Instituto Nazionale di Statistica - ISTAT

National Statistical Institute of Italy - ISTAT

Direzione centrale per le statistiche sociali e il censimento della popolazione
Dipartimento per la produzione statistica

Social Statistics and Population Census Directorate
Statistics Production Department

The SES 2014 is based on the Council Regulation (EC) No 530/1999 concerning structural statistics on earnings and on labour costs and the Commission Regulation 1738/2005. This quality report is compiled in accordance to the Commission Regulation 698/2006.

The target population is composed by all the enterprises and institutions belonging to the Private and Public sectors with at least 10 employees in the NACE Rev. 2 sections B to S.

With the current edition, for the first time, Italy provides also data related to section O (even if the extension to this section is still considered optional).

Statistical classification of economic activities in the European Community. NACE Rev. 2.

International Standard Classification of Occupation ISCO -08

International Standard Classification of education ISCED 2011

.

Enterprises and institutions belonging to the Private and Public sectors with at least 10 employees in the NACE Rev. 2 sections B to S.

In addition to individual characteristics of employee (sex, age, occupation in the reference month, highest level of education and training, contractual working time, length of service in the enterprise, type of employment contract); the main variables provided are related to working time (number of hours actually paid during the reference month, number of overtime hours paid in the reference month, annual days of holiday leave), and to information on earnings (gross earnings in the reference month and  year , annual bonuses and allowances not paid at each pay period, earnings related o overtime, special payments for shift work).

Employees and enterprises/institutions

The population of employees covered in the SES are those who received remuneration for the reference month (October), as requested by the  Regulation, in the enterprises and institutions belonging to the Private and Public sectors with at least 10 employees in the NACE Rev. 2 sections B to S

 Data cover the entire  country and aggregate estimates are disseminated at national level and NUTS 1 level

The reference year is 2014 and the reference month is October

Not applicable

The statistical processes for the Private and Public sectors are entirely different. In what follows brief explanations of both are reported. However, in different ways, they are based on intensive use of administrative and register data. It is then useful to briefly recollect the data situation at Istat.

Registers and administrative data

In recent years Istat has made considerable progress in exploiting administrative sources. As for the private sector, the processing of fiscal and social security sources has allowed the production of the statistical Employment Register with information at worker level that is also the micro base of the Business Register for what concerns information on employment. Regarding the employees it has been produced using, as a main source, the individual Social Security declarations remitted monthly by firms since 2010, for each of their employee. It contains information on characteristics of the employee, the job and the firm, with a Linked Employer Employee Data (LEED) structure.  The new RACLI Wage register is an extension of the Employment Register to a set of variables on wages and paid time at employee level, covering the entire population of employees and firms of the private sector. From 2014 the employment register also contain information on education derived from the the latest census information updated with data obtained by the Ministry of Education.

As for the public sector, the situation is still of work in progress toward the production of Statistical register. However, the collection of administrative data has been progressing and it has been possible, mainly for the purposes of of SES 2014, to produce a prototype of employment wage register at individual level for the public sector. Its construction has implied the use of different administrative sources at individual level (the social security declarations for the public sectors employees, payroll data for the employees of public schools and some central inistitutions -such as ministries-) and at aggregate level (the data from the Cost of Personnel of Public Administration -Conto Annuale-), a take all survey run yearly by the Ministry of Finance with detailed questions on number of employees,wages and its components, days of absences from works etc… Information on Education is obtained in the same manner as for the pirvate sector. The use of the combination of these sources has allowed to cover for the first time section O of the Nace rev 2. 

Private sector: Data sources and integration

The RACLI register does not provide all the informations requested by the SES regulation. In this situation a direct survey has been considered still necessary.

However, thanks to the availability of the RACLI wage register[1] the Structure of Earnings Survey 2014 has been completely redesigned, with respect to the 2010 edition, in terms of both methodological and technical aspects in order to increase data quality and to reduce the response burden on enterprises. The redesign aims at a mutual integration between survey and register, with the register data that assists the survey in some phases of the process, and the survey that has been planned in order to check the register data and provide details not available in the the Register. A new type of integration has been experimented introducing several solutions ex-ante, as a new sampling method and a new questionnaire which has also been completely redesigned. For all these reasons the private sector process is a mixed register-sample survey process, and the final data come from the integration of the two sources.

