Total

Male

Female

Age0-17

Age18-64

Age 65+

Cross sectional

·Over-coverage

·Under-coverage

·Misclassification

-  

Problems emerging in connection with integrating different data sources

The Finnish microdata are collected from administrative data and by computer-assisted telephone interviews (CATI, exceptionally CAPI). Administrative data are linked to the sample persons and co-residents by personal identification codes (ID codes). All person registers used for data compilation include ID codes. In the sampling phase, the codes are retrieved from the sampling frame. They are then used in all phases of linking different data sources. ID codes can be missing only from recently immigrated persons, members recently moved to the household or newly born co-residents.  

To help the interviewer in defining the household members of the selected sample person, the first survey year's questionnaires are pre-filled with the persons’ ID codes registered in the same dwelling with the sample person. The second and later waves'/replications' questionnaires are pre-filled using the information collected in the first year. The interviewer checks the co-resident’s position in the household deleting from the data base those who do not belong to it and adding the names, birthdays and ID codes of those who have moved into it. The ID codes to be added are either given by the respondent or searched later from the population register, using dates of birth, names and other information. If no ID code is found, the person is deleted from the data base. These cases are quite rare. In the 2012 gross sample operation, the ID code was missing for only 23 co-residents.

More importantly, reference periods in the administrative data sources may become a problem. In the Finnish SILC microdata, the variables on education attendance (PE010, PE020) are retrieved from registers referring to attendance in the educational institutes on 26 September pre-dating the survey year. Since most of the institutes start their terms in August or early September, and usually continue to May of the survey year, the coverage and reference period of this data source is reasonable.

Problems due to the need to keep up the national time series

According to the principles commonly agreed in the planning phase of the SILC survey, it was important to ensure coherence between the new instrument and the established national statistics. In Finland, the SILC survey was integrated into the national Income Distribution Survey (IDS), compiled yearly since 1976. A major problem in maintaining the national series was and still is in reconciling the variables on labour activity.  

In the IDS, the reference period for the labour information is the income reference year. In the SILC, labour information mainly refers to the current situation. Different reference periods in the IDS and SILC concern the variables PL031, PL040, PL050, PL111, PL130, PL140 and PL150. The SILC variables PL073 to PL090 are also in contradiction with similar IDS monthly activity variables: overlapping activities are permitted in the IDS, but in the SILC, one should define one's main activity for each month. Both reference periods cannot be collected; different reference periods would make it very hard for the interviewee and the interviewer to give accurate information, especially in cases where changes occurred during the income reference period (IRP) or currently. To make the fieldwork easier, from the very start of the EU-SILC, the reference periods were integrated. "Current" is operationalised as the December of the IRP.

Examples of labour information with different requirements in the IDS and EU-SILC

Source of measurement error: selection of the informant

Finland’s EU-SILC uses the selected respondent -model. Typically, only one member of the household is interviewed. As a rule, this interviewee should be the selected sample person. 

In the EU-SILC, it is important to interview subjective questions from selected respondents. He/she gives all the information to the household questionnaire and all personal questionnaires.

Often a choice has to be made between the selected respondent and other member of his/her household. A significant part of the data collected from other members is about the household’s affairs. The selected respondent (especially the youngest selected respondents who still live with their parents, or very old respondents) may not be aware of the household economy, household debts, child care, housing items, the other household members' activities, or many other items.

A proxy respondent is defined as the respondent who is not the selected respondent. A proxy represented 11 per cent of the 10,307 selected respondents in 2012. In 89 per cent of the households, the selected respondent was interviewed. The proxy respondent rate has been slowly decreasing through years: from 25 per cent in 2004 to 11 per cent in 2012.

Use of proxy respondents, the 2012 survey 

Source of measurement error: General fieldwork problems

The main data collection mode is computed-assisted telephone interview (CATI). According to interviewers' estimate, about half of the interviews are made through mobile phones. In about six per cent of those cases, interviews take place outside the respondent’s home. Telephone interviews are afflicted by a sense of rush. In large households, the interview is too long for the telephone. The interviewers are allowed to change the mode into CAPI, in the cases the respondent has no phone or has an exceptionally large household. The CAPI mode was used in 206 households, that is, in two per cent of all households in the 2012 survey. 

Interview duration. According to the Interviewers' Feedback Survey 2012, 42 per cent of the interviewers felt that the duration of the interview was too long. 49 per cent of them assessed it had an effect on the refusal rate and 14 per cent thought that it weakened the quality of responses.

Distribution of total duration of interview in the 2012 survey by rotational group (DB075)

Source of measurement error: Variable-specific problems

HS130 Lowest monthly income to make ends meet. The difficulty of this question for the respondents is well illustrated by the large number of item non-responses in the cross-section data. Very low and very high figures were also answered.

PL060, PL100 Number of hours usually worked per week in main job, Total number of hours in the second, third…jobs. Item non-response rate is rather high, obviously, due to proxy respondents’ inability to report the hours accurately. Missing values were imputed by the hot-deck method using the known information by gender, employee- or self-employed status, broad occupational group, broad age group and, where possible, information by the self-reported part-time/full-time employment. There were 470 records with missing values in PL060 and/or PL100 in 2012, from which 235 were edited using the previous year's information. PL060 was imputed to 282 employees or self-employed persons based on 10,623 donators, PL100 was imputed to 45 cases based on 535 observations with known information about second jobs.

