Labour force survey ad hoc module 2012 on transition from work to retirement

Transcription

1 UK Data Archive SN Labour Force Survey Ad Hoc Module Eurostat Dataset, 2012 Contents Labour force survey ad hoc module 2012 on transition from work to retirement Evaluation report - draft version

2 Contents Preface This report evaluates the Labour Force Survey (EU LFS) ad hoc module (AHM) 2012, which was on the topic transition from work to retirement. The main objective is to assess the implementation of the ad hoc module in 2012 by providing information about the quality of the data set and by giving the first results from the module. Recommendations for a possible repetition of the module are added as well. The EU LFS is a large sample survey among private households, which provides detailed quarterly and annual data on employment, unemployment and economic inactivity. The general regulation establishing the EU LFS with provisions on design, survey characteristics and decision making processes is the Council Regulation (EC) No 577/98 of 9 March 1998 on the organisation of a labour force sample survey in the European Union, with its amendments. The topic transition from work into retirement was covered by the 2006 LFS ad hoc module, and the topic was repeated in 2012 (Regulation No 365/2008). The lessons learned from the 2006 experience were integrated in the 2012 proposal as much as possible. The preparation of the 2012 module benefitted from the expertise of many labour market specialists from the national statistical offices, Eurostat and other General Directorates in the European Commission. An important documentation exercise involved all national statistical offices, as each of them drafted a mapping between the PENSTYPE variable (type of pension the person is currently receiving) and the national pension systems. The evaluation of the 2006 module, as well as documentation produced for the preparation of the 2012 module, are publicly available 1. They are meant to ease the task of researchers and the public in their understanding and use of the AHM data. They also show the administrative differences in the pension systems across the countries, and warn in cases where comparison between countries is not possible. This document includes a chapter with general information on the AHM It continues with a detailed description of each variable, its comparability across countries and with the applicable 2006 version and other information on data collection. Annexes include country abbreviations, the list of tables proposed for online dissemination as well as the text of Regulation No 365/2008 with the list of variables. This document makes use of the data sent to Eurostat by end of Minor revisions in the data set after this date are possible, but the data at this date was considered stable enough for analysis and interpretation. The quality reports provided by the participating countries were particularly useful in interpreting certain values, or in proposing suggestions for a potential future repetition of the module. Colleagues from many national statistical offices provided Eurostat with insight into the national circumstances, explaining specific results that do not fit the pattern. Eurostat thanks all contributors. This report was compiled by Diana Ivan and Håvard Lien of Eurostat s unit for labour market statistics (F3). Luxembourg, February

3 Contents Table of contents Preface... 2 Table of contents... 3 Chapter 1: General information on the module... 5 Executive summary for researchers... 5 Recommendations in case of a repetition of the module... 6 Description of the module... 6 Aims of the module and main findings... 6 Participating countries... 7 Target population... 7 Main findings... 9 Description of the variables... 9 Links with the AHM Links with the core LFS General measurement issues Sample size Non-response rates Other measurement issues Chapter 2: Quality analysis per variable PENSION: Person receives or does not receive a pension Short description Filter conditions and codes Analysis of the questionnaires Analysis of the results Conclusions and recommendations PENSTYPE: Type of pension(s) the person is currently receiving Short description Filter conditions and codes Analysis of the questionnaires Analysis of the results Conclusions and recommendations EARLYRET: Incidence of early retirement Short description Filter conditions and codes Analysis of the questionnaires Analysis of the results Conclusions and recommendations AGEPENS: Age at which a person first received an old-age pension Short description Filter conditions and codes Analysis of the questionnaires Analysis of the results Conclusions and recommendations REASNOT: Main reason for not staying longer at work Short description Filter conditions and codes Analysis of the questionnaires Analysis of the results Conclusions and recommendations WORKLONG: Wish to stay longer at work Short description Filter conditions and codes Analysis of the questionnaires Analysis of the results Conclusions and recommendations REDUCHRS: Reduced working hours in a move towards full retirement Short description... 52

4 Filter conditions and codes Analysis of the questionnaires Analysis of the results Conclusions and recommendations STAYWORK: Main reason for staying at work Short description Filter conditions and codes Analysis of the questionnaires Analysis of the results Conclusions and recommendations PLANSTOP: Plans to stop working in the future Short description Filter conditions and codes Analysis of the questionnaires Analysis of the results Conclusions and recommendations BUILDPEN: Information on pension rights the person has acquired so far Short description Filter conditions and codes Analysis of the questionnaires Analysis of the results Conclusions and recommendations CONTWORK: Expectations to continue working or looking for a job after receiving old-age pension Short description Filter conditions and codes Analysis of the questionnaires Analysis of the results Conclusions and recommendations Annexes Annex 1: Abbreviations Annex 2: Main AHM 2012 tables Annex 3: Commission Regulation (EU) No 249/ Contents

5 Chapter 1: General information on the module Executive summary for researchers The EU LFS sample size is about 1.5 million people every quarter. Only private households are included. The data is acquired by interviewing the sampled individuals directly. The interviewing mode varies across countries. In most countries, proxy interviews are allowed through a responsible person in the household. In most countries at least the first wave interview is conducted in person, while subsequent follow-up interviews can be conducted via telephone. Participation in the survey is compulsory in seven of the EU countries and in two of the participating EFTA countries. The list of variables collected by the LFS is available as Annex of the Regulation 377/2008, on codification and filters. These variables will be referred to into this document as core LFS variables, to distinguish them from the AHM variables. Explanatory notes on each of the variables are available 2 as well. Further legal basis 3 for the LFS AHMs is included in regulations on multiannual programmes of ad hoc modules and in regulations defining the list of variables that is to be collected in a specific year. The Commission Regulation (EU) No 249/2011 adopting the specifications of the 2012 ad hoc module defines the eleven variables collected by the AHM 2012 and describes the target population of the module and of each variable. The variables were proposed by a task force whose mandate was to define the list of variables to be collected, as well as the explanatory notes accompanying them. The frequently asked questions on the concepts to be measured were answered and presented in a collective document. All these documents are publicly available. Their main elements are summarized in this document, which adds further information on the data comparability across countries and across surveys. The first chapter explains in detail the structure of the target population of the AHM It clarifies which populations are included or excluded from the module. The fact that not all persons aged were interviewed shows that the focus of the module was only on the persons involved in the transition from work into retirement. The main side effect of the specific choice of the target population is that analysis by sex is limited, because the labour market participation of men and women is different. The AHM 2012 database does not include the non-applicable field (those not in the target population of the ad hoc module) for all countries which collected the data. The size of the non-applicable category for Germany, France, Austria, Sweden and Switzerland can be estimated by crossing the AHM data with the core LFS data, a method which was also used in the current report. In the first chapter the reader also finds the non-response rates by variable and country. All non-response rates higher than 15 % will be systematically flagged in the publication as unreliable. Awareness is therefore raised on the limits of the data quality for specific countries and/or specific variables of the module. The second chapter presents each variable in detail, from the data collection perspective as well as from the perspective of a future repetition of the module. At a glance, the quality of each variable can be summarised as follows: PENSION, AGEPENS are the variables with a good level of comparability; PENSTYPE, EARLYRET, WORKLONG, REASNOT, REDUCHRS, STAYWORK and are the variables where analysis should take into account the specific differences existing at national level; PLANSTOP, BUILDPEN and CONTWORK are the variables with the lowest quality, and their use should be restrained to national analysis only (provided that response rates allow even this). 2 _methodology#core_variables_.28user_guide.29_and_explanatory_notes 3 See, for more information:

6 Please note that names of variables will always be spelled in capital letters. Their definitions are available in the sub-chapters in chapter 2, along with the code lists. With the exception of the table 2 on sample size, table 4.1 and related graphs on percentile and mode, all data is weighted. Recommendations in case of a repetition of the module In the case of a repetition of a module with the same (or similar) topic, several recommendations made in the light of the 2012 experience are gathered here and can be taken into account when defining a new module. They were either formulated by the experts from the national institutes during the data collection or analysis at national level, or by Eurostat when analysing the dataset for all countries. (1) Consider shifting the topic of the module from transition from work into retirement into an analysis of the whole population within a certain age group. In general, it is considered difficult to collect detailed information on a complex phenomenon like transition from work into retirement by means of few questions that assume a linear evolution from work into retirement, while in practice a longer and atypical transition cannot be excluded. Inevitably, a module on a transition involves retrospective questions or questions about future intentions. Practice showed, in 2012 as well, that these questions are most difficult to implement in an AHM. (2) Chose simpler filters, both for the module and each of the variables. Complex filters are difficult to manage during the preparation phase, interviewing, data processing and data analysis. For example, an AHM having as filter only the age would have been easier to collect in a comparable way in all countries. Moreover, users would have benefitted from a richer dataset, allowing detailed analyses from both gender and country perspective. A broader filter also reduces the risk of a too small target population. The drawback of this approach would be a slight increase in the response burden, but this can be mitigated throughout the module if simpler variables are used. (3) Re-consider the definition of a pension in the context of the module. Because of specific national legal provisions, the borderline between pensions and other social benefits is not always easy to draw and is often leading to a situation when a similar social scheme is considered as pension in one country and another kind of benefit in another country. The 2012 AHM proposed to supplement the data collections with mappings of the pensions systems in each country. The advantage of the approach is the increase of the data collection transparency. However, little can be done for improving the comparability at European level for all levels of PENSTYPE collected by the module. Moreover, the documentation and update of the administrative information required a considerable effort from the countries participating in the exercise. Description of the module Aims of the module and main findings This ad-hoc module aimed at answering the following main questions: how people leave the labour market, why they left the labour market, why they did not stay longer and, how long the active population, aged 50 to 69, expects to be in the labour market. The results were intended for use in the framework of the Open Method of Coordination in the field of pensions, for the Joint Social Protection, Social Inclusion report, and in the framework of the Europe 2020 Strategy, in particular for the monitoring of the guideline on increasing labour market participation.

7 Participating countries The module was carried out in the EU-27, plus Croatia, Iceland, Norway and Switzerland. The aggregate EU-28 includes EU-27 and Croatia. Annex 1 includes the countries abbreviations used in this report. Target population The target population of the AHM 2012 are all persons aged 50 to 69 either working, or having worked after the age of 50. The target population is the same as for the AHM 2006 Transition from work into retirement. The target group excludes the persons that have not worked after the age of 50, for considerations linked to proximity to the labour market and reducing the response burden. It should however be analysed if this small reduction in the response burden did cause notable disadvantages for the data analysis. During the data collection, it was often difficult to identify the exact target population of the module. In particular, some of the countries collecting the data by paper questionnaire (Bulgaria, Greece, and Hungary) reported significant difficulties in ensuring that respondents respect the conditions imposed by each of the filters. Many countries (Malta, Switzerland, as well as other countries using paper questionnaires) had difficulties in retrieving data from the core LFS on the working experience after the age of 50, and had to re-collect all relevant information on age, working experience and working after the age of 50. In these cases, there were usually three questions (interviewer s check) added to the interview. In some cases, filters were applied only during codification, influencing the non-answering category or resulting in data not exploitable for the module. As the Austrian report noted, for detailed variables the additional filters reduced the number of observations. The specific target population makes both understanding and communication of the data more difficult, and results in a limitation of the possibilities of data analysis for several breakdowns, including by sex. The following graph shows the target population of the AHM 2012 compared to the total population aged 50-69, by sex, for EU-28. While at EU level the majority of the population aged are women (graph 1 shows that 52 % of the population aged are women), the majority of the target population of the AHM 2012 (aged 50-69, and currently working or having worked after the age of 50 years) are men. This is due to the fact that more than three in four persons in the group which was not part of the survey are women. Graph 1: Population aged by sex, employment status and participation in the AHM 2012, EU-28 (million persons)

8 Note: Not applicable values are estimated for: Germany, France, Austria and Sweden. In each of the EU-28 countries, there were more women than men who: (i) were not in employment and (ii) have not worked after the age of 50. However, there are significant differences among countries, as shown in graph 2. The group of persons who are not in employment and have not worked after the age of 50 is less likely to have access to pension rights, but the possibility of having pension rights cannot be excluded for persons who do have work experience. Unfortunately, the data set provides no information at all on persons who stopped working before the age of 50. The Hungarian quality report points out that the choice of the target population had additional disadvantages for countries with lower retirement age, because people who do not work after the age of 50 in general have high chances of receiving a pension either at the time of the interview, or in the future on the basis of their previous work experience. Moreover, it was unfortunate that although the module dedicated a special variable to the past contributions to several pension systems (BUILDPEN), the results are not available for those who have worked in the past, but not after their 50. The value of the results is therefore lower than what one could wish for. Graph 2: Non-applicable AHM 2012 population aged by sex (% in the population aged of the same sex)

