La economía bipolar (la "nueva normalidad" que la crisis nos legó) – Parte II

Anexo – Informes de Organismos Internacionales (en versión resumida)

Del Paper – Un análisis sobre la desigualdad de los ingresos (ganadores y perdedores de la crisis financiera mundial) – La Economía del Malestar (el fin de la cohesión económica y social), publicado el 15/7/11

– Eurostat Statistical Books

Income and living conditions in Europe – 2010

Capítulo 5 – Income poverty and income inequality

5.1 Introduction

5.1.1 Aim of this chapter

This chapter focuses on the financial dimensions of poverty and inequality.

Income is an important variable for Europe"s households. People are naturally concerned with how much they receive each month in the form of earnings (from employment or self-employment), pensions, government transfers (such as unemployment benefits, family benefits or sick pay), and from their savings. In this chapter, we examine the distribution of income in the 27 Member States of the European Union (EU-27). Are there large differences within and across countries? In which countries are the differences largest? Particular concern attaches to those households which, according to the EU definition, are "at-risk-of-poverty" as this is one of the three indicators that form the new EU Headline Target on social inclusion adopted by the June 2010 European Council in the context of the Europe 2020 Agenda.

The chapter has four main aims:

1. to identify (in the remainder of Section 5.1) the particular role of the EU-SILC data as a source of evidence about income inequality and poverty

2. to analyse (Section 5.2) headline indicators for income poverty and inequality that has been agreed at EU level, with particular reference to the cross-country patterns

3. to examine (Section 5.3) changes over time in income inequality and poverty

4. to consider (Section 5.4) how the EU indicators based on the EU-SILC data can be used in monitoring the Europe 2020 Agenda.

From the chapter, the reader will, we hope, learn about the income dimension of poverty and social exclusion in the EU-27, as shown in the EU-SILC data, and how this evidence relates to that from other sources. The chapter looks back in time, to see how (income) poverty and inequality have changed in recent years, and forward in time to consider the implications of the Europe 2020 Agenda.

5.1.2 Role of EU-SILC

As described in Chapter 2, EU-SILC is not a common survey across countries. In this respect, it differs from its predecessor, the European Community Household Panel (ECHP), which was based on a standardised questionnaire (the ECHP ran from 1994 to 2001 in most of the then 15 EU countries, providing comparative data on income and living conditions for the years 1993 to 2000). EU-SILC is a harmonised data framework involving ex ante standardisation but allowing countries a large degree of flexibility in the underlying source(s) and some flexibility in the concepts and definitions. For example, while in the ECHP the income reference period was the previous year, the EU-SILC income reference period may be a fixed 12-month period (such as the previous calendar year or tax year) or a moving 12-month period (such as the 12 months preceding the interview) or be based on a comparable measure. (2)

EU-SILC is not based on a common questionnaire used in all countries, but on a common ex ante framework that defines the harmonised "target variables" to be collected/produced and provided to Eurostat by the national statistical institutions. The aim of this procedure was to facilitate EU-SILC being embedded within the national statistical systems, allowing the results to be produced at a lower additional cost in terms of resources, while serving a common EU purpose. The intention in allowing a degree of flexibility is to secure, not input harmonisation, but output harmonisation. Output harmonisation in EU-SILC is sought through the use of common guidelines and procedures, common concepts (e.g. that of "household") and of the information produced. In this respect, it may be contrasted with ex post standardisation, where data from different sources are processed to put them as far as possible on a common basis, as in the Luxembourg Income Study (LIS). In this case, the aim is again output harmonisation, but without an ex ante framework. The scope for ex post standardisation is limited by the constraints imposed by the original survey designs or other sources (such as data from administrative/ register records).

Finally, EU-SILC may be contrasted with meta analyses that take, not the microdata, but the results from different sources and seek to put them in a common framework. In the study of income inequality, this approach was particularly developed by Simon Kuznets (1963). In the case of both income inequality and poverty, a lead was taken by the OECD, who published the study by Sawyer (1976), assembling results from some dozen countries, and later Atkinson, Rainwater and Smeeding (1995) which covered 17 countries.

The current OECD work involves "a regular data collection … (at around 5-year intervals) through a network of national consultants" (2008, p. 47). The national experts "apply common conventions and definitions to unit record data from different national data sources and supply detailed cross-tabulations to the OECD" (2008, p. 41). This procedure of "customising results" may be seen as lying between that of LIS, which produces microdata, and that of Kuznets, where the results are pre-defined. It has the advantage over meta-analyses of pre-imposing a degree of standardisation but "its disadvantage is that it does not allow accessing the original microdata, which constrains the analysis that can be performed" (OECD, 2008, p. 41); directly related to this disadvantage, it also seriously hampers the possibility of controlling the quality of the data received.

