Print

SILC_ESQRS_A_AT_2011_0000 - Version 1

National Reference Metadata in ESS Standard for Quality Reports Structure (ESQRS)

Compiling agency: Statistics Austria

Time Dimension: 2011-A0

Data Provider: AT1

Data Flow: SILC_ESQRS_A


For any question on data and metadata, please contact: EUROPEAN STATISTICAL DATA SUPPORT


1. ContactTop
1.1. Contact organisationStatistics Austria
1.2. Contact organisation unitDirectorate Social Statistics Unit Living Conditions, Social Protection
1.5. Contact mail addressGuglgasse 13 Vienna 1110 Austria


2. IntroductionTop
 

The production of quality reports is part of the implementation of the EU-SILC instrument. In order to assess the quality of data at national level and to make a comparison among countries, the National Statistics Institutes are asked to report detailed information mainly on: the entire statistical process, sampling and non-sampling errors, and potential deviations from standard definition and concepts.

This document follows the ESS standard for quality reports structure (ESQRS), which is the main report structure for reference metadata related to data quality in the European Statistical System. It is a metadata template, based on 13 main concepts, which can be used across several statistical domains with the purpose of a better harmonisation of the quality reporting requirements in the ESS.

For that reason the template of this document differs from that one stated in the Commission Reg. 28/2004.

Finally it is the combination of the previous intermediate and final quality reports therefore it is worth mentioning that it refers to both the cross sectional and the longitudinal data.


3. Quality management - assessmentTop

.


4. RelevanceTop
4.1. Relevance - User Needs

.
4.2. Relevance - User Satisfaction

.
4.3. Completeness

.
4.3.1. Data completeness - rate

.


5. Accuracy and reliabilityTop
5.1. Accuracy - overall
 

In terms of precision requirements, the EU-SILC framework regulation as well the Commission Regulation on sampling and tracing rules refers respectively, to the effective sample size to be achieved and to representativeness of the sample. The effective sample size combines sample size and sampling design effect which depends on sampling design, population structure and non-response rate.

 
5.2. Sampling error
 

EU-SILC is a complex survey involving different sampling design in different countries. In order to harmonize and make sampling errors comparable among countries, Eurostat (with the substantial methodological support of Net-SILC2) has chosen to apply the "linearization" technique coupled with the “ultimate cluster” approach for variance estimation. Linearization is a technique based on the use of linear approximation to reduce non-linear statistics to a linear form, justified by asymptotic properties of the estimator. This technique can encompass a wide variety of indicators, including EU-SILC indicators. The "ultimate cluster" approach is a simplification consisting in calculating the variance taking into account only variation among Primary Sampling Unit (PSU) totals. This method requires first stage sampling fractions to be small which is nearly always the case. This method allows a great flexibility and simplifies the calculations of variances. It can also be generalized to calculate variance of the differences of one year to another .

The main hypothesis on which the calculations are based is that the "at risk of poverty" threshold is fixed. According to the characteristics and availability of data for different countries we have used different variables to specify strata and cluster information. In particular, countries have been split into four groups:

1)BE, BG, CZ, IE, EL, ES, FR, IT, LV, HU, NL, PL, PT, RO, SI, UK and HR whose sampling design could be assimilated to a two stage stratified type we used DB050 (primary strata) for strata specification and DB060 (Primary Sampling Unit) for cluster specification;

2) DE, EE, CY, LT, LU, AT, SK, FI, CH whose sampling design could be assimilated to a one stage stratified type we used DB050 for strata specification and DB030 (household ID) for cluster specification;

3) DK, MT, SE, IS, NO, whose sampling design could be assimilated to a simple random sampling, we used DB030 for cluster specification and no strata;

 

In case Eurostat methodology is not accepted by your country, please describe the methodology used at national level for computing the estimates.
5.2.1. Sampling error - indicators
 

 

  AROPE At risk of poverty Severe Very low
-60% Material Deprivation work intensity
Ind. Stand. errors Half Ind. Stand. errors Half Ind. Stand. errors Half Ind. Stand. errors Half
               
value CI (95%) value CI (95%) value CI (95%) value CI (95%)
Total 26.5 1.7 3.4 11.4 1.0 2.0 7.5 1.0 1.9 4.7 1.1 2.1
Male 33.0 1.4 2.8 19.3 1.2 2.3 7.2 0.9 1.7 22.1 3.1 6.1
Female 14.0 1.2 2.4 12.0 0.5 1.0 0.8 0.3 0.6 2.8 0.7 1.4
Age0-17 16.0 1.3 2.5 10.9 0.6 1.3 4.2 0.8 1.5 1.1 0.4 0.9
Age18-64 8.4 1.6 3.1 11.0 0.6 1.2 0.5 0.3 0.6 NA NA NA
Age18-59 - - - - - - - - - 9.0 2.5 5.0
Age 65+ 17.1 3.4 6.6 16.0 0.9 1.8 2.0 0.4 0.7 NA NA NA

 NA: breakdowns not available due to definition of indicator
5.3. Non-sampling error
 

Non-sampling errors are basically of 4 types:

  • Coverage errors: errors due to divergences existing between the target population and the sampling frame.
  • Measurement errors: errors that occur at the time of data collection. There are a number of sources for these errors such as the survey instrument, the information system, the interviewer and the mode of collection
  • Processing errors: errors in post-data-collection processes such as data entry, keying, editing and weighting
  • Non-response errors: errors due to an unsuccessful attempt to obtain the desired information from an eligible unit. Two main types of non-response errors are considered:

Unit non-response: refers to absence of information of the whole units (households and/or persons) selected into the sample

Item non-response: refers to the situation where a sample unit has been successfully enumerated, but not all required information has been obtained
5.3.1. Coverage error
 

Coverage errors include over-coverage, under-coverage and misclassification:

  • Over-coverage: relates either to wrongly classified units that are in fact out of scope, or to units that do not exist in practice
  • Under-coverage: refers to units not included in the sampling frame
  • Misclassification: refers to incorrect classification of units that belong to the target population
5.3.1.1. Over-coverage - rate
 
 

Main problems

Size of error

Cross sectional

data

·Over-coverage

 

·Under-coverage

 

·Misclassification

  
1.4% (114 addresses) (DB120=23)
NA

NA
5.3.2. Measurement error
 

Cross sectional data

Source of measurement errors

Building process of questionnaire

Interview training

Quality control

Measurement errors are defined as the difference between the value of a certain variable (provided by the respondent) and the true, but unknown value of this variable. If the distribution of the error made at each single response is not random, the resulting statistic is biased. Elements affecting measurement are:

  • The questionnaire (e.g. the design, content, question wording, sensitivity of questions)
  • The interviewer (e.g. characteristics, behaviour, experience, workload, explanations, probing)
  • The respondents (e.g. problems arising during the cognitive response process, proxy interviews)
  • The interview situation (e.g. environment, presence of other persons, pressure of time)

The occurrence and effects of these errors is almost unavoidable. Nonetheless, Statistics Austria developed various routines to reduce these effects and errors.

 

 The questionnaire of EU-SILC is standardised and was developed according to EU-SILC regulations and EUROSTAT guidelines. The questionnaires for CATI and CAPI mode are identically implemented.

The standardised question wording should include all necessary information to answer the question. If respondents or interviewers need further information to answer the question additional definitions and explanations are integrated in the electronic questionnaire and written remarks for each question are allowed.

As was also the case for previous years, in EU-SILC 2011 pretests and consultations of experts were carried out for the questions of the ad-hoc module.

 

 In order to reduce interviewer effects it is necessary to provide interviewers with sufficient training and supporting measures. Overall, 147 CAPI interviewers and 13 CATI interviewers conducted the interviews for EU-SILC 2011. For EU-SILC 2011 interviewers which have already worked for previous EU-SILC waves did not receive a conventional training at Statistics Austria but were required to make a test interview on their laptop computer to learn about revisions of the questionnaire and the questions for the module 2011. Interviewers working for SILC for the first time were fully trained in small groups. Additionally, the interviewers received trainings materials and questionnaires on paper as well as a feedback of their last years’ work.

 EU-SILC in Austria applies several quality control measures during the fieldwork:

  • Plausibility checks are automatically implemented in the questionnaire. Since only computer assisted personal interviewing (CAPI) and computer assisted telephone interviewing (CATI) is used, the programming of the questionnaire ensures that values which are defined as unplausible or incorrect cannot be entered in the questionnaire.
  • The questionnaire is adapted yearly based on the experiences of previous years' fieldwork.
  • CATI interviewing is carried out in a seperate telephone studio at Statistics Austria allowing for a controlled and supervised interview situation.
  • Field control: if problems with certain questions or interviews arise, questionnaires are reviewed.