9 Note: Values for Germany, France, Austria, Sweden and Switzerland are estimated. For this reason, the AHM 2012 neither allows generalisations to the full population aged 50-69, nor to its transition into retirement. Only in the case of the group being in employment at the time of the interview is the analysis not affected by the special choice of the target population, neither from the gender, nor from the country perspective. However, this particular group represents only half of the population aged in the EU-28, with employment rates ranging in the EU from 63 % in Sweden to 34 % in Malta. Main findings For the target population at EU-28 level we find that: Average age of receiving an old-age pension for the first time is 59 years, for both women and men. The average age is the same for men and for women in 13 of the EU-28 countries. 42 % receive a pension. Statutory old-age pension is by far the most commonly used type of pension. 81 % of those who declared receiving some form of pension receive a statutory old-age pension. At EU level, roughly 4 in 10 persons receiving an old-age pension declared having used an early retirement path. The top two main reasons for not staying longer at work are: reached eligibility for a pension (37 % of the economically inactive persons receiving a pension) and own health or disability (21 %). Description of the variables A full description of the variables, filters, and coding, as defined in the Commission Regulation (EU) No 249/2011, is available in Annex 3. The module contains the variables PENSION (Person receives a pension), PENSTYPE (Type of pension(s)), EARLYRET (Early retirement), AGEPENS (Age at which person first received an old-age pension), REASNOT (Main reason for not staying longer at work), WORKLONG (Wish to stay longer at work), REDUCHRS (Person reduced working hours in a move

10 towards full retirement), STAYWORK (Main reason for staying at work), PLANSTOP (Plans to stop work), BUILDPEN (Pension rights built up so far), CONTWORK (Expects to continue working/looking for a job after receiving old-age pension). There are 11 variables in total. Links with the AHM 2006 The policy background for the AHM 2012 is similar to the one for the AHM 2006, namely monitoring the labour market participation. There are however no completely and directly comparable variables between the two surveys. In the evaluation report 4 from the task force that evaluated the AHM 2006 there are the following recommendations for changes in the variables, in case of a repetition of the topic or module. We list here only the six variables that deal in principle with the same content both in 2006 and (1) REDUCHRS caused problems for respondents in 2006 because it asked about their future plans, which they in many cases did not have at the time of the interview, especially due to changes in the legal basis regulating pensions. Also, the response item listing progressive retirement schemes was not used in several countries, because this was a non-existent concept for them. The task force evaluating the 2006 module recommended a simplification of the variable, which was followed in (2) PLAGESTP (linked to the AHM 2012 variable PLANSTOP) had problems for the same reason as REDUCHRS. Unfortunately, the fine-tunings proposed in 2012 did not solve the original problem of persons finding it difficult to answer about future plans. (3) REASRET (linked to AHM 2012 REASNOT) reported to have too many and too detailed response items, and caused trouble distinguishing between several of the items. There were changes proposed to the variable, but not necessarily towards its simplification. (4) AGEPENS had problems with the understanding of individual retirement pensions. This was accordingly changed in 2012 to old-age pension. (5) OTHBENF (linked to AHM 2012 PENSTYPE) had problems with the interpretation of individual benefits in the explanatory notes. In 2012 the survey asks for broad categories of pension types. (6) FININCTV (linked to AHM 2012 STAYWORK) found that the filter conditions in 2006 were too restrictive, leading to many missing values. This advice was followed in Links with the core LFS The target population of the module is in part based on the ILO employment status, which has three main subgroups: employed, unemployed, and economically inactive population. While the ILO status is obtained from the combination of several core variables (WSTATOR, SEEKWORK, AVAILBLE, METHODA to METHODM), the AHM 2012 module is based on a simplified version, derived from only two variables in the core LFS: WSTATOR (labour status during the reference week) and SEEKWORK (seeking employment during the previous four weeks), in the following way: Employed: WSTATOR in (1,2); Unemployed (simplified): WSTATOR in (3,5) and SEEKWORK in (1,2,4); Inactive (simplified): WSTATOR in (3, 5) and SEEKWORK = 3. In the schematic overview of the relation between the variables, in the last page of the Explanatory notes, the basic ILO terminology is used, without a clear indication of the intended simplification (by SEEKWORK). This is unfortunate. However, the regulation takes precedence over the addendum to the Explanatory notes. The fact that not all WSTATOR categories are included in the scheme has no impact on the module, as they refer to people who are either younger than 15 years, older than 75, or are doing compulsory military service. None of these are in the target group years. YEARPR (year in which person last worked) is the variable used to define the target population of the module, in combination with YEARBIR (year of birth). 4 The publication is available at:

11 General measurement issues This section gives detailed information on sample sizes and non-response rates. It also includes national characteristics of interviewing. Sample size The sample size of the AHM 2012 module is given by the number of persons interviewed during the labour force survey in the specific 2012 quarter(s) 5. Only persons aged 50 to 69 and fulfilling specific labour market status (see Target population chapter for more details) were interviewed. The following table gives the number of interviews conducted in each country, together with the calculation of the proportion of these interviews relative to the country s population in the target group. Table 1: Sample size of the AHM 2012, by country AHM sample (number of persons interviewed) AHM sample relative to the corresponding population in each country (%) EU % BE % BG % CZ % DK % DE % EE % IE % EL % ES % FR % HR % IT % CY % LV % LT % LU % HU % MT % NL % AT % PL % PT % RO % SI % SK % FI % SE % UK % IS % NO % CH % 5 See the Reference period of the module (in Table 3) for the specific quarter(s) by country.

12 PENSION EARLYRET AGEPENS REASNOT WORKLONG REDUCHRS STAYWORK PLANSTOP CONTWORK Note: The corresponding AHM2012 population is the weighted total of the yes, no and blank answers of the PENSION variable. Non-response rates The following table shows non-response rates. Values higher than 15 % will mean that the variable has a very limited user value. Only 9 of the 31 countries that did the survey have acceptable response rates (namely, higher than 85 %) for all variables. The two variables PLANSTOP and CONTWORK are close to useless for analytical purposes due to very high non-response at EU-28 level. Table 2: Non-response rates in the AHM 2012, by variable EU-28 2% 4% 5% 4% 6% 11% 12% 35% 19% BE 0% 0% 2% 19% 21% 12% 1% 2% 0% BG 3% 0% 0% 0% 0% 0% 0% 60% 52% CZ 0% 0% 0% 0% 0% 0% 1% 5% 6% DK 0% 0% 0% 0% 0% 0% 4% 0% 29% DE 8% 27% 30% 26% 33% 21% 42% 45% 19% EE 0% 0% 0% 0% 0% 0% 0% 19% 13% IE 5% 0% 2% 0% 1% 57% 2% 16% 8% EL 9% 0% 0% 0% 20% 19% 0% 40% 26% ES 0% 0% 2% 0% 5% 2% 1% 67% 17% FR 0% 0% 1% 0% 0% 1% 33% 26% 16% HR 0% 0% 0% 0% 0% 2% 18% 89% 0% IT 0% 0% 0% 0% 1% 0% 0% 63% 24% CY 0% 0% 0% 0% 0% 0% 0% 0% 0% LV 0% 0% 1% 1% 1% 0% 0% 11% 0% LT 0% 0% 1% 0% 0% 11% 0% 0% 46% LU 0% 1% 2% 0% 2% 3% 8% 14% 13% HU 3% 0% 0% 0% 0% 1% 0% 48% 0% MT 0% 0% 0% 0% 0% 0% 0% 0% 0% NL 1% 0% 1% 0% 1% 1% 4% 19% 15% AT 0% 0% 0% 0% 0% 0% 0% 16% 0% PL 0% 0% 0% 0% 0% 0% 0% 0% 0% PT 2% 0% 0% 0% 1% 1% 0% 22% 8% RO 6% 0% 0% 0% 0% 3% 0% 83% 61% SI 0% 0% 0% 0% 0% 0% 8% 0% 0% SK 0% 0% 0% 0% 0% 0% 0% 5% 3% FI 1% 1% 3% 0% 1% 2% 3% 8% 4% SE 0% 0% 0% 1% 1% 3% 1% 52% 17% UK 0% 0% 1% 0% 1% 43% 1% 29% 22% IS 2% 1% 3% 35% 37% 15% 13% 82% 32% NO 1% 62% 19% 0% 2% 11% 4% 25% 36% CH 2% 1% 9% 2% 1% 1% 17% 25% 31%

13 Other measurement issues The following table gives a summary of background information that can be expected to influence the quality and comparability of the results. Among the national characteristics of interviewing, the table below presents: reference periods (one specific quarter usually the second quarter: Q2 or more quarters in the year 2012), information on the nature of participation: compulsory or voluntary, direct or proxy answering. Presented here as well is the available information on the pilot testing before the full implementation. Finally, there is information included on the order of questions when collecting the LFS and the AHM data (either AHM questions were integrated in the survey by topic, or they followed after the LFS data collection). Table 3: General measurement issues in the AHM 2012 Reference period Participation Proxy answering Pilot survey / Testing type AHM questions after core LFS? BE Q2 Compulsory No Yes BG Q1-Q4 Voluntary Yes Yes, field testing Yes CZ Q1-Q4 Voluntary Yes No DK Q2 Voluntary Yes Yes DE Q1-Q4 Voluntary Yes Yes EE Q2, Q4 Voluntary Yes Yes IE Q2 Voluntary Yes Yes EL Q2 Compulsory Yes Yes ES Q1-Q4 Compulsory Yes Yes FR Q1-Q4 Compulsory No Yes, field testing (185 and 283 respondents) Yes HR Q2 Voluntary Yes Yes IT Q2 Compulsory Yes No CY Q2 Compulsory Yes Yes LV Q2 Voluntary Yes Yes LT Q2 Voluntary Yes Yes LU Q1-Q4 Voluntary Yes Yes, field testing Yes HU Q2 Voluntary Yes Yes MT Q1-Q2 Voluntary Yes No NL Q1-Q4 Voluntary Yes No AT Q1-Q4 Compulsory Yes Yes, field-testing (195 respondents) Yes PL Q2 Voluntary Yes Yes PT Q2 Voluntary Yes Yes RO Q2 Voluntary Yes Yes, field-testing (593 respondents) Yes SI Q2 Voluntary Yes Yes SK Q2 Voluntary Yes Yes FI Q1-Q4 Voluntary Yes Yes, cognitive testing (22 respondents) Yes SE Q1-Q4 Voluntary Yes Yes, cognitive testing Yes UK Q1-Q4 Voluntary Yes No IS Q2 Voluntary Yes No NO Q1-Q4 Compulsory No Yes CH Q1-Q4 Compulsory Yes Yes The reports summarizing the national interviewing phase give in general the picture of an overall positive experience. However, there are several issues that would need to be addressed in the case of the repetition of the module, like filters and the general routing. Challenging situations occurred during the interviewing phase. Further testing of the questionnaires, better mapping of national pension systems, and increased training or interviewers could have prevented the low-quality data that we see in some cases. In

14 other cases, complex, difficult to understand and even changing pension systems made the operationalization of the AHM difficult. The interviewing mode sometimes limited the options of the interviewers, and so did some variables which were too rigid for examples the ones not allowing the spontaneous I do not know answer. Questions asked twice (in the LFS and in the AHM), difficult administrative concepts, problems in recalling past events or in forecasting future decisions on retirement add to the list of issues that are likely to have had negative influence on the quality of the data. All these points should be re-assessed before collecting new data on the same topic. All these cases will be described in detail in the next chapter, by variable.

