Challenges in income comparability. Experiences from the use of register data in the Norwegian EU-SILC

Transcription

1 Challenges in income comparability. Experiences from the use of register data in the Norwegian EU-SILC Jon Epland Statistics Norway Paper prepared for VII International Meeting on Quantitative Methods of Applied Sciences, University of Siena, September 2006

2 1. Introduction The new instrument to monitor living conditions in Europe the EU-SILC encourages the member states to use the best national sources in collecting income data - whether they be surveys or registers. In the Nordic countries it is common practise to collect income data from various administrative registers. Register data have several advantages compared to traditional surveys for instance in respect to selective non response and erroneous reporting of income. However, register data may lack some of the flexibility that a household survey may offer. The aim of the paper is to give a description of some of the challenges faced in respect to comparability when using register data to construct the income module in the Norwegian EU- SILC. The paper also points at some of the differences in income distribution that may occur when using either register or survey as the method of data collection. The paper is organised as follows: In the next section we give a description on the numerous registers that are linked in order to construct income concepts that are comparable to the SILC guidelines. We share some of the experiences in post-harmonising register data in order to comply with Eurostat guidelines. In section 3 we focus on the difference in the mode of data collection. One important question to be addressed is whether it makes a difference if income data is collected from a survey of from registers. Section 4 summarizes the findings. 2. Constructing the SILC income module Data sources A prerequisite condition for being able to compare income distribution across time and countries is that the data have been harmonised, i.e. that the concepts and definitions are constructed in an identical way. Following recommendations from the Canberra-group (Expert Group on Household Income Statistics 2001) and the Eurostat Income Measurement Manual (Church & Verma 2001), the SILC guidelines (Eurostat 2003) give the data collectors a detailed recipe of which income items to collect and how these items are to be categorised into different components of income, e.g. income from work, capital income, social transfers etc. In a majority of countries in Europe these income data are collected from a household interview based on a harmonised questionnaire. However, in those countries which have the 2

3 opportunity to collect income data from various administrative registers the main challenge is to post-harmonise different income variables in order to compare with the Eurostat guidelines. In Statistics Norway the national income distribution statistics are constructed by linking a large number of administrative and statistical registers that cover different types of income data. Figure 1 gives a crude overview of all the inputs of the register based income statistics. The main data providers are the Tax Authorities and the National Insurance Administration. The single most important source is the Tax Return Register. This register give detailed information on all kinds of taxable income, e. g. wages and salaries, self-employment income, income from property and taxable pensions. Another important source is the Tax Register, where information on personal income taxes and social security contributions are collected. From the National Insurance Administration, all types of tax-free transfers (e.g. family allowance, support to single parents) are collected as well as different types of pension income (e.g. old age and disability). In addition to tax registers and social security registers some minor income items are collected from other administrative registers, for example dwelling support (The State Housing Bank) and scholarships (The State Educational Loan Fund). It should also be noted that register data are even used to collect some biographical data for individuals, such as highest level of completed education, formal marital status, citizenship, immigrant status and municipality of residence. Once the different registers have been linked to each individual by the use of the Personal Identification Number, there exists a comprehensive data source that can provide income data to all social surveys that Statistics Norway currently conducts, one of them being the EU- SILC. 3

4 Figure 1. The Construction and Utilisation of Register Based Income Statistics Population Statistics System The Tax Register for Personal Tax Payers The Tax Return Register The State Educational Loan Fund The Register for End-ofthe-Year Certificates Income Statistics for Persons and Families The State Housing Bank Statistics Norway s Educational Register Social Assistance Register The National Insurance Service The Income Distribution Survey EU-SILC Other Social Surveys Census

