Task Force on the development of the European Household Survey (EHS) Luxembourg - 7 and 8 December 2006

Transcription

1 Doc T_EHS/09-06 Task orce on the development of the European Household Survey (EHS Luxembourg - 7 and 8 December 2006 A Data Collection Strategy for the European System of Social Survey modules (E4SM

2 1

3 A Data Collection Strategy for the European System of Social Survey modules (E4SM Kevin McCormack (Ire John Kinder (UK 1

4 1. Background Pressure continues to grow at an EU level for more and more social information above that which is currently collected as part of the main EU population surveys (i.e. LS, SILC and HBS. Other surveys have been or are in the process of being developed in areas such as, ICT, adult education (AES and time use (TUS. The topics to be examined by these surveys have been developed without regard to data that is already being collected at an EU level and as a result they have or are proposed to be developed as stand-alone surveys. This uncoordinated development of the ICT, AES and TUS has raised concerns at the EU and Member State levels. As a result, an EU Task orce (T consisting of CZ, DE, ES, R, IE, IT, HU, SI, SK I, UK, DG EMPL and DG EAC was convened to establish if is possible to construct an efficient survey model that would meet the requirements of ICT, AES, TUS as well as other social topics such as the European Health Interview Survey (EHIS. 2. E4SM The T had its first meeting at the end of June 2006 where a proposal that outlined the construction of an efficient survey model to meet the requirements of ICT, AES, TUS as well as other social topics such as the European Health Interview Survey (EHIS was developed. This proposal was presented to the Directors of Social Statistics (DSS of the National Statistical Institutes (NSIs at their meeting held on 18 and 19 September. The DSS agreed on the need to take active steps to develop a system of social statistical survey modules along the broad outlines of the project in the form of the E4SM approach presented in the document prepared by the Task orce. The original T proposal distilled down to two main options: 1 do nothing and allow the development of numerous large stand-alone surveys or 2 develop a new survey model that has at its core modules on various topics (i.e. respondents are asked to complete a number of mini surveys. This second option has been given the title E4SM (European System of Social Statistical Survey modules. It was also decided that in order to achieve robust results a minimum of 10,000 persons must participate in whatever survey type is agreed. The T members from the MS agreed that the E4SM could meet the ICT, TUS, AES, EHIS, etc. requirements by one of three designs but only if the original proposals for the content of these surveys where substantially reduced to only the most important variables. The frequency and topics of the E4SM modules are listed in Table 1 below. One should note that this approach foresees two types of modules namely annual (undertaken each year and irregular (undertaken at intervals to be determined It is envisaged that the total interview duration for the core variables is to be 10 minutes, the ICT 20 minutes and the other modules 20 minutes thus giving a total interview duration of some 50 minutes for a E4SM survey. (See Appendix 2 document Doc. T_EHS/07-06 /EN 2

5 Table 1: requency and topics to be covered by the E4SM social survey design requency Module (name Topics to be surveyed Core variable (T to decide Annual ICT EHIS Current ICT Chronic illness, BMI, alcohol consumption, physical activity, smoking, consultations with doctors and dentists. AES(1 To consist of two parts 1. Main AES component (data on household, education & training completed or participation, main job. 2. AES variables on education or training not completed, second job, situation one year before survey, parental education, obstacles to participation in education Irregular AES(2 To consist of two parts 1. Main AES component (data on household, education & training completed or participation, main job. 2. AES variables on access to information on learning possibilities, language skills, attitudes towards learning AES(3 AES module on social, civic, cultural and sports related participation. Long-term To answer new policy demands (developed well in advance. Short term To cater for acute policy needs identified 3. E4SM 3 models Three options for the implementation of the E4SM model have been discussed by the T and they are presented in Appendix 2 of Doc. T_EHS/07-06 /EN and are reproduced below for information. The first option proposes a survey of 10,000 persons with the field-work undertaken over an 8-9 week period early in the year with a total interview duration of 65 minutes. The modules would consist of the core, ICT, EHIS annual indicators, one longterm and one short term. 3

