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.
The Swiss Earnings Structure Survey (SESS) has existed since 1994 and is carried out every 2 years. In 2012, the SESS was revised to ensure better matching with the Eurostat variables. However, the datasets delivered to Eurostat have fewer observations than the datasets used for the national publication due to missing values in some variables.
The Swiss Earnings Structure Survey (SESS) has existed since 1994 and is carried out every 2 years. In 2012, the SESS was revised to ensure better matching with the Eurostat variables. However, the datasets delivered to Eurostat have fewer observations than the datasets used for the national publication due to missing values in some variables.
2.2. Classification system
ISCO, ISCED, NACE
2.3. Coverage - sector
All enterprises of three or more employees are covered by the survey. The primary sector (agriculture) is not included.
2.4. Statistical concepts and definitions
The survey is based on a sample extracted from the Swiss Business and Enterprise Register. All enterprises of three or more employees are included in the universe.
2.5. Statistical unit
The enterprises are the statistical unit.
2.6. Statistical population
About 30,000 enterprises and 850,000 wages in the dataset sent to Eurostat (35,000 enterprises and 1.6 million employees for national publication).
2.7. Reference area
The survey covers all active enterprises from the secondary and tertiary sector in Switzerland.
2.8. Coverage - Time
The survey covers all wages of the month of October.
2.9. Base period
October 2014.
3.1. Source data
Sampling methodology: A stratified two-stage random sample
The basis for sampling was the Swiss Business and Enterprise Register. The universe for sampling includes all enterprises active in August 2014 on the territory of Switzerland, ranging from NACE Rev.2 division 2. The minimum size of the enterprise is 3 employees.
However the survey took into account apprentices and trainees and these two categories of workers were not included within the Eurostat datasets (nor in the national publication), because they have not yet been checked. The wage data of subcontracted workers had to be delivered by their employment agencies.
The SESS (private sector and public enterprises) is based on a random sample at two levels: the first level concerns the stratified enterprises, while the second level concerns employees in the enterprises. Enterprises are stratified according to three criteria: enterprise size (3 classes), economic branch (39 aggregations of NACE Rev.2 divisions) and geographical entity (7 major regions, 8 cantons and 1 city).
In total, the SESS stratification includes more than 1600 cells (strata). While for some strata the selection of enterprises is exhaustive (this is always the case for an enterprise with more than 49 employees, but not exclusively), for the majority of strata enterprises are sampled by random selection.
The public administrations are not included in the datasets delivered to Eurostat for the SESS 2014, but are covered by the SES survey and published in the national publication. The reason why we decided not to include them in the delivery to Eurostat was the insufficient quality of those observations (missing values, complexity of the structure) and therefore the difficulty in weighting a too limited number of observations.
The number of salary records of a sampled enterprise to be delivered depends on the size of the responding unit. Enterprises with fewer than 20 employees have to provide all salaries. For the size class 20 to 49 employees one out of two is to be reported, while in the size class 49+, at least one out of three is expected.
Sampling of the employees has to be done by the responding units themselves, according to the principles of random selection.
3.2. Frequency of data collection
Data are collected every two years.
3.3. Data collection
The Swiss Earnings Structure Survey 2014 was produced by 11 permanent statworkers (6 of them with a university education), assisted by 14 temporary employed statworkers, who were carefully trained and supervised. The production process comprises dispatching the questionnaires, counselling enterprises in case of questions; validation of the returned questionnaires and electronic data-files and, in case of doubt, calling back for clarification; furthermore reminding tardy enterprises; and finally entering the data into the provisional database.
3.4. Data validation
The completeness of the returned questionnaires was at first checked by eyes. The plausibility of the data was then checked by means of electronic tools, after entering the data in the data capturing system. The plausibility checks generally were partly constraining, partly not constraining (“warnings”); that means, besides some impossible categories, the warnings could, after verification, be ignored.
