Statistics Sweden

Unit for Innovation, Business sector production and Research

Statistics Sweden

Solna strandväg 86

SOLNA

Index of Service Production

Main purpose is to be able to calculate the production in the service sector,  both in total and by activity.

NACE Rev. 2.

NACE Rev. 2 Divisions 36-39, 45-63 and 68-96.

The frame is constructed using a cutoff where units over the cutoff together include 90-95 % (depending on industry sector) of the yearly turnover.

Service production calculated based on data collection of total turnover including value added tax and excluding exports (NACE Rev. 2 Divisions 45, 55, 56, 59, 60, 90, 91, 92, 93, 95, 96), total turnover excluding value added tax but including exports (the remaining activities in the service sector covered by the Index of Service Production except for NACE group 68.2, see concept 3.3). Production in NACE group 68.2 is calculated using data on produced quantities.

Planned changes in information collected: No planned changes.

Accounting conventions: The reference period is the calendar month.

Reporting unit: The reporting unit is kind of activity unit 

Observation unit (s): The observation unit is the activity unit

There are 750 000 kind of activity units in the reference population.

All regions of Sweden are covered.

From 2000.

Reference year is 2015.

The final result is an index with reference year equal to 100.

Reference period is the calendar month.

The Index of Service Production is governed internationally by the Regulation (EU) 2019/2152 of the European Parliament and of the Council of 27 November 2019 on European business statistics and the EBS general implementing act (EU) 2020/1197, and nationally by the Official Statistics Act of 2001 (SFS 2001:99) and Ordinance SFS 2001:100. The legal basis under which data is collected is specified in SFS 2015:4.

The units are obliged to provide data.

Data are not sent to other international organisations.

Statistics Sweden have policies in place in order to minimize the risk of disclosure and/or damage dealt to respondents in the survey.

The main rule is that groups with less than three enterprises are always treated as confidential. However that criteria may be weak in certain situations. Therefore Statistics Sweden have developed a program that estimates the probability of disclosure in each of the reporting groups. Based on the estimated probability of disclosure the potential damage of such an exposure is assessed by the statisticians involved in the production of data.

If data is under risk of disclosure, the first measure is cell suppression. If the risk of disclosure continues to recur in a certain cell, the second measure is aggregation of two or more similar cells.

Data are only published down to an activity level at which it is not confidential. Confidential data are transmitted to Eurostat but marked as "confidential data".

Dates for the monthly statistical releases are determined one year in advance. The dates can be found on Statistics Sweden's website, www.scb.se.

Anyone can access the release calendar on Statistics Sweden's website, Statistics Sweden's release calendar.

Data for Services production statistics is not simultaneously released. National Accounts in Sweden receive the data first, then Eurostat and other users (official release).

It is only the statistical office that has access to the results before they are released.

Data is transmitted to Eurostat each month. Data is in the SDMX format and transmitted through Edamis.

Monthly

No news release.

Data for selected activities are available in Sweden’s statistical databases and on the Statistics Sweden’s website.

The data is not disseminated in any publication.

Data for selected activities are available in Sweden’s statistical databases and on the Statistics Sweden’s website.

33 activities or aggregates of activities are published in Sweden’s statistical databases. Presentation as current prices, constant prices, constant prices working day/seasonally adjusted and trend depending on activity. Time series are available from Janyary 2000.

Micro-data making it possible to identify individual objects are not publicly released. Statstistics Sweden performs on request special processing of primary materials from previous surveys.

Researchers may apply to Statistics Sweden to get access to de-identified micro data for own processing purposes.

Data are transmitted to Eurostat to be published and to be used in the European aggregates.

Final results are sent to Eurostat to be used in the compilation of european aggregates and to be released as national data.

Final results are also sent to the National Accounts Department within Statistics Sweden to be used in the compilation of quarterly national accounts

Documentation about the survey is available in the form of Kvalitetsdeklaration (Quality declaration), Beskrivning av Statistiken - BaS (Description of the statistics), and SCBDOK (Documentation of statistics). Information on the final observation registries are stored in Statistics Sweden's database Meta Plus. All documentation is available in swedish on Statistics Sweden's website.