In EU-SILC 2011 administrative register data were used for the first time to collect income data (pensions and survivor's benefits). In order to evaluate the usage of register data a feasibility study was carried out with data from EU-SILC 2010. This study also allowed for record checks by comparing income data from the questionnaire and registers.
5.3.3. Non response error
 

Non-response errors are errors due to an unsuccessful attempt to obtain the desired information from an eligible unit. Two main types of non-response errors are considered:

1) Unit non-response which refers to the absence of information of the whole units (households and/or persons) selected into the sample. According the Commission Regulation 28/2004:

  • Household non-response rates (NRh) is computed as follows:

NRh=(1-(Ra * Rh)) * 100

Where Ra is the address contact rate defined as:

Ra= Number of address successfully contacted/Number of valid addresses selected

and Rh is the proportion of complete household interviews accepted for the database

Rh=Number of household interviews completed and accepted for database/Number of eligible households at contacted addresses

  • Individual non-response rates (NRp) will be computed as follows:

NRp=(1-(Rp)) * 100

Where Rp is the proportion of complete personal interviews within the households accepted for the database

Rp= Number of personal interview completed/Number of eligible individuals in the households whose interviews were completed and accepted for the database

  • Overall individual non-response rates (*NRp) will be computed as follows:

*NRp=(1-(Ra * Rh * Rp)) * 100

For those Members States where a sample of persons rather than a sample of households (addresses) was selected, the individual non-response rates will be calculated for ‘the selected respondent’, for all individuals aged 16 years or older and for the non-selected respondent.

2) Item non-response which refers to the situation where a sample unit has been successfully enumerated, but not all the required information has been obtained.
5.3.3.1. Unit non-response - rate
 

Cross sectional data

Address contact rate
(Ra)*

Complete household interviews
(Rh)*

Complete personal interviews
(Rp)*

 Household Non-response rate
(NRh)*		 Individual non-response rate
(NRp)*		 Overall individual non-response rate
(NRp)*

A*

B*

A*

B*

A*

B*

A*

B*

A*

B*

A*

B*
 
99.22
 
99.75
 
78.02
 
62.59
 
99.42
 
99.61
 
22.59
 
37.57
 
0.58
 
0.39
 
23.04
 
37.81

* All the formulas are defined in the Commission Regulation 28/2004, Annex II

A* = Total sample; B = * New sub-sample
5.3.3.2. Item non-response - rate
 

The computation of item non-response is essential to fulfil the precision requirements concerning publication as stated in the Commission Regulation No 1982/2003. Item non-response rate is provided for the main income variables both at household and personal level.
5.3.3.2.1. Item non-response rate by indicator
  Imputed rent Income from rental of property or land Family/ Children related allowances Social exclusion payments not elsewhere classified Housing allowances Regular inter-hh cash transfers received Interest, dividends, profit from capital investments in incorporated businesses Income received by people aged under 16 Regular inter-household cash transfer paid Repayments receipts for tax adjustment
(HY030N) (HY040N) (HY050N) (HY060N) (HY070N) (HY080N) (HY090N) (HY110N) (HY130N) (HY145N)
% of household having received an amount 0.0 95.3 99.6 97.8 98.0 95.2 89.0 91.8 95.0 98.6
% of household with missing values (before imputation) 0.0 3.4 0.0 1.2 0.7 1.1 6.7 8.2 1.7 0.6
% of household with partial information (before imputation) 100.0 1.3 0.4 0.9 1.3 3.7 4.3 0.0 3.2 0.8

In the tables below the first row (% of household having received an amount) refers to records were full information of a certain variable was available and imputation was not necessary. The components imputed rent (HY030) and interest payments on mortgages (HY100) are included in the tables although these variables are not directly collected from the respondents. Cash profits or losses from self-employment (PY050) are only collected on net-level, the gross value is computed from the net value.

  Total hh gross income Total disposable hh income Total disposable hh income before social transfers other than old-age and survivors benefits Total disposable hh income before all social transfers
(HY010) (HY020) (HY022) (HY023)
% of household having received an amount 51.8 87.6 88.3 87.0
% of household with missing values (before imputation) 5.7 0.4 0.8 2.0
% of household with partial information (before imputation) 42.6 12.0 10.9 10.9
  Imputed rent Income from rental of property or land Family/ Children related allowances Social exclusion payments not elsewhere classified Housing allowances Regular inter-hh cash transfers received Interest, dividends, profit from capital investments in incorporated businesses Income received by people aged under 16 Regular inter-household cash transfer paid
(HY030G) (HY040G) (HY050G) (HY060G) (HY070G) (HY080G) (HY090G) (HY110G) (HY130G)
% of household having received an amount 0.0 0.0 99.6 97.8 98.0 95.2 89.0 47.9 95.0
% of household with missing values (before imputation) 0.0 100.0 0.0 1.2 0.7 1.1 6.7 52.1 1.7
% of household with partial information (before imputation) 100.0 0.0 0.4 0.9 1.3 3.7 4.3 0.0 3.2

 

  Imputed rent Income from rental of property or land Family/ Children related allowances Social exclusion payments not elsewhere classified Housing allowances Regular inter-hh cash transfers received Interest, dividends, profit from capital investments in incorporated businesses Income received by people aged under 16 Regular inter-household cash transfer paid
(HY030G) (HY040G) (HY050G) (HY060G) (HY070G) (HY080G) (HY090G) (HY110G) (HY130G)
% of household having received an amount 0.0 0.0 99.6 97.8 98.0 95.2 89.0 47.9 95.0
% of household with missing values (before imputation) 0.0 100.0 0.0 1.2 0.7 1.1 6.7 52.1 1.7
% of household with partial information (before imputation) 100.0 0.0 0.4 0.9 1.3 3.7 4.3 0.0 3.2
                   

 

  Cash or near-cash employee income Other non-cash employee income Income from private use of company car Contributions to individual private pension plans Cash profits or losses from self-employment Pensions from individual private plans Unemployment benefits Old-age benefits Survivors benefits Sickness benefits Disability benefits Education-related allowances
(PY010G) (PY020G) (PY021G) (PY035G) (PY050G) (PY080G) (PY090G) (PY100G) (PY110G) (PY120G) (PY130G) (PY140
% of household having received an amount 62.5 77.8 0.0 95.6 0.0 56.5 92.9 90.6 83.7 40.2 58.7 93.1
% of household with missing values (before imputation) 32.0 20.7 0.0 4.4 100.0 43.5 2.6 7.8 12.3 51.3 26.7 6.9
% of household with partial information (before imputation) 5.4 1.5 100.0 0.0 0.0 0.0 4.5 1.5 3.9 8.5 14.7 0.0
  Imputed rent Income from rental of property or land Family/ Children related allowances Social exclusion payments not elsewhere classified Housing allowances Regular inter-hh cash transfers received Interest, dividends, profit from capital investments in incorporated businesses Income received by people aged under 16 Regular inter-household cash transfer paid Repayments receipts for tax adjustment
(HY030N) (HY040N) (HY050N) (HY060N) (HY070N) (HY080N) (HY090N) (HY110N) (HY130N) (HY145N)
% of household having received an amount 0.0 95.3 99.6 97.8 98.0 95.2 89.0 91.8 95.0 98.6
% of household with missing values (before imputation) 0.0 3.4 0.0 1.2 0.7 1.1 6.7 8.2 1.7 0.6
% of household with partial information (before imputation) 100.0 1.3 0.4 0.9 1.3 3.7 4.3 0.0 3.2 0.8
  Cash or near-cash employee income Other non-cash employee income Income from private use of company car Contributions to individual private pension plans Cash profits or losses from self-employment Pensions from individual private plans Unemployment benefits Old-age benefits Survivors benefits Sickness benefits Disability benefits Education-related allowances
(PY010N) (PY020N) (PY021N) (PY035N) (PY050N) (PY080N) (PY090N) (PY100N) (PY110N) (PY120N) (PY130N) (PY140N)
% of household having received an amount 91.8 77.6 0.0 95.6 93.6 91.9 94.0 97.8 98.1 89.1 97.4 93.1
% of household with missing values (before imputation) 3.4 20.7 0.0 4.4 5.6 8.1 2.1 1.7 1.9 9.7 1.2 3.0
% of household with partial information (before imputation) 4.8 1.7 100.0 0.0 0.8 0.0 3.9 0.5 0.0 1.2 1.5 4.0

 

 

 

 

 

 

 
5.3.4. Processing error
 
Data entry and coding Editing controls

Checks to detect processing errors have been implemented in the electronic questionnaire (programmed in Blaise), where the entry of a response is checked for ranges and inconsistencies. Problems are indicated to the interviewer and corrected if needed. Checks in the electronic questionnaire have to be commented by the interviewer, for example when according to the activity calendar the respondent has been employed during the last year but does not declare any employee-income. Correction of not accepted values and inconsistencies that are indicated to the interviewer during the interview is possible by repeating the question in case of a misunderstanding or re-entering the value in the questionnaire if it is correct. Another option is to comment the problem in a remark field which is accounted for during data-editing. The same applies to obligatory interviewer comments.

