Statistics Norway

Division For income and wage statistics

Statistisk sentralbyrå
Postboks 1400 Rasta
NO-2225 Kongsvinger

This report covers all the main points related to quality that are normally covered and commented on in connection with the publication of statistics, and in this case statistics on earnings. The aim is to supply information on the quality of the data and statistics from Norway that are reported to and distributed by Eurostat in connection with the Structure of Earnings Survey 2010.

Since the national statistics on earnings are the same as those forwarded in connection with the survey mentioned above, this report should also be of interest to users of national statistics on earnings published on Statistics Norway’s website.

Not available.
New concept added with the migration to SIMS 2.0.
Information (content) will be available after the next collection.

Not available.
New concept added with the migration to SIMS 2.0.
Information (content) will be available after the next collection.

Not available.
New concept added with the migration to SIMS 2.0.
Information (content) will be available after the next collection.

Not available.
New concept added with the migration to SIMS 2.0.
Information (content) will be available after the next collection.

Not available.
New concept added with the migration to SIMS 2.0.
Information (content) will be available after the next collection.

Not available.
New concept added with the migration to SIMS 2.0.
Information (content) will be available after the next collection.

Not available.
New concept added with the migration to SIMS 2.0.
Information (content) will be available after the next collection.

Not available.
New concept added with the migration to SIMS 2.0.
Information (content) will be available after the next collection.

The statistics are constructed by compiling several separate sample surveys. All of these surveys are subject to possible errors due to the sampling methods applied, the quality of the reporting and the quality of the source used as population.

The populations for all surveys on earnings are based on the Norwegian Central Register of Enterprises and Establishments. All enterprises with local units that have employees in the reference period are included as the population. The sample in each survey is limited by use of cut-off, which varies between industries (see chapter 6.2.1 Sampling errors).

a) Probability sampling

All of the aforementioned surveys are based on stratified random sampling of clusters (sampling unit), where the clusters are defined as enterprises and their local units by section according to Nace Rev. 1. Education (section M) and Health and social work (section N) industries in particular include both private enterprises and public enterprises. The latter are included in full, not as a sample. Weights are calculated by use of post stratification procedures, calculating the inverse inclusion probability, the aim being to estimate how many employed persons there are in the population in the reference period. The main aim is of course by using weights to make it possible to estimate earnings of the population.

Stratification

The variables used for stratification are industry and number of employees. As previously stated, each of the industries presented below, as rows in the table, represent a separate survey in the national earnings statistics. The stratification is different for each industry when defining size groups of enterprises and sub-groups of industry. The reason for this is that the distribution of small and large enterprises is different between industries, and furthermore that what can be called large may vary a great deal from one industry to another.

These properties and arguments are especially important if the level and distribution of earnings are actually different from one stratum to another within the same industry (more information in chapter 6.3.5).

See also the attached document Sample design which presents figures to Population and Sample size.

Several new subgroups were established within some industries to improve coverage in subpopulations. This, besides the new NACE-standard, explains an increasing number of strata for those industries compared to the Quality report on the Norwegian Structure of Earnings Survey 2010 (see chapter 6.3.1. Coverage errors).

b) Non-probability sampling

Not applicable

[Not requested]

[Not requested]

[Not requested]

Probability sampling

Estimation of weights

Below, all comments on population refer to the population as it is found and defined by use of  Norway’s Central Register of Establishments and Enterprises, for the relevant industries.

Notation

The weights in the Norwegian Structure of Earnings Survey are defined as:

1)     where     and   is the sample of enterprises in stratum b. is the inverse inclusion probability defined as

2)    , for all   . Thus       and as such

gives an estimation of the number of enterprises in a strata b in the population.

The preferred ideal for the weights in the earnings statistics is to be able to express

Post-stratification procedures to establish  are initiated when  

This implies that we wish the final weights to give an estimate on the number of employees in the population. The final weights can therefore be described as:

3)  

where 

A further inspection of the weights can be done through a comparison with other sources. This is covered in chapter 8.3 but comments here are ment to give an example of how the weights rearrange the distribution of the non-weighted figures.

