The Economic Journal, 109 (February), F91±F111. # Royal Economic Society 1999. Published by Blackwell Publishers, 108 Cowley Road, Oxford OX4 1JF, UK and 350 Main Street, Malden, MA 02148, USA.
POLICY IMPLICATIONS OF NEW DATA ON INCOME MOBILITY Karen Gardiner and John Hills The existence of income mobility may moderate concerns about growing inequalities, especially if income mobility has increased. However, the evidence for rising mobility is equivocal, and its extent is not enough to offset the growth of cross-sectional inequality. There is a case for greater concern for, and different policies towards, those persistently or recurrently poor than those only temporarily poor, but the data analysed here suggest that the bulk of low income observations come from the ®rst two categories. Analysis of income mobility may help understand why people's incomes follow different trajectories and how policy might affect this.
In a speech made the week after he resigned from the Government in July 1998, the former Minister for Welfare Reform, Frank Field, identi®ed understanding income dynamics as central both to the analysis of poverty and to policy-making surrounding it. Using data on snapshots of income as a way of understanding poverty was he argued, `similar to insisting in the age of digital TV that the Brownie is the most accurate means to recording the actions, say, of the contestants for the 100 metres Olympic medal' (Field, 1998). Continuing, he drew a distinction between those who do remain on a low income for very long periods of time, and those who move in and out of low income, and argued that, `policies need to match the dynamics of low incomes'. This kind of interest in dynamics can be argued to be an important feature of New Labour policy-making in its ®rst year in of®ce ± as, for instance, in its concentration on welfare-to-work measures rather than general increases in bene®t rates, but the phenomenon had not escaped its predecessors either. Politicians have for some time liked to talk about a `hand-up' rather than a `hand-out' approach to those with low incomes, or about `pathways out of poverty' as the late John Smith put it when leader of Labour in Opposition. Conservative ministers had also become interested in what newly available data were showing about income mobility. By the early 1990s a large body of evidence ± much of it based on of®cial surveys, like the Department of Social Security's Households Below Average Income (HBAI) analysis ± had accumulated showing the extent to which income inequality and relative poverty had grown in Britain between the late 1970s and start of the 1990s (DSS, 1994; Hills, 1995). For instance, the HBAI series (based on cross-sectional data from the Family Expenditure Survey) suggested that over the 1979 to 1991/2 period This paper is a revised version of one originally presented at Section F of the British Association for the Advancement of Science meeting at Leeds University in 1997. The authors are very grateful to participants in that conference, especially Tony Atkinson for comments and advice, to Stephen Machin and an anonymous referee, to the Data Archive at Essex University for access to data from the British Household Panel Survey and from the derived dataset kindly deposited by Sarah Jarvis and Stephen Jenkins, and to the Joseph Rowntree Foundation and Economic and Social Research Council for ®nancial support. The opinions expressed in the paper are, however, those of the authors alone. [ F91 ]
Copyright © 2000 All Rights Reserved
F92
THE ECONOMIC JOURNAL
[FEBRUARY
average real incomes grew by more than a third. However, those of the richest tenth were 60% higher in real terms in 1991/2 than those of their predecessors in 1979, while incomes at the bottom had barely grown at all (or had even fallen on the series giving incomes after housing costs).1 However, new information on income mobility, it was suggested ± notably by the then Secretary of State for Social Security, Peter Lilley, in a speech in Southwark Cathedral ± meant that ®gures of this kind were not so alarming: · First, analysis of newly available data from the ®rst two years of the British Household Panel Survey (BHPS) showed that there was a lot of mobility from year to year between income groups ± the poor do not stay poor long. As Mr Lilley put it, `Social mobility is considerable. Discussion about poverty is often based on the assumption that ®gures for households on low incomes describe a static group of people trapped in poverty, unable to escape and getting poorer. However this picture has been blown apart by recent studies. They show that the people in the lowest income category are not the same individuals as were in it last year, still less ®fteen years ago' (Lilley, 1996). · Second, and in stark contrast to other ®gures showing a dramatic widening in earnings dispersion, a major contributor to rising income inequality since the mid-1970s, new DSS analysis of panel data drawn from national insurance records suggested that the lowest paid had, in fact, been increasing their earnings fastest. Again, as Mr Lilley put it, `The results destroy the notion that people remain frozen in their place in the hierarchy of earnings. It challenges any contention that those on low earnings generally saw their earnings fall. Indeed, it showed that the lowest earners saw their incomes rise fastest' (Lilley, 1996). It might be added that the more ¯exible labour market and changed economy could mean that there is more mobility than in the past ± the gaps between rich and poor may have widened, but so may have movement between income groups. If so, averaging incomes over a longer period would give a slower growth in inequality than that shown in comparing cross-sections of incomes at one moment. The message of this kind of interpretation of income mobility appears to be that what is at work is a kind of `lottery model' of income determination. Each year a celestial income determination drum is twirled and, depending on the numbers which come up, each of us ends up randomly on Income Support or as a multi-millionaire. The next year we take our chances again. In this kind of world, a great deal of `poverty' will be a one-off phenomenon. If so, we may not need to worry so much about the increase in inequality. The difference between winners and losers may have increased, but as we all have the chance of being winners and losers, it does not matter much. In particular, provided 1 The most recent HBAI analysis, covering the period up to 1994/5 suggests that the pattern changed in the two years after 1992/3, with incomes for the lowest income groups growing faster than the average (DSS, 1997; Hills, 1998)
# Royal Economic Society 1999
Copyright © 2000 All Rights Reserved
1999]
F93
NEW DATA ON INCOME MOBILITY
that we all save or insure in the good times, we can cope with the bad ones unless we are particularly unlucky.2 Data on income mobility are therefore important to the policy debate in a number of ways. First, looking at panel data may give a different understanding of what has actually been happening to income distribution from that given by cross-sectional data. Second, if there is considerable income mobility, many of the effects of a widening gap between rich and poor may be smoothed out if averaged over several years. This would be of particular importance if income mobility has indeed become faster than it was when cross-sectional inequality was lower. Third, and perhaps most importantly, appropriate policy responses to particular groups may vary not just in terms of relative incomes but also in terms of the dynamic paths which their incomes are following ± or could follow. This article reviews what some of the recent evidence on income mobility in the UK has been showing on questions of this kind.
