Scholarly article on topic 'Health and Wealth on the Roller-Coaster: Ireland, 2003–2011'

Health and Wealth on the Roller-Coaster: Ireland, 2003–2011 Academic research paper on "Sociology"

0
0
Share paper
Academic journal
Soc Indic Res
OECD Field of science
Keywords
{""}

Academic research paper on topic "Health and Wealth on the Roller-Coaster: Ireland, 2003–2011"

Soc Indic Res

DOI 10.1007/s11205-014-0644-4

Health and Wealth on the Roller-Coaster: Ireland, 2003-2011

David Madden

Accepted: 24 April 2014

© Springer Science+Business Media Dordrecht 2014

Abstract The 2003-2011 period in Ireland can be viewed as a roller-coaster with years of high growth followed by years of deep recession. This paper reviews developments in income and health poverty in Ireland over that period using data from the survey of income and living conditions. Income and health poverty are examined both uni-dimensionally and bi-dimensionally using sequential stochastic dominance. Conventional poverty indices are also provided and the correlation between health and income poverty is also analysed Income poverty fell up to and including 2009, after which this trend is reversed. Health poverty shows less of a trend over the period though there is some evidence of a reduction in health inequality from 2006. Movements in bi-dimensional poverty are mostly driven by income poverty, but there is evidence of a reduction in the correlation between health and income poverty over the period.

Keywords Multidimensional poverty • Stochastic dominance • Sequential dominance

1 Introduction

It seems fair to describe Ireland's recent macroeconomic experience as a roller-coaster. The years from 2003 to about 2007 saw the last period of the era starting around 1995 when Ireland became known as the Celtic Tiger.1 However, from about 2008 onwards, Ireland experienced one of the sharpest declines in output in the OECD area as part of what

1 While the phrase ''Celtic Tiger'' is not favoured by all commentators [see for example Honohan and Walsh (2002), who prefer to label it the ''Irish Hare''] by this stage it has become the standard phrase used to describe this period of high growth.

D. Madden (&)

School of Economics, University College Dublin, Dublin 4, Ireland e-mail: david.madden@ucd.ie

Published online: 08 May 2014

© Springer

has become known as the Great Recession (see Nolan et al. 2012a). As Table 1 shows, both GNP and GNP per head showed falls approaching 10 % in 2009, and output continued to fall over the 2010-2011 period (we use GNP as opposed to GDP as net factor flows in Ireland are unusually large and so GNP is regarded as a more accurate measure of "National Income''). The fall in consumption per head has not been quite so dramatic but the turnaround since 2008 is still stark (as explained below our micro data only became available in 2003, so we date our analysis from that year). Unemployment has also increased dramatically from 4.6 % at end 2003 to 14.6 % at end 2011.

Given such dramatic developments in the macro aggregates, what has happened at a more micro level? In particular, how have indicators of living standards such as income poverty and health evolved? It is particularly interesting to look at developments in this area since the recent downturn began and since the first budgetary responses to the crisis were introduced in October 2008.

While it is natural to expect living standards as measured purely by income to fall during a recession, the relationship between economic recessions and health is not entirely clear, as there are a number of different pathways through which an economic downturn could affect health. Clearly an economic downturn implies that financial resources are fewer and this could imply lower spending upon health inputs e.g. on gym membership or other goods which might be expected to have a positive impact upon health. It is also possible that consumers who have private health insurance will lose such insurance if they lose their job. In a country like Ireland there is public health provision but a substantial fraction of consumers also purchase supplementary private health insurance. However, there is evidence that, following the recession, many consumers have chosen not to renew their private health insurance. For example, the number with private health insurance inpatient cover fell from a peak of about 2.3 m in December 2008 to 2.06 m in June 2013, a fall of about 10 %, not to mention the fact that some consumers may also have lowered their level of coverage (Health Insurance Authority of Ireland 2013). The high level of demand for health insurance in Ireland appears to relate to concerns over waiting times for elective procedures (see Watson and Williams 2001). Given such concerns and the reduced public resources available for health provision following the recession, this is a possible pathway whereby the economic cycle could affect health.

Lower incomes may also give rise to increased psychological stress, which in turn could affect other dimensions of health, albeit perhaps with a time lag (see Cooper 2011). However, the lower income arising from recessions may also lead to reductions in risky health behaviours e.g. excessive drinking and eating and smoking. In addition lower car use should also lead to reductions in motor accidents.

Recessions may also give rise to price effects. Higher unemployment and/or lower wages will lead to a reduction in the opportunity cost of leisure, and thus may lead to an increase in consumption of goods which are complementary to leisure and may also have beneficial health effects e.g. walking and other forms of exercise. Using the American Time Use Survey, Colman and Dave (2013) show that recessions can lead to an increase in physical exercise. However this is more than offset by the reduced physical exertion arising from unemployment.

Thus economic theory does not give us any clear indication of what the overall impact of a recession upon health will be, with various influences working in opposite directions. What does the empirical evidence say? Ruhm (2000) has claimed that recessions can be "good for your health'' as risky behaviours such as drinking, smoking and poor diet may be pro-cyclical. More recently however, he suggests that there is now little relationship between health (as measured by overall mortality) and the economic cycle (Ruhm 2013), a

Table 1 Ireland, key economic indicators, 2003-2011

Year GNP % GNP per head % Consumption per Unemployment

change change head % change rate (%)—end year s.a.

2003 4.9 3.2 1.6 4.6

2004 4.0 2.3 2.3 4.4

2005 5.0 3.5 4.8 4.3

2006 5.8 3.9 4.6 4.4

2007 4.2 0.8 2.9 4.7

2008 -1.8 -4.2 -2.5 8.5

2009 -8.1 -9.1 -6.4 12.8

2010 0.9 0.5 0.5 14.4

2011 -2.5 -2.9 -2.9 14.6

Source: Central Statistics Office "National Income and Expenditure" and Census 2006, Labour Force Survey, Quarterly National Household Survey

phenomenon which is mostly driven by a growing counter-cyclicality in cancer fatalities and also some categories of accidental death. He speculates that the pathway with respect to cancer is that modern treatments are expensive, and can be afforded by fewer people in times of recession. Regarding mortality from accidental poisoning it may be that recessions lead to greater mental stress and the medication which is prescribed for this distress may then play a role in accidental poisoning. Similar broadly neutral effects of the recent recession on health have also been found by Tekin et al. (2013) and also Asgeirsdottir et al. (2012).

However, there have also been studies which have suggested that the recent recession may have had an adverse effect upon health outcomes. Bucher-Koenen and Mazzonna (2013) find that the recession has had an adverse impact upon the self-assessed health of older people in some countries, while using aggregate time-series data, Chang et al. (2013) have claimed that the 2008 economic crisis has led to excess suicides. Walsh and Walsh (2011) reviewed the evidence on suicide for Ireland and found limited support for the Chang et al. results. Suicides for young males were significantly affected by unemployment, but this effect was less than the effect of alcohol consumption (which tends to fall in recessions). Barrett and O Sullivan (2013) in a study of people aged 50 and over in Ireland found that the Great Recession had little or no effect on a variety of health measures although there was greater pessimism (amongst the survey respondents) concerning future living standards.