PE010–PE040, variables on education are collected from different registers. Variables PE010 and PE020 refer to the autumn term (15 September) preceding the operation year. Since the school year and academic year usually last from August-September to May-June, the current situation is sufficiently covered. PE030 (Year when the highest level of education was attained) includes a large number of missing values due to register imperfection, while the coverage of PE040 (highest ISCED level attained) is satisfactory. PE040 is known to be imperfect in the case of migrants.

RB031 (Year of immigration) is retrieved from the population register. The data coverage is deficient in cases of immigration that took place before 1990.

General description of the fieldwork tools

The BLAISE-programmed questionnaire is divided into blocks of questions: a specific block for each household member aged 16+, blocks for child care, health, housing, household economics, household composition. The order of the blocks is optional to the interviewer. Only the household composition has to be fixed first, after that the interviewer is free to choose the blocks. In case he/she does not choose the blocks him/herself, the order is automatic. 

Questionnaire build-up has its starting point in the previous year's questionnaire, feedback from the field interviewers and feedback from the data editing process and users. The leading principle in the questionnaire build-up has been a gradual integration process of the SILC to the IDS, and avoidance of too many changes in the national IDS. Of course, another starting point of the questionnaire update is to check the changes in the SILC doc65. The changes were minimal in the 2012 survey.

During the process of BLAISE programming (autumn 2011) the questionnaire was table-tested several times by the team responsible for the IDS and EU-SILC. Six persons were involved. The focus was in the parts of the questionnaire undergoing some change. In the end, a group of professional interviewers checked the questionnaire against their experience. Finally, the technical functioning of the questionnaire was tested in the interviewer organisation before it was sent to the field.

The testing procedure makes use of the BLAISE-programmed questionnaire. The real field situation is simulated by a test sample being actual households from the preceding year's data base. Thus the test questionnaire is pre-filled with the information about the household composition and dates of birth. As in a real field situation, the second and consequent waves/replications have more information from the previous interview entered into the questionnaires. The testers fill in the questionnaire over again, trying all combinations of imagined situations, and likely errors (to disclose signalling), too. Testers are asked to pay attention to

- Spelling, language, formulations and conceptual correctness of the questions,

- Proper functioning of the routings and

- Adequacy of logical checks, signals and interviewing instructions on the screen.

Translations and formulations of certain questions on well-being were tested in the 2012 SILC questionnaire (collecting answers for five questions from nearly 10,000 selected respondents). The questions were useful in helping to choose the final formulations and the interviewer instructions for the next year's ad hoc -module on well-being.

Changes in the questionnaire

No major changes.

Building process of questionnaire: selection of the informant 

The contact letter is also sent to his/her parents or guardians, if the selected person is aged under 18. In the 2012 survey, 80 per cent of the selected respondents under the age of 18 have been represented by a proxy respondent.

Distribution of proxy interviews by their relationship to the selected person in age groups, the 2012 survey 

The questionnaire is built to enable the change of respondent during the interview. The questionnaire is programmed to accommodate the mode of addressing the respondent depending on whether the selected person him/herself or another member of the household is responding (interviewing the selected respondent about him/herself: Did you…; interviewing through a proxy respondent: Did N.N. …). This helps the interviewer and respondent to keep control of the member-specific data collection. 

Use of a proxy is denied only in the self-reported health questions (PH010-PH030). A special procedure and a reminder function has been built in the questionnaire to help interviewers contact the selected respondent separately for health questions in those cases where a proxy has answered the rest of the questionnaire. Some ad hoc modules (e.g. the 2013 module on well-being) also require a separate contact with the selected respondent.

Building process of questionnaire: General fieldwork problems

To help communication by telephone and shorten the interview duration, a number of pre-fills using the data from the previous wave are served on the questionnaire. Each year new ideas on how to relieve the respondent burden are applied to the questionnaire. According to the feedback from the interviewers, it was easier to manage the questionnaire in 2012 than in the previous year. In the longer run, the questionnaire has improved. The percentage of interviewers who felt that the questionnaire functioned technically badly fell from 13 per cent in 2005 to four per cent in 2012. The percentage of those who felt that the questionnaire functioned badly as to the substance fell from 18 per cent in 2005 to zero per cent in 2012.

Letters describing the purpose and contents of the survey are sent to the selected respondents. A special brochure describes the data construction, how the results are published and data protection issues. Respondents also get a small booklet with some results from the earlier survey years. Respondents who have refused to take part in the survey receive a special letter trying to persuade them to participate.

It is also possible for the interviewer to change the mode of collection from CATI to CAPI. CAPI is allowed as an exception in cases where the respondent cannot or does not want to be interviewed by phone. Especially households with no phones or large households use this opportunity. Since the CAPI questionnaire is identical with the CATI questionnaire, the possible differences in outcomes must be due to interaction differences, or, the fact that the CAPI-interviewed households differ from the average in many respects, e.g. income and family composition. 

Type of interview (n, %), the 2012 survey

Building process of questionnaire:  Variable-specific problems

HS130 The wording of the question is essential. The wording was reformulated at the Survey Laboratory of Statistics Finland for the 2006 operation but the high level of non-response prevails.

PL060, PL100 If the respondent could not give an exact answer, he/she was asked whether the hours exceed 30 hours per week.