5 Data quality In their hierarchical classification of income data, Atkinson, Rainwater & Smeeding (1995) suggest five conceptual levels of income reporting, where reporting in the upper level is considered more complete than the lower levels: 1. True Income 2. Administrative Record Income 3. Tax Reported Income 4. Edited Survey Income 5. Reported Survey Income From what has been shown, the Norwegian income data most likely belong to the second level. It is, however, difficult to measure the quality of income data. One frequently used method is to compare aggregate income from micro data with external sources, for instance the National Accounts. One obstacle to such comparisons is the large differences in income definitions between, on the one hand, the comprehensive National Accounts and, on the other hand, the more pragmatic cash income concepts available in micro data. Even after careful adjustments and reconciliation of concepts and classifications there remains large discrepancies that only can be explained by conceptual differences that are not possible to adjust for. Table 1 shows the result of such micro-macro reconciliations that recently have been conducted in three typical register countries namely Finland, Norway and Sweden. In these countries income from registers are compared to National Accounts estimates. It seems apparent that wages and salaries are well covered in the micro data from registers with an almost total overlap with the National Accounts. The deviations are a bit larger in respect to property income but still close (92-93 per cent in Finland and Norway). Self-employment income, however, show a larger difference when compared with mixed income from the National Accounts in particular for Sweden (register data comprise less than a third of the NA estimates). One explanation why self-employment income from tax registers show a larger discrepancy may be the fact that this kind of income to a large extent is defined by current tax rules, for example variations in the legal possibility to withdraw expenses from gross income. It is therefore easy to agree with Church & Verma (2001) when they state that obtaining accurate information about the income of the self-employed is one of the more difficult challenges in survey research. This holds even when such data are collected from tax registers. 5

6 Table 1 Comparison of factor income in Income Statistics and National Accounts after adjustments for differences in concepts and classifications. Finland (2000), Norway (1998) and Sweden (1998). Billion FIM, NOK & SEK Income National Ratio Statistics Accounts IS/NA Finland Wages 293,0 293,9 100 SE-Income 70,8 70,0 101 Property income 15,6 17,0 92 Factor income 379,3 380,9 100 Norway Wages 463,3 465,7 99 SE-Income 49,4 53,2 93 Property income 30,5 32,9 93 Factor income 543,2 551,8 98 Sweden Wages 778,3 787,6 99 SE-Income 26,2 99,9 26 Property income 31,0 27,7 112 Factor income 835,5 915,2 91 Sources: Finland: Kavonious & Törmälehto 2002 Norway: Epland & Frøiland 2002 Sweden: Statistics Sweden 2002 Challenges in post-harmonising register data Although there exist a number of different registers available to producers of income statistics these registers do not always cover the exact income component that is requested in the guidelines, or more often, register data may include more than just one specific type of income. The obvious reason for this is that most registers are established not for statistical purposes, but for public administration. We will illustrate this lack of flexibility in register data, with reference to the task of post-harmonising the income module of the EU-SILC. The by far most important income component for the average household is employee cash income. The SILC guidelines give a detailed description on how this component is to be defined. Since almost all income from employment is taxable income, this income component 6

7 is covered well in tax registers. It is, however, a problem that register data sometimes may be to crude, i.e. they do not allow for a more detailed break-down of income components. One example illuminates this problem. In Norwegian tax data sickness benefits are included in wages and salaries. The reason is that these income items are taxed in a similar way. Consequently, there is no reason for the tax authorities to make a distinction between these types of income. It is only from the income year 2004 that a new register from the National Insurance Service became available covering all sickness payments, thus making it possible to treat this income component according to the guidelines by deducting sickness payments from wages. 1 This may have an impact on comparability but only at component level - with previous waves of EU-SILC, where a large part of sickness benefits could not be distinguished from employee income. During the last years we have seen a steady increase in the availability of register data due to more computerization in public administration. As a consequence more and better data have become available for statistical purposes. 2 The producers of income statistics are but one of those who has benefited from this development. There are, however, a few examples of the opposite development too, i.e. where data that were once available from registers no longer are available or have been substantially changed over the years. One such example with relevance to the SILC income concept concerns paid child maintenance from former spouse (inter-household transfer received). Up until 2003 received child maintenance was considered taxable income in Norway (and the contributor was allowed a tax deduction). In 2004 this income item is tax-free income for the receiver (and no deduction allowed for the contributor). This information was thus no longer available from tax registers. The solution was once more to collect such data from the social security registers. However, data from the social security registers are less comprehensive compared to tax data in respect to this particular income item, because they do not include paid child maintenance that are totally privately organised. For most households the impact of such loss of data will be 1 The process of deducting one income component from register A from another income component from register B isn t always a straight forward one. The two administrative sources may for example differ in respect to the treatment of repayments (income earned previous year, but paid out the next year) and prepayments (income for next year paid out in advance). Generally, social security registers sometimes include both repayments and prepayments, while tax registers generally include only income received during the tax year (i.e. the calendar year). Consequently, it may occur that some cases in the data set have a negative amount of for example employee income in cases where sickness benefits (including repayments or prepayments) deducted from wages is larger than sum of wages from tax registers. 2 In Norway there is a Statistics Act that allows Statistics Norway to make use of all public registers for statistical purposes. 7