This paper analyses these issues for Ireland over the 2003-2011 period, using individual level data, but compared to Barrett and O Sullivan we have a wider range of age and also a longer time period, though a less detailed set of health measures. Our specific contribution is that not only do we examine developments in income and health in isolation, we also examine them together, paying particular attention to the correlation between the two. This is important in terms of coming to some form of understanding of what has been happening in Ireland to broader measures of welfare over the 2003-2011 period. If the earlier Ruhm hypothesis was correct, then if falls in income are accompanied by improvements in health, then the overall effect on welfare may be ambiguous. On the other hand, if the fall in income is accompanied by either no change or a deterioration in health, then it is highly likely that overall welfare will have fallen, unless some other, unobserved dimension of welfare has risen to offset declines in income and health. It is also important to bear in

mind that we also analyse the correlation between income and health, since even if overall indices of income and health were to be unchanged, if there was an increase in the correlation between health and income poverty, then this could be regarded as an increase in bi-dimensional poverty. It is also important to note that rather than employing mortality or morbidity rates as our measure of health we instead use a broader measure of self-assessed health which we describe in more detail below.

Developments in income poverty alone over the period are analysed using standard poverty indices, while developments in self-assessed health are analysed using the approach pioneered by Allison and Foster (2004). In our analysis of income and health poverty together, we employ the dominance analysis of Duclos and Makdissi (2005) and also calculate bi-dimensional poverty indices. Finally, we also specifically investigate the extent to which the correlation between income and health poverty has changed over the course of the boom and recession.

Before describing the overall layout of the paper, it is useful to review some other work in the area of income poverty for Ireland. The evidence appears to be that since the onset of the crisis in 2008, budgetary policy at least has, in relative terms, been progressive. Callan et al. (2013) provide a review of developments in overall income inequality for the 2003-2011 period and also examine the specific contribution of budgetary policy. They find that overall inequality has changed very little over a long period (from 1994 up to the current period) and this also applies to the more recent recession. However they also find a drop in inequality specifically in 2009, which was reversed almost immediately in 2010. Over the recession period of 2008-2011 they find that budgetary policy has been broadly progressive with greater reductions in incomes for those in the upper half of the distribution and within that half, greater reductions again for the better-off. Developments in the lower half of the income distribution are driven to a large extent by the degree to which old-age pensions have not been cut during the recession, which has relatively favoured those in the 2nd and 3rd deciles of the distribution. The main focus of the Callan et al. analysis is developments in inequality but they also provide some data on changes in poverty and their results are consistent with our findings below.

Nolan et al. (2012a) look at the Irish experience with respect to distribution of household income during the Great Recession. They conclude that while the macro figures as illustrated in Table 1 identify Ireland as one of the countries most affected by the downturn, in relative terms the principal part of the burden has fallen on higher income groups. They also point out that, in common with other OECD countries, though to a more pronounced extent, the impact of the recession on the household sector was considerably less than on the economy as a whole, with the company sector bearing the brunt. Relative poverty, as measured by the fraction below 60 % of median income, fell quite substantially between 2007 and 2009. Of course in the same way that a relative poverty measure can be misleading in times of rapid economic growth (where it can understate improvements in living standards), so too it can be misleading at a time of economic contraction. This is because relative poverty rates can fall even though overall living standards are also falling. Poverty measures based upon an absolute poverty line (i.e. fixed in real terms and not expressed as a fraction of a measure of central tendency) can give quite a different picture, and this is explored in more detail below.

The main conclusion of the Nolan, Callan and Maitre paper is that the initial budgetary responses to the recession and the fiscal deficit in Ireland were mainly concentrated on higher taxes. Since many in the lower part of the income distribution do not pay tax, in relative terms this had a greater impact upon the more wealthy. The same was true of the pension levy charged on public sector workers and the public sector paycuts, since most

public sector workers are in the middle or upper part of the income distribution. Going into 2011 and 2012 the focus of adjustment turned to social welfare payments. While reductions in these payments would be expected to have a greater impact further down the income distribution, Nolan, Callan and Maitre speculated that, even allowing for this, in broad terms the recession would still have had an equalising effect on household incomes.

In a further contribution in this area Nolan et al. (2012b) cover much the same ground as Nolan et al. (2012a) with the advantage of having an extra year of data. Their conclusions for the 2007-2009 period are very similar but they also find that the trend of declining relative poverty was reversed in 2010. It should be noticed that this analysis relies on data from the 2010 Survey of Income and Living Conditions (SILC), data which was subsequently revised. The revised data (which is used in this paper) typically shows that the reversal in the downward trend of poverty witnessed in 2010 was less pronounced than originally thought. However, the qualitative results remain more or less unchanged. It should be noted that a situation where relative poverty is static or rising, combined with falling incomes in general, implies that poverty with respect to a fixed absolute line can rise quite sharply.

Most recently, the Department of Social and Family Affairs (2013) have released a social impact assessment of the main tax and welfare changes introduced in the 2013 budget. As SILC data only runs as far as 2011, their analysis uses the microsimulation model SWITCH. They show that the distributive impact of the combined direct tax and welfare measures depend upon assumptions made regarding the deferral or otherwise of the local property tax. If deferrals of property tax are treated like waivers or exemptions from the charge, then the impact of the welfare and direct tax measures in the 2013 budget could be regarded as broadly neutral as the impact was greatest (in percentage terms) on the middle quintile, with the least impact upon the highest and lowest quintiles. However, if no deferral of the tax is assumed, then the combined impact is regressive, with the greatest percentage loss for the poorest quintile and the least percentage loss for the richest quintile. It should be noted that deferral is not the same as exemption and that it must be assumed that at some stage the tax will be paid. Thus while timing of payment does complicate distributional analysis, it does seem fair to conclude that the impact of the combined measures of the 2013 budget was regressive. Those households who fared relatively the worst were those with children (reflecting changes in child benefit).

So far we have concentrated on developments in household income distribution. But household welfare and poverty are arguably multidimensional concepts and one of the contributions of this paper will be to examine what has happened to multidimensional poverty. Madden (2011) examined the development of the Bourguignon-Chakravarty bi-dimensional poverty index defined over income and health over the 2003-2006 period. He found that uni-dimensional income poverty fell, while health poverty rose and then fell. Movement in the bi-dimensional indices reflected movements in the individual indices and depended upon the relative weights assigned to income and health poverty. However, his analysis did not cover the recent recession.

More recently Whelan et al. (2012) have examined a snapshot of multi-dimensional poverty in Ireland in 2009. They apply the recently developed Alkire and Foster (2011) methodology to a wide sample of European countries, including Ireland. Their measure of multidimensional poverty includes relative income poverty and also various measures of deprivation, including health. However, given that this is just a snapshot of a single year, it is not possible to see how this multidimensional measure has evolved over the boom and subsequent recession. Walsh (2011) examines time-series data on the relationship between

macroeconomic conditions and various indicators of life satisfaction over the 1975-2011 period. While he finds evidence of a positive relationship over the years 1975-1990, he also finds evidence that the relationship has weakened in more recent times, echoing the findings of Ruhm (2013).

Before proceeding to the analysis, it should be stressed that this paper is primarily an exercise in measurement. We measure poverty in income and in self-assessed health, in addition to analysing the correlation between the two. We do not attempt to "explain" developments in poverty using, say, multivariate analysis. While such an exercise would be of considerable interest, it is beyond the scope of this paper and would also have to address non-trivial issues of simultaneity and endogeneity.2

The layout of the remainder of this paper is as follows: in Sect. 2 we describe the income poverty and health measures we use as well as the approach of sequential stochastic dominance. In Sect. 3 we discuss our data and present results, while Sect. 4 offers some concluding comments.

