Do Self-Insurance and Disability Insurance Prevent Consumption Loss on Disability?

Finance and Economics Discussion Series Divisions of Research & Statistics and Monetary Affairs Federal Reserve Board, Washington, D.C. Do Self-Insurance and Disability Insurance Prevent Consumption Loss on Disability? Steffan G. Ball and Hamish W. Low 2009-31 NOTE: Staff working papers in the Finance and Economics Discussion Series (FEDS) are preliminary materials circulated to stimulate discussion and critical comment. The analysis and conclusions set forth are those of the authors and do not indicate concurrence by other members of the research staff or the Board of Governors. References in publications to the Finance and Economics Discussion Series (other than acknowledgement) should be cleared with the author(s) to protect the tentative character of these papers.

Do Self-Insurance and Disability Insurance Prevent Consumption Loss on Disability? (cid:3) Ste⁄an Ball Hamish Low y z Federal Reserve Board University of Cambridge and IFS June 2009 Abstract Inthispaperweshowtheextenttowhichpublicinsuranceandself-insurance mitigate the cost of health shocks that limit the ability to work. We use consumption data from the UK to estimate the insurance provided by the government disability programme and account for the e⁄ectiveness of alternative self-insurance mechanisms. Individuals with a work-limiting health condition, but in receipt of disability insurance, have 7% lower consumption than those without such a condition. Self-insurance through savings and a working partner each provide some insurance bene(cid:133)t, improving outcomes from 2% to 4%. Reductionsinthegenerosityofincapacitybene(cid:133)tafter1995areassociatedwith increases in the consumption loss associated with disability. Keywords: disability insurance, living standards, consumption, liquidity constraints JEL Classi(cid:133)cation: D91, J14, H53 WewouldliketothankJamesBanks, TomCrossley, andparticipantsofseminarsatCambridge (cid:3) University, theRoyalEconomicSocietyandtheFederalReserveBoard. Theviewsinthispaperare solely the responsibility of the authors, and should not be interpreted as re(cid:135)ecting the views of the BoardofGovernorsoftheFederalReserveSystem,oranyotherpersonsassociatedwiththeFederal Reserve System. BoardofGovernorsoftheFederalReserveSystem,20thandCSt.,NW,WashingtonDC20551. y Email: ste⁄an.g.ball@frb.gov Faculty of Economics, University of Cambridge, Sidgwick Avenue, Cambridge, CB3 9DD. z Email: hamish.low@econ.cam.ac.uk 1

1 Introduction There is now signi(cid:133)cant evidence of the growth in claimants of disability bene(cid:133)ts since the mid-1980s in the UK, and convincing evidence that, although these bene(cid:133)ts have become less generous over time, they have nonetheless become increasingly generous relative to other public insurance programmes. The (cid:133)nancial cost of incapacity bene(cid:133)t is now almost three times the cost of providing government unemployment insurance, and for the last decade the number of registered disabled has far exceeded theunemployed(DWP,2007b). Despitethis, therehasbeenlittleattempttomeasure the welfare bene(cid:133)t of the insurance provided by state disability support. The aim of this paper is to (cid:133)ll this gap. We use consumption data to estimate the costs of disability and the insurance bene(cid:133)t provided by the government disability programme, and to account for the e⁄ectiveness of alternative self-insurance mechanisms. Much of the analysis of disability insurance has focused on the incentive e⁄ects of disability insurance on labour supply (Parsons, 1980a, b; Bound, 1989), and on false applications by those who are not truly disabled (Benitez-Silva, Buchinsky, and Rust, 2004). Similarly, theliteratureexploringthegrowthinclaimantspointstoacombination of changes to disability bene(cid:133)t generosity (Disney and Webb, 1991; Huddleston, 2000; Beatty et al, 2000; Bell and Smith, 2004) and screening intensities (Berthound, 1998), in addition to falling demand for low skilled workers (Huddleston, 2000; Bell and Smith, 2004; McVicar, 2008). These explanations focus on the incentive e⁄ects and the costs of providing disability insurance. Our focus instead is on the value of disability bene(cid:133)t to those who receive it. The value of this insurance will depend on the seriousness of the disability and the availability of alternative mechanisms for self-insurance. UsingdataonconsumptionexpendituresfromtheBritishHouseholdPanelSurvey (BHPS) we show the extent to which disability leads to lower consumption and the 2