General description of the interviewer training routines

Statistics Finland's interviewer services employs about 160 field interviewers on a permanent work contract. They work mostly part-time. They are given basic training on interviewing and questionnaire standards and codes of practices when they start working. They collect most of Statistics Finland's survey data, for the Labour Force Survey, Household Budget Survey, Time Use Survey and Adult Literacy Survey, for example. In other words, they are experienced. 

The questionnaire changes were introduced to the interviewers in a separate written report and, of course, in the instructions book. The instructions book is rewritten every year and it is also under constant development. 

Newly recruited interviewers were trained separately. They had two day's training about the SILC. The training programme included a lecture on the planning of the survey, including a description of Eurostat's process, legislation and future uses of the data, and Eurostat guidelines on data protection. Concern over international comparability was underlined. Instructions on the fundamental rules of central data collection were given and discussed, such as the definition of the target population, household definition and its implementation in practice, different concepts and classifications of activity, especially labour market activities, child care questions, housing costs and mortgages. A major part of the training time was used on going through the videoed BLAISE questionnaire with the aid of three lecturers. The panel design and the future modules were described. The last part of training consisted of data transferring, data protection and other practicalities.

During the whole fieldwork period, interviewers' information desk is open for them. They can ask for support from the IDS-SILC team. The interviewers, who are distributed all over the country, also have organised district meetings with each other to discuss professional matters.

Fieldwork management and data reception

The interviewers collect the data and transmit them to the central unit. At Statistics Finland, there is a separate organisation, the interviewers' services unit, to control, monitor and supervise the fieldwork. The central unit transmits the fieldwork tools to the field and organises interviewer training at the beginning of the project, follows the fieldwork progress, and receives the output from the field, checks that all the sampled units are adequately processed and transmits the data to the IDS-SILC team. It also collects feedback from the interviewers with a standardised questionnaire. All data contents are processed by the IDS-SILC team either using the BLAISE system or SAS. Mainly the IDS and SILC data processing is integrated.

Quality control: selection of the informant

Problems arising from the use of proxy respondents concentrate on the subjective questions: control in terms of which household member answers the questions involving subjective assessments, depends on the interviewer. The denial to use proxy respondents for the health questions produces item non-response. In 2012, the item non-response percentage was 5.9 (variable PH010_F = -1 weighted by PB060). This percentage reflects the share of the selected respondents who could not be reached at all in the survey due to denials or other reasons, but whose household respondents were contacted. 

The high percentage of proxy interviews guarantees a higher quality of the household information. Most of the proxy respondents are parents or spouses. Proxies are mostly (45 % in 2012) the 1st or 2nd persons responsible for the accommodation, which also indicates their competence regarding knowledge of the household affairs.

Quality control: General fieldwork problems

HB100, PB120 - Household and personal interview duration - In Finland's selected respondent model, the duration of the interview is measured as the duration for both household- and personal interview in variable HB100. Variable PB120 is empty.

In fewer than two per cent of the households, the duration of the interview has exceeded the 60 minutes maximum stipulated in the framework regulation.

Quality control: Variable-specific problems

HS130 None

PL060, PL100 Distributions check

PE010–PE040, missing values are checked and imputed. 

• Fieldwork management and data reception
• Missing identification information of sample persons
• Incomplete interviews
• Missing information, outliers, illogical and inconsistent information
• Inconsistency between self-reported months of activity and registered income sources in particular
• Database construction
• Processing register data
• Comparison of aggregates

Fieldwork management and data reception

The Interviewers Services Unit in the Data Collection Unit of Statistics Finland (unit has been valid since 2013) manages, supervises and monitors interviews’ fieldwork, provides interview tools to the field and organises interviewer training. This unit employs 160 trained interviewers operating in all parts of the country. Four field supervisors administer this network. The interviewers’ average period of service is 11 years. Communication between the Interviewers Services Unit and the interviewers is arranged through a job monitoring system. The system enables continual follow-up of the fieldwork by each interviewer. The accruing paradata include contacts, completed interviews, non-response, workload and costs. Interviewers collect the data and transmit them to the Data Collection Unit, which checks the interviewed data quality (e.g. sample units are adequately processed) and transfers them to the SILC team. Occasional controlling is made by re-interviewing. Feed-back from the central unit to the interviewers is regular. A standardised questionnaire on interviewers’ evaluations of the operation is in regular use.

Missing identification information of sample persons 

Missing identification information of sample persons has been checked and edited by using register information from the Population Information System. The process takes usually from mid-February to the end of the June, when the interviewed data have been received.

Incomplete interviews

After the fieldwork period, the SILC team looks through incomplete interviews and makes a decision on their acceptance. Some of the incomplete interviews are rejected. Since the register income data are nearly perfectly available, the acceptance decision is based on the sufficient completeness of the main activity and housing information. In the 2012 operation, 23 incomplete interviews were excluded from the received sample and treated as non-responses.

Missing information, outliers, illogical and inconsistent information 

The BLAISE programming system of the CATI and CAPI questionnaires primarily controls data quality. In addition, interviewed information and interviewers’ remarks on the questionnaires have been checked and edited against other interviewed and register data for the sample unit. Errors of objective type of information have been corrected, subjective and opinion questions are left without editing. Errors contain such as missing answers, denials and don't knows, clear mistakes, outliers, illogical and inconsistent answers. Special emphasis has been put on the questions of activity months, occupation, NACE, housing costs and childcare. They are checked against other information (including register information). Occupation and NACE are processed through automatic coding. Some of the cases remain open, which are processed manually.

