Reference metadata describe statistical concepts and methodologies used for the collection and generation of data. They provide information on data quality and, since they are strongly content-oriented, assist users in interpreting the data. Reference metadata, unlike structural metadata, can be decoupled from the data.
Statistical Office in Gdańsk - Centre for Education and Human Capital Statistics
1.3. Contact name
Piotr Jaworski
1.4. Contact person function
Chief specialist
1.5. Contact mail address
Statistical Office in Gdańsk
Danusi 4
80-434 Gdańsk
POLAND
1.6. Contact email address
p.jaworski@stat.gov.pl
1.7. Contact phone number
+(48) 58 76 83 100
1.8. Contact fax number
Not applicable.
2.1. Metadata last certified
10 January 2024
2.2. Metadata last posted
10 January 2024
2.3. Metadata last update
10 January 2024
3.1. Data description
The Adult Education Survey (AES) covers adults’ participation in education and training (formal - FED, non-formal - NFE and informal learning - INF). The 2022 AES focuses on people aged 18-69. The reference period for the participation in education and training is the twelve months prior to the interview.
Information available from the AES is grouped around the following topics:
Participation in formal education, non-formal education and training and informal learning
Volume of instruction hours
Characteristics of the learning activities
Reasons for participating
Obstacles to participation
Access to information on learning possibilities and guidance
Regulation of the Council of Ministers of 19 November 2021 on the program of statistical surveys of official statistics for the year 2022 (Journal of Laws 2021, item 2303)
6.2. Institutional Mandate - data sharing
Not applicable.
7.1. Confidentiality - policy
All statistics collected and published by Statistics Poland are governed by the Act on Official Statistics. This Act establishes the statistical independence of Statistics Poland.
Statistics Poland cannot publish, or otherwise make available to any individual or organization, statistics that would allow the identification of data for any individual person or entity.
The question of confidentiality is explained in the art. 10, 38 and 38a of the Act on Official Statistics:
Article 10.
Identifiable individual data collected in statistical surveys are subject to absolute protection. Such data may only be used to prepare statistical studies, compilations and analyses, as well as to create a statistical survey sampling frame by the President of Statistics Poland; making such data available or using them for purposes other than those specified herein is prohibited (statistical confidentiality).
Article 38.
1. Identifiable unit data obtained in statistical surveys may not be published nor made available.
2. Statistical data obtained in statistical surveys that can be linked and identified as data concerning a specific natural person, as well as information and statistical data characterising economic and financial results of national economy entities conducting economic activity, may not be published nor made available if the given aggregation consists of fewer than three entities or the share of one entity in a given compilation is greater than three-fourths of the whole.
3. In the case of national economy entities, the information and statistical data referred to in sec. 2 may be published if the person authorised to represent a given entity has consented to the publication of specific data characterizing the economic and financial results of that entity.
Article 38a.
1. The President of Statistics Poland, at the request of entities referred to in Art. 25 25 sec. 1 point 9, justified by the preparation of specific programmes, forecasts and analyses, may provide these entities with identifiable unit data of public finance sector entities within the meaning of Art. 9 of the Act of 27 August 2009 on public finances.
2. The data made available in accordance with sec. 1 may be used only for the purpose indicated in the application, subject to the rules referred to in art. 38 sec. 1 and in Regulation No. 223/2009.1. Identifiable unit data obtained in statistical surveys may not be published nor made available.
7.2. Confidentiality - data treatment
There are various measures to ensure confidentiality. The exchange of microdata between interviewers and statistical offices is performed by a central server. Data are rendered anonymous. Aggregated data are disseminated only after its control. Individual data are only available to those developing results. Statistical confidentiality is applied.
8.1. Release calendar
Data are disseminated nationally according to a publicly available advance release calendar.
Article 14 point 2 of the 'Act on Official Statistics' states that shall provide equal, non-discriminatory and simultaneous access to statistical information, and especially to major figures and indicators. Data are released simultaneously to all users and are available on Statistics Poland's website (http://stat.gov.pl/en/) in the form of announcement.
Every 6 years.
10.1. Dissemination format - News release
January 2024: News release (Polish/English version).
