Statistics Sweden (SCB)

Department of social statistics and analysis

Section for salary and employment statistics

Statistics Sweden

701 89 Örebro

The purpose of this report is to describe how the Labour Cost Survey (LCS) 2020 was conducted and the quality of the survey in Sweden, according to the instructions in the Commission Regulation No 698/2006 regarding quality of labour cost statistics.

The LCS is an EU-regulated survey that is conducted every four years. The Swedish LCS2020 was carried out during 2021, basically in full compliance with the Commission Regulation. LCS2020 was conducted in a similar way as LCS2016, but section O (public administration and defence) was included for the first time and covid-19 affected LCS2020 in different ways.

Eurostat publish results by economic activity, size class and region for the member states. The labour cost is presented in total per year, per month and per hour, as well as per employee and per full-time equivalents. The distribution between the different labour cost items is also presented. The number of employees, hours worked and hours paid is presented for full- and part-time employees.

The results of the survey are used for analyzing economic growth, employment and labour costs for different industries in different countries and regions. One of the main results from the survey is the average hourly labour cost.

The classification systems that are used to present statistics by economic activity is NACE Rev.2 on section and division level, and for regional breakdown NUTS Level 1. Statistics are also presented by size class according to Eurostat standard. In appendix 6 the NACE sections and divisions are titled.

Economic activities defined by NACE Rev.2 sections B to S at division level in private and public sectors are covered. Enterprises with less than 10 employees are optional and have not been included. 

The purpose of LCS is to measure the level and structure of the labour costs; within/between countries, between branches and within/between regions. Labour costs refer to the total expenditure borne by employers for the purpose of employing staff. That include employee compensation, which is mainly comprised of gross wages and salaries in cash and in kind and employers social security contributions. Vocational training costs, other expenditure-, such as recruitment costs and spending on working clothes, and employment taxes are also regarded as labour cost. Finally, subsidies received and related to labour cost are deducted.

For the private sector, county councils[1] and some governmental authorities the enterprise (legal unit) is the statistical unit. For the public sector (municipalities and some governmental authorities), local unit is the statistical unit.

The target population for LCS is defined as all local units in NACE sections B-S belonging to an enterprise with at least 10 employees who have been active during 2020 in private or public sector. In LCS2020 section O was included for the first time in the Swedich LCS, even if it still is optional. Enterprises with less than 10 employees are optional and was not included. 

Sweden

The reference year is 2020. The survey has also been carried out for the reference years 1997, 2000, 2004, 2008, 2012 and 2016.

Eurostat produce some forecast estimates of the hourly labour cost annually. Results from the latest conducted LCS is used as a base and updated by the change in the Labour Cost Index (LCI).

The LCS is a tailored sampling survey with data collected directly from the employers. Their payroll and accounting systems is usually the main sources they use when they put together the requested information and respond to LCS.  

LCS is conducted every four years.

The samples were planned to be drawn in March 2020, but because of covid-19 the sample selections were postponed until August and pre information was sent out in September 2020 to the units in the samples. Since LCS only is conducted every four years and respondents have indicated that LCS is quite difficult, Statistics Sweden (SCB) finds it necessary to give pre information about the survey, so they have the possibility to prepare and save data for the reference year.

Some discussions were held about the optimal time period for data collection. Normally, the enterprises are occupied with balancing their accounts in January and February. To avoid disturbing this important work, the data collection started on March 15^th 2021 and respondents were given five weeks to respond.

A modified and up-dated version of the web questionnaire from the previous survey was used for LCS2020. The respondents  (payroll-/financial managers) received instructions including the web address, user id and a password from Statistics Sweden and were asked to log in and respond to the questionnaire. The respondent filled in the data on the web and some logical controls were made before the questionnaire was transmitted to SCB and the respondent was asked to give comments to the reported figures if they were outside certain limits. The English version of the paper questionnaire and instructions can be found in appendix 4 and 5.  

The non-response rate was high, 70%, at the time of deadline 20^th April. After deadline a lot of effort was made on collecting the missing questionnaires. Reminders were sent to all non-responding units and telephone reminders were also made. The companies are obliged to respond by law and may need to pay a fine if they fail to report. However, to not burden the respondents to much during the pandemic, Statistics Sweden took a general decision to not actively enforce this legislation. In previous LCS an order to pay a fine was sent to the largest enterprises failing to respond. Instead, a letter was sent directly to all CEO to underline that it is important and mandatory to respond.

From middle of March and onwards there was an on-going process of verifying and editing questionnaires and contacting respondents when indications of possible errors were detected. The data collection stopped in the beginning of 2022. During January – May 2022, the data were evaluated further at micro and macro level. Results of the Swedish LCS were forwarded to Eurostat in June 2022 as requested.

The graph shows that only around 30% had responded by the date of deadline, but that was not a surprise based on experiences from previous surveys where the inflow patterns were similar. At the end 89.8% of the questionnaires were received, compared to 90.9% in LCS2016. (The response rate including over-coverage was 91.6%, compared to 92.9% in LCS2016.)

Before the results were sent to Eurostat, validations of the result were done, using the recommended checks; relationship checks between variables, consistency checks and cross checks between tables.

