeration areas were selected. These enumeration areas were unevenly spread across States; some states in the North West (Kano, Katsina, and Jigawa), and a few in the South South (Akwa Ibom and Delta) had over 100 enumeration areas selected while others such as Imo and Abia in the South East, and Borno, Gombe and Taraba in the North East, had as few as 20 enumeration areas selected. This selection partially reflected the relative population distribution and number of Local Government Areas in the component states. Annex Table B shows details of the states and geopolitical regions, their shares in population of the country, the number of Local Government Areas and enumeration areas in each state and the number of enumeration areas given in the NBS list that formed the frame for the study.
B. The Sample for the Migration Survey
a. Sample Selection of States, Local Governments and Enumeration Areas
Originally, the intention was to have proportionate allocation across all states, using the population of each state in the 2006 Census to select the number of households to be included in the sample. But it was later recognized that this would not yield enough migrant households, particularly those with international migrants, especially as the total number of households that could likely be covered in the sample to was limited to 2000. Consequently, a disproportionate sampling approach was adopted, with the aim of oversampling areas of the country with more migrants. According to Bilsborrow (2006), this approach becomes necessary because migrants are rare populations for which a distinct disproportionate sampling procedure is needed to ensure they are adequately captured. Given the relative rareness of households with out-migrants to international destinations within the 10 year reference period (selected by the World Bank for all countries) prior to the planned survey, sampling methods appropriate for sampling rare elements were desirable, specifically, stratified sampling with two-phase sampling at the last stage.
Establishing the strata would require that there be previous work, say from the most recent Census, to determine migration incidence among the states. However, the needed census data could not be obtained from either the National Bureau of Statistics or the National Population Commission. Therefore, the stratification procedure had to rely on available literature, particularly Hernandez-Coss and Bun (2007), Agu (2009) and a few other recent, smaller studies on migration and remittances in Nigeria. Information from this literature was supplemented by expert judgement about migration from team members who had worked on economic surveys in Nigeria in the past. Information from the literature and the expert assessment indicated that migration from households is considerably higher in the South than in the North. Following this understanding, the states were formed into two strata- those with high and those with low incidence of migration. In all, 18 States (16 in the South and 2 in the North) were put into the high migration incidence stratum while 19 states (18 in the North and 1 in the South) were classified l into the low migration incidence stratum (column C of Appendix Table 1).
The Aggregate population of the 18 states in the high migration incidence stratum was 67.04 million, spread across 10,850 Enumeration areas. Thus, the mean population of an EA in the high migration stratum was 6179. In turn, the aggregate population of the 19 states in the low migration incidence stratum was 72.95 million spread across 12,110 EAs yielding a mean EA population of 6024. These numbers were close enough to assume the mean population of EAs was essentially the same. To oversample states in the high stratum, it was decided to select twice as high a proportion of the states as in the low stratum. To further concentrate the sample and make field work more efficient in being oriented to EAs more likely to have international migrants, we decided to select randomly twice as many LGAs in each state in the high stratum states as in the low stratum states.
Thus, 12 states were randomly selected with probabilities of selection proportionate to the population size of each state (so states with larger populations were accordingly more likely to fall in the sample) from the high stratum states. Then two LGAs were randomly selected from each sample state and 2 EAs per sample LGA (one urban, one rural) to yield a total of 12 x 2 x 2 or 48 EAs in the high stratum states. For the low stratum, 6 states were randomly selected. From each of these, 1 LGA was randomly picked and 2 EAs were selected per sample LGA to give a total of 6 x 1 x 2 or 12 EAs in the low stratum. This yielded a total of 60 EAs for both strata. Given the expected range of 2000 households to be sampled, approximately 67 households were to be sampled from each local government area or 34 households from each enumeration area.
So far, the discussion has assumed two groups of households - migrant and non-migrant households. However, the study was interested in not just lumping all migrants together, but rather in classifying migrants according to whether their destination was within or outside the country. Migrant households were thus subdivided into those with former household members who were international migrants and those with former household members who were internal migrants. Three strata of households were therefore required, namely:
1. Households with an international migrant: at least one person who was a member of the household since Jan. 1, 2000 left to live in an international destination and has remained abroad;
2. Households with an internal migrant: at least one person who was a member of the household since Jan. 1, 2000 left to live elsewhere in Nigeria (outside the sample LGA) and has not returned to the LGA; and
3. Households with no migrant: No member of the household has left to live elsewhere either within or outside the country since Jan. 1, 2000.
The selection of states to be included in the sample from both strata was based on Probabilities of Selection Proportional to (Estimated) Size or PPES. The population in each stratum was cumulated and systematic sampling was performed, with an interval of 12.16 million for the low stratum (72.95 million divided by 6 States), and 5.59 million for the high stratum (67.04 million divided by 12 States). This yields approximately double the rate of sampling in the high migration stratum, as earlier explained. Using a random start between 0 and 12.16, the following states were sampled in the low stratum: Niger, Bauchi, Yobe, Kano, Katsina, and Zamfara. In the high stratum, states sampled were Abia, Ebonyi, Imo, Akwa Ibom, Delta, Edo, Rivers, Lagos, Ondo, Osun and Oyo. Given its large population size, Lagos fell into the sample twice. The final sample, with LGAs and EAs moving from North to South (i.e. from the low to the high stratum states) is presented in Table 1 below.