1. Are the Low Paid Catching Up? Before discussing the evidence on income mobility per se, it may be useful to touch brie¯y on the second of Mr Lilley's propositions, that panel data suggest that the low paid have been catching up. The results on which this assertion was based are illustrated in Table 1. This is drawn from Department of Social Security analysis of the Lifetime Labour Market Database (LLMDB) by Nicholls et al. (1997) (although the authors are not, of course, in any way responsible for its interpretation by others). The database consists of the earnings in ®fteen successive years of those aged 25±44 at the start of the period (to avoid the effects of periods in full-time education and of retirement). The table compares the growth in earnings between 1978±9 and
Table 1 Changes in Median Earnings by 1978±79 Earnings Quintile Group All aged 25±44 1978±9 earnings group (®fths)
Median earnings £ p.a. 1978±9 (Sept 1995 prices)
1992±3 (Sept 1995 prices)
% change (Real)
9,300 11,900 13,900 16,200 20,900
13,200 15,700 17,800 21,000 27,900
42 32 28 30 33
1 (lowest) 2 3 4 5 (highest)
Note: Median earnings in 1992±3 are calculated for those who were in the relevant quintile group in 1978±9 and were in class 1 employment in 1992±3. Source: Nicholls et al. (1997), Table 5. 2 If people are risk averse, or if insurance itself has a cost beyond its actuarial value, there are of course costs to income instability.
# Royal Economic Society 1999
Copyright © 2000 All Rights Reserved
F94
THE ECONOMIC JOURNAL
[FEBRUARY
1992±3 of those with earnings in both years, arranged by initial earnings quintile groups. It shows that those who started in the bottom ®fth had the fastest growth of incomes, 42%, while the other groups had increases of between 28 and 33%. At ®rst sight, this might seem to contradict, for instance, Gosling et al.'s (1994) analysis of hourly male wages from the Family Expenditure Survey, showing that between 1978 and 1992, the bottom decile (the cut-off for the bottom tenth) hardly changed in real terms, while the top decile grew by 50%, or similar ®ndings of widening wage dispersion from the New Earnings Survey (Hills, 1995, Fig. 24). It is not that there is anything wrong with the LLMDB ®gures in themselves, but the interpretation that they imply that the lowest paid were `catching up' in any meaningful way is surprising. There are three problems. First, the data only relate to those with earnings in both years. But a substantial proportion of the lowest paid in the ®rst year do not have earnings in the ®nal year. For instance, of those aged 25 to 34 at the start, only 23% of the original lowest paid ®fth end up in the bottom ®fth. However, 28% of them end up in the `credits' category, for instance being unemployed. Looking at the older, 35 to 44 year-old, group, 41% of the lowest paid ®fth end up in the `credits' category. The 42% increase in earnings for those ending up in work leaves out the large numbers at the bottom who drop out altogether. Furthermore, we know that those who start at the bottom are more likely to have dropped out than those who initially had earnings higher up the distribution. Second, there is a problem resulting from both measurement error and short-term variability in earnings. Results of the kind shown in Table 1 are vulnerable to the `regression fallacy' (Friedman, 1992). Where incomes are either subject to measurement error or have a transitory component, division by starting income (rather than, say, average income over the period) biases the results. The individual observations initially found at the bottom will disproportionately tend to be those with negative measurement errors or negative transitory components of income. This produces more apparent convergence in later observations than there is in the true underlying earnings relativities (see Atkinson et al., 1983, Fig. 5.2). Third, there is an important age effect which is apparent from the same analysis of the LLMDB. This shows the way in which earnings tend to rise with age, reaching a relative maximum when workers are in their early forties, and then tend to decline (Ball and Marland, 1996, Chart 11). Given the particular age group in the LLMDB, the lowest paid at the start will tend to have been the youngest, and this cohort will have reached its relative peak of earnings at the end of the period. By contrast, the highest paid group at the start will have included many of the older workers in their forties, and their earnings will have tended to decline in relative terms as a result of the age effect. This is, of course, important and does suggest that some low incomes are a life-cycle phenomenon which people will `grow out of'. Other low incomes are also a life-cycle phenomenon, for instance, declining incomes for the elderly, and these will not be grown out of. Rowntree (1902) told us about this nearly # Royal Economic Society 1999
Copyright © 2000 All Rights Reserved
1999]
F95
NEW DATA ON INCOME MOBILITY
one hundred years ago, and it remains important. However, using this phenomenon to imply that the low paid in general are in some sense catching up on the rest of us ± so that low pay is not a problem ± does not seem helpful as a way of understanding the changing labour market.