6 The second option again proposes a sample of 10,000 persons in total surveyed over a year with an interview duration of 50 minutes. The total sample is to be divided into 4 waves consisting of 2,500 persons with one wave surveyed each quarter. The core variables are surveyed each quarter; the ICT and EHIS are surveyed in two quarters while the long & short-term modules vary each quarter. This approach would mean that 10,000 persons will answer the core questionnaire; of these 5,000 would answer the ICT and the other 5,000 the EHIS. And of the 5,000 answering the ICT or EHIS half of them (2,500 will answer one set of the long & short-term modules and the other half a different set. The third option is termed dual data collection by the T. This option proposes a sample of 20,000 persons surveyed over a year with an interview duration of 50 minutes. The sample is to consist of two sub-samples of 10,000 persons. Each subsample is to be divided into 4 waves consisting of 2,500 persons and two waves will be surveyed at different periods within each quarter. Each sub-sample will survey the core variables. One sub-sample will have an ICT focus and the other an EHIS focus. Again individuals in the different waves will be asked varying sets of long & shortterm modules. Applying the modules to existing surveys The modules plus core variables could be delivered by incorporating them into existing surveys run by member states. The variable (or irregular modules would comprise three modules on AES (Adult European Survey and different modules, which will be divided into long term and short term modules. Long term would be planned in advance and short term would meet acute policy needs. It will be for member states to identify how the core variables and modules can be included in an existing survey. However, member states will be required to provide the modules together with their associated core variables. It is recognised that appending different modules to different surveys will not allow cross referencing of variables from different modules. 4. Modules The procedure for choosing the themes of the modules, prioritising them, and for developing their content is to be clearly defined. Several ideas were advanced how this could be done. By way of example, the following procedure is suggested by Eurostat. Each year, during the bilateral meeting at high level between Eurostat and the client DGs of the Commission (the annual Hearings meeting, a list of possible candidates for modules of the E4SM would be proposed by the DGs. A consolidated list based on the different proposals would be prepared by Eurostat and discussed by the E4SM Task orce who would establish a document with proposed priority topics for the following years. This last document would be discussed at a meeting of the Strategic Development Group 1 that would decide on priorities and on the T/Working Group that would take responsibility for developing the variables/questions to be included in the module. Early involvement would be necessary both for operational and administrative reasons. 1 The Strategic Development Group is considered an appropriate forum for this discussion as it comprises users, funders and producers of data - combining member state (DSS and Commission representatives (notably DGs for Employment, Social Affairs and Equal Opportunities; Education and Culture; Health and Consumer Protection; Information Society and Media; Justice, reedom and Security alongside Eurostat. 4

7 Alongside the Core Variables, ICT, EHIS annual indicators and AES, a range of topic areas were presented as example modules for discussion by the T. Subject to comments on detailed technical/methodological content, the T broadly classified these as shown in the following table: Category Module Reasoning 1. easible Core Variables; ICT; EHIS annual indicators; irst, second and third AES modules; Care services; Health 2. easible, with difficulties 3. Impossible in the short term, but could be considered in longer term. Victimisation and security; Consumer protection; ood intake; ertility ; Youth Discrimination Sample size constraints (need to cumulate or oversample; Inclusion of multiple attitudinal questions; Existing or administrative sources may be more suitable. Practical and legal constraints on fieldwork (other than age/sex discrimination 5

8 Illustrative scenario 1a: a single data collection instrument organised at a fixed period in the year In this illustrative scenario, the whole data collection is carried out over a limited time period (for example over a two-month period in Spring each year. This could be thought of as a traditional survey approach. A sample of around individuals in each country is interviewed during the same period and all respondents are asked exactly the same modules (Core Variables, ICT, EHIS annual indicators and one to three short/long term modules to answer new policy requests. ieldwork is to take place in Spring given the constraints of the ICT project to have the data available in October each year. Possible variants of scenario 1a: The total duration in this illustrative example is based around a target of 65 minutes and uses the available information on possible duration of the various content elements. If adequate results could be obtained, a more flexible interpretation of this model might include asking half the sample the ICT questions and half the sample the EHIS annual indicators questions, reducing the total interview duration for each individual respondent. ieldwork Sample Duration Content * Nature Spring Core variables 20 ICT ixed each year 15 EHIS annual indicators 20 Module(s: 65 min (long-term (short-term Variable each year *Note: Long-term module = planned in advance; short-term module = acute policy need identified. 6