During post-data-collection-processing the checks included in the questionnaire are repeated and additional checks are conducted. They include formal data checks (e.g. checking of completeness of data copies, correctness of routings and ranges, ratios and balances of entered or computed values, frequencies of new variables) but also checks which use cross-sectional, longitudinal or external information to evaluate plausibility and consistency. Interviewer comments are also taken into account. If necessary, collected values are altered or the value is deleted and thus marked to be imputed later on. Interviewer remarks also can give background information which supports the collected value. Repeated description of the same constellation indicates the necessity of adapting either the question or the check in the next survey.

Distributions and frequency tables of main variables are produced after each major step in the processing to assess the impact of each procedure and to check that the distribution did not become biased. For the evaluation of extensive changes in procedures or newly integrated features dissemination of documentation and reports to all team members and their review and discussion prove to be useful. Final distributions of income variables, European and national indicators are compared with various data sources (e.g. previous EU-SILC waves, ECHP, Microcensus, LFS, HBS, tax statistics and national accounts; see also chapter 4) to identify implausible distributions. As the last step the EUROSTAT target variables are checked by the EUROSTAT SAS checking program to detect errors in computation and coding. Cases which are identified by the checking program but are considered correct are commented and sent to EUROSTAT with the first data transmission. Nevertheless, EUROSTAT’s checks after receiving the datasets mostly identify some remaining problems.

Processing error that arises during post-data-collection-processing mostly can be corrected by adaptation of existing procedures which are repeated after being modified. After correction checks should not identify any more errors or implausible cases and EUROSTAT receives clean datasets.

For the Austrian EU-SILC cross-sectional data 2011 two data transmissions were made, because some data problems were only detected after the first transmission.
5.3.4.1. Imputation - rate

Not requested by Reg. 28/2004.
5.3.4.2. Common units - proportion

Not requested by Reg. 28/2004.
5.3.5. Model assumption error

Not requested by Reg. 28/2004.
5.3.6. Data revision

Not requested by Reg. 28/2004.
5.3.6.1. Data revision - policy

Not requested by Reg. 28/2004.
5.3.6.2. Data revision - practice

Not requested by Reg. 28/2004.
5.3.6.3. Data revision - average size

Not requested by Reg. 28/2004.
5.3.7. Seasonal adjustment

Not requested by Reg. 28/2004.


6. Timeliness and punctualityTop
6.1. Timeliness

Not requested by Reg. 28/2004.
6.1.1. Time lag - first result

Not requested by Reg. 28/2004.
6.1.2. Time lag - final result

Not requested by Reg. 28/2004.
6.2. Punctuality

Not requested by Reg. 28/2004.
6.2.1. Punctuality - delivery and publication

Not requested by Reg. 28/2004.


7. Accessibility and clarityTop
7.1. Dissemination format - News release

Not requested by Reg. 28/2004.
7.2. Dissemination format - Publications

Not requested by Reg. 28/2004.
7.3. Dissemination format - online database

Not requested by Reg. 28/2004.
7.3.1. Data tables - consultations

Not requested by Reg. 28/2004.
7.4. Dissemination format - microdata access

Not requested by Reg. 28/2004.
7.5. Documentation on methodology

Not requested by Reg. 28/2004.
7.5.1. Metadata completeness - rate

Not requested by Reg. 28/2004.
7.5.2. Metadata - consultations

Not requested by Reg. 28/2004.
7.6. Quality management - documentation

Not requested by Reg. 28/2004.
7.7. Dissemination format - other

Not requested by Reg. 28/2004.


8. ComparabilityTop
8.1. Comparability - geographical

Not requested by Reg. 28/2004.
8.1.1. Asymmetry for mirror flow statistics - coefficient

Not requested by Reg. 28/2004.
8.1.2. Reference population
 

Reference population

 Private household definition Household membership
 The reference population of EU-SILC is all private households and their current members residing in the territory of Austria at the time of data collection. Persons living in collective households and in institutions are generally excluded from the target population. There is no difference to the standard EU-SILC concept.  Accommodations with at least one person aged 16 or older who has her/his main residence (Hauptwohnsitzmeldung) in these accommodations. Institutional housing facilities, dwelling units where no person with his/her main residence in the dwelling is 16 years or older are not included. There is no difference to the standard EU-SILC concept. Person living permanently in the accomodation, i.e. at least three months. If a person has her/his main residence not in the household they are treated as members of the household. There is no difference to the standard EU-SILC concept.
 
8.1.3. Reference Period
 
Period for taxes on income and social insurance contributions Income reference periods used Reference period for taxes on wealth Lag between the income ref period and current variables
 The reference period was 2010, accordingly the repayments and receipts of tax adjustments are recorded if the money was paid or received in this year. There is no difference to the standard EU-SILC concept.  The income reference year was 2010. There is no difference to the standard EU-SILC concept.  There are no taxes on wealth in Austria.  This refers to the lag between the income reference period and the household interview date. The fieldwork lasted from 1st March to 30th July. One last interview was transmitted on 8th August. Hence, the gap between the income reference period and the interview date was within the required eight months.
 
8.1.4. Statistical concepts and definitions
 
Total hh gross income Total disposable hh income Total disposable hh income before social transfers other than old-age and survivors' benefits Total disposable hh income before all social transfers
(HY010) (HY020) (HY022) (HY023)
F F F F

 

Imputed rent Income from rental of property or land Family/ Children related allowances Social exclusion payments not elsewhere classified Housing allowances Regular inter-hh cash transfers received Interest, dividends, profit from capital investments in incorporated businesses Interest paid on mortgage Income received by people aged under 16 Regular taxes on wealth (HY120) Regular inter-hh transfers paid
(HY030) (HY040) (HY050) (HY060) (HY070) (HY080) (HY090) (HY100) (HY110) (HY130)
F F F  F F F F F F NC F

 

In EU-SILC 2011 Statistics Austria used register data for the calculation of old-age benefits for persons older than the standard retirement age. The data source to calculate the variable was the wage tax register of 2010. The link between wage tax register is done via the sector specific personal identifier bPK (bereichsspezifisches Personenkennzeichen) which enables the practicable, secure and pseudonymised exchange of information about individuals between administrations.

Old-age benefits also include other income not elsewhere classified if plausible and proportional lump-sum payments if the person is retired (at least 2 monthly regular payments, up to the total lump-sum payment). Since the standard retirement age in Austria is 65 years for men and 60 years for women, it contains all pension benefits paid to persons aged 65/60 years or older. This complies with the EUROSTAT definition.

In EU-SILC 2011 survivors’ benefits were calculated on the basis of register data (c.f. PY100). The source for the calculation were the wage tax register and the register of accident benefits (these latter benefits are paid by Austrian Workers' Compensation Board (the general Austrian accident insurance) and are not taxable and therefore not included in the wage tax registers.

Employee cash or near cash income non-cash employee income company car Employers social insurance contributions Cash profits or losses from self-employment Value of goods produced for own consumption Unemployment benefits Old-age benefits Survivors benefits Sickness benefits Disability benefits Education-related allowances Gross monthly earnings for employees Disability benefits Education-related allowances Gross monthly earnings for employees
(PY010) (PY020) (PY030) (PY030) (PY050) (PY070) (PY090) (PY100) (PY110) (PY120) (PY130) (PY140) (PY200) (PY130) (PY140) (PY200)
F F F F F F F F F F F F F F F F

 

 

The source or procedure used for the collection of income variables The form in which income variables at component level have been obtained The method used for obtaining target variables in the required form
The information on income components is asked from the respondents, register information is used to obtain income information on pensions and survival benefits. To collect the required information to fill the EU-SILC target variables, the income components are split into more differentiated sub-components. These sub-components are defined according to the Austrian regulations and benefit system. For some components only the receipt was asked and the amount was calculated. For example, the respondents were not asked to give the amount of the family allowance, because the amount was calculated on the basis of the information about the family situation (number and age of children). For all variables the net and the gross values were asked from the respondents, except for self-employment incomes, for which only the net income was asked.
For all variables the net and the gross values are in the dataset. If either the net or the gross value was missing for PY010 or PY100, the missing value was calculated on the basis of a net-gross conversion and vice versa. Missing gross values for incomes from self-employment (PY050) were calculated on the basis of the tax payments and social contributions stated by the respondents, missing values for income from employment (PY010) or pension incomes (PY100) are calculated on the basis of the wage tax statistics. For persons over the standard retirement age (women 60; men 65) the values for PY100 were taken from wage tax register, all values for PY110 were taken from the wage tax register and the accident benefits register.