15 Chapter 2: Quality analysis per variable This chapter continues the assessment of the AHM 2012 with an analysis per variable. The eleven variables of the module will be presented in the same order as in the Regulation. This is the order of columns in the database 6, but it does not imply that variables were collected in this order in all countries. National questionnaires used to collect the AHM 2012 data are available 7, often in several languages. 1. PENSION: Person receives or does not receive a pension Short description The purpose of the variable is to classify the target population in two groups, one composed of those currently receiving a pension and the other of those not receiving one. Respondents were expected to decide, via their own judgment, whether the type of benefit they receive is a pension. One general condition was required: the payment considered as pension had to be a regular and periodic benefit in cash apart from salaries or wages. Lump-sum payments and benefits in kind were excluded from the definition 8. A list of benefits which are not considered as pensions was drawn up 9 as well. Symbolic payments, even if labelled as pensions, were excluded 10. Filter conditions and codes Asked to all persons in the module, that is: aged and are either currently working (WSTATOR=1, 2) or have worked after the age of 50 (WSTATOR=3, 5 and (YEARPR-YEARBIR)>49)). See the section Target population for more information. Code 1 Yes 2 No Description 9 Not applicable (not included in the filter) Blank Analysis of the questionnaires No answ er or does not know This variable has a particular importance, first of all because it makes the distinction between receiving and not receiving a pension, and secondly because it acts as a filter for the rest of the module. Ideally, the same questionnaire was used in all countries, to minimise the risks of introducing national differences in a variable of which the outcome was crucial for the whole AHM. In general, the recommendation of asking respondents to judge themselves whether they receive a pension was followed. In some questionnaires however (see for example the Belgian and Swiss questionnaire), this question was repeated for several types of pensions mainly the ones from the PENSTYPE variable while in others (Spain, Portugal, Romania) more types of pensions were already given as examples in the opening question. In Bulgaria and Hungary, it was first checked whether respondents receive an old-age pension, and afterwards the questionnaire addressed the other types of pensions. It can be assumed that these variations did not fundamentally affect the comparability of the variable; however the real impact cannot be assessed. 6 AHM 2012 columns ranged from 197 to 218, with one or more digits per variable. 7 See description of the 2012 AHM here: _ad_hoc_modules 8 Explanatory notes, page 4. 9 Idem, page Frequently asked questions document, page 4.

16 The role that the variable played in the routing of the questionnaire is important. On one hand, those who answered no (code 2) were routed outside the variables PENSTYPE, EARLYRET, AGEPENS, REASNOT, WORKLONG, STAYWORK, and PLANSTOP. They were routed instead towards REDUCHRS (if in employment) and BUILDPEN and CONTWORK. In most cases, those who wrongly answered no the first subjective question did not receive follow-up questions on the same topic. On the other hand, those who answered yes (code 1) were routed towards the PENSTYPE question. It can be assumed that the combination of a subjective variable (PENSION) and of a yes/no variable based on fixed administrative schemes and which moreover does not accept a does not know answer (PENSTYPE) was not perfect in all cases. It can be expected that inconsistencies between the two variables triggered on the spot decisions of interviewers and as a result, the routing of the questionnaire was not always as consistent as it was hoped for. The variable PENSION was not collected in 2006, and we do therefore not have any time series data to compare it to. It is, however, possible to compare the results to the LFS variable MAINSTAT (Persons self-perception regarding their main activity status) 11. Comparison to the LFS variable LEAVREAS (Main reason for leaving the last job or business) was made as well. The testing of the model questionnaire had anticipated a high quality variable, useful as a start for the module. Results from the full implementation generally support this view. Analysis of the results Response rate is good for all participating countries. The univariate analysis of a categorical variable gives rather limited options on methods, but we do see that this variable is the one with the highest response rate in the module, at 98 % (table 2). Also visible is the impact of the target population for the possible interpretation of the data. Analysing the existence of pension rights on a sub-group of the population aged (moreover a sub-group with an unequal gender distribution) is not without risks. This aspect was detailed in the section Target population, where a breakdown by country and sex is available. Graph 1.1 provides an overview of the distribution of this variable for the population aged 50 to 69, showing for each country how the yes and no categories of PENSION relate to the non-applicable category. Graph 1.1: Distribution of the AHM 2012 variable PENSION (% of the population aged 50-69) 11 MAINSTAT is an optional variable in the LFS, measuring the (perceived) main labour status.

17 When focusing the analysis on the specific target group of the AHM, we note that the EU-28 average for responding yes, that is, those in the target group who receive a pension, is at 42 % (see graph 1.2). Nineteen of the thirty-one countries are within a ± 5 percentage point range from this average. Three countries are on the exact average (Finland, Greece and Portugal). Nine EU member states, as well as Switzerland, Norway and Iceland are lower than the EU average. The yes-no curve in graph 1.2 is rather smooth; consequently we do not have any outliers. The highest value of yes (53% in the Czech Republic) is two and a half times larger than the lowest value (in Iceland). As for geographical patterns, the countries in the upper part of yes are Eastern European (the Czech Republic, Slovenia, Slovakia, Romania, Poland), whereas the lower part is more mixed, with Mediterranean as well as North-Western countries (Iceland, Ireland, the Netherlands, Cyprus, Spain, Norway), but also including Hungary. Graph 1.2: Distribution of the AHM 2012 variable PENSION (% of the target population aged 50-69)

18 A sub-group that is not affected by the choice of the target population is the one of persons in employment. Graph 1.3 shows the presence or absence of a pension among persons in employment. The height of each bar represents the overall employment rate for persons aged in each country. In all countries there is only a small part of the population that combines employment with pensions. In general, countries with a relative high proportion of yes answers among persons in employment have at the same time a high employment rate see for example Sweden, Estonia, the United Kingdom or Norway. However, there are several countries where employment rates are high and the proportion of persons with pension among the employed is relatively low (in Germany, Denmark and the Netherlands). Graph 1.3: Distribution of the AHM 2012 PENSION variable for persons aged and in employment (% of the entire population)

19 Graph 1.4 shows that actually the majority of persons who declared receiving a pension are not in employment (anymore). At EU-28 level, 34 % of the AHM 2012 population receive a pension and are not in employment, while only 7 % receive a pension and are in employment. Graph 1.4: Distribution of the AHM 2012 PENSION variable for persons aged by employment status (% of the AHM 2012 population)

20 In the light of the previous graphs, we can conclude that there is a stong link between the employment status and the variable PENSION, but there are many other labour market features that play a role in the final distribution of the PENSION variable. Some countries (like Malta, Greece, Italy and Belgium - see graph 2) have a significant population who did not work after the age of 50, and that population was not interviewed. Other countries (like Sweden, Estonia and the United Kingdom see graph 1.3) have a high number of persons combining employment and pension, but in general persons receiving a pension are not in employment anymore, as seen in graph 1.4. Among persons receiving a pension, there are important variations at country level regarding their employment status: more than one third of those receiving a pension are still in employment in Sweden, Estonia and the United Kingdom, while only 3 % of those receiving a pension are still in employment in Greece. Finally (see graph 1.2), more than half of the respondents in the Czech Republic and Slovenia do receive a pension, while in Ireland and the Netherlands less than 30 % do so. The next section will look into the aggregated PENSION data, and analyse the links between the PENSION variable and the age of respondents, for EU-28. Graph 1.5: Distribution of the AHM 2012 variable PENSION by age groups and employment status, EU-28 (million persons) Unsurprisingly, the variable PENSION has a distribution strongly changing with the age of the respondents. Before the age of 60 a majority of respondents are still in employment and do not receive a pension. After the age of 60, a majority of respondents are not in employment and do receive a pension. The groups of persons combining employment and pension, or not in employment and without pension are less prominent. While the age of 60 could be taken as a natural turning point for the EU-28 population in its transition from work to retirement, all analysis by pension needs to take into account the significant differences at country level from the demographic perspective. The following graphs looks into more detail at the population aged 55 to 64, in an attempt to describe the actual transition from the labour market towards employment in the age group when many countries show this turning point from work to retirement. The age groups and have respectively low proportions of persons receiving pensions (6 % at EU level with some countries variations), or very high

21 proportions averaging 95 % at EU-28 level. Ireland and the Netherlands are the only EU countries where the proportion of respondents aged 65 to 69 receiving a pension is under 90 %; in Greece, Romania and Spain the proportion is 90 %. There are as well countries with full pension provision for the AHM population aged (Estonia, Malta, Latvia), showing that the pension right becomes universal with age for the population having worked after the age of 50. The graph below shows the breakdown of the PENSION variable by 5-year age groups, of respondents below and above 60 years of age. Differences in the proportion of yes answers are significant among countries, both for the age group 55-59, and for the age group. Some countries, like Denmark, Hungary, and Malta, have important increases in the proportion of PENSION =yes just after the age of 60. It is not surprising that countries with relatively low proportions of pensions in these age groups (Ireland, the Netherlands, Spain and Cyprus) will find themselves in the lowest part of the overall PENSION distribution (see graph 1.2). However, the Swedish average for PENSION=yes is superior to the EU average, a result driven by the high proportion of respondents receiving a pension in the age group (98 %). Graph 1.6: Proportion of yes in AHM 2012 variable PENSION by respondents age group (% in the corresponding age group) We conclude the section dedicated to the age group 55 to 64 with a graph showing the gender differences among the AHM population. In most countries, women complete their transition towards retirement before men. Countries where women retire earlier than men are the countries likely to show an overall high proportion of PENSION=yes occurrences. Graph 1.7: Proportion of yes in AHM 2012 variable PENSION for the AHM population aged by gender (%)

22 Since this variable was not included in the module in 2006, the best we can do for evaluating the quality is to follow the recommendation in the Explanatory notes, of comparing it to the LFS variable MAINSTAT. The two variables are related, PENSION being about receiving a pension, and MAINSTAT collecting information about the self-perceived main activity status, with one option (code 4) being: in retirement or early retirement or has given up business. By definition, they cannot overlap completely, but we would expect these two variables to show the same picture to a large degree. The following graph confirms these expectations, with most of those who define themselves as retired also receiving a pension, and with persons in employment or unemployment less likely to receive a pension. It also shows that, at aggregate level, there are more respondents in PENSION=1 than in MAINSTAT=4. Graph 1.8: Proportion of yes in AHM 2012 variable PENSION for selected categories of the LFS variable MAINSTAT, EU-28 (%)

23 Note: EU-28 average excluding Germany and the United Kingdom. The following graph confirms that the relationship is not random. It shows, for the age group 50 69, the correlation between those who have the value 4 in the core LFS variable MAINSTAT (in retirement or early retirement or has given up business) and the value 1 in the LFS AHM variable PENSION (yes, receives a pension), per country. The clusters of countries are recognisable from graph 1.1, with Iceland and Ireland at the lower end, and the Czech Republic and Slovenia at the other. Germany and the United Kingdom are not displayed in graph 1.8 because they did not provide data for MAINSTAT in No country is massively off the trend line. We retrieve as well the systematic pattern having the share of PENSION=yes higher than MAINSTAT=4; it holds in all countries, with the exception of Hungary and Croatia. The Hungarian exception is explained by the recent pension reform 12 that has transformed several of the previous types of pension into social benefits. This example shows the sensitivity of the variable to national administrative definitions, and constitutes a good warning on the errors that can occur if data is generalised and interpreted outside the AHM 2012 context. Graph 1.9: Correspondence between LFS variable MAINSTAT and AHM 2012 variable PENSION (%) 12 More information is available in the document mapping the national pension systems and the AHM 2012 variable PENSTYPE.