Aggregations of statistical data not included in the programme of statistical surveys of official statistics may be conducted (under statistical confidentiality) at individual requests (observing statistical confidentiality), on a commission. Unidentifiable individual data sets are disseminated by Statistics Poland for a payment for scientific purposes to universities and other higher education establishments and to scientific institutes conducting research. The datasets prepared for the ordering party shall be verified for disclosure of any confidential data. The reviewed sets are transferred to the ordering party on an electronic storage device (CD) or through the TransGUS platform – a safe communication and information channel designed to transmit data in electronic form. Social statistics (including AES data) and census data are also made available on the Scientist Workstand, an IT environment dedicated to in-depth, specialised analyses conducted by scientific and research institutions or universities.
Methodological information will be included in the Polish publication 'Adult education 2022'. Please refer also to the attached (see section 18.3 Data collection) interviewers instruction (in Polish only).
10.6.1. Metadata completeness - rate
Not applicable.
10.7. Quality management - documentation
According to article 3 of Act on Official Statistics - the official statistics shall ensure reliable, objective and systematic information for the society, the state and public administration bodies and economic entities.
Documentation on procedures applied for quality management and assessment is available on the website: Quality in statistics.
11.1. Quality assurance
The legal basis of the quality measurement and assessment is the internal Regulation of Statistics Poland's President No 35 of 28th December 2011 which came into force in the 4th quarter of 2012. The quality assessment of statistical surveys is conducted according to an annual quality program for the official statistics, which is in line with the Regulation of Statistics Poland's President No 35 of 28th December 2011.
Procedures and rules applied in quality assessment and monitoring are based on the ESS Quality Standards, i.e. the ESS Quality Declaration, recommendations of the ESS Leadership Expert Group on Quality (LEG) and the European Statistics Code of Practice (CoP). The quality of statistical surveys is evaluated by means of standard quality reports and quality indicators recommended in the ESS Handbook for Quality Reports, as well as by self-assessment checklist and quality audits and reviews.
11.2. Quality management - assessment
The quality of statistical surveys is evaluated by means of standard quality reports and quality indicators recommended in the ESS Handbook for Quality Reports, as well as by self-assessment checklist and quality audits and reviews. Quality review of the previous edition of the survey (AES 2016), was conducted in 2018. A detailed assessment has not yet been made for the AES 2022. The overall quality of the survey can be characterised as follows:
Relevance is maintained every year while drafting the 'Statistical Survey Program of Official Statistics'.
The overall reliability and accuracy of the survey results can be assessed as high due to the large sample size. Nevertheless, there is a systematic increase in the non-response rate (especially among young people), which reduces the precision of the generalised survey results for some indicators.
Concerning timeliness and punctuality, all releases are delivered on time.
Comparability of series is maintained geographically and over time.
General coherence is observable with others cross domain surveys, but there may be some discrepancies with the results of the LFS. These are mainly due to the different objectives and sample design of these surveys.
12.1. Relevance - User Needs
User needs are systematically monitored and taken into consideration while drafting the 'Statistical Survey Program of Official Statistics' for the next year. The organs of the state power, government and self-government administration, organizations of employers and other institutions, provide opinions on statistical surveys and recommend implementation of the new ones.
Data users, e.g. universities, research institutes, enterprises or individuals may indicate their needs by data inquiries and orders.
A classification of users with some indication of their importance:
The most important user group - policymakers at European level (all topics covered by AES).
Other user groups:
Policymakers at national level (all topics covered by AES)
Researchers, students (mainly microdata access)
Enterprises: for own market research activities or for consultancy services in the information sector (some topics)
Media (some topics)
12.2. Relevance - User Satisfaction
No user satisfaction survey has been conducted.
12.3. Completeness
Final dataset covers all variables as requested in the 2022 AES legislation.
12.3.1. Data completeness - rate
Not applicable.
13.1. Accuracy - overall
We consider AES estimates are characterized by a high degree of accuracy due to the large sample size and use of the electronic questionnaire containing plausibility and consistency checks ensuring the data accuracy to a greater extent. However, due to a particularly high non-response rate among young people (aged 18-24) there is a problem in meeting the precision requirements of the regulation for the indicator 'Participation rate in formal education and training'. Efforts were made to reach out to young people, but there was a great lack of willingness among them to take part in the survey. The Polish AES is based on dwelling sample. According to the instructions for the interviewers, persons aged 18-24 were surveyed first in the dwellings, which had been drawn. Nevertheless, young people were often away from their permanent place of residence, mainly for educational purposes. This made it even more difficult to make contact for an interview, despite the possibility of telephone interviews being allowed.