 Private sector:The sampling design

The availability of an employee level register has triggered, for the edition 2014, a study to modernize the sample. Until the SES 2010 edition the sampling design consisted in a two stage process. In the first stage a sample of enteprises was selected from the Business register. Then, Istat instructed each of the enterprises to select a sample of workers from their books according to a systematic rule. The number of employees to be sampled for each enterprise was predetermined and increasing with the enterprise size. For the 2014 edition the study on the sampling design at first explored the possibility to use a single stage design, in which workers are selected directly from the register. Due to the very large efficiency gains associated with this design it might ensure a huge reduction in the number of the sampled employees. After some experimentation this design was abandoned since the  number of enteprises, that in this sampling scheme is random, that were involved was too large and has been judged not sustainable for the collecting structure. The decision was to revert back to a two stage design. However, compared to the past, the presence of the Register has allowed to stratify the enteprises on the basis of employee level variables and to drive the allocation among the stata with the wage variables available. Since both the stratification variables and the allocation driving variables are correlated to the target earning variables this has ensured efficiency gains and to limit the sample size in terms of employees. In this design to enterprises of the same size may be requested a different number of sampled employees if the wage variability is quite different. Moreover in this case the employees are selected by the NSI. This has ensured that the enterprises could not cheat on the sampling rule choosing, for instance, not to provide information on some kind of workers (e.g very high wage employees, managers, etc…).

The sample of the private sector has been studied to be representative of the population of workers active even for very short periods of the year 2014. In this way the sample fulfills, on one side, the requirements of the SES regulation that requests data only for the workers paid for the entire month of October and, on the other side, produces auxiliary information for the Register also for workers present in other subperiods of the year. 

In details, the design of the two stage sampling is as follows.

In the first stage, the enterprises are selected within strata obtained by crossing 2 digit Nace sectors, size classes and NUTS1 regions. The entperprises with at least 250 employees belongs to take all strata and have been selected with certainty. For the remaining enterprises the allocation has been performed using the Bethel  multivariate technique that determines the optimal number of units per strata by minimizing the expected error of given target variables. In this case the driving variables used for the allocation are the enterprise average of the hourly wage and the total wage paid to each worker  respectively in October and in the Year.  The allocation has been obtained by constraining the precision of the estimates of these target variables to a maximum CV of 3.6% per estimation domain. The sampled enterprises are 24,465 out of the 168,290 entperpises with at least 10 employees contained in the Business Register ASIA.

In the second stage, the employees were selected in strata built by splitting up each enterprise into groups defined by Nuts1 localization of worker, working time, Blue collar/white collar. The same allocation technique has been used using as target variables the hourly wage of October and the annual wage at employee level. The overall second stage sample size has been of 212,722 sampled jobs out of 10,546,683 paid for at least one hour in 2014. A further adjustment step has been performed by forcing the final sampling size to vary from a minimum of 3 to a maximum of 500 of workers per enterprise.

The final sample of workers that have been paid for the entire month of October is 136650 out of 6492413 in the population.

Public sector: Data sources and integration

The public sector data  is derived ony from available registers and data sources. Referring to public institutions the frame list is built using the latest available list of Public Institutions (S13 list). Within the Education sector a complete list of schools (and the personnel working for each) has been compiled starting from the archives of the Ministry of Public Education. The data at the individual level on the core variables are derived for all employees from the social security monthly declarations for the public sector (module DMA 2). More individual level variables are derived for the sector education and central institutions -such as ministries- from payroll data.

The data necessary for computing more detailed variables are obtained through models using aggregate, altough with a deep breakdown by qualifications, information derived from the Conto Annuale.

The data on occupation according to ISCO are obtained by transcoding the available information on DMA.

Public sector: The sampling design

The entire statistical process is performed on the population of public employees. From it a sample of employees is selected to provide the necessary file of microdata.

The sample of employees for the public sectors has been planned to mimic the design of the private sector. In this case however only a sample of workers paid for the entire month of October has been drawn, resulting in a sample size of 52.571 workers out of 3.121.614 in the target population.

 The data collection occurs every four years.

As regards the data collection phase, the SES data for the private sector has been gathered mainly through a web based questionnaire (CAWI) with some built in checks (first level checks). In addition to the web based questionnaire, in order to make the task easier, especially for the employers with a large number of sampled employees, the enterprises could upload into the electronic system a data file in a format compliant with the questionnaire structure. In order to avoid measurement errors, the respondents who adopted this secondary mode were asked to check the data inserted using the web application with the rules built in it.

The questionnaire redesign

With respect to the previous edition, the questionnaire was redisegned to complement the informatuion available in the register. 

In fact, in the register there are some variables that fulfill perfectly the SES definitions and others that are the core components to calculate what is required. Some more detailed variables are not available in the register.