8 insignificant. However, for a particular group such as single mothers a likely consequence may be a certain under-estimation of household income. Another aspect of register data concerns accuracy. Despite the fact that register data are considered to be close to true income and compares well with the National Accounts, it will be naïve not to think that even such data suffer from under-reporting. The most serious weakness in register data probably concerns income earned from informal, or black, work. Although the most recent estimates suggest that the shadow economy in Norway may be on the decline, the same figures indicate that income from this kind of work may constitute as much as 5-10 per cent of GDP (Shima 2004). Even the reported tax data may to some extent suffer from under-reporting. Once more the self-employed stand out as one group with a unique possibility to lowering their actual taxable income, due to the existence of several legal constructions. There is also a general consensus that most fringe benefits given to employees have an estimated tax-value that is below the actual market value. In respect to EU-SILC this only concerns the value of a company car so far, but from 2007 this will also be the case for other fringe benefits like free (mobile)telephone, internet access, newspapers, holidays etc. There is also trouble ahead regarding the value of imputed rent that is to be included in the SILC income definition in The assessed tax value of own home that is available in the Norwegian tax return register only covers about 20 % of the actual market value of the dwelling. The discrepancy between tax value and market value furthermore vary widely and wildly in respect to regions and age and size of the building. It is therefore highly unlikely that the tax registers is a plausible source to collect this income item. Most likely a kind of imputation based on general information on the dwelling from other types of registers may be used to calculate an estimate of housing value and imputed rent. Based on the experiences of the first waves of the EU-SILC we have to conclude that it is probably impossible to harmonise register data on income to be 100 % in agreement with Eurostat guidelines at least at component level. Register data on employee income may for example include income items that should be classified as social security benefits, while selfemployment income may include income that should be classified as capital income, and vice-versa. The opportunity to use register data has, however, improved significantly in later years due to more computerization in public administration. Often one is in the favourable situation to have access to more than just one register covering the same income item (e.g. tax 8

9 records and social security records). However, when one depends on administrative data one also depends on definitions and concepts used in registers. Although the situation so far has been favourable, one should not turn a blind eye to the possibility that statistical producers one day may experience loss of important data caused by administrative changes. Consequently, income comparisons may be harmed. In particular the tax system is vulnerable to such changes, for instance in respect to how the tax rules of the day define property income, fringe benefits or income from self-employment. 3. Comparisons of register data and survey data In their effort to produce common indicators for social inclusion, Eurostat so far has made use of several different data sets from the member states. For the majority of countries the ECHP has been the primary source of data, while for others countries different national data sources have been used, e.g. the Household Budget Survey, Living Condition Surveys or Income Distribution Surveys. Furthermore, the data collection methods also vary, where some countries collect income data from surveys while others use register data. The new instrument, EU-SILC, will clearly improve the situation in respect to data comparability but the mode of data collection still differ between countries, e.g. some countries collect income data from surveys while others use registers. One important question to be raised is whether it makes a difference in cross-country comparisons if we use register data or data collected from a household interview? Based on comparisons of aggregate estimates there is a strong reason to believe that this is the case. Van der Laan et al (1996) claim for instance with reference to income data from the Netherlands that survey data suffer from selective non-response and that even the most elaborate income question underestimate income statistics from tax registers by as much as 10 per cent. The probably best way to answer this question, however, is to make one-to-one comparisons between data collected in an interview with similar data extracted from administrative records. So far there has only been conducted a handful studies of this kind, and in the following we will presents some of the main findings from two such studies conducted in Finland and Norway, respectively. 9