The sample was concentrated in the South since that is where it was expected that more households have international migrants. It was expected that the survey would still also be reasonably representative of the whole country and of both internal migrant and non-migrant households through weighting the data. To this effect, field teams were asked to keep careful track at all stages of the numbers of people and households listed compared to the number in the actual sample in each stratum, at all stages of sampling; from the first stage of sampling states (the Primary Sampling Units, or PSUs), to Local Governments (LGs), and finally Enumeration Areas or EAs (see below). It is worth noting that the number of EAs to be selected from each sample LG is miniscule in every state compared to the total number of EAs in the state. Overall, the intended sample of about 2000 households would yield about 13 thousand persons, or only 0.00008 of the huge population of Nigeria. Eventually, a total of 2,251 households with 13,415 individuals were actually sampled.
The next steps, then, were to select the local government areas (LGs) and enumeration areas (the last stage or ultimate sampling units, or UAUs), and then finally select households from the sample EAs in the selected states in each stratum. To select local government areas, it was decided a priori to maintain a balance between rural and urban areas. In the Northern States where there was to be only one LG selected per sample state, this selection was made randomly. Fortunately, with the urbanization rate at approximately 50 percent, there is a near balanced distribution of rural and urban areas in the country. The distribution between urban and rural areas of the selected local government areas was therefore fairly equal, as expected. For the Southern States, initially only two Local Governments were provided for in the sample, so an attempt was made to select one rural and one urban centre in each state. In some cases though, the distinction between urban and rural areas was blurred by insufficient information. It was also considered useful to ensure that the two local government areas selected fall in at least two different senatorial zones to improve representation of cultural and other differences among peoples in each state. Where such differences were considered not too significant to warrant special attention, equal representation of rural and urban areas was prioritized.
c. Sample Selection of Households
The mean population size of the EAs in Nigeria in the 2006 census was around 6000 persons in both strata (and assumed to be slightly larger by 2009). The number of households is therefore near 1000 on average, and always more than 100. Given available resources, it had been determined that it would be possible (and sufficient) to list only 100 to 150 households in each sample EA. To do this, each sample EA therefore had to be partitioned using a defined procedure into an average of 6 to 10 segments before, one of which is randomly selected. This ordinarily would require local maps or landmarks, perhaps a listing of dwellings, consultations with local government officials or police, etc. But there were no adequate maps for most areas. In fact, only major cities such as Lagos and Abuja had such maps, mainly street maps. Here, the NBS sample of 3188 enumeration areas was useful since maps were not available from previous surveys. The NBS sample contained sample listings of about 10 households in each of the 3188 enumeration areas in its national sample frame, obtained with the the intention that in a proportionate random sample across all states all ten would be interviewed. But in the present context, with the adoption of disproportionate sampling, the team found the randomly generated list of households useful only in locating a segment of each enumeration area. Thus for each sample enumeration area, about 90 adjoining households (the number depending on the team and the distribution of settlements in each enumeration area) were added to the list of 10 NBS selected households to make a total of 100 or more households to be listed. This made it unnecessary to develop complex partitioning directions for field teams. The 10 original households in the NBS list for each enumeration area might or might not be in the final sample given that they are only 10 of 100 or so listed households, and only a maximum of 34 could ever be sampled (see below). The teams were allowed to take on the nearest adjoining 90 households to the 10 selected. The count could go in any direction for highly populated places. In some cases where the population of an EA is small, teams could take on other households that may not be exactly adjoining to make a list of 100 or so households. This way, it became irrelevant what the partitioning system could be for all households within the enumeration area and no new randomization or selection procedure for the partitioned enumeration area was needed.
To simplify weighting, it was decided to evenly allocate the sample of 2000 or so households across local government areas in the sample. For states with 2 local government areas, that meant having twice as much households in the sample compared to the states with only one local government area. Since most of the states with two or more local government areas are in the South with potentials for higher migration incidence, this achieves the basic purpose of oversampling households with migrants (international and internal). Dividing the total expected number of 2000 questionnaires by the 30 local government areas in the list gave an average of 68 questionnaires per local government area, i.e., 34 per EA. This was the original sampling plan, called Procedure A.
d. Household Listing and Sampling Procedures
Actual sampling of the households in the last stage of the 4-stage sample involved 2-phase sampling that, in the first phase, lists all households in a randomly selected part of the EA with about 100 occupied households (in both urban and rural EAs). Once a 'partition' was selected, the team listed the households in the partition, to show how many of the households were in the three strata - households with no (out-)migrant since year 2000, households with one or more internal migrants, and households with one or more international migrants. Households with no one in the eligible age group of interest, viz., thought to be involved in making their own migration decisions, taken to be persons aged 15 to 59, were excluded from the list of eligible households to be sampled. Thus only households with someone who had actually been aged 15 to 59 at the time of migration and who had left since January 1, 2000, and were still living away at the time of interview in 2009, were classified as households with migrants for the study. Households with only children under age 15 or persons aged 60 or more were recorded in the supervisor listing sheet summary but not eligible to be sampled even as non-migrant households. The reason is that the study of the determinants and consequences of migration-part of the goal of the survey was to provide data for that-would require that non-migrant households contain some adult aged 15-59, for comparison with the households which had someone leave in that adult age group. It was thus necessary to list more than 100 households, say 105 or 110 households, to have 100 at risk of having a relevant out-migrant or non-migrant. Even though such households are rare in Nigeria, this procedure was adopted to ensure an adequate sample size. The procedure also makes sense to allow for non-response.
Listing sheets and supervisor field control sheets were provided to team leaders. In addition, "supervisor sampling sheets" were prepared and provided to team leaders and supervisors for selecting (sampling) households from the three strata of households designated above.
Early in the fieldwork, it was determined that Procedure A based on the selection of 34 households from a listing of 100 to 150 per EA would not yield sufficient households with international migrants, and would, moreover, involve listing more households than necessary and hence involve longer and more costly fieldwork. This led to replacing Procedure A with Procedures B and ultimately C in the final stage of selecting households for interview in Enumeration Areas, the Ultimate Area Units (UAUs). Sections IVB,C describe and compare the three procedures below.