13.2. Sampling error
Overall assessment of the sampling error for 2022 AES:
For most of the 2022 AES key indicators, the sampling errors are relatively small. The problem with meeting the requirements of the regulation is for the indicator 'Participation rate in formal education and training' for population aged 18 to 24 years – this is a consequence of the high non-response rate in this age group.
Sampling method:
The sample draw was realized on the Sampling Frame for Social Surveys (OBS), i.e. the database for which the main source of information is TERYT - Register of Territorial Division of the Country, containing, inter alia, the list of territorial statistical units and the address list of dwellings. The OBS database also contains additional information from the most important administrative registers.
A two-stage sampling scheme with different selection probabilities at the first stage was used. Primary sampling units (PSU) were enumeration census areas. At the second stage, dwellings were selected. All the households from the selected dwellings were supposed to enter the survey.
PSU units were drawn using a stratified sampling scheme with probabilities of selection proportional to the number of dwellings in the respective PSU. Prior to the draw, the PSU were stratified, separately within each region (according to NUTS 2 classification) and in the regions according to the six size classes of the locality.
The strata were defined as a cross-section of: NUTS 2 region, class of localities and 3 classes determined on the basis of share of dwellings with persons aged 18-24 within the PSU. The method of determining the 3 PSU classes was based on available unit data on dwellings with persons aged 18–24 is as follows:
for each PSU, the share of the number of dwellings with persons in the determined age group is calculated in relation to the number of dwellings in a given PSU.
on the basis of the distribution of the calculated share in the whole PSU population, each PSU was divided into 3 approximately equal parts; in this way, we separate the set of enumeration census areas which contain significantly more addresses with persons aged 18 to 24.
The final set of strata is obtained by dividing each stratum and each PSU into two parts of similar size; this division is determined by sorting all dwellings in a given PSU according to the characteristic of the number of people working; this approach results in a number of strata of the order of 400.
The sample allocation was determined according to the algorithm described in the paper: Wesołowski, Wieczorkowski (2017) [Wesolowski J., Wieczorkowski R. (2017), An eigenproblem approach to optimal equal-precision sample allocation in subpopulations, Communications in Statistics - Theory and Methods, 46: 5, 2212-2231]; this algorithm uses the eigenvalues of a specially constructed matrix to determine the optimal allocation of the sample in a two-stage scheme between subpopulations and strata; the optimality of the solution consists in obtaining the smallest possible relative standard error (coefficient of variation) of the estimator of the given variable (i.e. of the variable number of persons employed, available in the frame) at the level of established subpopulations (i.e. NUTS 2 regions), within the imposed constraints on the sample sizes at each stage of the draw (we assume that 4 dwellings should be drawn from one PSU on average). In addition, the allocation algorithm will take into account factors related to expected non-response occurring at the second draw stage. As an additional element to increase the efficiency of the sampling design in terms of examining an adequate number of dwellings with 18-24 year old persons, it is proposed to use some over-representation in strata with a significantly higher proportion of dwellings with 18-24 year old persons.
The units of the second-stage draw (SSU) were dwellings. From each pre-drawn PSU, addresses were drawn by simple random sampling without replacement according to the predetermined allocation. As a result, the new sample drawn for the survey consisted of 6646 census areas and, 26148 dwellings. In these dwellings there were 26303 households (gross sample). All households and persons aged 18-69 years were surveyed in the dwellings from the drawn sample (22542 eligible households).
Estimation method:
The applied weighting procedure included the following elements:
1. probability of dwelling selection,
2. level of completeness of the interviews in households,
3. level of completeness of individual interviews.
13.2.1. Sampling error - indicators
See table 13.2.1 “Sampling errors - indicators for 2022 AES key statistics” in annex “PL - QR tables 2022 AES (excel)”.
Method used for the calculation of the coefficients of variation, the standard errors and the confidence intervals - 'ultimate cluster method' implemented in R package 'vardpoor'.
13.3. Non-sampling error
See items 13.3.1 - 13.3.5 below.