2. How Much Mobility is There? The most familiar way of looking at mobility is perhaps a transition matrix showing what percentage of, say, each tenth of the population end up in each income group a year later. If there was complete immobility, all the positive entries would be in the leading diagonal and would equal 100. By contrast, if movements from year to year are random ± as in a `lottery' model ± then all of the entries in a ten by ten matrix will equal 10. The less mobility there is, the greater the numbers on the leading diagonal will be. How the actual results compare with these extreme cases gives an idea of how much mobility there is. Whether you think it represents `a lot' or `a little' depends to some extent on prior expectations. For instance, Table 2 shows Jarvis and Jenkins' (1997a) results comparing the income group of individuals between the ®rst and second waves of BHPS. 36% of the sample are on the leading diagonal (in bold), compared to 10% in the random model or 100% with complete immobility. 71% are on the leading diagonal or the two neighbouring diagonals. In other words, nearly threequarters of the sample are either in the same income group or a neighbouring
Table 2 Transitions between 1991 and 1992 Income Groups Income decile group, wave 1
Income decile group, wave 2 (% of initial group) 1
2
3
4
5
6
7
8
9
10
All
1 2 3 4 5 6 7 8 9 10
46 23 12 7 2 3 3 2 4 2
21 39 19 9 4 5 1 1 2 1
15 20 28 19 11 5 2 2 2 1
5 11 22 27 15 10 4 2 2 1
4 4 8 20 30 17 11 2 2 1
2 1 3 9 22 25 20 11 6 2
3 1 3 5 7 18 36 19 8 3
1 1 2 2 5 10 14 34 23 7
0 0 2 0 2 5 6 17 41 24
2 1 1 2 1 2 3 6 13 58
100 100 100 100 100 100 100 100 100 100
All
10
10
10
10
10
10
10
10
10
10
100
Note: Group 1 contains the poorest tenth: group 10 the richest tenth. Income equivalent net household income (£ p. w., January 1991 prices), distributed amongst individuals. Source: Jarvis and Jenkins (1997a). # Royal Economic Society 1999
Copyright © 2000 All Rights Reserved
F96
THE ECONOMIC JOURNAL
[FEBRUARY
one a year later. In the random model only 28% would be in this position. As Jarvis and Jenkins (1997b) show more generally, while there is considerable income mobility, most of it is short-range. In particular, for the poorest tenth only 46% are also in the poorest tenth a year later. In that sense more than half of them `escape'. However, 67% of them remain either in the poorest or next-poorest tenth. Only a third move further than this. For the poorest ®fth as a whole, 65% are still there one year later and 85% are either in the poorest or next-poorest ®fth. Taking these ®gures together they suggest that around one third of low income is in some sense transient, but two-thirds is not. Even so, repeating this process often enough could mean that the income groups will become completely mixed up. After all, if 54% of the poorest tenth leave it each year, then with repeated attrition less than 10% would be left by Year 4. However, there are two problems: · First, some people come back again, even if the process is random. In fact, low-income escapers are more likely to drop back into the poorest tenth than those who started with higher incomes. · Second, the escape rates of those who stay at the bottom for more than one period seem to decline. Either people get stuck and ®nd it increasingly hard to escape, or there are two different populations: `bouncers' and `stickers' (that is, there is either state dependency or heterogeneity). Jarvis and Jenkins' (1997b) results for the transition between Waves 1 and 4 show the combined effect of these two processes ± that 37% of the poorest tenth in Wave 1 are also in the poorest tenth in Wave 4, compared to the 9% that would be expected if there was simply repeated attrition with more than half leaving each year. DSS analysis of BHPS ®nds that 36% of the poorest tenth in Year 1 are also in the poorest tenth in Year 5 (DSS, 1997, Table 7.4), compared to under 5% in a repeated attrition model. Looking at the poorest ®fth in Wave 1, only a third are in the poorest ®fth in all of the next three years, but including returners 54% end up in the poorest ®fth in Wave 4 (and 53% in the poorest ®fth in Wave 5). That only a minority of the poor are continuously so does not stop most of them being repeatedly poor. What such results tell us is that low income is not a random phenomenon. Low income observations are linked. From Jarvis and Jenkins' analysis of the ®rst four waves of BHPS, one can plot how often different individuals (with income data for all four years) are found in the poorest ®fth. Table 3 compares these ®ndings with what one would expect from a `lottery model' over four years. The BHPS data are very different from those generated at random. In the BHPS 64% of individual cases never enter the poorest ®fth (compared to 41% if the process were random), and 14% of individuals are in the poorest ®fth three or four times (compared to only 2.7% at random). In terms of the proportion of low income observations (being in the poorest ®fth in any given year), 61% of these observations are accounted for by the individuals who are in the poorest ®fth three or four times, compared to only 10.4% in the random model. In other words, repeated low income accounts # Royal Economic Society 1999
Copyright © 2000 All Rights Reserved
1999]
NEW DATA ON INCOME MOBILITY
F97
Table 3 Repeated Low Income Observations: `Lottery Model' and BHPS data Number of low income (poorest ®fth) observations
Lottery model (4 years)
BHPS data (Jarvis and Jenkins)
(a) Percentage of individuals with each number None 41 1 41 2 15 3 2.5 4 0.2 (b) Percentage of low income observations accounted for None ± 1 51 2 36 3 9.6 4 0.8
64 13 9 7 7 ± 17 22 26 35
Source: Own calculations and Jarvis and Jenkins (1997b), Table 1.
for a far greater proportion of the low income seen in cross-sections than a lottery model would suggest. Looking at it in another way, Jenkins (1998) shows that the Gini coef®cient of income inequality averaged over observations in six years of BHPS is about 88% of the coef®cient for single-year cross-sectional inequality. Averaging over six years like this reduces the coef®cient by about 3 percentage points. By comparison the growth in the equivalent index for cross-sectional inequality between 1977 and 1993 was about 10 percentage points (Goodman et al., 1997). Even if increased mobility did mean that averaging over several years had more of an effect in the 1990s than the 1970s, this could clearly only offset a small part of the growth in cross-sectional inequality. It clearly is true that those with low income in Britain do not make up an unchanging `underclass'. Equally, whether one is poor in one year greatly increases the chance of being poor in later ones: the existence of mobility does not mean that income inequalities are ironed out if one waits long enough.
3. Is Mobility Increasing? Unfortunately, there is no equivalent of the BHPS ± which started at the beginning of the 1990s ± for earlier periods, with which one could compare the recent pattern of mobility to give direct evidence on whether mobility has indeed risen during the period through which cross-sectional inequality increased. There are, however, three sources of indirect evidence, the ®rst two on components of income, rather than income itself. 3.1. Earnings Richard Dickens (1997) of the LSE's Centre for Economic Performance has looked at earnings mobility, using panel data from the same source (the annual # Royal Economic Society 1999
Copyright © 2000 All Rights Reserved
F98
THE ECONOMIC JOURNAL
[FEBRUARY
New Earnings Survey) as the DSS analysis discussed above. He constructs a mobility index for year-to-year movements in the earnings distribution (abstracting from age effects) over the period 1974±94. The results show that there has been a very clear decline in earnings mobility over the period. However, they also show that the mobility index is correlated with in¯ation. This has also declined over the period. One might expect such a correlation given that the pattern of annual settlements could lead to much greater variation in people's position in the earnings distribution from observation to observation at times of rapid in¯ation. If one observation was just after a settlement, but the next shortly before the next one, the comparison might suggest a signi®cant decline in relative earnings, even though this was about to be corrected. With the lower in¯ation rates of the late 1980s and 1990s than in the 1970s, the effect would be smaller in more recent years and this could affect the trend in the value of the index over time. Allowing for this, Dickens still ®nds that the mobility index has fallen by 22% for men, and by 11% for women. There is certainly no evidence of an increase in mobility. He reaches the overall conclusion that, `The low paid are worse off both in terms of the relative wage they receive, and in terms of the opportunity to progress out of the low pay trap' (Dickens, 1997).