2 Unidimensional and Multidimensional Poverty

The last 10 years or so has seen substantial developments in the measurement of multidimensional poverty. This reflects the fact that poverty (and by corollary welfare) can be viewed as occurring in a number of different dimensions, apart from the most typically used ones of income or expenditure.3 The approach to multi-dimensional poverty analysis can also differ with some authors choosing to calculate multi-dimensional poverty indices (Bourguignon and Chakaravarty 2003) and others looking for more robust multi-dimensional poverty orderings for broader classes of measures (Duclos et al. 2006). One of the more influential recent contributions is that of Alkire and Foster (2011) which provides multidimensional indices which encompass both union and intersection approaches to poverty. Ravallion (2010) in contrast has questioned the need for multidimensional indices at all and suggests instead a ''dashboard'' of multiple indices. A review of this literature which advocates an eclectic approach can be found in Ferreira and Lugo (2012).

Our approach in this paper is to first of all search for robust income and health poverty orderings in one dimension. In the case of health, given that our health measure is ordinal and categorical, we will instead be looking for dominance relationships across the whole of the health distribution, following the approach of Allison and Foster (2004). We then look for sequential poverty dominance following the approach of Duclos and Makdissi (2005) and Duclos and Echevin (2011). Sequential poverty dominance is suited to finding robust poverty orderings in two dimensions when one of the dimensions is an ordered categorical measure. Essentially it involves looking for income poverty dominance between two periods for that group with poorest health, followed by a search for dominance for the two poorest groups and so on. Along the way we also provide evidence on standard unidimensional (for income) and bi-dimensional (covering income and health) poverty indices and we also investigate the correlation between health and income for those who are poor and non-poor.

2 I am grateful to an anonymous referee for this point.

3 For a review of work in this area, see Lustig (2011).

2.1 Univariate Poverty Dominance

We now give a brief account of dominance in the areas of income poverty and health, followed by an account of sequential dominance. We will first briefly run through the analysis of poverty dominance for the single dimension of income (or whatever measure of household resources we are using). Let x be the measure of household resources and let z be the poverty line. Following the exposition in Duclos et al. (2006), when dealing with poverty in a single dimension the stochastic dominance curve for x is given by

Pa(z) = J (z - x)adF(x) 0

where F(x) is the cumulative distribution of x.4 For first order stochastic dominance, poverty does not increase for any possible choice of z when moving from a distribution A to a distribution B, if the incidence of poverty under distribution A is never greater than under distribution B (i.e. where a = 0). If this condition for first order dominance is not met, then second order dominance (a = 1) may be investigated which requires that normalised poverty deficits should not increase for any possible choice of z when moving from A to B. Thus, in general, orders of dominance s = a + 1 can be examined. It is also possible to search for dominance over a more restricted range of poverty lines.5

2.2 Health Dominance

The approach in the previous section is suitable when the variable x is cardinal. However, when dealing with health, it is often the case that cardinal measures which cover general health will not be available.6 More typically, measures which address general health may come in the form of an ordered, categorical self-assessment of health. In that instance the analyst has two choices: either to employ a dominance approach which is specifically designed to deal with ordered categorical data, or else to transform the ordered measure into a cardinal measure. The transformation of ordered categorical data into cardinal data has been discussed extensively by Van Doorslaer and Jones (2003), but as Madden (2010) points out, results obtained can be sensitive to the approach adopted.

As we wish to retain our data in its original, ordered categorical format, we choose to adopt a dominance approach specifically designed to deal with this sort of data, the approach of Allison and Foster (2004), henceforth AF. While this approach has the advantage of being specifically designed to deal with ordered categorical data, it carries a disadvantage in that it is ill-equipped to deal with poverty dominance. This is because of the difficulty in identifying a poverty line for categorical data. Thus the AF approach could be best described as a welfare dominance (with respect to health) approach.

The measure of self-assessed health we have is the following: individuals answer a question of the form: what is your general health status? The possible answers are: very bad, bad, fair, good and very good. While this measure appears to give a good indicator of

4 In common with most of the literature in this area, we choose to work with the continuous as opposed to the discrete version of poverty measures when dealing with dominance issues.

5 Note that poverty dominance was first introduced by Atkinson (1987).

6 While there may be cardinal measures for certain dimensions of health such as blood pressure or BMI, in the absence of health indices such as the SF-36 it is usually the case that general health measures are ordinal and categorical.

overall health (Idler and Benyamini 1997) it is not cardinal, and with only five categories, it is not suited to the application of standard poverty and inequality indices. This is because standard measures of the spread of a distribution which use the mean as a reference point, such as the Gini, are inappropriate when dealing with categorical data, since the inequality ordering will not be independent of the (arbitrarily chosen) scale applied to the different categories.

The AF approach assumes we have a measure of SAH with n different categories which can be clearly ordered 1,..., n. Let m denote the median category and let R and T denote two cumulative distributions of SAH with Ri and Ti indicating the cumulative proportion of the population in category i, in each distribution, where i = 1, ..., n. In this case distribution R will dominate distribution T if the cumulative frequency at each point on the ordinal scale (as we go from lower to higher) is always higher in T than in R. This is equivalent to the first order stochastic dominance condition referred to above. For an example of application of this approach to a comparison of SAH between different social classes, see Dias (2009).

AF also provide a partial inequality ordering based on a median-preserving spread of the distribution (analogous to the partial ordering based on a mean preserving spread provided by say a Lorenz comparison). For the case where both R and T have identical median states m then R has less inequality than T if for all categories j < m, Rj < Tj and for all j > m, Rj > Tj. What this is effectively saying is that distribution T could be obtained from distribution R via a sequence of median-preserving spreads.

In Sect. 3 we will present results for health and health inequality dominance using the AF approach. First we give a brief account of sequential poverty dominance.

2.3 Sequential Poverty Dominance

In terms of producing poverty indices, multidimensional analysis works best when attributes can be measured cardinally. As discussed above, general health measures are usually only available on an ordinal basis, such as measures of self-assessed health. Since these measures typically give health status on the basis of a discrete number of categories, there is a clear ranking and it is this clear ranking which permits the application of multidimensional analysis to situations where one of the attributes is ordinal in nature.

As outlined in Duclos and Makdissi (2005) suppose that the population can be divided into K mutually exclusive and exhaustive subgroups with population share defined by /(k), k = 1,...,K and Y1 K= 1 /(k) = 1. The subgroups could be defined over a wide range of characteristics but what is most important is that, for a given measure of the continuous attribute (income), certain groups can be reasonably viewed as having lower overall well-being than others. Thus, for a given level of income, an individual with, say zero literacy, has lower overall well-being than someone with 100 % literacy, or someone with "very bad'' health has lower well-being than someone with ''very good'' health.7 We assume that, for given alternative indicators of well-being, the K subgroups can be ordered in decreasing value of needs, so that group 1 has greater needs than group 2 etc.