 

 

 

 

 

 

 
8.2. Comparability - over time

EU-SILC 2011 was the eighth regular wave of EU-SILC in Austria with a rotational design. In 2011, the fieldwork was done by the fieldwork organisation of Statistics Austria.

The following comparison focuses on the income target variables in EU-SILC 2011 and EU-SILC 2010. The table presents the median, the number of receiving households/persons and the sum of each income component.

The following tables compare the income components of EU-SILC 2010 and 2011. The median total household income and the total disposable households income of EU-SILC 2011 is by about 5% higher than in 2010.

Most of the other income components feature a higher median in EU-SILC 2011, apart from taxes on income and social contributions (HY140G).

Income target variables on household level: EU-SILC 2010 and EU-SILC 2011
    EU-SILC 2010 EU-SILC 2011
    Median
(in €)		 House-holds Sum
(in Mio €)		 Median
(in €)		 House-holds Sum
(in Mio €)
hy010  Total household gross income* 40,834 3,624,300 182,125 41,342 3,647,513 184,505
hy020  Total disposable household income* 31,125 3,624,300 132,643 31,772 3,647,513 136,402
hy030n  Imputed rents 5,075 2,518,355 12,728 5,077 2,545,815 12,921
hy040n  Income from rental of property or land 3,600 212,109 2,060 4,200 206,944 2,199
hy050n  Family/children related allowances 4,735 1,147,630 6,133 4,735 1,167,386 6,410
hy060n  Social exclusion not elsewhere classified 250 207,933 366 250 206,829 433
hy070n  Housing allowances 1,440 204,608 317 1,536 188,977 335
hy080n  Regular inter-household cash transfer received 3,000 275,290 1,224 3,132 302,858 1,381
hy090n  Interest, dividends, Profit 147 2,693,023 1,741 150 2,701,429 1,954
hy100n  Interest repayments on mortgage 1,173 802,499 1,479 1,295 806,234 1,591
hy110n Income received from people aged under 16 1,452 34,779 96 2,100 43,057 180
hy130n  Regular inter-household cash transfer paid 2,880 422,113 1,637 3,000 399,160 1,585
hy145n  Repayments/receipts for tax adjustment -300 1,771,274 -619 -330 1,734,048 -645
hy140g Tax on income and social contributions 8,926 3,549,352 47,834 8,915 3,558,063 46,518
Source: Statistics Austria, EU-SILC 2010 and EU-SILC 2011

On personal level the income from employees – the main source of income of households – in 2011 is about 3% higher than in 2010. As on household level, the median of the most income components are higher in 2011 than in 2010, the exceptions are sickness benefits (PY120N) and disability benefits (PY130N). The sum of old-age benefits decreased significantly due to fact that for the EU-SILC 2011 operation all survivors’ benefits are gathered in the respective variable PY110N (This was done according to the ESSPROS manual of 2008, c.f. Doc 65, Operation 2011, p.348-349).

Income target variables on personal level: EU-SILC 2010 and EU-SILC 2011
    EU-SILC 2010 EU-SILC 2011
    Median
(in €)		 Persons Sum
(in Mio €)		 Median
(in €)		 Persons Sum
(in Mio €)
py010n  Employee cash or near cash income* 17,500 3,867,343 73,984 18,000 3,903,566 74,888
py020n   Non-cash employee income* 600 819,128 1,125 600 766,620 850
py035n  Contribution to individual pension plans 800 1,792,673 1,979 818 1,852,974 2,214
py050n  Cash benefit or losses from self-employment 9,596 802,591 11,382 12,000 767,003 11,862
py080n  Pension from  individual private plans 3,600 28,646 160 5,400 32,051 257
py090n  Unemployment benefits 3,000 671,757 2,771 3,250 644,314 2,980
py100n  Old-age benefits 15,760 1,753,052 29,849 15,759 1,677,305 28,304
py110n   Survivor' benefits 6,908 75,272 571 7,885 359,305 2,936
py120n  Sickness benefits 1,215 222,309 461 1,200 208,770 573
py130n  Disability benefits 12,372 183,402 2,246 11,821 194,449 2,334
py140n  Education-related benefits 1,200 133,369 301 1,320 120,256 315
py200g Gross monthly earnings for employees 1,900 3,379,753 7,557 1,941 3,411,401 7,547
Source: Statistics Austria, EU-SILC 2010 and EU-SILC 2011
8.2.1. Length of comparable time series

Not requested by Reg. 28/2004.
8.3. Comparability - domain

Not requested by Reg. 28/2004.


9. CoherenceTop
9.1. Coherence - cross domain

(a) Wage tax statistics 2010 – cross annual incomes of employees

Since the income reference period of EU-SILC 2011 was 2010 a comparison with the wage tax statistics of 2010 was carried out.

The wage tax statistics (WTS) records the incomes of employees and pensioners if the income is gained at source in Austria. Here, the WTS is used to validate the distribution of the most important income component on personal level, the income from employment (PY010).

Since in EU-SILC 2011 the calculation of old-age benefits is based on the wage tax register a comparison for pension income would not be particularly helpful. Therefore the comparison of pensions in the WTS and EU-SILC is omitted and only incomes from employment are compared.

For this comparison conceptual differences between WTS and EU-SILC are to be considered. An important share of these differences can be explained by the different coverage of EU-SILC and the WTS. The following lists the main differences:

  1. EU-SILC does not cover persons outside private households;
  2. EU-SILC cannot cover persons who have died or moved to another country between the tax reference period and the fieldwork period;
  3. EU-SILC does not cover employment incomes received by persons who are aged 15 year or younger (Incomes received by people aged under 16 are recorded in variable HY110N/G on household level and it is not differentiated between income from employment and other means of income);
  4. Sum lump-sum payments are registered in the WTS but only partially in EU-SILC;
  5. WTS includes an unknown number of fictitious income records by which taxpayers attempt to achieve a more advantageous tax base.

As in the last years the distribution of employees’ income from the wage tax statistics and EU-SILC are quite similar. The number of employees in EU-SILC is slightly lower than in the wage tax statistics. This underreporting of employees is maybe due to coverage differences between EU-SILC and the WTS as well as a possible underestimation of short employment spells in EU-SILC.

Underreporting of shorter employments spells with lower annual wage is also a possible explanation for the overestimation of wage at the lower fringe of the income distribution in EU-SILC. While overall the match between the two statistics is quite satisfying, EU-SILC data tend to underestimate higher incomes and overestimates lower incomes. Thus the income distribution of EU-SILC overestimates the equality of the income distribution of employees’ income.

Comparison of gross annual income of employees 2010 - wage tax statistics 2010 and EU-SILC 2011 (employed for at least 1 month)
  WTS (in Euro) EU-SILC 2011 (in Euro)
Total Male Female Total Male Female
10% … 4,298 6,205 3,170 5,263 8,731 4,013
20% … 10,090 15,400 7,140 11,808 17,731 8,482
25% … 13,008 19,368 9,486 14,588 20,533 10,728
30% … 15,747 22,536 11,629 17,130 22,999 12,525
40% … 20,703 27,102 15,491 21,140 26,927 16,785
50% … 25,471 31,147 19,015 25,200 30,800 19,300
60% … 30,063 35,556 23,006 29,400 35,200 22,919
70% … 35,349 41,424 27,738 34,720 41,300 26,900
75% … 38,753 45,302 30,610 37,671 43,650 29,400
80% … 42,993 50,282 34,066 42,000 48,566 33,041
90% … 56,990 66,714 44,902 53,584 63,153 42,000
Mean 29,882 36,476 22,540 29,074 35,542 21,926
Persons 3,743,311 1,971,989 1,771,322 3,598,610 1,889,216 1,709,394
Source: Statistics Austria, EU-SILC 2011 and Wage Tax Statistics 2010

 

 Following the assumption that short employment spells are underreported in EU-SILC a restriction to employments lasting the entire year (at least 11 months) should improve the comparison. The comparison is presented in the table below. The match of the distribution is improved for the lower half of the distribution but not for the higher percentiles. Particularly incomes of male employees are underestimated at the top of the income distribution.