The new design aimed at building a bridge between the available data and the statistical requirements and at reducing the statistical burden. Two main innovations have been introduced. First, all questions regarding information available from the register and not necessary for checking purposes have been removed. In this way the questionnaire has not included any question at firm or local unit level such as the geographical location of the local unit in which the worker is employed, economic activity, and the enterprise size. This has allowed to eliminate altogether the section at firm level. Moreover, in the employee level section, the questions on sex, age and the number of paid weeks in the year have been removed.

The second innovation has been to prefill some information available from the register with the possibility for the respondents to confirm or correct the value if considered wrong. Contractual working time, share of full-time hours, the number of workable hours in the month, the social security wage for the month of October and for the entire year are the prefilled variables. Not all cases have been prefilled due to uncertainty in the measurement in the register. The pre-filling not only allows to build questions that bridge available data with the information required, but it provides solid benchmarks to detect errors in the sub-items.

Finally the variables required by the Regulation not available in the Register, such as  overtime hours and earnings and special payments for shift works, have been added to the questionnaire.

The contact with the enterprises

The contact with the enterprises has adopted the Certified email (PEC) system, introduced for the first time with the LCS 2012, to send the recruitment letters and up to three reminders to solicit the filling of the questionnaire. Moreover, during the filed work  communication with the enterprises, through phone calls and emails, was performed by a two level contact service. The first level operated by the external contact service dedicated to the Istat Web Access for the Enterprises (Portale delle imprese)  supported the respondents in the registration phase and on the use of the Access platform and provided information on the deadlines, the obligations etc.. The second level, operated by internal experts, provided support on technical issues regarding the contents of questionnaire.

Editing and imputation procedures

Since also the SES 2014 for the private sector is a mixed Register-Survey process it is important to describe how the editing process has been carried out. For the private sector most of the variables have been derived from the RACLI register. Some of them have been used to prefill the questionnaire and could be modified by the respondent. The variables not available in the registers have been moreover added to the questionnaire.

All the variables have then been passed to the editing and imputation procedures and edited when considered incorrect. The variables derived from the register have been edited only in very few cases when considered clearly incorrect. For most of the cases, instead they have been considered correct and been used as pivotal variables for the E&I procedures in several ways. As for the respondents they have been used in the edit rules that imply the consistency between a core component (from the register) and a subcomponent (from the survey), in the formation of more homogenous edit groups and to provide the matching variables in the minimum distance imputation.

A Fellegi-Holt type of procedure has been used to identify the errors. 

In this framework, the records for the employees belonging to non respondent enterprises, have been compiled on one side from the variables available from the registers and on the other with values imputed through minimum distance donor methods. Also in this case the register variables have been  used as the basis to impute the remaining required information.

 All these procedures have been applied to the questionnaire scheme. Starting from it, a reclassification procedure has derived the variables requested from the Regulation.

 As for the public sector a set of checks has been performed during the processing and compilation of the data and virtually no record has been edited. 

A full file covering both the private and the public sector has also been used to reweight the sample at the end of the statistical process. This has allowed to calculate weights that produce estimates aligned with the final version of the registers, while at the time of the sampling selection, only a provisional version of the RACLI register was available. Moreover it has been possible to use, in the calibration procedure, variables not available in the provisional version of Racli such as the level of education.

More specifically, after the editing and the imputation phase, only the subsample of workers paid for the entire month of October has been reweighted using known totals from the wage registers. The direct weights have been calibrated to reproduce the totals of the number of jobs, the number of paid hours and earnings for the population of workers paid for the entire month of October. The calibration groups have been formed by crossing sectors of economic activity (somewhat in between the division and sections of NACE rev2)  and the sex of the worker with, alternatively, the size class of the enterprise, the NUTS1 region where the employee works, age of the worker in classes, the working time and ISCED groupings.

[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 core users of SES 2014 data are analysts, including researchers from public and private institutions and from universities, and journalists.

At the moment, Istat collects no data on the level of satisfaction, gaps and redundancies in the information provided.

[Not requested]

[Not requested]

-

The main sources of error in the SES process are the sampling errors and the measurement errors. On the contrary due to a quite high response rate, an imputation procedure  of non responses and a calibration procedure based on very correlated auxiliary variables should minimize the impact of biases due non responses, at least for the estimates at aggregate level.

The Coefficient of variation of the monthly earnings and the hourly earnings are presented in the table 1.