Duclos and Makdissi (2005) demonstrated that if we start off with the neediest group (k = 1) then for first order poverty dominance to hold for distribution A over distribution B

7 Note that it is possible that responses to the self-assessed health question may differ systematically, so that, for example, an older person may view a given health state as ''very good'' while a younger person might view it only as "good". This issue is typically addressed via anchoring vignettes and is beyond the scope of this paper. For a general discussion, see King and Wand (2007).

we first require that the headcount for this group should be higher in distribution A compared to B. We then require that the cumulative headcount for the two neediest groups (k = 1, 2) should be higher in A than B and so on. Thus the sequential cumulative headcount for all groups up to where we reach group K should be higher for A compared to B, where the sequence is carried out starting with the neediest group etc. Note that it does not require that each subgroup k have more poverty independently in A. Poverty for, say, group 3, could be lower in A compared to B, as long as cumulative poverty for groups 1-3 is higher in A. For a recent application of this to health and income data from Canada and the US, see Duclos and Echevin (2011).

For our example here we investigate income poverty for 2003 through to 2011 where the population is partitioned into four groups on the basis of self-assessed health. Even though we have five categories of self-assessed health, as the fraction of the population reporting ''very bad'' health is so small (typically less than 1 %), we combine the two lowest categories.

3 Data and Results

Our data comes from nine consecutive cross-sectional surveys (2003-2011) which are the Irish part of the European Union Survey of Income and Living Conditions (EU-SILC).8 This survey is the successor to the European Community Household Panel survey. After allowing for missing observations for certain variables the sample sizes are between 13,000 and 14,000 for each year. However, in Ireland there was only 6 months of data collection for 2003 (as opposed to 12 months collection for the other years) hence the sample size for 2003 is only about half of that for the other years (see CSO 2007).

As our income measure we use equivalised income after social transfers, using the EU definition of income (details of this measure are included in the ''Appendix'') and the modified OECD equivalence scale (1.0 for first adult, 0.5 for subsequent adults and 0.3 for children aged less than 14). The ordinal health measure we use is based on responses to a question concerning self-assessed health. The self-assessed health question asks: ''in general, how good would you say your health is?'' The possible answers are: very bad, bad, fair, good and very good. We confine our analysis to those aged 16 and over, as the health question was not put to those aged younger than 16. This reduces our sample size by around 2000-3000 each year. Average incomes are slightly higher for the reduced sample, and while the moments of the distributions appear to be quite close together, the Kolmogorov-Smirnov tests for equality of distributions rejects the null for each year.9 However, when we analyse the FGT Pa poverty indices for income only for the two samples, the trends over time are practically identical. We also carried out the analysis on the data excluding under 25 s (this was only possible for the years 2006-2011 as SILC did not provide a more detailed age breakdown for 2003-2005) and the qualitative results were very similar. While it is regrettable that health data for the under 16 s is not available, on balance we do not believe that our qualitative results are greatly affected by working with the smaller

8 For details of the Irish part of EU-SILC see CSO (2007) and the documentation at http://www.cso.ie/ eusilc/default.htm.

9 Details of the comparison between the distribution of income with and without those aged 16 and younger are available on request.

sample, and if we wish to include health in our poverty analysis, then unfortunately we have no other option.

In Table 2 we provide summary statistics for the frequencies of the categories of self assessed health and for mean equivalised income. Equivalised income is presented in 2010 prices. Note that in order to remove the influence of outliers we trim the data of the top and bottom 0.5 % (by non-equivalised income).10 Table 2 shows that mean equivalised income rose continuously from 2003 to 2007. It was essentially constant between 2007 and 2009 but then dropped quite sharply in 2010 and 2011.

3.1 Health Dominance

The position with respect to self-assessed health is a little more complicated. Between 2003 and 2011 the fraction declaring very good health dropped from about 50 % to about 43 %. For the most part this has been offset by an increase in the fraction declaring good health. However as explained above, unless dominance is observed, it is not possible to state that health has improved, without assigning an arbitrary scale to each health category. In Table 3 we present a grid which shows whether dominance applies. As the frequencies in the category ''very bad'' are so small, we combine the two lowest categories together. Recall that if year A stochastically dominates year B then the cumulative frequency for year B at each point on the ordinal scale (as we go from lower to higher) is always higher in B than in A. Note also that entries which are above the main diagonal indicate a situation where the earlier year (the row year) dominates the later one (the column year), while entries below the diagonal indicate where the later year dominates the earlier one. In Table 3 ''F'' refers to first-order dominance and we can see there are a number of instances of first order dominance over the period. The greatest predominance is for the years 2007 and 2008 indicating that health in these years dominated other years, particularly so for 2008. 2011 is dominated by 2008 and 2009, which suggests some deterioration since the onset of the recession, but in truth, the differences in the percentages for each category are so small that it is hard to argue that they are economically meaningful.

What about the spread measure introduced by AF? Recall that spread dominance, as denoted by ''S'' indicates greater spread in the row year compared to the column year. Thus the distribution for year A has greater spread than that for year B if (a) year A and year B both have the same median category, m, (b) for all categories below the median the cumulative frequency in A is at least as great as in B and (c) for all categories greater than or equal to the median the cumulative frequency in B is at least as great as that in A. Thus instances of S below the main diagonal indicate that health inequality has been decreasing over time, and this seems to be the case. For example, in 2011 there was less health inequality than all years from 2003 to 2006 inclusive.

Thus to summarise, the grid in Table 3 indicates first order dominance for 2007 and 2008 over many of the preceding and succeeding years. However changes since the onset of the recession have been very marginal. The data also indicate that health inequality appeared to decline over the period. What is also noticeable is that health inequality did not increase as the recession began, as is indicated by the greater spread in 2004-2006 compared to the 2008-2011 period.

10 We also carried out the analysis without the data trimming and the qualitative results were very similar. Results available on request from author.

Table 2 Summary of self-assessed health and equivalised income (mean, 2010 prices)

Very bad Bad Fair Good Very good Equiv Y

2003 (N = 6,101) 0.00869 0.0259 0.1335 0.3267 0.5053 214.65

2004 (N = 10,896) 0.0094 0.0283 0.1349 0.3567 0.4707 222.55

2005 (N = 11,915) 0.0083 0.0276 0.1353 0.3576 0.4712 228.97

2006 (N = 11,360) 0.0072 0.0243 0.1373 0.365 0.4662 237.60

2007 (N = 10,778) 0.0051 0.0209 0.1326 0.3739 0.4675 254.24

2008 (N = 10,013) 0.0035 0.0222 0.1296 0.3728 0.4719 250.84

2009 (N = 9,800) 0.0045 0.0232 0.1377 0.3966 0.4379 254.12

2010 (N = 8,704) 0.0071 0.0249 0.1363 0.3801 0.4517 237.39

2011 (N = 8,116) 0.0046 0.0238 0.1375 0.4002 0.4339 224.65

Source: Central Statistics Office, Survey of Income and Living Conditions (SILC), 2003-2011

Table 3 Health dominance

2003 2004 2005 2006 2007 2008 2009 2010 2011

2003 F F

2005 F

2006 S S

2007 S S F F F F

2008 F F F F F F F

2009 S S S F

2010 S S

2011 S S S S

3.2 Income Poverty Dominance: Fixed Poverty Line

What about poverty dominance in the area of income? First we will look at the case where the poverty line is fixed in real purchasing power terms. We now need to fix what we could view as a "reasonable" range for the poverty line to lie within. We chose a range from zero to 80 % of median equivalised income for the year when equivalised income was highest (2007), noting that a typical value for a poverty line is 50-60 % of median equivalised income. That gives a figure (in 2010 prices) of just over €170 and provided we observe dominance for one distribution over another up to the upper limit of our poverty line, we regard that as poverty dominance.11

Once again we analyse this using a grid. Table 4 shows the grid for poverty dominance. If poverty is declining over time, then we expect to see entries below the main diagonal. We use "F" to indicate first order dominance and "S" to indicate second order dominance.

