Instituto Nazionale di Statistica - ISTAT

National Statistical Institute of Italy - ISTAT

Dipartimento per la Contabilità Nazionale e le Statistiche Economiche

Unità Operativa Statistiche su Retribuzioni, Costo e Domanda di Lavoro

National Accounts and Economic Statistics Department

Earnings, Labour Costs and Labour Demand Statistics Unit

The first Structure of Earnings Survey (SES) took place for reference year 1995 covering all sectors of economic activity from C to K of the NACE Rev. 1.

Since then, according to Council Reg.530/99 SES was carried out every 4 years, starting with year 2002 and covering all mandatory scope. In 2002, in derogation to Council reg.530/99 the compilation of statistics was based only on enterprises; moreover a pilot Survey on three key sectors of Italian economy, namely Textile, Retail trade and Services to business, for micro-enterprises (business with at least 1 employee to 9 employees) has been carried out to test the burden and quality of results of the extension of SES to micro enterprises; results for this pilot survey showed evidence of reduced quality and excessive burden for small business.

Two major improvements have been implemented in the 2006 round: data referring to Local Unit and Public Institution in sectors of economic activity (in the Nace Rev.1.2) M (education) and N (health) have been covered using administrative, fiscal and data from other statistical archives, given by the Labour Force Survey and the Eu_Silc survey.

With regards to the session 2010, the passage to the NACE Rev. 2 has been implemented, covering the whole required scope from B to S (O excluded, that is optional). From a methodological point of view, a final optimization of the integration of the different archives has been assured and finally structured in order to estimate the final data on the sector of the public Instruction. On the other side, a strong change on the direct survey has been implemented by establishing a on line data acquisition, together with a complete re-engineering of the statistical process that assured a better result in terms of quantity and quality of the supplied microdata.

In accordance with the Commission Regulation No 698/2006 of 5 May 2006 concerning quality evaluation of Statistics on labour costs and earnings, Istat provides Eurostat with required information on Quality Indicators and Specific Methodological Notes, concerning Structure of Earnings Survey year of reference 2010.

The framework of the quality report is as follows: descriptive analysis of each applicable quality evaluation criteria and specific methodological notes are reported in the text, while required tables with quality indicators on Accuracy and Response are described in attached documents.

Reference period

Structure of Earning Survey refers to year 2010 for annual variables and to Month of October 2010 for monthly data in line with Commission Reg. 1738/2005.

Statistical unit

Reporting units consisted of enterprises and institutions; results relate to local units and to employees.

Scope of the survey

SES 2010 covers all sectors of economic activity from B to S (O excluded) NACE Rev 2 for unit with at least 10 employees

Variables and definitions

Each variable is defined according to Commission Regulation 1738/2005.

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 main source of data on Structure of earnings is the SES 2010 survey, based on a two stage sample design: first stage is a sample of enterprises employing at least 10 people, afterwards a sample of employees is drawn from each enterprises belonging to the first sample. As regards the enterprises belonging to the register ASIA, a sample has been drawn on the enterprise with less 250 employees and a census on the ones with more than 250 employees. As regards the public sector (excluded Section O and P), a Census of all the institutions has been implemented.

As far as Public Institutions in Education (Section P), an integrated system of both administrative and fiscal files) and statistical sources (Labour Force Survey and Eu_Silc) have been used.

Sample design, selection procedure, stratification and size of the sample from the register ASIA

The Structure of Earnings survey sample design is based on a two stage sample; a primary units consists of enterprises, while final units are employees. The target population of the SES2010 was identified by three different lists of statistical units, two of which previously gathered and one on purpose built internally in the survey, using administrative, fiscal data and statistical archives from other surveys. In this view, the features of the sample design will be described according to the list of units. In this regard, the sample design has been organized in a way to have a different scheme according to the list:

1. a sample for the private enterprises with a direct survey;
2. a census on the public institutions belonging to the Istat S13 list (that included Universities), whose data was collected by direct survey as well;
3. the sector of Education (except the Universities) were estimated only with data from other archives (administrative, fiscal and from other surveys).

The sample design varies according the different frame with regards to the first stage, but about the second stage the design is the same in every case. Within each statistical units, belonging to the final list, the number of units to be included in the sample of employees was established according to a predetermined size proportional to the size class of the enterprise (from a minimum of 10 to 200 employees). Every unit could choose the employees to include into the sample, Istat gave proper instruction in order to guarantee the randomization of the sample.