3.5. Data compilation
Once the data were definitively cleared, specialised research associates attributed the final weights for national publication. After further checking, the final database (“Swiss variables”) was set up.
3.6. Adjustment
Specialised research associates cleaned the data again to match Eurostat’s requirements and attribute the new weights.
4.1. Quality assurance
Enterprises could use three channels to send their data. Each channel had its own quality controls:
Paper questionnaire: the completeness is checked by qualified statworkers and a set of rules controls the relevance of the content
Online questionnaire (esurvey): a set of rules controls the content and drop-down menu guarantee the exactitude of some variables, missing values are also not allowed
Electronic delivery (xlsx, xml): two sets of coherence rules control the format and the content, and missing values are also not allowed
Some other controls are made based on registers, for example:
The Swiss Business and Enterprise Register is used to check the number of delivered employee, the address, etc.
We also used the population register for some personal controls.
4.2. Quality management - assessment
The revision in 2012 involved a process on quality control that is still under construction.
5.1. Relevance - User Needs
The main users of the Swiss Earnings Structure Survey (SESS) are government bodies, trade unions and employers’ associations, international organisations, courts, researchers, the mass media and the public.
The main needs of users in Switzerland consist in knowledge about earnings according to economic activity, region, skill level required by the job, gender, age, citizenship, level of education, ISCO, managerial position, length of stay in the enterprise, collective pay agreement and marital status.
Usually users want to know the medians, sometimes also the quartiles and the arithmetic means. In addition, they are often interested in the distribution of wage classes, the proportions of wage components, and finally the comparability of earnings in specific settings of variables across the years.
5.2. Relevance - User Satisfaction
No specific survey about the satisfaction of users has been conducted. Based on some direct feedback we have obtained, we assume that their needs are largely being met.
Users’ unanswered demands mainly concern earnings according to a very detailed level of disaggregation (either concerning economic activity or regionalisation), often coupled with a combination of too many variables. From the public there is sometimes a demand for earnings in specific professions, which cannot be satisfied due to our data protection policy.
Users generally made no statements concerning lacking or redundant variables. In a few cases, it was regretted that information about the total length of professional experience is lacking.
We used the common tools to measure the accuracy, i.e. the standard deviation and the coefficient of variation.
6.2. Sampling error
See point 6.2.1 Sampling error - indicators
6.2.1. Sampling error - indicators
The measures of precision is shown in the attached document. Sampling errors were calculated in accordance with the sampling design – taking into account both the first and second stage. All monetary estimators are expressed in the national currency, i.e. Swiss francs (CHF). Monthly earnings relate to the variable B42, hourly earnings to B43.
Non-sampling errors are divided into coverage errors, measurements and processing errors, non-response errors and model assumption errors. They are described below.
6.3.1. Coverage error
Over-coverage:
No over-coverage exists.
Under-coverage:
Economic Activity:
The Datasets EARNINGS_SES_A_A4 and EARNINGS_SES_B_A4 delivered to Eurostat cover NACE Rev.2 sections B to S, including O, but without the public administrations.
Size of enterprises:
No under-coverage regarding the regulation. Sampling covers enterprises with three or more members of staff, according to the Business and Enterprise Register (“staff”, because the Business and Enterprise Register actually only provide information on the number of persons working in the enterprise, regardless whether they are owners or employees). Answering enterprises with fewer than three employees in the reference month were kept in the data set. With the goal of covering employees throughout the whole economy (NACE Rev.2 sections B to S), there is an estimated under-coverage of around 1-2% due to the fact that the smallest enterprises with fewer than three staff are not sampled (estimation source: Business and Enterprise Register)
Specific categories of employees:
Regarding the COMMISSION REGULATION (EC) No 1738/2005, there is under-coverage for the following categories of employees, which in 2014 were excluded from the datasets:
apprentices
trainees
Apprentices account for about 5.1% (2013) of under-coverage in relation to the reference population (Source: https://www.bfs.admin.ch/bfs/fr/home/statistiques/education-science/indicateurs-formation/systeme-formation-suisse/degre-formation/degre-secondaire-ii/places-apprentissage.html, accessed April 26, 2017).