Table 1. Sampling errors for selected domains

-

The target population on which the survey has been designed is defined by the employees of the private businesses and institutions and public institutions with at least 10 employees belonging to the section B-S Nace Rev. 2. including the section O.

As said before since the public sector data are register based, there is no list or coverage errors by definition. As for the private sector, although a direct survey was issued, no information related to the target population (economic activity sector, size of the firm, status of employees) was asked to the firms, since in this production framework this kind information from the register are reputed the most reliable. The only causes of coverage errors may arise since the sampling frame is a provisional frame. However the differences due to the updating of the frame are negligible. Moreover, since the weights of the sample are calibrated to the final version of the frame, any possible misalignement between the selected sample and the final frame should be adjusted with the reweighting procedure.

For the type of statistical process used it is considered that the data has no over-coverage.

[Not requested]

Measurement errors are defined as errors occurring at data collection time, while processing errors are those occurred in post-data-collection processes.

While little information is gathered on processing errors some considerations can be drawn on measurement errors. First, we define measurement errors those considered such by our Editing and Imputation procedures. As above sketched, as for the public sector, all variables have been derived from  registers directly or through estimation methods. In this section thus we focus on the private sector.

From what has been said in the Data validation section, some of the target variables are derived completely from the register, some completely from the questionnaire, and some, those on total earnings and hours paid, are composed from a part derived from the register and a part derived from the questionnaire. This implies that there are very different item imputation rates among the variables. In particular, the variables with the highest imputation rates are those completely not available in the register or those composed partially from the register and partially through the survey, since it is on them that the imputation of non respondents impacts. However for the second group, since the part not available from the register is very small compared to the total,  the imputation change the available value only slightly.

For what has been said before non response errors affect only the process for the private sector. 

Since some of the variables, those derived from the register, are available for the entire sample, the unit response rate assume the meaning of percentage of non respondents to the direct survey not that of units whose all values must be imputed.

Unit response rates are calculated, following the indication of regulation n. 698/2006 as follows.

r=100x (number of in scope respondents / number of in scope sample units)

The units in the sample that are deemed in scope are those belonging to the of the new statistical register on wages (RACLI) with data on the entire population of employees of the private sectors referred to 2014.

The non-response rate in terms of enterprises is 35,7%, while in terms of  sampled employees is about 27,0%. These non response rates presented are related to employees who received remuneration during the reference month.

In table 2, unit non-response rate in terms of enterprises and employees are showed

Table2. Private sector: unit non - response rates in terms of enterprises and employees.

Since the questionnaire could not be sent even if only a single variable was not filled in, there are no item non responses for the respondents and the item non response rates correspond to unit non responses.

-

The Item imputation rate is calculated as the ratio of values imputed for a specific variable over the total number of values for that variable.

For the variable Gross earnings in reference month, code B42 in SES classification, the item imputation rate is 9.5%.

The overall imputation rate is calculated as the ratio of values imputed for any SES mandatory variable over the total values for mandatory variables.

Among the 21 variables measured at the level of employees it amounts to 15%.

Model assumption errors may affect editing and imputation rules. The imputation of  the records for the employees belonging to non respondent enterprises, have been compiled on one side from the variables available from the the RACLI register  and on the other with values imputed through minimum distance donor methods

[Not requested]

[Not requested]

[Not requested]

[Not requested]

-

The data appeared for the first time on Eurostat database in October 2016 and have been published with a national press release on 30 December 2016. Thus the length of time between the end of the reference year (2014) and the first publication of data is of 20 months as for the Eurostat release and 22 months as for the national release.

[Not requested]

[Not requested]

The release of data was not announced in any release calendars for national purposes. As for the EU deadlines Italy delivered the data at the end of June 2016 respecting the Regulation deadline (30 June 2016). However further data trasmissions have been necessary due to review in a data subset. This delay was due to mainly to some difficulties in adopting the new data gathering system, that have caused a remarkable delay in the start of data collection. Due to these delays the sending of the cover letter was completed on 12 January 2016. The data collection has been officially closed on 24 April 2016 resulting in a period of fieldwork of less than 4 months. With  respect to the previous edition of SES the data collection period has been reduced of more than half and at the same time the response rate has been largely increased. However the delay in  the start of the data collection, together with the fixed deadline, has reduced the period of time available for editing, imputation, estimation and validation.

[Not requested]

-

The SES 2014 complies with the standard set up on the Council Regulation (EC) No 530/1999 concerning structural statistics on earnings and on labour costs and with the definitions of variables adopted in the Commission Regulation 1737/2005. The sampling unit in the first stage is the enterprises, but the register information allows to localize the worker in the region in which she is employed.

