Maltese National Statistics Office (NSO)

Labour Market Statistics

Labour Market Statistics

National Statistics Office

Lascaris

Valletta CMR 02

Malta

This is the fourth time the Maltese National Statistics Office (NSO) is carrying out the Structure of Earnings Survey (SES).

The SES 2014 was based on Commission Regulation 1022/2009 of 29 October 2009 and covered NACE Rev 2 Sections B to S and units employing at least 10 employees.

This report is intended to cover the following items as per Commission Regulation 698/2006:

• Relevance
• Accuracy

- Sampling errors

- Non-sampling errors

• Punctuality and timeliness
• Accessibility and clarity
• Comparability
• Coherence

The economic activity is coded in NACE Rev. 2 (General industrial classification of economic activities within the European Communities).

Occupation is coded according to the Commission Recommendation of 29 October 2009 on the use of the International Standard Classification of Occupations (ISCO-08).

Information on the 'Highest successfully completed level of education and training' variable is classified using the International Standard Classification of Education, 2011 version (ISCED11).

The statistics cover all economic activities defined in NACE Rev. 2 sections B to S including Section O. National information for section D was not provided since in 2014 there was a change in the operations of the national utility company also implying a reclassification in nace from section D to section N. During that year, the number of persons employed under section D was less than 10 and as a result the unit was not eligible for the SES 2014.

The enterprises included employ at least 10 employees and the size classes (corresponding to the number of employees) available are 10 to 49, 50 to 249, 250 to 499, 500 to 999 and more than 1 000.

An employee is defined as a person who has a direct employment contract with the enterprise or local unit and receive remuneration, irrespective of the type of work performed, the number of hours worked (full or part-time) and the duration of the contract (fixed or indefinite).

The statistical unit is the enterprise and the employees working within the selected unit.

The SES 2014 statistics refer to enterprises with at least 10 employees in the areas of economic activities defined by NACE Rev. 2 sections B to S including Section O.

Malta (NUTS 3).

2014

not applicable

Sampling methodology

The Sampling Frame

The sampling frame for this survey was the Business Register which is maintained by the Structural Business Statistics Unit within NSO. This register contains data regarding legal units which are recognized as having autonomous management and an independent accounts system at NUTS 1 level. In this regard, the target population for SES could be chosen from this database.

The total number of enterprises operating in NACE Sections B to S employing 10 or more was 2932. According to Commission Regulation 1022/2009, SES data has to be collected for enterprises operating in NACE B to S split in the following size classes: 10 to 49, 50 to 249, 250 to 499, 500 to 999 and 1000+.

After stratifying enterprises by NACE and size class a sample of 1709 enterprises was chosen. 

The file attached below includes two tables which illustrate the distribution of the sample and the probability of selection for each strata.

Every four years.

A custom made application which enabled the automatic emailing of questionnaires to the different respondents along with pre-filled variables for the selected employees was used. 

The SES questionnaire was made available in excel version since this software is widely available and since it facilitates the copying and dragging of information for different employees. In cases when emails of enterprises were not available, the questionnaires were sent by post.  These units were given a deadline and following the elapse of the deadline a number of interviewers were employed to collect SES data for non responding units.  This was especially necessary for units which are key players in their respective sectors and which consequently determine the develpment taking place within the sector in which they operate and hence influence the representativity of the results.

The NSI also provided additional assistance to respondents mostly via telephone and email.

To minimize processing errors, each incoming questionnaire was thoroughly checked by trained statisticians using a number of validations. These validations included consistency checks between information provided for the reference month and information given for annual earnings. Moreover any NACE coding which had to be carried out was to be in line with the Business Register classification. In terms of processing errors emanating from data entry procedures, NSO split this task in two. For data which was provided by respondents in soft copy format, information was directly uploaded into the data entry programme, thus minimizing any data entry errors. For manually entered questionnaires inbuilt validations were applied to reduce data entry errors.

not applicable.

not applicable.

Refer to concept 4.2.

The SES provides a unique opportunity by which users can be provided with data on labour costs, as reported by the employers themselves. Data on earnings derived from the employers themselves is generally deemed more reliable than that collected from employees. Information derived from the SES is based on a large sample of employees, and therefore high accuracy of results is expected.  