9 Illustrative scenario 1b: a single data collection instrument organised in waves In this illustrative scenario, the whole data collection is carried out in different waves. An advantage of this approach is that it allows data collection to be spread over the year, which may be less of a burden for some countries. In the table provided below as an example, the annual sample of individuals is divided into four sub-samples of individuals interviewed in waves (ebruary, April, September, November. The individuals belonging to the same wave will answer the same module questions, but individuals from different waves will be asked either the same questions or different modules questions depending on the needs, although all will answer the Core Variables module. Thus in the example provided, the individuals of the ebruary wave and those of the separate April wave will each answer the same ICT questions but may be asked questions relating to different modules. Similarly, in the example in the appendix the September and November waves will both be asked the EHIS questions, but may be asked different modules. Possible variants of scenario 1b: In this illustrative scenario, the total duration for any respondent is 50 minutes. The scenario gives great flexibility to countries to organise fieldwork as they consider appropriate. or example, in a given year it might be desirable to combine sample waves together and effectively conduct a single survey in Spring (something like scenario 1a, whilst the following year it might be preferable to spread the samples across the year in waves as suggested, then in another year it might be optimal to keep some waves together but spread others. 7

10 Illustrative scenario 1b: a 'single' data collection instrument organised in waves ieldwork Sample Duration Content * Nature ebruary Core variables ixed each year 20 ICT ixed each year 20 Module(s: (long-term Variable each year (short-term 50 min. April Core variables ixed each year 20 ICT ixed each year 20 Module(s: 50 min. (long-term (short-term Variable each year September Core variables ixed each year 15 EHIS annual indicators ixed each year 25 Module(s: (long-term Variable each year (short-term 50 min. November Core variables ixed each year 15 EHIS annual indicators ixed each year 25 Module(s: 50 min. (long-term (short-term Variable each year *Note: Long-term module = planned in advance; short-term module = acute policy need identified. Wave approach allows modules to be conducted repeatedly, or separate modules organised at separate periods of the year bearing in mind that each wave comprises a sample of different persons. 8

11 Illustrative scenario 2b: a dual data collection instrument organised in waves In this illustrative scenario, a dual model is developed with two strands - one strand is centred upon the Core Variables and the ICT module, and the other is focused around the Core Variables and the EHIS annual indicators module. In the specific example presented, the whole data collection is carried out in different waves. In the table provided below as an example, the annual sample of individuals is divided into two strands, each comprising individuals. Strand one is organised into four sub-samples of individuals interviewed in waves (e.g.. March, April, September, November. Strand two is organised into four sub-samples of individuals interviewed in different waves (e.g.. Jan/eb, May/June, October, December. The individuals belonging to the same wave will answer the same module questions, but individuals from different waves will be asked either the same questions or different module questions although all will answer the Core Variables module. Strand one samples will all answer the ICT module questions. Strand two samples will all answer the EHIS annual indicators questions. Thus in the example provided, the individuals of the March wave and those of the separate April wave will each answer the same ICT questions but may be asked questions relating to different modules. Similarly, in the example in the appendix the October and December waves will both be asked the EHIS annual indicators questions, but may be asked different modules. Possible variants of scenario 2b: In this illustrative scenario, the total duration for any respondent is 50 minutes. This scenario gives great flexibility to countries to organise fieldwork as they consider appropriate. or example, in a given year it might be desirable to combine sample waves together and effectively conduct a single large survey, or two parallel surveys, in Spring (something like scenario 1a again, whilst in other years it might be preferable to spread the samples across the year in waves as suggested, then in another year it might be optimal to keep some waves together but spread others. 9