13.3.1. Coverage error
Information on sampling frame:
Social Survey Sampling Frame - register of dwellings, based on National Official Register of Territorial Division of the Country (TERYT) and main administrative registers. Sampling frame is updated once a year (with quarterly updates for TERYT database information). Information gathered from conducted social surveys is used to assess the quality of the sampling frame; on this basis, the average level of overcoverage can be estimated to be in the order of 10%.
Undercoverage is rather low, but difficult to assess.
13.3.1.1. Over-coverage - rate
See table 13.3.1.1 “Over-coverage - rate” in annex “PL - QR tables 2022 AES (excel)”.
13.3.1.2. Common units - proportion
Not applicable.
13.3.2. Measurement error
Main sources of measurement errors:
data entry errors
conversion of national questionnaire into EU variables
design, content and wording of the questionnaire
Measures taken to prevent the measurement errors:
methodological and linguistic consultations (external and internal) in order to better reflect in the AES questionnaire the Polish specificities of non-formal education and its forms and improve consistency with LFS questionnaire
pilot testing of the questionnaire
the electronic questionnaire containing plausibility and consistency checks
conversion of national questionnaire into EU variables underwent control and validation process
training of the supervisors and the interviewers in regional statistical offices
methodological consultations by phone or e-mail during a fieldwork
analysis and validation of the control tables and possible data correction at the local and at the country level
13.3.3. Non response error
Assessment on the level of unit non-response:
There is a systematic increase in the non-response rate (especially among young people) in comparison to the previous waves of AES. This reduces the precision of the generalised survey results, but for most of the 2022 AES key indicators, the sampling errors are still relatively small. High non-response rate among young people (aged 18-24) particularly affects the level of sampling error for the indicator 'Participation rate in formal education and training', for which the level of precision is not satisfactory. The main reason of unit non-response are refusals to participate in the survey (67.5% of total unit non-response).
Variables that are more subject to item non-response:
NFEPAIDVAL (expenditure on non-formal education)
HHINCOME (net current monthly household income)
Both variables relate to financial matters and are therefore inherently sensitive in character. Also, the complexity of the concepts causing a problem in determining household income and expenditure on non-formal education contributes to the relatively high item non-response rate for these two variables.
Measures taken to reduce unit non-response:
official letter explaining importance of the survey and special leaflet containing main results of AES 2016 that were sent to the sampled dwellings before the fieldwork
posting the information about the survey on the Portal of the Statistics Poland.
in order to increase the response rate among young people, a recommendation for interviewers was introduced to first interview persons from 18-24 age group, as well as allowing the possibility of a telephone interview
recommendation for interviewers to come back to the household in case of absence of person at home during a first visit or try to make repeated attempts of contact by telephone
13.3.3.1. Unit non-response - rate
See table 13.3.3.1 “Unit non-response - rate” in annex “PL - QR tables 2022 AES (excel)”.
13.3.3.2. Item non-response - rate
See table 13.3.3.2 “Item non-response rate” in annex “PL - QR tables 2022 AES (excel)”.
13.3.4. Processing error
The data entering process was done via the electronic questionnaire which contained plausibility and consistency checks as well as pop-up notifications in case when improbable or incorrect data values were entered. The raw AES data was imported to the 'Data validation software of the survey (SIB)'. The software for data processing enabled rechecking, correction and validation of AES data both on macro and micro level by supervisors in the local statistical offices and also in methodology department.
The electronic questionnaire allowed to substantially reduce the number of processing errors at the data entry stage.
Main processing errors were detected and corrected during conversion of national data set to the default structure of EU AES file (coding errors).
13.3.5. Model assumption error
Not applicable.
14.1. Timeliness
Two data releases. News release (within 11 months after the end of the fieldwork period) and analytical publication (within 16 months after the end of the fieldwork).
14.1.1. Time lag - first result
News release (Polish/English version) - release in January 2024, 11 months after the end of the fieldwork.
14.1.2. Time lag - final result
Analytical, bilingual (Polish/English version) publication - release in May 2024, 16 months after the end of the fieldwork.
14.2. Punctuality
The date of transmission of the final dataset was on 29/09/2023, i.e. within six months of the end of the national data collection period (deadline in accordance with the Regulation).