On the other hand, information relating to education levels is of less quality when sourced from enterprises since the individual would be more accurate on this matter.  Another limitation of this survey is that it does not cover micro-enterprises, which may have different earnings patterns when compared to larger units. 

Main users of this data include:  (1) International organisations (such as Eurostat, UNESCO, OECD, EU's Directorate General for Employment), (2) Public Entities (such as Ministries, Authorities) (3) Private entities (research organisations, unions, businesses), and (4) Research Institutes (5) Market Research Companies (6) Universities (7) Individuals.

No user survey to determine the needs of SES users has been carried out.

All requirements of the regulation are met. 

refer to 5.3

Every effort is made to reduce non-sampling errors, nevertheless a small element of these errors is inevitable in all variables. These include: 

- Recall Bias

- Data Entry Errors

- Response Error (definitional differences, misunderstanding... etc.)

Probability sampling - Bias

Chapter 3.1 Source data provides information on the sampling methodology used for this survey. 

The sample was determined into two stages. In the first stage NSO identified the enterprises which were to be included in the SES, (1709 enterprises). In the second stage, the Office identified the names of employees on whom SES data for Part B was to be collected. This method was adopted for those companies where the Office managed to obtain a match between the Business Register references and the company references available at the National Public Employment Office (PES).

For those enterprises which were not provided with a list of employees to cover in the SES and thus had to choose the sample of employees themselves, a set of instructions for the selection process were provided in order to minimize any bias.  In addition, employers were encouraged to provide information for all the variables using internal data from their databases or registers.  Whenever such data was not available because it was not compiled, employers were encouraged to provide estimates. NSO believes that since employers have a better and more in depth knowledge of their enterprise, any estimates are bound to be of better quality if provided by the respondents themselves. 

For enterprises which had a sample of employees identified, NSO provided instructions that if an employee no longer worked for the company s/he had to be replaced, preferably, by another employee working in the same occupation.

Each completed questionnaire has in addition been analysed and various consistency checks were applied at a micro-level. Furthermore, if data provided was not deemed to be of sufficient quality or did not make sense when compared to other information available, employers were contacted by telephone or email in order for them to clarify estimates and figures provided.

The Office also tried to reduce respondent burden by obtaining a number of variables from administrative sources for those units where a match between BR references and PES references was possible. The variables obtained from administrative sources were the following:

• Form of economic and financial control (Variable 1.4)
• Sex (Variable 2.1),
• Date of Birth (used for Variable 2.2) and
• Date of Entry into Service with the enterprise (used for Variable 2.6) 

For this wave of SES, the Office also made use of other administrative sources, particularly for sampled public sector enterprises. In the case of these units all information was retrieved from three different sources:

• the national PES
• a national database for public sector employees which is administered by the Management Personnel Office
• the Inland Revenue Department

Since information for these public sector units was going to be obtained from administrative sources, a decision was taken to keep all employees working in these units and do away we a sub-sample selection. Hence in terms of grossing up procedures, each employee represented him or herself.

Important to note that in the majority of the cases (80%), the transmission mode of the questionnaire was through email correspondence.

Non-Probability sampling

Probability sampling has been used for SES and therefore there are no non-probability errors.

The SES was carried by the 2-stage sampling design. In addition, sampling proportion of employees was not the same for all enterprises since the Office obtained full data pertaining to public sector employees. As a result the majority of public sector employees were surveyed. The following table illustrates the required employee sample size according to the enetrprise size class:

Probability sampling

This section of the report presents a number of tables on the coefficient of variation for variables identified in EC Regulation No. 698/2006. One should note that Estimation of variance is not taking into account the sampling methodology since we are assuming Simple Random Sampling but it is taking into account the variance of weights. The coeffcient of variation is further adjusted by the finite population correction factor. The following link gives more information on the theory and formuale used to compute the indicator : http://www.stat.wisc.edu/courses/st224-iltis/notes8.pdf

Please refer to the attached document Coefficient of variation.

refer to 6.3.1

Misclassification errors

Coverage errors which have been encountered in SES include errors relating to misclassification and over coverage.