Relative distribution of employees in the SES weighted and non-weighted by industry, comparison with National Accounts and LFS. 2010

It is clear that the final weighted distribution of the SES has a closer resemblance to the distributions found in the National Account (NA) and the Labour Force Survey (LFS). This does not prove the accuracy of the weights as far as estimation of earnings is concerned, but the improvement of the distribution caused by the weights is a claim that they at least make for a better understanding of the actual composition of the population.

[Not requested]

Not available.
New concept added with the migration to SIMS 2.0.
Information (content) will be available after the next collection.

[Not requested]

The purpose of the statistics is to provide an overview of levels and changes in earnings for all employees (wage and salary earners) independent of industry or working hours, and in accordance with user needs. Statistics are provided for each industry separately, broken down by sex, occupational group, age, and educational level in order to meet the demands of public and private users.

Major users outside Statistics Norway are the Technical Reporting Committee on the Income Settlement, research and policy institutes, employee and employer organisations, Eurostat, ILO, OECD, the media, enterprises, and private persons. The statistics are also used in Statistics Norway's Labour Accounts and in quarterly wage indices.

Based on the extensive use and feedback concerning Statistics Norway's earnings statistics, it is generally perceived that the statistics meet most user needs. Expressed needs for more statistics from the source are always an integral part of planning annual work programmes within the field.

[Not requested]

[Not requested]

-

[Not requested]

a) Probability sampling

Bias

The statistics on earnings are, as with all other sample based statistics, subject to bias, which arises when the distribution on some variables in different parts of the sample is not the same as the corresponding distribution in the population. Dividing the population into groups (strata) according to certain stratification variables reduces the possibility of imbalances in the sample. Partial non-response in several of the items collected by form and used in the wage statistics can normally be logically calculated on the basis of other information given on the form or imputed from earlier years.

Post-stratification adjusts any imbalances arising in the distribution between the stratification variables due to non-response. The weights are additionally adjusted for any imbalances due to non-response.

Non-response that is not randomly distributed may bias the separate samples for the different sections, and this may have some influence on these statistics. Non-response in the wage statistics is between 1 and 9 per cent. Possible sample bias in the individual statistics will be of less importance for these statistics due to the considerable quantity of data it is based on (table in 6.3.3).

The use of cut-off may be a source of bias. In most industries, the sample consists of sampling units with five or more employees.

b) Non-probability sampling

Not applicable

a) Probability sampling

Variance

Variance of interest in this case is variance that arises due to the size and composition of the sample, more specifically the sampling model, so-called sample variance. Statistics on earnings make use of random sampling of clusters (enterprises by industry), however the samples are large and this therefore results in relatively low variance. (See also the attached document Variance tables). The coefficient of variance varies theoretically between 0 and 1, and is in some cases used as a percentage. A low value presents the argument that very little of the variance derives from the sample. (Appendix A Coefficient of variance)

Coefficient of variance for monthly earnings of full-time employees in the SES, by industry

Statistics Norway have some challenges in the wage statistics with some non-sampling errors

The population consists of all enterprises in Statistics Norway's Central Register of Establishments and Enterprises, with the exception of small enterprises with fewer than three, four or five employees according to industry. Each enterprise covers one or more local units grouped by industrial category. The sample in each section consists of enterprises drawn from the population, dependent on activity code and the number of employees. The wage statistics data are obtained for each person employed in the local units in the reference period covered in the industrial sectors according to Nace.

Errors in the stratification variables, activity (Nace ) and number of employees could be a source of errors. Additionally, actual differences between the population and the sample may lead to problems such as over-coverage or under-coverage in sub-populations.

In order to deal with this potential problem, the local units in the sample are asked to control the preprinted code of activity on the form. If this code is believed to be incorrect, the local units are asked to describe their activities in order to correct this code. In each specific case, this information is assessed in order to come up with a correct classification of the unit.

In the wage statistics, some under-coverage may be expected due to a time lag in the registration of new units in the Central Register of Establishments and Enterprises. Over-coverage may also be present for the same reason, i.e. the time lag in the registration process when enterprises no longer have employees because the business has been closed, sold or taken over by new owners, has gone bankrupt or has been merged in the time period between the selection of the sample and the time of the census. As long as these errors are fairly constant, the effect on the statistics is minimal.