degree to which consumption losses are insured by positive asset holdings, having a spouse who is working, own labour force participation, and disability insurance. None of these insurance mechanisms provide full insurance, and they are imperfect substitutes. We (cid:133)nd that individuals experiencing a work-limiting health condition, whose only support is the state social security programme, experience a 7% drop in food expenditures. Given that food is a necessary and nondurable good, this represents a substantial fall in total consumption. Each self-insurance mechanism o⁄ers some mitigation for this fall in food expenditures, ranging from 2% to 4% improvement; holding positive assets is the most e⁄ective of these mechanisms. Among the disabled, individuals not receiving disability insurance, as a group, are observed to have higher food expenditure than those in receipt of insurance payments. This arises because the consumption level of insurance recipients is determined by two o⁄setting mechanisms: (cid:133)rstly, food expenditure will be higher from the direct e⁄ect of individuals receiving bene(cid:133)ts; and secondly, the group(cid:146)s average food expenditure will be lower from the selection e⁄ect of bene(cid:133)ts being targeted at the most severely disabled. Our results indicate that the selection e⁄ect dominates, providing some support for the e⁄ectiveness of the screening mechanism onto state disability support. Further, we use a fully insured group to establish that issues of non-separability are not important in our sample(cid:150)that is, disability a⁄ects food consumption primarily through the channel of income,not by a⁄ecting the marginal utility of consumption. Our average food expenditure drop of 7% is in line with similar econometric studies carried out on US data: Stephens (2001) and Meyer and Mok (2006) (cid:133)nd 5% and 8.6% lower food expenditures for the disabled, respectively. However, these studies do not analyse the e⁄ect of di⁄erent self-insurance mechanisms on expenditure, nor do they investigate the selection e⁄ect that we (cid:133)nd signi(cid:133)cant in the UK data. Theseconsumptionlossesmaskheterogeneityovertime. Wepresentevidencethat 3

the consumption loss associated with disability is greater after 1995, when the generosity of government insurance in the UK was reduced. This reduction in generosity was part of an attempt to reduce the incentive costs of the disability programme, but our results show the implications of this in terms of the reduction in insurance bene(cid:133)ts provided. The next section describes the disability bene(cid:133)t programme in the UK and how it has changed in the recent past. Section 3 describes our data source and sample selection, and provides information on the characteristics of the disabled. Section 4 details the estimation strategy. Section 5 shows our results, estimating the e⁄ect of health shocks on food expenditure in three stages: (cid:133)rst, for a broad measure of disability; second, allowing for di⁄erent types of disability; and (cid:133)nally, looking at the dynamics of consumption loss after the onset of disability. Section 6 provides evidence on the implications of the 1995 reforms for consumption insurance, and section 7 concludes. 2 The Disability Insurance Programme in the UK The Current Bene(cid:133)t Programme There are three main types of public insurance provided to support individuals su⁄ering from a work-limiting health condition. Firstly, there exist four di⁄erent bene(cid:133)ts that are targeted at replacing lost earnings: incapacity bene(cid:133)t, statutory sick pay, carer(cid:146)s allowance and severe disablement allowance. Incapacity bene(cid:133)t (previously called invalidity bene(cid:133)t) is designed to insure individuals against long-term sickness or disability. It is a contributionary, earnings replacement bene(cid:133)t and accounts for the vast majority of disability bene(cid:133)t claimants. Incapacity bene(cid:133)t requires a work history, in the form of su¢ cient accumulated credits, in order to be eligible and usually constrains claimants not to work.1 Individuals 1Though some work may be permitted if earnings remain low. 4