3.2. Bene®ts Earnings are, however, only a part of the picture of how incomes as a whole have developed. Many of the poorest groups do not have earnings. In fact, most of the poorest depend on bene®ts like Income Support. If mobility at the bottom has been increasing, one might expect to see people receiving Income Support for shorter periods. Administrative data from the DSS's Social Security Statistics show that the numbers of Income Support recipients who have been receiving it for over two years increased substantially from 1.9 million in 1979 to 3.3 million in 1995. On the other hand, there are also many more short-term IS recipients as well, so long-term cases declined as a proportion of all IS recipients, from 65% in 1971 to 58% in 1995.3 There are thus many more people receiving Income Support ± almost by de®nition amongst the poorest ± than before, but a slight decline in the percentage who are long-term. A rise in the absolute number of non-pensioner long-term IS cases suggests falling mobility at the bottom. However, fewer pensioners are amongst the poorest now, with the opposite effect. The overall result of these effects is unclear, but again there is little evidence for a rise in mobility at the bottom.
3 The bene®t system has changed considerable over the period, but not in ways which make this comparison invalid. Features like compulsory `Restart' interviews for those unemployed over six months may have led to some increase in people cycling between short-term bene®t receipt and marginal employment, rather than becoming a `long-term' claimant.
# Royal Economic Society 1999
Copyright © 2000 All Rights Reserved
1999]
NEW DATA ON INCOME MOBILITY
F99
3.3. Incomes While there is no 1970s equivalent of the BHPS, one earlier panel survey does give limited evidence on some income transitions over the period from 1978 to 1979. This is the combination of the then Department of Health and Social Security's Family Finances Survey and its follow-up Family Resources Survey (not to be confused with the current survey of the same name). The results from these were analysed by Hancock (1985). The surveys looked at families with children only, focussing on those with `Relative Net Resources' (essentially income after housing costs as a percentage of each family's Supplementary Bene®t scale rate) of under 140% in the ®rst year. These represented the poorest 15% of all families with children. The ®rst two columns of Table 4 compare Hancock's ®ndings ± which can be expressed as the rates of mobility out of the poorest 5, 10, and 15% ± with Jarvis and Jenkins' (1996) ®ndings for one year transitions out of the poorest 10 and 20% between the ®rst four waves of BHPS. The two sets of results show the importance of the size of the group from which escape is being measured: the smaller the group, the more likely is escape within a year (partly re¯ecting measurement error and short-term variability as discussed above). This comparison does suggest that mobility was greater in the early 1990s than the late 1970s, for instance with only 43% of the poorest tenth escaping after a year in the earlier surveys, but between 49 and 55% escaping in the later survey (depending on which waves are examined). However, the results are not straightforwardly comparable. The earlier results are in terms of numbers of families (and then only those with children), the later ones in terms of numbers of individuals. If one counts only families with children within the BHPS, the escape rate from each group is somewhat faster. The third column shows, for instance, that the escape rate from the poorest tenth is between 52 and 60% (depending on the years chosen, being faster between the earlier waves), signi®cantly more than the 43% in the earlier
Table 4 Has Income Mobility Risen? 1978±9 (families with dependent children, income after housing costs) % leaving income group:
Poorest 5%: 58 Poorest 10%: 43 Poorest 15%: 31
Early 1990s (all individuals, one year transitions in BHPS) Poorest 10%: 49±55 Poorest 20%: 35±39
Early 1990s (families with dependent children, one year transition in BHPS) Poorest 5%: 62±75 Poorest 10%: 52±60 Poorest 15%: 45±51
Sources: First column is from Hancock (1985) Table 2.2 reanalysed (transitions between Family Finances Survey and Family Resources Survey in terms of Relative Net Resources, with cut-offs of 100%, 120% and 140% relative to SB scales); second column is from Jarvis and Jenkins (1996), Table 3 (transitions between ®rst 4 waves of BHPS); third column is from reanalysis of the dataset based on BHPS derived by Jarvis and Jenkins. # Royal Economic Society 1999
Copyright © 2000 All Rights Reserved
F100
THE ECONOMIC JOURNAL
[FEBRUARY
survey. This does suggest a rise in mobility at the bottom for this particular group, families with dependent children, although the increase in mobility does not appear to be of the scale required to offset the kind of growth in cross-sectional inequality seen in the 1980s. Caution is also needed in interpreting the comparison, given the differences between the methods used in the surveys and indications that mobility varies over the economic cycle (which was in a boom in the late 1970s, but in recession in the early 1990s).4 The rise in cross-sectional income inequality over the period means that differences between income quantiles are greater, and so the mobility shown in the BHPS ®gures corresponds to greater absolute movements than in the earlier comparison. Some increase in short-term income mobility would be consistent with Blundell and Preston's (1998) ®nding from analysis of income and consumption patterns in the Family Expenditure Survey between 1968 and 1992 that transitory income variance has increased, particularly in the late 1980s.
4. Examining Transitions Given that some people do escape from periods of low income, but also that income movements do not happen at random, it would be helpful for policymaking to understand why different people follow different trajectories, and then perhaps what factors or policies might help some onto a more positive trajectory. To do that we need to understand the data in a more complex way than can be derived from two-dimensional transition matrices, useful as they are: · Are those who leave poverty really escaping, or do they soon drop back again? · Do those who escape go far? · How much of the movement observed is accounted for by life-cycle changes ± for instance, young workers whose pay rises rapidly with experience? · Are the movements over time chaotic, or do they follow recognisable patterns? · Are poverty and low income observations simply one-off `blips' (maybe measurement error or short-term unemployment), aberrations in otherwise stable patterns of higher income? · Do the observed movements re¯ect large changes, or might they just be `wobble', with income just changing enough to take people across some dividing line between one income group and another, but not representing any substantial change? In other words, it is important to distinguish the genuine `movers' from those who are simply `shakers'. Analysis by both the Institute for Fiscal Studies and the ESRC Research Centre on Micro-Social Change at Essex University has begun to give some information about the variety of trajectories which people follow. Fig. 1 shows 4 Schluter (1998), p. 26, presents analysis suggesting that mobility fell within the ®rst four years of the BHPS, that is during the ®rst half of the 1990s. This could re¯ect the state of the economic cycle.