Criteria for quality assessment in swedish official statistics are found in in the documentation Kvalitet för den officiella statistiken – en handbok (A Handbook on Quality for Official Statistics of Sweden), available at at Kvalitet för den officiella statistiken – en handbok, version 2:2 (scb.se)

In general, Statistics Sweden applies the European statistical Code of Practice.

Statistics Sweden uses the EFQM Excellence Model from the European Foundation for Quality Management as a framwork for quality assurance.

Overall Statistics Sweden has a high level of quality. The Index is highly relevant and the level of accuracy is high. Comparability over time is somewhat affected by methodological changes to increase the accuracy. It is comparable with other statistical domains primarily the quarterly national accounts.

Primary users and their needs are:

• National Accounts Department within Statistics Sweden: As input data in the quarterly national accounts compilation.
• Ministry of Finance, governmental agencies and the national central bank: Input data into forecasts of the national accounts as well as analysis of the economic development.
• Eurostat: Input data into analysis of the economy as well as the compilation of indicators and other statistics.
• Banks and financial institutes: Input data in analysis and forecast of the economic development.

Secondary users and their needs are:

• Researchers and students: Input data in research and education.
• Social actors: Analysis of the economic development.
• Media: Publishing and release of data.
• Enterprises/businesses: Analytical uses.

Statistics Sweden has general satisfaction surveys but no specific survey for the short-term statistics.

Views and opinions of the users are collected through user councils with 2 to 4 meetings per year. The members of the council represents the users of the statistics which is funded by grants produced by Statistics Sweden. The council has an advisory role regarding questions concerning new statistics, development and improvement of existing statistics and priorities for the coming financial year.

Direct contact with users is also a source for evaluating user satisfaction.

The requirements of the European STS regulation are fulfilled.

There are a number of sources for inaccuracy in the services production index. This is an attempt to structure the possible sources for inaccuracy and to qualitatively describe the effects.

Coverage: The frame is constructed yearly using a "frozen" version of the business register, so enterprises entering into the business register after this freezing moment is considered undercoverage, and enterprises exiting the frame after this moment are considered overcoverage.

Sampling: The sample method is stratified random sampling. Stratification variables are NACE Rev.2 activity and size of yearly turnover. The sample consists of eight size classes where size class 1-4 is sampled, 5-6 and 8-9 is censused. Size classes 1-4 consists of the smallest units and 5-6 of the largest. Size class 8 consists of complex units consisting of two or more legal units. Size class 9 consists of various special cases. Size classes are created using the Dalenius-Hodges rule. Sample size in each strata is determined using Neyman-allocation.

Nonsresponse: Nonresponse is handled by imputation methods for both item and object nonresponse. The imputation method consists of a list of ranked alternatives where the first alternative is used if it meets a set of criteria’s. If the first alternative is not possible due to failure to meet the criteria’s the next alternative is used and so on. The process is automated using the BANFF-system in SAS.

Measurement: As always there may be measurement errors with sampling. The statistics is cross-checked with sources such as VAT-data, in order to identify measurement errors and correct them. Questionnaire data for units in size class 1-4 is replaced with VAT-data on a quarterly basis for all units in the frame which decreases measurement errors. However the VAT-data also contain measurement errors and may have periodicity problems. But since each unit only represents itself the measurement errors are limited compared to a "normal" estimation.

Data processing: Turnover is deflated using various price indices. Errors and inconsistencies in this data source may affect accuracy of the turnover index.

Model assumptions: The frame is constructed using a cutoff where units over the cutoff together include 90-95 % (depending on sector) of the yearly turnover. The turnover for the units under the cutoff is assumed to have the same development in turnover as the units over the cutoff.

Data is revised so data changes after the first release.

The sample method is stratified random sampling. Stratification variables are NACE Rev.2 activity and size of yearly turnover. The sample consists of seven size classes where size class 1-4 i sampled, and 5-6 and 8 is censused. Size class 1-4 consists of the smallest units and 5-6 of the largest units. Size class 8 consists of complex units consisting of two or more legal units. Size classes are created using the Dalenius-Hodges rule. Sample size in each strata is determined using Neyman-allocation. The two largest strata are censused and the four smallest one are sampled.

The turnover survey is a a probability sampling survey for monthly estimates but a census on a quarterly basis. On a quarterly basis, data for units that are sampled using probability sampling is replaced by VAT-data for the part of the frame i size class 1-4. Data for censused size classes are never replace with VAT-data. Monthly estimates bases on survey data only, is then replaced. The sampling error for the "census" results on a quarterly basis is zero since it is a census. The sampling error for the estimates based on survey data can be quantified.