10 The most comprehensive study of this subject was conducted as part of the CHINTEX project, where German and Finnish researchers compared the income data collected in the Finnish ECHP with comparable income data from registers (Rendtel et al 2004). The results from this study seem to confirm that there is a certain under-reporting of income in surveys. When comparing average equivalent household income based on the two modes of data collection they found that the size of under-reporting varied from 8 to 5 per cent for the years 1995 and A similar result was found for Norwegian data. When comparing average gross household income from the tax registers with the same income concept reported in the Level of Living Survey, Epland and Kirkeberg (2002) found that the latter was roughly 8 per cent lower than the former. It should however be noted that the Norwegian question on household income was asked to only one household member (the most capable person on financial matters) and not based on the individual reporting from all household members. Generally, the quality of income data collected by proxy are of a lesser quality compared to individual reporting, c.f. Rendtel et al (2004). A probably more important aspect concerns income distribution. In table 2 we present some figures on income distribution from the two aforementioned studies. The table shows the percentile ratios of income collected from interview and registers. It seems likely that income collected from the survey have a slightly more unequal distribution than register data. Both the Norwegian and the Finnish data show that the percentile ratio of P90/P10 is larger for survey income compared to income from registers. Table 2 Selected percentiles of the income distribution in Finland (FIM) and Norway (NOK). Norway 1996 Finland 1999 Equivalent household Gross household income income Register Survey Register Survey P P P p90/p10 5,7 6,0 2,9 3,3 p50/p10 3,2 3,2 1,8 1,9 p90/p50 1,8 1,9 1,6 1,8 Source: Rendtel et al (2004) Table 8, and Epland and Kirkeberg (2002). 10

11 There may be more than one explanation for why survey data tend to show more inequality that register data. One explanation may be that people in the lower tail of the income distribution fail to report small or irregular income in the survey, but which will be recorded in administrative registers. One group that has been identified to clearly under-report their income is old-age pensioners. One important source of income for this group is income from capital, but this kind of income may be easily forgotten when they are asked about their household income in a survey. In addition some types of social transfers tend to be underreported in the survey like unemployment benefits and social assistance. The impact on cross-country comparability One of the main arguments for launching the EU-SILC was the need for comparable statistics in order to combat poverty and social exclusion. To this end a number of statistical indicators have been developed, for example on income distribution and level of poverty. If, however, it turns out that income data collected from registers have a different distribution compared to income data collected in a survey what are then the likely consequences when it comes to international comparisons and rankings? One disturbing finding of the CHINTEX study was that poverty results for Finland in fact varied substantially depending on which data collection method was applied. While the proportion living in poverty (less than 50% of median equivalent income) based on register data was 5.9 per cent, the poverty figure increased to 8.4 per cent when based on income data from the survey. Consequently, the ranking of countries may change depending on the mode of data collection. Figure 2 illustrates this problem. In this figure we rank a number of European countries according to the proportion at-risk of poverty (below 50 % of median equivalent income) for the period The figures for Finland are presented both based on register data and survey data. As the figure shows, the ranking for Finland changes when the mode of data collection changes. From being one of the countries with the lowest poverty rates when income data from register data are used, the figure based on survey data places Finland more or less in middle of the European ranking not far from the EU-15 average (9 %). 11