must also pass a (cid:147)personal capability assessment,(cid:148)testing the degree to which they are unable to undertake certain mental and physical tasks. The level of payment individuals receive while claiming incapacity bene(cid:133)t is not earnings related or means tested; bene(cid:133)ts are paid at three di⁄erent (cid:135)at rates, depending on the claimant(cid:146)s length of inactivity. Statutory sick pay is paid by employers and only covers the (cid:133)rst six months of illness. Carer(cid:146)s allowance (previously called invalid care allowance) provides (cid:133)nancial support to spouses for loss of income resulting from leaving employment in order to careforadisabledpartner. Carer(cid:146)sallowancehasbeenplaguedbyeligibilityproblems (for example, married women were only deemed eligible in 1987) and low take-up. Severe disablement allowance is aimed at those with insu¢ cient credit for invalidity bene(cid:133)ts; there have been no new claimants allowed since 2001. Secondly, in addition to these earning replacement bene(cid:133)ts there are a number of means tested bene(cid:133)ts. These include income support (which may come with a disability premiumif the claimant passes the (cid:147)personal capability assessment(cid:148)), working tax credit for the disabled in low paid work, housing bene(cid:133)t, and council tax bene- (cid:133)t. Finally, there are additional cost bene(cid:133)ts intended to compensate for the extra costs associated with disability: for example, transportation expenses, sensory aids, special clothing, or modi(cid:133)cations to household appliances. These include attendance allowances and disability living allowance. Changes in Bene(cid:133)ts over Time Since 1980, increases in disability bene(cid:133)ts have been linked to prices rather than average earnings. Given that wage growth has outstripped in(cid:135)ation over the last three decades, even for the lowest decile of the income distribution (Machin, 2003), average replacement ratios have been falling. Sweeping reforms in 1995 reduced the generosity of disability bene(cid:133)ts further still. However, since incapacity bene(cid:133)t is a very progressive programme, average replacement rates 5

hide signi(cid:133)cant heterogeneity in the generosity of bene(cid:133)ts. The age-related generosity of the programme adds further heterogeneity. While absolute disability bene(cid:133)t replacement ratios have fallen over the last 25 to 30 years, there is evidence that generosity has declined less than for other public insurance payments. In particular, many researchers have drawn comparisons to unemployment bene(cid:133)t, arguing that disability insurance has become relatively more liberal overtheyears (DisneyandWebb, 1991; Bell andSmith, 2004). Bell andSmith (2004) document how the this increased relative generosity is concentrated on older claimants. They show that the ratio of disability bene(cid:133)t to unemployment bene(cid:133)t has remained roughly constant for those under 40 from 1984 onwards, whereas among the older age groups (ages 45-49 and 55-59) the relative replacement ratio has risen from around 150% in 1984 to 200% in 1995. This striking growth came to an end in 1995 when the previous programme of invalidity bene(cid:133)t was renamed incapacity bene(cid:133)t and, crucially, the additional pension bene(cid:133)t was removed. At this time the relative generosity for older claimants fell back to levels similar to those of the early 1980s and has remained at these low levels since then. Coinciding with this growth in relative generosity, there has been signi(cid:133)cant increases in the number of disability bene(cid:133)t claimants during the last few decades. This enrolment growth was particularly pronounced between the mid-1980s and mid- 1990s, withincapacity(invalidity)bene(cid:133)tclaimantsdoublingfromaroundonemillion in the mid-1980s to two million in the mid-1990s (McVicar, 2008). During the last ten years this growth has continued, although at a slower rate, and there are currently almost 2.5 million claimants (DWP, 2007b). The co-movement of claimant numbers and relative generosity has spurred discussion on possible causation (Disney and Webb, 1991; Beatty et al, 2000; Bell and Smith, 2004). These authors argue that the increased generosity of incapacity bene(cid:133)t relative to unemployment insurance has directly caused the increases in disability bene(cid:133)t rolls. See McVicar (2008) for a re- 6

view of the literature attempting to explain the reasons behind the observed growth in disability bene(cid:133)t rolls. In addition to the reduction in absolute generosity in 1995, the government has triedtoreducetheincentivecostsofprovidingdisabilityinsurancethroughthe(cid:147)pathways to work(cid:148)programme, which tries to remove barriers to working for those claimingdisabilitybene(cid:133)t. Priortothisinitiative, whichbeganin1997, verylittlehadbeen done to promote disability claimants moving back into the workforce. In addition to advice on obtaining work and job focussed interviews, tax credits are now available to encourage claimants to return to the labour force, and payments can di⁄er depending on the severity and permanence of the health condition. 3 Data We use data from the British Household Panel Survey (BHPS) between 1991 and 2004. This is a longitudinal data set starting in 1991 with a sample of approximately 10,000 individuals each year. The survey is designed to be representative of the UK population and has information on a wide number of variables such as spending, health,anddemographics. Inthissectionwedetailourchoiceovervariablesofinterest and our sample selection criteria. Disability Status Wecategorisethedisabledusingresponsestothefollowingquestion: (cid:147)Does your health limit the type of work or amount of work that you do?(cid:148) Answers: {yes, no} Those who answer positively to this (cid:147)work limitation(cid:148)question are asked (up to) two further questions from the survey to identify the severity of the disability: 7