The frame is constructed using a "frozen" version of the business register, so enterprises entering into the business register after this freezing moment is considered undercoverage, and enterprises exiting the frame after this moment are considered overcoverage.

The survey instrument is an electronic questionnaire. The questionnaire has been evaluated using cognitive interviews and expert evaluations in order to reduce the possibilities for non-sampling errors.

Nonresponses are handled by imputation methods. The imputation method consists of a list of ranked alternatives where the first alternative is used if it meets a set of criteria. If the first alternative is not possible due to failure to meet the criteria the next alternative is used and so on. The process is automated using the BANFF-system in SAS.

Turnover is deflated using various price indicies. Errors and inconsistencies in this data source may affect accuracy.

The frame is constructed using a cutoff where units over the cutoff together include 90-95 % (depending on industry sector) of the yearly turnover. The turnover for the units under the cutoff is assumed to have the same development in turnover as the units over the cutoff. Data is revised so data changes after the first estimation. All non-responses are handled by imputation so the imputation method affects the accuracy of the final estimates.

Non-sampling errors may arise due to different factors such as:

Frame coverage: Undercoverage consists of newly established enterprises that, at the time of drawing of a sample, does not yet exist in the business register. Analog to undercoverage, overcoverage consists of enterprises that are closed down but have not yet been removed.

Measurement: Although sampling design is done at the enterprise level, measurement is done at the Kind-of-activity level. Enterprises are asked to provide information on the parts active in the industrial sector. Enterprises may therefore mistakenly include information on total turnover instead of the turnover generated from services.

Survey-instrument: The Services Production Index is estimated from deflated deliveries. Deliveries are collected through a survey of enterprises. For the surveyed enterprises a large majority of the surveyed units report through an electronic questionnaire, although a small number still report on a paper questionnaire.

Mean weighted unit response rate for total services at first estimate of a reference period in 2022 was 83,7 percent. Turnover is used as weights.

Service production statistics are published approximately 35 – 40 days after the end of the reference period.

Data collection

Questionnaires are sent to the reporting units at the end of the month of the reference period asking for replies by the 15th of the following month.

All releases have been done on time.

The same statistical concepts are applied in the entire Swedish national territory. No geographical discrepancies exist.

Data are comparable with other EU countries thanks to the use of common definitions (Commission Regulation No 1503/2006).

Comparable data are available from 2001 and there are no breaks in series.

Comparability over time is mostly affected by changes in changes in NACE classification system. The transistion from NACE Rev.1.1 to NACE Rev.2 meant that other publishing industry and  the recycling industry are not included in the industry.

Comparisons can be made with the Swedish value added tax register. For most of the enterprises and for figures for each activity, a comparison can only be made afterwards because the service production statistics are available quicker than the value added tax register.

Ad-hoc confrontation with other statistical sources covering turnover (Value added tax register and Structural Business Statistics).

Data are internally coherent as higher aggregates are based on lower level data also for seasonally adjusted series

No internal incoherence exists in the Services Production Index.

Every year, appr. 67 500 respondents use in average 1/4 of an hour to answer the survey. Hence the respondents burden is approximately 17 000 hours.

In total Statistics Sweden estimates that 500 working hours per year is associated with the work at Statistics Sweden for the collection, estimation, calculation and presentation of the Services Production Index.

Routine revisions are made for two to four months prior to the current month depending on which month in the quarter the reference month is. The whole previous qaurter is revised in connection with the publishing of the first and second month in a quarter. Revisions for previous periods are released at the same time as the release of a new period.

The same revision policy is applied nationally and in transmissions to Eurostat.

The Index is a chain-index, so no revisions occur due to regular base year changes (i.e. the rebasing that is carried out every fifth year due to STS requirements).

Major revisions may occur due to methodological changes. Methodological changes are announced in advance.

Non-scheduled revisions, i.e. unexpected revisions may occur due to errors discovered in the input data after results are considered definitive, if deemed necessary due to the magnitude of the error. Non-schedule revisions may also occur between normal releases if errors of a greater magnitude are discovered shortly after release.

There is no public vintage database for services production.

No benchmarking is made.