11 We analyse dominance using the "dompov" command in the DASP package of Ararr and Duclos (2012).

We note many instances of F below the diagonal up to about 2009. However, after 2009 we note that entries appear above the main diagonal, along the 2010 and 2011 columns. This implies that these years are poverty dominated by earlier years and indicates a pattern whereby poverty was falling up to and including 2009, but then this was reversed in 2010 and 2011.

Note that we indicate a number of entries as F*. By this we mean that poverty dominance applies across the range of income up to the poverty line except for a crossing at very low levels of income (below the first percentile), crossings which we believe are more likely to reflect measurement error rather than genuine crossings of the poverty incidence curves. We also include a category which we label WF which indicates weak first order dominance. This is the situation where the difference between poverty incidence curves was not statistically significant over some part of our poverty range, but that it was statistically significant over another part. We do not observe dominance if either there is no part of the poverty range where there is a statistically significant difference or if over our range we observe two (or more) instances of a statistically significant difference but of opposite sign [this is essentially the criterion suggested by Bishop et al. (1991)].

Thus Table 4 indicates falling poverty for the first 2 years of the recession, 2008 and 2009, but a reversal of this in later years. This is consistent with the findings of Callan et al. (2013) and Nolan et al. (2012b).

For completeness sake in Tables 5 and 6 (and Figs. 1, 2) we present poverty indices for the well known Pa measures of Foster et al. (1984, henceforth known as FGT) for fixed and relative income poverty lines. We show the results for three values of a, 0, 1 and 2. An a value of 0 refers to a headcount measure, a value of 1 gives a measure which is sensitive to the depth of poverty while a value of 2 indicates a measure which places a greater weight on the poverty shortfalls of the very poorest. These tables bear out the previous results whereby poverty falls pretty consistently up to and including 2009, but then this trend is reversed quite sharply in 2010 and 2011, so that poverty indices return to values previously seen around 2005-2006, for the absolute poverty line. It is also worth noting that the rise in poverty starting in 2010 is proportionately greater for the P2 measure, which is more sensitive to income distribution within the poor.12 For the case of a relative income poverty line (60 % of median equivalised income), the pattern is very similar, except that the decline and subsequent reversal is not quite so pronounced.

Thus to summarise so far in terms of univariate analysis: income poverty fell consistently up to and including 2009 and this was reversed in 2010 and 2011. Thus for the early part of the recession at least the evidence is that income poverty did not rise. Owing to the categorical nature of self-assessed health it is more difficult to come to any definitive conclusion here. There is a fall in the proportion reporting very good health. However there are also falls in the proportions reporting very bad and bad health. The proportions involved moving out of the lowest two categories are quite small compared to those moving out of the highest category, so unless there was a considerably higher weight attached to the improvements in health in the lower categories, the likelihood is that most scales would probably record a deterioration in health, although most of that deterioration had occurred by 2006. What also seems clear is that since about 2007 there has been a reduction in health inequality.

12 To economise on space we do not present standard errors for these statistics but they are available on request.

Table 4 Equivalised income poverty dominance (fixed poverty line)

2003 2004 2005 2006 2007 2008 2009 2010 2011

2004 WF

2005 F WF WS

2006 F* F* WF S

2007 F* F* F* WF S WF F

2008 F F* WF WF WF F

2009 F F* F* WF WF F F

2010 WF WF F

WF no statistically significant dominance of column over row, statistically significant dominance for row over column for some range of poverty line, F first order dominance, S second order dominance

Table 5 Poverty indices, 2003-2010—fixed poverty line, 60 % of median 2007 income

Year P0 P1 P2

2003 0.273480 0.076970 0.032063

2004 0.263301 0.063769 0.022254

2005 0.244228 0.056857 0.019733

2006 0.220867 0.047143 0.015028

2007 0.170925 0.034029 0.010609

2008 0.158499 0.032806 0.011404

2009 0.138416 0.030122 0.011621

2010 0.177784 0.042139 0.018765

2011 0.211248 0.057624 0.027463

Table 6 Poverty indices, 2003-2010— relative poverty line, 60 % of median income

Year P0 P1 P2

2003 0.212015 0.052117 0.021452

2004 0.214403 0.045720 0.015244

2005 0.201265 0.044659 0.015002

2006 0.187219 0.036553 0.011260

2007 0.170925 0.034029 0.010609

2008 0.157737 0.032591 0.011330

2009 0.149943 0.031861 0.012209

2010 0.147275 0.035399 0.016432

2011 0.154679 0.042398 0.021561

Fig. 2 Pa measures, income, 2003-2011, relative poverty line (2003 = 100)

3.3 Sequential Poverty Dominance

We now move on to sequential poverty dominance for income and health. Recall that this involves looking for income poverty dominance between 2 years for the most needy group, followed by looking for dominance for the two neediest groups, then the three neediest groups and so on. In our case we identify ''needy'' with self-assessed health. However as indicated above, since the proportions in the neediest group (very bad health) are so small, typically less than 1 %, we combine the two lowest health categories. Thus our neediest group is those with very bad and bad health, the next neediest is those with fair health and so on.

As before, we present our dominance results in Table 7 using a grid. The pattern is very similar to that in Tables 3 and 4. We observe many instances of weak sequential dominance below the main diagonal and the only instances above the diagonal are for 2010 and 2011, once again indicating that even when we look at developments exploiting the distribution of two dimensions, 2010 shows a major reversal of what had been happening in earlier years, and this continues in 2011.

The similarity in results between Tables 4 and 7 is notable. It is helpful to examine in more detail one instance where the dominance results differ. This is in the comparison between 2005 and 2006. Table 4 indicates that for income dominance only, 2006 weakly dominates 2005. Figure 3 shows the difference in poverty incidence curves (along with

Table 7 Sequential stochastic dominance, income and health

2003 2004 2005 2006 2007 2008 2009 2010 2011

2005 WF WF

2006 WF WF

2007 WF WF WF WF

2008 F WF WF WF WF WF WF

2009 F WF WF WF WF WF WF WF

2010 WF WF

WF no statistically significant dominance of column over row, statistically significant dominance for row over column for some range of poverty line, F first order dominance

95 % confidence intervals) between the 2 years. As can be seen, the difference is negative (indicating lower incidence in 2006 compared to 2005) and it is statistically significant for much of the range of the poverty line. Figure 4 however shows the difference between poverty incidence curves for the neediest group (those with very bad and bad health). There is no range of the poverty line where we observe statistically significant dominance and hence dominance does not apply. Thus while poverty dominance was to be observed for the population as a whole, it was not observed for the neediest, indicating that improvements in income between 2005 and 2006 were more concentrated amongst the healthier (or less needy). Analysis of income by health group show that average incomes for those with ''very bad'' and "bad" health (our neediest group) fell between 2005 and 2006, while average incomes for the three less needy groups all rose. A corollary of this is that polarisation of income by health group increased between 2005 and 2006, with the less healthy becoming poorer while the relatively healthier groups experienced income increases (this is confirmed by the change in the Duclos, Esteban and Ray polarisation index which increased between 2005 and 2006 from 0.199 to 0.203, a change which is significant at the 10 % level).