Comparison of gross annual income of employees 2010: wage tax statistics 2010 and EU-SILC 2011 (employed for the entire year)
  WTS (in Euro) EU-SILC 2011 (in Euro)
Total Male Female Total Male Female
10% … 10,968 18,742 7,724 11,970 18,200 9,532
20% … 17,200 25,035 12,763 17,674 23,800 13,578
25% … 19,641 26,971 14,684 19,480 25,200 15,523
30% … 22,027 28,749 16,430 21,620 27,034 17,080
40% … 26,190 32,145 19,522 25,200 30,740 19,460
50% … 30,090 35,800 22,931 28,700 33,970 22,444
60% … 34,300 40,386 26,711 32,780 38,500 26,196
70% … 39,678 46,529 31,297 37,800 43,500 30,000
75% … 43,172 50,630 34,128 41,434 47,600 32,760
80% … 47,596 55,779 37,600 44,520 52,400 35,180
90% … 61,857 72,808 48,538 57,500 67,200 44,051
Mean 35,195 43,086 26,548 33,374 40,127 25,514
Persons 2,913,811 1,523,438 1,390,373 2,914,290 1,567,549 1,346,740

Source: Statistics Austria, EU-SILC 2011 and Wage Tax Statistics 2010

(b) Microcensus 2011 - Tenure status and rent-payments

The Austrian Microcensus is a quarterly household survey with a sample of more than 22,000 randomly selected households. The Microcensus operates like EU-SILC with a rotational longitudinal design. The Microcensus is the basis of the Austrian labour force survey (LFS) and because of the size of the sample it is one of the most important sources for socio-demographic information in Austria.

In this report Microcensus data are used to compare information on the legal status of housing and housing costs with the information recorded in EU-SILC. Since the Microcensus is one of the main data sources on housing statistics in Austria it is a valuable basis for comparisons. Furthermore, the information used for the calculation of imputed rents in EU-SILC is taken from the Microcensus. Thus, the comparison is not only of importance for the variables taken into account but also – at least indirectly – for the validity of imputed rents.

However, the Microcensus and EU-SILC apply different concepts and use different variables. For example, the definition of the tenure status is different in EU-SILC and the Microcensus. Hence, some categories of the tenure status of the original variable in EU-SILC and the Microcensus are merged to allow for the comparison.

Comparison of tenure status - microcensus 2011 and EU-SILC 2011  
  Microcensus 2011 EU-SILC 2011
  Households in % Households in %
Total 3,650,398 100.0 3,650,398 100.0
House owner 1,434,392 39.3 1,442,256 39.5
Owner of apartment 392,338 10.7 384,474 10.5
Tenure: community housing 276,796 7.6 293,389 8.0
Tenure: cooperative society 585,855 16.0 522,052 14.3
Tenure: other 634,684 17.4 664,843 18.2
Subtenancy 38,988 1.1 52,593 1.4
Rentfree house / apartment 287,345 7.9 290,792 8.0
Source: Statistics Austria, EU-SILC 2011 and Microcensus 2011.

 

The following table compares the rent payments and the costs of services and charges by the size of usable living area and the number of inhabitants in the region. Overall, the housing costs are only slightly overestimated, the median of the monthly housing costs are only about 15 Euro higher than in the Microcensus. The overestimation is slightly larger for other tenancies. In general, the differences between EU-SILC and the Microcensus are higher in categories with few cases like large apartment from community housing or community housing outside Vienna, but also for larger apartments in general.

Comparison of rent payments and costs of services and charges by size of usable living area and number of inhabitants in the region - microcensus 2011 and EU-SILC 2011
    Microcensus 2011 EU-SILC 2011
    Total Community housing  cooperative society Other tenancies Total Community housing  cooperative society Other tenancies
Total Median (in €) 404 320 410 448 420 329 427 464
  Number 1,495,861 276,638 585,568 633,656 1,479,418 293,389 522,052 663,978
Usable Living area                
under 60 m2 Median (in €) 300 250 295 350 310 260 295 387
  Number 579,810 142,665 176,264 260,880 543,443 145,242 145,843 252,359
60 to 120 m2 Median (in €) 471 429 467 510 490 420 495 540
  Number 842,934 131,856 397,303 313,774 866,063 146,724 371,349 347,990
120 and more m2 Median (in €) 700 657 710 700 740 800 670 740
  Number                
Inhabitants in the Region 73,118 2,117 12,000 59,001 69,913 1,423 4,860 63,630
Vienna Median (in €) 400 322 465 420 400 329 476 430
  Number 646,398 202,816 162,952 280,630 637,508 200,017 146,882 290,609
> 100,000 Median (in €) 424 360 388 480 450 420 400 530
  Number 201,221 16,152 93,764 91,305 213,132 18,795 98,425 95,913
> 10,000 Median (in €) 410 310 396 480 400 300 400 460
  Number 284,576 27,660 160,897 96,020 270,116 42,281 121,449 106,386
<= 10,000 Median (in €) 400 278 399 436 450 309 433 498
  Number 363,666 30,009 167,955 165,702 358,663 32,296 155,296 171,071
Source: Statistics Austria EU-SILC 2011 and Microcensus 2011.

 
9.1.1. Coherence - sub annual and annual statistics

Not requested by Reg. 28/2004.
9.1.2. Coherence - National Accounts

National Accounts 2010

The Austrian National Accounts (NA) provide data on the income approach of the GDP. The sector accounts are available only for the combined sectors S14 and S15 (private households and non-profit instiutions serving households (NPISH). The disposable income in that sector can be used for comparison with EU-SILC total income amounts.

For the comparison the values of the national accounts have to be adjusted. This means that the following amounts and estimates have to be deducted from the basic value of the national accounts:

  1. The estimated income value of NPISHs (sector S15) in the case of disposable income. Separated figures for sector S14 (private households) and sector S15 are only calculated for gross income. The total amount of individual consumption of NPISHs (account P3)      is used as a proxy for disposable income of NPISHs and therefore deducted here.
  2. The estimated income value of persons not living in private households. The proportion of persons not living in private households is estimated 1.25 (104,500 of 8,388,000      persons in 2010).
  3. The estimated income value of transfers from reserves. This value is estimated on the basis of the household budget survey (HBS) 2009/10 as 1.1% of the total expenditures of private households.
  4. The income relevant part of imputed rents. These data also come from the NA (account B2N).

Moreover, other relevant conceptual differences between the income concepts of the NA and EU-SILC cannot be quantified:

  1. Non-cash income and lump-sum payments are included in the NA but not to the same extend in EU-SILC.
  2. The NA uses estimates for black economy, income from tips for employees in the hotel, restaurant and cab driver sector, missing incomes due to time lags in the registers, value of self production for construction sites, car repair and housekeeping. The total of the estimates was 8.0% of the GDP in 2008 (~19,900 million Euro). The proportion relevant for disposable income of private households was not estimated in this comparison but might explain some differences.
  3. Self employed income in the NA is a balancing item. There are some difficulties to differ between self employed income for private households and not withdrawn gains from enterprises.
  4. Charity donations and membership fees are estimated in the NA and deducted from the disposable income but not in EU-SILC.
  5. Transnational transfers are included in the NA.
  6. For the net lending/net borrowing for NPISHs no estimate was available and was assumed to be zero.
  7. Property incomes paid (account D4) are 2010 3467,2 Million Euros. These incomes refer in particular to interests for mortgages and are not reflected in the income target variables of EU-SILC (HY010 and HY020).

In parallel to the results of the previous years, the differences between National Accounts and EU-SILC are significant. Again, if property incomes are not considered, the difference is smaller. Though this hints to the problem of collecting and estimating incomes from property, the difference between NA and EU-SILC is still about 7%.

Comparison between National Accounts 2010 and EU-SILC 2011 (in Mio. Euro)  
  Gross incomes of private households Disposable income
  Total Without property income  
Basic Value from national accounts 224,746 207,450 171,533
       
Deduction for non-profit organisations 1) - - 4,143
Deduction for persons not living in private households 2) 2,800 2,585 2,137
Deduction for value of goods self-consumption 3) 2,472 2,282 1,887
Deduction for imputed rents 4) 8,798 8,798 8,798
       
Estimate from National Accounts 210,676 193,785 154,568
       
Estimate from EU-SILC 2010 184,505 179,274 136,402
       
Difference between NA and EU-SILC 2011 17.9 13.6 20.5
Source: Statistics Austria EU-SILC 2011 and national accounts 2010    
1) estimated value, as for disposable income only one estimate is produced for NPOs and private households  
2) estimated on the basis of the population prognosis; 1.25% in 2010    
3) estimate for 1.1% of the total consumption expenditures, HBS 2009/10    
4) NA 2010      
9.2. Coherence - internal

Not requested by Reg. 28/2004.


10. Cost and BurdenTop

Not requested by Reg. 28/2004.


11. ConfidentialityTop
11.1. Confidentiality - policy

Not requested by Reg. 28/2004.
11.2. Confidentiality - data treatment

Not requested by Reg. 28/2004.


12. Statistical processingTop
12.1. Source data

The sampling frame of the first wave households of EU-SILC 2011 was, as in the previous years, the ZMR. In 2011, 2,946 addresses were selected at the beginning of the fieldwork to constitute the rotational group 3, where two addresses turned out to contain more than one household, so that the gross sample ultimately consists of 2,949 houeseholds. The ZMR is a continuously updated population register based on the registration of the main residence. It contains information on the person (date and place of birth,etc.) and on the address(es) of a person. The ZMR is administrated by the Federal Ministry of the Interior (BMI). Data of the ZMR are delivered quarterly to Statistics Austria. The reference date for the sampling of EU‑SILC 2011 was the 30th of September 2010. Households of the previous waves of EU-SILC (2007‑2010) were excluded from the sampling frame.