See table 14.2 “Project phases - dates” in annex “PL - QR tables 2022 AES (excel)”.
14.2.1. Punctuality - delivery and publication
Not applicable.
15.1. Comparability - geographical
See table 15.1 “Deviations from 2022 AES concepts and definitions” in annex “PL - QR tables 2022 AES (excel)”.
No additional variables related to COVID-19 were collected.
15.1.1. Asymmetry for mirror flow statistics - coefficient
Not applicable.
15.2. Comparability - over time
There have been some changes in comparison to AES 2016, but not enough to warrant the designation of a break in series:
introduction of the CATI interview method (in addition to CAPI and PAPI)
allowing proxy interviews (but only in exceptional cases, e.g. if members of the household have reliable information that the respondent could not participate in any form of learning for some objective reason, such as health status)
See also table 15.2 “Comparability - over time” in annex “PL - QR tables 2022 AES (excel)”.
15.2.1. Length of comparable time series
Not applicable.
15.3. Coherence - cross domain
There may be some discrepancies with the results of the LFS regarding indicator 'Participation rate in formal and non-formal education and training' and structure of population. These are mainly due to the different objectives and sample design of these surveys.
See table 15.3 “Coherence - cross-domain” in annex “PL - QR tables 2022 AES (excel)”.
15.3.1. Coherence - sub annual and annual statistics
Not applicable.
15.3.2. Coherence - National Accounts
Not applicable.
15.4. Coherence - internal
AES results for a given data collection round are based on the same microdata and results are calculated using the same estimation methods, therefore the data are internally coherent.
Complete and exact information on the number of the staff involved in administering the survey in full-time equivalent and real costs of the field work is not available. ICT is used at all stages of the statistical process, including data entry. Also, the introduction of the CATI method has reduced the cost of the survey and the burden on respondents in comparison to the AES 2016.
Estimated average time of the interview for the household = 19 minutes.
Estimated average time of the interview for personal questionnaire = 15 minutes.
17.1. Data revision - policy
Not applicable.
17.2. Data revision - practice
Not applicable.
17.2.1. Data revision - average size
Not applicable.
18.1. Source data
AES is a sample survey in the domain of education.
Sampling frame:
Social Survey Sampling Frame - register of dwellings and persons, based on National Official Register of Territorial Division of the Country (TERYT) and main administrative registers.
Type of sampling design:
Stratified random sampling: Two-stage sampling with different sampling probabilities at the first sampling stage. The sample size for Poland was determined at 26148 dwellings.
See table 18.1 “Source data” in annex “PL - QR tables 2022 AES (excel)”.
18.2. Frequency of data collection
Every 6 years.
18.3. Data collection
Sampling method:
Two-stage sampling with different sampling probabilities at the first sampling stage. First stage units were stratified before sampling. The first stage units comprised statistical enumeration districts in urban areas, and census enumeration areas in rural areas. Enumeration statistical region comprises at maximum 9 census areas. Dwellings were selected on the second stage of sampling.
Survey method:
Mixed mode: CAPI, CATI and PAPI.
Implementation of the telephone interviews (CATI) in AES 2022 was a reaction to unforeseen external factors (e.g. Covid-19) and also aimed at reaching specific groups of respondents (young, mobile and economically active) more easily. Reduction of survey costs and the burden on interviewers was an additional benefit.
See also table 18.1 “Source data” in annex “PL - QR tables 2022 AES (excel)”.
Statistical interviewers were equipped with electronic devices functionally prepared for surveying with electronic form, data control and sending correct data to the server, at the same time retaining any requirements associated with data security. Data collection software was developed using Progressive WebApplication (PWA) technology. This allowed the application to be run on any type of device (desktop, laptop, tablet, smartphone) with any operating system (e.g. Android, Windows, Linux, iOS). Entering data process was done via mobile application. The software enabled work in CATI and CAPI mode (entering data directly during the interview) or in PAPI mode (entering data from paper questionnaires). The electronic questionnaire contained plausibility and consistency checks as well as pop-up notifications for when improbable data values (outliers, logical inconsistencies etc.) were entered. All statistical classifications used in the survey were embedded directly in the questionnaire, aimed at helping interviewers with classifications of the given phenomena.