12 Illustrative scenario 2b: a 'dual' data collection instrument organised in waves ieldwork Sample Duration Content * ** ieldwork Sample Duration Content * ** March Core variables Jan./eb Core variables 20 ICT 15 EHIS annual indicators 20 Module(s: 25 Module(s: (long-term V (long-term V (short-term (short-term 50 min. 50 min. April Core variables May/June Core variables 20 ICT 15 EHIS annual indicators 20 Module(s: 25 Module(s: (long-term V (long-term V (short-term (short-term 50 min. 50 min. September Core variables October Core variables 20 ICT 15 EHIS annual indicators 20 Module(s: 25 Module(s: (long-term V (long-term V (short-term (short-term 50 min. 50 min. November Core variables December Core variables 20 ICT 15 EHIS annual indicators 20 Module(s: 25 Module(s: (long-term V (long-term V (short-term (short-term 50 min. 50 min *Note: Long-term module = planned in advance; short-term module = acute policy need identified. Wave approach allows modules to be conducted repeatedly, or separate modules organised at separate periods of the year bearing in mind that each wave comprises a sample of different persons. **Note: = fixed each year; V = variable each year 10

13 5. Survey Units 5.1 Target Population Most micro-level income statistics are confined to the population living in private households. This is the case with EU-SILC, and is expected to be so for E4SM. Excluded from the target population of private households are all persons living in collective households or in institutions on a permanent or long-term basis, and persons temporarily in collective households or institutionalised but not included as members of any private household on the basis of criteria described below. A collective household refers to a non-institutional collective dwelling such as a boarding house, dormitory in an educational establishment or other living quarters shared by more than five unrelated persons. Also included are persons living as lodgers in households with more than five lodgers. However, unrelated individuals (even if in large groups sharing accommodation under a private arrangement are included in the private household sector. An institution refers to old persons home, health care institutions, religious institutions (convents, monasteries, correctional and penal institutions. Basically, institutions are distinguished from collective households, in that in the former, the resident persons have no individual responsibility for their housekeeping. In some cases, old persons home can be considered as collective households on the basis of this last rule. Generally, the target population covered should aim to include all private households throughout the national territory of each country. However, often there have to be accepted some minor exclusions from full geographical coverage. Examples from EU- SILC, for instance, are overseas departments and territory in rance, Ceuta and Melilla in Spain, overseas departments and territories (DOM-TOM in rance, and minor islands in Italy, Ireland and Britain. The completeness of coverage is determined by the quality of the available sampling frame used for selecting this sample. or instance the use of an electoral roll, which does not cover households recently arrived in the country and those not on the roll for some other reason, results in the exclusion of such groups from the survey. Similarly for new residences constructed since the frame was last created or updated; or the use of population registers which are incomplete or out-of-date. 5.2 Household: the basic unit of sampling, data collection and analysis Households form the basic units of sampling, data collection and data analysis. It is important to clearly define and consistently implement criteria for the grouping of individuals into households. This requirement is common to all surveys using households (or other such units for sampling, so as to ensure that individuals in the population of interest are correctly covered in the survey, without omission or doublecounting Living arrangements The first issue is to choose a harmonised definition of household for E4SM based on sharing of living arrangements and resources. A number of alternative concepts are available. 11

14 Income Sharing In the context of income, the assumption is made that household income gives an accurate reflection of the living circumstances and financial well-being of all members of the household. This implies that there is income pooling within the household. or the purpose of the present survey the household would optimally be defined as: someone living alone, or a group of people who live together in the same private dwelling and pool their income. The income pooling criterion is far from clear-cut, however. The strict adoption of the income pooling criterion, then, leads to anomalies and difficulties in operationalising the household concept at the fieldwork stage. It leaves open the question of the extent of income pooling that is to be required. If this definition is strictly applied, then a married couple with two separate bank accounts might be considered as two separate households for our purposes, since not all of their income is pooled. Similarly, grown children who contribute to household costs from their own income but have separate savings accounts, do not pool all of their income. In such cases, even families might contain several income-pooling units, creating huge practical problems with survey administration. Co-residence There are two concepts of the private household that are frequently used in different countries and surveys: co-residence and common consumption (the housekeeping concept. A household defined by the housekeeping concept will be a subset of a household defined by the co-residence concept. It should be noted that that for census purposes, countries which use the co-residence concept are able to aggregate households defined using more restrictive criteria into the more general household defined simply on the basis of co- residence. However, the reverse is not necessarily the case, in that the countries using the wider co-residence definition may not be able to produce statistics based on a more restrictive definition. or instance, in countries using registers the more restrictive housekeeping-household cannot readily be identified. or this reason, it is normally recommended that the co-residence concept be used for priority statistical tables in the Census context. It will not be possible to aggregating housekeeping units into households defined in terms of co-residence in sample surveys, however, unless all members of a particular dwelling are sampled. In any case, since E4SM will most likely require the household as a unit for analyses with the assumption of shared resources within the household the dwelling concept is too broad and unsuitable, since there must be some economic sharing among members to make the assumptions inherent in the design of E4SM valid. The co-residence concept may well be (and often is suitable for the definition of ultimate sampling units, nevertheless, within each such unit substantively more meaningful households have to be identified for the purpose of the survey data collection and analysis. This applies even when all households at a sample address are taken into the survey. This requirement is of course essential if households are to be sampled within addresses. Co-residence and sharing of provisions A definition for the household has already been developed in recommendations for the EU Household Budget Surveys and the EU-SILC surveys and for consistency 12