A frozen version of the Statistical Business Register (SBR) was used for setting-up the frame. Two independent probability samples were then drawn; one for the Private sector (in this context defined as private enterprises, county councils and some state authorities) and one for the Public sector (municipalities and some state authorities). Enterprises were sampled from the private sector and in the public sector, local units constituted the sampling units. 

Comparability between the national and the European concept:

Statistical units

The statistical unit should be a local unit belonging to an enterprise with 10 or more employees. For private sector, county councils and some state authorities, local units are not the statistical unit for LCS2020 in Sweden. In these cases, the legal unit is the statistical unit. This is a problem when it comes to producing estimates by region and NACE. For estimates by regions a model is used to compensate for that. No adjustments have been done to estimates by NACE, which makes those estimates less accurate. For the public sector (municipalities and some state authorities), local unit is the statistical unit.

Population

Enterprises with less than 10 employees and NACE section O are optional according to the EU-regulation. In LCS2020 section O is included in the Swedish LCS for the first time, but small enterprises (1-9 employees), are still not included. It is a general request from Eurostat to include section O and most countries do.

Some variables should have been collected separately for apprentices. This has not been done since apprenticeship are not common in the Swedish labour market.

Reference time

2020 is the reference year. Some respondents reported for the fiscal year instead and in cases when the fiscal year where shorter or longer than 12 months Statistics Sweden adjusted from fiscal year to calendar year. For instance, if the fiscal year was 15 months, all figures except those concerning the average number of employees have been divided by 15 and multiplied by 12.

Classification of variables

Most variables are collected in LCS. In chapter 6.3.5 information regarding some model assumptions can be found. For example are Payments for days not worked (D1113) and Direct remuneration paid in each period (D11111) asked for as an aggregate in LCS in Sweden and are divided using a model. The model was improved for LCS2016 and the share that refers to D1113 became larger. In 2020, the model had to be adjusted further because of covid-19.

-

SCB adheres to the European statistics code of practice. 

There are reoccurring reviews by international experts to ensure that SCB follows the European statistics code of practice.

The results from LCS contribute with statistics for different economic analyses of the labour market, which is useful for both international and national users and policymakers.

Main users are Eurostat and other EU institutions. Other users are for example OECD, IMF, ILO, central banks, research institutes, media, employer’s associations, trade unions, private companies, the National Mediation Office and Statistics Sweden (SCB). The National Mediation Office is responsible for the official salary and labour cost statistics in Sweden and SCB produce surveys like LCS on their commission.

The statistics on salaries and wages in Sweden are quite well comprehensive, with both monthly and yearly surveys. However, statistics on the total labour cost are not so comprehensive, but LCS contributes with statistics in this area every four years. LCS is used for economic analyses on labour costs and no other survey than LCS measure the total level and structure of labour costs in such detail. The data on national level broken down by branch of industry are used for comparison with other countries.

The national users seems to be satisfied with LCS and no major gap has been reported. SCB also assumes that the main user, Eurostat, is satisfied with the quality of the overall results of the Swedish LCS. Eurostat has though requested statistics for section O (Public administration and defence) and most of the countries include that, even if it is not mandatory according to the regulation. Because of this, Sweden has included section O in LCS2020.

The survey was carried out basically in full compliance with the Commission Regulation. With the survey design in use it though becomes difficult to calculate the number in universe(N) and number in the sample(n) for tables containing regional data and data by size class that is requested. A legal unit can operate in more than one region and on the other hand a local unit is to be attributed to the size class according to the legal unit to which it belongs. Because of this the number in universe(N) and number in the sample(n) have not been reported for tables containing region or size class.

94% of the requested cells were delivered. 

See sub-secitons

The assessment made by Statistics Sweden is that the overall accuracy is high.

The following can affect the accuracy of the statistics: sampling, measurement, coverage, non-response, model assumptions and data processing. For this survey the main sources of errors are likely to be sampling errors and measurement errors. The other errors are considered minor.  

Sampling errors is a consequence of not surveying the whole population.

More information about the sampling design can be found chapter 3.5.

The following tables contain coefficients of variation (cv) for the key variables of the survey, Annual labour cost and Hourly labour cost[1]. The estimated cv’s are produced using CLAN[2] under the H-T estimator. The cv’s are estimated at the population level and by breakdowns according to NACE sections, size bands and regions respectively. The cv’s are small on both NACE level, size level and region for the two variables. The cv’s for the combination of NACE and size are generally higher, see appendix 1.

Section Q (Human health and social work) is the largest section, containing 18% of the Annual labour costs and section C (Manufacturing) is the second largest with 14%. 49% of the Annual labour cost comes from enterprises with more than 1 000 employees. About 14% of the Annual labour costs can be found in the north region (SE3) of Sweden, 39% in the south region (SE2) and 47% in the east region (SE1).