# Royal Economic Society 1999
Copyright © 2000 All Rights Reserved
1999]
‘High’ income (top 80%)
F101
NEW DATA ON INCOME MOBILITY
Wave 1
Wave 2
Wave 3
8,668
7,819
7,256
618 417 377
849
‘Low’ income (bottom 20%)
2,167
849
563
1,318
941
231 432
Fig. 1. Trajectories in First Three Waves of BHPS (IFS analysis) Source : Goodman et al. (1997), Figs. 9.2 and 9.3.
the pattern of movement between the ®rst three waves of BHPS revealed by Goodman et al.'s (1997) analysis. The ®gure shows how people move between `low income' (poorest 20%) and `high income' (the other 80%) groups. This generates eight possible trajectories, each followed by at least 2% of the sample (the boxes are scaled in proportion to the numbers following each trajectory). The pattern is already becoming hard to follow, but it can be seen that those who escape from the low income group between Waves 1 and 2 have a greater propensity (more than a quarter) to drop back into it in Wave 3 than those with `high' income in the ®rst two waves (only 7% of them do so). Also, those who stay with low income in the ®rst two waves have a lower escape rate in Wave 3 (29%) than the original low income group in Wave 1 (39% of them were outside it in Wave 2). Fig. 2 shows the more complex pattern of transitions between the ®rst four waves of BHPS given by Jarvis and Jenkins (1997b). Again, this looks at movements in and out of the bottom ®fth. There are now sixteen possible trajectories, so the ®gure separates out those starting in the bottom ®fth in the lower panel from those starting outside it. Again, each trajectory is followed by at least 0.9% of the sample. The picture con®rms the declining escape rates for those who remain at the bottom: 39% of the Wave 1 poorest escape by # Royal Economic Society 1999
Copyright © 2000 All Rights Reserved
F102
THE ECONOMIC JOURNAL
[FEBRUARY
Wave 1
Wave 2
Wave 3
Wave 4
80
72.3
67.5
63.9
(a) Starting in top 80%
Top 80%
2.9 2.6 1.5
3.8
Bottom 20%
7.8
3.6 0.9 2.2 2.5
4.8 4.0
(b) Starting in bottom 20% Top 80%
5.4
4.2 2.2 1.0 1.9
7.8 3.4
Bottom 20%
20
1.2 1.2 1.4
2.4 12.3
8.9
7.0
Fig. 2. Trajectories in First Four Waves of BHPS (Jarvis and Jenkins analysis) Source : Jarvis and Jenkins (1997b)
Wave 2, but only 21% of those in the poorest ®fth in all three of the ®rst three waves are outside it in Wave 4. Despite the complexity of the patterns shown, there are two limitations to this kind of analysis as a way of capturing the movement which is occurring. First, the `movements' tell us simply that people have moved across a particular threshold. We have no way of telling whether these are large or small movements. Second, the diagrams give little impression of what is happening within the upper part of the distribution, which may also be of interest ± if only so that one could tell whether patterns of mobility lower down the distribution were similar to those higher up. To try to get round this problem, we have reanalysed the dataset of # Royal Economic Society 1999
Copyright © 2000 All Rights Reserved
1999]
NEW DATA ON INCOME MOBILITY
F103
equivalised incomes5 over four waves of BHPS created by Jarvis and Jenkins.6 The observations are divided into groups in each wave. In the ®rst wave, these are simply percentile groups, with a hundredth of the (weighted) sample in each group. In subsequent waves the boundaries are taken from the initial percentiles, increased in line with average income growth, so we refer to them as quasi-percentiles. If the overall shape of the distribution remained unchanged, there would still be a hundredth of the sample in each group in later waves. On the other hand, if, for instance, those with low incomes had above average income growth, relative poverty would fall and so would the number in the bottom groups. This seems preferable in principle to simply taking actual percentile groups in each subsequent wave, although it does not in fact make much difference over this particular period.7 An individual could therefore be in any of a hundred groups in each of four waves (giving a total of 100 million possible combinations). We then divide the sample cases between ®ve different broad kinds of trajectory which these combinations might represent, as illustrated in Fig. 3, ignoring the effects of small variations in someone's position in the distribution between one year and another: 1. Flat trajectories, where the individual spends the four periods in the same income group or its near neighbours. This means that a small `wobble' would not prevent someone's trajectory being counted as ¯at. Within this category, individuals are classi®ed as poor ¯at if two or more observations are within the bottom 20 quasi-percentile groups. 2. Rising trajectories, where the individual moves a signi®cant amount across the distribution, and all movements from wave to wave are either upwards or ¯at. Those starting in the bottom 20 groups would be rising out of poverty. 3. Falling trajectories, where the individual moves a signi®cant amount, and all movements are downwards or ¯at. Those ending in the bottom 20 groups would be falling into poverty. 4. `Blips', where the basic trajectory would be de®ned as ¯at (within nearby groups for three out of the four periods), except that one observation is further away (excluding those already de®ned as `rising' or `falling'). This group includes blips out of poverty (where the ¯at part of the trajectory is in the bottom 20 groups, or at least is so for two out of the three observations), blips 5 The income variable created by Jarvis and Jenkins from the BHPS is comparable with the Department of Social Security's HBAI `Before Housing Costs' net income measure. For more detail on the income de®nition, see, for example, Jarvis and Jenkins (1997b). 6 The analysis presented here is re®ned from that in an earlier presentation in Hills (1998) in that, for reasons explained below, trajectories are de®ned in relation to percentiles of the original distribution rather than deciles. 7 An alternative would be to take the percentiles of the initial wave of data, and adjust them for in¯ation only. This would give a measure of mobility against absolute income standards, as opposed to the movements against a relative income standard shown here.