(cid:147)Does your health keep you from doing some types of work?(cid:148) Answers: {can do nothing, yes, no} (cid:147)For work you can do, how much does your health limit the amount of work you can do?(cid:148) Answers: {a lot, somewhat, a bit, not at all} We categorise the severely disabled as those who answer (cid:147)can do nothing(cid:148)to the (cid:133)rst additional question, or answer (cid:147)a lot(cid:148)or (cid:147)somewhat(cid:148)to the second additional question. All others are categorised as mildly disabled. In addition to the severity of disability, we also di⁄erentiate according to the duration of the work-limiting condition. We de(cid:133)ne a short-term disabled spell to be one that lasts up to three periods in total, with long-term disability representing a work-limiting condition for four or more periods. Using these two dimensions we can disaggregate down to four di⁄erent categories of disability: short-term mildly disabled, long-term mildly disabled, short-term severely disabled and long-term severely disabled. There is some debate as to the reliability of self-reported responses to questions about health (see Banks et al, 2005), with some authors preferring to use disability bene(cid:133)t receipt to de(cid:133)ne work limitation (see, Bound et al, 2006). However, the use of disability bene(cid:133)t rolls is also subject to signi(cid:133)cant biases. Many of those truly disabled may not apply for disability bene(cid:133)t, for reasons ranging from stigma to ignorance to ineligibility. Also, many applicants are denied, and such rejection does not necessarily mean that these individuals are not su⁄ering work limitations. Further, the use of self-reports is now becoming commonplace in the literature (see, Meyer and Mok, 2006; Stephens, 2001; Burchardt, 2000) and a study by Benitez-Silva et al (2004) has shown that these responses provide unbiased estimates of disability bene- (cid:133)t eligibility decisions. Given this paper(cid:146)s focus on analysing both self-insurance and 8

public insurance, we are interested in capturing all types of work-limiting conditions, not just those covered by the disability bene(cid:133)t programme. Therefore, even though these self-reports are not without their limitation, we believe such responses are the best available criteria for assigning disability onset and duration. Insurance Mechanisms We analyse both public and self-insurance mechanisms. Respondents are classi(cid:133)ed as receiving disability insurance if they ever received the main class of government provided insurance, namely incapacity (invalidity) bene(cid:133)t. We consider three distinct self-insurance mechanisms: (cid:133)rstly, individuals can use personalsavingtobu⁄erworklimitations. TheBHPScontainslimitedinformationon household asset allocations, and we take households who report positive investment earnings2 in the month prior to the interview as those able to use savings to selfinsure. Secondly, we look at individuals with a working partner.3 Finally, we consider individuals who remain in the labour force while experiencing a work limitation. We include both self-employed and the employed in this de(cid:133)nition. Consumption The BHPS contains data on food expenditure. In each wave, respondents have been asked the following question: Please [...] tell me approximately how much your household spends each week on food and groceries? Include all food, bread, milk, soft drinks etc., exclude pet food, alcohol, cigarettes and meals out. For all but the (cid:133)rst wave the interviewer asks the respondent to assign their expenditure into one of twelve bands, rather than giving a precise (cid:133)gure.4 However, in the (cid:133)rst wave a precise amount was reported. We assign this value into the same bandsusedforallsubsequentyears,andtreatallthedataasthoughtheywerebanded. 2This includes earnings from rents, savings and investments. 3The classi(cid:133)cation (cid:147)married(cid:148)includes those who report that they are (cid:147)living as a couple(cid:148). 4Answers (waves 2-13): {Under £19, £10-£19, £20-£29, £30-£39, £40-£49, £50-£59, £60-£79, £80-£99, £100-£119, £120-£139, £140-£159, £160 and over} 9