The number of households to be interviewed in a sample EA was originally set at 34 in Procedure A, with a fixed maximum of 12 households allocated to each of the two groups of households with international or internal migrants, and 10 to the households without migrants. Where the number of international or internal migrants was up to 12 in each group, all were selected automatically for interview. If the number was more than that, then 12 of each type of households were randomly selected from whatever the aggregate number was in the listing for each group. It was ensured that this was done while the team was in the EA, usually by the team leader, using a table of random numbers or other convenient random selection process. The balance, to achieve a total of 34 in the three strata combined, would be selected from the non-migrant household stratum. It was expected that there would always be sufficient number of non-migrant households. So, in an extreme case in which the quotas of 12 households each with recent international and internal migrants (as defined) were both filled, then only 10 households having no migrants need be selected for interview, to reach the maximum of 34 per EA. In such situations, the procedure would result in oversampling households with international migrants compared to those with internal migrants and non-migrants. But there was no a priori reason to expect that situations in which there were at least 12 households with recent international migrants would dominate. Thus if the listing produced, say, 2 households with international migrants, 40 with internal migrants, and 108 with no migrants, Procedure A would lead to samples from the three strata, respectively, of 2, 12 and 20, and hence few international migrants despite the large listing efforts.
With some concerns about non-response, it was allowed for teams to select up to 14 each from strata 1 and 2, and 11 from stratum 3, to ensure getting at least 12, 12, and 10 completed responses in most situations. Shortfalls in stratum 1 and 2 would be made up by increasing the number interviewed in stratum 3. This means a final range of 0 to 12 households in stratum 1, 0 to 12 households in stratum 2, and 10 to 20 households in stratum 3, in each EA (the maximum in stratum 3 being fixed at 20). The result of this process would be to select and hopefully interview up to 34 households in each sample EA. The total number of households in the survey would then be about 2000 or more. However, a major problem with this approach is that it guaranteed sampling and interviewing more internal migrants than international migrants, since there would be far more cases in which there would not be 12 households listed with qualified international migrants than 12 households with internal migrants.
Deviations from the Sample Design
A. Fieldwork in the Different Zones - Matters Arising
The fieldwork was originally scheduled to be completed between September 7 and September 30, 2009. It actually began on 14 September but significant changes in the methodology warranted that the teams be withdrawn from the field while the sample and implementation strategy was re-drawn (see above, the evolution from Procedure A to Procedure B to Procedure C). Following this and the adoption of the new sampling framework, the teams headed back to the field on the 5th of October, with a mandate to finish within three weeks, but the difficulties associated with getting to some of the sample communities were greater than anticipated, leading most of the teams to spend far more time in the field than was projected.
For states in the South East - Imo, Ebonyi and Abia states - led by Chioma Onwumelu, the interviews were ultimately conducted on a total of 405 households: 106 non-migrant households, 160 households with internal migrants, and 139 households with international migrants. (Details are found in Annex A.) These households were selected from a sampling frame near evenly distributed between the population living in urban and rural areas. The number sampled represents approximately 24 percent of the 1679 households listed by teams in the selected sample states of the zone. Almost all targeted households responded. A few cases required the intervention of the supervisor with the help of community authorities to encourage sample households to respond. Given the different teams' understanding of the local language and socio-cultural terrain of each region, only in rare cases were there challenges in communication between the team and respondents. Cooperation and understanding within the team was appreciably high. For example, when the team in Imo finished earlier than those in Abia and Ebonyi, they agreed to be posted to those states to assist with the work at no extra cost to the project. This was also due to the sterling leadership of the regional coordinator who had meetings with the teams before they departed to the field. During the meetings, each field worker was brought to see the success of the region as the success of his own state and therefore the success of his/her own fieldwork, creating both collegiality and collective responsibility. This proved invaluable at the different stages of changes in the field methodology described above, when exchanges of information among teams enhanced overall performance of teams in the region.
Despite this, the team confronted a number of challenges. For example, in some instances there was difficulty in obtaining correct information in sections of the questionnaire seeking personal (particularly financial) data. Understandably, many households were apprehensive about divulging information relating to household finances. In some cases, letters of introduction and identity cards issued to the enumerators and supervisors were not sufficient to convince these households; it took lengthy pleas and oaths from the fieldworkers to convince them to respond, even though most teams had already met with traditional local community heads and paid for publicity using town criers. In a few others when practicable, inferences were made from information on daily expenditures and what could be gleaned from the residence and its contents to estimate weekly or monthly expenditures.
Accessibility of some selected communities was also an issue, particularly in Ebonyi state where infrastructure is weaker than in the other states of the region, making it difficult to access some enumeration areas. Ezillo community in Ishielu Local Government (of Ebonyi State) was originally selected but involved in tribal wars so had to be dropped and replaced by another community with Oriuzo in Ezza North from the sample in order to not jeopardize the security of interviewers. Some areas were not easily accessible by car so interviewers had to use motor bikes or trek very long distances.
In the four states of the South South - Akwa Ibom, Edo, Delta and Rivers - led by Eric Onyebalu, a total of 1,642 households were listed out of which 523 or 32% were interviewed (see details for each state in Annex A). The teams were directed by the supervisor to first conduct physical identification of the enumeration areas. Thereafter, they had comprehensive listings of the households in the first set of enumeration areas (this was before the 50 percent increase in Local Government and Enumeration area coverage). Like the team in the South East, the South South team also demonstrated high mutual respect for one another which made it possible to overcome the many challenges they faced. Adherence to the provisions of the codes of conduct and mutual desire to resolve differences in team members' perceptions of the work proved critical in keeping them together and yielding regular progress.