None

[Not requested]

Measurement errors are defined as a discrepancy between the value of a variable reported by the respondent and the "true" value. Such errors mainly arise because the respondent lacks the information or finds it difficult to calculate the value. This may be due to the following:

• In his daily work, the respondent uses other unit definitions than those used as a basis for the statistics for example other payment periods
• The respondent does not have the information that is requested
• The respondent himself has incorrect information
• The respondent misunderstands or fails to read the instructions. The respondent may misinterpret the content of the variables, or is imprecise in checking off on the form that will be read optically

However, the increasing use of the electronic standard for reporting statistics has reduced the amount of measurement errors in reporting. This standard basically retrieves wage data directly from the enterprises’ wage and personnel systems, thus eliminating several possible sources of error that arise when using traditional forms. On the other hand, new problems arise when making use of new methods of collection and processing. In general however, these problems have been more easily identified and corrected when making use of electronic solutions in data collection and processing.

Measurement errors are identified and corrected both by logical, automated computer controls, as well as manual checks of extreme outliers and conspicuous changes in wage levels, and number of employees in the enterprises, compared to the previous years.

Distribution of collected observations by source. Per cent by industry, 2010

Unit non-response

Unit non-response refers to the fact that the respondent, in this case an enterprise, has not completed and returned the statistics questionnaire. In the statistics, the unit non-response is between 0.1 and 13.3 per cent (table below). The main reasons for non-response are that units have ceased to exist, been sold or transferred to a new owner, gone bankrupt or have been merged. Furthermore, there is also a small group reporting too late to be included in the statistics, or providing data of a quality that cannot be used for statistical purposes. In the case of unit non-response, the weights of the units on which the statistics are based are adjusted to compensate for the non-response.

Imputation is only done for the variable occupation.

Response rate by industry

Partial non-response

In the case of methods making use of clusters as sampling units, it is necessary to make distinctions between two types of partial non-response. The first and most typical type of non-response for a sample survey is that the sample unit, enterprise in this case, has not reported all employees. The second major type of non-response would be the traditional type, where elements of information regarding the unit of analysis are missing. Some of the items can often be calculated on the basis of other information and possibly imputed from previous years.

[Not requested]

[Not requested]

Processing errors are errors that can arise during the course of computer processing of the reported data from the respondent and up to the point the statistics are completed. This applies to factors such as data transmission, registration, encoding, and error correction. Reported forms are registered either optically or manually, while electronically reported data are either downloaded to the database through the internet portal for public reporting (Altinn), or entered directly into the tables where data information is compiled.

Imputation rate of occupation, by industry.

Statistics Norway has chosen to use September and October as the reference months for the annual wage statistics. These months are believed to be less affected by holidays and the most stable regarding wages and therefore also considered representative.

The accounting and fiscal year is identical to the calendar year in Norway. Hence, this is not subject to any errors regarding the wage statistics.

The sample model used for all sections is based on stratified samples. Dividing the population into groups (strata) according to certain stratification variables reduces the possibility of imbalances in the sample and assures a better coverage of certain units or groups of units in the wage statistics.

The sample consists of enterprises drawn from the population. The population is basically all active enterprises in the section, with the exception of small enterprises with fewer than five/three employees, which are not included in the sample. Large enterprises (sample units), where the definition of large varies between industries, receive a sampling probability of 1. Strata that cover small and medium-sized sample units are given a lower sampling probability.

The stratification is made according to industry and the size (number of employees) of the enterprises, on the assumption that wages and composition of occupations in large enterprises differ from those in small ones, and that there are differences according to industry. In each stratum, this sample model ensures a minimal dispersion in the main variables measured, i.e. wage.

The number of employees is an important feature with regard to the stratification. Some assessment of this size is made through to the sampling process and serves as guidance for ongoing improvement. In each stratum, the mean number of employees is calculated along with the standard deviation. This is done to ensure an optimum stratification that reflects the differences between the strata.