[Not requested]

The SES 2014 is broadly comparable with the previous edition of SES (2010) for what concerns large breakdowns on earnings variables. However, since this edition has introduced a brand new questionnaire, a new sampling design and the procedures of E&I and estimation have been thoroughly modified from the previous edition some of estimates may have some problem of comparability.

An area with an issue of comparability is the estimates on the number of employees. The availability of the registers has allowed for this edition to better comply with the regulation and to provide weights that sum up to the target population of the regulation: the employees paid for the entire reference month (October). This was not possible in 2010. The result is a large decrease in the number of estimated employees.

Another area in which there might be problems of comparability is the public sector since for this edition the entire process is based on registers and administrative data. Together with the sources all the methods of derivation of the target variables have been changed.

In this edition of SES the hourly earnings (SES code B43) has been calculated, for the employees with a standard monthly payroll (that are the large majority of employees), dividing the monthly earnings (B42) by a measure of the hours that has been calendar adjusted, and not for the number of hours paid in the month (B32). This is allowed by the regulation that requests only that the hourly earnings be consistent with the ratio of earnings and hours, not equal to that ratio, and compliant with the implementation arrangement 2014 that allows for a discrepancy between the hourly earnings and the ratio of earnings and hours of up to +-10%. At the same time this way of providing the data is a) compliant with the regulation that states in relation to the Number of hours actually paid during the reference month that “What is required here is the number of hours actually paid during the reference month, not the number of hours in a standard working month”; b) more respectuful of Italian payment system that for the category of employees above mentioned establishes that the monthly earning does not depend on the calendar c) more in line with the spirit of the regulation that requires, according to us, a hourly earning that should not be dependent on the calendar. This innovation should ensure a better comparability of the hourly earnings between SES 2010 and SES 2014 despite the fact that October 2014 has two workable and payable days more than October 2010.

The innovations in the use of registers and the sampling design should have improved the estimates for some sectors. For instance in the Nace Rev 2 Section R, the data now includes athletes and sports staff that before was missed. This has implied a steep increase in the earnings due to remuneration of very high wage sportsmen, In nace rev 2 section N the data includes temporary workers hired by temporary work agencies.

[Not requested]

[Not requested]

[Not requested]

This paragraph reports on the comparisons between the SES variable ‘Gross annual earnings in the reference year’, expressed per employee, and the National Account variable ‘Wages and salaries’ per employee. The data are showed by NACE Rev 2 Section in the table 3.

Wages and salaries presents values lower than Gross annual earnings in almost all the Nace sections. The main reason is the difference in coverage (all the enterprises in NA; enterprises with 10 and more employees in SES). Being the remuneration increasing with enterprise size, SES estimates appear higher in almost all NACE sectors. A further factor explaining the difference in favour of SES is the fact that the NA statistics include also irregular employees which have compensation much lower than regular employees. The difference appear infact less relevant in sectors (as D) where the share of small units is lower and the percentage of irregular workers too.

Another reason could be that the reference population in SES variables is related to employees that have actually receveid remuneration in the reference month while NA estimates are related to all employees including less stable ones with lower remunarations.

Table 3. – Gross annual earnings in the reference year (SES) and‘wages and salaries per employee (NA).

Compensation of employees per employee: NA vs SES.

[Not requested]

-

[Not requested]

Beyond the Eurostat database, aggregate data are disseminated as follows.

A first publication of SES 2014 data integrated with Racli data limited to the private sector is disseminated with a press release in the series Statistica report on December, 30 2016. The report is released with a number of tables in spreadsheet format and a methodlogical note. They are available at the following internet link: http://www.istat.it/it/archivio/194951 .

An e-book focused on detailed analysis concerning earning inequalities based both on registers and the survey data is scheduled by July  2017.

 .

More detailed data are going to be released through Istat data warehouse I.Stat.

[Not requested]

As for the micro data they will be released, anonimyzed, by Eurostat as SES scientific use files. Moreover access to confidential data will be ensured through the Eurostat Safe center. Similar accesses will be ensured at national level, by Istat.

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

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

The availability of this quality report is advertised in the methodological note published with the first publication of the data.

[Not requested]

[Not requested]

[Not requested]

-

[Not requested]

[Not requested]

Improvements proposed for Structure of Earnings Survey 2018

Further development in exploiting both administrative and fiscal sources, is planned for the next edition. 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 variables, from administrative and fiscal data. This project would imply  the strong reduction of the statistical burden on the units and of the cost of the data collection and processing.