In short, we have a "hierarchy" of degrees of standardisation:

1. common survey instrument (ECHP);

2. ex ante harmonised framework (EU-SILC);

3. ex post standardised microdata (LIS);

4. ex post customised results (OECD);

5. meta-analyses of results (Kuznets).

Presenting them in this rank order may seem to imply a quality ranking (with 1 at the top). However, it should be borne in mind that tighter requirements of standardisation may have a cost in terms of reduced accuracy in the final statistical outcomes. In particular, a common set of variables may have differing significance in different countries, and a degree of flexibility may allow national statistical institutions to provide data better suited to purpose. Input harmonisation does not necessarily ensure output harmonisation. Different sources may be appropriate in different countries. For example, the use of tax records may allow superior income data to be collected in some countries but may not be possible or reliable in other countries. The ultimate validity of the results may be greater where countries are allowed to make use of register data, and not constrained to take income data from survey interviews.

The EU-SILC procedure may therefore be seen as a balance of considerations. There is a cost in that greater flexibility may lead to lower comparability, but this may allow data to be drawn from different sources including sources other than household surveys. It may also have been instrumental in allowing Member States to reach agreement that EU-SILC could be adopted on a continuing annual basis. In this respect, there is an important difference between EUSILC, on the one hand, and the LIS and OECD data, on the other hand. The results in the OECD report Growing Unequal? (OECD, 2008) relate to the mid-80s, mid-90s, and mid-2000s. Such decadal observations are valuable but of limited use to policy-makers. LIS has more frequent observations, approximately semi-decadal:

Waves I (around 1980), II (around 1985), III (around 1990), IV (around 1995), V (around 2000), and VI (around 2004). But the data are not annual.

(2) In practice, except for Ireland and the United Kingdom, the income reference period is for all EU countries the calendar year prior to the Survey Year. In Ireland, the survey is continuous and the reference period is the last 12 months. In the UK, current income is collected and annualised with the aim of referring to the current (survey) year – i.e. weekly estimates are multiplied by 52, monthly estimates by 12, etc. (Eurostat, 2009).

The essential requirement of (timely) annual data is apparent from the recent economic and financial crisis. The occurrence of such events will only by chance correspond to the decadal or semi-decadal measurements. Data for 2004, the central year for Wave VI in LIS, and the year taken for 23 of the 30 observations analysed by the OECD in their 2008 report (2008, Table 1.A2.3), are too far distant to provide a benchmark for monitoring the impact of the crisis and the subsequent recession. (Indeed, even annual data may not always be sufficient for monitoring purposes – see the discussion on timeliness and frequency at the OECD March 2009 Roundtable on Monitoring the effects of the financial crisis on vulnerable groups of society (3) and Section 18.2.3 of Chapter 18.)

EU-SILC has therefore a distinctive role on the international scene. At the same time, it is important to examine how the findings relate to those in other cross-country sources. The OECD in its 2008 report makes exactly such a comparative analysis, and the present chapter uses this analysis in Section 5.2 when comparing the EU-SILC evidence on income inequality and poverty with that in other international sources.

5.2 Income poverty/inequality across countries and comparison with international sources

5.2.1 Evidence from EU-SILC on the risk of poverty

The chapter begins with the key income-based indicators from EU-SILC Survey Year 2008.

"Income" refers here to the total household disposable income; it includes cash transfers and is net of income taxes and social insurance (4)

In order to reflect differences in household size and composition, total household income is divided by an equivalence scale (called the modified OECD scale), which gives a weight of 1 to the first adult, 0.5 to other household members aged 14 and over and 0.3 to each child aged under 14. This means that, for a couple and 2 children, income is divided by 2.1 (1 + 0.5 + 0.3 + 0.3), so that an annual income of € 10.500 becomes an equivalised income of € 5.000 which is artificially assigned to each of the four household members (i.e. also to each of the two children). As explained above, the data in the 2008 Survey are based on the income reference year 2007 (except in Ireland and the United Kingdom). The reader should bear in mind that we are considering annual income in 2007 in relation to the household circumstances at the time of interview in 2008. There may have been changes in these circumstances, such as the arrival of a new baby.

(3)See:http://www.oecd.org/document/2/0,3343,en_2649_33933_42507906_1_1_1_1,00.html.contributions.

(4) The definition of income used here excludes imputed rent, i.e. the money that one saves on full (market) rent by living in one"s own accommodation or accommodation rented at below-market rent. It also excludes non-cash transfers, such as education and healthcare provided free or subsidised by the government. Finally, as explained in Chapter 2, it also excludes pensions from private plans (which as from the second half of 2010 will be incorporated in the EU-SILC income definition for all -past and future- waves) and most non-monetary income components. Income is neither top-coded nor bottom-coded.