For trainees, no reliable information is available to quantify the resulting under-coverage.
There is concern about possible under-coverage of top-earning managers. Indeed, it was noticed that their earning data often figure in separate payrolls with restricted access; thus, it is possible that such categories may sometimes not have been delivered.
6.3.1.1. Over-coverage - rate
0%
6.3.1.2. Common units - proportion
0%
6.3.2. Measurement error
The Swiss Earnings Structure Survey starting in 1994, with respect to some variables, did not fully correspond to COMMISSION REGULATION (EC) No 1738/2005. With the revision in 2012, the Swiss variables match better with the Eurostat variables. However, some remarks should be made:
Variable
Measurement Errors
Remarks
A13
In deviation of COMMISSION REGULATION (EC) No 1738/2005, data mostly could not be delivered according to local units. For this reason, this kind of economic activity normally refers to that of the majority in the enterprise.
A15
Switzerland actually asks for only 4 categories instead of the 7 categories of A15
Codes A, C and E are not on the dataset, because not surveyed separately. Code B comprises also the Categories A and C, while E is subsumed under D.
B31
As yearly earnings are not surveyed, but computed (for a full year), B31 is not surveyed either. Hence, to be consistent with B41, B31 was hardcoded 52.14
B321
The number of overtime-hours are not surveyed but computed on the basis of overtime-pay – on the assumption that a supplement of 25% is normally paid for overtime-hours.
B32
The precision of this variable is theoretically affected by some imprecision of B321.
As the share of B321 in B32 is only 0.3%, the effect of the inaccuracy of B321 on B32 can be ignored.
B33
In case of an unrealistic number, we imputed 24 days of holidays leave (Swiss average)
NACE Rev.2 section P was treated in the same way as the other sections, with the assumption that in fact teachers’ effective holiday entitlement does not exceed that of the general public: a great amount of the time off school is dedicated to the preparation of lessons.
B41
This variable is computed, not surveyed, according the formula B41 = B42*12
B41 does not correspond to the reality where there was a change in the occupational degree; or for part-timers occupied on an hourly compensation basis if their working time in the reference month deviates from their monthly average. It can be assumed that such deviations level out.
6.3.3. Non response error
See below.
6.3.3.1. Unit non-response - rate
The gross unit response rate for 2014 is not yet available. For the SES 2012, 82% of enterprises responded.
6.3.3.2. Item non-response - rate
The variable B23 was partly missing, therefore some imputation has been made (see point 6.3.4.1).
6.3.4. Processing error
Data scanning is also a potential risk regarding the correct reproduction of handwriting.
One possible source of potential processing errors might be the manual capturing of the data from questionnaires (exceptionally applicable). As regards data delivery by electronic means, an Excel – xml sheet with numerous incorporated validation checks was designed and put on the internet for download. The barriers were constraining, but some exception could exceptionally pass the rules, which could lead to some processing error risk.
6.3.4.1. Imputation - rate
The only imputation was a correction of the variable B23 (occupation in the reference month). Enterprises filled in the form with a free text for this variable, and then an automatic process completes this text with the national occupation nomenclature (which is linked to ISCO). When it was not possible to find a correspondence with the national nomenclature, the observations were excluded.
Some observations could only be coded at the level ISCO-1, so those observations (13% of all observations) were imputed using the following method in each ISCO-1 category:
Calculate the percentage of observations in each subcategory of ISCO-1
Random distribution of the observations in each subcategory according to the previous calculated percentage
6.3.5. Model assumption error
Concerning correction for the smallest enterprise size class (1 or 2 employees – not sampled) no model was applied, because
the information about size class smaller than 10 is optional
it would be difficult to establish a reliable model
until now, applying models for correction is not common usage in SESS
the effect of deficiency on earnings for the total of size classes and economic activities is estimated to have only a small impact
6.4. Seasonal adjustment
Not applicable
6.5. Data revision - policy
The Swiss SES data could be revised only in exceptional case.