As regards the Business Register ASIA , the target population included 176,295 enterprises, of which 3380 belong to the size class 250 and more employees; 40,6% of employees (3,296,097) out of a total of 8,119,388 is employed in big enterprises. As far as the first stage is concerned a stratified sample of enterprises with 10 to 249 employees has been designed, while a census survey of big business (with 250 and more employees) was carried out. Sampling size for enterprises with 10 to 249 employees, which fulfils the target of 3% for the coefficient of variation of a transformed function of monthly gross earning per employee, resulting from the multivariate allocation procedures adopted, has been increased considering the response rate recorded in previous SES rounds. Units have been drawn according to Pareto sampling, a specific case of probability proportional to size selection. Where possible, an increase of 17% of sample was performed in order to prevent the negative effect of total non-response.

A further improvement has been implemented in order to guarantee, where possible, that the second stage sample would represent the actual structure of the target population by age and gender. Indeed, administrative fiscal files has been used as auxiliary information, so that for units with more than 100 employees the stratification by gender and age has been possible. For the units where the record linkage with those fiscal data was robust, a list of VAT code of their employees representing their structure was drawn and proposed to the units.

The final first level sample includes 19,535 businesses.

Expected numbers of second level units was 479,718.

The SES sample design on the public institutions consists of a census for units with 10 or more person in employment, from the S13 list (statistical Register of the Public Institutions, including Universities). This list includes about 1,100 units, it did not contain any information on the employment, that have been properly estimated. During the survey 113 units of the list (about the 12%) resulted to be out of scope and therefore were excluded.

The list of institutions belonging to the Education sector - except Universities - has been estimated to be of 13.030 statistical units by the integration of administrative and statistical archives. For the second stage sample, the integrations with data from the statistical archives supplied by the Eu_Silc and the Labour Force Survey has been used. Once the due number of units belonging to the second stage sample was established, one part of the set could be fully gained by the integration among the survey microdata and the administrative data, that allowed to estimate the full microdata on a subset of the final employees sample. At last, on the remaining part of the sample only the quantitative variable were available, gathered by the administrative and fiscal data, on which the imputation of item no response was run.

SES 2010 – Summary table: Main features of sampling design

The table below has been completed using the following explanatory notes:

• “1/1 for more than …. employees” represents the size (in terms of the number of employees) of the first level units which were covered exhaustively, i.e. 100% covered rather than sampled.
• “Sampling rate” represents the number of first level units in the sample survey expressed as a % of the total number of units in scope (i.e. in the population).
• “Response rate” represents the number of first level units whose responses have actually been used, expressed as a % of the number of first level units in the sample.
• “Coverage rate” represents the total number of employees in the first level units actually covered by survey expressed as a % of the number of employees in the units in scope (i.e. in the population).

[Not requested]

[Not requested]

[Not requested]

Grossing up

As regards the data covered by the direct survey, the final grossing up have been done adjusting the direct weights in each strata have to take into account the non response pattern.

Responding units have been post-stratified and the final weight resulted from the standard procedure implemented in Istat according to the calibration estimators. Numbers of business and employees in the frame provided by ASIA register (with reference year 2010) have been used as auxiliary variables as regards the private sector, numbers of business in the frame provided by the S13 list (with reference year 2010) has have been used as auxiliary variables as regards the public sector .

Final weights guarantee the convergence of the results to the known totals of the auxiliary variables of the following estimation domain:

• Subsection of economic activity (Nace Rev.2) and employees class (10-19, 20-49, 50-99, 100-249, 250-499, 500 and more)
• Subsection of economic activity (Nace Rev.2) and 5 macro regions (Nuts1 2005)

On the other hand, as regards the grossing up of the data related to the public part of the sector P, the procedure is completely different from the one used in the other context. The elaboration of the final weights at the local unit level takes into account the ones already elaborated for the same record for the survey from which they have been extrapolated, i.e. Eu_Silc and LFS. Distinct pondering has been obtained for the local unit (first stage sample) and for the employees (second stage sample).

[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 SES 2010 survey results to provide a important information for decision makers in national and European social policy and to establish reliable and harmonized comparisons of earnings and composition of earnings between European countries or regions. The Italian institute of Statistics (Istat) disseminates those information through three main channels:

• a public release of the main aggregates and indicators by a statistical report;
• the Istat Research data centre ADELE that assures a similar access to the one offered by the Eurostat safe centre (where also the privacy and the statistical anonymization of data are assured);
• as a MFR file specifically anonymized data for the research;

The main national users of the SES survey data are academics, the universities students, typically for any kind of degree and/or PhD thesis, field experts and sociological and economic research institutions.

[Not requested]

[Not requested]

[Not requested]

-

In the general statistical sense, accuracy denotes the closeness of computations (or estimates) to the exact values; it is defined by the size of possible gap between measurement and true but unknown population parameters. This possible gap consists of the so-called errors.