Misclassification of NACE

Misclassification errors refer to incorrect NACE classifications which were assigned to units present in the target population. The table below provides the percentage distribution within each NACE division before and after the data collection process.  A total of 6 enterprises were misclassified.  Of these none had to be excluded from the sample since they all operated in economic activities which were nevertheless being covered for the purpose of SES. The remaining misclassifications were corrected before grossing up the data to represent the whole population.

Misclassification of size class

Another aspect of misclassification concerned size class. The following table indicates the extent of this misclassification. A comparative analysis is provided showing the sample classified by the six size classes prior to data collection and after information was returned to the Office.

One can note that a number of units resulted to be ineligible for SES since they employed less than 10 employees (refer to footnote 1).  These units were excluded from the sample, whereas the rest of the units which had been assigned a different size class prior to the data collection were reclassified and weights were worked out accordingly.

Over-coverage errors 

Over-coverage errors found in SES mainly related to misclassified units which were not within the scope of the survey or units which were no longer active during the reference period identified for SES.  In this regard, as highlighted above, a total of 107 units were excluded from the initial population because they had less than 10 employees or because they ceased to operate before October 2014 which was the reference month for the SES.

To correct for this error, NSO reclassified units in the categories in which they were actually operating in the reference period, excluded ineligible units and applied grossing up factors accordingly.

[1] Includes enterprises which were ineligible for the survey either because they employ less than 10 employees or because enterprises ceased to operate before the reference month for the SES (October 2014).

refer to 6.3.1

not applicable

NSO tried to minimise measurement errors during different stages of the data collection.

Accounting of measurement errors from the questionnaire 

NSO’s initial but very important objective was to have a questionnaire which was easy to understand by respondents without creating excessive respondent burden.  For SES variables the Office opted for a combined questionnaire for all sampled employees.  This was deemed to be a more suitable option for the local context since employers could fill in data on various employees simultaneously.

The final version of the questionnaire contained detailed explanations on which data had to be provided. Furthermore, NSO took on board feedback provided by respondents on the structure and content of the SES questionnaire from the 2006 wave. In this regard, the Office changed and modified where necessary, in order to have a clearer and a more user friendly version.

Respondents were also provided with additional assistance by staff working within the Labour Market Statistics Unit.  Such assistance was mostly provided via telephone and email.  In a number of cases on-site meetings were also held with enterprises in order to explain the method in which data had to be collected and to assist in the compilation of information. 

Accounting for Respondent errors 

Since SES data had to be filled in by respondents themselves, NSO tried to make questions easy to understand so that if the respondent refrained from reading the supporting explanatory notes, s/he would still answer correctly.  In addition, where possible, NSO tried to retrieve a number of variables from administrative records not to increase the response burden. In fact, where possible, the following variables where retrieved using administrative records and thus were not asked in the SES questionnaire:

• Form of economic and financial control (Variable 1.4)
• Sex (Variable 2.1),
• Date of Birth (used for Variable 2.2) and
• Date of Entry into Service with the enterprise (used for Variable 2.6) 

Data for public sector employees which were included in the SES sample, was totally derived from a combination of administrative sources.

In terms of errors resulting from lack of information on respondents’ registers or databases, respondents were asked to provide estimates.  In this regard the following variables were found to be difficult to retrieve from enterprises’ records:

• education of employees
• annual days of absence
• annual bonuses and allowances 

In line with the previous SES version, NSO tried to minimise respondents’ burden by trying to collect data in a way which was easier for the respondent to provide.  From information given by respondents, the NSO was aware that data on salaries was potentially available in 4 different categories, namely, weekly, fortnightly, monthly or for four week periods.   In certain cases different pay periods were applied within the same company with the most common being fortnightly pay period for manual workers and monthly pay period for the clerical and managerial staff.  As a result the Office took note of this information and adjusted the questionnaire to make it easier for the respondent to answer the part of the questionnaire relating to earnings for the representative month.

Additional efforts intended to reduce respondent errors concerned the variable Length of Service in the enterprise (Variable 2.6).  Since this variable was bound to produce biases in information provided, respondents were asked to provide the Office with the Date of Entry into service with the enterprise and the Date of Termination (if applicable). The difference between these two dates was in turn used to work out the variable Length of Service in the enterprise (Variable 2.6).