The EU headline indicator of (income) poverty/inequality is the proportion of the population living "at-risk-of-poverty", defined as those living in households whose total equivalised income is below 60 per cent of the median national equivalised household income. It is thus a relative concept. The equivalised income of € 5.000 for the four members of the family described above is compared with 60 per cent of the median in the Member State in which they live. Table 5.1 provides the value of the national income poverty thresholds for each Member State for a family consisting of 2 adults and 2 children below 14. To make them more comparable, because the cost of living can vary a lot from one country to the next, these thresholds are expressed in Purchasing Power Standards. (5) So, if we take our example above and assume that this family has an income of 10 500 Purchasing Power Standards (rather than euros), then the four members of this family would not be considered at risk of poverty in eight EU countries (all of them are New Member States: Bulgaria, the three Baltic States, Hungary, Poland, Romania and Slovakia); in the remaining 19 EU countries, they would be considered income poor.

Figure 5.1 shows the standard bar chart for the percentage of people living in households at risk of poverty. The countries covered are those in EU-27. The average for the EU-27 as a whole is 16.6 per cent, which means that 1 in every 6 of EU citizens are at risk of poverty, or around 80 million people. (6) The rate for the 12 "new" Member States (NMS12) was 17.3 per cent, a little but not much higher than for EU-15 with a rate of 16.4 per cent. It is certainly not the case that those at risk of poverty on the EU definition are mostly to be found in the New Member States: of the 80+ million at risk of poverty in EU-27, 64 million are to be found in the EU-15. In Germany, alone, there are 12½ million; in the United Kingdom 11½ million; in Italy 11 million; and France and Spain together account for a further 17 million. In the largest New Member State, Poland, the number of people at risk of poverty is about 11½ million.

On this relative poverty measure, New Member States are to be found at both ends of the national figures, which range from 9-11 per cent (in the Czech Republic, the Netherlands, and Slovakia) to 20 per cent or more in Lithuania, Greece, Bulgaria, Romania and Latvia. The picture shows that, in terms of cross-country variation, there is a relatively continuous gradation. It is not easy to draw sharp dividing lines on the basis of income poverty performance. There are only four jumps from an adjacent country in excess of 1 percentage point: Finland/ Malta (1.1), Poland/ Portugal (1.6), Bulgaria/ Romania (2), and Romania/ Latvia (2.2).

(5) On the basis of Purchasing Power Parities (PPP), Purchasing Power Standards (PPS) convert amounts expressed in a national currency to an artificial common currency that equalises the purchasing power of different national currencies (including those countries that share a common currency).

(6) This "EU-27 average" is a weighted average of the 27 EU Member States" percentages, in which each country percentage is weighted by the country"s population size. EU-15, NMS10 and NMS12 averages presented in this chapter are calculated in the same way. For the countries included in the various geographical aggregates, see the list of "Country official abbreviations and geographical aggregates" (Appendix 2). of income poverty performance. There are only four jumps from an adjacent country in excess of 1 percentage point: Finland/ Malta (1.1), Poland/ Portugal (1.6), Bulgaria/ Romania (2), and Romania/ Latvia (2.2).

From Figure 5.1, we can assess the ambition of the Europe 2020 Agenda "to lift at least 20 million people out of the risk of poverty and social exclusion" (European Council, 2010). Measured in terms of the at-risk-of-poverty rate, (7) it would mean reducing poverty and social exclusion by 4 percentage points. The EU-27 as a whole would have to match the performance of Austria. It is also clear that attainment of this ambition requires, as far as the at-risk-of-poverty indicator is concerned, action by the six largest Member States. France, Germany, Italy, Poland, Spain and the United Kingdom cannot stand aside. If they were to do so, then reaching the 20 million targets would require the virtual elimination of income poverty in the other 21 Member States.

Who is "at-risk-of-poverty"? EU-SILC allows income poverty rates to be calculated for many groups within the population. Here we focus on just one group which has (rightly) received a great deal of attention in recent years: the proportion of children living in households at risk of poverty. (8) This is referred to for short as "child poverty", although it should be emphasised that what is being measured is the status of the household where the child lives (see above example). It should also be emphasised that no account is taken of the possibly unequal sharing of income within the household. Figure 5.2 shows the child poverty risk rate in each country compared with the overall poverty risk rate for Survey Year 2008. Countries lying on the heavy line have the same rate of child poverty risk as overall population poverty risk. The cause for concern about child poverty is that relatively few (only about a quarter of the 27 EU Member States) are below this line. For seven Member States, the child poverty rates are more than 5 percentage points above the overall rate – shown by those above the dashed line in Figure 5.2. So that while in Hungary child poverty rate is slightly below the EU average (19.7 vs. 20.1 per cent), it is 7.3 per cent higher than the overall population poverty rate. Above the dashed line are Luxembourg and Italy, but the other 5 countries are New Member States. The overall child poverty rate for the 12 New Member States is indeed 4 percentage points higher than for EU-15 (23.1 vs. 19.3 per cent).