15 purposes across all EU household surveys it is, therefore, logical to adopt the same definition. The definitions of what constitutes a household is achieved by adding criteria from 1 to 4 below: 1. Co-residence (living together in the same dwelling unit 2. Sharing of expenditures including joint provision of essentials of living 3. Pooling of income and resources 4. The existence of family and emotional ties or Household Budget Surveys and EU-SILC surveys, Eurostat recommend defining a household using the first two of these criteria as we have seen, the third can be ambiguous to apply and the fourth would be unduly restrictive. Joint provision of the essentials of living includes primarily housing costs but also food, clothing etc. to each other. The HBS and EU-SILC recommendations define the household as: a person living alone or a group of people who live together in the same private dwelling and share expenditures including the joint provision of the essentials of living. This household definition retains the criterion of persons sharing a common accommodation or address, but although the members are not required to be related to each other they may be required to fulfil one or more conditions of living together which would indicate a social unit of some kind. Such conditions should provide a plausible proxy for the assumption of shared economic well-being. Sharing in housing facilities and other living expenses would normally be the determining criterion. This includes people who benefit from housing costs paid by others (particularly children and persons with no income as well as persons contributing to housing costs or other living expenses from their own income. However, some countries use other indicators of joint provision of the essentials of living, such as sharing at least one meal each week. To some extent, such indicators of housekeeping are country-specific and cannot be fully harmonised Usual residence The general definition of the household given above needs to be elaborated in concrete terms. In particular, it is necessary to specify how individuals with different living arrangements are to be classified. There are a number of categories of people whose membership of a household needs to be defined in a harmonised fashion across the EU, because the way in which they are treated influences level and distribution of the responses to household surveys. These include boarders, people temporarily away in education, and the unrelated persons living in the household. It is important to have a harmonised treatment of such persons. The basic issue is to determine what constitutes an individual s usual place of residence as the concept of place of usual residence is a necessary tool for allocating persons to households and regions. This variable captures where the respondent actually lives, the de facto approach. It should reflect neither where some legal rules or instruments define place of residence of a person, (often called the de jure ap- 13

16 proach, nor subjective measures like where the person feels to belong or has his/her centre of interest. The EU-SILC survey recommends defining a person who has stayed most of his/her time in the household over the past six months would be as a household member, while a person who has stayed away most of the time over the past six months would be excluded as a non-member. It is recommend that the E4SM adopts the above EU- SILC recommendation for determining household membership. This is a general rule, which may be modified in particular circumstances as described below Intention to stay Account has also to be taken of what may considered as permanent movements into or out of households. Thus a persons who has moved into a household with the intention to stay for a period such as 6 months or more could be considered a household member, even though the person has as yet not stayed in the household for, say, 6 months, and has in fact spent a majority of that time at some other place of residence. Similarly, a person who has moved out of the household to some other place of residence with the intention to stay away indefinitely or for, say, 6 months or more, would no longer be considered a member of the previous household. In the application of these criteria, the intention would be to minimise the risk that individuals who have two private addresses at which they might potentially be enumerated are not double-counted in the sampling frame. Similarly, the intention would be to minimise the risk of some persons being excluded from membership of any household, even though in reality they belong to the private household sector Correct coverage of the target population Correct treatment of persons with different living arrangements is necessary both for substantive reasons and for ensuring complete and correct coverage of the population of interest. Complete and correct coverage can only be achieved if each person in the target population is included in the survey with a known and non-zero chance. The achievement of this objective is also affected by the sampling procedures adopted. These requirements can be violated if, for instance, the same survey unit (person, household can be selected through more than one sampling unit (e.g. addresses, or if some eligible units are not associated with any sampling unit at all. Appropriate sampling frames, designs and selection procedures have to adopted to ensure coverage of the population, in fact to achieve what is called a probability sample. Attention needs to be paid to certain special situations which may affect the completeness and correctness of the coverage. These problems arise when one-to-one correspondence is lacking between the ultimate units of sample selection (such as addresses and the substantive units of interest (households and persons, or when some units of interest are missed from the frame. Examples of the former include giving households a chance of selection from more than one addresses (from the main residence as well as from secondary residence, holiday home etc; or sub-sampling households within multiple-occupancy addresses but without compensating for that in the weighting of the sample data. Groups subject to the risk of under-coverage include unrelated persons (such as students co-residing at an address under a private arrangement, but miss-classified as a collective household and hence exclude from scope of the survey of private households. Another category of persons subject to 14