Note: Annual labour cost = Code D (total labour cost), sum of the values of code D1, D2, D3, D4 minus D5 in appendix 1 regulation No.  1726/1999

Note: Annual labour cost = Code D (total labour cost), sum of the values of code D1, D2, D3, D4 minus D5 in appendix 1 regulation No. 1726/1999

NUTS 1:

SE1 – Östra Sverige: Stockholm, Uppsala, Södermanland, Östergötland, Örebro, Västmanland

SE2 – Södra Sverige: Jönköping, Kronoberg, Kalmar, Gotland, Blekinge, Skåne, Halland, Västra Götaland

SE3 – Norra Sverige: Värmland, Dalarna, Gävleborg, Västernorrland, Jämtland, Västerbotten, Norrbotten

The Hourly labour cost was 399 SEK for B_S. For B_S excl. O the hourly labour cost was 398 SEK, a 11% increase from 2016. Section K (Financial and insurance) has the highest Hourly labour cost (672 SEK) and section I (Accommodation and food service) the lowest (268 SEK). This means that the Hourly labour cost is 151% higher in section K than in section I.

Enterprises with 10-49 employees have the lowest Hourly labour cost and enterprises with 250-499 employees have the highest. 

In 2020 the statutory social contributions were lower for up to 30 employees in an enterprise. This affected the labour cost in the small enterprises more than the big ones.

The east region of Sweden (SE1), which includes the capital of Sweden, has around 17% higher Hourly labour cost compared to the north region (SE3).

Note: Hourly labour cost = Code D (total labour cost), sum of the values of code D1, D2, D3, D4 minus D5, divided by the value of code B1, in appendix1 to regulation No. 1726/1999

 Note: Hourly labour cost = Code D (total labour cost), sum of the values of code D1, D2, D3, D4 minus D5, divided by the value of code B1, in appendix 1 to regulation No. 1726/1999

NUTS 1:

SE1 – Östra Sverige: Stockholm, Uppsala, Södermanland, Östergötland, Örebro, Västmanland

SE2 – Södra Sverige: Jönköping, Kronoberg, Kalmar, Gotland, Blekinge, Skåne, Halland, Västra Götaland

SE3 – Norra Sverige: Värmland, Dalarna, Gävleborg, Västernorrland, Jämtland, Västerbotten, Norrbotten

See sub-sections.

A lot of effort was made to prevent empty strata and response was given in all sampled strata in LCS2020. To keep the sample sizes down, some small NACE divisions were not sampled. The under-coverage because of this is less than 0.1% of the total number of employees, see appendix 3.

The samples for LCS2020 were drawn in August[1] 2020, instead of March as in previous LCS, because of the pandemic. This means that units that started between August-December 2020 were not in the frame or in the sample and might have contributed to the under-coverage error. A general rate of under-coverage has not been calculated and studied, but there is no reason to believe that this is a major source of error.

In cases of over-coverage, new units have not been sampled.

The following cases have been treated as over-coverage: 

- Enterprises/local units who died before 2020 or during the early part of 2020.

- Enterprises/local units who were sleeping during 2020.

- Enterprises/local units who did not have any employees during 2020.

- Enterprises that was incorporated into another enterprise in the frame.

The overall rate of over-coverage was 2.0%. In the private sector the rate was 0.7% and in the public sector 6.8%. 

The table below shows the over-coverage in the samples.

-

In 2019, SCB was undertaken a project to prepare, improve and up-date the LCS-questionnaire, software production system and guidance that was used in previous survey. The web-based questionnaire and its logical controls was updated and adjusted for LCS2020. If a control was triggered, the respondent had to correct or make a comment before the questionnaire could be submitted to SCB. Once submitted, the questionnaire is subject to further controls at SCB.

A minor cause of measurement error might have been that information about apprentices were not collected separately. The reason is that this form of employment is very rare in Sweden. A few apprentices can be found in some branches, for example in the construction industry, and are in those cases probably treated as regular employees. Since they are few, they will likely not affect the estimates and it is not worth collecting separately. This can change for next LCS, and will of course be considered if the situation change.

The table contains information about the number of units responding, not responding or are classified as over-coverage in the two samples.

The response rate can be defined in different ways depending on how over-coverage is treated and the response rate is calculated in three different ways below.

  - If the over-coverage is considered as non-response, the response rate is 89.8% using the following formula. (For 2008 the figure was 87.4%, for 2012 86.9% and for 2016 90.9%)

- If the over-coverage is considered as response, the response rate can be expressed by this formula.

For 2020 the overall, non-weighted, response rate was 91.6% including 2.0% over coverage. Appendix 2 contain tables of unit response rates, broken down by the stratification used for the two samples. For the private sector the rate was 92.2% and for the public sector 90.6%. The above formula was used to calculate the response rate in previous quality reports for LCS and the rate over time is presented in this table.

- If the over-coverage is excluded in both the numerator and the denominator, the response rate can be expressed by the formula below. The response rate is calculated in this way in many of the labour market surveys in Sweden. The response rate is in this case 91.7 % which can be compared to 87.2 % for 2004, 89.9% for 2008, 89.2% for 2012 and 92.8% for 2016.

The method that has been used to reduce the size of the error resulting from non-response is re-weighting within strata, i.e. imputation of mean value within the strata. This method has been used for each stratum where non-response has occurred. If this method is to work satisfactorily, non-response has to occur randomly within stratum. In the largest size class, one has to study the results carefully in case of non-response, because of possible large differences in the number of employees of enterprises concerned.