All electronic devices working with the mobile application were synchronized with the 'Surveys management system (CORStat)' which enabled ongoing data transmission to the server.
The raw AES data was imported to the "Data validation software of the survey (SIB)". The software enabled rechecking (data control option), correction and validation of AES data by supervisors in the local statistical offices. Validation of control tables (containing basic aggregated data) meant the work completion at the local level.
The control tables were also analysed and validated at the country level in the methodology department.
The Polish AES questionnaire differs from the EU questionnaire in the design of questions, coding and categories of variables. The data collected from the national questionnaire therefore has been converted into the variables in the EU questionnaire. The process for creating the EU variables has been checked in order to ensure that this process does not create errors. Checking of valid values for the main EU variables has been done. Also, checking that categories of variables in the Polish questionnaire were correctly coded into categories in the EU questionnaire was done.
At the end, the AES database was validated with the Eurostat’s STRUVAL/CONVAL software.
18.5. Data compilation
Weighting:
The applied weighting procedure included the following elements:
1. probability of dwelling selection,
2. level of completeness of the interviews in households,
3. level of completeness of individual interviews.
The above factors comprised general level of completeness on the level NUTS 2 and were used for preliminary correction of design weights. Design weight is defined for each selected dwelling as inverse of probability of selection and equals the ratio: number of dwellings in given NUTS 2 region and stratum in the frame divided by product of number of primary sampling units (PSU) sampled from given region and stratum and the number of dwellings drawn from each PSU in given stratum.
The design weights for dwellings (and households) were then adjusted for the non-responses related to refusals, temporary absence of household members, unable to contact the dwelling drawn, etc.
These weights were adjusted, for each NUTS 2 region, separately in each of the five propensity score groups by correction factors computed as ratios: sum of design weights for eligible dwellings to sum of design weights for successfully interviewed dwellings. The propensity score probabilities used to create 5 groups, were derived from a machine learning model (using the h2o_automl function from the lares package); in this model, the following characteristics available in the sampling frame were used, among others, as predictor variables: NUTS 2 region, class of locality, whether the dwelling contains persons aged between 18 and 69 (and also between 18 and 24 years), number of employed persons, and income decile group (based on tax registers).
The level of completeness of individual interviews was about 65%, the lowest being in the group of people aged 18 to 24 (46%). Adjustment of the weights (from previous stage of correction for households) for missing individual questionnaires was done in the following cross-sections: urban/rural * sex * (18-24 group and nine 5-year age groups).
As a result of taking into account the above two factors related to completeness, adjusted weights for individual interviews were obtained.
Then, these weights were calibrated with the use of data from current demographic estimates concerning sex, age (18-24 group and nine 5-year age groups), age2 (18-24 group and four 10-year age groups), place of residence (urban or rural areas), and region (NUTS 2). Weights were calibrated according to marginal distributions in the following form:
urban/rural areas x sex x age groups,
NUTS 2 x urban/rural areas x age2 groups,
NUTS 2 x sex x age2 groups,
NUTS 2 x urban/rural areas x age groups.
Calibrated weights were calculated with the use of the specially created program in the R system. We used one of calibration algorithms described in the article: Matthew R. Williams, Terrance D. Savitsky 'Calibration of Survey Weights under Large Number of Constraints', arXiv 2019. A variant of the calibration algorithm was used which ensures range restrictions for weights i.e. the spread of the resulting weights in relation to the weights before calibration is maintained within a range between 1/3 and 3.
18.5.1. Imputation - rate
No imputation was used in the survey.
See table 18.5.1 “Imputation - rate” in annex “PL - QR tables 2022 AES (excel)”.
The Adult Education Survey (AES) covers adults’ participation in education and training (formal - FED, non-formal - NFE and informal learning - INF). The 2022 AES focuses on people aged 18-69. The reference period for the participation in education and training is the twelve months prior to the interview.