3.4 Multidimensional Poverty Indices

As a complement to the sequential analysis above, we also calculate multidimensional poverty indices. As discussed above an important issue in the choice of index is whether to adopt an intersection or union approach. The advantage of the Alkire and Foster (2011) methodology is that the union and intersection choices are extreme cases of a more general approach, and a compromise or intermediate position is possible. If we have m possible dimensions of poverty then the compromise involves counting poverty in k dimensions where k lies between one (a union approach) and m (an intersection approach). However, when working with only two dimensions, no such k exists and so it is easier to present results for both the union and intersection approaches. Bearing in mind also the criticisms of Ravallion (2010), in addition we present results showing the degree of dependency

Difference between FGT curves

(alpha = 0)

Poverty line (z)

Confidence interval (95 %) - Estimated difference

Fig. 3 Difference between poverty incidence curves 2005-2006

Difference between FGT curves

Poverty line (z)

Confidence interval (95 %) - Estimated difference

Fig. 4 Difference between poverty incidence curves 2005-2006 for neediest group

between the distributions of health and income. Given that we have already provided information on the marginal distributions of health and income, information on the dependence between these distributions could be regarded as constituting the final element in the ''dashboard'' (see Ferreira and Lugo 2012).

Table 8 (and Figs. 5, 6) give the headcount poverty rates for both intersection and union approaches for the period 2003-2011. Note we only present headcount measures as gap or weighted gap measures would not be appropriate given that our health measure has no cardinal interpretation i.e. the nature of our health data tells us that someone is below the health poverty line, but cannot tell us how far below the poverty line they are in any sort of

meaningful unit. We also include two different income poverty lines, one representing a fixed amount in real purchasing power (60 % of median 2007 income), and the other simply 60 % of median income of the year in question. There is also a choice to be made concerning the health poverty line. Given the classifications available the two obvious candidates are those with health less than or equal to ''bad health'' or the category of ''fair health''. Thus in Figs. 5 and 6, "I" and ''U'' refer to the intersection and union approaches respectively while ''3'' and ''4'' refer to the location of the health poverty line.

For the fixed income poverty line and intersection approach we observe sharp declines in poverty up to and including about 2009. We then see a levelling out for the lower of the health poverty lines (HPov = 3) but an increase for the higher of the lines (HPov = 4). This suggests an increase in income poverty for those with health in the ''fair health'' category. The union approach sees quite a sharp pick-up in poverty from 2010. Given the parallel movement in income poverty it seems likely that people who previously had neither income nor health poverty may now be falling into income poverty.

Results for the relative income poverty line are quite similar, except in the intersection case with the higher health poverty line, where multidimensional poverty continues falling through 2010 and 2011. What this seems to indicate is that the increase in relative income poverty which occurred between 2009 and 2011 was more concentrated amongst the more healthy, since the intersection measure shows a fall between the 2 years, suggesting that the households who moved into income poverty (as evidenced by Table 6) were not, for the most part, households which were health poor.

We further explore this issue by looking at FGT poverty indices for each health grouping. One of the attractive features of the FGT index is that if the population can be decomposed into mutually exclusive and exhaustive subgroups then any of the FGT indices can be expressed as the population weighted sum of the FGT indices for each subgroup. We divide our population into four subgroups corresponding to the self-assessed health categories (where once again we combine the categories ''bad'' and ''very bad''). A correlation between health and income poverty would suggest that the FGT index for the less healthy subgroups should be higher than for the more healthy subgroups. Figures 7, 8 9 shows the FGT Pa measures for a = 0, 1, 2 for the four subgroups. As expected the index is higher for the less healthy groups, but what is also noticeable is how the difference in the indices between the subgroups has been narrowing over the years. This further confirms the conjecture that the correlation between health and income poverty is weakening.

As a final check on this weakening correlation between health and income poverty, in Table 9 we present measures of dependence between income and health. The measures we use are the polyserial correlation coefficient, the Spearman rank correlation coefficient and the Kendall tau-b rank correlation coefficient. These measures were chosen in preference to the standard Pearson correlation coefficient, since the value of this measure would depend upon the (arbitrary) scale used in the ordinal health measure. We present them for the complete distribution and also for those observations below the income poverty line and below the health poverty line. For the income poverty line we use the same upper limit as was used in the dominance analysis (80 % of median income in 2007) and for the health poverty line we define as health poor those with health less than or equal to ''fair''.

Looking at the dependence measures for the distribution as a whole we see that dependence has been falling since about 2007. Between 2003 and 2007 there was some fluctuation but no real trend evident. However, since 2007 all three measures of dependence show a decline. This decline is consistent with developments below the poverty lines. Just looking at the subset of people who are income poor, we can identify three phases. From about 2003 to about 2005 the Spearman and Kendall tau correlations hover

Table 8 Bi-dimensional poverty indices

Fixed income poverty line Relative income poverty line

Intersection Union Intersection Union

HPov = 3 HPov = 4 HPov = 3 HPov = 4 HPov = 3 HPov = 4 HPov = 3 HPov = 4

2003 0.020 0.089 0.288 0.353 0.014 0.068 0.233 0.312

2004 0.021 0.086 0.280 0.350 0.018 0.071 0.235 0.316

2005 0.017 0.073 0.263 0.342 0.014 0.062 0.223 0.310

2006 0.015 0.063 0.237 0.326 0.013 0.054 0.206 0.302

2007 0.010 0.051 0.187 0.279 0.010 0.051 0.187 0.279

2008 0.006 0.038 0.178 0.276 0.006 0.038 0.177 0.275

2009 0.006 0.034 0.160 0.270 0.006 0.037 0.171 0.279

2010 0.007 0.040 0.203 0.306 0.006 0.032 0.173 0.283

2011 0.007 0.045 0.232 0.332 0.005 0.030 0.178 0.291

Fig. 5 Intersection and union bi-dimensional poverty indices, fixed income poverty line, 2003-2011 (2003 = 100)

Fig. 6 Intersection and union bi-dimensional poverty indices, relative income poverty line, 2003-2011 (2003 = 100)

Fig. 7 FGT decomposition by self-assessed health, a = 0, 2003-2011

around about 0.1, and all are statistically significant. For 2006 and 2007 the correlation drops to about 0.05, still statistically significant, though arguably not economically very significant. Since 2008 however, the correlation has vanished and by 2010/2011 it had even

Fig. 9 FGT decomposition by self-assessed health, a = 2, 2003-2011

turned negative. For the subset of people who are health poor, the series is somewhat more volatile, with generally lower significance levels. Nevertheless, the trend of declining dependence in recent years is still evident.