The response rate is slightly lower 2020 compared to 2016. Main reasons is probably problems because of the pandemic, letter to pay a fine were not used and priorities because of limited resources. Despite the difficult circumstances 2020, the response rate is still high and at the same level as in previous LCS. The response rate for LCS is also high compared to other similar surveys.

Possible reasons for high response rate: 

• Some respondents have experience from participating in previous surveys.
• Because of the small sample size, it has been a lot of work and focus on reminding and urging the respondents to participate in the survey, to increase the overall response rate and reduce the risk to end up with empty strata.
• The change in the sample design, used from 2008 and onwards, has probably had a positive impact on the response rate. Respondents usually find it easier to answer on enterprise level and the 20 county councils are sampled on enterprise level, instead of local unit. 
• Instructions for each variable together with account guidelines were on the website and easy for respondents and payroll system providers to find. That might have contributed to high response rate.
• Statistics Sweden cooperate with the payroll system providers. Since LCS2016 they could get newsletters with information about the survey from SCB. In some cases meetings and presentations were organised. Many of them offered assistance to the respondents and some have also created standard reports for LCS.
• The survey is mandatory, companies are obliged by law to respond and may need to pay a fine if they fail to report. However, to not burden the respondents to much during the pandemic, SCB took a decision to not actively enforce the legislation. Instead increased the reminders, sent a letter directly to the CEO, and relied on the good will of the respondents.
• LCS is only conducted every four years with a long collecting period. The possibility to extend the data collection period and give the respondent more time, contribute to a high response rate at the end. The fact that the survey only is conducted every four years, is also an argument that managed to persuade respondents to participate in same cases. On the other hand, it is also challenging when respondents are not used to the survey and sometimes find it time consuming and complicated.
• Pre information was sent to the units in the sample as in previous surveys. That gave the respondents the possibility to save data for 2020 and time to prepare. SCB also got information about some over-coverage and contact persons by this procedure. This has most likely contributed to the high response rate. However, the pre information was sent out later than usual since the sample was drawn in August instead of March, that might be one explanation for the increase in time it took to respond to LCS2020. 

See Appendix 2 - Response rate 2020

Item non-response were handled in the editing process using manual imputation. In some cases, the national monthly and yearly surveys for salary and wages have been used for comparison of average salary and used for imputation when there have been partial non-response and possible outliers. Also received data within a strata have been useful.

SCB has tried to minimise the risk for processing errors. For example by ensuring that critical steps in the production process are documented and that production of the results were compiled in close cooperation between subject matter specialist and the methodologist.

After careful consideration no unit non-response were imputed in LCS2020.

Estimates by region - NUTS1

Sweden is divided into three regions according to NUTS level 1 since 2008. Regional data was there­fore produced for LCS2008 for the first time. LCS-data are collected at enterprise level for the private sector and not on local unit. This is a problem when it comes to producing estimates by region and a model is needed. For enterprises with local units in more than one region, the number of employees at local unit level from the Business register (BR) was used to perform the allocation of the enterprise data to the local units. This model is likely to function well for variables related to number of employees. For variables related to costs, it is likely to produce some bias. An enterprise with local units in more than one region will have the same average labour cost in all regions. Other surveys show that the average salary is higher in the Swedish capital, Stockholm, than the rest of Sweden. Labour costs for the region containing Stockholm are likely to be under­estimated, while the other regions are likely to be overestimated by this model.

For LCS2012 the model for producing regional data was improved and also used for LCS2016 and LCS2020. This model is likely to give less bias in the regional estimates. The yearly survey for salaries and wages for private sector (SLP) was used for enterprises included in both SLP and LCS. In 2020 about 83% (2 895) of the enterprises were in both surveys. SLP has information for September about each employee in the enterprise and in which local unit the employee works. Using this information, the average monthly salary and number of employees by region for each enterprise were calculated. The cost variables in LCS were then distributed by the average salary per region for each enterprise. The number of employees and number of hours worked and paid were divided using the distribution of number of employees from SLP. For enterprises that were not in SLP, the previos model was used. 575 enterprises did not match and 465 of them were located only in one region. This means that the ’old’ model was used for 110 enterprises. The table below shows the number of enterprises in LCS, which matches SLP, by the number of regions the enterprises have local units in.

The following table show the number of employees (according to the BR) in each region, divided by whether they stem from a single, two or three region unit. This is done in order to give a rough idea on how model dependent the regional estimates are.

NUTS 1:

SE1 – Östra Sverige: Stockholm, Uppsala, Södermanland, Östergötland, Örebro, Västmanland

SE2 – Södra Sverige: Jönköping, Kronoberg, Kalmar, Gotland, Blekinge, Skåne, Halland, Västra Götaland

SE3 – Norra Sverige: Värmland, Dalarna, Gävleborg, Västernorrland, Jämtland, Västerbotten, Norrbotten

The table below show that 42 381 enterprises have local units in only one region. There are only 8% (3 437=2 141+1 296) of the enterprises that have local units in more than one region, but they employ about 1.3 million persons. This means that 43% of the employees work in an enterprise with local units located in more than one region and are therefore included in the model used for the estimates on region. This indicate that the large enterprises tend to be located in more than one region more often than small enterprises. In the sample 1 003 (430+573) enterprises (29%) were in more than one region.