The different products in the wage statistics are separated according to section. Each section in the wage statistics represents one part of the total population, and these are therefore also to be considered as a part of the stratification. This stratification ensures that each section is fully covered, and that no major enterprises are left out. Also the coverage of other related sections is ensured through this model.

By collecting individual employees with person identification numbers it is possible to add information from administrative sources. The quality of these identification keys is stressed both in the questionnaires and in the administrative data and reduces the relative level of errors to a minimum.

The purpose of the sample model selection process is basically to get samples that ensure a representative basis for the wage statistics and avoid burdening all enterprises in the industry with forms to fill in. This limits the size of the samples while focusing on main variables. Another objective is to ensure that the burden of reporting obligations is minimized as much as possible for the smallest enterprises. Statistics Norway likes to believe that all these purposes are well fulfilled. Additionally, the effects of any known and unknown model errors are reduced to an acceptable minimum through the use of this model.

[Not requested]

[Not requested]

[Not requested]

[Not requested]

The wage statistics for all employees 2010 were published on March 31, 2011. These statistics are produced by using the wage statistics for several industrial sections with the following publishing dates:

• Section A. Employees in Fish Farming. Published February 17, 2011.
• Section B. Employees in Oil and Gas Extraction and Mining. Published December 17, 2010.
• Section C. Employees in Manufacturing. Published January 27, 2011.
• Section D. Employees in Electricity Supply. Published March 17, 2011.
• Section E. Wage statistics. Employees in Water Supply, Sewerage, Waste Management and Remediation activities. Published March 19, 2011.
• Section F. Employees in Construction. Published January 13, 2011.
• Section G. Employees in Wholesale and Retail Trade. Published December 15, 2010.
• Section H. Employees in Transport and Communication. Published February 11, 2011.
• Section I. Employees in Hotels and Restaurants. Published March 3, 2011.
• Section J. Employees in Information and Communication. Published January 19, 2011.
• Section K. Employees in Financial Intermediation. Published November 18, 2010.
• Section L. Employees in Real Estate. Published November 25, 2010.
• Section M. Employees in Professional, Scientific and Technical activities. Published December 2, 2010.
• Section N. Employees in Business activities. Published November 26, 2010.
• Section O. Central Government employees. Published February 3, 2011.
• Section P. Employees in Private Education. Published March 4, 2011.
• Section Q. Employees in Private Health and Social Work activities. Published February 25, 2011.
• Section R. Employees in Arts, Entertainment and Recreation. Published March 11, 2011.
• Section S. Employees in Social and Personal Service activities. Published March 3, 2011.
• All employees. Published March 31, 2011.

[Not requested]

[Not requested]

The reference period for the surveys is September 1 for sections G, K and N and October 1 for the remainder. The statistics are collected by way of the mandate given through "The Statistics Act of 1989",which, for the statistics presented here, makes response mandatory.

Key dates in the data collection process:

September 1:

• Questionnaires sent: August 20, 2010
• Date for delivery: September 8, 2010
• Reminder: September 20, 2010
• Final date for delivery: October 8, 2010

October 1:

• Questionnaires sent: September 17, 2010
• Date for delivery: October 12, 2010
• Reminder: October 19, 2010
• Final date for delivery: October 29, 2010

In addition to the forms, large enterprises are phoned during this period to ensure that the questionnaires are returned. The post-collection phase begins as soon as questionnaires are received. Working deadlines are set as the process takes place, and in accordance with priorities given by the pre-planned list for publication. There are no given and explicit deadlines for the different elements in the post-collection phase, except for the final deadline; the statistics for the different sections are finished and approved one week prior to the publishing dates.

[Not requested]

The Norwegian earnings statistics are collected annually and comply with most mandatory points drawn up in the council regulation 530/1999, and subsidiary commission regulations. Specific exceptions are stated in the EEA agreement Annex XXI - p.25. Some special features for Norway do however apply:

1. The variable "4.2.2 Special payments for shift work" will include payments for shift work and other irregular payments.
2. The variable "3.3 Total annual days of holidays leave" will only include number of days of holiday and not absence due to sickness or absence for vocational training.

The statistical units are identical to the units used in other countries; the reference population is basically all active enterprises in the section, with the exception of small enterprises with less than 3, 4 or 5 employees (depending on section), which are not included in the reference population.