Information available from the AES is grouped around the following topics:
Participation in formal education, non-formal education and training and informal learning
Volume of instruction hours
Characteristics of the learning activities
Reasons for participating
Obstacles to participation
Access to information on learning possibilities and guidance
We consider AES estimates are characterized by a high degree of accuracy due to the large sample size and use of the electronic questionnaire containing plausibility and consistency checks ensuring the data accuracy to a greater extent. However, due to a particularly high non-response rate among young people (aged 18-24) there is a problem in meeting the precision requirements of the regulation for the indicator 'Participation rate in formal education and training'. Efforts were made to reach out to young people, but there was a great lack of willingness among them to take part in the survey. The Polish AES is based on dwelling sample. According to the instructions for the interviewers, persons aged 18-24 were surveyed first in the dwellings, which had been drawn. Nevertheless, young people were often away from their permanent place of residence, mainly for educational purposes. This made it even more difficult to make contact for an interview, despite the possibility of telephone interviews being allowed.
Number, EUR.
Weighting:
The applied weighting procedure included the following elements:
1. probability of dwelling selection,
2. level of completeness of the interviews in households,
3. level of completeness of individual interviews.
The above factors comprised general level of completeness on the level NUTS 2 and were used for preliminary correction of design weights. Design weight is defined for each selected dwelling as inverse of probability of selection and equals the ratio: number of dwellings in given NUTS 2 region and stratum in the frame divided by product of number of primary sampling units (PSU) sampled from given region and stratum and the number of dwellings drawn from each PSU in given stratum.
The design weights for dwellings (and households) were then adjusted for the non-responses related to refusals, temporary absence of household members, unable to contact the dwelling drawn, etc.
These weights were adjusted, for each NUTS 2 region, separately in each of the five propensity score groups by correction factors computed as ratios: sum of design weights for eligible dwellings to sum of design weights for successfully interviewed dwellings. The propensity score probabilities used to create 5 groups, were derived from a machine learning model (using the h2o_automl function from the lares package); in this model, the following characteristics available in the sampling frame were used, among others, as predictor variables: NUTS 2 region, class of locality, whether the dwelling contains persons aged between 18 and 69 (and also between 18 and 24 years), number of employed persons, and income decile group (based on tax registers).
The level of completeness of individual interviews was about 65%, the lowest being in the group of people aged 18 to 24 (46%). Adjustment of the weights (from previous stage of correction for households) for missing individual questionnaires was done in the following cross-sections: urban/rural * sex * (18-24 group and nine 5-year age groups).
As a result of taking into account the above two factors related to completeness, adjusted weights for individual interviews were obtained.
Then, these weights were calibrated with the use of data from current demographic estimates concerning sex, age (18-24 group and nine 5-year age groups), age2 (18-24 group and four 10-year age groups), place of residence (urban or rural areas), and region (NUTS 2). Weights were calibrated according to marginal distributions in the following form:
urban/rural areas x sex x age groups,
NUTS 2 x urban/rural areas x age2 groups,
NUTS 2 x sex x age2 groups,
NUTS 2 x urban/rural areas x age groups.
Calibrated weights were calculated with the use of the specially created program in the R system. We used one of calibration algorithms described in the article: Matthew R. Williams, Terrance D. Savitsky 'Calibration of Survey Weights under Large Number of Constraints', arXiv 2019. A variant of the calibration algorithm was used which ensures range restrictions for weights i.e. the spread of the resulting weights in relation to the weights before calibration is maintained within a range between 1/3 and 3.
AES is a sample survey in the domain of education.
Sampling frame:
Social Survey Sampling Frame - register of dwellings and persons, based on National Official Register of Territorial Division of the Country (TERYT) and main administrative registers.
Type of sampling design:
Stratified random sampling: Two-stage sampling with different sampling probabilities at the first sampling stage. The sample size for Poland was determined at 26148 dwellings.
See table 18.1 “Source data” in annex “PL - QR tables 2022 AES (excel)”.
Every 6 years.
Two data releases. News release (within 11 months after the end of the fieldwork period) and analytical publication (within 16 months after the end of the fieldwork).
See table 15.1 “Deviations from 2022 AES concepts and definitions” in annex “PL - QR tables 2022 AES (excel)”.
No additional variables related to COVID-19 were collected.
There have been some changes in comparison to AES 2016, but not enough to warrant the designation of a break in series:
introduction of the CATI interview method (in addition to CAPI and PAPI)
allowing proxy interviews (but only in exceptional cases, e.g. if members of the household have reliable information that the respondent could not participate in any form of learning for some objective reason, such as health status)
See also table 15.2 “Comparability - over time” in annex “PL - QR tables 2022 AES (excel)”.