Inconsistency between self-reported months of main activity and registered income sources in particular

The months of main activity have traditionally been heavily edited to comply with register data, especially with income data. As a result of comparisons between interviewed and registers, some activities of the respondents' answers are rejected and replaced with the corrected ones to be in coherence with the register information about salaries, entrepreneur income, pensions, unemployment and other benefits. Editing was started from the 2009 SILC operation. In the 2004 to 2008 operations, months of main activities (PL070, PL072, PL080, PL085, PL087, PL090) were collected and dealt as subjective responses given by respondents as defined in the EU-SILC document 065.  

Database construction

Simultaneously with the checking process, a database is opened and variable formation begins by SAS-programming. Variable construction based on interviews and registers is started. Interviewed variables are transferred from the questionnaires to the database. Variables that need constructing – i.e. combined interview- and register information and complex questionnaire items – are added into the database after all the checks and edits have been carried out. Imputations are done.

Processing register data

Register data - that have been subscribed from the register authorities with a special procedure - arrive in electronic form to the IT and Statistical Methods units. In 2012, use was made of eleven registers. The incoming data are checked technically and substantially. Possible defects are notified to the administrative authority in charge of the data source. The registers cover all units - population, dwelling units, income receivers, etc. The data are linked to the sample persons and transmitted into the database. The data are compared with available external data, i.e. those of the tax authority, pensions authority and other statistics.

Comparison of aggregates

Routines have been developed to compare the results on variable level with external sources such as the Labour Force Survey, National Accounts, wage statistics and statistics on different social transfers and taxation produced by the National Pensions Institute, National Board of Taxes and National Research and Development Centre for Welfare and Health. Standard comparisons are routinely made each year. These comparisons also have an effect on error detection.

The private household includes a person residing alone, or all the persons, related or not, who reside and have their meals together or otherwise use their income together. The definition equals with the obligatory EU-SILC definition on "shares in household expenses", but uses other words "use income together" in the interview.

See the private household definition.

For persons who were temporarily absent from the household’s main dwelling and from home no specific time duration (6 months) was set for the absence in the interview provided that the criteria of household formation and membership (shares in household expenses) were fulfilled. Such persons have close family ties to the household and they do not form a household of their own. Therefore, the following persons are also counted in household members:

- Persons conducting military service or conscript service
- Persons residing and working in another locality or abroad if they are involved in the acquisition and use of household income
- Persons residing and studying in another locality if they use income received mostly from their parents
- Persons temporarily in institutions, on holiday or travelling.

The following persons form a household of their own:
- Subtenants
- Domestic staff
- Students living on their own if they live mostly on their own income or on a student loan
- Students residing in dormitories, unless they are married or officially cohabiting.

Stratification and sub-stratification criteria

Strata contain socio-economic groups grouped by income class (13 strata in the new sample). The register information used for strata construction is about all members of the selected sample person's household-dwelling unit available at the time of sample selection (updated by the final population register data later). Strata are created only for those who do not belong to the over-coverage. The stratification normally takes the highest earning person as a categorising person for a stratum, but an exception to the rule is an entrepreneur who does not have to be the highest earning one in the category of entrepreneurs. The income class division allocates the sample more to high-earners.

The strata of the new sample/rotational group (DB075=3)

The strata of the older rotational groups in the data refer to the equivalent groups of the new sample as follows: 2nd replication (DB075=2), DB050-12; 3rd replication (DB075=1), DB050-24; 4th replication (DB075=4), DB050-36.

Sample selection schemes

The new sample is selected with a two-phase stratified sampling design. In the first phase, a master sample of persons (50,000) is selected with systematic sampling from the population register data. In the second phase, a sample of persons (5,000) in the household-dwelling units is selected from the stratified master sample with simple random sampling without replacement within every stratum and using non-proportional allocation. The new sample selection scheme is consistent with the first years' sample selection of the old rotational groups. 

The old rotational groups (3 groups) are included in the set of responded households (DB135=1, including initial sample person) from the previous survey year to be interviewed (gross sample). Those households in that the initial sample person is no longer in-scope (moved to a collective household or institution in the country, moved outside the country or dead) are omitted for the final net sample of the survey year as over-coverage. The final household composition is defined during the interview. In the survey year, all old rotational groups are supplemented by an extra sample of persons aged 16 (selected from the stratified master sample with simple random sampling without replacement within every stratum and using non-proportional allocation, Section 12.1.3).

The sample does not contain substitutions.

Sample distribution over time

The reference population is defined as the population registered as permanently resident in Finland on 31 December 2011. Household composition is also dated to 31 December 2011. The income reference period is constant for all households and persons: the calendar year 2011.

In the SILC 2012 operation, the fieldwork period stretched over five months; it started at the beginning of January 2012 and ended in May 2012. The cross-sectional sample of the EU-SILC consists of four rotational groups from which old groups (from the 2nd to 4th replications) are also for the longitudinal component. The fieldworks of old rotational groups were started at the beginning of January and were completed mostly by the end of April. Only a few households were interviewed after April. The new sample households were interviewed in February to May.  