One possible explanation for the reduced correlation between health and income is the experience of pensioners (those aged 65 and over). As has been documented by Nolan et al. (2012b) the relative position of pensioners in income terms has improved significantly in recent years. Since this group in general have poorer health than the non-pension population, their relative improvement in income terms could explain the apparent decoupling between health and income poverty. In Table 10 we re-calculated Table 9 for the nonpension population and found that in qualitative terms the results were quite similar. This is illustrated in Figs. 10 and 11 (for ease of visual interpretation we just include the Spearman correlation coefficient). For the under-65 group as a whole (i.e. including the non-poor) there is a slight decline in the correlation. However for the health and income poor from about 2007 there is a clear fall in the correlation (albeit from a fairly low level). Perhaps the main difference compared to the population including older people is that the negative correlation between income and health below the income poverty line in 2010/2011 which is evident in Table 9 is not present in Table 10. Thus while the relative improvement of the pension population over the period under review may explain part of the reduced correlation between health and income poverty, it is not the complete story.

Table 9 Dependence measures between health and income

Total distribution Below income Pov line Below health Pov line

PS Sp KT PS Sp KT PS Sp KT

2003 0.293 (0.015) 0.275* 0.212* 0.058 (0.025) 0.137* 0.104* 0.105 (0.051) 0.045 0.036

2004 0.295 (0.015) 0.296* 0.228* 0.044 (0.019) 0.120* 0.092* 0.167 (0.04) 0.089* 0.071*

2005 0.294 (0.011) 0.299* 0.230* 0.039 (0.018) 0.115* 0.088* 0.059 (0.035) 0.051+ 0.041+

2006 0.273 (0.013) 0.274* 0.212* 0.017 (0.020) 0.059* 0.045* 0.179 (0.043) 0.096* 0.077*

2007 0.306 (0.014) 0.296* 0.229* 0.009 (0.021) 0.05* 0.038* 0.149 (0.049) 0.067* 0.054*

2008 0.262 (0.013) 0.261* 0.201* -0.008 (0.020) -0.013 -0.01 0.013 (0.04) 0.001 0.001

2009 0.248 (0.013) 0.243* 0.188* -0.017 (0.021) -0.035+ -0.027+ 0.060 (0.04) 0.028 0.023

2010 0.214 (0.013) 0.210* 0.162* -0.015 (0.021) -0.035* -0.027* 0.030 (0.043) 0.007 0.006

2011 0.216 (0.013) 0.199* 0.153* -0.020 (0.022) -0.045* -0.035* -0.003 (0.044) -0.007 -0.006

4 Summary and Conclusions

We have presented a considerable number of results in the previous section, so it is useful to try to draw together our conclusions. Looking at the univariate analysis, we see that income poverty fell quite consistently up to about 2009, but there was a sharp reversal of this trend in 2010, which continued in 2011. Developments in health poverty are more difficult to assess, as it is more difficult to arrive at a poverty line when data is ordinal. But the dominance analysis carried out suggest that health improved up to around 2008, and has deteriorated slightly since then. Health inequality has also been falling since about 2007. Overall though, and acknowledging the difficulty in assessing changes over time in an ordinal measure, what changes we have observed in self-assessed health over the period, and particularly since 2008 have been very modest.

Turning now to the bi-dimensional analysis, the dominance analysis for the most part mirrors what happened with univariate income poverty analysis. Once again this is consistent with a situation where developments in health have been less dramatic. Perhaps the most interesting development in this area has been the reduced correlation between income and health over the period, both for the population as a whole and also for those experiencing health and income poverty. This is consistent with much of the literature cited earlier and also consistent with the Irish analysis of Barrett and O Sullivan.

To some extent our findings can be explained by the experience of the 65 years and over age group, but only partly. It is possible that developments in health will eventually follow those in income with a lag (though it is likely that causality runs both ways). It is also possible that budgetary changes which affect the provision of health care will also feed into health over time and that this may have different impacts across the income distribution. For the present however, given this apparent decoupling, the likelihood is that developments in bivariate poverty will for the most part be driven by developments in income poverty.

Are there any policy conclusions which can be drawn from this analysis? In one sense we are confronted with the phenomenon of the "dog which doesn't bark'' in the sense that health is relatively unchanged despite the economic cycle. However, as other studies have shown, the message is somewhat more complex than this. While measures of overall health such as self-assessed health and overall mortality do not seem to be affected by the economic cycle, this is partly a reflection of individual factors cancelling each other out. Thus should economic growth resume again, it is quite likely that adverse health outcomes arising from diet and lifestyle may increase, as well as mortality from motor accidents. In terms of other dimensions of health, to the extent that treatments for conditions such as cancer become more expensive (though producing better outcomes) there will be complex issues in terms of determining access to such treatments. If treatments are primarily determined by public spending upon health, then fatalities from these sources are likely to be counter-cyclical. Even if treatment is financed from a mixture of public and private resources, it seems likely that the socioeconomic gradient of access to such treatment may become steeper in times of recession, perhaps reversing the trend which seems to have emerged in recent years. It is important here to bear in mind the distinction between the overall relationship between health and income over the economic cycle, and the socioeconomic gradient within a given cross-section.

In terms of policy response, for those dimensions of health where morbidity and mortality appear to be pro-cyclical, the challenge will be to provide information and incentives which will limit the degree of risky behaviour. Already the Irish government has considered some such policies with debate over the possible introduction of a food-related

Table 10 Dependence measures between health and income for population aged under 65

Total distribution Below income Pov line Below health Pov line

PS Sp KT PS Sp KT PS Sp KT

2003 0.229 (0.018) 0.197* 0.154* 0.067 (0.030) 0.164* 0.126* 0.137 (0.065) 0.083+ 0.066+

2004 0.233 (0.014) 0.224* 0.175* 0.048 (0.022) 0.141* 0.108* 0.192 (0.049) 0.108* 0.087*

2005 0.237 (0.013) 0.232* 0.181* 0.041 (0.021) 0.131* 0.101* 0.058 (0.046) 0.067+ 0.053+

2006 0.222 (0.015) 0.219* 0.171* 0.026 (0.023) 0.088* 0.067* 0.229 (0.059) 0.136* 0.109*

2007 0.259 (0.017) 0.245* 0.192* 0.020 (0.025) 0.092* 0.070* 0.169 (0.068) 0.089* 0.072*

2008 0.220 (0.015) 0.216* 0.169* 0.005 (0.025) 0.028 0.021 0.045 (0.053) 0.020 0.017

2009 0.221 (0.005) 0.203* 0.159* -0.009 (0.024) -0.014 -0.011 0.083 (0.058) 0.055++ 0.045++

2010 0.195 (0.016) 0.198* 0.154* 0.007 (0.025) 0.030 0.023 0.028 (0.054) 0.011 0.009

2011 0.206 (0.015) 0.189* 0.146* 0.006 (0.025) 0.023 0.018 0.079 (0.057) 0.036 0.027

Fig. 11 Measure of dependency, under 65 s, 2003-2011

health tax (for a discussion of this see Madden 2013) although no decision has yet been made in that regard. However, a decision has been made to introduce a legal minimum price for alcohol (for a discussion of the issues involved here see Griffith et al. 2013).

Policy issues with respect to morbidities and mortalities from other causes such as cancer mostly centre around access. Ireland currently has a mix of private and public health provision, though as noted in the introduction, the recent recession has led to a reduction in private health insurance. If this trend continues then clearly there may be a greater demand for public health resources to provide treatment and the level and quality of such treatment is likely to be affected by public budgetary considerations which in turn will be affected by the economic cycle.