6.6. Data revision - practice
No revision occurs for the SES2014.
6.6.1. Data revision - average size
Not applicable
7.1. Timeliness
The reference periods considered for the Swiss Earnings Structure Survey 2014 were:
October 2014 for monthly earnings
The whole year 2014 for irregular payments (B411)
Yearly earnings (B41) were computed and relate to the year 2014.
The national results were published in two steps:
30th November 2015: Private sector
12th April 2016: Whole economy, i.e., private sector and all public sectors (federal, cantonal and communal level).
7.1.1. Time lag - first result
30th November 2015
7.1.2. Time lag - final result
12th April 2016
7.2. Punctuality
The harmonised Datasets EARNINGS_SES_A_A4 and EARNINGS_SES_B_A4 were delivered to Eurostat in time, i.e. in June 2016. Only a slight revision was made to the delivered data due to the imputation of observations with only ISCO 1 digit (see point 6.3.4.1).
7.2.1. Punctuality - delivery and publication
The milestones for producing the Swiss Earnings Structure Survey 2014 were:
Time
Action
September 2014
Sampling process
November 2014
Dispatching the questionnaire
March 2015
Deadline for response
April 2015
First reminder to enterprises having not responded
Mai 2015
Second reminder
August 2015
End of data recording, plausibility checks
September 2015
Data cleaned
October 2015
Weighting process for private sector
November 2015
Data analyses and production of tables
November 2015
Dissemination of first results (“Swiss variables”) by media release: media conference and online publication on private sector results
April 2016
End of weighting process for the public sector
April 2016
Public and public+private sector online publication; Swiss salary calculator “Salarium”
June 2016
End of computing Datasets EARNINGS_SES_A_A4 and EARN-INGS_SES_B_A4; validation and transmission to Eurostat
April 2017
Transmission of the quality report to Eurostat
8.1. Comparability - geographical
The Swiss datasets EARNINGS_SES_A_A4 and EARNINGS_SES_B_A4 delivered to Eurostat are mainly in line with the relevant regulations.
Deviations were discussed in chapter 5.3 (Non-sampling errors).
The most important differences are the following ones:
Absence of apprentices, trainees and home workers
NACE Rev.2 of enterprise instead of local unit
Annual Earnings (B41) are not the actual annual earnings, but the computed annual earning for a full year (consequently, B31 is uniformly hardcoded 52.14).
8.1.1. Asymmetry for mirror flow statistics - coefficient
Not applicable
8.2. Comparability - over time
As we revised of the methodology, in part to meet the requirements of Eurostat, we cannot guarantee full comparability. However, we did the maximum to guarantee the main indicators (median for Switzerland and for each NACE 2, etc.).
8.2.1. Length of comparable time series
For the Eurostat data, data are comparable from 2010 (except that the universe is not exactly the same between 2010 and 2014).
For the national publication, the comparability is guaranteed from 1994 for the private sector. The public sector was added progressively (a key is needed to analyse the principal activity).
8.3. Coherence - cross domain
The comparison with the LFS, is not possible because of differences in the universe of each of these statistics.
8.4. Coherence - sub annual and annual statistics
No information
8.5. Coherence - National Accounts
As regards the National accounts, the annual amount for D11 in 2014 for the whole economy (including the first sector) was CHF 320,605 million. National Accounts do not break down this figure according to economic sections, and consequently it is not possible to limit it to NACE Rev. 2, B to S.
8.6. Coherence - internal
To check the consistency of the data, the main tool is comparison with previous years, at first in general and then by activity (NACE).
