Back to top
National reference metadata

Latvia

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.

For more information, please consult our metadata website section.

Close

Adult Education Survey 2022

National Reference Metadata in Single Integrated Metadata Structure (SIMS)

Compiling agency: Central Statistical Bureau of Latvia

Need help? Contact the Eurostat user support


Short metadata
Full metadata

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
  • Employer financing and costs of learning
  • Self-reported language skills

For further information see the 2022 AES legislation (http://ec.europa.eu/eurostat/web/education-and-training/legislation) and the 2022 AES implementation manual (http://ec.europa.eu/eurostat/web/education-and-training/methodology).

18 December 2023

Definitions as well as the list of variables covered are available in the 2022 AES implementation manual (http://ec.europa.eu/eurostat/web/education-and-training/methodology).

Individuals, non-formal learning activities.

Individuals aged 18-69 living in private households.

All the territory is covered.

Fieldwork for AES 2022 was form 01.09.2022.-15.01.2023

Despite the main sources of error being non-response and over-coverage, there is no substantial evidence of their impact on the obtained estimates. During data collection process response levels and response representativity was monitored using R-indicators as set out in Schouten, Cobben and Bethlehem (2009) and Shlomo, Skinner and Schouten (2012) to help to identify potential bias by measuring the degree of difference between responding and non-responding sample groups. Based on the monitoring and analyses of the R-indicators, active interventions are implemented during data collection process to increase the chances of obtaining a representative set of final response units. Design weights were adjusted accordingly to fieldwork results. The calculation of sampling errors is provided for key indicators in various cross-sections.

Number, EUR.

Imputation was made for calculation of variable HHINCOME. Imputation was made for respondents who did not provide any income or provided their income as interval. Hot deck imputation - a method for handling missing data in which each missing value is replaced with an observed response from a "similar" unit.

Grossing of the net sample of individuals is done with the help of weighting procedure. The weighting procedure was carried out as follows: at first, the design weights were adjusted for unit non-response and then adjusted design weights were calibrated taking into account demographic data.

To select optimal weights, four different weights were calculated. Final weights were compared by using the estimated accuracy of the target variable estimators. The standard errors of the required variable estimators were compared to the limits of the standard errors and 2016 AES results, thus ensuring that further the sample has better quality.

Four options for pre-calibration weights were calculated:

  • first pre-calibration weights where sample design weight adjusted to unit non-response by homogeneity group (stratum);
  • second pre-calibration weights where first pre-calibration weights divided by estimated response propensities;
  • third pre-calibration weights where second pre-calibration weights multiplied by q coefficient, to calibrate weights so that the sums are equal to those of the basic design weights in each stratum;
  • fourth pre-calibration weights where design weight adjusted for unit non-response dividing by response propensities for valid response set.

Response propensity was estimated with binomial regression. Response propensity model was developed the same way as for the ASD. Variables were selected by minimizing AIC and excluding insignificant regressors from the model. At the threshold of 0.51, which maximizes the number of correct predictions, modification of response propensities to the predicted answer matches the actual response in 67.4% of all cases. The auxiliary variables used to estimate response propensities are:

  1. sex (male, female) and age group (18–24, 25–34, 35–44, 45–54, 55–69);
  2. statistical region (Rīga, Pierīga, Kurzeme, Vidzeme, Zemgale, Latgale);
  3. nationality (1, 13, 17, 21, 45, 9999);
  4. citizenship (Latvian, other);
  5. level of income (20th percentile groups of income);
  6. educational attainment level (groups of ISCED levels: 0–2, 3–4, 5–8).

Estimated response propensities were further used in weighting.

Calibration to population size was applied to each of the four pre-calibration weights in the following groups:

  • sex (male, female) and age group (18–24, 25–34, 35–44, 45–54, 55–69);
  • statistical region (Rīga, Pierīga, Kurzeme, Vidzeme, Zemgale, Latgale);
  • educational attainment level (groups of ISCED levels: 0–2, 3–4, 5–8).

Sample errors in the main indicators also are calculated in different breakdowns. Accuracy of the four final weights was compared among target variables (Participation rate in formal education and training, Participation rate in non-formal education and training, and Everyday learning) thus finding out the best weights. Compared to others, the estimates of the variables Everyday learning and Participation rate in non-formal education and training is slightly higher when using the first weights, but the estimate Participation rate in formal education is slightly lower when using the first weights. The evaluation of the estimates for the variable of interest shows smooth consistency among all weights, precision and confidence intervals are very close among methods.

The sampling frame was built from the register data available in April 2022, and it covered the whole target population, i.e., usual residents of Latvia who at the start of the survey (1 September 2022) were aged 18–69 and lived in private households. In total sample frame included 1 035 816 people.

To implement sampling procedures, also a sample frame as list of units (persons) was prepared. CSB usually uses multi-mode research method, but due to COVID-19 and thus cancellation of face-to-face interviews it was decided to shift to CATI in 2020 (as CSB has phone number of the most of sample units). As for now, CAPI has not returned to the previous level.

Sample allocation was calculated based on 2016 AES data using two variables – FED (formal education) and NFENUM (non-formal education). In 2022 AES, a one-stage stratified systematic sampling of persons is used and several stratification options were tested. Territorial stratification is made according to the registered place of residence. In AES there are 48 strata divided by statistical region (Rīga, Pierīga, Kurzeme, Vidzeme, Zemgale, Latgale), sex (male, female) and age group (18–24, 25–34, 35–54, 55–69). Based on the scope differences between 2016 AES and 2022 AES, strata allocation in AES 2022 was following: for age 18–24 were duplicated with correction on lower response rate from estimation of age group 25–34, 55–69 were estimated based on 55–64.

To distribute the sample evenly over the whole territory and among levels of education, systematic sampling was used, and units were sorted within strata, survey polygons (small territories), administrative and territorial units, educational attainment levels.

Sampling procedure resulted in the gross sample volume of 8 764 persons.

See table 18.1 “Source data” in annex “LV - QR tables 2022 AES (excel)”.

Every 6 years.

AES was conducted and data was produced according to Eurostat’s implementation monitoring table for AES. The fieldwork ended on January 15 2023, and production of microdata was completed by July 15th 2023. Data is planned to be published in the October 2023.

See table 15.1 “Deviations from 2022 AES concepts and definitions” in annex “LV - QR tables 2022 AES (excel)”.

No additional variables related to COVID-19 were collected.

For AES 2022 several changes and improvements were made. The questionnaire and electronic questionnaire were improved. Data collection modes were used slightly different – in AES 2016, CATI and CAPI modes were used more equally, however, in the AES 2022, CATI was used mostly, and CAPI was used for respondents without correct or not available phone numbers. Also sampling was changed and improved based on AES 2016 data and the one-stage sample was used instead of two-stage. Improvements to weighting schemes was made as well.

See table 15.2 “Comparability - over time” in annex “LV - QR tables 2022 AES (excel)”.