(7) This is in fact only one of three indicators.

(8) See, for instance: Frazer and Marlier (2007), Social Protection Committee (2008), Tárki (2010), Frazer, Marlier and Nicaise (2010).

So far, we have been counting the number of people, or the number of children, at risk of poverty. But how far do they fall below? The final EU indicator considered here is the total poverty risk gap. What is the total income shortfall? Figure 5.3 shows, in addition to the at-risk-of-poverty rate, the median percentage by which households fall below the income poverty line. For EU-27, the figure is 22 per cent, which means that half of the at-risk-of-poverty population is living on less than 78 per cent of the income poverty threshold. Since the threshold is 60 per cent of median income, this means that the shortfall is some 13 per cent of median income. What is of interest is that the graduation is now much less smooth as we move across countries. For half the Member States (those to the left of Germany in Figure 5.3), the shortfall is between 15 and 20 per cent, but for Germany and countries to its right the gaps range from 16.5 to 32.3 per cent.

EU-SILC contains much further rich data about the risk of poverty, but the evidence presented above from the 2008 Survey (income year 2007) shows that the risk is pervasive, affecting all Member States. New Member States are not concentrated at the top of the scale. Looking to the future, achievement of a 20 million reduction requires action by the large Member States: the largest six accounts for nearly three-quarters of the total at risk of poverty.

5.2.2 Evidence from EU-SILC on income inequality

To this juncture, we have focused on the bottom of the income distribution. What is the overall extent of inequality? Many are concerned that inequality was a factor contributing to the economic crisis; others are concerned that the crisis will exacerbate inequality. But just how unequal are incomes? The two main indicators of income inequality used at EU level are shown in Figure 5.4. The first is the ratio of the share of income going to the top 20 per cent of the population (referred to as the top quintile share) to that going to the bottom 20 per cent (the bottom quintile share).

This ratio, also called S80/S20, varies from 3.4 to 7.3 across the EU Member States. There is an interesting geographical pattern. The lowest ratios are found in some of the New Member States (Slovenia, Slovakia, the Czech Republic and Hungary) as well as in Austria and the Nordic countries. Then come Malta, Benelux, Cyprus and France. In Southern Europe (except Cyprus and Malta), Poland, the United Kingdom and Lithuania, the ratios are between 5.1 and 6.1, and they are 6.5 or more in Bulgaria, Romania and Latvia. For the EU-27 as a whole, the S80/ S20 ratio is 5. It should be noted that the latter is the weighted average of the 27 national ratios, in which each country ratio is weighted by the country"s population size; it is thus not the same as the ratio of the top to bottom quintile shares in the EU-27 as a whole, which can be expected to be higher.

The second indicator of income inequality shown in Figure 5.4 is the Gini coefficient, a summary measure, based on the cumulative share of income accounted for by the cumulative percentages of the number of individuals, with values ranging from 0 per cent (complete equality) to 100 per cent (complete inequality). The Gini coefficients vary a lot across countries, from 23 per cent in Slovenia to 38 per cent in Latvia. (9) For the EU-27 as a whole, the (weighted) averaged value is 31 per cent. What do such values mean? The following hypothetical calculation may be helpful. Suppose that the tax and transfer system is approximately of the form of a uniform tax credit and a constant tax rate on all incomes, that the government spending on goods and services absorbs 20 per cent of tax revenue, and that the Gini coefficient for disposable income is 48 per cent in the absence of redistribution. Then, an increase in the tax rate of 5 percentage points would be needed to reduce the Gini coefficient by 3 percentage points. (10) Since a tax rise of 5 percentage points would be a challenge for any Finance Minister, this suggests that a 3 point difference would be salient. This means that moving across a vertical division in Figure 5.4 represents a significant -in economic terms- difference.

(9) The scales for the two inequality indicators in Figure 5.4 are different but the indicators move very closely together. There is no reason why this should necessarily be the case. A redistribution that affected only those between the bottom quintile and the top quintile would have no impact on the S80/S20 ratio but would affect the Gini coefficient as this indicator considers the entire income distribution and not just the top and bottom quintiles.

(10) See Atkinson (2003), p. 484. The Gini coefficient is equal to half the mean difference divided by the mean. Taxation with a constant marginal tax rate implies that the mean difference is reduced by (1-marginal tax rate); the mean is reduced by (1-average tax rate). 1 minus the average tax rate is what is left for households after paying for government goods and services: in this example, 80 per cent. With no redistribution, the tax rate would be 20 per cent. So that the Gini coefficient for disposable income would be the same as for pre-tax income. If the marginal tax rate is raised to 25 per cent to finance redistribution via a uniform tax credit, then (1-marginal tax rate) becomes 75 per cent, while the average tax rate (allowing for the credit) is unchanged. The Gini coefficient is therefore reduced to 75/80 of its previous value: i.e. from 48 per cent to 48 per cent times 75/80, which equals 45 per cent.