24 Another possibility open to the analysis is to compare PENSION to the core LFS variable LEAVREAS. The variable LEAVREAS (Main reason for leaving the last job or business) is collected yearly from all persons not in employment, but having a previous employment experience and who had stopped work within eight years. The overlap between this variable and the PENSION variable is represented by the group of persons aged not in employment but having a previous employment experience after the age of 50 and still within previous eight years. Graph 1.10: Proportion of yes in AHM 2012 variable PENSION for selected categories of the LFS variable LEAVREAS, EU-28 (%)

25 As the graph 1.10 shows, the two variables are correlated as expected. A wide majority of persons for whom the reason for leaving the last job is retirement (be it normal or early) do declare receiving a pension in This holds for persons leaving the labour market for health reasons as well. Among the other reasons from the LEAVREAS variable, we can mention: dismissal (code 00), end of a contract of limited duration (code 01) or other personal reasons and in all cases the distribution of the PENSION variable is according to the expectations. Conclusions and recommendations The form of the variable suggests that it can be collected by a very simple yes/no question, with no complicated routing or transcoding. On the other hand, it can be argued that the answer on whether one receives or not a pension (as defined by the AHM 2012) can be a more laborious cognitive process. Most difficulties were reported to have occurred in situations when respondents receive some benefits (in particular unemployment benefits or disability benefits) that might even be called pensions in the common language, but which are not considered as pensions in the framework of this module. The opposite situation was reported to have occurred as well, if persons do receive pensions (usually not oldage pensions) but answered spontaneously no to the question on pensions. Countries with recent changes in their pension systems (notably Bulgaria, Hungary, and Poland) had to ensure consistency among the old and new systems. In order to prevent errors resulting from these situations, national mappings were prepared, establishing the links between the existing national terminology and the PENSTYPE types of pension. It is implicit that PENSION is the sum of all the cases foreseen for PENSTYPE. However, it has to be noted that this conceptual complexity is not fully compatible with an intentionally subjective and intentionally simple yes/no question. In several countries (Bulgaria, Hungary for example), the variable was not collected through a yes/no question, but derived from other answers on the type of pensions received. The complexity of the opening question varied as well, from a simple Do you receive a pension? version towards questions listing several national types of pensions. Depending on the flow of questions, on the interview mode as well as the specific training and experience of interviewers, PENSION could sometimes have been re-coded in the light of the information collected by the PENSTYPE variable. It can also be expected that in other cases the fully consistent coding could not be ensured. Very few of the participating countries reported any problems with measuring this variable, and the

26 response rate is good. The Netherlands report that they underestimate the number of pensioners, because it was in some cases difficult to make the respondents understand the module s broad definition of pensions. Greece reports a situation where one is formally entitled to a pension, but the actual payment can be delayed for more than a year, thus making it hard to know if one should answer yes or no to the question of receiving a pension. No country reports that the variable differs from the Commission regulation (249/2011) in the national implementation of it. We have a good response rate, no outliers, and good correlation with MAINSTAT and LEAVREAS. It is therefore safe to conclude that quality of the variable is good. However, the variable is based on national definitions and classifications of pensions and other social benefits. Conceptually its comparability is unfortunately limited at a broader than national level. In case of a repetition of the module it should be analysed whether one single variable is sufficient for the ambitious purpose of classifying the target population in two groups: those receiving and those not receiving a pension. If a similar variable is used again as a filter for the rest of the module, the risks of inconsistencies or misclassifications and their impact on the overall data collection should be assessed as well. Moreover, the concept of pension was defined in 2012 at national level, as specific provisions on entitlements to selected social benefits. The possibilities of increasing its comparability at European level could be further analysed. 2. PENSTYPE: Type of pension(s) the person is currently receiving Short description The purpose of the variable is to know, for those having answered yes in the previous question (PENSION), which type of pension(s) he or she is receiving. However, not all participating countries had this question flow in their questionnaires. PENSTYPE is a variable that, as one can see in the provided code list, has eight possible answers: four old-age types of pension by scheme, unemployment pension, disability pension, survivor s pension and other pension or type of pension unknown. More than one yes answers could be given, as one person can receive several types of pensions at the same time. The answer does not know was not foreseen for the individual pension types. Lists of the different pensions and benefits available at national level and their assignment to the relevant code were prepared 13. Filter conditions and codes Asked to all persons who answered yes (code 1) to PENSION. Code Description PENSTYP1 1: Yes; 0: No Old-age pension. Statutory scheme PENSTYP2 1: Yes; 0: No Old-age pension. Occupational scheme PENSTYP3 1: Yes; 0: No Old-age pension. Personal scheme PENSTYP4 1: Yes; 0: No Old-age pension. Scheme unknow n PENSTYP5 1: Yes; 0: No Unemployment pension PENSTYP6 1: Yes; 0: No Disability pension PENSTYP7 1: Yes; 0: No Survivor's pension PENSTYP8 1: Yes; 0: No Other pension(s) or type of pension unknow n Not applicable (not included in the filter) Analysis of the questionnaires It was recommended to ask the PENSTYPE question only after the PENSION variable. Most countries followed this recommendation and collected first information on pension benefits, and afterwards 13 A detailed document presenting a mapping of the pension types from this variable and the national types of pensions is available online at:

27 information on the type of pensions received. In other countries, the structure was slightly different, as in Belgium, Lithuania and Switzerland. The risks of inconsistencies among the two variables for the overall questionnaire were discussed in the PENSION section. For the PENSTYPE variable, pension mappings were prepared at national level. They were in general helpful, but involved substantial documentation. Countries reported that in the case of extremely complex national systems (Denmark, Italy) or, moreover, changing systems (Hungary, Poland) it was difficult to ensure the robustness of the variable for each of its eight dimensions. Bulgaria and Sweden report minor deviations in the implementation of the variable. For Bulgaria this is due to a change in the Social Security Code in 2012, which split social disability pension from disability pension (PENSTYP6), with an estimated impact on the results of PENSTYP6 of 12 %. Sweden reports to have not used PENSTYP4, but included it in PENSTYP8. In some countries (Finland, the Netherlands) it was noted that the terminology of this AHM is not always known and respondents had difficulties to answer. For example, in Finnish the old-age pension concept is not always understood correctly. In some situations (in Finland, Italy and it is likely to have occurred in other countries as well), when the respondent receives a benefit, the whole label has in some cases changed over time, for example when a disability pension is automatically converted in an old-age pension after a certain age. In these cases, respondents were not always precise in indicating the correct type of pension. Errors in the PENSTYPE variable had an impact on the overall routing of the questionnaire, as well as on the precision of the AGEPENS variable. It can be however assumed that the impact of these cases on the overall data collection was minor. A somewhat similar variable was collected in 2006, as discussed in the chapter Links with the AHM In Hungary there was a major overhaul of the pension system in 2012, which makes any attempt at comparison to the 2006 module meaningless. Analysis of the results This is a case where eight variables are fitted into one. Each of the eight sub-variables had yes and no as answering categories, and as a result multiple types of pensions could have been selected per person. Respondents receiving two (or more) types of pensions are therefore counted as many times as they receive pension types. Unfortunately, data is not easy to interpret at EU level, as very similar types of pensions could be classified in different countries under different schemes. Moreover, for some types of pensions (PENSTYP 5 or 6), similar social rights are classified as pension scheme in some countries or social benefit not collected by the module in other countries. Below we provide a table showing the proportion of the yes answers in the total number of answers (yes answers and no answers) for each sub category of the PENSTYPE variable, and by country.

28 PENSTYPE1 PENSTYPE2 PENSTYPE3 PENSTYPE4 PENSTYPE5 PENSTYPE6 PENSTYPE7 PENSTYPE8 Table 2.1: Proportion of yes answers in total answers in each of the AHM 2012 variables PENSTYP1 to 8, EU-28 (% of those with PENSION=1) EU-28 81% 16% 8% 2% 1% 9% 8% 5% BE 73% 7% 7% 7% 16% 9% 5% 0% BG 90% 1% 0% 0% 0% 11% 3% 0% CZ 90% 0% 4% 0% 0% 9% 12% 0% DK 83% 37% 14% 1% 0% 13% 2% 0% DE 86% 19% 17% 12% 0% 11% 16% 15% EE 81% 0% 6% 0% 0% 19% 0% 0% IE 46% 49% 5% 2% 0% 6% 7% 2% EL 95% 1% 1% 0% 1% 3% 1% 0% ES 67% 3% 2% 0% 10% 15% 11% 4% FR 88% 3% 5% 0% 1% 12% 10% 0% HR 67% 0% 0% 0% 0% 27% 3% 7% IT 92% 3% 1% 0% 0% 7% 0% 4% CY 89% 3% 2% 0% 0% 6% 7% 0% LV 88% 0% 0% 0% 0% 12% 0% 0% LT 85% 0% 0% 0% 0% 12% 12% 5% LU 80% 12% 7% 6% 0% 13% 8% 4% HU 97% 0% 0% 0% 0% 0% 14% 0% MT 83% 18% 0% 1% 1% 6% 3% 2% NL 57% 54% 10% 2% 0% 17% 12% 2% AT 87% 7% 2% 0% 2% 14% 10% 1% PL 81% 0% 0% 0% 0% 0% 3% 10% PT 74% 2% 1% 2% 5% 14% 14% 1% RO 85% 0% 0% 0% 0% 14% 1% 0% SI 88% 0% 0% 0% 0% 8% 5% 5% SK 84% 0% 1% 1% 5% 8% 20% 0% FI 78% 5% 9% 0% 1% 15% 7% 1% SE 84% 74% 47% 0% 0% 6% 6% 6% UK 65% 60% 20% 0% 0% 0% 4% 3% IS 18% 49% 5% 0% 0% 37% 9% 0% NO 50% 31% 4% 0% 0% 26% 5% 11% CH 73% 58% 10% 6% 0% 9% 7% 4% The main result is that statutory old-age pension (PENSTYP1) is by far the most commonly used type. 81 % of those who declared receiving some form of pension receive statutory old-age pension in the EU- 28. It is only in Ireland and Iceland that we find this pension type being surpassed by another type of pension, and this is in both cases by the occupational old-age pension. If we look at the relation between old-age statutory pension and the sum of all other pension types, often old-age statutory pensions outnumber the use of all other types of pensions combined. We only find Germany, Ireland, the Netherlands, Sweden, the United Kingdom and the EFTA countries being the ones not fitting the main pattern here. Seven countries, Bulgaria, Greece, Italy, Hungary, Poland, Romania, and Croatia, show very low use of other types of pension other than old-age statutory scheme. The following graph gives an indication on the frequency of combinations of pensions received by the same respondent. Unlike in the previous variable PENSION, where each respondent receiving a pension

29 is counted only once, this variable allows the quantifications of the number of received pensions falling into the eight different schemes. The graph below shows that in many countries the stacked bar combining different types of pensions is not far from the value of 100 %, indicating that the combination of several pensions for one person is rather rare in these countries. Graph 2.1: Distribution of the PENSTYP1 to PENSTYP8 AHM 2012 variable (% of yes answers in total answers given for each PENSTYPE) Note: The maximum of the scale of the graph would be 800 % if each respondent received all 8 types of pensions at the same time. Concerning PENSTYP5, unemployment pension, the data reflects the national mappings well, where a majority of countries have mentioned that this specific scheme does not exist at national level. At EU level, only in the case of Ireland, Italy and Lithuania (probably small-scale) unemployment pensions were declared as existing, but they are not reflected in the data collection. As PENSTYP1 is the most frequent type of pension, we will look into how PENSTYP1 relates to the other categories of the variable PENSTYPE, in other words how frequently persons receive only PENSTYP1, or PENSTYP1 in combination with other types of pensions, or one (or more) of the PENSTYP2 to 8. Graph 2.2: PENSTYP1 and its relationship with PENSTYP2 to PENSTYP8 (% of persons with PENSION=1)

30 The previous graph shows that, in the EU-28, in most cases persons receiving a pension receive only an old-age pension from the statutory scheme. However, there are important differences at country level, as this indicator ranges from above 90 % in Greece to less than 5 % in Iceland. Except in a few countries (Sweden, Germany, the Netherlands and the United Kingdom), the phenomenon of combining different pensions from different schemes is not very common. In the next section we will focus the analysis on those persons who do not receive any of the old-age pensions (PENSTYP1 to 4 = no). They represent 12 % of all persons with PENSION=yes. In this group, when aggregating the results from PENSTYP5 to 8, we observe a rather constant share among the EU countries. The fact that the values shown in the graph 2.2 are often neighbouring 100 % shows that in most countries, when a person does not receive an old-age pension, then it is unlikely that that person combines several pensions of types PENSTYP5 to 8.Germany is the only exception. Data is ordered by PENSTYP=6, but a full analysis on each PENSTYP5 to 8 should also include persons receiving old-age pensions. It can be concluded that the analytical potential of an analysis on combinations of pensions received is limited, because national schemes follow special patterns that speak more about the organization of the pension system rather than about the respondents themselves. Graph 2.3: Distribution of the PENSTYP5 to PENSTYP8 AHM 2012 variable if PENSTYP1 to PENSTYP4 = no (% of yes answers in answers given for each PENSTYP)