# Royal Economic Society 1999
Copyright © 2000 All Rights Reserved
F104
THE ECONOMIC JOURNAL
1. Flat Remaining within same income group or its near neighbours
2. Rising Crossing more than a minimum number of income group boundaries (upwards or flat between waves)
3. Falling Crossing more than a minimum number of income group boundaries (downwards or flat between waves)
4. Blips ‘Flat’ except one wave
[FEBRUARY 5. Other Other cases
Fig. 3. Trajectory-type Classi®cations
into poverty (where the `blip' observation is in the bottom 20 groups and the others are higher).8 5. Other trajectories, covering all possibilities not included in the four types described above. These sub-divide into trajectories with repeated poverty (two or more observations in the bottom groups), one-off poverty (one observation in the bottom two groups), and non-poor cases. As will be apparent from this description, how individual cases are allocated between groups will depend on just how wide a band of movement is allowed for a trajectory still to be classi®ed as `¯at'. If only a narrow band across the distribution is allowed, few cases will be `¯at' ± even a small movement would be enough to rule this out. On the other hand, a very wide range of allowable movement would lead to many more trajectories being counted as `¯at'. In an earlier presentation of these ®ndings (Hills, 1998, Figure 24) we divided the incomes up more crudely, into ten groups, and classi®ed trajectories as ¯at unless they crossed more than one boundary (quasi-decile). The effects of doing this are summarised in the left-hand column of Table 5 ± 41% of cases are allocated to the ¯at category. On average this seems a reasonable criterion 8 Where only a narrow band of movement is allowed for those categorised as `¯at', it is possible that a trajectory could be categorised as a `blip' , but all the observations counted as poor. Strictly speaking this would be a `blip within poverty'. In the results presented in Tables 6±8 we have included a very small number of trajectories which turn out to be of this kind within the `blip out of poverty' group, re¯ecting the way in which these cases are also generally poor and hence problematic.
# Royal Economic Society 1999
Copyright © 2000 All Rights Reserved
1999]
F105
NEW DATA ON INCOME MOBILITY
Table 5 Trajectory Classi®cation and Criteria for Assessing Movement Percentage of cases assessed as:
Two quasideciles
Flat Rising Falling Blip Other All
40.9 10.3 8.3 23.3 17.3 100
Trajectory assessed as `¯at' unless it crosses: Ten Fifteen Twenty quasiquasiquasipercentiles percentiles percentiles 27.5 7.7 5.1 29.1 30.5 100
41.2 6.1 4.0 29.8 19.0 100
53.2 4.8 3.1 27.0 12.0 100
Note: 1. Results based on weighted results from ®rst four waves of BHPS. 2. Quantiles for classi®cation of trajectories taken from Wave 1 distribution, uprated with average income growth in later waves.
± if people move beyond the neighbouring tenth of the distribution, their incomes are not `¯at'. However, the ¯at group could include those who had started near the bottom of one tenth and moved almost to the top of the next, while others who had, in fact, crossed less of the distribution would be allocated to one of the other more mobile categories. To avoid this inconsistency, the results here are based on the ®ner quasi-percentiles described above. The other three columns of the table show the effects of restricting in different ways the number of quasi-percentiles which can be crossed by those in the ¯at group. The middle column ± allowing less than ®fteen to be crossed ± is roughly equivalent to the earlier rule (as ®fteen percentiles is the average movement which would still qualify as `¯at' when using decile groups). This gives 41% of cases categorised as ¯at, and it is results based on this criterion which we use below. It should immediately be recognised that even at random, a proportion of cases will fall into the ®rst four trajectory groups. If there are enough cases, some of them will apparently be following a consistent pattern over four periods, even if this has only been generated by something like the lottery model. Table 6 therefore compares the results which we obtain from the four waves of BHPS data with the pattern which would be seen in a hypothetical dataset containing 10,000 cases, with incomes distributed at random between the different income groups in each of the four waves. Within this dataset, 73% of cases would fall into one of the chaotic `other' categories. One might take low income observations which result from the `rising out of poverty', `blips into poverty', and `other one-off poverty' trajectory types as being less problematic than the others ± they would be consistent with either transitory low income (one period out of four) or low income which the individual had apparently escaped from by the fourth wave. Other low income observations are either from trajectories where the individual is consistently # Royal Economic Society 1999
Copyright © 2000 All Rights Reserved
F106
[FEBRUARY
THE ECONOMIC JOURNAL
Table 6 Four Wave Trajectories by Type 15 percentile criterion for movement Random dataset (10,000 cases)
Trajectory type
% of cases
1. Flat:
Poor Non-poor 2. Rising: Out of poverty Non-poor 3. Falling: Into poverty Non-poor 4. Blips: Out of poverty Into poverty Non-poor 5. Other: Repeated poverty One-off poverty Non-poor By low income observations None One Two Three Four
First four waves of BHPS
% of low income observations % of cases
% of low income observations
0.3 1.1 2.6 2.0 2.5 1.8 2.6 4.5 10.1 13.6 32.6 26.3
1.1 0.1 4.4 ± 4.4 ± 8.9 5.8 ± 34.5 40.9 ±
8.7 32.5 2.6 3.5 1.5 2.5 5.6 5.1 19.2 5.2 5.8 8.0
40.5 1.6 5.4 ± 3.6 ± 20.6 7.1 ± 13.7 7.5 ±
41.0 41.0 15.4 2.6 0.2
± 51.2 38.4 9.6 0.8
64.4 13.7 8.7 6.5 6.7
± 17.7 22.6 25.2 34.5
Includes some cases where all observations are in poorest 20 quasi-percentile groups. Source : First four waves of BHPS.
poor or only moves temporarily out of low income (spending at least two out of four periods with low income), or where the individual's income appears to be on a downward trajectory ending up in the poorest ®fth.9 Within the hypothetical dataset, more than half of low income observations would be generated by the less problematic trajectories. The third and fourth columns of the table show the patterns we actually found in the data. Rather than 73% of cases (and 75% of poverty observations) coming from the `other' groups, these accounted for only 19% of cases (and 21% of poverty observations). The four other categories capture the majority ± four-®fths - of what is going on: the income movements are not as chaotic as might have been thought. Overall, over 70% of the BHPS cases are within the `¯at' or `blip' categories. The trajectories are, if anything, ¯atter than one might have expected from single year transitions of the kind seen in Table 2. Of the poverty observations in the actual dataset, 41% come from the `poor ¯at' group, 21% from the `blip out of poverty' group, 4% from the `falling into poverty' group, and 14% from other cases with repeated poverty. Less than a quarter of poverty observations come from the less problematic trajectories. 9 Of course, with a larger number of waves of data, observed trajectories may be found to be part of more complicated sequences.