Total

Male

Female

Age0-17

Age18-64

Age 65+

Cross sectional

·Over-coverage

·Under-coverage

·Misclassification

-  

Problems emerging in connection with integrating different data sources

The Finnish microdata are collected from administrative data and by computer-assisted telephone interviews (CATI, exceptionally CAPI). Administrative data are linked to the sample persons and co-residents by personal identification codes (ID codes). All person registers used for data compilation include ID codes. In the sampling phase, the codes are retrieved from the sampling frame. They are then used in all phases of linking different data sources. ID codes can be missing only from recently immigrated persons, members recently moved to the household or newly born co-residents.  

To help the interviewer in defining the household members of the selected sample person, the first survey year's questionnaires are pre-filled with the persons’ ID codes registered in the same dwelling with the sample person. The second and later waves'/replications' questionnaires are pre-filled using the information collected in the first year. The interviewer checks the co-resident’s position in the household deleting from the data base those who do not belong to it and adding the names, birthdays and ID codes of those who have moved into it. The ID codes to be added are either given by the respondent or searched later from the population register, using dates of birth, names and other information. If no ID code is found, the person is deleted from the data base. These cases are quite rare. In the 2012 gross sample operation, the ID code was missing for only 23 co-residents.

More importantly, reference periods in the administrative data sources may become a problem. In the Finnish SILC microdata, the variables on education attendance (PE010, PE020) are retrieved from registers referring to attendance in the educational institutes on 26 September pre-dating the survey year. Since most of the institutes start their terms in August or early September, and usually continue to May of the survey year, the coverage and reference period of this data source is reasonable.

Problems due to the need to keep up the national time series

According to the principles commonly agreed in the planning phase of the SILC survey, it was important to ensure coherence between the new instrument and the established national statistics. In Finland, the SILC survey was integrated into the national Income Distribution Survey (IDS), compiled yearly since 1976. A major problem in maintaining the national series was and still is in reconciling the variables on labour activity.  

In the IDS, the reference period for the labour information is the income reference year. In the SILC, labour information mainly refers to the current situation. Different reference periods in the IDS and SILC concern the variables PL031, PL040, PL050, PL111, PL130, PL140 and PL150. The SILC variables PL073 to PL090 are also in contradiction with similar IDS monthly activity variables: overlapping activities are permitted in the IDS, but in the SILC, one should define one's main activity for each month. Both reference periods cannot be collected; different reference periods would make it very hard for the interviewee and the interviewer to give accurate information, especially in cases where changes occurred during the income reference period (IRP) or currently. To make the fieldwork easier, from the very start of the EU-SILC, the reference periods were integrated. "Current" is operationalised as the December of the IRP.

Examples of labour information with different requirements in the IDS and EU-SILC

Source of measurement error: selection of the informant

Finland’s EU-SILC uses the selected respondent -model. Typically, only one member of the household is interviewed. As a rule, this interviewee should be the selected sample person. 

In the EU-SILC, it is important to interview subjective questions from selected respondents. He/she gives all the information to the household questionnaire and all personal questionnaires.

Often a choice has to be made between the selected respondent and other member of his/her household. A significant part of the data collected from other members is about the household’s affairs. The selected respondent (especially the youngest selected respondents who still live with their parents, or very old respondents) may not be aware of the household economy, household debts, child care, housing items, the other household members' activities, or many other items.

A proxy respondent is defined as the respondent who is not the selected respondent. A proxy represented 11 per cent of the 10,307 selected respondents in 2012. In 89 per cent of the households, the selected respondent was interviewed. The proxy respondent rate has been slowly decreasing through years: from 25 per cent in 2004 to 11 per cent in 2012.

Use of proxy respondents, the 2012 survey 

Source of measurement error: General fieldwork problems

The main data collection mode is computed-assisted telephone interview (CATI). According to interviewers' estimate, about half of the interviews are made through mobile phones. In about six per cent of those cases, interviews take place outside the respondent’s home. Telephone interviews are afflicted by a sense of rush. In large households, the interview is too long for the telephone. The interviewers are allowed to change the mode into CAPI, in the cases the respondent has no phone or has an exceptionally large household. The CAPI mode was used in 206 households, that is, in two per cent of all households in the 2012 survey. 

Interview duration. According to the Interviewers' Feedback Survey 2012, 42 per cent of the interviewers felt that the duration of the interview was too long. 49 per cent of them assessed it had an effect on the refusal rate and 14 per cent thought that it weakened the quality of responses.

Distribution of total duration of interview in the 2012 survey by rotational group (DB075)

Source of measurement error: Variable-specific problems

HS130 Lowest monthly income to make ends meet. The difficulty of this question for the respondents is well illustrated by the large number of item non-responses in the cross-section data. Very low and very high figures were also answered.

PL060, PL100 Number of hours usually worked per week in main job, Total number of hours in the second, third…jobs. Item non-response rate is rather high, obviously, due to proxy respondents’ inability to report the hours accurately. Missing values were imputed by the hot-deck method using the known information by gender, employee- or self-employed status, broad occupational group, broad age group and, where possible, information by the self-reported part-time/full-time employment. There were 470 records with missing values in PL060 and/or PL100 in 2012, from which 235 were edited using the previous year's information. PL060 was imputed to 282 employees or self-employed persons based on 10,623 donators, PL100 was imputed to 45 cases based on 535 observations with known information about second jobs.