There are two possible kinds of errors:

1. SAMPLING errors related to the sampling design, measured by the relative error (coefficient of variation=CV)
2. NON SAMPLING errors related to other aspects of the data production process quantified by a set of specific indicators dealing with the effects of list status, total and item non response, editing and imputation criteria.

The SES2010 microdata on employees is the result of an integration of sources and modes, at the basis of which a designed based method has been established, that involves a two stage sample procedure. At the first stage, a sample and a census of the statistical units, respectively on the private and on the public sectors, have been implemented. At the second stage, a sample of employees has been drawn in both cases. Furthermore, a mixed mode strategy of collection and the integration of sources, between direct survey and administrative data, has been used. Hence, the indicators of accuracy about the sampling errors will be given with regards the whole domain of estimation, while the indicators of accuracy about the non-sampling errors will be given with regards that part of data collected through the direct survey, that covered most of the final domains but not all.

Variance estimation and therefore coefficient of variation have been calculated for the two interest variables “Average gross hourly earnings in the reference month” and “gross earnings in the reference month” providing data broken down by:

• NACE Rev.2 Section;
• occupation (ISCO-08 at the 1-digit level);
• age band (16-24, 25-54, 55-64,>= 65 years old);
• size class (10 – 49, 50 – 249, 250 – 499, 500 – 999, > 1000);
• NUTS1;
• ISCED.

Coefficient of variation refer to whole observed units.

Please refer to the attached document Coefficients of variation.

-

The target population of the survey is defined by the private businesses and institutions and public institutions belonging to the Section B-S Nace Rev. 2, except the Section O. In order to identify the statistical units to be surveyed the whole universe has been divided into three type of units, mainly according to the available information about the starting list of units. In this view, two lists were already available that could cover the whole private sector and a part of the public one; for the remaining part of the public sector of the Section P Education an ad hoc reconstruction of the starting frame has been done because a proper list was non available.

Hence, the three registers were:

1. ASIA (Statistical Register on Active Business) Register, year of reference 2009. It consists of a statistical Business Register resulting from the logical and physical combination of data resident in statistical and administrative sources (Tax Register, Register of Enterprises and Local Units provided by the Chambers of Commerce, Social Security Register, Work Accident Insurance Register, Register of the Electric Power Board) treated with statistical methodologies. It supplies official information about the economic activity and the average employment in the reference year. A sample of the private business has been drawn from this list (reference year 2009) and data were due to be collected only through a direct survey.
2. S13 list (Statistical Register of the Public Institutions), year of reference 2009, This is the list of units included by Istat in the S13 domain that is the “General government” institutional sector in European System of National and Regional Accounts, see ESA (1995). The information from this list does not supply anything about employment of the statistical unit. In this case, a census has been done in order to evaluate also the frame of the population according the number of employees, to identify the survey target population of business with at least 10 employees. Data were due to be collected through a direct survey. This list could not cover the institutions operating in the Section P, under the public control, (except the universities) since they have specific characteristics and are not included in the list S13.
3. the list of units operating in Section P, not included in the previous list S13, that had to be estimated. To this purpose, a specific model using administrative and fiscal data has been applied to estimate the target population of the statistical units of the public schools (local units). Furthermore, the complete universe of employees of this sector has also been obtained through the record linkage of administrative and fiscal data, from which the second stage sample has been drawn, optimizing the use of data on the same people derived form the statistical archives LFS and EuSilc.

A more detailed description of this ad hoc method can be found in Cardinaleschi, Spinelli (2012)[1].

[1] Cardinaleschi S.- Spinelli V. Integration of Administrative Registers for Statistics Eurostat”, European University SIS VPS , April 19-20, 2012, Rome.

As a consequence, the measure of any problem of misclassification can be related only to the first two registers. As regards the list on the private sector, the out of scope units added to wrong addresses and to change of status sum up to a final list error of 8.2% of total initially sampled units. This value is due to different type of error: a. out of scope units about the 1% of the sample; b. added to wrong addresses the 4.7% of the sample; c. change of status 2.5% of the sample.

In this regard, an update of the starting list has been made, in order to take into account the new units resulted from change on the units belonging to the starting list which percentage accounts to the 0.6% of the starting frame.

On the other side, as regards the list S13 the percentage of error list is higher, about the 14.2%, even though with a completely different composition in comparison to the one about the other list. In the case of the public sector, most of error list are due to the units out of scope with regards to the size class, indeed the 12% of errors are due to this reason (less than 10 employees) and only the 2,2% are due to wrong address and different status than the one declared in the starting frame.