Given that the communities and enumeration areas selected for the project in the states were randomly chosen, they were far apart, leading to significant logistic problems. In a number of cases, sample communities were not covered by mobile telephones, raising difficulty in communication between teams and the supervisor(s), on the one hand, and among the teams, on the other. Some of the states in the South South (the Niger Delta area) are volatile because of activities of militants. So security was a major concern in this zone. There were also challenges with transportation emanating from swampy neighbourhoods. At one point, a sample EA (and consequently the Local Government, Uvwie) had to be replaced with another one (Isoko South) on account of the combined challenges of security and navigability. Fortunately, the survey took place in the dry season, reducing logistic difficulties so interviews were successfully held in the rest of the LGs and EAs.
As in the South East, many respondents were not comfortable with some questions on personal incomes and family finances. The teams therefore had to use tact to extract relevant responses. To this end, examples cited and instances from role playing helped significantly to deal with such challenges in the field. Meanwhile, following the understanding that international migrants should be oversampled, the team most often employed their knowledge of the local environment and the reconnaissance listings to explore means of reaching households with international migrants. As all teams were required to deliver soft as well as hard copies of the responses, the South South team spent a lot of time following the end of fieldwork to translate responses on papers into soft copies. The reason for this is that it was not possible for them to key in data from the questionnaires at night due to the rigorous work of interviews. After observing the difficult work for the first few days, the regional supervisor excused the teams from the task of daily keying in responses. Fortunately, the team was also as diligent after the fieldwork to use the first couple of days in filling out the soft copies of the instruments. Overall, the fieldwork in the region was a success.
Sample states in the North included Kano, Katsina and Zamfara (in the Northwest), Bauchi and Yobe (in the Northeast) and Niger (in the North Central) geopolitical zones. Given the much lower out-migration rate in this part of the country, analyses on this study often treat them as a homogenous group. Fieldwork in the three regions was coordinated by Uchenna Amaeze. Much more than other parts of the country, the Northern countryside is characterized by states with large land masses and great distances between towns and communities. Some are so far apart that moving from one sample EA to the next could mean several hours' drive, requiring a relocation of the entire team and materials. Invariably, the result was a lag in work that could be up to three days before teams could settle down well enough to resume listing and interviewing in a new EA. As in other parts of rural Nigeria, some communities in the sample were not accessible by motor vehicle, so teams had to severally use motorbikes in areas not accessible to cars.
Teams in the North also experienced additional challenges with the relatively lower rate of literacy in sample areas. In some EAs, the use of local consultants notwithstanding, it took an unusually long time to finish a single interview as the interviewers had to spend a lot of time explaining to respondents the information they were asked to supply and why. This resulted in longer interview completion times, return visits to households for reconfirmations. Otherwise, respondents were generally very co-operative, not unconnected to the fact that most interviewers working in the region were from the region and had considerable experience in conducting surveys there. The teams also made things easier by paying prior courtesy calls to the local authorities and chiefs, who then introduced them to the community, soliciting that the people give them maximum co-operation.
The teams in the North also showed tremendous understanding given the myriad challenges that cropped up during the exercise. Five members of the team (independently and at different times and in different states) were involved in serious road accidents as they travelled through their designated EAs; one required medical attention and had to be replaced, others took a few days off, after which they were able to return to their work. Being far from project headquarters and far from one another in remote and hard-to-reach communities made it difficult for the regional supervisor to reach each team on time when the sampling methodology changed from Procedure A to B to C. Therefore, a number of the teams in the North simply went ahead with the original Procedure A listing, as against the changing procedures in other regions. In fact, by the time Procedure C was being implemented in Southern States, most teams in the North had already finished the interviews (see sections IVB, C below).
The Southwest sample consisted of three states besides Lagos - Osun, Oyo and Ondo. The fieldwork there was coordinated by Franklin Agbai. Anticipating a tedious and painstaking survey, the team quickly had a review meeting in Ibadan following the conclusion of training. Despite this, the supervisor was continually stretched as he attended to the diverse needs of teams for intervention on gray areas of the work and unexpected experiences with respondents and communities. Teams had to continually invent ingenious means of dealing with challenges, including household members (especially younger ones) asking for tips before allowing the administration of questionnaires, team members being mistaken for government workers on prying missions despite producing identity documents, etc. Fortunately most of the interviewer were young or could pass for students. So where convenient and could ease interaction with households, enumerators who were affiliated with universities in one way or the other would produce identity cards to that effect instead identity card and introduction letter from Zibah Consults.
Owing to the stress of the daily interviews, again it was not practicable to input data from the interviews on a daily basis as hoped. Teams were therefore allowed to wait till the end of the field work to start the data input process. However, team leaders and where possible the supervisor prior to that had to crosscheck the entries in the hard copies each day to ensure that mistakes were corrected while the team was still within reach of the respondents. Teamwork among the members of the region was appreciable. Owing to the high workload, Oyo had to engage one extra interviewer while Lagos engaged two more. They were trained in the field and in the listing and interviewing procedures. It is noteworthy that the work in Lagos was perhaps the most intensive of all the states in the country, given the double sampling of the state, leading to large number of local governments that had to be covered. Consequently, the team in Lagos first concentrated on listing households in the first set of local government areas assigned to them before beginning interviewing. They put in very commendable efforts to cover the work in good time.