Applying the criterion that 3 percentage points represents a "salient" difference in the Gini coefficient, we obtain a partial ranking of Member States. We cannot say that inequality is different in France from that in Germany (in Survey Year 2008), but there is a salient difference between the Gini coefficients for France and the United Kingdom, as there is between those for Sweden and France. On this basis, income inequality is higher in Latvia than in any other country apart from Romania, Bulgaria and Portugal. Income inequality can be said to be lower to a salient degree in Slovenia than in all Member States apart from Slovakia, the Czech Republic, Austria, Hungary and the Nordic countries.

How is inequality in income related to income poverty? Do the same countries have both low at-risk-of-poverty proportions and low income inequality? There is no reason why this should necessarily be the case. The share of the bottom 20 per cent may reasonably be taken as closely linked to the incidence of income poverty, but this leaves considerable room for differences in the other quintile group shares. A country may for example have a share for the bottom 20 per cent of 11 per cent, which -if equally distributed- would ensure an income equal to 55 per cent of the mean. (11) Since the mean is typically higher than the median, this could well be above 60 per cent of the median and the poverty risk score could be zero. Such a (low poverty risk) bottom quintile share could however be combined with a relatively unequal distribution, such as 12, 13, 14 per cent for the second to fourth quintile groups and 50 per cent for the top 20 per cent. The S80/ S20 ratio would then be 4.55, which is not much lower than the EU-15 average (4.88).

(11) The figure of 55 per cent is obtained by dividing 11 per cent by the group"s proportionate share (20 per cent): 11/20 = 0.55.

In fact, as may be seen from Figure 5.5, the at risk-of-poverty rate is closely correlated with the degree of income inequality as measured by the S80/S20 ratio (the same is true with the Gini coefficient in place of the S80/S20 ratio, although this is not shown here). There do not appear to be countries with medium/high inequality and low poverty risk. A simple regression shows that the inequality ratio explains 85 per cent of the variance in the poverty rate, and that an increase in the ratio from 3.5 to 4.5 is associated with a 3.4 percentage point increase in the poverty rate.

5.2.3 Comparison with other cross-country sources

There are now a variety of sources of internationally comparative data on income inequality and income poverty. The best known is perhaps the World Bank"s World Development Indicators (WDI), which shows in its 2009 edition estimates of the distribution of income or consumption for 136 countries in the form of the Gini coefficient and the shares of income quintile groups (World Bank, 2009, Table 2.9). The values for 24 out of the 27 EU countries (data for Cyprus, Luxembourg and Malta are not included in the WDI table) are shown in Table 5.2, together with the sources. There are two evident problems. The first is that the data come from two different sources. It is stated that data for "the high-income countries" are income data taken from the LIS database, and this applies for 16 of the countries. But for eight countries, all New Member States, the data relate to expenditure and come from other sources. Secondly, as explained earlier, the LIS data are not annual, and those used in the 2009 WDI relate mostly to the year 2000 or, in seven cases, even earlier. This latter point reduces significantly the value of the WDI compilation. It certainly appears a little odd that the data in the 2009 WDI table for Liberia and Morocco relate to 2007, whereas the data for France are no more recent than 1995. The former problem limits the comparability within the EU, although the expenditure data may be more comparable with those for middle-income and developing countries.

The question naturally arises as to why the WDI does not employ the EU-SILC data, which would have the definite advantages of being more current and of not mixing income-based and expenditure-based estimates? The answer may depend on the comparison of this new source with the longer established LIS and with official sources such as the OECD. Here we may turn to the OECD report (OECD, 2008), which contained a most helpful comparison of the OECD estimates with EU-SILC (2005 data, income reference year 2004) and LIS (mostly relating to years around 2000). There is relatively little discussion of the findings of the comparison in the OECD report, perhaps because the results appear reassuring. Their figures for the at-risk of- poverty definition based on 60 per cent of the median are reproduced in Figure 5.6. (12) The three bars show the estimates for each country for the OECD, EU-SILC and LIS (in some cases one of the latter two is missing).

(12) The comparison also includes four non-EU countries: Iceland (IS), Norway (NO), Switzerland (CH) and Turkey (TR).

In almost all cases, the estimates of poverty risk in the three sources are close. Only for 9 of the 57 possible comparisons is the difference equal to 3 percentage points or more (although the estimates are rounded to the nearest integer, so that some of the differences may be only 2.1). Three countries (Germany, the Netherlands and the United Kingdom) account for six of these discrepancies, and these differences are identified by the OECD as a matter for concern. The differences in the case of Germany are four (LIS/OECD) and five (EU-SILC/OECD) percentage points. These differences are among those discussed further in Section 5.3. It should also be noted that only one of the nine discrepancies (for Sweden) concerns the comparison of the EU-SILC and LIS estimates, which are generally closer.