Though the ZMR is expected to provide an updated image of the resident population of Austria, the sample nevertheless contained obsolete units, mainly due to changes that occurred between the reference date and the fieldwork. These changes are for example persons who emigrated or died since the reference date or persons who did not report changes of their main residence in time. Other units, for example accommodations newly built since the reference date, were not included in the sampling frame.

One problem connected with the sampling frame is the construction of the connection of persons living in one dwelling unit. The entries of the ZMR comprise information on individuals and there is no key or link to identify all persons that are living in a dwelling. So the connection of dwelling units has to be constructed by the individual address characteristics. The connections constructed in this way are not always correct, mainly because of spelling errors or differences of the spelling of the addresses. However, the ZMR is regarded as the most reliable source for drawing representative samples and is also used in other surveys in Austria like the Microcensus/Labour Force Survey.
12.1.1. Sampling design and procedure
 

Type of sampling design
 EU-SILC in Austria uses an integrated rotational design meaning that each year about one fourth of the sample is replaced by a new rotational group. Beginning in 2004, EU-SILC 2011 was the eights year of EU-SILC in Austria as a panel. Each rotational group of the sample 2011 entered the survey in a different year: 2008 (R4), 2009 (R1), 2010 (R2) and 2011 (R3).

Stratification and sub stratification criteria

 The first wave sample of EU-SILC 2011 is a one-stage stratified probability sample. The sample of the first wave was stratified according to 206 interviewer units (Sprengel). These are regional divisions of federal territory which may be approximately combined to Austrian provinces (NUTS 2 units).

Sample selection schemes

The first wave sampling process was carried out according to a stratified one-stage probability sample with disproportional allocation and without replacement. It was planned to select 2,946 addresses for the first wave rotational group of 2011 (R3/11). The number of selected households was determined as approximately 0.08% of all eligible addresses. The starting point in the development of the first wave sample was a proportional allocation by province. However, different expected response rates should be taken into account by the sampling design. The expected response rates of the first wave sample of 2011 were estimated with the response rates of the first wave sample of EU‑SILC 2010. For example, more about 8.4% more addresses were drawn in Vienna because in Vienna the response rate tends to be lower than the average response rate on national level. For provinces with comparatively low response rates an oversample was applied and for provinces with comparatively high response rates an undersample was carried out. So the resulting sample selection scheme facilitated a disproportional allocation in order to compensate for different response rates in different provinces.

Sample distribution over time
 The fieldwork of EU-SILC 2011 was done exclusively by Statistics Austria. The fieldwork for the operation 2011 started in March and ended in August.
 
12.1.2. Sampling unit

Sampling units are dwelling units registered in the ZMR. The sampling frame consisted of all accommodations with at least one person aged 16 or older who has her/his main residence (Hauptwohnsitzmeldung) in these accommodations. Institutional housing facilities, dwelling units where no person with his/her main residence in the dwelling is 16 years or older were excluded from the sample as well as units that have been selected for the prior samples of EU-SILC.
12.1.3. Sampling rate and sampling size
 

Concerning the SILC instrument, three different sample size definitions can be applied:

- the actual sample size which is the number of sampling units selected in the sample

- the achieved sample size which is the number of observed sampling units (household or individual) with an accepted interview

- the effective sample size which is defined as the achieved sample size divided by the design effect with regards to the at-risk-of poverty rate indicator

Given that the effective sample size has been already treated in the section dealing with sampling errors, in this section the attention focuses mainly on the achieved sample size.

 The necessary sample size for Austria was determined in view of framework regulation (1177/2003) to guarantee an effective sample size of 4,500 households. The quantity of the effective sample size is dependent on the so called “design effect” (deff) of the at-risk-of-poverty rate. The design effect is a measure of the change in variance that occurs if a sampling design different to simple random sampling is used. If the design effect is larger than one, more than 4,500 households have to be interviewed in order to achieve the aspired effective sample size. For the survey year 2007 a design effect of approximately 1.33 was estimated by Statistics Austria. In order to estimate the at-risk-of-poverty rate with the same precision that a simple random sample would provide, the sample had to be enlarged by one third. Therefore a sample of about 6,000 households had to be drawn in 2011 to achieve an effective sample size of 4,500. Using the resulting response rates of the last year’s survey the expected response rates for 2011 were determined as 62.5% for the first wave sample and 85% for the follow-up wave samples. In view of these expected response rates a first year gross sample of 2,791 households (at existing addresses) and a follow-up gross sample of 5,015 households would lead to a net cross-sectional sample of about 6,000 households. 

In order to compensate for ineligible elements in the sampling frame (e.g. address no longer existent) - which was estimated as 5.3% (Estimated value based on the quantity of eligible households in the first wave sample of EU‑SILC 2009) - the size of the first wave sample was determined 2,946 addresses. Two addresses turned out to contain more than one household so, ultimately, 2,949 households are in the first wave sample.

Including the 125 split-off households the total number of addresses in the sample amounted to 8,106 (one address couldn’t be edited). 114 of these addresses turned out to be nonexistent (not a proper dwelling unit, dwelling unit is not occupied etc). From the remaining 7,991 addresses in the gross sample 7,930 addresses could be contacted, 61 households could not be contacted. For 6,204 of the successfully contacted addresses interviews could be conducted, 1,726 households refused to or couldn’t take part in an interview. 17 household interviews had to be excluded because of poor quality which led to 6,187 completed household questionnaires which could be used for analysis.

  Total First wave addresses Follow-up addresses
  N % N % N %
Gross sample EU-SILC 2010* 8,106 100.0 2,949 100.0 5,157 100.0
Address edited 8,105 100.0 2,949 100.0 5,156 100.0
Address not edited 1 0.0 0 0.0 1 0.0
             
Used Addresses 8,105 100.0 2,949 100.0 5,156 100.0
Addresses existent 7,991 98.6 2,835 96.1 5,156 100.0
Addresses not existent*** 114 1.4 114 3.9 0 0.0
             
Contacted Addresses 7,991 100.0 2,835 100.0 5,156 100.0
Adresses successfully contacted 7,930 99.2 2,828 99.8 5,102 99.0
Adresses not successfully contacted 61 0.8 7 0.2 54 1.0
             
Successfully contacted addresses 7,930 100.0 2,828 100.0 5,102 100.0
Household questionnaire completed 6,204 78.2 1,774 62.7 4,430 86.8
Refusal to co-operate 1,099 13.9 806 28.5 293 5.7
Entire household entirely away for the duration of fieldwork 413 5.2 169 6.0 244 4.8
Household unable to respond 89 1.1 53 1.9 36 0.7
Other reasons 125 1.6 26 0.9 99 1.7
             
Successful household questionnaire 6,204 100.0 1,774 100.0 4,430 100.0
Interview accepted for the database 6,187 99.7 1,770 99.8 4,417 99.7
Interview rejected** 17 0.3 4 0.2 13 0.3
Source: Statistics Austria, EU-SILC 2011            
* Including split-households in follow-up addresses          
**17 household interviews had to be excluded due to quality issues and were coded as "refusal to cooperate" in db130, because these households will not be approached again in any further wave of the survey.
 
12.2. Frequency of data collection

The fieldwork of EU-SILC 2011 was done exclusively by Statistics Austria. The fieldwork for the operation 2011 started in March and ended in August.

 

  Total First wave Interview Follow-up interviews
Interviewed in % cum. % Interviewed in % cum. % Interviewed in % cum. %
Total 6,187 100.0 100.0 1,770 100.0 100.0 4,417 100.0 100.0
März 2,222 35.9 35.9 829 46.8 46.8 1,393 31.5 31.5
April 1,584 25.6 61.5 465 26.3 73.1 1,119 25.3 56.9
Mai 1,589 25.7 87.2 414 23.4 96.5 1,175 26.6 83.5
Juni 604 9.8 97.0 59 3.3 99.8 545 12.3 95.8
Juli 187 3.0 100.0 3 0.2 100.0 184 4.2 100.0
August 1 0.0 100.0 0 0.0 100.0 1 0.0 100.0
Source: Statistics Austria, EU-SILC 2011

 
12.3. Data collection
 

Mode of data collection

EU-SILC in Austria facilitates computer assisted personal interviews (CAPI) and computer assisted telephone interviews (CATI). 57.4% of all interviews are carried out with CAPI, 42.6% are done with CATI. For the first wave of a rotation only CAPI is used, for follow-up waves both CAPI and CATI interviews are carried out. If a respondent is either away from the household, incapacitated or ill and this status is sustained for the duration of the fieldwork, proxy interviews are carried out as an exception. This means that another household member responds to the questionnaire. Overall 10.8% of all interviews in EU-SILC 2011 were proxy interviews. Compared to EU-SILC 2010 (proxy rate 13.7%) this means a reduction of the proxy rate by about 1/5. The table below displays the mode of data collection according to the target variable RB260:

2-CAPI 3-CATI 5-proxy
(% of total) (% of total) (% of total)
52.5 36.6 10.8

The mean interview duration

The mean interview duration per household is calculated as the sum of the duration of all household interviews plus the sum of the duration of all personal interviews, divided by the number of household questionnaires completed. Only households accepted for the database have to be considered.

