Sénégal, France, Italie, Espagne - Enquête sur les migrations entre l'Afrique et l'Europe - Sénégal (2008)
ID de référence | SEN-INED-MAFE-2008 |
Année | 2007 - 2008 |
Pays | Sénégal, France, Italie, Espagne |
Producteur(s) | Cris Beauchemin - Institut national d’études démographiques (France) |
Bailleur(s) | Septième programme-cadre de la Communauté européenne - - Institut national d’études démographiques (France) - INED - Agence nationale de la recherche (France) - ANR - Région Ile- de-France - IDF - Programme FSP "Migrations int |
Collection(s) | |
Métadonnées | Documentation au format PDF |
Créé le
Sep 12, 2017
Dernière modification
Sep 12, 2017
Affichage par page
663993
Echantillonnage
Méthode d'échantillonnage
SENEGAL
A three-stage stratified random sample was used. At the first stage, primary sampling units (census district) were selected randomly with varying probabilities. At the second stage, households were selected randomly in each of the selected primary sampling units (PSUs). At the third stage, individuals were selected within the households.
a) Selection of primary sampling units (first stage)
In the Senegal survey, the sample was designed to be probabilistic and representative of the Dakar region, and at the same time to maximize the chance of reaching households 'affected' by international migration (rare population). The sampling frame used to select the primary sampling units was the 2002 Population Census. The census districts (CD) -which are usually used as the primary sampling units in surveys in Senegal - have an average size of 100 households in urban areas. 60 primary sampling units were randomly selected at the first stage. This number of primary sampling units allows reaching a balance between a large dispersion of households (which decreases sampling errors) and a more concentrated sample (which reduces costs).
The region of Dakar was divided into 10 strata of equal size, according to the % of migrant households within each of them (in average, 11.6% of the households were 'migrants'). 6 CD's per stratum were drawn, with a probability proportional to the number of households within each CD. In other words, census districts with a large number of migrants were more likely to be selected than those with low numbers of migrants. This approach increases the number of migrants interviewed in the individual survey, while still having a probabilistic sample representative of the target area.
The listing of the households in the 60 selected primary sampling units was updated in order to select the sample of households. This stage was essential because a lot of changing occurred in some large neighbourhoods of Dakar since the previous census (2002), especially in suburban areas. This counting also allowed distinguishing between households with and without migrants.
b) Selection of households (second stage)
The following approach was used in MAFE-Senegal:
- Households were selected randomly (using systematic random sampling) from the updated list of households in the selected PSUs. Two strata were distinguished: the households with migrants and those without migrants. A maximum of 50% of households with migrants were drawn in each district. Selected households that could not be reached (absence, refusals,…) were not replaced during the fieldwork. Replacement would distort the computation of sampling weights, and could also lead to bias the sample. To take account of refusals and absences of households, 22 households were selected to reach an effective sample size of 20 households per CD on average (a total of 1 200 households in Dakar region was to be reached).
c) Selection of individuals (third stage)
Next, individuals were selected within households for the life history survey. In each household, individuals were classified into 3 strata (which do not overlap):
- Return migrants, who were aged 18 or over at their (first) departure (or whose age at departure is unknown) ;
- Spouses/partners of migrants (if the spouse/partner is not a return migrant himself/herself) ;
- Other people.
Then, a simple random sample was done in each household to select:
- Up to 2 return migrants (random selection if more than two in the households, all the return migrants were selected if not more than two in the household)
- Up to 2 Spouses/partners of migrants (random selection if more than two in the household)
- Another individual.
In Senegal, an additional condition was that people had the Senegalese citizenship at birth. This condition was dropped in the Ghana and DR Congo surveys.
Two types of questionnaires were used in the departure countries: the household questionnaire and the individual life history questionnaire.
- The first questionnaire was used among a representative sample of households in the target region.
- The second questionnaire was used among a sample of individuals in the selected households, targeting both return migrants and non-migrants. The household questionnaire was thus used as the sampling frame for the selection of individual respondents.
EUROPE
The objective of the survey was to obtain a sample 'as representative as possible' of the African populations (Congolese, Ghanaian, Senegalese) in the destination countries (150 individuals per origin and destination country). The way the sample was constituted may vary across countries, but some common principles were respected:
- The composition of the sample should be as close as possible to the population of (Congolese, Ghanaian, Senegalese) migrants in the country in terms of gender, geographic distribution, age, socio-economic category or occupation.
- One exception: the sample should be gender balanced. Males and females should be equally represented in order to allow gender analyses.
- Samples in origin and destination may be linked, but migrants with weak or no relationships at origin should not be excluded from the sample.
- Both documented and undocumented migrants should be represented in the sample.
For MAFE Senegal, 200 Senegalese were to be sampled in each country (France, Spain and Italy).
The quota method was preferred in France and Italy to collect information on Senegalese migrants. In Spain, a sample of Senegalese migrants was drawn in the Padron. This source appears as a unique sampling opportunity in Europe since it is annually updated and includes all migrants, even the undocumented ones.