17 high risk of exclusion are persons who do not fully meet all the requirements of household membership at the place where they live (such as live-in domestic workers, but have no other place of residence from which they could have been selected Use of criteria in combination Some categories of persons may have to be included as members of a household even though they do not strictly meet the residential and presence criteria noted above, because of other considerations, because they are eligible for inclusion but would not be included at any other place of residence, or because they clearly have strong ties with the household concerned such that it would be illogical or clearly counter-intuitive not to include them. or instance, as noted in the guidelines for EU- SILC, certain categories of persons would normally be included irrespective of the length of absence: a person with spouse a household member, who considers this address to be his/her main residence; an unmarried child under a certain age (say <21 away for full-time education, but retaining close contact and ties with their parents household; or a person who has been a resident household member in the past, is presently institutionalised, but definitely expects to return to the household within a certain period. It was also recommended in EU-SILC to include other past members, who are expected to be away for a limited period (less than a total of one year and continue to consider this their main place of residence Treatment of borderline cases Students Particular difficulties are posed by the position of students living away from home. If they live away from home and do not share in the household expenses of the parental household (including benefiting from those expenses then they should not be regarded as household members. The criterion of length of absence may also need to be applied in the case of students who are financially dependent on their parents, to avoid the situation where they have the potential to be selected at two addresses. As with the other categories of people who are temporarily absent, a financially dependent student is enumerated at the parents household if he or she is away for 6 months or less, or if he or she does not have a private address elsewhere (for example, she may be living in student collective accommodation. Since people living in collective accommodation will generally be omitted from a household income survey, this special treatment for students is required in order to ensure consistency of coverage across countries and within countries despite differences in living arrangements. Other borderline cases Table 5.1 sets out how the multiple criteria discussed above should be interpreted including the treatment of the borderline cases. 15

18 Table 5.1: Household Membership: Treatment of Specific Situations Type of member Usually resident, related to other members Usually resident, not related to other members Recommended treatment Rationale Member if shares household expenses * Member if shares household expenses * If expenses are not shared, then the person constitutes a separate household at the same address Resident Boarder, lodger, tenant Visitor Live-in domestic servant, aupair Resident, absent from dwelling in the short-term (e.g. holiday, work, education Children of household, in education away from home Residents away for longer periods: Long-term absence with household ties: working away from home Member if: shares household expenses* and either currently has no private address elsewhere or duration of stay is 6 months or more Member if: (a shares household expenses* and either (b currently has no private address elsewhere or (c duration of absence is 6 months or more Considerations are whether the person shares in household expenses and the person s potential to be selected as a member of another household Residents away from home in institutional or collective accommodation would be considered household members, as long as there is some financial tie to the household. If the person who is temporarily absent is in private accommodation, then whether they are members of this (or their other household depends on the length of their absence Long-term absence with household ties: in hospital, nursing home, boarding school or other institution * Shares in household expenses includes benefiting from expense (e.g. children, persons with no income as well as contributing to expenses. 16