12 One interesting question would be if the findings for Finland will have relevance for other countries as well. Generally, when international comparisons on distribution and poverty are made, for instance by Eurostat, the OECD or research papers based on the Luxembourg Income Study, the Nordic countries usually end up among the countries with lowest poverty rates. However, one common feature of these countries is that they all collect their income data from registers. With reference to the case of Finland there may be reasons to believe that even the other Nordic countries might have higher poverty estimates and place themselves more close to the EU-average, had their income data been based only on surveys. The flipside of the coin would of course be that the poverty rates for countries collecting income data from a survey might be too high. Maybe the poverty figure of 9 % for the EU-15 might actually be lower if all countries had access to comparable register data? Figure 2 At risk of poverty rate (cut-off point: 50% of median equivalised income) in selected European countries Sources: Eurostat and Rendtel et al (2004) (data for Finland) Denmark Sweden Germany Luxembourg Hungary Netherlands Slovenia Finland (Register) Belgium Austria France Malta Finland (Survey) Bulgaria EU-15 Poland Latvia Lithuania Romania Ireland United Kingdom Estonia Italy Spain Portugal Greece 12

13 4. Conclusion The aim of the paper is to share some of the experiences made in constructing the income module of EU-SILC on the basis of register data. We have demonstrated some of the lack of flexibility that is one of the drawbacks of register data, for instance in the inability to cover the exact income definition that international guidelines require. In addition register data may be vulnerable to administrative changes that may harm comparability over time. The main reason this is the fact that that the primary purpose of administrative registers is public administration and not to provide data for statisticians. There are, nevertheless, reasons to believe that register data have a clear advantage over survey data, for instance in respect to the problem of non-response and erroneous reporting of income, that more than compensates for this lack of flexibility. It should also be mentioned that response burden are significantly reduced when all questions related to income are dropped from the questionnaire and in stead collected from registers. This may boost response rates, particular in panel survey, and saves time and money. Comparisons of data collected from survey and data collected from registers suggest that results may differ in respect to income distribution, where register data show less inequality and lower poverty figures than survey data. This again may have an impact on ranking of countries, for instance by the number of poor people. Statistical producers in countries where there is a tradition to use register data are thus facing a dilemma of whether to use the best national data (i.e. register data), or a probable more comparable data source (i.e. survey data). The EU-SILC Regulation is, however, quite clear in favour of using best national data source and from a national point of view this clearly strengthens the authority and relevance of EU-SILC. The problem of comparability remains, nonetheless, but could be reduced if less emphasis is put on ranking countries by various economic indicators and more focus is put on describing trends and evaluation of progress made. 13

14 References Atkinson, A. B., L. Rainwater & T. M. Smeeding (1995): Income Distribution in OECD Countries, Social Policy Studies No. 18. OECD. Paris. Church, J. & V. Verma (2001): Income Measurement Manual. Working Group Statistics on Income, Poverty & Social Exclusion, Luxembourg, April Epland, J. & G. Frøiland (2002): Husholdningenes inntekter. En sammenligning av nasjonalregnskapet og inntektsundersøkelsens inntektsbegreper. Notater 2002/76, Statistics Norway. Epland, J. & M. I. Kirkeberg (2002): Comparing Norwegian income data in administrative registers with income data in the Survey of Living Condition. Paper presented at the International Conference on Improving Surveys (ICIS), Copenhagen, Denmark. Eurostat (2003): Description of Target Variables: Cross-Sectional and Longitudinal, EU- SILC 065/03. Expert Group on Household Income Statistics (2001): Final Report and Recommendations. Ottawa, Canada. Kavonius, I. K. & V-M. Törmäletho (2002): Contrasting Factor Income of the Income Distribution Survey to National Accounts Primary Income in Finland. Paper prepared for the 27 th General Conference of the International Association for Research in Income and Wealth, Stockholm, August Rendtel, U, L. Nordberg, M. Jäntti, J. Hanisch & E. Basic (2004): Report on Quality of Income Data. Results of Work Package 5 of the CHINTEX project. Final Version: January Shima, I. (2004): The Shadow Economy in Norway: Demand for a Currency Approach, Memorandum 10/2004, Department of Economics, University of Oslo. Statistics Sweden (2002): Disponibel inkomst i nationalräkenskaperna och i inkomststatistiken, Van der Laan, P. & H. T. van Tuinen (1996): Increasing the Relevance of Income Statistics. Experiences and Plans in the Netherlands. Paper prepared for the first meeting of the Expert Group on Household Income Statistics, Canberra, Australia, 2-4 December