31 Note: Sweden: PENSTYP8 includes PENSTYP4 as well. The maximum of the scale of the graph would be 400 % if each respondent received all 4 types of pensions. As for comparison to the core data, or to the AHM 2006, the possibilities are quite limited. PENSTYP1 to 4 (different varieties of old-age pension, in the 2012 module) are to some extent overlapping with AGEPENS from 2006, but the 2006 variable covers a larger range of pensions than just old-age pension, and its main purpose is to determine the age of the recipients, not the type of pension. PENSTYP6 (disability pension) has some overlap with OTHBENF=1 in 2006, but here again, the fit is far from perfect, as the 2006 variable also covers sickness benefits as well as disability pensions. For unemployment pension and survivor s pension (PENSTYP5 and PENSTYP7) there are no parallels to 2006 at all. Conclusions and recommendations PENSTYPE is a variable that in reality combines eight variables, and deals with a topic that is hard to compare across countries, due to different legal frameworks and pension systems. Furthermore, some of the types of pensions do not exist in all of the countries, which makes cross country results hard to interpret in some cases. There were some countries that had problems because very complex pension systems led to difficulties in coding the variable. The experience in the national statistics institutes on this variable is nevertheless for the most part good. We have no other LFS variable to compare to, so a hard data based assessment of the quality is not available. As the Austrian quality report noted, comparisons with administrative data sources would be difficult as well, given the particular conventions applied in the collection of this data set. In case of a repetition of the module it should be analysed whether all types of pensions proposed in 2012 can be collected by an AHM and if all types proposed can provide comparable and meaningful information at EU level. Merging eight yes-no variables into one, where each variable had to be accompanied by a list of possible pension schemes was a considerable effort for the data collection. In addition, in the absence of the code no answer/does not know, respondents and interviewers sometimes had to take difficult yes/no decisions. Unfortunately, this variable offers low possibilities for analysis at EU level. Some countries (Austria, Slovenia, and Norway) made a suggestion to spend less energy on a classification that differs a lot among countries, and which at the same time is inconsistent with

32 administrative data. If some simplification of the variables would be wanted for a possible repetition of the module, it could be explored whether the analysis of only old-age pensions would be sufficient, providing both ample analysis opportunities without an unnecessary burden on respondents. The current breakdown by scheme (statutory, occupational, personal, unknown) shows how different the organisation of pension systems are from country to country, even for a basic scheme like old-age pensions. 3. EARLYRET: Incidence of early retirement Short description The purpose of the variable is to know if old-age pensioners retired via an early retirement path or not. Early retirement includes the following possibilities: anticipated old-age pensions, disability pensions, early retirement pensions in case of reduced ability to work, early retirement pensions for labour market reasons, early retirement pensions due to seniority (long career or long contribution period), early pensions due to family reasons. In specific situations, full unemployment benefits and partial retirement pensions can be considered as early retirement. Early retirement is defined in relation to the standard retirement age for a given sex, occupational group, etc. Not all persons with low retirement ages should systematically be coded as receivers of early retirement pension, unless they were affected by further early retirement measures going beyond the normal rule for their profession 14. In the case of a sequence of pensions, it is the first one that is checked against early retirement 15. The variable does not have the same definition as the ESSPROS 16 indicator anticipated old-age pension beneficiaries. There are differences as well compared with the definition of early retirement in the LFS LEAVREAS variable, as the focus of the LEAVREAS variable is on economic factors as for instance difficulties in specific sectors of the economy, while EARLYRET combines personal and labour market reasons. Filter conditions and codes Asked to all persons who receive old-age pension (PENSTYP1=1 or PENSTYP2=1 or PENSTYP3=1 or PENSTYP4=1). Code 1 Yes 2 No Description 9 Not applicable (not included in the filter) Blank No answ er or does not know Analysis of the questionnaires This is a yes/no variable on a concrete measure (early retirement), that respondents would be assumed to remember and answer easily. It was not asked in In Italy, and also other countries, complex pension systems was the source of the uncertainties on if, in a given period and given age and given profession and for a given case, the retirement was early retirement or not. In the same countries, AGEPENS was collected after EARLYRET, and interviewers could not assist respondents nor check the consistency of responses (this is why in the Italian report there is a suggestion to consider collecting AGEPENS before EARLYRET). In some countries the distinction 14 Explanatory notes, pages Frequently asked questions document, page ESSPROS (European system of integrated social protection statistics) is built on the concept of social protection, or the coverage of precisely defined risks and needs including health, disability, old-age, family and unemployment; it records the receipts and the expenditure of the organizations or schemes involved in social protection interventions. See: for more information.

33 between the statutory scheme on one hand and the occupational scheme on the other hand (both old-age pensions) was not easily made by respondents, especially those thinking that in both cases employers do make contributions to the pension systems. In Bulgaria it was not easy to communicate to respondents what measures going beyond the normal rule for their profession are, in particular for professions with a lower retirement age (e.g. for policemen). Finland reported difficulties in interpreting early retirement in relationship with part-time pensions. Analysis of the results Due to high non-response the data from Germany and Norway must be used with care. In the case of Norway this is explained by the question not being asked to persons older than 66 years of age. The nonresponse rate at EU-28 level was at 4 %. Graph 3.1 shows the countries where it is most common to go into early retirement, based on the relation between those who answered yes and no. Graph 3.1: Distribution of the AHM 2012 variable EARLYRET (%) We see a very varied picture in this graph, with early retirement being close to non-existent in Bulgaria and the Czech Republic, whereas almost three in four pensioners are early retired in Italy and Ireland. Both Bulgaria and Italy report some difficulties in collecting this variable, as detailed in the previous section. At EU-28 level, roughly 4 in 10 persons receiving an old-age pension declared having used an early retirement path. The split line is increasing gradually throughout the graph, and there are no striking outliers. No clear geographical pattern is evident. Graph 3.2 will further analyse early retirement by sex. It shows that at EU level there are more men than women going into early retirement (the EU-28 value is 60 %). The high value for Malta can be explained by the specific gendered target population of the module. In general, the fact that more men than women are early retired can be explained by the higher retirement ages for men in many of the countries, which, once fixed, turns any retirement which takes place before that age into a relative early retirement. The result can also be explained by specific measures taken for certain occupations, in which the proportion of men and women could be unequal. In 13 of the EU countries the early retirement is however more frequent among women, with Hungary, Latvia and Estonia having the proportion of women among the

34 early retired higher than 60%. Graph 3.2: Proportion of males among EARLYRET=1 (%) One could very well expect an obvious correlation between using an early retirement scheme and the age of receiving an old-age pension for the first time. Graph 3.3 shows that this is not the case; there is a very weak pattern to be found between these two variables. Graph 3.3: The variables AGEPENS and EARLYRET plotted against each other (% and age in years)

35 The LFS variable LEAVREAS provides information on the main reasons for leaving the last job or business. Even if the definitions and the target of the two variables are different, the comparison allows an assessment of the comparability of data at EU level. Graph 3.4: AHM 2012 EARLYRET=1 among respondents who left the last job or business for early retirement reasons (LEAVREAS=06), (%)

36 Note: Data from Germany and Norway is excluded from the analysis because of high non-response rates in the EARLYRET variable. Data from Bulgaria and Iceland is not available for LEAVREAS=06. This graph confirms that data should be interpreted with care, as the match between the two variables (EARLYRET and LEAVREAS) differs significantly among countries, with Denmark, the Czech Republic, UK and Finland showing the smallest overlap among the otherwise related concepts. In the Danish case the reduced overlap is explained by a filter problem, as the persons with PENSTYP1=yes did not receive the question on early retirement. In Finland the low match is probably explained by the different time horizon and precision of the two variables: EARLYRET was asked after having collected precise information on several types of pensions, while LEAVREAS was a more general variable, and probably easier to answer. Conclusions and recommendations The main difficulty of the variable is to provide meaningful information at aggregate level, given the relative component it includes. Depending on the country, provisions with similar outcomes could either be considered regular policy measures or early retirement paths. For these reasons, the characteristic of early retirement can only be meaningful within a certain country, occupational group, age cohort or sex. On the other hand, the sample is too small to allow detailed views on such sub-groups. The variable shows great heterogeneity, with early retirement adding to the already different national situations and practices. The variable cannot be easily benchmarked against other (administrative or not) data sources. As a result, it cannot be interpreted in isolation, but only in the broader context of the national data on pensions, pension age, occupation, sex, etc. In case of a repetition of the module it should be further analysed whether this variable can provide comparable and meaningful information at EU level. 4. AGEPENS: Age at which a person first received an old-age pension Short description The purpose of the variable is to know at which age the person started to receive the first old-age pension. Filter conditions and codes Asked to all persons who receive old-age pension (PENSTYP1=1 or PENSTYP2=1 or PENSTYP3=1 or PENSTYP4=1). Code 2 digits Description 99 Not applicable (not included in the filter) Blank Analysis of the questionnaires No answ er or does not know It is the only numerical variable in the module, which opens up for more analysis than for the other ones. The data collection had some slight variations: some countries asked for the age when the person started to receive the first old-age pension, others collected the year and month, and then used the year and month of birth in order to derive this information. Those variations are not expected have a significant impact on the quality of the variable. The variable is identical in content, but not in coding, with the parallel variable in the 2006 AHM module. Analysis of the results

37 Due to high non-response the data from Germany and Norway must be used with care. Except for Germany, the response rate is very good at EU-28 level, at 95 %. There were no major reported problems in implementation, although a few respondents had difficulties recalling the date of receiving their first old-age pension, especially if old-age pension came after an unemployment pension. Graph 4.1: Distribution of the AHM 2012 variable AGEPENS (average age in years) The EU-28 average of receiving an old-age pension is 59 years. At the lower end we find Bulgaria, Poland, Romania, Slovenia, and Slovakia, all at 57 years. The highest average among the member states is in Sweden, at 64 years. Among all participating countries, Iceland and Norway come out on top, both with 65 years. The spread across the countries is quite high, with eight years separating the highest and lowest results. We see that the geographical pattern looks much like the pattern found in graph 1.1 (PENSION). This correlation is not surprising, as when the average age for being a pensioner is lower, the chance of being a pensioner and at the same time in the target group of the AHM increases. Graph 4.2: Distribution of the AHM 2012 variable AGEPENS (average age in years), by sex

38 In 13 of the EU countries (46 %) and 2 of the EFTA countries (66 %) the average age is the same for men and for women. In 13 other EU countries the average pension age is lower for women than for men. Only two countries show pension ages for men being lower than for women, those being Cyprus and the United Kingdom. A very clear west east divide appears when we look at the difference between average age of receiving the first old-age pension for men and women. The gender gap is most pronounced in Croatia, Slovakia, Slovenia, Poland and the Czech Republic, with smaller but still clearly visible differences in Romania, Hungary, Estonia, and Latvia and Lithuania, making all the 10 highest gender gaps eastern European. The countries that show no difference at all between men and women, or a higher age for women, are western European or Mediterranean (the United Kingdom, Sweden, Malta, Cyprus).We also see that for those countries with the largest gender gaps, it generally (in three out of five cases) is the combination of national average pension age for women being lower than the EU-28 average and at the same time the national average pension age for men being higher than the EU-28 average. However, in Slovenia and Romania both for men and women averages are lower than the corresponding EU ones. Another pattern is that as the pension age of women increases, it becomes closer to that of men. When the pension age of women is 60 years or more, the gender gap disappears, with the sole exception of Iceland. One possible way of assessing the quality of AGEPENS is to compare it to the expected duration of working life. The duration of working life indicator (DWL) 17 measures the number of years a person aged 15 is expected to be active in the labour market throughout his or her life. Graph 4.3: LFS AHM AGEPENS average by country plotted against LFS 2012 working life indicator (duration of working life), age expressed in years 17 This indicator is derived from demographic data and labour market data. It is published online at the Eurostat data base page as table reference lfsi_dwl_a.

39 Graph 4.3 shows a very reasonable correlation here, which is a support for thinking that AGEPENS is measured accurately: as the length of the working life increases, so does the pension age. Graph 4.4: LFS AHM AGEPENS average by country plotted against LFS AHM PENSION (years and %) Graph 4.4 shows the unsurprising pattern that as the pension age in a country increases, the number of

40 persons in the target group of the AHM that receive a pension decreases. This points towards a safe assumption that the quality of the AGEPENS variable is good. Graph 4.5. Difference in number of years of average AGEPENS by answer categories of EARLYRET (EARLYRET=yes minus EARLYRET=no), by sex (years) Graph 4.5 shows the difference in number of years between the average age for AGEPENS for those who did and those who did not use early retirement schemes. We see here, that in Germany, Denmark, Belgium, Finland, and Norway the average age when receiving the first old-age pension of those who did not use an early retirement scheme is lower than for those who did use such a scheme. The previous section has already explained the difficulties encountered in Germany, Denmark, Finland and Norway when collecting the EARLYRET variable. This is only an indirect argument to support the AGEPENS quality. Data should be used and interpreted with care. The following section, relating to the results in table 4.1 and graphs 4.6 and 4.7, deals with un-weighted data.