The experiences of the different teams provide important lessons for the design and administration of future surveys. As one of the supervisors noted, time pressures led to insufficient work on design issues and experimentation before the actual fieldwork. The result was the need to later make costly changes after the teams were already in the field. This required more work of supervisors to ensure that not only were agreed-upon standards maintained by every team, but that this was done in a consistent manner. Communication problems led to significant avoidable cancellations and repetitions of procedures in the field. Any deficiencies in the time for adequate planning lead to costly adjustments in the field. This makes it imperative to conduct thorough and detailed preparatory work before going out to the field. Of course, no amount of preparatory work can completely eliminate chance occurrences in the field, but it minimizes them and enhances pre-emptive reaction to them. In particular, challenges came with the changes in sampling methodology. The changes in the sampling procedure, the coverage and selection of sample areas (states, local governments and enumeration areas) when teams were already in the field and had started fieldwork negatively affected the fieldwork. It took time to cascade the new instructions to the teams in the field, including pointing out the differences between the old and new methodologies and explaining the rationale behind the changes.
Dealing with the question of confidentiality and by extension obtaining sensitive information from respondents remains a continual challenge in household surveys of this nature. The Project Team would not say that it found the magic wand but it is proper to assert that understanding cultural sensitivities and winning the confidence of respondents is critical to obtaining meaningful results. The Nigerian survey recorded only one non-response from all teams. This was possible because every supervisor was made to appreciate the need to seek the approval and assistance of traditional local leaders. Consequently, local publicity including town criers backed by the traditional ruler and his chiefs helped secure the cooperation of the respondents. In one instance, the team was treated to a feast by the chief, who then gathered the sample households for interviewing. To a large extent then, not only was the sampling easy, but there were greater grounds for placing faith in the information obtained. On the other hand, in many urban areas, particularly in the south where the traditional governance institutions have been weakened, as in Lagos, Ibadan and other urban areas, teams had to resort to appeals, diplomacy and persuasion to obtain results from sample households.
A significant number of supervisors and interviewers appealed for reduced and more focused questions in the instrument. But the project management team in Nigeria made it clear it did not have the latitude for such significant changes as the project was continent-wide and had to be made comparable to other countries. So, changes were limited to minor adjustments in wording that contextualized the questions for Nigeria and yielded a better understanding of each question. While this procedure sufficed for the present project, the demand for greater clarity in the instrument (which we verified did not simply stem from the desire for reduced work) indicates the necessity of more extensive deliberations with potential participants in fieldwork before commencement of the survey in the future. While participants cannot be allowed to decide the contents of the instrument, it might help to have an outline of targeted outcomes and then have a wider audience make inputs into the nature of questions to yield better results in each country.
B. Modifications to the Initial Sampling Procedure in the Last Stage (Sampling Households)
As the fieldwork progressed, it was found desirable to make adjustments. The pilot study of two local government areas in two states - Enugu and Kogi - had showed the incidence rate of international migration to be about 10 percent (5 households out of 50 interviewed), 64 percent for internal migration (32 households) and 26 percent for non-migrant households (13 households). However, in the course of the actual fieldwork, some early household listings in states in the South (East) indicated that the relative number of households with international migrants was higher than observed in the pilot, with a few extreme cases being in excess of 20 percent. So it was decided that it would be excessive to list as many as 100-150 households to select 34. To address this, the number of households to be listed by teams was reduced to between 50 and 100, and the number to be interviewed pegged at about 20 for each EA, but including specific guidelines on how to oversample households (hhs) with international migrants relative to others, and households with internal migrants relative to those with no migrants. This took into account the expected relative prevalence of the three types of households. Thus it was decided, in a new Procedure B, to select a maximum (to reduce clustering effects, from large clusters) of 10 households with migrants in each sample EA. This meant that if there were fewer than 10 listed, all would be sampled and accepted to be interviewed, while in the few cases expected to have more than 10, the 10 would be selected randomly, keeping the maximum at 10. In stratum 2, the maximum would be fixed at 7, recognizing that most sample communities would have a tenth or more of their households with an out-migrant in the past 9 years to some internal destination. Then for stratum 3, we set a minimum of 3 non-migrant households and a maximum of 12. Thus in an EA with many out-migrant households, the maximum numbers of households with international migrants and internal migrants selected in the sample would be 10 and 7, leaving 3 non-migrant households to sum to 20. If there were a shortfall in the number of households with international migrants, the number in stratum 3 would be increased. For example, if 60 households were listed in an EA, one having an international migrant, 5 having internal migrants, and 54 having non-migrants, then the sample would comprise 1 international migrant household, 5 internal migrant households, and 12 non-migrant households, for a total of 17 households in the EA. If there were 0 international migrant households, 10 internal migrant households, and 60 non-migrant households, the numbers sampled in the three strata would be 0, 7 and 12, for a total of 19. The understanding was that there would usually be a higher number of households with no migrants than internal migrants and almost always a larger number of internal migrants than international migrants leading to higher sampling fractions for the latter in each case. Even with this, the total number of households with internal migrants in the final sample interviewed was still expected to be larger than the number with international migrants because the latter were far rarer. But generally, the difference would be less with this Procedure B than with the earlier Procedure A. Note this meant there would always be a range of 0 to 10 households to be interviewed in stratum 1, 0 to 7 in stratum 2 and 3 to 12 in stratum 3. Where there were no households in strata 1 and 2, the number in stratum 3 could be increased in each EA. This procedure was named Procedure B to distinguish it from the original procedure (termed Procedure A, explained earlier).