# Royal Economic Society 1999
Copyright © 2000 All Rights Reserved
1999]
NEW DATA ON INCOME MOBILITY
F107
Those rising out of poverty account for 5% of the observations, those `blipping into poverty' for only 7%, and other one-off cases for 8%. In some sense, therefore, nearly 80% of poverty observed at any one time still represents a `problem' case, despite the dynamics. At the same time, any cross-section will show some people whose general trajectories are, in fact, unfavourable who are not in poverty at that moment. For instance, in the BHPS dataset in any single year more than 1% of people would be in the `out' part of a `blip out of poverty' trajectory, and more than 2% would be in one of the `out' years of a trajectory involving repeated poverty. This adds more than a tenth to the numbers of concern. Allowing for these cases, the size of the `poverty problem' is therefore more than 90% of the amount observed at any one time. Understanding the dynamics helps us distinguish between the kinds of problem people are affected by; it does not do very much, if anything, to suggest that the problem is smaller than that seen in a snapshot. For policy-targeting, it would be very useful to know more about the characteristics of those found in the different groups. For instance, how do they vary in terms of age, education, family circumstances, or the kind of neighbourhood in which they live? What events are associated with people taking different trajectories? We already know a great deal about the characteristics of those with low incomes at any one time, but it would help in designing intervention if we were able to differentiate between those who start in similar circumstances, but whose lives go in different directions. Tables 7 and 8 present some initial information on this. The ®rst sets the context, by showing the proportion of low income observations in the ®rst four waves of BHPS accounted for by groups with different characteristics in the ®rst wave (as before, taking a low income cut-off of the bottom quintile in Wave 1 uprated by average income growth). The pattern is a familiar one. Groups disproportionately affected by low income are: females; single pensioners; lone parents and their children; all the economic groups without a full-time worker in the household, especially the unemployed and `others' (including long-term sick and disabled and lone parents); social tenants; and those who are non-white (although response rates to this question in BHPS are very poor). Within these, those starting the four years as lone parents, with an unemployed head or spouse (and no earner), and as council tenants are more than twice as likely as the average to have low incomes on this de®nition. Table 8 takes this further, showing the distribution of those with each initial characteristic between four groups of the trajectory types shown in Table 6. Within each breakdown the categories have been ranked by prevalence of `poor ¯at' cases. Because of the poor response in the BHPS we have omitted the breakdown by ethnic group from this table. Of the weighted sample as a whole, 9% fall into our `poor ¯at' group (using the 15 percentile criterion for movement) and a further 12% into one of the other `problematic' trajectories involving either only temporary escape from poverty, or where someone is falling into poverty. The other 79% of cases either never fall below the low income cut-off, do so only temporarily, or appear to have escaped from it by the last wave. # Royal Economic Society 1999
Copyright © 2000 All Rights Reserved
F108
[FEBRUARY
THE ECONOMIC JOURNAL
Table 7 Low Income Observations and Initial Characteristics
Gender Family type
Economic Status
Housing tenure
Ethnic group
Female Male Single pensioner Pensioner couple Couple with children Couple, no children Single with children Single Self-employed Full-time worker Part-time worker only Head=spouse 60 Head=spouse unemployed Other Outright owner Mortgagor Council tenant Housing association tenant Private tenant White Non-white
% of cases
% of low observations
52.3 47.7 10.2 9.3 39.6 21.4 6.6 13.1 11.4 50.8 6.7 19.2 5.4 6.5 19.3 50.4 19.8 3.3 6.2 97.4 2.6
57.5 42.5 18.2 10.0 37.5 9.4 16.4 8.4 11.5 17.7 10.6 29.7 14.7 15.7 20.5 25.9 39.2 5.6 8.1 95.8 4.2
Note: Low income is below Wave 1 bottom quintile uprated by average income growth in later waves. 24% of sample have this value classed as `missing'. Source: First four waves of BHPS
Looking at the different categories in this way some interesting differences emerge. More than half of lone parents and their children fall into the `poor ¯at', or `problematic poor' groups, but only 14% into the groups affected by low income, but in a less problematic way. By contrast, single pensioners are also more than twice as likely than the average to be in the poor ¯at group, but otherwise tend more to be in the less problematic groups affected by low income than the more problematic ones. While Table 7 showed that pensioner couples do not have poverty rates much higher than the average, this analysis shows that they are more likely to fall into the `poor ¯at' group: if they are poor at all, it is most likely to be persistent poverty. By contrast, single people and couples without children are not only less likely to be affected by low income than others, but if they are affected at all they tend to have less problematic trajectories. Looked at by economic category, those starting with a head or spouse unemployed (and no-one in paid work) overwhelmingly have unfavourable trajectories over the four years as a whole, as do those in the `other' category. More than this, they are particularly concentrated in the `poor ¯at' and `problematic poor' trajectories. By contrast, even when someone from one of the more favoured economic groups ± starting with someone in full-time work or self-employment ± is affected by poverty, their trajectory is most likely to fall into one of the less problematic groups. For those with no earner and a head # Royal Economic Society 1999
Copyright © 2000 All Rights Reserved
1999]
F109
NEW DATA ON INCOME MOBILITY
Table 8 Trajectory Types and Initial Characteristics 15 percentile criterion for movement % of cases
Gender Family type
Economic status
Housing tenure
All Female Male Single with children Single pensioner Pensioner couple Couple with children Single Couple, no children Head=Spouse unemployed Other Head/spouse 60 Part-time worker Self-employed Full-time worker Council tenant Housing association tenant Outright owner Private tentant Mortgagor
Poor ¯at
Problematic poor
Less problematic poory
Nonpoor
8.7 10.2 7.1 27.1 21.0 13.3 6.9 2.9 2.3 30.0 23.5 18.8 12.2 4.5 1.3 23.0 14.9
12.3 13.0 11.2 26.8 14.0 9.5 14.0 9.2 6.2 26.8 27.9 12.2 19.3 16.4 6.6 19.1 17.3
13.5 13.3 13.7 13.6 21.4 12.3 12.3 16.0 10.9 23.4 15.4 18.5 23.8 19.3 7.6 17.6 24.7
65.7 63.3 68.2 32.2 43.5 65.0 66.9 71.9 80.7 19.9 33.2 50.7 44.8 60.0 84.5 40.2 43.1
10.6 7.4 2.2
12.3 17.9 8.6
12.6 18.1 10.7
64.6 56.8 78.6
Note: Falling into poverty, blip out of poverty and other repeated poverty. y Rising out of poverty, blip into poverty and other one-off poverty. Source: First four waves of BHPS.
or spouse aged 60 or more, there is something of a divide, with nearly 20% in the `poor ¯at' and `less problematic poor' groups, but about average in the `problematic group'. Finally, by tenure, mortgagors are not only unlikely to be affected by low income at all, but if they are, it is most likely to form part of a less problematic trajectory. By contrast, social tenants of either kind are much more likely than others to be affected by low income at all, but it is council tenants who are most likely to be in the `poor ¯at' group. The previous table suggested that council tenants were twice as likely as the average to have a low income at any one time; this analysis suggests that they are almost three times as likely to fall into the `poor ¯at' group. Prospects for housing association tenants look slightly brighter.