These expenditure data are far from ideal. It would be better to have detailed expenditure data on goods other than food, as total expenditure may not respond in the same way as food expenditure to changes in health status. However, we cannot use the Family Expenditure Survey, for example, primarily because that source does not have a disability question and is not a panel. Food has the advantage of being a nondurable, necessary good with a small income elasticity. Recent work by Browning andCrossley(2008)hasshownthatmanyhouseholdssmoothconsumptionbycutting back on the purchase of durables, leaving nondurable expenditure almost unchanged. This implies that any test based solely on nondurable expenditures (such as food) is not very sensitive. The restrictions imposed on us from the BHPS data mean that we are putting into force a weak test of the extent of consumption loss, and any e⁄ect we (cid:133)nd of disability on food expenditures can be interpreted as a lower bound to the e⁄ect on total expenditures. Despite their drawbacks, food expenditure data are the basis of much empirical work, andrelatedstudieshaveusedverysimilarmeasuresforconsumption(Stephens, 2001; MeyerandMok,2006). Apotentiallylargerproblemoriginatesfromthebanded nature of our food data. This means that in our food expenditure regressions we do not have a continuous variable as our dependent variable, so we cannot use standard OLS estimators (see section 5). Other Individual Characteristics We construct dummy variables for gender, marital status, educational achievement, home ownership, and ethnicity. We also use variables on respondents(cid:146)age and the number of persons in the household. Sample Selection WeusetheentireBHPSunbalancedpanelfrom1991throughto the 2004wave inthis study. We dropthe oversamplingof lowincome individuals, and keep the original sample that was designed to be representative. This gives us 16,082 10

respondents before our sample selection. We select male and female respondents of workingage,restrictedtobebetween25and60yearsold. Weuseindividualresponses and input food expenditure and household characteristics fromthe data on household responses. No health question was asked in 1999, compelling us to drop this wave entirely. We only consider individuals for whom we have at least four years of data, with a minimum of three of these running consecutively. In order to control on past observables, werequirethatindividualshaveatleastonedatapointpriortotheir(cid:133)rst reporting of disability. Following Burkhauser (1999), we re-classify those respondents whoonlyreportoneperiodofdisabilityasnotdisabled.5 Finallywedropobservations where key variables (demographics, health status, disability bene(cid:133)t receipt, region) are missing. This leaves us with a sample of 5,985 individuals over twelve waves (1991-2004), with an average of just over nine responses per individual. 3.1 Characteristics of the Disabled This section provides information on the characteristics of individuals with healthrelated work limitations. We focus on three aspects: (cid:133)rst, we provide information on the prevalence of disability in the sample, and the prevalence of disabilities of di⁄ering severity and duration. Second, we analyse the individual characteristics of the disabled. Finally, we provide information on the characteristics of those receiving disability insurance. Prevalence of Disability The (cid:133)rst striking point about disability is the sheer number of individuals who su⁄er at least some work limitation. In our sample, over a quarter of respondents report a health-related work limitation at least once. Of 5We later test this assumption and (cid:133)nd no signi(cid:133)cant consumption loss for individuals experiencing a single period work limitation, suggesting these shocks are not serious or are contaminated by measurement error. 11

those who become disabled within our sample period (and have a spell of more than one year), just over 50% su⁄er from long-term poor health, almost 22% experience a severedisablement,and12%aresubjecttoahealthshockthatisbothsevereandlongterm. Thus, these work limitations a⁄ect signi(cid:133)cant proportions of the population, often with acute and long-term consequences. Individual Characteristics Given the evidence on the prevalence of work limitations it is important to determine the characteristics of individuals a⁄ected by these conditions. To gain insights into the attributes of the disabled, we run a series of probit regressions. We use a pooled probit estimation, and correct the standard errors for dependence across the panel structure. Table 1 shows the marginal e⁄ects from estimating disability status as a function of age, income, time trend and a number of individual characteristics. The (cid:133)rst column of Table 1 shows the results from a probit on the probability of being disabled for our whole sample. We (cid:133)nd that the disabled are more likely to be old, female, married without a working spouse, non-white and less educated. This details the well-documented relative depravation of those su⁄ering from work limitations. In the second column of table 1 we show a probit regression on the probability of being long-termdisabled, conditional on being disabled, using the same explanatory variables as before. We (cid:133)nd that out of those su⁄ering from a work limitation, individuals experiencing longer duration tend to be signi(cid:133)cantly older, non-white, and less likely to own their own home. The third column documents the probability of severe disability conditional on being disabled. Again we see increased likelihood of older individuals but many of the other variables are not signi(cid:133)cant. Recipients of Disability Insurance A (cid:133)rst step to understanding how well these negative health shocks are insured by public insurance is to identify who actually 12