But after reducing the number of households to be selected per EA to 20, a new challenge emerged namely; the overall number of questionnaires to be generated given the total number of enumeration areas and the number of households within each enumeration area to be interviewed would yield less than the desired 2000 households. Thus, it was decided to increase the range of households to be interviewed in each household stratum and peg the number of households to be interviewed per EA at a maximum of 25 instead of 20. For households with international migrants, the maximum number was accordingly increased to 12, with the number of internal migrant households increased from 7 to 10. The minimum of 3 was retained for households without migrants, with the maximum increased to 25. This modified procedure was termed Procedure C, to distinguish it from the first and second procedures (A and B) for sampling outlined earlier.
Thus, household sampling was carried out using 3 different procedures, depending on when the EA was covered, in the search for an optimum oversampling process to identify the (relatively rare) sample households with recent out-migrants to international destinations. Procedure A involved sampling about 34 randomly selected households from a list of 100 to 150 households in each sample enumeration area. The number of enumeration areas completed with this depended on the speed of the team in each state (see discussion below in sections IV B, C). Most enumeration areas in the North were sampled using procedure A: all in Bauchi, Kano, Katsina, Niger and Yobe. Only Zamfara did not use A, and used B instead. Under procedures B and C (see discussion below in section IV B, C), between 50 and 100 households were listed in each sample EA, from which about 20-25 were selected by team leaders in the field using the supervisor sampling sheets.
In addition to the adjustments in household listing and sampling procedures, the number of local government areas in sample states was increased by 50 percent across the two strata of (high and low migrant) states. This had to be done in a manner that maintained the ratio of 2:1 in favour of households in the high stratum states to ensure oversampling of this group. This implied adding one local government area in each of the sample states in the South and adding one local government area for three of the selected six states in the North. For the latter, consideration was given to the existing coverage of the three geopolitical zones in the region i.e., North West, North Central and North East. Given its huge population, Kano State in the North West was automatically selected to have an additional Local Government from among Kano, Katsina and Zamfara in the zone. Niger state was the only state in the North West included in the sample, so was also selected to have an additional LG. Finally, Yobe was selected from between Yobe and Bauchi from the North East.
For the south, it was agreed that as many states as budgetary constraints could allow for should have an additional LG. Eventually, with prudent resource management, it was possible to add one LG to every state in the sample in the high stratum South (including 2 in Lagos to reflect its double sampling). Except where impracticable, the new set of LGs was again to be drawn from the original NBS list, but this was not possible on two occasions. The first was in Ebonyi State, where one of the communities selected was involved in communal clashes with a neighbouring community. The second was in Delta state where one of the selected communities was prone to activities of militants, crisscrossed by creeks, and had difficult swampy terrain. On both occasions, the supervisor for the region was allowed to randomly select another LG which did not necessarily have to be in the NBS list but which increased equitable coverage across the senatorial zones of the state.
C. Listing and Sampling in the States
As noted in the previous section, there were eventually three procedures for listing and sampling for the teams. The points of communication of these changes for the teams determined the relative number listed and sampled in each enumeration area. As at the time of the transition from Procedure A to Procedure B, a number of teams had completed a certain proportion of interviews, while other teams, which undertook to complete all the listings first, with the intent to sample households and begin interviews afterwards, were still able to adapt to Procedure B to sample households. Teams in the second group were instructed to not bother re-listing, but instead to adjust the numbers in the three lists to be sampled from. For those teams that were both listing and interviewing, the speed of each team determined the point at which it was instructed to change household listing and sampling procedures. The same issues arose at the time of the second change, from Procedure B to C, though the adaptations necessary were much less since the listing aspects were identical.
Within the ambit of each procedure, teams were given a range of number of households to list and sample. For example for all the Procedures (A, B and C), there was a range of about 50 households between the maximum 150 and 100 or 100 and 50 that enumerators could list. Likewise, a recommended maximum number was given for sampling. Ultimately, the number listed by each team depended roughly on the proportion of households with international migrants. For areas with low incidence of international migrants, more households were listed, to increase the probability of finding the population of interest, i.e., households with international and internal migrants (where the latter was also rare). In places where out-migration rate was high, only around 50 households were listed since the population of interest was more easily available for sampling. The ultimate number listed or sampled by each team depended in part therefore on the prevalence of international migrants being encountered during listing.
As such, in nearly all states in the south, listing and sampling proceeded with a combination of procedures. For example, Abia, Delta, Ebonyi, Imo, Ondo, Osun, Oyo and Rivers worked with a combination of procedures B and C, since the first change from procedure A to B happened while they were still listing. Thereafter, Procedure C was introduced when the additional Local Governments were added. In contrast, all the Northern States, where the LGs and EAs to be covered were fewer in number, the listing had proceeded quickly with procedure A partly due to the very low proportion of households with migrants (international and internal). All states in the North (except Zamfara) therefore used procedure A, which involved listing more households. The team in Lagos had also started with Procedure A in the first set of EAs, and was still doing listings when the announcement for the change from procedure B to C was made. So the sampling switched from Procedure A directly to Procedure C, leaving them without any EA sampled with procedure B. Edo state was also listing when the first change was made and therefore finished up its listing within the framework of the newly adopted procedure (B).
The above scenarios produced a wide range of listing and sampling procedures across states, but all consistent with the specified methodologies. The number of households listed ranged from 660 in each of Ondo and Osun (Lagos had 1084, but this is because it was sampled twice) to 102 in Zamfara (having only one Local Government and two EAs). Likewise the number interviewed ranged from 150 in Akwa Ibom to 70 in Katsina and 49 in Zamfara. The average proportion of households interviewed relative to those listed is shown in the last column as 30 percent.