 Model for dividing the aggregate of the variables (D1113) and (D11111)

One experience from the first run of LCS was that the respondents found it difficult to separate their costs between D1113 (payments for days not worked) and D11111 (direct remuneration etc. paid in each pay period). Therefore, a decision was made for LCS2000, to stop collecting these items separately, but instead ask for the sum of D11111 and D1113. A model was developed to separate the sum into the two variables, and the same model was used until LCS2016, when the model was improved. Employees in Sweden have the legal right to 25 vacation days per year. Holiday pay is statutory by 12% of the employee's wages. Collective or individual agreements may stipulate more vacation days and higher percentages. The number of vacation days have increased over time in private sector, but that was not reflected in the model. Therefore, the model was modified leading to a re-distribution from D11111 to D1113. The main differences between the models is how the payment for vacation is estimated and how public holidays are treated. The new model use new information about number of vacation days from the yearly salary survey, which enables us to differentiate the calculations according to both NACE and institutional sector. The new model favors days not worked with around two percentage points compared to the old model.

In 2020 the model was adjusted again, because of the pandemic. In 2020, new support to employers were introduced to reduce labour costs because of covid-19, such as reduced employer social contributions, compensation for sick pay costs and short-time work allowance. Some companies had employees on short-time work during part of 2020. A full-time employee may for example worked 40% of full-time, and was on short-time leave at 60%. The employer had to pay 92.5% of full-time salary to the employee for 40% hours worked. The enterprise could apply for compensation from the state as a subsidy (D5). This means that payments for days not worked increased in those enterprises and the model had to be adjusted. The short-time work affects the distribution of D1113 and D11111. About 1/3 of the enterprises in the LCS-sample for private sector received short-time work allowances for 2020.

  Other model assumptions errors

- Small enterprises with less than 10 employees are optional and have not been included or accounted for. About 15% of the employees work in an enterprise with less than 10 employees.

- No data is given for apprentices. The reason is that this form of employment is very rare in Sweden. So rare, that it was not considered worthwhile to ask about apprentices separately.

- Respondents were asked to report data for 2020, but as a second choice they were given the possibility to report data for the fiscal year. Adjustments from fiscal year to calendar year have been made by SCB, when needed. For instance, if the fiscal year was 15 months, all figures except those concerning the average number of employees have been divided by 15 and multiplied by 12.

- In 2020, D11144 (stock options) remained optional. Once again the question about stock options (D11144) was integrated with D11112 (direct remuneration, bonuses and allowances not paid in each pay period). Some respondents found it difficult to answer the question about stock options. Statistics Sweden does not know how many enterprises included stock options nor the magnitude of the value. The general opinion is still that stock options is a small part of D11112.

- Individually agreed social security contributions are quite common in Sweden. As in previous surveys, this variable was asked for separately. This cost has then been added to variable D1212 (collectively agreed, contractual and voluntary social security contributions), just like in the results of the previous surveys.

- A model to adjust Subsidies and Social contributions was used in LCS2020. Subsidies containing subsidies for both salaries and social contributions have been asked for as a lump sum, since it is difficult for the respondent to split. This type of subsidies was common 2020 and SCB have tried to adjust the variables Subsidies (D5) to only include subsidies for salaries (as Eurostat want) and reduced the social contributions by the subsidies for social contributions.  

-

-

-

-

Statistics Sweden awaiting final results from Eurostat, and then plan to publish results with comparison with the other countries, similar to previous LCS.

-

-

The tables for the Swedish LCS2020 were forwarded to Eurostat in June 2022 as stipulated by the regulation.

- 

All EU-member states and some other European countries conduct LCS. Eurostat produce results for the different countries and for EU as a whole. The countries that have conducted LCS have not been the same over time which is worth having in mind when analysing and comparing results.

Region 

Regional data on NUTS Level 1 is required. During the work with the first LCS 1997, Statistics Sweden found that one of the major difficulties for the respondents was to report data at local unit level. Analyses was carried out to assess how different the Swedish LCS results would have been if data instead had been at enterprise level. There are basically two ways such a change can affect the results. First, data broken down by regions might be incorrect if data is given at the enterprise level. Before 2008 this problem did not exist since Sweden was one region at NUTS 1 level. Secondly, data broken down by NACE might be affected. However, the analyses indicated that this problem was minimal. Therefore, with the intention of making life easier for the respondents and thereby increasing the quality of the data, it was decided to sample enterprises instead of local units for the private sector. Eurostat was in­formed and this sampling design has been kept since 2000. From 2008 also county councils have been sampled at enterprise level. For the public sector, local units are the sampling units.

Sweden is divided into three regions according to NUTS level 1 since 2008 and regional tables were produced for the first time for LCS2008. Since data for the private sector are collected on enterprise level and not on local unit level in Sweden, a model was used to split the enterprise data by region. For LCS2012 the model was improved utilizing information from the yearly salary and wage survey. The same model has been used since then and is described in chapter 6.3.5. Below is information about the counties that belongs to the regions.  