41 P25 P50 P75 Mean Mode Shape P25 P50 P75 Mean Mode Shape P25 P50 P75 Mean Mode Shape Table 4.1: LFS AHM AGEPENS un-weighted quartiles, mode, and mean, in years and curve shape, by sex All Men Women EU Neg BE BG CZ Norm DK Pos DE EE Norm IE Neg Neg Neg EL ES FR HR Norm IT CY LV Neg LT Norm LU Pos HU Norm MT NL Neg Neg AT PL Pos PT RO Norm Neg SI Norm Norm SK Norm FI SE UK Neg IS Neg NO CH The placement of the 25, 50 (median), and 75 percentiles, along with the mode and the mean, show what shape the age distributions have. If the mean, median, and the mode are equal to each other we have a normal distribution, as in the case of Lithuania and Romania. For a positively skewed distribution, the mean will always be the highest estimate of central tendency and the mode will always be the lowest estimate of central tendency. This we find in Poland. The exact opposite (the mean is lowest and the mode is highest), creates a negatively skewed curve, and this is the more common occurrence, with Ireland, Latvia, and the Netherlands. Only for two countries do we find the same curve shape both for men and for women (Ireland and Slovenia). Graph 4.6: Box-and-whiskers plot for the AHM 2012 variable AGEPENS

42 A box-and-whiskers plot is a powerful tool to describe both the central tendency as well as the spread of the values of AGEPENS. The length of the box represents the interquartile range (the distance between the 25th and 75th percentiles). The square symbol inside the box represents the mean. The horizontal line inside the box represents the median, and the vertical lines issuing from the box extend to the minimum

43 and maximum values. The most obvious difference between the countries in this respect is that some of the lines are very long, and some are quite short. The data range is almost four times larger for the country with the longest line than the shortest one. In Malta, the highest age at which anyone started to receive a pension was 65 years, whereas twelve other countries are at 69 years of age, which is the highest possible maximum point given that the data collection was made on a population aged 50 to 69. Spain, France, Italy, Sweden, and the United Kingdom have respondents who received their first old-age pension in their early thirties. 30 years of age was set as a lower limit in the data editing for the age of first receiving old-age pension; anything below this would be very implausible and highly likely an error in data coding. One should note that the spread of the data ranges are not correlated very much with the average ages in each country. Iceland is at the other extreme, with the shortest box-plot line, showing that the youngest age when reception of a pension started was in the late fifties. Sizes of the interquartile range boxes do also differ. As it also can be seen from table 4.1, the longest interquartile range, 8 years, is in Portugal. The shortest, 3 years, appear for Cyprus, Finland, France, Lithuania, Luxembourg, Malta, and Switzerland. One can also analyse relationships among the elements of the plots by noticing for example that France, which has a very long line issuing from the box, has a very short range, whereas Greece has the very opposite. This means that most of the data from France is concentrated in the middle range of age, with a few but very pronounced outliers, and Greece has no real outliers, but the age of receiving a pension is much more concentrated in the area between the 25 th and 75 th percentile. Graph 4.7: Histogram with imposed Gauss curve for the variable AGEPENS, all participating countries We see in graph 4.7 that AGEPENS is not normally distributed. The far end of the tails behave close to a Gauss curve, but we observe two interesting tops at each middle section, that fall again on both sides

44 towards the middle of the distribution, and with a very pronounced high point for 60 years of age. The graph clearly shows us that on European level there is a substantial chance of receiving a pension for the first time at the age of 60, and the two other main occurrences are at around 55 and 65 years of age. Graph 4.8: Distribution of the AHM 2012 variable AGEPENS for all participating countries, by sex (% in the population of the same sex receiving an old-age pension) Graph 4.8 shows that although the clear peak point for both men and women is 60 years, there are clearly more women than men at this point of the figure. Men have declared most frequently the ages of 60 followed by 65 as age of receiving their first old-age pension, while women declared most frequently the ages of 60 followed by 55 as age of receiving their first old-age pension. These three main data points constitute almost half of mass of the data. Almost four times more women get their first pension at the age of 60 than at the age of 65, whereas for men this relation is at only 1.4 times. Looking at the ages 55 and 65 we see a similar pattern, in that for women it is more common to be a pensioner at 55, whereas for men it is 3 times more likely to become a pensioner at 65 than at 55 years of age. Graph 4.9: AHM variable AGEPENS in 2006 and in 2012 (average age in years)

45 This variable was also used in the 2006 ad hoc module, although it then asked for the starting age of receiving an individual retirement pension, and not for an old-age pension. Graph 4.9 provides a scatter plot of the two results against each other. A deviation of plus/minus one year must be allowed for. Slovenia wrote in their quality report for 2012 that the pension system has been under revision since 2006, so comparability has been compromised, but for an explainable reason. This means that the results are consistent for most countries, but one should take care when using and interpreting data on this variable from Latvia, Slovenia and Norway, as table 4.2 shows. Table 4.2: Confidence limits for LFS AHM 2006 and LFS AHM 2012 variable AGEPENS, 2006 and 2012, for the outliers in graph 4.9 (average age in years) Mean estimate 95 % Confidence limits Mean estimate 95 % Confidence limits Latvia Slovenia Norway As we see in the overview of confidence limits, the changes observed for Latvia and Norway cannot be explained by wider confidence intervals either. Therefore comparative analysis between 2006 and 2012 should be avoided for these countries. Conclusions and recommendations The variable had overall very good response rates. None of the national quality reports point at major deviations or problems that have impact on comparability. Suggestions for improvement came from Austria, whose report suggests that personal and occupational pensions should have been excluded from the filter of this question, or at least asked separately from the other old-age-pensions, as this would facilitate comparison with administrative data. France and Greece report that some respondents had problems determining or recalling the exact age when respondents started to receive the pension. These

46 ideas can be taken into consideration if the variable is used again in a future module, but the general assessment of this variable remains that it is of good quality and with few reported problems. 5. REASNOT: Main reason for not staying longer at work Short description The purpose of the variable is to identify the main factor that determined the person to leave the labour market. The moment of leaving the labour market is defined as the moment of leaving the last job. The list of possible reasons for not staying longer at work included the following: favourable financial arrangements, impossibility to find another job, reaching the maximum retirement age, reaching pension eligibility, job-related reasons, personal reasons, etc. Among them, only the main reason was collected. The spontaneous answer early retirement was followed by the repetition of the question in order to collect the reason for not staying longer at work 18. Filter conditions and codes Asked to all inactive persons (simplified ILO status: WSTATOR=3, 5 and SEEKWORK=3) receiving a pension (PENSION=1). Code Description 1 Favourable financial arrangements to leave 2 Lost job and/or could not find a job 3 Had reached the maximum retirement age 4 Had reached eligibility for a pension 5 Other job-related reasons 6 Ow n health or disability 7 Family or care-related reasons 8 Other 9 Not applicable (not included in the filter) Blank No answ er or does not know Analysis of the questionnaires A similar variable was collected in 2006, the AHM 2006 variable REASRET. In spite of the different name ( Main reason for not staying longer at work - REASNOT in 2012 and Main reason for retirement or early retirement - REASRET in 2006), the objective of the 2006 variable was the same as of the 2012 variable: to get the main factor that made person exit from work using categories of responses more adapted/oriented to older persons 19. In 2006 however, the exit from the labour market was understood as the retirement moment and this could impact on the comparability with the 2012 data. The fact that in 2012 the variable refers to the moment of leaving the last job decouples to a certain extent the variable from the analysis of pensions, and could in principle render the data less sensitive to variations resulting from national definitions of pensions. For example, persons who took a side job even after receiving a pension should have referred to the moment of leaving the last job in their answer, even if the moment of retirement could have been defined as the moment of receiving a pension. This nuance limits the comparability of the 2006 and 2012 module only to those persons who stopped work at the same age as when they first received a pension. As we saw in the end of the AGEPENS chapter, only seven of the 31 participating countries show no difference on average between the age of leaving the last job (core 2012 data) and the age of first receiving a pension (AHM 2012 data). In practice however, the 18 Explanatory notes, page

47 overlap between the 2006 and 2012 variables might have been greater than intended: because in most countries variables on age of first pension and reason for leaving the last job were consecutive, and because several answering categories did refer to the pension or retirement, it is likely that some respondents actually referred to the moment of leaving the job that provided them the access to a pension. The 2012 variable has therefore an identical purpose as the LFS LEAVREAS variable on main reason for leaving the last job or business. Only the populations receiving this question are slightly different. For those respondents that had to answer both the LFS LEAVREAS variable and the AHM REASNOT variable, some level of confusion could be understandable. Questions on the difference between codes 3 and 4 have been raised during the preparation 20 of the module. The intention of codes 3 and 4 was to distinguish those who are forced to leave the labour market because of their age, and those who leave the labour market as soon as their age allows it. In several countries, as for instance the Czech Republic, code 3 was not relevant. However, most countries have included the response category in their questionnaire, because it could in principle have been applicable at low scale, or in specific sectors of activity or occupations, for example as a result of specific collective agreements between trade unions and employers 21. Finland reported translation problems with codes 3 and 4. In general, categories 3 and 4 were cognitively demanding. Austria and Italy report that they added answer categories. In Italy the category fear of pension age being raised was collected and is included in the other reason category. Even without added answering categories, the variable was rather demanding. There is a clear risk of low data comparability. There is as well a subjective component that cannot be ignored, because in the case of a mix of reasons for not staying longer at work, it is the respondent who decides which was the main reason. Analysis of the results Germany, Belgium and Iceland have important shares of non-response, and their data should be interpreted with care. In Belgium this is due to the data collection methods used. A variable with ten possible values and a distribution that differs markedly among the participating countries is challenging to interpret in any form. Graph 5.1: Distribution of answer categories for the AHM 2012 variable REASNOT (%) 20 See Frequently asked questions document, page Explanatory notes, on page 14, provides the longest explanations for code 3.

48 Graph 5.1 shows the answering categories in their decreasing order of importance at EU-28 level. Two main reasons present themselves clearly as the main ones: had reached eligibility for a pension (37 % for EU-28 also used for ordering the countries) and own health or disability (20.9 % for EU-28). Not counting the non-response category and the own health or disability, we see that reached eligibility for a pension comes out as the same size as the sum of the remaining six answer categories. That being said, the graph shows a spread on this value of the variable across the EU countries from above 85 % in Malta and Czech Republic to 12.9 % in Estonia. Thirteen of the 31 countries show higher values on own health and disability than on reached eligibility for a pension. Moreover, graph 5.1 shows considerable differences among the categories related to age (categories 3 and 4). This can be the reflection of the different realities, but can also be influenced by the imperfect collection of cognitively demanding categories. There is no comparable 2006 data to use for benchmarking. The existing LFS variables refer to the last job, but when retirement does not coincide with the last job it is not easy to check the previous occupation, nor the activity sector. There are not many options available for confirming this at European level. In this light, there is a risk that the variable has failed to fully provide comparable information on the role that the age factor plays in the retirement decision. Graph 5.2 provides an analysis of the frequency of main reasons for not staying longer at work, by sex. At EU-28 level, answers were quite balanced with two exceptions: family or care related reasons was a more frequent answer among women (and not only at EU level but in each of the EU countries), while the favourable financial arrangement to leave was more frequent among men at aggregated level. These results match the distribution by sex in other core variables on reasons for reduced participation on the labour market. Graph 5.2: Distribution by sex of the answer categories for the AHM 2012 variable REASNOT, EU-28 (%)

49 Unfortunately, there is no possibility to compare the 2012 and 2006 data, mainly because the answering category had reached eligibility for a pension was not collected in 2006, while being the most frequent category in the EU-28 in Conclusions and recommendations Overall the variable shows a good response rate, with the exceptions of Belgium, Germany and Iceland. There are no reports from the participating countries on major problems in implementation. Two answer categories are clearly predominant: reached eligibility for a pension and own health and disability, and this could point towards similar implementation across the countries in Moreover, the analysis by sex provides the same main results as similar core LFS variables. However, country differences are significant. Given that the ambition of the variable was to collect eight main reasons which are similar, but different from the LEAVREAS ones, full comparability among countries cannot be guaranteed. Moreover, comparability to 2006 is not possible. In case of a repetition of the module it should be re-assessed whether the variable can add a lot of relevant information to the already collected LFS LEAVREAS variable. Recent analysis performed in the overall exercise of evaluating the current system of AHMs has stated that it is not a recommended practice to add a variable in the AHM with the main purpose of increasing the target population of an LFS variable. 6. WORKLONG: Wish to stay longer at work Short description The purpose of the variable is to establish whether the person would have preferred to stay longer at work (be it in the last or any other job) at the moment of leaving the labour market / last job. The variable expresses the preference of the respondent at the moment of leaving the last job, so spontaneous answers I would have liked to continue working but it was not possible for me to stay longer at work should be coded 22 as yes. 22 Explanatory notes document, page 16.