9.1. Dissemination format - News release
A press release on gender pay gap was published in March 2017:
The results of the Swiss Earnings Structure Survey 2014 are published on the internet pages of the Swiss Statistical Federal Office.
So far, only data relating to the “Swiss variables” have been made available. These results are not directly comparable to those obtainable with the harmonised Datasets EARNINGS_SES_A_A4 and EARNINGS_SES_B_A4.
Access to Swiss results based on the harmonized variables within the framework of Commission Regulation (EC) No 1738/2005 is currently only possible via the internet pages of Eurostat.
Salary calculator (“Salarium”): an interactive application based on a model allowing the calculation of the gross monthly earning and its dispersion (interquartile range) for an individual profile (only available in French, German and Italian)
9.3.1. Data tables - consultations
The last available reporting for the downloading of tables related to SES (national publication) is for the year 2014. The table below shows the number of tables downloaded in each month of 2014.
Month
Number of download
January
1199
February
1181
March
1105
April
1569
Mai
1813
June
1696
July
1703
August
1834
September
1851
October
2276
November
2238
December
1884
9.4. Dissemination format - microdata access
On demand, researchers or institutions can access our data. This service is charged.
No systematic feedback on their results is given to the reporting units – this is only available upon request.
9.6. Documentation on methodology
Explanations concerning the scope, the legal aspects, the processing and the methodology of the Swiss Earnings Structure Survey are accessible on the internet pages of the Swiss Federal Statistical Office.
Users are provided information about links to the data through press releases, newsletters, print publications, letters, e-mails, by phone and through personal contacts.
Metadata is implicitly contained in the tables, where definitions and codes are explained. Until recently, the questionnaire with its annex containing explanations for the Swiss Earnings Structure Survey 2014 was available on the internet. All important nomenclatures are accessible on
Methodological aspects that are comprehensible to the general public for the 2012 survey are summed up on pp. 22-26 of the latest print publication (only in French and German – available for free download): The Swiss Earnings Structure Survey 2012 – commented results, Neuchâtel, 2012.
Several methodological reports, addressed to the scientific reader, have been published in the past and are available for free download on the internet:
Andrade, Beatriz; Graf, Monique / BFS : Enquête sur la structure des salaires 2006 - Aspects métho-dologiques du modèle des salaires «Salarium». Neuchâtel, 14.7.2008
This methodological report concerning the salary calculator is available only in French
Graf, Monique ; Ferrez, Jacques / BFS: Enquête suisse sur la structure des salaires - Programmes R pour l'intervalle de confiance de la médiane. Rapport de méthodes. Neuchâtel, 2007
This methodological report concerning the calculation of the confidence interval based on the median is available only in French
Graf, Monique ; Matei, Alina / BFS: Enquête suisse sur la structure des salaires 2002 – La précision du salaire brut médian. Rapport de méthodes. Neuchâtel, 2005
This methodological report concerning the calculation of the confidence interval based on the median is available only in French
Graf, Monique / BFS : Enquête suisse sur la structure des salaires 2002 - Plan d'échantillonnage et extrapolation pour le secteur privé. Neuchâtel, 2004
This methodological report concerning sampling and extrapolation is available only in French
Graf, Monique / BFS : Enquête suisse sur la structure des salaires 2000 - Plan d'échantillonnage, pondération et méthode d'estimation pour le secteur privé. Neuchâtel, 2002
This methodological report concerning sampling and extrapolation is available only in French
The application of the FStatA is guaranteed by the fact that no publication for an aggregated number of fewer than 60 employees or 5 enterprises is possible. Among other things, this constraint is also an obligation for all recipients of our microdata, recorded in the data protection contract that every recipient has to sign before the microdata is delivered.
Not all points of this report can be filled due to the methodological revision in 2012 and the informatics revision in 2014. Those revisions was partly developed to match better with the Eurostat requirements.