The Gini coefficients of income inequality from the three sources are compared in Figure 5.7. The general pattern is similar. It has to be borne in mind, and this applies to both the poverty risk figures (Figure 5.6) and the Gini coefficients (Figure 5.7), that the definitions are not identical. The EU-SILC estimates use the modified OECD equivalence scale described above, whereas, a little strangely, the OECD does not use the scale that bears its name, but uses a square root equivalence scale, as in the LIS data. Use of this latter scale means that income is divided by the square root of the household size (two in the case of the four person household example), which means that the relative position of different households will be affected. This may well affect the comparison, as may the fact that the OECD and EU-SILC data refer mostly to 2004, whereas the LIS data refer to a variety of years around 2000.

All in all, there appears to be a high level of coherence between the cross-country datasets. The data for certain countries needs to be examined, but data created by the EU-SILC framework approach do not seem to be out of line with those assembled by the LIS or OECD methods.

5.3 Changes in income poverty and inequality over time

5.3.1 Monitoring trends in EU-SILC

In the previous section, we have described the situation in the EU in 2007 (the 2008 Survey Year related in nearly all countries to incomes in 2007). But much of the interest of the figures lies in how inequality and poverty are changing over time. In this respect, it is frustrating that we can say little about what has happened since 2007. At a time of economic crisis, everyone, citizens and politicians alike, wants to be able to monitor what is happening to living standards following the financial crisis and the subsequent world recession. Who is bearing the burden?

It is also important, however, to understand what was happening before the economic crisis. How far had the EU been successful in its 2000 declared ambition of achieving a significant reduction in poverty and social exclusion? Was it the case that there had been rising inequality, a factor which some commentators have treated as a contributing to the crisis? Here too we are limited as to what we can say. EU-SILC was launched in 2003, with income reference year 2002, on the basis of a "gentleman"s agreement" in six Member States. The official starting date for EU-SILC was Survey Year 2004 for EU-15 (minus Germany, the Netherlands and the United Kingdom, plus Estonia), with income reference year 2003. The New Member States that joined the EU in 2004 (apart from Estonia) as well as Germany, Netherlands and the United Kingdom, started with respect to Survey Year 2005. Bulgaria entered in Survey Year 2006, and Romania in Survey Year 2007. This means that there are data for between 2 and 6 years -see Table 5.3. (As indicated previously, the income reference year is different for Ireland and the United Kingdom.)

Can we identify from this short EU-SILC time series countries where income poverty and inequality are decreasing or increasing? In the case of year-to-year changes, sampling errors are clearly relevant. In the case of the at-risk of- poverty rate, Lelkes et al (2009, Figure 1.10) show for Survey Years 2004-2006 10 countries where there were changes outside the 95 per cent confidence interval for the preceding year. (13) The countries are equally divided in their direction of movement. The "improvers" were Estonia, Ireland, Netherlands, Poland and Slovakia. Those moving towards higher poverty risk were Finland, Italy, Latvia, Luxembourg and Sweden.

Year-to-year variation on account of sampling error certainly means that we should not attach weight to modest changes in the at-risk-of-poverty rate over time. The sampling errors reported for the 2005 EU-SILC for the proportion at-risk-of poverty imply a one-sided 95 per cent confidence interval of less than 1 percentage point for 11 of the 23 countries analysed and in all cases it is less than 2 percentage points (Eurostat, 2008). Account has also to be taken of non-sampling errors, as has been discussed in Chapter 3. These considerations refer to the "supply side": the accuracy of the estimates supplied by EUSILC (or other sources). It is indeed a prerequisite that the observed performances are different. But we have also to ask about the "demand" side. What differences are of interest to the user? Here the Europe 2020 targets provide a point of reference. The ambition of the EU is to reduce those at risk of poverty and social exclusion by 20 million. In terms of the at-risk of- poverty rate, this would mean a reduction of approximately a quarter (20 million out of 80 million) or, put differently, a reduction of about 4 percentage points for the EU-27 as a whole. Applied at the level of individual countries, a reduction of a quarter would mean between 2½ and 6½ percentage points. Taking account of both supply and demand side considerations, we pay particular attention in what follows to changes of 2 percentage points or larger. (13) We have here excluded Hungary on the grounds explained by Lelkes et al, that there appear to be problems with the estimate for 2006 (Survey Year).

(13) We have here excluded Hungary on the grounds explained by Lelkes et al, that there appear to be problems with the estimate for 2006 (Survey Year).