Unit response rate and non response

A number of units resulted to be ineligible due to misclassification and were thus excluded from the survey as discussed in section 6.3.2. The table below provides the unit non response rate for enterprises and the reasons for the non response.

The unit response rate for the Structure of Earnings Survey 2014 stood at 65%.

refer to 6.3.3

refer to 6.3.3

To minimize processing errors, each incoming questionnaire was thoroughly checked by trained statisticians using a number of validations. These validations included consistency checks between information provided for the reference month and information given for annual earnings.

Moreover any NACE coding which had to be carried out was to be in line with the Business Register classification.

In terms of processing errors emanating from data entry procedures, NSO split this task in two.  For data which was provided by respondents in soft copy format, information was directly uploaded into the data entry programme, thus minimizing any data entry errors.  For manually entered questionnaires inbuilt validations were applied to reduce data entry errors.

Following the completion of the data entry process, analysis of data was performed at micro level and where necessary any data inconsistencies which were found when checking the questionnaires were directly clarified with the respondents themselves.  In this regard each company had to provide two contact persons who had direct knowledge of the questionnaire for any eventual follow ups made by NSO.

Item imputation rate

The item imputation rate for the variable Gross earnings in the reference month resulted to be 28.1%.

Overall imputation rate

The overall imputation rate resulted to be 16.4%.

Ensuring the selection of a representative month 

For the SES 2014, October was selected as the representative month.  The same considerations applied for the SES 2010 wave were considered when choosing the representative month for the 2014 SES.  These are outlined below:

• October is not considered a vacation period and therefore is not likely to be characterized by a lot of absences (example the Education Sector is not affected in October and no public holidays happen to be during this month)
• The month is also not characterized by any irregular payments (for instance no statutory bonuses are paid in October)
• From the previous SES waves, October proved to be a good representative month.

Adjusting accounting or fiscal year to calendar year

All units operated on a calendar year basis and therefore this was not a major issue.

Ensuring the good coverage of the target population

In order to ensure that the target population of enterprises was well covered, stratified random sampling was applied for those enterprises operating in NACE sections B to N and P to S and employing 10 persons or more. The sampling selection for Nace section O was not carried out since all units operating in section O were chosen for the survey. Sample selection was at NACE two digit level and size class in order to ensure adequate representatively at all stages. 

Combination of data from administrative sources

For a number of enterprises which were successfully matched with a database kept by the national employment agency, a number of variables relating to employees did not have to be asked since these could be retrieved from administrative sources.  In order to be in a position to retrieve such information, a person unique identification number was used.  Such a number is commonly used in various databases at a national level.  The same number was also used to identify records relating to employees working in public sector units which were selected for the SES.

not applicable

NSO revisions policy can be downloaded from:  http://www.nso.gov.mt/docs/Revisions_of_Official_Statistics.pdf

not applicable.

not applicable

refer to 7.2

refer to 7.2

refer to 7.2

The following table illustrates the various stages between data collection and analysis.

From the table above one can note that a lot of time was taken to complete data collection. This was mainly due to the fact that key enterprises which had not forwarded their data and which had a probability of selection of one had to be chased numerous times until they submitted their questionnaire. The small size of the country necessitates the use of a census for most business surveys which are carried out amongst enterprises employing 50 employees or more.  This fact therefore places a huge response burden on selected organizations who are constantly being contacted for various surveys which are held by various units within NSO and also by other entities in general. Since a number of these organizations are key players in their respective sectors one cannot afford to leave them out of any business survey since most often they determine the developments which are taking place within the sector in which they operate and consequently influence the representativity of the results.

In addition, the use of administrative sources placed additional challenges for this wave of SES.  Despite the fact that a unique identification number is applied across the different agencies which forwarded the NSO with administrative information, the person number on its own was not enough to retrieve the necessary variables since a link had also to be established between the person and his/ her employer.  When a person had more than one job during the reference period, and hence had a link with more than one employer, the NSO had to determine which employer was to be used to retrieve the information which was required for the SES in order to avoid matching earnings data with the wrong employer.