[Not requested]

Measurement errors are the ones that occur at the time of data collection because of the survey instrument, the respondent, the information system, the mode of data collection.

A structured scheme of control of such errors has been established, first during the data collection with interactive editing rules on the web questionnaires. Indeed, on the web questionnaire the main rules have been implemented, first of all in order to avoid missing data on the mandatory variable, furthermore also the easiest kind of logical control were required to be satisfied.

 Afterwards, once the collection was ended, a further process of Editing and Correction has been used that followed several steps:

1. deterministic control in order to detect that error which solution could be given directly;
2. a selective editing procedure has been run, to detect influential data that could be potentially errors controlling for the variable per capita earnings. The identification of such data has been based on a statistical software produced by the Institute itself (Selemix, in R language) [1];
3. a complete check on all the variable, specifically the software BANFF has been used on the set of quantitative variable required on the employees, in order to identify inconsistencies among the data and to correct according to the principle of the minimum change and to localize the variable to correct with the best donor.

[1] Buglielli M.T., Di Zio M., Guarnera U., (2010). Use of Contamination Models for Selective Editing, Q2010, European Conference on Quality in Survey Statistics, 4-6 May 2010, Helsinki.

Unit response rate

First level unit (business and public institutions) response rates by economic activity, size class and macro-region have been calculated as “valid usable compiled questionnaires“ divided by “number of units being contacted”, hence it is evaluated only on the units on which the direct survey has been implemented. The total response rate was 43.9%.
Results are presented in the attached document Unit response rates according to the following list:

• 7 - Structure of Earnings Survey - year 2010 - Response rate by Nace Rev.2 division (two digit)
• 8 - Structure of Earnings Survey - year 2010 - Response rate by Nace Rev.2 division and size class
• 9 - Structure of Earnings Survey - year 2010 - Response rate by Nace Rev.2 division and macro-region (Nuts5)

Item response rate

Item response rate has been calculated for interest variables: monthly hours, monthly earnings, and annual earnings; the table below shows absolute figures and percentage.

Around 15.000 (6.1%) records has one missing value in at least one of those variables. More in details, 4.7% of the record had a missing response on the item concerning monthly hours, 5.7 on the variable monthly earnings and of 5.9% on the variable annual earnings.

Interest variables monthly hours, monthly earnings, and annual earnings - Item non response: absolute figures and percentage

The effective response rate taking into account only the valid returning questionnaire is about 44.5 (36.3% in SES 2006) As regards the units belonging to the first frame ASIA, the response rate assured the 54.3% of total expected employees (only on this frame is possible to do such evaluation, because it is the only one supplying an a priori information on the employees).

Handling of non-responses: Unit non-response

An over-sampling of 30% has been considered to prevent the negative effects of unit non-response in those strata where there was a response rate lower than 22% in previous SES round. So as to minimize the non response rate, special telephone lines and personalised email addresses were set up to help information providers in answering the questions. One reminder together with a new copy of the questionnaire was sent by mail two months later, while special follow-up of major enterprises has been carried out to ensure minimum response in “ critical cells”; nearly all business with 250 or more employees which had not responded received telephone calls.

A further step has been taken in this session of the SES, indeed administrative data gathered by the social security institution (INPS) have been used to estimate the variable about earnings and hours worked of a specific set of enterprises belonging to the register ASIA, that had been identified belonging to the most delicate empty stratum. As second step, the remaining variables have been imputed according to the Editing and Imputation design, considering them as item no response. The final result is: 272 firms estimated, for a total of 7.180 employees records estimated.

Finally, reweighting with post stratification model has been adopted to deal with unit non-response considering the employment measured by the ASIA business register (average of year 2010).

Item non response

Key variables on monthly and annual earnings have been estimated according to a deterministic model based on auxiliary variables available from Social Security data base (year 2010); auxiliary variables consisted in the average monthly and annual earning by position in employment in the business.

[Not requested]

[Not requested]

-

All item non responses have been imputed. The variable Gross earnings in the reference month has been imputed in 5.7% of cases. For more information on non responses see section 6.3.3.

Model assumption errors may affect the editing and imputation rules. 

[Not requested]

[Not requested]

[Not requested]

[Not requested]

The updated microdata to Eurostat were officially sent in September 2012 while aggregate estimates were released nationally in February 2013.

[Not requested]

[Not requested]

The statistical units have been contacted by a letter, sent on the 11th of September 2011, that was an invitation to compile the given questionnaire on line, where it could be possible to find all the necessary instructions to proceed with sending the data via web. All the legislation and the instruments to assure the security of the web questionnaire were given (proper user name and password). Afterwards, an official recall has been made to the not respondents, by sending them the whole paper questionnaire as a further mode to supply the requested data. Data collection went on till June 2012.