Weighting the data is critical in this project given that the sample selected is not an epsem sample, in which all elements (households) in the sample had an equal probability of being selected. Rather, the sample was explicitly designed to oversample areas and households with international migrants. Therefore, it is not a self-weighting sample in which the (unweighted) sample mean of the values for sample households is a reliable estimate of the population mean. Instead, the sampling procedures used aimed at selecting a sample that would have sufficient numbers of households with recent migrants to international destinations, which are relatively rare elements in the large population of Nigeria (or most any national population). This required the use of special sampling procedures (see Bilsborrow et al., 1997; Groenewold and Bilsborrow, 2008) to ensure that the actual sample of households interviewed included sufficient households with migrants, i.e., much larger than their share in the national population.
Overall, the sample selection involved four-stages, the selection of states, the selection of Local Government Areas, the selection of Enumeration Areas, and the Selection of Households. Though they have been severally referred to and described above, we briefly review each selection stage as follows. The Selection of States was the first stage, which required stratifying all states in the country according to the expected proportion of households in each state with international migrants. Due to the lack of access to data on migration from the 2006 census (the most recent), expert opinion and available literature were relied upon to classify the 36 states and the Federal Capital Territory into high and low migration strata. Eighteen states were classified in the high migration stratum (16 in the South, 2 in the North), and 19 states were classified in the low migration stratum (18 in the North, 1 in the South). The proportion of the population in the high migration stratum to be sampled was then taken to be double that of the low stratum. This yielded a sample of 12 states out of the 18 in the high stratum and 6 states out of the 19 in the low stratum, with an approximate relative weight of 3/1 for all households in the low stratum compared to 3/2 for households in the high stratum, or double the weight for households in the low compared to the high stratum. To be more precise, the total population of the high stratum states in the 2006 census was 67.04 million, compared to 72.95 million in the 19 states in the low stratum. Therefore, stage I weights for households in the two strata are 67.04/12 = 5.5867 in the high stratum compared to 72.95/6 = 15.4917 in the low stratum. This approach achieves an additional benefit in that it stratifies the entire country by population so that the population of the sample can represent the population of the entire country
The second stage of the sample was the selection of Local Government Areas (LGAs), which constitute the next lower administrative units in states. To further concentrate the sample of areas among those expected to have a higher proportion of international migrants; it was decided to select two LGAs from each sample state in the high stratum and one LGA from each in the low stratum. As explained in Section IVB, it was realized that the initial number of LGAs was insufficient to produce the total sample size desired, which was corrected for by (1) increasing the number of LGAs in all sample areas by 50 percent, and (2) increasing the number of households to be selected at the last stage from 20 to 25 (moving from Procedure B to Procedure C, explained earlier). This translated to selecting 3 LGAs instead of 2 in each high stratum state sampled and 1.5 instead of 1 in each low stratum state. The final result was that All states in the South had three LGAs in the sample (Lagos had six as it was sampled twice). Kano, Niger and Yobe were selected from the low stratum states to have 2 Local Governments each instead of 1.
At the Local Government sampling stage, it was considered important to explicitly incorporate the rural/urban divide of the sample in order to better represent these two groups of the population. This was particularly important given that it was not incorporated into the selection of states and in any case, there is no exclusively rural or exclusively urban state in the country; each state has a share of both. Ordinarily, it could make sense to divide the LGAs into rural and urban, and then purposively sample one of each from each sample state. But according to Oluwasola (2007) and the United Nations (2008), Nigeria is one of the most rapidly urbanizing countries in the world. In 2006, the urban population was estimated to be 46 percent of the total, up from 11 percent in 1952. This implies that about 65 million of the country's 140 million lived in urban centres in 2006; so the urban population was estimated to be about half by the time of the survey in late 2009. Given the above, a random sample of Local Governments Areas would produce approximately the same statistical result as a purposive sampling if the target was to have equal numbers of urban and rural LGAs (discussed also above).
To weight the LGAs, the number selected in each sample state as a proportion of the total number in the state indicates the probability of selection, so the inverse is the LGA weight for the state. Thus every selected household in the sample LGA will have the weight indicated attached to it, in the final computation of household weights.
The third stage was the selection of Enumeration Areas from LGAs. The total number of EAs in Nigeria in the 2006 census is about 23,000, with the number per state varying from 180 to 1350 and the number in sample states from 360 to 1350. The reason for this variation is that the number of LGAs per state varies so much though the number of EAs per LGA remains 30. So it should have been desirable to obtain the census population of each sample LGA and of each EA in each sample LGA to know what the population in the sample EA represents. These data were also not available from the National Bureau of Statistics; only information on the population of each LGA was available. We thus assumed that the population of each EA in the sample LGA was approximately the same. Thus Given that 2 EAs were selected out of 30 EAs in every sample LGA, the probability of the selection of an EA was 2/30, and the weight for all sample EAs in the survey is straightforward, i.e., the inverse of 30/2 or 0.06666.
The last stage was the selection of households within each sample EA. As has severally been referred to in this report, the survey involved three distinct phases of selection of households from EAs, referred to as Procedures A, B and C. Procedure A involved listing about 150 households in an EA, identifying their migration status, and selecting randomly a total of 34. The instruction given at that time was that any sample household which refused the interview or for any other reason did not provide the data requested should be replaced by another household selected randomly from the list guaranteeing 34 completed households per EA. However, there was no actual replacement during the fieldwork using procedure A. Since quotas of a maximum of 12 households with international migrants, 12 with internal migrants, and 10 non-migrants were to be selected, with the sum being 34, the weights were to be computed as follows for those EAs in which Procedure A was used (mainly in the North). So if X1, X2 and X3 are the actual numbers of households listed in the three strata, and n1, n2 and n3 are the numbers actually sampled (following the procedures described in III B above), then the weight for each international migrant household selected in the EA is X1/n1, while the weights for each internal migrant and non-migrant household are respectively X2/n2 and X3/n3.