This task proved to be difficult and time consuming since there is no unique employer number which is being used by the different administrative sources.

The issues above also had a direct effect on the timings of data analysis and weighting procedures.

refer to 7.2

National concepts applied for SES were in line with European concepts since the definitions outlined in Commission Regulation 1022/2009 were applied in the local context. 

The target population for SES was units which operated in NACE Sections B to S and employed 10 or more persons. 

In terms of the statistical units which were covered for SES, data was collected from legal units which are recognized as having autonomous management and an independent accounts system.  At NUTS 1 level, the whole country is represented; therefore information could be collected from enterprises which were recognized to be legal units by the Business Register.

A number of classifications had to be applied for this survey.  These included NACE and ISCO, both of which are applied for other enquiries.  In terms of education data, this was collected using national breakdowns and subsequently information was reclassified to be in line with ISCED levels.

not applicable.

All the variables for SES 2014 did not deviate from the Community legislation.

[Not requested]

Coherence with National Accounts data

National Accounts data is being compared to SES data. One is to note however that National Accounts information relates to all enterprises operating in the sector whereas SES data refers to enterprises which employ 10 or more employees.

Gross Annual Earnings per employee (€)

Variations between National Accounts and Structure of Earnings Survey estimates are the result of the micro business effect (under 10 effect) which is taken into account in the National Accounts averages but is missing in the SES estimate. The largest difference in earnings relates to NACE S.  Upon additional checks with National Accounts, the change between the two estimates is deemed to be the result of seasonal changes in employment since the sector increases its employment during the summer months and hence this matter lowers the per capita value when spread across one year.

[1] Source: National Accounts data as at June 2016. However please note that data is subject to revisions.

not applicable.

refer to 8.3

[Not requested]

At national level, results are intended to be published in 2017. These results will be published in the form of a news release and will be disseminated to the media and via the office’s website.

not applicable.

not applicable.

not applicable.

Micro data may be provided upon request. NSO has a specific set of regulations on the issue and data goes through a process of annonymisation before it is disseminated. (http://www.nso.gov.mt/site/page.aspx?pageid=682)

Results have not been published at a national level yet. However, main results from this survey can be obtained from Eurostat's online database: http://epp.eurostat.ec.europa.eu/portal/page/portal/statistics/search_database. In addition users can ask for customised information directly to the office.

Methodological notes will be accompanying the news release.

The methodological manual provided by Eurostat is constantly being consulted to ensure the full conformity to Eurostat definitions. All methods are documented in a quality report which is updated annually.

NSO recognises that the production of high quality statistics from this survey is paramount for policy making purposes. Many efforts were made during the data collection and data analysis stages in order to ensure accuracy of results. Great importance is also given to the production of harmonised results, so as to ensure comparability of results with those produced by other European national Statistics Institutes.  

The following is a list of concrete measures that were taken in order to ensure high quality of results:

> Thorough checking of all paper questionnaires prior to data entry

>  Data entry program had a number of in built validations in order to avoid data entry errors

>  Data was validated during the data analysis stage in order to identify misleading errors, using a program that was designed specifically for this purpose. Misleading data was verified with the respondents, or else suppressed to be later imputed using valid mathematical methods

>  Further checks at aggregate level were made in order to ensure consistency of results

> Aggregate statistics were compared with auxiliary sources in order to ensure consistency of results.

not applicable

not applicable.

not available.

Micro data is collected in terms of the Malta Statistics Authority, in which Part VIII - Use of Records of Public Authorities and protection of collected information stipulates that:
• All information furnished by any person, undertaking or public authority under this Act shall be used only for the purpose of statistical compilation and analysis.
• No information obtained in any way under this Act which can be related to an identifiable person or undertaking shall, except with the written consent of that person or undertaking or the personal representative or next-of-kin of that person, if he be deceased, be disseminated, shown or communicated to any person or body except -
     • for the purposes of a prosecution for an offence under this Act, or
     • to officers of statistics in the course of their duties under this Act.

Statistics based on less than 3 counts is not published. On the other hand, micro-data is fully anonymised before being disseminated to researchers.

no comments.