PE010–PE040, variables on education are collected from different registers. Variables PE010 and PE020 refer to the autumn term (15 September) preceding the operation year. Since the school year and academic year usually last from August-September to May-June, the current situation is sufficiently covered. PE030 (Year when the highest level of education was attained) includes a large number of missing values due to register imperfection, while the coverage of PE040 (highest ISCED level attained) is satisfactory. PE040 is known to be imperfect in the case of migrants.

RB031 (Year of immigration) is retrieved from the population register. The data coverage is deficient in cases of immigration that took place before 1990.

General description of the fieldwork tools

The BLAISE-programmed questionnaire is divided into blocks of questions: a specific block for each household member aged 16+, blocks for child care, health, housing, household economics, household composition. The order of the blocks is optional to the interviewer. Only the household composition has to be fixed first, after that the interviewer is free to choose the blocks. In case he/she does not choose the blocks him/herself, the order is automatic. 

Questionnaire build-up has its starting point in the previous year's questionnaire, feedback from the field interviewers and feedback from the data editing process and users. The leading principle in the questionnaire build-up has been a gradual integration process of the SILC to the IDS, and avoidance of too many changes in the national IDS. Of course, another starting point of the questionnaire update is to check the changes in the SILC doc65. The changes were minimal in the 2012 survey.

During the process of BLAISE programming (autumn 2011) the questionnaire was table-tested several times by the team responsible for the IDS and EU-SILC. Six persons were involved. The focus was in the parts of the questionnaire undergoing some change. In the end, a group of professional interviewers checked the questionnaire against their experience. Finally, the technical functioning of the questionnaire was tested in the interviewer organisation before it was sent to the field.

The testing procedure makes use of the BLAISE-programmed questionnaire. The real field situation is simulated by a test sample being actual households from the preceding year's data base. Thus the test questionnaire is pre-filled with the information about the household composition and dates of birth. As in a real field situation, the second and consequent waves/replications have more information from the previous interview entered into the questionnaires. The testers fill in the questionnaire over again, trying all combinations of imagined situations, and likely errors (to disclose signalling), too. Testers are asked to pay attention to

- Spelling, language, formulations and conceptual correctness of the questions,

- Proper functioning of the routings and

- Adequacy of logical checks, signals and interviewing instructions on the screen.

Translations and formulations of certain questions on well-being were tested in the 2012 SILC questionnaire (collecting answers for five questions from nearly 10,000 selected respondents). The questions were useful in helping to choose the final formulations and the interviewer instructions for the next year's ad hoc -module on well-being.

Changes in the questionnaire

No major changes.

Building process of questionnaire: selection of the informant 

The contact letter is also sent to his/her parents or guardians, if the selected person is aged under 18. In the 2012 survey, 80 per cent of the selected respondents under the age of 18 have been represented by a proxy respondent.

Distribution of proxy interviews by their relationship to the selected person in age groups, the 2012 survey 

The questionnaire is built to enable the change of respondent during the interview. The questionnaire is programmed to accommodate the mode of addressing the respondent depending on whether the selected person him/herself or another member of the household is responding (interviewing the selected respondent about him/herself: Did you…; interviewing through a proxy respondent: Did N.N. …). This helps the interviewer and respondent to keep control of the member-specific data collection. 

Use of a proxy is denied only in the self-reported health questions (PH010-PH030). A special procedure and a reminder function has been built in the questionnaire to help interviewers contact the selected respondent separately for health questions in those cases where a proxy has answered the rest of the questionnaire. Some ad hoc modules (e.g. the 2013 module on well-being) also require a separate contact with the selected respondent.

Building process of questionnaire: General fieldwork problems

To help communication by telephone and shorten the interview duration, a number of pre-fills using the data from the previous wave are served on the questionnaire. Each year new ideas on how to relieve the respondent burden are applied to the questionnaire. According to the feedback from the interviewers, it was easier to manage the questionnaire in 2012 than in the previous year. In the longer run, the questionnaire has improved. The percentage of interviewers who felt that the questionnaire functioned technically badly fell from 13 per cent in 2005 to four per cent in 2012. The percentage of those who felt that the questionnaire functioned badly as to the substance fell from 18 per cent in 2005 to zero per cent in 2012.

Letters describing the purpose and contents of the survey are sent to the selected respondents. A special brochure describes the data construction, how the results are published and data protection issues. Respondents also get a small booklet with some results from the earlier survey years. Respondents who have refused to take part in the survey receive a special letter trying to persuade them to participate.

It is also possible for the interviewer to change the mode of collection from CATI to CAPI. CAPI is allowed as an exception in cases where the respondent cannot or does not want to be interviewed by phone. Especially households with no phones or large households use this opportunity. Since the CAPI questionnaire is identical with the CATI questionnaire, the possible differences in outcomes must be due to interaction differences, or, the fact that the CAPI-interviewed households differ from the average in many respects, e.g. income and family composition. 

Type of interview (n, %), the 2012 survey

Building process of questionnaire:  Variable-specific problems

HS130 The wording of the question is essential. The wording was reformulated at the Survey Laboratory of Statistics Finland for the 2006 operation but the high level of non-response prevails.

PL060, PL100 If the respondent could not give an exact answer, he/she was asked whether the hours exceed 30 hours per week.