5.3.2 Changes in poverty risk

What do we learn from Table 5.3 if we run our cursor over the figures identifying cases where the Survey Year 2008 data represent a change of 2 percentage points of more in the proportion at-risk-of-poverty relative to an earlier year? For six Member States, we have EU-SILC data covering six years. For only one -Ireland- did an earlier year have a proportion that differed by 2 percentage points or more. Between 2003 and 2008, Ireland moved from having an above EU- 27 average at-risk-of-poverty rate to one that is below it. In the other five countries there were falls, but these were smaller and in some cases reversed: for example, in Greece the proportion fell, then rose, and then fell.

For the countries with five years of data, Finland saw an increase in the at-risk-of-poverty rate in each year and ended with a figure 2½ percentage points higher – an increase of nearly a quarter. In the opposite direction, Portugal, with an initially high at-risk-of-poverty poverty rate, showed a reduction of 2 percentage points. Sweden showed both falls and rises of at least 2 percentage points, but ended in 2008 with an at-risk-of-poverty rate less than 1 percentage point different from that in Survey Year 2004.

There is some tendency for convergence, with high poverty risk countries tending to show reductions in income poverty rates (although not universally) and for there to be slippage in the opposite direction among the previous better performers. This is illustrated by the fall between Survey Years 2005 and 2008 in the at-risk-of poverty rate for the NMS10 group, i.e. the 10 countries that joined the EU in 2004, where the rise in Latvia was more than offset by the falls in Poland and Slovakia.

In sum, the picture prior to 2008 was not a static one. Some countries have achieved sustained reductions in the proportions at-risk-of-poverty, but in the EU as a whole this progress has been offset by reversals in other Member States.

5.3.3 Changes in income inequality

It is widely believed that income inequality has been on the increase. This belief is much influenced by the experience of the United States, but has the same happened in Europe?

The EU-SILC data suggest that the EU picture is more nuanced. Tables 5.4a and 5.4b show the EU-SILC results for the two inequality indicators used in the previous section. Overall the weighted-average indicator for EU-27 hardly changed between Survey Years 2005 and 2008. (Again it has to be remembered that this is the average of national inequalities, not the overall EU inequality taking account of between-country differences.) This did not reflect stasis. There were country changes, and indeed some degree of convergence. The average for the 10 New Member States showed a reduction in inequality: the S80/S20 ratio went from 5.6 to 4.6, and the Gini coefficient fell by nearly the 3 percentage points that we described as a "salient" change in the previous section. There were falls of more than 3 percentage points in the Gini coefficient in Estonia and Poland.

If we look at EU-15, then among the larger countries there is little evidence of change in France, Italy, Spain and the United Kingdom. The most evident change in the EU-SILC data is the rise in the S80/S20 ratio (from 3.8 to 4.8) and in the Gini coefficient (from 26 to 30 per cent) in Germany. (During the same period, the at-risk of- poverty rate measured on the basis of EUSILC also increased sharply in Germany, from 12.3 per cent to 15.3 per cent; we come back to these estimates in Section 5.3.4.)

These country differences underline the need to compare the EU-SILC findings with those from national sources, to which we now turn.

5.3.4 Comparison with national sources: a case study

The provision of data on income inequality and poverty has a long history in individual Member States. Whereas in some countries the launching of ECHP, and now EU-SILC, was a stimulus to collect distributional data on a regular basis, and the EU reference data provide the main national source, in quite a number of countries there are long running regular series, typically annual, for income inequality and poverty. In the latter cases, it is important to compare the findings from EUSILC with those from the national sources. (14)

Differences between the results from EU-SILC and from national sources do not imply that one source is necessarily in error or that one source is to be preferred. Differences may arise for several reasons, including the following ones:

• differences in the population covered (for example, the exclusion in EU-SILC of the non-household population, whereas national sources may cover people living in collective households or institutions);

• differences in the definitions adopted (for example, of the unit of analysis or of total income or of the equivalence scale);

• differences in timing (for example, in the definition of the income reference period or

in the scheduling of the interviews).

On the other hand, differences may be attributable to identifiable shortcomings. Response rates may be different, particularly where there is attrition from a panel survey. The extent of reporting may vary, as may be indicated by checks against known income totals.

In this section, we take one comparison with national sources as a case study. The case study is that of Germany. There are three reasons for this choice. First, Germany is the largest Member State. Secondly, the EU-SILC findings show that Germany was one of the countries to exhibit rising income poverty and inequality. Thirdly, there have been a number of academic studies making comparisons between the EU-SILC results and those from other sources.

(14) It would also be possible to use the findings from the ECHP – see Lelkes et al (2009). The issue of the continuity of indicators during the transition between ECHP and EU-SILC is considered by Eurostat (2005).