The delay was mainly due to several facts, first of all the complete re-ingeneration of the survey involved a late starting of the survey; then some problems in managing a new instrument has caused, in certain periods, some problem to be solved, that in a second session would be already performed.

In order to optimise the contacts and hence the quality of the data, special follow up of non respondent units were implemented and a strong back office job has been assured by all the personnel involved in the survey, in order to help out the statistical units in any conceptual or practical problem they met.

Official delivery of updated microdata to Eurostat dated September 2012.

[Not requested]

Statistical units

Reporting units consisted of enterprises and public institutions; results relate to local units. No restriction on NACE coverage was performed. All mandatory variables have been surveyed and provided
to Eurostat.

[Not requested]

The SES 2010 is broadly comparable with the previous edition of SES (2006) on private sector. 

[Not requested]

[Not requested]

[Not requested]

An analysis on links and coherence is presented in the following part, by comparing the variable ‘gross annual earnings in the reference year’, expressed per employee, and the variable ‘wages and salaries’, per employee, of the NA.

Ideally, they belong to the same statistical system internally coherent, but differences in coverage, statistical unit used and concepts in defining the items may produce some differences. This analysis aims to identify discrepancies in the data provided and to evaluate the causes of potential differences.
National Accounts (NA) are the implementation of comprehensive and consistent accounting techniques to quantify the economic activity of a nation, combining data from many statistical source. As a matter of fact, techniques of data combination and integration vary, depending on the compilation approach, the statistical source available, the particular account in the system of accounts and the timeliness of data release and other factors. NA provides a coherent and consistent macroeconomic picture of the economy as they broadly present the production, income and expenditure activities of the economic actors (corporations, government, households). All stages of production, distribution and use of income are covered, focusing on aggregates by industry and institutional sectors. The comparison between the variable “gross annual earnings expressed per capita” in the reference year supplied by the SES and the variable “wages and salaries per employee” estimated by the NA is presented (Prospect 6). To have a meaningful comparison the national Account data have been chosen to have the same coverage of the SES. The NA data refer only to regular economy and excluding less than 10 employees firms but it has not been possible to exclude public and private institutions units with less than 10 employees. Therefore in the last column of the prospect 6 there is the percentage incidence of the employment related to units with less than 10 employees which can help to understand differences deriving only from this coverage factor.

Taking into account the differences in definitions in the greater part the Nace Sections the coherence is very high. High difference arise in Sections I, N, P, Q and S. But it must be taken into account that the Sections P to S have a high incidence of units with less than 10 employees. In I the tips and remuneration in kind make explain the differences. In fact National accounts labour cost aggregates, differently from the SES of information available, include Tips and remuneration in kind, the latter is included entirely only in structural business surveys wages and salaries. Wages and salaries in large enterprises and labour cost series include only fiscally taxable in kind remuneration. This can heavily affect the comparability of specific industries (for example, hotels and restaurants for tips).

The data are broken down by NACE Rev 2 Section.

SES vs. NA 2010, NACE rev. 2 Sections B-S (O excluded), annual earning per capita – euro and % diff.

[Not requested]

[Not requested]

Online dissemination of SES 2010 main results will be soon available in the Istat web site www.istat.it (http://www.istat.it/it/archivio/83362).

Furthermore, as already mentioned, the data will be available through the secure Research data centre ADELE and on the data warehouse I. Stat, that is directly available on the Istat web site www.istat.it aimed at the production of tables with possibility of multiple choice.

[Not requested]

Special anonymized microdata files for research purpose will be disseminated in the future months, according to Reg. 831 on confidentiality.

No results are sent to reporting units included in the sample.

Every way of diffusion of data, above mentioned, includes the proper information on the metadata and on the methodological scheme and the glossary of every definition implemented by the survey.

Dissemination of national data on quality by Eurostat

No national data on quality shall be published without the agreement of Istat.

[Not requested]

[Not requested]

[Not requested]

[Not requested]

[Not requested]

Improvements proposed for Structure of Earnings Survey 2014

Further development of the use of already existing archives, both administrative and fiscal and from statistical surveys, have been studied and planned for the next session. The aim is to optimize the integration of other sources in order to achieve as much information as possible on the available information on the economic variable, from administrative and fiscal data, together hast is possible to exploit about the personal characteristics of the employees, where possible from the other statistical sources. Once all the set of information is established, it is possible to study a new direct survey aimed only at the reaming part of the needed information and with a much smaller sample of enterprises and institution to contact. This project would imply a the strong reduce of the statistical burden on the units and of the cost for the institute.