NUTS 1:                                          County                            Area code

SE1 – Östra Sverige:                        Stockholm                                         01

          (East Sweden)                         Uppsala                                             03

                                                          Södermanland                                   04

                                                          Östergötland                                      05

                                                          Örebro                                               18

                                                          Västmanland                                     19

SE2 – Södra Sverige                         Jönköping                                         06

          (South Sweden)                      Kronoberg                                         07

                                                          Kalmar                                               08

                                                          Gotland                                             09

                                                          Blekinge                                            10

                                                          Skåne                                                12

                                                          Halland                                              13

                                                          Västra Götaland                                14

SE3 – Norra Sverige                         Värmland                                          17

          (North Sweden)                      Dalarna                                              20

                                                          Gävleborg                                         21

                                                          Västernorrland                                  22

                                                          Jämtland                                            23

                                                          Västerbotten                                      24

                                                          Norrbotten                                         25

- 

LCS2020 in Sweden was conducted in a similar way as LCS2016. However, 2020 was a special year and the pandemic has affected the survey in many ways and it was more complicated to compare, verify and edit the reported data. Comparing statistics over time can be challenging and knowledge about different circumstances over time is useful. Below are some changes that might have impact on the comparability.

NACE coverage

The first LCS was conducted for the reference year 1997. The sample was drawn at local unit level for section C-K in NACE Rev.1. In 2000, two independent samples were drawn. One at enterprise level for NACE C-K in the private sector and one at local unit level for the public sector. In 2004 the sample was drawn in the same way, but the following sections were included for the first time; Education (M), Health and social work (N), and Other community, social and personal service activities (O) according to NACE Rev.1. The public sector represents about one third of the total economy and is dominating in those sections.

Since LCS2008 the NACE Rev.2 standard is used, which was a major change and data before 2008 are not comparable by NACE. (Although, LCS2008 were double coded and results on section level were also produced by NACE Rev.1.) The number of sections as well as the number of divisions increased, 81 divisions are asked for in sections B-S excl. O. In NACE Rev.1 this number was 54, an increase by 27 divisions. In one of the requested tables, containing size and division, this means an increase from 270 to 405 groups. The increase in number of cells, for which estimates are required, put more strain on the survey design. Also, the sample size had to be smaller from LCS2008 and onwards due to budget restrictions and response burden.

In LCS2020 section O (Public administration and defence) is included for the first time in the Swedish LCS. To include O is still optional, but most countries do, since it is a general invite from Eurostat with the intention to increase the coverage and comparability.

Sampling time-point

Since LCS2004 the sample has been drawn in March the reference year. For LCS 1997 and 2000 the sample was drawn in November the year before the reference year. To draw the samples in the same year has resulted in less over coverage. The plan for LCS2020 was to also draw the samples in March but had to be postponed due to the pandemic and were instead drawn in August 2020. This meant that the respondents had less time to save data for 2020 and prepare their systems. It also might have caused more under coverage, less over coverage and increased time spent responding to the survey for some respondents. However, we don’t believe this has affected the comparability over time significantly.

Web-based questionnaire

The collection method changed fundamentally for LCS2008 when a web-based questionnaire was introduced. The respondents were given a web address, user id and pass­word. They filled in their data and some logical controls were made before the questionnaire was sent to SCB. 88% of the questionnaires were collected this way for LCS2008 and 98% for LCS2012 and only a few respondents did not use the web for LCS 2016 and 2020. The number of integrated logical controls in the questionnaire have increased and been adjusted each year. Whether the change in collection method and controls has affected the statistics has not been thoroughly studied.

New auxiliary information

Employers has begun to report information about the paid-out salaries for each employee to the Tax authority monthly. Based on this information, number of employees by salary intervals for each enterprise was derived. SCB had also got access to information if an enterprise had applied for short-time allowances 2020. The access to the new auxiliary information were useful when validating collected data for LCS2020 and has most likely contributed to increase the quality of the statistics.

Covid-19

The pandemic has affected LCS2020 in many ways. Beside moving the sampling point from March to August and challenges collecting and validating the reported data, it has also affected the statistics itself. Statistics on hours worked and paid, social contributions, salaries, subsidies and guarantee pay­ment in event of sickness are examples of variables that have been affected. Also, the model to create the variables D11111(Direct remuneration, bonuses and allowances paid in each period) and D1113 (Payments for days not worked), from the collected aggregate, had to be adjusted. The model to split the aggregate was improved 2016 using new informa­tion about vacation days from the yearly salary survey. In 2020 the model was adjusted further because of covid-19, especially because of short-time work. These variables are therefore not completely comparable between years.

In 2020, many countries introduced different types of governmental support to reduce the labour costs for the employers, with the intention to mitigate the financial consequences of covid-19. Reduced social contributions, short-time work allowance and compensation for sick pay were common in Sweden. Almost all employers had compensation for sick pay and reduced social contributions. Enterprises could also, instead of dismissing employees, apply for support for short-time work for their permanent staff. This meant that the costs were shared between the employees, the company and the state. (To get short-time work allowances, temporary staff were not allowed at the enterprises at the same time.) Working hours could be reduced by 20%, 40% or 60% (May-July by 80%), but wages and employers’ contributions was not reduced to the same extent. For example, at 60% layoff, hours worked is 40% of full-time, and paid hours 92.5% of full-time since the employer paid 92.5% of the full-time salary to the employee. They also paid social contributions on that salary. The enterprise could apply for short-time work allowances and receive subsides containing support for both salaries and social contribu­tions. SCB asked for the total amount of subsidies as a lump sum in LCS2020, since respondents usually had this information. Later, SCB tried to adjust the subsidies to only include subsidies for salaries and deducted the social contributions, since subsidies (D5) should not include social contributions according to Eurostat. 