The Senegalese sample in Europe was made up from two types of samples:
- a linked sample of respondents, whose contacts were obtained in Senegal through the household survey. The module D of the HH questionnaire was designed to help the collection of contacts. It had to be entered and transmitted securely to the destination countries before data collection started in Europe. The contacts validity was checked before the fieldwork began. For Senegalese in France, 156 contacts were collected but only 54% of them were actually usable (correct contacts but a large amount were not interviewed). In Italy, almost no contact could be used. The efficiency of this method varies a lot according to the context ;
- an additional sample made up in each country to reach the expected sample size. The size of the additional sample is unknown until the contacts in the origin country have been obtained. The method of selection of this sample varied across countries (selection in a municipal register, through migrant associations, street recruitment, by snowballing methods…).
Regions were selected to cover the largest possible population of Senegalese in the country.
* France *
- Target areas: 3 regions comprising 64% of Senegalese people in France (Ile de France, Rhône-Alpes and Provence-Alpes-Côte d'Azur)
- Sample size: 201
- Quotas: By age, gender and socioeconomic status
- Recruitment methods: Selection from contacts obtained in Senegal, public spaces, migrant associations, snowballing, interviewers' contacts
* Italy *
- Target areas: Lombardia, Emilia Romagna, Toscana, Campania
- Sample size: 205
- Quotas: By age and gender
- Recruitment methods: Selection from contacts obtained in Senegal, public spaces, migrant associations, snowballing, interviewers' contacts
* Spain *
- Target areas: 12 provinces: Almería (Andalucía), Alicante & Valencia (Comunidad Valenciana), Barcelona, Lérida, Tarragona & Gerona (Cataluña), Madrid (Comunidad de Madrid), Zaragoza (Aragón), Las Palmas (Islas Canarias), Murcia (Comunidad Autónoma de Murcia), Baleares (Islas Baleares)
- Sample size: 200 (+ an additional sample of 400 people expected in 2010)
- Quotas: Random sample from Padron
- Recruitment methods: Population register (Padron), contacts obtained in Senegal, interviewers' contacts
- Note: a quota was imposed and respected: the proportion of interviewed migrants living in areas with a large concentration of Senegalese residents had to be equivalent to the real proportion of Senegalese migrants living in those areas in the selected regions
Taux de réponse
SENEGAL
In the MAFE Senegal survey, the quantitative objectives were not completely fulfilled:
- The selected number of households was 1320 in Senegal, but only 1141 were successfully interviewed, including:
- Non-migrant household : 458
- Household with at least 1 returnee: 205
- Household with at least 1 current migrant: 617
- Household with returnee(s) and current migrant(s): 139
This represents a response rate of 86.4% (90% was expected).
For the biographic questionnaire, 1 387 individuals were selected, only 1 062 were interviewed, including:
- Returnees: 193
- Partners left behind: 101
- Other non-migrants: 768
At the end, the drop was of 23.6%, while a drop of 10% was expected.
The overall response rate in Senegal is 66.1%.
EUROPE
600 Senegalese migrants were successfully interviewed: 200 in France, 198 in Spain, and 202 in Italy.
Pondération
SENEGAL
The computation of sampling weights relies on computing sampling probabilities at each stage. The product of sampling probabilities at each stage gives the overall sampling probability. Taking the inverse of the sampling probability gives the inflation factor. These factors are adjusted (taking into account non-response and by trimming the weights). They are then normalized, so that the sum of weights is equal to the sample size.
The initial "household weighting" takes into account the sampling probability of the DR and then the households. It was corrected by the non-response by multiplying it by the opposite of the rate of answer, estimated by strata of the DR and strata of the household (only available variables for the answering and non-answering households).
The initial "individual weighting" is equal to the product of the rectified household weight and of the opposite of the sampling probability in the household. It also has corrected by the non-response by multiplying it by the opposite of the rates of answer of the individuals, calculated by homogeneous groups obtained by logistic regression. The explanatory variables kept for this regression were the strata of the individual, sex, age, the size of the household, and the rank of contact.
EUROPE
In the European countries, similar sample sizes were selected for males and females, resulting in an overrepresentation or underrepresentation in the MAFE samples. Similarly, older people were usually oversampled. For these reasons, post-stratification weights are computed to give each observation its proper weight and to match the samples as closely as possible to selected population characteristics.
Senegalese migrants in France and Italy were selected using quota sampling with quotas set by age groups and sex, while the sample in Spain was a probabilistic sample from the population register, augmented by Senegalese respondents identified through contacts in the household survey in Senegal. Given this mix, weights for Spain were also computed by poststratification and not by computing sampling probabilities.
The procedure for constructing post-stratification weights in the case of MAFE-Senegal consisted of the following steps:
- Computation of target population estimates
- Computation of initial inflation weights
- Computation of poststratification weight
In the MAFE data, all survey weights have been rescaled (normalized) so that the sum of weights corresponds to the sample sizes of households and individuals respectively while the mean of the weight variables equals one.
For further details about weights, please read the MAFE methodological note 6 entitled "Sampling and Computation Weights in the MAFE Surveys" (see related materials).