General description of the interviewer training routines

Statistics Finland's interviewer services employs about 160 field interviewers on a permanent work contract. They work mostly part-time. They are given basic training on interviewing and questionnaire standards and codes of practices when they start working. They collect most of Statistics Finland's survey data, for the Labour Force Survey, Household Budget Survey, Time Use Survey and Adult Literacy Survey, for example. In other words, they are experienced. 

The questionnaire changes were introduced to the interviewers in a separate written report and, of course, in the instructions book. The instructions book is rewritten every year and it is also under constant development. 

Newly recruited interviewers were trained separately. They had two day's training about the SILC. The training programme included a lecture on the planning of the survey, including a description of Eurostat's process, legislation and future uses of the data, and Eurostat guidelines on data protection. Concern over international comparability was underlined. Instructions on the fundamental rules of central data collection were given and discussed, such as the definition of the target population, household definition and its implementation in practice, different concepts and classifications of activity, especially labour market activities, child care questions, housing costs and mortgages. A major part of the training time was used on going through the videoed BLAISE questionnaire with the aid of three lecturers. The panel design and the future modules were described. The last part of training consisted of data transferring, data protection and other practicalities.

During the whole fieldwork period, interviewers' information desk is open for them. They can ask for support from the IDS-SILC team. The interviewers, who are distributed all over the country, also have organised district meetings with each other to discuss professional matters.

Fieldwork management and data reception

The interviewers collect the data and transmit them to the central unit. At Statistics Finland, there is a separate organisation, the interviewers' services unit, to control, monitor and supervise the fieldwork. The central unit transmits the fieldwork tools to the field and organises interviewer training at the beginning of the project, follows the fieldwork progress, and receives the output from the field, checks that all the sampled units are adequately processed and transmits the data to the IDS-SILC team. It also collects feedback from the interviewers with a standardised questionnaire. All data contents are processed by the IDS-SILC team either using the BLAISE system or SAS. Mainly the IDS and SILC data processing is integrated.

Quality control: selection of the informant

Problems arising from the use of proxy respondents concentrate on the subjective questions: control in terms of which household member answers the questions involving subjective assessments, depends on the interviewer. The denial to use proxy respondents for the health questions produces item non-response. In 2012, the item non-response percentage was 5.9 (variable PH010_F = -1 weighted by PB060). This percentage reflects the share of the selected respondents who could not be reached at all in the survey due to denials or other reasons, but whose household respondents were contacted. 

The high percentage of proxy interviews guarantees a higher quality of the household information. Most of the proxy respondents are parents or spouses. Proxies are mostly (45 % in 2012) the 1st or 2nd persons responsible for the accommodation, which also indicates their competence regarding knowledge of the household affairs.

Quality control: General fieldwork problems

HB100, PB120 - Household and personal interview duration - In Finland's selected respondent model, the duration of the interview is measured as the duration for both household- and personal interview in variable HB100. Variable PB120 is empty.

In fewer than two per cent of the households, the duration of the interview has exceeded the 60 minutes maximum stipulated in the framework regulation.

Quality control: Variable-specific problems

HS130 None

PL060, PL100 Distributions check

PE010–PE040, missing values are checked and imputed. 

• Fieldwork management and data reception
• Missing identification information of sample persons
• Incomplete interviews
• Missing information, outliers, illogical and inconsistent information
• Inconsistency between self-reported months of activity and registered income sources in particular
• Database construction
• Processing register data
• Comparison of aggregates

Fieldwork management and data reception

The Interviewers Services Unit in the Data Collection Unit of Statistics Finland (unit has been valid since 2013) manages, supervises and monitors interviews’ fieldwork, provides interview tools to the field and organises interviewer training. This unit employs 160 trained interviewers operating in all parts of the country. Four field supervisors administer this network. The interviewers’ average period of service is 11 years. Communication between the Interviewers Services Unit and the interviewers is arranged through a job monitoring system. The system enables continual follow-up of the fieldwork by each interviewer. The accruing paradata include contacts, completed interviews, non-response, workload and costs. Interviewers collect the data and transmit them to the Data Collection Unit, which checks the interviewed data quality (e.g. sample units are adequately processed) and transfers them to the SILC team. Occasional controlling is made by re-interviewing. Feed-back from the central unit to the interviewers is regular. A standardised questionnaire on interviewers’ evaluations of the operation is in regular use.

Missing identification information of sample persons 

Missing identification information of sample persons has been checked and edited by using register information from the Population Information System. The process takes usually from mid-February to the end of the June, when the interviewed data have been received.

Incomplete interviews

After the fieldwork period, the SILC team looks through incomplete interviews and makes a decision on their acceptance. Some of the incomplete interviews are rejected. Since the register income data are nearly perfectly available, the acceptance decision is based on the sufficient completeness of the main activity and housing information. In the 2012 operation, 23 incomplete interviews were excluded from the received sample and treated as non-responses.

Missing information, outliers, illogical and inconsistent information 

The BLAISE programming system of the CATI and CAPI questionnaires primarily controls data quality. In addition, interviewed information and interviewers’ remarks on the questionnaires have been checked and edited against other interviewed and register data for the sample unit. Errors of objective type of information have been corrected, subjective and opinion questions are left without editing. Errors contain such as missing answers, denials and don't knows, clear mistakes, outliers, illogical and inconsistent answers. Special emphasis has been put on the questions of activity months, occupation, NACE, housing costs and childcare. They are checked against other information (including register information). Occupation and NACE are processed through automatic coding. Some of the cases remain open, which are processed manually.