Big and important revisions are announced. Smaller revisions and revisions for a single activity are not announced. Methodological changes are announced at the time of change. Sometimes the description is that the revisions have occurred but not a description of what the changes are.

Data in the questionnaire is collected for the last three periods which introduces revisions. Enterprises revising previously submitted data also introduces revisions.

Mean Absolute Revision (MAR) and Mean Revision (MR), computed values for the last 36 monthly (Jan 2020 - Dec 2022) for total services production:

Growth rates for calendar/working day adjusted data series (YoY): MAR = 1.02 , MR = -0.54

The source is a statistical survey. For final calculations data from the Swedish VAT registry is also used.

Frame on which the source is based is Swedish Business Register and Swedish value added tax register. All at Statistics Sweden.

It is a stratified sample survey. As a base, activity and employee size are used for stratification but also total turnover is considered. Enterprises are also classified into size strata if they belong to a consolidated business group (concern).

There are approximately 750 000 kind of activity units in the reference population.

The sample is updated once a year.

Data collection is made monthly and quarterly.

Collected through random sampling survey. Most enterprises use an on-line questionnaire. New enterprises in the survey are contacted through regular mail with information and login information. Each month, after the end of the reference period, login information is sent to the enterprises through regular mail reminding them of the survey. A reminder is sent out around 15 days after the first contact. Important and/or larger enterprises are reminded especially if they have not yet responded at the end of the collection phase. During the production phase enterprises are contacted in order to get data from important and/or larger enterprises. Contact is primarily through phone and e-mail. The survey has a sanction system with a penalty for not responding.

VAT data for final calculations (see section 13.1) is collected quarterly from the Swedish VAT registry.

The online questionnaire contains a number of controls to check for inconsistencies. The controls are soft and respondents are asked to correct or comment.

The data is then validated using a number of tools and checks:

• Value set error - Checks for small enterprises with very large levels of turnover. It is fairly common that an enterprise responds in SEK rather then thousand SEK creating very large figures.
• Accumulated turnover - Checks for enterprises that leaves an accumulated turnover. This is an uncommon problem.
• Major enterprises - Checks major enterprises and their development compared to same month last year. Some enterprises are investigated more closely and also contacted to double check the figures.
• Identical enterprises - A tool that compares that enterprises that was part of the survey last year with both the new and the old enterprises. This tool also gives individual development numbers, weights and impact on the development numbers for each industry.
• Graphical data validation - Checks for outliers in a diagram.
• Outliers in each size group- An SQL script where we examine the highest and lowest value in each size group and industry to check for outliers.
• Development numbers - The development numbers are re-calculated using other formulas to see if there is something off. Same with price changes.

An independent investigator also checks the data using different calculations then the survey. If something out of the ordinary is found the STS staff is notified.

At the end of a validation cycle a meeting is held where the results of the different parts is presented. Any issues that haven't been solved yet is discussed at this meeting.

Missing observations from unit and item non-responses are dealt with by using automatic imputations carried out by the estimation programs.

The estimator is an Horvitz-Thompson type of estimator combined with VAT-information. Sampled enterprises are replaced with turnover based on VAT-data when quarterly turnover data is available. Censuses units are never replaced with turnover from VAT.

Turnover based on VAT does not only apply sample units in stratas that are sampled, but all units in the frame so the survey is a census on a quarterly basis.

The index method is a chained Laspeyre-index with weights updated each year using annual overlap.

The grossing up is based on the relation between the number of enterprises in the population and the number of responding enterprises in each activity and strata.

A program developed by Statistics Sweden fetches data from the internal databases and converts it to GESMES/TS in order to transmit the data to Eurostat.

The services production index is deflated using different price indicies. For working day and seasonal adjustment: Software: Proc X12 in SAS 9.3 and ETS 12.1. Method: X-12-ARIMA.

The seasonal adjustment use an additive model.

We use a Partial Concorrent Revision principle for the seasonal adjustment. We make a revision every year and use it for the whole year until the next revision. Parameters are reestimated every time. The verification is checked in the annual revision.

Decomposition method is supposed to be decided by each time series. They are all multiplicative so far.

A national calender is used and we adjust for moving holidays (e.g. Easter). We also adjust for leap year.

in the original time series are linked using backcasting methods. 

Only direct SA is used. 

No further comments.