5. Conclusion Returning to the three ways identi®ed in the introduction in which a better understanding of income mobility might be relevant to the policy debate, panel # Royal Economic Society 1999
Copyright © 2000 All Rights Reserved
F110
THE ECONOMIC JOURNAL
[FEBRUARY
data surveyed here do not contradict the impression of rapidly growing inequality in both wages and household incomes in the United Kingdom between the late 1970s and early 1990s. Nor is there much in the data surveyed here which suggests that allowing for income mobility reduces the scale of the problem represented by the associated rise in relative poverty. The evidence is equivocal, but even if income mobility is now greater than it was at the end of the 1970s (as one of the comparisons presented here suggests), albeit for a restricted group, the size of the increase is not enough to do any more than moderate the effects of the rise in cross-sectional inequality. Similarly, only a small fraction of observations of low income are accounted for by people whose overall income trajectory over a four-year period is relatively unproblematic. When thinking about the scale of the `poverty problem' it is as important to remember the `temporarily not poor' as it is to put less weight on the temporarily poor. However, on the third issue this area of analysis has much to offer. It is important to distinguish between at least three groups: the persistently poor, the recurrently poor and the temporarily poor. For the last of these, ensuring that people are covered by social or private insurance and have adequate savings to `tide them over' bad times may be enough. This kind of policy may be most relevant to relatively favoured groups like mortgagors and those starting with full-time workers in a household. However, most low income is not of this kind. Much current government policy ± particularly approaches like the New Deal ± is aimed at getting people who are currently out of work into work. The extent of recurrent poverty ± our problematic trajectories ± suggests that policy needs to pay attention not just to the ®rst transition, off bene®t and into work. It also needs to focus on subsequent transitions, stopping the same people simply cycling between bene®ts and work, much of it low paid. Finally, 40% of low income observations are accounted for by people who are not only poor, but whose position in the income distribution does not change signi®cantly over a four year period. For many of these ± disproportionately low income pensioners and lone parents and their children ± it is the level of social security bene®ts which will have the greatest effect on their standard of living. London School of Economics and Political Science
References Atkinson, A. B., Maynard, A. K. and Trinder, C. G. (1983). Parents and Children: Incomes in two generations. SSRC/DHSS Studies in Deprivation and Disadvantage 10. London: Heinemann. Ball, J. and Marland, M. (1996). Male Earnings in the Lifetime Labour Market Database, DSS Analytical Services Working Paper No. 1. London: Department of Social Security. Barclay, Sir Peter (chair) (1995). Joseph Rowntree Foundation Inquiry into Income and Wealth. Volume 1: Report. York: Joseph Rowntree Foundation. Blundell, R. and Preston, I. (1998). `Consumption inequality and income uncertainty.' Quarterly Journal of Economics, vol. 113, pp. 603±40. Department of Social Security [DSS] (1994). Households Below Average Income: A statistical analysis 1979± 1991/92. London: HMSO. DSS (1997). Households Below Average Income: A statistical analysis 1979±1994/95. London: The Stationery Of®ce. # Royal Economic Society 1999
Copyright © 2000 All Rights Reserved
1999]
NEW DATA ON INCOME MOBILITY
F111
Dickens, R. (1997). `Caught in a trap? Wage mobility in Great Britain: 1975±94'. Centre for Economic Performance, LSE (mimeo). Field, F. (1998). `The great divide: the future of welfare reform'. Speech to the Social Market Foundation, 6 August 1998 (mimeo). Friedman, M. (1992). `Do old fallacies ever die?', Journal of Economic Literature, vol. 30 (December), pp. 2129±32. Goodman, A., Johnson, P. and Webb, S. (1997). Inequality in the UK. Oxford: Oxford University Press. Goodman, A. and Webb, S. (1994). For Richer and Poorer. The changing distribution of income in the United Kingdom 1961±91, IFS Commentary No. 42. Institute for Fiscal Studies. Gosling, A., Machin, S. and Meghir, C. (1994). What has Happened to Wages?, IFS Commentary No. 43. Institute for Fiscal Studies. Gregg, P. (ed.) (1997). Jobs, Wages and Poverty: Patterns of persistence and mobility in the new ¯exible labour market. Centre for Economic Performance, LSE. Hancock, R. (1985). Explaining Changes in Families' Relative Net Resources: An analysis of the Family Finances and Family Resources Surveys, TIDI discussion paper no. 84. STICERD, London School of Economics. Hills, J. (1995). Income and Wealth, Volume 2: A survey of the evidence. York: Joseph Rowntree Foundation. Hills, J. (1998). Income and Wealth: The latest evidence. York: Joseph Rowntree Foundation. Jarvis, S. and Jenkins, S. (1996). Changing Places: New evidence about income dynamics from the British Household Panel Survey, Occasional Paper 95-2, ESRC Research Centre on Micro-Social Change, University of Essex. Jarvis, S. and Jenkins, S. (1997a). `Income dynamics in Britain: new evidence from the British Household Panel Survey' in Gregg (1997). Jarvis, S. and Jenkins, S. (1997b). `Low income dynamics in 1990s Britain', Fiscal Studies, vol. 18, no. 2, pp. 123±42. Jenkins, S. (1998). `Modelling household income dynamics', Presidential Address to the European Society for Population Economics Twelfth Annual Congress, Amsterdam, June. Lilley, P. (1996). `Equality, generosity and opportunity: welfare reform and Christian values', speech in Southwark Cathedral, 13 June, Department of Social Security (mimeo). Nicholls, M., Ball, J. and Marland, M. (1997). `The Department of Social Security Lifetime Labour Market Database', in Gregg (1997). Rowntree, S. (1902). Poverty: A study of town life. London: Nelson. Schulter, C. (1998). Income Dynamics in Germany, the USA, and the UK: Evidence from panel data, CASE paper 8, Centre for Analysis of Social Exclusion, London School of Economics.
# Royal Economic Society 1999
Copyright © 2000 All Rights Reserved