19 5.2.7 Recommended household definition The following set of recommendations are made for E4SM. (1 Complete and correct coverage of the population: The coverage will be confined to the population living in private households. Persons in collective households or various types of institutions, will generally be excluded. rom the sampling point, the fundamental requirement is to ensure that every individual in the target population (residing in a private household is associated with one and only one household and is selected into the sample in association with one and only one ultimate sampling unit (be it an address, a household or a person, with known and non-zero probability of selection. (2 Definition of a household: In line with the 2003 EU-SILC Recommendations, it is recommended that a household be defined as: A person or a group of people who live together in the same private dwelling and share expenditures including the joint provision of the essential of living (3 Usual residence: A person (sharing in the joint provision of the essentials of living will be considered a normal resident (hence a member of the household if he/she spends most of his/her daily night-rest there, evaluated over the past six-months (4 Changes in place of residence: Persons forming new households or joining existing households will normally be considered members at their new location; similarly, those leaving to live elsewhere will no longer be considered members of the original household. The above mentioned past six-month criteria would be replaced by the intention to stay for a period of 6 months or more at the new place of residence (5 Special situations: The above criteria should be used to deal with a vast majority of cases. However, especially to ensure that the sampling requirement (1 is met: O The above criteria may be applied in combination when not all are strictly satisfied; and O Additional or alternative considerations may be evoked to deal with border-line and other special situations. 17

20 5.3 Constructing a sample of households and persons In terms of the units involved, four types of data are involved in the E4SM: (i variables measured at the household level; (ii information on household size and composition and basic characteristics of household members; (iii complex variables measured at the personal level, but aggregated to construct household-level variables; and (iv variables collected and analysed at the person-level. or set (i-(iii variables, a sample of households including all household members is required. Households and household members Among these, sets (i and (ii will normally be collected from a single, appropriately designated respondent in each sample household using a household questionnaire for set (i and a household member roster for set (ii. Alternatively, some or all of these may be compiled from registers or other administrative sources. Household and personal complex variables Set (iii concerning mainly but not exclusively the detailed collection of household and personal data on ICT, etc., must be collected directly at the person level, covering all persons in each sample household. In most countries, these variables will be collected through personal interviews with all adults aged 16+ in each sample household. This collection will be normally combined with that for set (iv variables, since the latter also must also be collected directly at the person level. These are the socalled survey countries. Personal variables Set (iv variables will normally be collected through direct personal interview in all countries. These are too complex or personal in nature to be collected by proxy; nor are they available from registers or other administrative sources. or the survey countries, this collection will normally be combined with that for set (iii variables as noted above consequently both normally based on a sample of complete households, i.e. covering all persons aged 16+ in each sample household. However, from the substantive requirements of the E4SM, it is not essential that in contrast to set (iii variables set (iv variables be collected for all persons in each sample household. It is possible to do this collection on a representative sample of persons (adult members aged 16+, such as by selecting one such person per sample household. It is expected that this option will normally be followed in register countries, since for these countries interviewing all household members for set (iii is not involved. In countries which choose to do so, the sampling process involved will be the selection of persons (usually one adult member aged 16+ per household directly or through a sample of households. The selected individuals may be termed selected respondent ; the probability of selection of a selected respondent will be that of selection of its household, divided by the number of eligible persons (adult members aged 16+ in the household. Randomised selection procedures must be used to ensure that a representative sample of persons is obtained from the representative sample of households. Table 5.2 summarises the type of survey units for sampling, analysis and data collection involved in the E4SM. The ultimate units used in the sample selection may be addresses, households or persons, each unit selected with a known probability. rom 18