 

Average interview duration =43,03 minutes

 
12.4. Data validation

Not requested by Reg. 28/2004.
12.5. Data compilation

.
12.5.1. Weighting procedure
 

Design factor

Non-response adjustments

 Adjustment to external data

Final cross sectional weights
 The design factor, also knwon as the "design weight" was calculated with reference to the design of the sample to take into account the different inclusion probabilities of the selection units in the first wave sample of EU‑SILC 2011. The idea was that if the inclusion probability of an element is low, it should be assigned a higher weight. The design weight then was calculated as the inverse of the inclusion probability of the selection unit. Since the selection probability ps is the same within each stratum, the design weights ds are also constant within each stratum s (of K=206 strata).

ds=1/ps (s ϵ {1,2, …, K})

 

The   aim of non-response weights is the reduction of the bias caused by unit   non-response on household level. The correction of this bias ideally requires   knowledge of the response probability of each of the responding households.   The households could then be re-weighted by the inverse of this probability.   The estimation strategy applied for the first wave households by Statistics   Austria was similar to the strategy for the first wave households in 2010.
Sample selected in 2011 (first   wave)
For the estimation of weights   a logistic regression model was set up to estimate the response probabilities   rhof each household with explanatory   variables known prior to the questionnaire.
 
rh= P(Resp=1|Xj)=(exp(β0+β1X1+…+βJXJ)/(1+ exp(β0+β1X1+…+βJXJ))
 
The final model was obtained   by using a stepwise optimization algorithm to exclude insignificant   explanatory variables. For example, the age of the oldest person in the   household (according to the administrative records) did not appear to be a   sufficiently reliable predictor for non-response. The final model consisting   of twelve significant predictors and the intercept (total final model   chi2=19.4, df=23; final model maxed-rescaled R2=0.0534).
The non-response weights are   calculated as the inverse of the estimated response probability rh. The non-response adjustment of the design weights ds is carried out by multiplying the design weights by the   non-response weights. This way the loss of design weights caused by   households refusing to take part in the questionnaire can be compensated.
 
bh=ds/rh  (s ϵ   {1,2, …, H})
 
Non-response adjustment   between 2010 and 2011 (follow-up waves)
Unlike the non-response   weighting in the initial first wave sample, weighting for longitudinal   non-response is oriented towards individuals. Between two waves a certain   amount of respondents could not successfully be traced, even if their former   households remained in the sample. Those individuals who left the target   population due to natural mortality or migration were of no further concern   for weighting since these processes reflect true changes in the target   population (i.e. residents in private households in the reference   period). 
What was of concern,   however, is the selectivity of participation in the survey over time either   due to refusals or difficulties in tracing particularly mobile individuals.   In essence, the procedure distributed the base weights of these attritors   among similar individuals in the sample. These longitudinal non-response   weights are multipliers for the previous waves’ weights (i.e. non-response   adjusted design weights).
The weighting procedure was   based upon a logistic regression model which predicts response probabilities   among those individuals who were enumerated in the previous wave (t-1) and who were eligible in the   current wave (t). Given   the vast information available in the personal and household questionnaire   such a model could be reasonably sophisticated. Again the rationale is to   distribute previous year's base weights RB060p(t-1)= RB060p(t-1)/rp(t) for the attritors   among similar respondents remaining in the sample. Like in the case of   adjusting for non-response in the first year wave, a logistic regression   model was used to estimate the response probabilities of the persons eligible   in the follow‑up waves of 2011. To compensate for the loss of weights caused   by attrition, the previous year's base weights RB060p(t-1)were multiplied by the inverse of the estimated response rates rp of each person and thus leading to the current year’s base   weights RB060p(t).
A few methodological   refinements were implemented for the preparation of such a model. In order to   include all eligible respondents some explanatory variables had to be   imputed, using a straightforward hot deck procedure using age and the   household as stratification variables. Given the vast number of potential   explanatory variables a stepwise optimisation algorithm was employed to   identify significant predictors in a logistic regression model in which   predictors were recoded into dichotomous dummy variables. Normally, when the   objective of a model is to identify the dimensions according to which a   phenomenon can be best characterised, categorical variables are treated   blockwise, i.e. the respective dummy variables are entered into or removed   from a model simultaneously. Categories with too few observations to produce   significant differences in response rates would then usually be collapsed by   eyeballing the data. With a large number of predictors it becomes a   cumbersome and time consuming task to choose between competing alternatives,   involving decisions each time. Further, the optimization algorithm model   would automatically select variables with many categories which combine the   predictive power of several dummies.In order to avoid this problem all categorical variables were   automatically transformed into dummy variables. Hence the degrees of freedom   for each predictor were equal. Then all the potential dummy predictors were   entered separately into the stepwise algorithm, filtering only those   categories which appeared to significantly improve the chi square statistic.   The parameter estimates obtained from such a model are somewhat difficult to   interpret as they do not necessarily have clear-cut reference categories.   While these kinds of models are certainly not ideal to improve the   understanding of the substantial process leading to non-response, it could   still be held as a useful reduction of the vast number of potential   predictors to obtain a reasonable ratio between the model’s degrees of   freedom and its chi square statistic. Furthermore, it involves hardly   substantial intervention by the researcher and could be fully automated.
In principle, the procedure   to obtain longitudinal non-response weights was identical for all follow-up   waves (R4/08, R1/09, R2/10), only that it would be advisable to estimate   response probabilities separately because the reasons (and thus relevant predictors)   for attrition may shift away from deliberate refusals to more mobility   related problems the more mature the panel becomes. In practice however,   weighting the initial sample of the two year panel, the three year panel and   the four year panel became slightly more complex. The tracing rules imply   that respondents who were missed in one year remained eligible in one   subsequent wave. In the case of the 2009 first wave sample this referred to   individuals who did not respond in 2010 but re-entered the sample in 2011.   For the four year panel another problem arose. Since respondents who refused   to answer the questionnaire for two consecutive waves were not followed up,   two scenarios of re-entries were possible. That is an absence in 2009 or in   2010. Thereby EUROSTAT’s recommendations distinguish clearly between those   individuals who were absent in the target population (e.g. temporarily   abroad, or institutionalized) or those who were not in the sample for other   reasons. The former case inevitably augments the total of weights as it will   augment the population total and can be treated analogously to newborns by   receiving the weight of another household member or the average of other   household members. In practice the population status of absent individuals   was difficult to determine as respondents currently do not provide such   retrospective information. The second case is somewhat more complex since the   weight of temporary attritors had already been distributed among other sample   persons. If such returnees should regain their weight this could only be   achieved by reducing other respondent’s weights. According to EUROSTAT’S   guidelines this could be solved by sharing the weights within the household   into which the returnee enters. In the Austrian situation however returnees   are practically always complete households and there are no weights to be   shared. Assigning these households a zero weight would come next to a massive   waste of effort and money spent to collect information of themany returning individuals   concerned. 
The alternative solution   followed in the Austrian survey was to stop following up persons who   re-entered the sample in 2011. Thus the longitudinal non-response adjustment   could be done on the basis of respondents who were interviewed in 2010 and   were enumerated again in 2011. 
The model for response   probabilities between 2010 and 2011 in Rotation 2/10 produced 50 coefficients   which differed significantly (α = 10%) from zero (total Wald-chi2=384.4; df=50; model maxed-rescaled R2=0.1617).The   models for the non-response rates in R1/09 (total Wald-chi2= 402.5; df=47; model maxed-rescaled R2=0.2582) and   R4/08 (total Wald-chi2= 337.8; df=41; model maxed-rescaled R2=0.2998) yielded similar results. 
Trimming
After response probabilities   were estimated, the attrition weights were trimmed such that the condition   stated on p. 40 in EU-SILC Doc 65 (2011 operation) with   C=2.
Base weight
The base weights RB060p(t) for all further   calculation were produced by multiplying the design weights ds with the inverse of the estimated response probabilities rh. The non-response adjusted weights bh of the first wave sample were calibrated to reliable external   data in order to establish coherence according to important marginal   distributions of the population. These calibrated weights of the first year   wave are the base weights for next year’s second wave. The basis for the   cross-sectional weights had to be on household level. In order to achieve   this, the mean of the personal base weights within a household had to be   assigned to each individual. However, before this could be done, non-sample   persons, i.e. new-borns and new entrants, had to receive personal base   weights too.
Newborns and new entrants
Following EUROSTAT’s   guidelines, individuals who were newly born received their mother’s weight   or, alternatively the average weight of sample persons in the household. In   principle new entrants from outside the target population should be treated   analogously. In absence of the required information of their former   population status all other cohabitants were assigned zero base weights. 
Weight sharing
After every person in each   household of the follow-up waves had received a personal base weight, the   average over all persons m in   each household h was   calculated.
These new household weights wh(t) are the basis of all   further calculations for the cross-sectional weights belonging to the   follow-up waves. Weight sharing is not necessary   for households of the first wave sample, because the non-response adjusted   weights bh are already on household level and are available for every   person in all first-wave households.