About 1/3 of the enterprises in the LCS-sample for private sector received short-time work allowances for part of 2020. The graph shows the share of enterprises that had short time work allowences in 2020 by section, and in section I(hotel and restaurants) this was most common, almost 80%. Divisions with high share was for example division 51(Air transport), 55(Accommodation) and 79(Travel agencies).

Social contributions

How social contributions are formed and changed over time have impact on the labour cost statistics. Employers’ social contributions constitute a large proportion of the total labour cost in Sweden compared to many other countries, and therefore have a large impact when looking at statistics on labour costs.

The employer pays statutory social contributions on the salary to the tax authority. The rate was 31.42% in 2020 and been the same for many years as can be seen in the table. But there have been exceptions from the rate, for example for small enterprises, specific activities and young employees. The intention has been to make it easy to start and run enterprises, strengthen/support/encourage specific activities and make it easier for young people to get a job and enter the labour market.

In 2020 the rate was lower for employees under 18 and over 65 years old, as well as for employees working with research and development. The rate was also temporarily lower between 1 March – 30 June due to covid-19 for up to 30 of the employees in an enterprise. The employers only needed to pay 10.21% during that period for gross wages up to SEK 25 000 per employee and month. On amounts exceeding this, full contributions applied (31.42%). The temporary reduction had more im­pact on the small enterprises than the bigger ones, since the reduction was applied for up to 30 employees.

Knowledge about the social contribution design and the age structure in the labour market can be useful when analysing the data. The rate for social contributions depending on the age of the employees has varied over time. In 2012 the statutory social contributions were 15.49% for employees younger than 26 years, and in 2016 the rate for young employees was 25.46% during January-May and from June full rate was paid. In 2020 the rate was 10.21%. for employees under 18 and over 65 years. The collective social contribu­tions are also often lower for young staff and from the age of 23 or 26 full collective contributions is usually paid. Most common age for retirement in Sweden is at 65, but it has become more common to work after that. In 2020 about 1% of the employees were between 65-67. Most of them worked in section H (Transportation and storage) and Q (Human health and social work) where the share was over 2%. The share for different age groups can be estimated from the yearly national salary survey. In the graph below the share of young employees (18-25 years) by NACE is shown. The highest share, about 35%, can be found in section I (Accommodation and food). In almost all sections the share of young employees has decreased 2020 compared to 2016. It might be an effect of covid-19. When enterprises got financial challenges and need to reduce the number of employees, temporary staff and the ones that did not work that long usually have to leave first, which in many cases are young staff.

Exchange rate

Results are presented both in Euro and in Swedish Krona (SEK). The results in Euro are in­fluenced by the exchange rate and the graph below shows the exchange rate the LCS years and how the rate has varied over time.

When comparing the change of the labour cost over time it is important to know if the change is calculated in Euro or in national currency. Below is an example that shows the increase of the Hourly labour cost for NACE section B-S excl. O from 2004 to 2020. Between 2016 and 2020 the Hourly labour cost increased by 11% when calculating in national currency and only 1% when calculating in Euro. 

Since LCS2008 the NACE Rev.2 nomenclature is used, which was a major change and data before 2008 are not comparable by NACE.

Below are comparisons between LCS and some other surveys.

Labour Cost Survey vs. Labour Force Survey

The graph below shows a comparison of Hours actually worked[1]expressed per employee during 2020 in LCS and the average actual hours worked in the main job per employee 2020 in LFS (Labour Force Survey). Differences between LFS and LCS is that LFS use the population register as a frame and cover the whole labour market and it is the employees that respond to the survey. LCS only covers enterprises with 10 and more employees and it is the employer that respond to the survey for all employees in the enterprise/local unit.

Labour Cost Survey vs. Structural Business Statistics

The graph below shows the Wages and salaries[2], expressed per employee from LCS compared to SBS (Structural Business Statistics). When comparing the LCS and the SBS one must know that there are a couple of significant differences between the two surveys. Firstly, enterprises with less than 10 employees are excluded in the LCS. Secondly, the public sector is included in LCS but not in SBS. Section O, P, Q and R and some industries within K and S are not covered in SBS. These sections are therefore excluded for SBS in the graph.

Labour Cost Surey vs. Labour Cost Index

The graph below shows the Average annual growth rates[3] in national currency (SEK) for the Hourly labour costs[4]by NACE between year 2016 and 2020 in LCS and LCI (Labour Cost Index). LCI does not include Vocational training costs (D2), Other expenditure paid by the employer (D3) or Subsidies received by the employer (D5). Enterprises with less than 10 employees are included in the LCI-estimates, but not in LCS. The temporally reduction of the social contributions that were in place during 4 months 2020, benefitted the smaller enterprises in a larger extent. That is one explanation why LCI increased somewhat less than LCS in most sections.