50 Filter conditions and codes Asked to all inactive persons (simplified ILO status: WSTATOR=3, 5 and SEEKWORK=3) receiving a pension (PENSION=1). Code 1 Yes 2 No Description 9 Not applicable (not included in the filter) Blank No answ er or does not know Analysis of the questionnaires The question is a binominal character variable, asking for yes or no answers. The purpose of the variable was self-explanatory enough and there were no questions on it during the implementation of the survey. However, the variable is subjective and hypothetical, because it assumes that in principle it was possible for everybody to stay longer in employment. There is an additional complexity brought by the any other job mentioned in the definition. It was most likely intended to cover the cases where a last job was not easy to identify, but there was rather a succession of shorter contracts or spells of inactivity combined with periods of employment. When interpreting the data, users need to have in mind that for most respondents the variable corresponds to the last job they had, and for them it measures the general wish of continuing working in the same job. For those with a less clear transition moment, the variable reflects the general wish to have a job, or the general wish to continue working. One country (Italy) reports some problems in answering the question for persons who are economically inactive and receiving only survivors pensions. We do however see that the Italian non-response is 1%. Even if the likelihood of these cases is low, the example is useful for depicting the imperfect overlap between the labour market and pension receivers. This concerns (to a small extent) other variables (like REDUCHRS or REASNOT) as well. In Romania, the variable was not collected for those who answered had reached eligibility for a pension code 4 in the REASNOT variable. The reason is that code 4 in REASNOT, when translated as preference for leaving the labour market when reaching eligibility for a pension 23 can be seen as equivalent to WORKLONG=2: does not wish to stay longer at work. As the question was very sensitive to wordings, data should be interpreted with care. It is plausible that in some cases the REASNOT=4 triggered a negative answer to the question on the wish to stay longer at work, and indeed at EU-28 level (see graph 6.2) approximately 12 % of those leaving the labour market when reaching eligibility would have liked to stay longer at work. As detailed in the previous section, it is the REASNOT variable that is likely to suffer from lack of comparability among all its answering categories, rather than WORKLONG. Analysis of the results Due to high non-response the data from Belgium, Germany, Greece, and Iceland must be used with care. In Belgium this is due to the data collection methods used. In Greece the high non-response rate is more related to the difficulty for the respondents of being sure on answers corresponding to situations that were complex (and sometimes difficult) at the time of leaving the labour market. In graph 6.1 we see that a majority of pensioners in Europe did not wish to stay longer at work at the time of leaving their last job. The EU-28 average for wanting to continue working is slightly above one quarter (28 %) of the respondents, whereas two thirds said no. The remaining 6 % are non-respondents. Only two countries have a yes proportion over 50 %, those being Portugal and Estonia, thus having a majority of their pensioners specifically stating that they would have liked to continue working when they left the labour market. On the other side of the spectrum we find Poland and Slovenia, with a proportion of yes of less than one in ten respondents. There is no obvious geographical pattern. 23 As it was formulated in the Romanian questionnaire

51 Graph 6.1: Distribution of answer categories for the AHM 2012 variable WORKLONG (%) Even if the variable is not completely correlated with the REASNOT variable, it can be expected to be a certain correspondence among a wish to stay longer in employment and the real reasons for having left. An analysis at EU-28 level (see graph 6.2) shows that, in broad lines, this correspondence exists: those who left the labour market because they could not find a job are more likely to have wished to stay in employment than the ones leaving as soon as they could reach eligibility for a pension. There are no noticeable gender patterns. However, it has to be said that those correspondences are weaker than expected and they rarely hold at country level. For example, it was expected to have a higher proportion of WORKLONG=1 among those who were forced by law to leave their job (REASNOT=3). As this inconsistency results from the combination of two variables, it is more likely that the variable REASNOT has a limited comparability, because of the reasons exposed in the previous section. Graph 6.2: Percentage of AHM 2012 variable WORKLONG=1 by answering categories of AHM 2012 variable REASNOT and by sex, EU-28 (%)

52 There was no parallel variable in the 2006 module, and there is no link to any of the core variables. We therefore do not have alternative ways of making a comparative analysis of this variable. Conclusions and recommendations The variable was relatively easy to collect. However, it was subjective, hypothetical and very sensitive to translation wordings, and for these reasons its comparability at EU level cannot be fully guaranteed. It is recommended to further test the variable if it will be repeated. 7. REDUCHRS: Reduced working hours in a move towards full retirement Short description The purpose of the variable is to know whether the person reduced working hours in a move towards full retirement, and in case of receiving old-age pension, whether this happened before or after receiving the first old-age pension. Both voluntary and involuntary, and both formal and informal reduction of the working hours are collected. A transition from a full-time to a part-time job is considered a reduction of working hours. For the economically inactive persons the reference is to the past, before leaving their last job. For the employed persons, the reference period is the present. For persons who are in employment but do not receive a pension code 2 cannot be applicable, and as a result the question is of a yes/no type for this category. The same yes/no question should apply for persons who are inactive, receive a pension (PENSION=1) but their pension is not an old-age pension. Filter conditions and codes Asked to all employed (ILO status: WSTATOR=1, 2) aged or to all inactive (simplified ILO status: WSTATOR=3, 5 and SEEKWORK=3) receiving a pension (PENSION=1) and aged

53 Code Description 1 Yes, before receiving the first old-age pension 2 Yes, since or after receiving the first old-age pension 3 No 9 Not applicable (not included in the filter) Blank No answ er or does not know Analysis of the questionnaires In the review phase it turns out that the variable had a too complicated wording and a too complicated filter. It was often difficult to translate in a move towards full retirement, and this sometimes resulted in an expression which was too abstract for the respondents to understand. In Slovakia, Latvia and Croatia, the reduced working hours are not an existent form of leaving the labour market, and therefore REDUCHRS=3. Furthermore, distinguishing between codes 1 and 2 depending on the moment of receiving his/her first old-age pension was considered demanding. The variable was addressed to employed persons, with or without a pension, aged and to inactive persons with a pension, aged The answering categories should have been different for those not receiving a pension and for those receiving a pension other than an old-age pension. Managing these filters proved to be difficult. A suggestion was to expand the filters (age, type of pension) in order to facilitate the data collection. Analysis of the results Due to high non-response the data from Germany, Ireland, Greece, and the United Kingdom must be used with care. The high non-response in Ireland is due to a filter problem omitting those persons that were working without receiving a pension. The high non-response in the United Kingdom is also due to a filter problem: inactive persons were excluded. The non-response rate in the EU-28 is 11 %, a rather highvalue. Reduced working hours in a move towards full retirement is uncommon at EU level. The next answering category was either don t know or no answer. In this circumstance, further break-down of the yes answers is troublesome. The remaining categories are so small that they are close to meaningless to analyse. Data should be interpreted with care. Graph 7.1: Distribution of answer categories for the AHM 2012 variable REDUCHRS (%)

54 Graph 7.1 shows the distribution of the variable in the participating countries, ordered by the importance of the yes answers. What we do see is that only five of the 31 countries have values of 10 % or more for Yes, before receiving the first old-age pension (the Netherlands, Belgium, Denmark, Finland, Switzerland), with the Netherlands totalling 18 % of the yes answers, followed by Belgium, with 15 %. Another seven countries have yes answers representing from 5 % up to 9 %. For the remaining 19 countries this scheme is practically non-existent. For the yes, after receiving the first old-age pensioncategory there are four countries who show values larger than 5 %, with Finland and the Czech Republic on top with 8 % each. 20 countries report 2 % or less. Graph 7.2 further confirms that the breakdown of the variable was too detailed and that the variable was very dependent on translation nuances. It is difficult to interpret that the reduced working hours took place before or after receiving the first old-age pension, as the values at EU-28 level are too low. In addition, it is relatively surprising that among persons having a part-time job there is still an clear majority (82 % in the EU-28) answering no to the question on reducing working hours in a move towards full retirement. The result is plausible, as for these respondents the reference period is the present, and in many cases the reduced working hours took place before the retirement plans. However, question marks remain on the value added from this variable. Graph 7.2: The distribution of the AHM 2012 REDUCHRS variable by economic activity and type of job, EU-28 (million persons)

55 In 2006 there was also a variable with the same name, but the filter conditions were different, and so were the answer categories. It is not possible to entirely set up the 2012 filter for the 2006 data, or vice versa, because the 2006 data does not have the PENSION variable that 2012 uses, and the 2012 data has a more restricted age group than the 2006 variable. For the sake of comparison we can however collapse the answer categories to plain yes and no, and blank, and obtain more comparable, broader categories. Graph 7.3: AHM variable REDUCHRS in 2006 and in 2012, proportion of no answers (%)

56 We find a clear cluster in the upper right corner, which simply confirms the scarcity of the reduced working hours in a move towards retirement at EU level both in 2006 and in However, we do also have some pronounced outliers (Bulgaria, Slovakia, Germany, Poland, Slovenia, and Estonia) that confirm that time comparison should be avoided. Conclusions and recommendations The variable provides very little comparable information at EU level. Its form was clearly not adapted to the expected distribution of the variable. In the case of a repetition, it is recommended to not use any of its previous forms for this variable. 8. STAYWORK: Main reason for staying at work Short description The purpose of the variable is to identify the main factor that makes a person who receives a pension to stay in employment. Filter conditions and codes Asked to all employed (ILO status: WSTATOR=1, 2) receiving a pension (PENSION=1). Code Description 1 To establish or increase future retirement pension entitlements 2 To provide sufficient personal/household income 3 Combination of 1 and 2 4 Non-financial reasons, e.g. w ork satisfaction 9 Not applicable (not included in the filter) Blank No answ er or does not know Analysis of the questionnaires

57 The variable has the purpose to find out why persons receiving a pension want to continue working. In several countries the question on main reason for staying at work was asked without any explanatory test, so most respondents could not see the link between this question and the fact that they already declared receiving a pension. In other countries (Austria, Cyprus, Germany, Spain, Hungary, Italy, Luxembourg, the Netherlands, Portugal, Romania, Switzerland) the context of the question was clearly explained, with a question similar to the following: you are receiving a pension or a pension-type benefit. Why are you still working?. In Estonia, the question was formulated differently Do you have a financial incentive that makes you as pensioner stay at work? If yes, name it. It can therefore be concluded that the questionnaires are not fully comparable among countries, and this can be expected to have some impact on the answers. The reasons proposed to the respondents are classified either as being of a financial nature (codes 1 to 3) or of a non-financial nature: code 4. No countries report substantial problems with this variable, but there is a suggestion that code 4 should have been split, and have asked for work satisfaction as a separate answer category. Analysis of the results Due to high non-response the data from Germany, France, Croatia, and Switzerland must be used with care. At EU-level, the non-response rate is at 11 %, which invites to analyse with care. Graphs 1.3 and 1.4 have shown the proportion of persons who are both employed and receive a pension, for EU-28 and at country level. Less than 7 % of the EU population aged is in employment while at the same time receiving a pension. This value corresponds to 8 % of the population of the module. At EU-28 level, there are important differences among countries: more than 15 % of the AHM 2012 population combines work and a pension in Estonia, Sweden and the United Kingdom, while the same percentage is under 4 % in Greece, Spain, Hungary and Belgium. The interpretation of the variable s outcome should me made bearing in mind not only the different labour market participation, but also the existing differences among pension systems. Graph 8.1: Distribution of answer categories for the AHM 2012 variable STAYWORK (%)