The Swiss Earnings Structure Survey (SESS) has existed since 1994 and is carried out every 2 years. In 2012, the SESS was revised to ensure better matching with the Eurostat variables. However, the datasets delivered to Eurostat have fewer observations than the datasets used for the national publication due to missing values in some variables.
The Swiss Earnings Structure Survey (SESS) has existed since 1994 and is carried out every 2 years. In 2012, the SESS was revised to ensure better matching with the Eurostat variables. However, the datasets delivered to Eurostat have fewer observations than the datasets used for the national publication due to missing values in some variables.
Not Applicable
The survey is based on a sample extracted from the Swiss Business and Enterprise Register. All enterprises of three or more employees are included in the universe.
The enterprises are the statistical unit.
About 30,000 enterprises and 850,000 wages in the dataset sent to Eurostat (35,000 enterprises and 1.6 million employees for national publication).
The survey covers all active enterprises from the secondary and tertiary sector in Switzerland.
Not Applicable
We used the common tools to measure the accuracy, i.e. the standard deviation and the coefficient of variation.
Not Applicable
Once the data were definitively cleared, specialised research associates attributed the final weights for national publication. After further checking, the final database (“Swiss variables”) was set up.
Sampling methodology: A stratified two-stage random sample
The basis for sampling was the Swiss Business and Enterprise Register. The universe for sampling includes all enterprises active in August 2014 on the territory of Switzerland, ranging from NACE Rev.2 division 2. The minimum size of the enterprise is 3 employees.
However the survey took into account apprentices and trainees and these two categories of workers were not included within the Eurostat datasets (nor in the national publication), because they have not yet been checked. The wage data of subcontracted workers had to be delivered by their employment agencies.
The SESS (private sector and public enterprises) is based on a random sample at two levels: the first level concerns the stratified enterprises, while the second level concerns employees in the enterprises. Enterprises are stratified according to three criteria: enterprise size (3 classes), economic branch (39 aggregations of NACE Rev.2 divisions) and geographical entity (7 major regions, 8 cantons and 1 city).
In total, the SESS stratification includes more than 1600 cells (strata). While for some strata the selection of enterprises is exhaustive (this is always the case for an enterprise with more than 49 employees, but not exclusively), for the majority of strata enterprises are sampled by random selection.
The public administrations are not included in the datasets delivered to Eurostat for the SESS 2014, but are covered by the SES survey and published in the national publication. The reason why we decided not to include them in the delivery to Eurostat was the insufficient quality of those observations (missing values, complexity of the structure) and therefore the difficulty in weighting a too limited number of observations.
The number of salary records of a sampled enterprise to be delivered depends on the size of the responding unit. Enterprises with fewer than 20 employees have to provide all salaries. For the size class 20 to 49 employees one out of two is to be reported, while in the size class 49+, at least one out of three is expected.
Sampling of the employees has to be done by the responding units themselves, according to the principles of random selection.
Not Applicable
The reference periods considered for the Swiss Earnings Structure Survey 2014 were:
October 2014 for monthly earnings
The whole year 2014 for irregular payments (B411)
Yearly earnings (B41) were computed and relate to the year 2014.
The national results were published in two steps:
30th November 2015: Private sector
12th April 2016: Whole economy, i.e., private sector and all public sectors (federal, cantonal and communal level).
The Swiss datasets EARNINGS_SES_A_A4 and EARNINGS_SES_B_A4 delivered to Eurostat are mainly in line with the relevant regulations.
Deviations were discussed in chapter 5.3 (Non-sampling errors).
The most important differences are the following ones:
Absence of apprentices, trainees and home workers
NACE Rev.2 of enterprise instead of local unit
Annual Earnings (B41) are not the actual annual earnings, but the computed annual earning for a full year (consequently, B31 is uniformly hardcoded 52.14).
As we revised of the methodology, in part to meet the requirements of Eurostat, we cannot guarantee full comparability. However, we did the maximum to guarantee the main indicators (median for Switzerland and for each NACE 2, etc.).