 

  
In   accordance with the guidelines of EUROSTAT described in the EU-SILC Doc 65   (2011 operation) all the four rotational subsamples were adjusted to external   marginal distributions in 2011. Like in EU‑SILC 2010 the calibration was   done using the SAS macro "CALMAR" developed by INSEE. 
As in previous years the   main data source for calibration was the Microcensus, a quarterly household   survey with a sample of about 23,000 randomly selected households. As a   reference data base the average of the four quarters of the Microcensus 2011   was chosen. The Microcensus operates with a rotational design like EU-SILC.   The Microcensus incorporates the Labour Force Survey, and due to the size of   the sample it is also one of the most important sources for socio-demographic   information in Austria. Additionally data from the main association of   Austrian Social Security Institutions (Hauptverband der   österreichischen Sozialversicherungsträger) were   used to provide an accurate number of people who were receiving unemployment   benefits (People who received benefits for more than one month during the   income reference period were counted).
The adjustments were carried   out on household level and on individual level and were done with reference   to the following variables: 
Household level: the household   size (four categories: 1, 2, 3 household members and households with 4 and   more household members), tenure status (two categories: rented flat/house or   owned), and region (nine categories: Nuts II level). 
Individual level: Sex, age
In addition to these variables   adjustments were implemented to achieve coherence in 
the number of foreign citizens   using Microcensus data
the number of recipients of   unemployment benefits for a duration of more than one months
An “integrative”   calibration design was applied with the target that on individual level every   person of the household should be assigned the same weight. The individual   characteristics were aggregated on household level, and dummy variables were   constructed for every parameter of the individual adjustment characteristics.   Using CALMAR to carry out these adjustments, a bounded method (logit method)   of CALMAR was used, which defined lower and upper values for the weight   adjustment factors and thus avoiding too extreme weights. Finally adjusted weight Wh for each household h were obtained.

 

Combination   of the four subsamples
The three subsamples of the   follow-up waves were representative of slightly different target populations,   since the initial samples of 2008, 2009 and 2010 could not represent   individuals who were not in the target population at the time the sample was   drawn. This can be referred to as “IN-Population” and   consists mostly of migrants of the years 2008, 2009 or   2010. Their weights   need to be inflated accordingly to give an unbiased representation of the   population in scope. Consequently, when subsamples   were combined those parts of the population which entered the population   needed to be given higher weights.In the case of four subsamples the inflation factors were 3.41,   1.78 and 1.24 respectively if the new entrants were represented in two, three   or four subsamples (These factors take into account the sizes of the   subsamples compared to the whole cross-sectional sample and therefore are   more accurate than the factors 4/1, 4/2 and 4/3 proposed in the guidelines of   EUROSTAT described in the EU-SILC Doc 65 (2011 operation)). All initial   samples were drawn from a population register which contains information on   the previous population status. So it was possible to identify that part of a   sample which could not have been selected into earlier samples as these   individuals were only later added to the sampling frame. 
Final calibration
Adjustments in general were   done to reduce bias in the data. At this stage household weights of the   combined subsamples were again adjusted to external marginal distributions   using the procedure described in chapter 2.1.8.3 yielding the final   cross-sectional weights DB090h on household level and RB050p on personal level respectively.
DB090h=RB050p=gh ₓ wh

 
12.5.2. Estimation and imputation
 
Imputation procedure used Imputed rent Company car
 

As far as item non-response is concerned, Statistics Austria in general only imputes net income variables, missing gross variables are calculated by the net-gross conversion. Item non-response of income variables occurs because of three reasons: either the information whether an income of a particular type is received or not is missing, or the information about the months an income component is received is missing, or the amount of the income is missing.

If the information whether an income component is received is missing, Statistics Austria tries to deduce this information from other variables (e.g. the information on main activity). If it is not possible to derive this information from other questions of the questionnaire (e.g. the activity calendar), it is assumed that no income of this kind was received.

If the information about the number of months is missing, Statistics Austria again tries to derive the length of a period an income component is received from other variables of the survey. If this is not possible, a conditional random value is imputed. This means that the random value does not range automatically from 1 to 12, but that the range of the value is limited by additional information given in the questionnaire.

The question of missing income values receives special attention. Basically, the respondents have more than one possibility to provide information about their income: they can provide either the gross or the net income amount, or they can provide information about their income by declaring an income category. The latter possibility is foreseen to reduce the number of missing income values. The interviewer presents show cards to support the respondent to identify the approximate range, and in case of unwillingness to respond, to reduce the burden to give an answer. If an income variable is missing but either the gross or the net amount is declared, the corresponding missing value is computed according to a model based on Austrian tax data. If the respondent declares an income category to give the information about the income received, Statistics Austria then assigns an income value by selecting a random value from the distribution of valid cases from within this income category.

If the respondent refuses to give any information about the income, Statistics Austria applies deductive, stochastic and deterministic methods of imputation. Deductive methods are applied when the “correct” value can be calculated from information from the questionnaire or the legal regulations. Estimations made by these methods produce comparatively exact results that are relatively close to the missing true value. For other missing income information Statistics Austria applies two approaches: longitudinal and cross-sectional imputation. The longitudinal method is used when the person with the missing information has declared a value in previous waves. For all other cases the cross-sectional imputation method is used.

The longitudinal imputation procedure is based on the row-and-column-method of Little and Su[1]. As suggested by the name, the method uses the row effects and the column effects of the data to identify an appropriate donor case. The row effect, then, is the development of the variable between waves, and the column effect quantifies the relation of one case to all other observations in the sample. This results in a total effect that is used to sort the data file. The nearest neighbour is then used as a donor value.

For cross-sectional imputation Statistics Austria uses regression models as estimation procedures. The estimated values are added with a residual term to prevent the reduction of variance. This estimation procedure requires the specification of several regression models per income component to ensure that a value can be estimated in case of missing values in predictor variables in the most sophisticated models.

The predictors are selected according to their predictive capability (variation of the R2) and / or according to theoretical assumptions about the response variable. In cases where no regression model can be specified the missing information is estimated by using the group means or the group median of the distribution added with a random residual term.


[1] Little, Roderick J.A. / Su, Hong-Jin (1989), Item Non-response in Panel Surveys. In: Duncan, G./Kalton, G./Kasprzyk, D./ Singh, M.P. (1989), Panel Surveys. New York, p. 400-425

 

  Households living in a self-owned dwelling or in a rent-free dwelling or in a dwelling that is rented at a reduced rate enjoy a financial advantage compared to households living in a rented dwelling. The idea of imputed rents is, then, to quantify and estimate that financial advantage and consider this financial advantage for the computation of household incomes. The aim, then, is to estimate the virtual rent for self-owned dwellings (and rent-free dwellings and dwellings rented at a reduced rate), that a household would have to pay on the free market for its dwelling. This virtual rent, then, is used as a proxy for the financial advantage and is calculated as the imputed rent. In EU-SILC 2011, the imputed rent is in short calculated on the basis of the data of the Austrian microcensus. On the basis of the microcensus data linear regression models are used to estimate the rent for those dwellings, for which no rent information is available (including those dwellings that are rented at a reduced price). This estimate is than used as imputed rent. For dwellings that are rented at a reduced rate, the imputed rent equals the difference between the actually paid rent and the estimated virtual rent for the dwelling. The value of the use of a company car is included in PY010. The number of month where a company car was available for private use during the income reference period was collected in the questionnaire of EU-SILC 2011. For each month 300 Euro were assumed as money value for the private use of a company car.
 
12.6. Adjustment

Not requested by Reg. 28/2004.


13. CommentTop

EU-SILC 2011 in Austria is fully comparable with Eurostat guidelines in terms of the 2011 ad-hoc module.

The entire questionnaire of EU-SILC 2011 in Austria (including the ad-hoc module) in German is attached in the annex.


AnnexesTop
Questionnaire of EU-SILC 2011 in Austria (German language) (2011_Questionnaire_AT (DE).pdf)