58 Graph 8.1 is ordered by the two main categories of reasons: financial (income, future pensions, and the combination of the two) and non-financial (as for instance job satisfaction). The magnitude of the individual answer categories within the financial groups is visible as well, starting with the most important category: to provide sufficient income (code 2). In all countries, current income situation is more important than the future pension situation, with an EU-28 average of 37 % as main reason for continuing work in order to provide sufficient income, and only 7 % in order to establish or increase future retirement pension entitlements. The EU-28 sum of financial reasons is 59 %. Non-financial reasons are at 29 %. Geographically the split is almost a perfect East and South versus West and North divide: financial reasons in the East and South and non-financial reasons in the North and West. The only exception from this pattern is Slovenia that has placed itself in the North-West group. A more detailed analysis can be made at EU-28 level by the type of job (full-time versus part-time) of persons in employment and receiving a pension. Graph 8.2 shows that part-time work is more associated with staying in employment mainly for non-financial reasons, while full-time work is more linked to the financial reasons (and among them, the purpose of establishing future pension entitlements). This result is not surprising, and it is driven by those countries where part-time work among elderly is a frequent phenomenon, which is not the case in all EU countries. Graph 8.2: Distribution of the AHM 2012 STAYWORK variable by type of job (full-time or part-time), EU-28 (million persons) In 2006 the variable FININCTV (main financial incentive to stay at work) collected information on reasons for continuing work after receiving a pension, with a similar split financial / non-financial, for a reasonably similar filter condition: 2006 filtered on receive an individual retirement pension, 2012 on receives a pension. However, the variable was different, because it was focused on financial incentives in 2006: for a future pension, for current income, none. This aspect, combined with the small group that received the question (less than 10 % at EU level) discourages the analysis of trends from 2006 to 2012 at EU level. Plotting per cents of answers of non-financial reasons against each other for the two years further confirms this message. Graph 8.3: FININCTV 2006 versus STAYWORK 2012, non-financial reasons (%)

59 Conclusions and recommendations The variable has the ambition of shedding light on the reason why persons continue to work even after receiving a pension or pension-type benefit. The picture that emerges is that financial reasons are more important in South-Eastern European countries and less important elsewhere. However, pensions are not always comparable as they can either be old-age pensions but also of other types. Being in employment is neither fully comparable, as some persons work full-time and others work part-time. The fact that these respondents represent only 7 % of the EU population does not allow for deeper analysis on comparable pensions or comparable working times. In these circumstances, we can conclude that the variable adds little information on the patterns of working while receiving pension. It is recommended to further test the variable (also paying attention to the nuances of the questionnaire) before re-proposing it in a future module. 9. PLANSTOP: Plans to stop working in the future Short description This is a forward looking variable, intending to estimate the planned time span for stopping all work for pay or profit. Spontaneous answers like there is no planned age / not yet decided should have been collected as blank answers. Filter conditions and codes Asked to all employed (ILO status: WSTATOR=1, 2) receiving a pension (PENSION=1).

60 Code Description 1 In up to 1 year 2 In more than 1 year up to 3 years 3 In more than 3 years up to 5 years 4 In more than 5 years up to 10 years 5 More than 10 years 9 Not applicable (not included in the filter) Blank No answ er or does not know Analysis of the questionnaires The most significant answering category is the non-response one, which makes the variable less useful than the others. 35 % of all EU-28 answers are blank answers. This is as well reflected in the comments from the participating countries, where eight countries stated having problems either due to proxy answers, or simply because the respondents did not plan or could not plan (due to a reform of the pension system or because of other reasons) their retirement age. Unfortunately, the blank category cannot distinguish among those who meant no planned age for retirement and those who refused to answer. Moreover, there are countries where the blank category represents 0 %, hinting at different approaches in dealing with or coding the no response. Therefore, the analytical potential of this variable is very limited. Having bands of years as answer categories might have helped a bit, but assuming the existence of a plan going beyond the next decade might have been one of the triggers of non-response. In any case, forward looking and hypothetical questions are usually not the best strategy for a survey. Analysis of the results Due to high non-response the data from Bulgaria, Germany, Estonia, Ireland, Greece, Spain, France, Croatia, Italy, Hungary, the Netherlands, Austria, Portugal, Romania, Sweden, the United Kingdom, Iceland, Norway, and Switzerland must be used with care. Graph 9.1 shows that the most common category is not subject-related, but the blank one. Croatia has the highest non-response rate, at 95 %, followed by Romania, at 83 %. Only twelve of the participating countries have a response rate higher than 85 %. It is clear that an analysis is, at best, only possible at country level. Graph 9.1 Distribution of answering categories of the AHM 2012 variable PLANSTOP (%)

61 Conclusions and recommendations One in three respondents in the EU have chosen not to answer the question about future retirement plans, either because they do not have a plan or because they did not feel comfortable to speak about such plans. The failure could also be due to a badly chosen target group for the question. Italy suggests that the question should rather be asked to all persons in employment, and not just to those receiving a pension. This is clearly the least successful variable in the module, and should therefore not be repeated. As a general rule, variables on future plans of the respondents should be avoided. 10. BUILDPEN: Information on pension rights the person has acquired so far Short description The purpose of the variable is to get information on the pension rights the person has built-up so far. The concept to be captured is whether the person is currently acquiring pension rights, or has acquired some in the past (of one or more of the types indicated in the variable). If so, an affirmative answer should be given even if the respondent has not yet the right to receive the pension. Filter conditions and codes Asked to all persons aged who do not receive an old-age pension (PENSION=2, blank or (PENSTYP1 to PENSTYP4=0)), and who either are working or have worked after the age of 50. Code BUILDPEN1 1: Yes; 0: No BUILDPEN2 1: Yes; 0: No BUILDPEN3 1: Yes; 0: No BUILDPEN4 1: Yes; 0: No Description Old-age pension. Statutory scheme Old-age pension. Occupational scheme Old-age pension. Personal scheme Old-age pension. Scheme unknow n 9999 Not applicable (not included in the filter)

62 Analysis of the questionnaires Several of the participating countries noted that this is a very complex issue to measure and reported difficulties in implementation. The most problematic aspect proves to be the understanding of pension rights the person has acquired so far. The variable aimed at collecting information on any contributions made in the past for a pension in the future. When analysing the questionnaires used in different countries it became obvious that the variants differ considerably. For instance, in Belgium, Denmark, and Luxembourg the question was on variants on gained pension rights. The UK questionnaire asked for expectation of future pension rights. Bulgaria, Spain, France, and Portugal asked the intended question, that is, on contributions to the pension system. In Poland the question posed was rather being able to retire at the time of the question than having built pension rights (for the future retirement), which of course had a major impact on the results. Some respondents thought that the pension rights are in force only at the moment when a pension is received. The question was however asked to all those not receiving an old-age pension. Other respondents could not know if their past contributions will result in real pension rights. Finally, some respondents considered uncertainty in the pension system as a lack of guarantee of their rights acquired. The breakdown by scheme was detailed and difficult to collect due to employees often not knowing what form of pension rights they have (Bulgaria, Greece, Lithuania), or several of the answer categories not being applicable (Czech Republic, the United Kingdom, Greece). Persons in occupational schemes are not always aware of being covered by this, or they may be confused with public sector old-age schemes. Also, the filter conditions were seen as less than optimal (Austria), giving very high values on BUILDPEN1, which does not open for very interesting analysis. It is also a weakness that there was no don t know response item in the answering categories. Analysis of the results Due to high non-response the data from Bulgaria, Denmark, Germany, Greece, Spain, France, Italy, Lithuania, Romania, Sweden, the United Kingdom, Iceland, Norway, and Switzerland must be used with care. Graph 10.1: Proportion of yes answers for the AHM 2012 variable BUILDPEN, by oldage pension scheme (%)

63 The lay-out of graph 10.1 is the same as in graph 2.1, in that it shows the sums of pension rights. As one person can have more than one type of pension rights, the totals exceed 100 %. We see that most countries have good coverage on at least one type of pension types, typically statutory. Denmark, the Netherlands, Sweden, Iceland, and Switzerland show high occurrence of occupational pension rights as well. Poland shows close to no coverage on any type, due to the already explained reason. Unfortunately, analysis at EU level is hindered by differences in translating and adapting the question to national circumstances. It is recommended to keep the analysis at national context, linking each time results and the exact wording of the question used. Conclusions and recommendations The variable should be reviewed critically in case of a repetition of the module. Future explanatory notes should aim at providing examples for the abstract pension rights that the person (otherwise not receiving an old-age pension) has acquired so far, or a different wording focussing the variable on past contributions rather than on hypothetical rights. The implementation in 2012 can serve as a starting point, and identification of best practices should be possible. We also re-emphasise what was already said in the introductory chapter on the target population: the choice of the target population has additional disadvantages for analysis of countries with a lower retirement age. This is because people who did not work after the age of 50 nevertheless have reasonable chances of receiving a pension, either at the time of the interview or in the future, on the basis of their work experience before the age of 50. It is unfortunate that the survey excluded those who stopped working before the age of 50 from the analysis of contributions to the pension system (BUILDPEN). 11. CONTWORK: Expectations to continue working or looking for a job after receiving old-age pension Short description The variable is forward-looking, on whether those respondents building up old-age pension rights expect to continue staying on the labour market when they receive a pension. Presence on the labour market includes both working as well as looking for a job. Moreover, in the case of expected continuation of

64 work/looking for a job, the variable distinguishes among the financial reasons for working and nonfinancial reasons for working. Filter conditions and codes Asked to all active persons (simplified ILO status: WSTATOR=1, 2 or (WSTATOR=3, 5 and SEEKWORK=1, 2, 4) who are either building up or have built up old-age pension rights (BUILDPEN 0000, 9999), and who are not receiving a pension (PENSION=2, blank). Code 1 Yes, for financial reasons 2 Yes, for other reasons Description 3 No, stop immediately w hen receiving old-age pension 4 No, stop before receiving old-age pension 9 Not applicable (not included in the filter) Blank No answ er or does not know Analysis of the questionnaires Many national reports (Bulgaria, Switzerland, Estonia, Italy, the Netherlands, and Slovakia) reported difficulties on collecting the information on expectations to continue working, because the question was hypothetical and proved in many cases to be impossible to answer. Furthermore, the variable intended to combine not only intentions to work, but also intentions to look for work, which made its collection difficult. The complicated filter added to the existing problems of this variable s collection. Analysis of the results The collection of the variable has resulted in a high non-response rate in many countries: the rate is 19 % in the EU-28, with values above 50 % in two of the EU countries, namely Romania (61 %) and Bulgaria (52 %). Non-response rates higher than 15 % were as well recorded in the following: Lithuania, Iceland, Norway, Switzerland, Denmark, Greece, Italy, the United Kingdom, Germany, Spain, Sweden and France. Graph 11.1: Distribution of answer categories for the AHM 2012 variable CONTWORK (%)

65 Graph 11.1 shows the distribution of the CONTWORK variable, with most of the respondents answering no to the question on expectations to continue working after receiving old-age pension: the EU-28 sum of the two possibilities for no is at 56 %. Except for the Czech Republic, the group answering stop immediately when receiving old-age pension is clearly larger than stop before receiving old-age pension. For those who do plan to continue either working or looking for a job after receiving an old-age pension, financial reasons (EU-28 at 16 %) are more important than other reasons (EU-28 at 9 %). Continue working for financial reasons is most important in Estonia, Latvia, Cyprus, Slovakia and Lithuania and least important in Slovenia, Austria, Spain, Denmark and Norway. Graph 11.2 compares the importance of financial reasons for continue working (CONTWORK) with the importance of financial reasons for staying at work (STAYWORK). The numbers shown are how many of those who answered yes to the question on expecting to continue work or look for work do this for financial reasons (CONTWORK), and how many of those who stay at work do this for financial reasons We expect a high correlation between these variables, although they both were affected by high nonresponse rates. In this light, the interpretation of graph 11.2 should be made with care. Iceland and Croatia are suppressed from the graph due to very small answer groups on some of the values. Graph 11.2: Financial reasons in answers declaring expectations to continue working for CONTWORK and STAYWORK (%)