Considering these two policy areas together, once again we are faced with a situation where there are offsetting forces. Thus, the balance of probability suggests that the recent phenomenon of no clear link between overall health and the economic cycle may persist for some time yet.

Appendix: Definition of Income

Definition of Income: The income measure we use is equivalised income after social transfers using the EU definition of income and the modified OECD equivalence scale. The EU definition of income consists of:

• Direct income (employee cash and non-cash income)

• Gross cash benefits or losses from self-employment

• Other direct income (but not pensions from individual private plans, value of goods produced for own consumption, employer's social insurance contributions)

• All social transfers (e.g. unemployment benefits, housing allowances, sickness allowances etc.).

Tax on income and contributions to state and occupational pensions are deducted from this to give disposable income, which is then adjusted to equivalised income by applying the modified OECD scale (1.0 first adult, 0.5 other adults, 0.3 children aged less than 14). For details see CSO (2007). The unit of analysis is all adults (i.e. those aged 16 and over) in the household.

References

Alkire, S., & Foster, J. (2011). Counting and multidimensional poverty measurement. Journal of Public Economics, 95, 476-487.

Allison, R. A., & Foster, J. (2004). Measuring health inequality using qualitative data. Journal of Health Economics, 23, 505-524.

Ararr, A., & Duclos, J. Y. (2012). DASP: Distributive analysis STATA package. Universite Laval and World Bank.

Asgeirsdottir, T., Corman, H., Noonan, K., Olafsdottir, P, & Reichmann, N. (2012). Are recessions good for your health behaviours? Impacts of the economic crisis in Iceland. NBER working paper 18233.

Atkinson, A. (1987). On the measurement of poverty. Econometrica, 55, 759-764.

Barrett, A., & O Sullivan, V. (2013). A short note on the wealth, health and well-being of ireland's older people before and during the economic crisis, SHARE Working Paper Series, WP 15-2013.

Bishop, J. A., Formby, J., & Smith, W. J. (1991). Lorenz dominance and welfare: Changes in the U.S. distribution of income, 1967-1986. Review of Economics and Statistics, 73, 134-139.

Bourguignon, F., & Chakaravarty, S. (2003). The measurement of multidimensional poverty. Journal of Economic Inequality, 1, 25-49.

Bucher-Koenen, T., & Mazzonna, F. (2013). The recent economic crisis and old-age health in Europe. In A. Borsch-Supan, M. Brandt, H. Litwin & G. Weber (Eds.), Active ageing and solidarity between generations in Europe. First results from SHARE after the economic crisis, pp. 233-242. De Gruyter.

Callan, T., Nolan, B., Keane, C., Savage, M., & Walsh, J. (2013). Crisis, response and distributional impact: The case of Ireland. ESRI working paper, No. 456.

Chang, S., Stuckler, D., Yip, P., & Gunnell, D. (2013). Impact of 2008 global economic crisis on suicide: Time trend study in 54 countries. British Medical Journal, 347, f5239.

Colman, G., & Dave, D. (2013). Exercise, physical activity and exertion over the business cycle. Social Science and Medicine, 93, 11-20.

Cooper, B. (2011). Economic recession and mental health: An overview. Neuropsychiatry, 25, 113-117.

CSO. (2007). EU survey on income and living conditions. Dublin: Stationery Office.

Department of Social Community and Family Affairs. (2013). Social impact assessment of the main welfare and direct tax measures of the 2013 budget. http://www.welfare.ie/en/downloads/2013-03_ SIABudget2013_Final.pdf.

Dias, P. R. (2009). Inequality of opportunity in health: Evidence from a UK cohort study. Health Economics, 18, 1057-1074.

Duclos, J. Y., & Echevin, D. (2011). Health and income: A robust comparison of Canada and the US.

Journal of Health Economics, 30, 293-302.

Duclos, J. Y., & Makdissi, P. (2005). Sequential stochastic dominance and the robustness of poverty orderings. Review of Income and Wealth, 51, 63-87.

Duclos, J. Y., Sahn, D., & Younger, S. (2006). Robust multidimensional poverty comparisons. Economic Journal, 116, 943-968.

Ferreira, F., & Lugo, M. (2012). Multidimensional poverty analysis: Looking for a middle ground. IZA Policy Paper, No. 45.

Foster, J., Greer, J., & Thorbecke, E. (1984). A class of decomposable poverty measures. Econometrica, 52, 761-766.

Griffith, R., Leicester, A., & O Connell, M. (2013). Price based measures to reduce alcohol consumption, IFS Briefing Note BF138.

Health Insurance Authority of Ireland. (2013). Market statistics, September 2013.

Honohan, P., & Walsh, B. (2002). Catching up with the leaders: The Irish hare. Brookings Papers on Economic Activity, 1, 1-77.

Idler, E., & Benyamini, Y. (1997). Self-rated health and mortality: A review of 27 community studies. Journal of Health and Social Behaviour, 38, 21-37.

King, G., & Wand, J. (2007). Comparing incomparable survey responses: Evaluating and selecting anchoring vignettes. Political Analysis, 15, 46-66.

Lustig, N. (2011). Multidimensional indices of achievement and poverty: What do we gain and what do we lose? Journal of Economic Inequality, 9, 227-234.

Madden, D. (2010). Ordinal and cardinal measures of health inequality: An empirical comparison. Health Economics, 19, 243-250.

Madden, D. (2011). Health and income poverty in Ireland, 2003-2006. Journal of Economic Inequality, 9(1), 23-33.

Madden, D. (2013). The poverty effects of a fat-tax in Ireland. Health Economics, published inline Oct 17, 2013.

Nolan, B., Callan, T., & Maitre, B. (2012a). Country case study—Ireland. In S. Jenkins, A. Brandolini, J. Micklewright, & B. Nolan (Eds.), The great recession and the distribution of household income. Oxford: Oxford University Press.

Nolan, B., Maitre, B., Voitchovsky, S., & Whelan, C. (2012b). Inequality and poverty in boom and bust, GINI Discussion Paper.

Ravallion, M. (2010). Troubling tradeoffs in the human development index. World Bank Policy Research Working Paper 5484.

Ruhm, C. (2000). Are recessions good for your health? Quarterly Journal of Economics, 115, 617-650.

Ruhm, C. (2013). Recessions, healthy no more? NBER Working Paper 19287.

Tekin, E., McClellan, C., & Minyard, K. (2013). Health and health behaviours during the worst of times: Evidence from the great recession, NBER Working Paper 19234.

Van Doorslaer, E., & Jones, A. (2003). Inequality in self-reported health: Validation of a new approach to measurement. Journal of Health Economics, 22, 61-87.

Walsh, B. M. (2011). Well-being and economic conditions in Ireland, UCD Centre for Economic Research, Working Paper, WP11/27.

Walsh, B. M., & Walsh, D. (2011). Suicide in Ireland: The influence of alcohol and unemployment. Economic and Social Review, 42, 27-47.

Watson, D., & Williams, J. (2001). Perceptions of the quality of health care in the public and private sectors in Ireland. Dublin: The Economic and Social Research Institute.

Whelan, C., Nolan, B., & Maitre, B. (2012). Multidimensional poverty measurement in Europe: An application of the adjusted headcount approach, Geary Institute Working Paper WP/2012/11.