Statistics Norway also uses international standards with regard to classifications of different variables. Some national adaptations are made, but these are not present in the transferred SES files.

[Not requested]

Comparable annual statistics on earnings were established for most industrial sections in 1997, a few sections were included later. The statistics are comparable from 1997 and are uniform and comparable among the sections. There has not been any change in the definitions of variables since 1997. The applied methods and models have however been subject to ongoing improvement based on increased knowledge and new requirements since they were established. These ongoing improvements have not affected comparability.

[Not requested]

Coherence with the Labour Force Survey (LFS) 3rd quarter 2010

The following is a short presentation and comparison of the Norwegian SES and the Norwegian LFS surveys. It is important to point out basic differences that possibly could be the cause of differences between the surveys as they are observed in the following tables. Statistics from the LFS are based on published figures.

Comparison of basic information on model assumption, sampling, units and purpose

In the following three short chapters, several basic aspects of the LFS and SES are compared. One of the main reasons for different surveys is to meet different needs. Consequently, the statistics are based on assumptions that meet these specific user needs. The LFS survey monitors and documents quarterly changes in the composition and distribution of the work force. It is based on a sample survey covering individuals (the sample unit is family) that report on their status in the work force.

The earnings statistics on the other hand are structured to answer questions concerning the level and distribution of earnings. As described earlier, the source is a sample of enterprises that reports on employees. There is significant overlap between the populations of the two surveys, but the source of information is different and so are the sampling models. Furthermore, the two surveys have different reference periods and utilize different sources for control, verification and finally dissemination.

Both statistics are nonetheless used for explaining different properties of the same field of interest and in this capacity we can use the LFS to understand the distribution and composition of jobs and employees as they are described in the earnings survey. Discrepancies should, where they occur, be explained and understood as a consequence of overlapping information.

Tabular results and comparisons with the LFS

See the attached document Coherence with LFS

For the tables that refer to distributions of full-time and part-time employees respectively by age, discrepancies are small. Most of the differences between the two sources might very well be a result, at least to some extent, explained by the differences described in the previous chapters. Differences in the definitions of full-time employees in particular may contribute to some of the observed discrepancies even though these should be viewed as small to minimal in this case.

The same factors mentioned above will also explain discrepancies between the tables that show the distribution of full-time employees by industry.

In general it seems that the distribution of employees by sex and industry and sex and age are very similar. This also gives more credit to the assumptions presented in connection with chapter 6, especially concerning the sampling model and hence model assumptions and bias.

Coherence with National Accounts

In the two tables in the attached document Coherence with National Accounts, comparisons between National Accounts and the SES are shown. The first table gives the distribution of wages by industry as the estimated sum of annual wages from the SES and compensation of employees in the National Accounts. The other table compares the distribution of employees by industry. Discrepancies can mostly be explained through differences in definitions, reference periods between the two sources and revision of the NA data. All statistics from the National Accounts are based on published figures for 3rd quarter 2010.

[Not requested]

[Not requested]

[Not requested]

[Not requested]

References to the statistics:

• All employees: http://www.ssb.no/english/subjects/06/05/lonnansatt_en/arkiv
• Section A: http://www.ssb.no/english/subjects/06/05/lonnfisko_en/arkiv
• Section B: http://www.ssb.no/english/subjects/06/05/lonnolje_en/arkiv
• Section C: http://www.ssb.no/english/subjects/06/05/lonnind_en/arkiv/
• Section D: http://www.ssb.no/english/subjects/06/05/lonnkraft_en/arkiv/
• Section E: http://www.ssb.no/english/subjects/06/05/lonnvar_en/arkiv/
• Section F: http://www.ssb.no/english/subjects/06/05/lonnbygganl_en/arkiv/
• Section G: http://www.ssb.no/english/subjects/06/05/lonnvare_en/arkiv/
• Section H: http://www.ssb.no/english/subjects/06/05/lonnsamf_en/arkiv/
• Section I: http://www.ssb.no/english/subjects/06/05/lonnhotell_en/arkiv/
• Section J: http://www.ssb.no/english/subjects/06/05/lonnikt_en/arkiv/
• Section K: http://www.ssb.no/english/subjects/06/05/lonnfinans_en/arkiv/
• Section L: http://www.ssb.no/english/subjects/06/05/lonneiendom_en/arkiv
• Section M: http://www.ssb.no/english/subjects/06/05/lonnfagvit_en/arkiv
• Section N: http://www.ssb.no/english/subjects/06/05/lonnef_en/arkiv
• Section O: http://www.ssb.no/english/subjects/06/05/lonnstat_en/arkiv
• Section P: http://www.ssb.no/english/subjects/06/05/lonnprivund_en/arkiv/
• Section Q: http://www.ssb.no/english/subjects/06/05/lonnhelse_en/arkiv/
• Section R: http://www.ssb.no/english/subjects/06/05/lonnkult_en/arkiv
• Section S: http://www.ssb.no/english/subjects/06/05/lonnsos_en/arkiv/