Inconsistency between self-reported months of main activity and registered income sources in particular

The months of main activity have traditionally been heavily edited to comply with register data, especially with income data. As a result of comparisons between interviewed and registers, some activities of the respondents' answers are rejected and replaced with the corrected ones to be in coherence with the register information about salaries, entrepreneur income, pensions, unemployment and other benefits. Editing was started from the 2009 SILC operation. In the 2004 to 2008 operations, months of main activities (PL070, PL072, PL080, PL085, PL087, PL090) were collected and dealt as subjective responses given by respondents as defined in the EU-SILC document 065.  

Database construction

Simultaneously with the checking process, a database is opened and variable formation begins by SAS-programming. Variable construction based on interviews and registers is started. Interviewed variables are transferred from the questionnaires to the database. Variables that need constructing – i.e. combined interview- and register information and complex questionnaire items – are added into the database after all the checks and edits have been carried out. Imputations are done.

Processing register data

Register data - that have been subscribed from the register authorities with a special procedure - arrive in electronic form to the IT and Statistical Methods units. In 2012, use was made of eleven registers. The incoming data are checked technically and substantially. Possible defects are notified to the administrative authority in charge of the data source. The registers cover all units - population, dwelling units, income receivers, etc. The data are linked to the sample persons and transmitted into the database. The data are compared with available external data, i.e. those of the tax authority, pensions authority and other statistics.

Comparison of aggregates

Routines have been developed to compare the results on variable level with external sources such as the Labour Force Survey, National Accounts, wage statistics and statistics on different social transfers and taxation produced by the National Pensions Institute, National Board of Taxes and National Research and Development Centre for Welfare and Health. Standard comparisons are routinely made each year. These comparisons also have an effect on error detection.

The private household includes a person residing alone, or all the persons, related or not, who reside and have their meals together or otherwise use their income together. The definition equals with the obligatory EU-SILC definition on "shares in household expenses", but uses other words "use income together" in the interview.

See the private household definition.

For persons who were temporarily absent from the household’s main dwelling and from home no specific time duration (6 months) was set for the absence in the interview provided that the criteria of household formation and membership (shares in household expenses) were fulfilled. Such persons have close family ties to the household and they do not form a household of their own. Therefore, the following persons are also counted in household members:

- Persons conducting military service or conscript service
- Persons residing and working in another locality or abroad if they are involved in the acquisition and use of household income
- Persons residing and studying in another locality if they use income received mostly from their parents
- Persons temporarily in institutions, on holiday or travelling.

The following persons form a household of their own:
- Subtenants
- Domestic staff
- Students living on their own if they live mostly on their own income or on a student loan
- Students residing in dormitories, unless they are married or officially cohabiting.

Stratification and sub-stratification criteria

Strata contain socio-economic groups grouped by income class (13 strata in the new sample). The register information used for strata construction is about all members of the selected sample person's household-dwelling unit available at the time of sample selection (updated by the final population register data later). Strata are created only for those who do not belong to the over-coverage. The stratification normally takes the highest earning person as a categorising person for a stratum, but an exception to the rule is an entrepreneur who does not have to be the highest earning one in the category of entrepreneurs. The income class division allocates the sample more to high-earners.

The strata of the new sample/rotational group (DB075=3)

The strata of the older rotational groups in the data refer to the equivalent groups of the new sample as follows: 2nd replication (DB075=2), DB050-12; 3rd replication (DB075=1), DB050-24; 4th replication (DB075=4), DB050-36.

Sample selection schemes

The new sample is selected with a two-phase stratified sampling design. In the first phase, a master sample of persons (50,000) is selected with systematic sampling from the population register data. In the second phase, a sample of persons (5,000) in the household-dwelling units is selected from the stratified master sample with simple random sampling without replacement within every stratum and using non-proportional allocation. The new sample selection scheme is consistent with the first years' sample selection of the old rotational groups. 

The old rotational groups (3 groups) are included in the set of responded households (DB135=1, including initial sample person) from the previous survey year to be interviewed (gross sample). Those households in that the initial sample person is no longer in-scope (moved to a collective household or institution in the country, moved outside the country or dead) are omitted for the final net sample of the survey year as over-coverage. The final household composition is defined during the interview. In the survey year, all old rotational groups are supplemented by an extra sample of persons aged 16 (selected from the stratified master sample with simple random sampling without replacement within every stratum and using non-proportional allocation, Section 12.1.3).

The sample does not contain substitutions.

Sample distribution over time

The reference population is defined as the population registered as permanently resident in Finland on 31 December 2011. Household composition is also dated to 31 December 2011. The income reference period is constant for all households and persons: the calendar year 2011.

In the SILC 2012 operation, the fieldwork period stretched over five months; it started at the beginning of January 2012 and ended in May 2012. The cross-sectional sample of the EU-SILC consists of four rotational groups from which old groups (from the 2nd to 4th replications) are also for the longitudinal component. The fieldworks of old rotational groups were started at the beginning of January and were completed mostly by the end of April. Only a few households were interviewed after April. The new sample households were interviewed in February to May.