21 these, it is always necessary to construct a sample of households, the probability of each household in the sample being determined through its association (or identity, as the case may be with units in the sample selected. The analysis units can be households, all members, adult members, or possibly a sub-sample of adult members; these are the units to which the information collected pertains. Their probabilities of selection (or the corresponding sample weights are determined through their association with the sample household. The collection unit refers to the person or source providing the information. Table 5.2: Survey units for sampling, analysis, and data collection Sampling unit Analysis units Collection unit/source selected constructed survey country register country Address or Household Household Set (i: household Set (ii: all household members Set (iii: household and personal & related Household respondent (HR Household respondent* Personal interview (all members 16+ Registers +HR(? Registers +HR(? Registers (all members 16+ or Person (aged 16 + * combined with set (i household interview ** combined with set (iii personal interview Set (iv: personal variables ( social All members 16+ Selected respondent Personal interview** Personal interview Constructing a sample of households or analysis units in sets (i-(iii, as well in the first case of set (iv with all household members aged 16+ taken, the selection probability of all types of units is the same as that of their household. Hence the various analysis requirements are served best by having basically an equal probability ('self-weighting' sample of households. Variations in selection probabilities by region, household size, or whatever mostly results in reduced sampling efficiency. However, for the same reason, the second case of set (iv with only a sub-sample of members aged 16+ taken, it is desirable to aim at an equal probability ('selfweighting' sample of these units, rather than an equal probability sample of households. Consequently, in the situation pertaining to a majority of the countries (the survey countries in Table 2.2, the E4SM primary objective is served best by having an equal probability ('self-weighting' sample of households, taking all persons in a selected household into the sample for the personal interview, and hence obtaining an equal probability (self-weighting sample of persons as well. By contrast, in the register countries it is more efficient to aim at a self-weighting sample of person to be interviewed in detail; household variables and personal data then may not be based on a equal probability sample, but that is less critical. Sample of household from selection of addresses Constructing a sample of households from a selected sample of addresses or similar units is normally straightforward. In most cases, there is in fact a one-to-one correspondence between the two types of units, so that, for instance, an equal probability sample of households is obtained by taking an equal probability sample of ad- 19

22 dresses. Exactly the same applies when all households are taken into the sample from any selected addresses containing more than one households. However, in the presence of some addresses containing large numbers of households, it is desirable for technical and practical reasons to limit the maximum number of households which will be taken for the survey from any selected address. This changes the probabilities of selection of the households. Sample of household from selection of person Now consider the situation when a sample of households is constructed from a sample with individual persons as the selection units. A household is selected through its association with one or more individuals. Normally, the latter will be selected from a list of adults. In so far as each eligible household contains at least one such person in the list, the household receives a non-zero probability of selection; consequently, a probability sample of persons will yield a probability sample of households. This will be the case for instance if the sampling frame (list covers all persons aged 16+, in so far as it can be assumed that practically every household contains at least one adult. However, this might not be so in the case of certain other types of lists of persons. Or instance if an electoral roll is used as the frame, only those household which contain at least one eligible and registered voter will have any chance of being selected into the sample. The population of households not containing such a persons will not be covered in the survey. In any case, the main consideration in constructing a sample of households through the selection of persons is that the selection probability of a household would vary in direct proportion to the number of persons in the list through which the household could have been selected. If, for instance, persons are selected with equal probabilities, larger households will be selected with higher probabilities. These differences in household selection probabilities have to be compensated for by applying weights to the data. 6. Mode of data collection Member states can decide the mode of data collection from the following; Computer assisted Personal Interview (CAPI Computer assisted Telephone Interview (CATI Computer assisted Self Interview (CASI Paper-based and personal interview (PAPI Internet self completion It is proposed that, where possible, Computer Assisted Personal Interview (CAPI techniques should be used. However, a mixed mode approach, comprising CATI with CAPI, PAPI or CASI, would also be acceptable dependent on the content and requirements of specific modules to be included in the survey. It is proposed that all questions in the core module should be suitable for both personal and telephone interviewing modes. Therefore, at a minimum, in the event of interview non-response in the field, the core module could be re-issued for a telephone interview to be attempted, to maximise response rates. 20

23 Internet self-completion is included in the list above. Although at present this medium may not be suitable for this sampling frame, response, and data quality at present, it may be considered in the future, 7. Response rates, call back and proxy interviews Each Member State shall follow appropriate procedures to maximise the response rates achieved, in accordance with its own "best practices". In the case of an interview survey, at least three call-backs (at different times of the day shall be made before a household or individual is accepted as a non contact, unless there are conclusive reasons (such as a definite refusal to co-operate, circumstances endangering the safety of the interviewer, etc. why this cannot be done. The minimum response rates for an E4SM module will be dependent on the type of information being collected and so will vary by module. They will be specified in Commission Regulation (EC No [ ] of [ ]. Target response rates for the modules should be 70% with a minimum response rate of 60%. More detailed analysis of the expected response rates will need to be undertaken once the content of the main modules are set. 21