The statistics are published on the Internet

[Not requested]

[Not requested]

No results are sent to the reporting units.

The same Internet addresses as mentioned in 9.2 apply for references to methodical documents; these documents can be found using the link "About the statistics" in the left margin.

References to the statistics:

• All employees: http://www.ssb.no/english/subjects/06/05/lonnansatt_en/arkiv
• Section A: http://www.ssb.no/english/subjects/06/05/lonnfisko_en/arkiv
• Section B: http://www.ssb.no/english/subjects/06/05/lonnolje_en/arkiv
• Section C: http://www.ssb.no/english/subjects/06/05/lonnind_en/arkiv/
• Section D: http://www.ssb.no/english/subjects/06/05/lonnkraft_en/arkiv/
• Section E: http://www.ssb.no/english/subjects/06/05/lonnvar_en/arkiv/
• Section F: http://www.ssb.no/english/subjects/06/05/lonnbygganl_en/arkiv/
• Section G: http://www.ssb.no/english/subjects/06/05/lonnvare_en/arkiv/
• Section H: http://www.ssb.no/english/subjects/06/05/lonnsamf_en/arkiv/
• Section I: http://www.ssb.no/english/subjects/06/05/lonnhotell_en/arkiv/
• Section J: http://www.ssb.no/english/subjects/06/05/lonnikt_en/arkiv/
• Section K: http://www.ssb.no/english/subjects/06/05/lonnfinans_en/arkiv/
• Section L: http://www.ssb.no/english/subjects/06/05/lonneiendom_en/arkiv
• Section M: http://www.ssb.no/english/subjects/06/05/lonnfagvit_en/arkiv
• Section N: http://www.ssb.no/english/subjects/06/05/lonnef_en/arkiv
• Section O: http://www.ssb.no/english/subjects/06/05/lonnstat_en/arkiv
• Section P: http://www.ssb.no/english/subjects/06/05/lonnprivund_en/arkiv/
• Section Q: http://www.ssb.no/english/subjects/06/05/lonnhelse_en/arkiv/
• Section R: http://www.ssb.no/english/subjects/06/05/lonnkult_en/arkiv
• Section S: http://www.ssb.no/english/subjects/06/05/lonnsos_en/arkiv/

[Not requested]

[Not requested]

[Not requested]

[Not requested]

[Not requested]

[Not requested]

References

• Council regulation (EC) 530/1999, of 9 March 1999: Concerning structural statistics on earnings and labour costs
• Commission regulation (EC) 72/2006, of 16 January 2006: on implementing council regulation (EC) 530/1999 as regards quality evaluation of structural statistics on earnings.
• Commission regulation (EC) 1738/2005, of 21 October 2005: on implementing council regulation (EC) 530/1999 as regards quality evaluation of structural statistics on labour costs and earnings.
• Grini, Knut Håkon: Notater 74/2003 Lønnsstatistikk 1997-2006. Dokumentasjon av utvalg og beregning av vekter. Statistics Norway
• NOS D 362: Lønnsstatistikk 2005, Statistics Norway
• Särndal, Swensson, Wretman: Model Assisted Survey Sampling, Springer. Corrected fourth printing, 1997.