Procedure B involved listing 50 to 100 households in the sample EA, and recording the numbers in each stratum. In the first step, up to 10 households with one or more (recent) international migrants were selected. Where the total number of households with international migrants in the list was less than or equal to 10, all were selected, meaning each of those households had the same 100 percent chance of being selected. Where the number selected was less than the total number of households with international migrants listed in the EA, then the probability of selection was 10/X1, where X1 is the total number listed, and the weight for each household in stratum 1 is X1/10. In the second step in Procedure B, households with internal migrants were selected with the probability calculated as the ratio of the number selected to the total number in the stratum, allowing a maximum of 7. Again, if there were more households listed than 7, say Z, the probability of selection of each is 7 divided by that number, and the weight is the inverse, or Z/7 or Z/n2. The last step was to sample non-migrant households in the third stratum. For any numbers fewer than 10 in stratum 1 or fewer than 7 in stratum 2, there would be a corresponding increase in the sample size selected from stratum 3. This number was determined before drawing the sample, but would always be at least 3 since a maximum total sample size from each EA was set at 20, meaning that even with a maximum selected in strata 1 and 2 (17), three non-migrant households would be selected. Again, the weight is X3/n3, as before, but following the supervisor sampling sheet for Procedure B. Finally, Procedure C is similar in methodology to Procedure B except that the total number or “sample take” per EA was increased from 20 to 25 - with the numbers in the sample in strata 1 and 2 (international and internal migrant households) allowed to increase from 10 to 12 and 7 to 10, respectively. This left the residual for non-migrant households at the same minimum of 3 for stratum 3, but a maximum of 20. Note that in Procedures B and C, there was no replacement of non-responding sample households, so the weights in all cases in both procedures B and C (and as is normal in surveys where there is no replacement of non-responding sample households) can appropriately take into account that some households selected in the sample from the list may refuse, not be at home, etc.
To bring all the weights together from the four-stage sampling procedure, we multiplied the four weights for sample state, sample LGAs within the state, sample EAs within the LGA, and the appropriate stratum weight of the household in its sample EA. So for a particular household H in a state j denoted by Hj, the probability of selection would be the consolidated probability of the migration stratum in which the state it falls into multiplied by the weight of the LGA within the State multiplied by the weight of the Enumeration area within the LGA multiplied by the household weight based on its selection within the stratum of households that it falls into in its EA. So if weights for selecting states in stage 1 in the high and low strata are indicated by W1m where the subscript m=1,2 stands for either high or low strata; weights for LGAs in each sample state in stage 2 are given by W2L with subscript L indicating the sample LGA in the state; and weights for selecting EAs from sample LGAs in stage 3 are designated by W3e with subscript e representing 30/2. Finally, weights for households in the last stage are given by W4S, with the subscript S representing any one of the three household strata (international, internal and non-migrant households).
The total weight W applied to a household (or individual migrant in a household) in Stratum S and EA designated by e, within Local Government L in State m, is given by:
WmLeS = W1m * W2L * W3e * W4S
Dates of Data Collection
Data Collection Mode
The project management team at Zibah Secretariat, headed by a project coordinator provided overall intellectual coordination for the project and in consultation with the World Bank Migration team, took final decisions on all matters relating to the project. The project coordinator was assisted by a project manager responsible for day to day management of logistics and technical aspects the project. There was also an overall fieldwork coordinator who oversaw data gathering and later provided support for data cleaning and analysis. A regional coordinator provided oversight for each of the zones covered in the work.
Data Collection Notes
Interviewers were trained over a period of 2 days at the First Bank Learning Centre, Enugu. Training facilitators consisted of accomplished researchers who had all been previously involved in national surveys of similar magnitude. Participation by resource persons and potential enumerators was quite intense and appreciable - all but one of the enumerators (who had problems on the journey) completed the training. The training reviewed the background, objectives and rationale of the household survey for the project, overview of the proposed scientific methodology for the survey, and an item-by-item familiarization with the instrument for fieldwork, contextualization of issues raised and potential responses and case studies. The training also outlined the project management structure, including reporting lines, roles and conduct of each team member, potential challenges and implications, documentation and field reports as well as relationships with respondents. Reports from the pilot study also featured.
The fieldwork used an integrated system of interview and data entry with each team entering the information from the hard copy into soft copies at the end of each work day and submitting both hard and soft copies at the end of the fieldwork. However, a second set of data specialists was convened at the end of the fieldwork to re-enter all data and crosscheck all data for consistency. Field teams were mandated to provide daily feedback through the regional coordinators and team leaders. Clarifications were sought and rendered on ongoing basis.
· Questionnaire was designed by the World Bank; updated and contextualized by Zibah Consults
· Tested in a pilot survey conducted in Enugu and Kogi States of Nigeria
· Language of design was English. Enumerators however came from the local areas covered in the survey and were trained to elicit information as contained in the questionnaire in local languages
Data editing was conducted through a period of nearly two months. After data entry, professional data editors worked with the Zibah team to clean up the data and check for inconsistencies. The team corresponded with a number of other persons including Prof Richard Bilsborrow of the University of North Carolina and Prof Mario Navarrete in identifying and correcting all inconsistencies in the original data.
Data was originally entered in MS Excel but later transferred to STATA for analyses. Double entry and crosschecks were performed on the data at each stage.
Use of the dataset must be acknowledged using a citation which would include:
- the Identification of the Primary Investigator
- the title of the survey (including acronym and year of implementation)
- the survey reference number
- the source and date of download
Disclaimer and copyrights
The user of the data acknowledges that the original collector of the data, the authorized distributor of the data, and the relevant funding agency bear no responsibility for use of the data or for interpretations or inferences based upon such uses.