Labour Cost up-dates - Annual data

Since LCS is only conducted every four years, Eurostat forecast the labour cost levels by NACE section for the years in between using LCI. The levels from the latest conducted LCS is used as a base and updated according to the change in LCI. The forecast for 2020 have LCS2016 as a base. When comparing that forecast with results from the conducted LCS there are of course differences. Things that can explain the differences are for example that both LCI and LCS are sample surveys with sampling errors. LCI include small companies but does not included items like costs for vocational training, other expenditure or subsidise. The total hourly labour cost were 391 SEK from the annual estimate and 398 SEK from the conducted LCS2020 (B_S excl.O). The differences tend to be higher in the different NACE sections.

Labour Cost Surey vs. National Accounts

The graph below shows Compensation per employee[1] during 2020 in LCS and National Accounts (NA). Compensation per employee includes salaries and social contributions but no taxes and the main explanation why the LCS-bars are higher than the NA-bars is differences in the definition of tax vs social contributions. NA changed the definition of social contributions vs. taxes because of ESA 2010 in 2014 back to 2012. In connection with the transition to ESA 2010, a stricter interpretation of what should be included in social contributions was introduced. Henceforth, only payroll taxes earmarked for its purpose are recognized as social contributions. Only pension contributions are now considered as social contributions in NA. A large part of the statutory social contributions, like parental and sick insurance contributions are now considered as taxes instead and the compensation of employees is therefore lower in NA. LCS has not changed the definitions and only the special wage tax is considered as tax. Eurostat discuss the distinction between social contributions and taxes in NA and will hopefully come up with recommendations how this should be treated. Another explanation why the LCS-bars are higher than NA-bars is that LCS excludes enterprises with less than 10 employees, and larger enterprises are considered to have slightly higher compensation per employee.

 

 [1] Code D1, divided by the value of code A1, in appendix 1 to Regulation No 1726/1999.

LCS-data transmitted to Eurostat are internally coherent. Before the results were sent to Eurostat, validations of the result were done, using the recommended checks; relationship checks between variables, consistency checks and cross checks between tables.

-

On the website at Statistics Sweden is a link to Eurostat’s website where the results for all countries can be found[1]. Some tables and graphs for LCS 2004, 2008, 2012 and 2016 are also published on the website at Statistics Sweden and results for 2020 will be published there as well. 

Eurostat publish results for LCS in their database. 

-

-

The plan is to also publish articles with results from LCS2020.

Results will not be sent to the reporting units. However, in the pre information, all sampled units received the main result of last LCS. The idea is to also give the sampled units in LCS2024, some results from LCS2020.

Beside this quality report, Statistics Sweden has additional documentation. Some of the metadata documentation is available for the users on the website. There is also a link to Eurostat where users can find the results and quality reports for other countries. 

-

-

-

A voluntary question about the time spent completing the questionnaire has been included in the questionnaire since LCS2008. To include this question has become common practice in most of Statistics Sweden’s surveys as a way of measuring response burden.

The question on the amount of time respondents spent on fulfilling the questionnaire has changed between the collection of LCS2016 and LCS2020. Also, the population for LCS2020 has another composition. This makes it difficult to analyse the difference in time spent on LCS2016 and LCS2020. However, since the change in the question has been made in all business surveys conducted by Statistics Sweden, we can safely say that the new formulation of the question generally has led to an increase in the stated time spent on fulfilling the questionnaires. Unfortunately, it is not possible to determine whether this is the sole explanation or if in fact the response burden have increased since LCS2016.

About 60% of the respondents answered this question for LCS2020 and in previous years only around 40%. The non-weighted average time for completing the questionnaire for LCS2020 was 7 hours and 47 minutes. For the private sector, county councils and some governmental authorities this figure was 7 hours and 32 minutes and for the municipalities and some governmental authorities 9 hours and 15 minutes.

The response burden seems to have increased each year according to the answers to this question. The main reason is probably that more controls have been gradually integrated in the web questionnaire and the respondent are asked to correct or give comments to their figures in larger extent. On the other hand, it has in many cases been possible for Statistics Sweden to correct/adjust the reported figures according to comments and information that was provided without recontacting the respondents.

LCS2020 was extra difficult for some enterprises, because many variables were affected by the pandemic. Short-time work and sick leave have for example affected hours worked, hours paid, sick-payment, subsidies etc. This has probably contributed to the increased average time spent to respond to LCS2020.

The graph below shows the answers from the respondents grouped into time intervals 2016 and 2020. The median was 4 hours and 30 minutes in 2020, which mean that 50% needed less than that time to complete the questionnaire. About 25% needed more than 8 hours to complete the questionnaire. 

The information that is reported to Statistics Sweden is protected by confidentiality according to chapter 24 section 8 of the security Act (2009:400). The treatment of confidentiality is in line with the policy adopted by Statistics Sweden.

Primary confidentiality flags have been set using the same methodology as for previous LCS. A cell is given a primary confidentiality flag if one or more of the following conditions occur: 

• The number of contributing units is less than four
• One observation account for more than 70 percent of the total estimate of the number of employees (A1)
• Two observations account for more than 95 percent of the total estimate of the number of employees (A1)

Eurostat request secondary confidentiality flags. This has been done where needed, in such a way that the estimates with the smallest value of the number of employees (A1) have been flagged with secondary confidentiality.

-