The main national sources of household data in Germany are the Microcensus, the Income and Expenditure Survey and the German Socio- Economic Panel (GSOEP) conducted by the Deutsches Institut für Wirtschaftsforschung (DIW). The relationship between these sources has given rise to considerable discussion. Hauser (2008) has compared the EU-SILC results for 2005 with the Microcensus and GSOEP. He noted that two features of the German EU-SILC (reliance on a postal survey and delay in developing a fully random sample) led there to be ex ante doubts about the EU-SILC German data. He reported that there were "significant deviations in the coverage of poorly integrated foreigners, small children and the level of education, as well as the ratio of house/apartment owners and the employment ratio" (2008, p. 2).

The implications for the EU commonly agreed indicators have been discussed by Lelkes et al. Drawing on Frick and Grabka (2008), they note that "the proportion of the population at risk of poverty is about 5 percentage points lower when calculated from the EU-SILC data than when calculated from (GSOEP)" (2009, p. 44). They cite figures from GSOEP (EU-SILC figures in brackets) of income poverty rates of 16.3 per cent for Survey Year 2004, 16.7 (12.0) per cent (15) for Survey Year 2005, and 18.0 (12.7) per cent for Survey Year 2006. These are large and disconcerting differences, but since then the GSOEP methodology has been revised with regard to weighting and the imputation of missing income. The estimates given by Frick and Krell (2010, Table 2) show income poverty rates of 13.9 per cent for Survey Year 2005 and 14.3 per cent for Survey Year 2006. For these two years, the difference is now reduced.

If that were the end of the story, then one might be reassured. However, a correspondence between the aggregate (income) poverty rates does not imply that the constitution of the poverty population is the same. We need to go further and examine, for example, the household composition. We need to consider the implications of the differences in the degree of mobility found in the longitudinal data by Frick and Krell (2010). Moreover, the EU-SILC data for Survey Year 2007 show (see Table 5.3) a rise in the income poverty rate by 2.5 points (to 15.2 per cent), maintained as 15.3 per cent in Survey Year 2008; by contrast, GSOEP estimates decrease between these two years (from 14.3 to 13.6 per cent). Not only is the direction of movement in the opposite direction from the GSOEP figures, but the magnitude of the increase in the EU-SILC values is hard to understand.

(15) The figure of 12.0 from EU-SILC corresponds to that of 12.3 in Table 5.3.

In the same way, for the income inequality measures, the GSOEP (calculations of Frick and Krell, Table 2) show a broadly stable S80/S20 ratio (4.4 for Survey Year 2006 and 4.3 for Survey Year 2007), whereas the EU-SILC data show a rise from 4.1 to 5.0. Frick and Krell comment that the size of the latter increase is "exceptionally difficult to comprehend or explain based on the evolution of income inequality in Germany over the last few decades – particularly given the positive labour market conditions at the end of the period" (2010, p. 18). They go on to explore the sources of the discrepancy in the sample composition and weighting methods.

The issues raised by this comparison with national sources are technical ones, but there is clearly need to invest in their resolution. Such comparisons are necessary to secure acceptance of the EU reference source at the national level. Results that indicate income poverty rates very different (whether higher or lower) from those reported nationally are likely to raise questions and potentially generate political debate. Where levels and/or trends over time are different in EUSILC and in national sources, it becomes difficult to draw conclusions about the effectiveness of policy measures taken to reduce income poverty and inequality.

5.4 Monitoring progress

EU-SILC data play a central role in the promotion of the Social Agenda of the EU. (16) In this section, we consider the use of EU-SILC data in forensic policy analysis, particularly for monitoring the Europe 2020 Agenda. As we emphasised earlier in this chapter, the significance of changes in income inequality and poverty depends on both supply and demand side considerations. The suppliers of the data can advise on the statistical validity of observed changes, and the demanders can calibrate the policy significance of the changes. Both of these are relevant to monitoring, but we focus here on the less discussed side: the criteria stemming from the use of the EU-SILC data.

5.4.1 An at-risk-of-poverty target

The original proposal by the Commission was of a Headline Target set in terms of the numbers at-risk-of-poverty, with the aim of reducing these by 20 million, and we begin by considering this case. As we have seen in Section 5.2, such a target is ambitious; it is also in need of further amplification. We discuss two aspects here. First, it needs to be anchored in time. (17) The 80+ million figure for those at risk of poverty relates to Survey Year 2008, typically income year 2007. Even though it is still being discussed, it is likely that this is to be taken as the base figure. This -perfectly reasonable- choice would imply that, in the early years of monitoring, performance will be affected by the economic crisis. The lags mean that the incomes of the present year (2010) will only enter the assessment based on EUSILC Survey Year 2011 whose data will become available at the end of 2012. Does this mean that the at-risk-of-poverty percentage will initially rise? The implications are not in fact clear. The economic crisis has affected both the incomes of those at the bottom of the income distribution and the median income against which poverty risk is being measured. If, for example, pensions have been maintained but incomes in work have fallen, then fewer pensioners may be below the income poverty threshold. On the other hand, there are reasons to fear that the unemployed living in households where there is a single earner have suffered falls in income.