Table 1: Probit estimation for disability status Pr[Disabled] Conditional on being disabled Pr[long-term dis] Pr[Sev dis] Age 0.008 0.077 0.031 (0:002) (0:014) (0:016) (cid:3)(cid:3)(cid:3) (cid:3)(cid:3)(cid:3) (cid:3)(cid:3) Age squared/1000 -0.063 -0.841 -0.322 (0:020) (0:157) (0:179) (cid:3)(cid:3)(cid:3) (cid:3)(cid:3)(cid:3) (cid:3) Male -0.014 0.006 0.033 (0:004) (0:029) (0:036) (cid:3)(cid:3)(cid:3) Married, working spouse -0.032 -0.049 -0.09 (0:005) (0:031) (0:037) (cid:3)(cid:3)(cid:3) (cid:3)(cid:3) Single -0.014 -0.052 -0.059 (0:005) (0:044) (0:047) (cid:3)(cid:3)(cid:3) College -0.022 -0.087 -0.083 (0:005) (0:050) (0:048) (cid:3)(cid:3)(cid:3) (cid:3) High school -0.008 -0.006 -0.062 (0:005) (0:035) (0:04) (cid:3) Home owner -0.046 -0.079 0.009 (0:006) (0:030) (0:039) (cid:3)(cid:3)(cid:3) (cid:3)(cid:3)(cid:3) Household size 0.002 -0.001 -0.006 (0:002) (0:011) (0:013) White -0.016 -0.069 -0.052 (0:006) (0:032) (0:045) (cid:3)(cid:3)(cid:3) (cid:3)(cid:3) Year dummies Yes Yes Yes Mean Value 0.143 0.503 0.218 N 5985 859 859 Pseudo R-squared 0.07 0.05 0.02 Coe¢ cients show marginal e⁄ects. Robust standard errors in parentheses. (cid:3) signi(cid:133)cant at 10%; signi(cid:133)cant at 5%; signi(cid:133)cant at 1%. (cid:3)(cid:3) (cid:3)(cid:3)(cid:3) 13

receives disability bene(cid:133)t. In this section, we (cid:133)rst report the raw correlations of disability insurance receipt by self-reported disability status. We then present the results of probit regressions of disability receipt on observable characteristics. Table 2: Disability insurance receipt by health status Work limitation No work limitation Total Disability bene(cid:133)t 0.60 0.40 0.36 0.04 0.09 (311) (211) (522) No disability bene(cid:133)t 0.10 0.90 0.64 0.96 0.91 (548) (4,915) (5;463) Total 0.14 0.86 (859) (5;126) Row percentages in bold, column percentages in italics, cell sizes in parentheses. Table 2 shows the relationship of disability insurance receipt to disability status. An individual is classi(cid:133)ed as being work limited if they have ever reported su⁄ering from a work-limiting condition, and similarly classi(cid:133)ed as being on disability bene(cid:133)t if they have ever received disability payments.6 The (cid:133)rst issue is the fraction of false positives: 40% of the sample have received disability bene(cid:133)ts despite never su⁄ering a work limitation in our sample period. The second issue is the fraction of those with a work limitation who never receive disability insurance. 64% of those reporting a work limitation never bene(cid:133)t from any state disability insurance, although this falls to 35% if we condition on those reporting a severe disability. We cannot infer from this fraction that the disability application process is rejecting legitimate claimants becausewedonotobservewhohasappliedforbene(cid:133)ts. Indeed, giventherequirement 6Very similar correlations are obtained if contemporaneous measures of disability and bene(cid:133)ts are used. 14

that recipients are allowed to work only a very limited amount, this high percentage of unhealthy individuals not receiving bene(cid:133)ts may simply re(cid:135)ect a large number not applying for disability insurance. Without data on the disability application decision, we will not be able to disentangle this e⁄ect. On the other hand, the fraction does tell us that disability bene(cid:133)t is not providing insurance to a large fraction of those who have su⁄ered a shock to their health. To determine the characteristics of the recipients of disability bene(cid:133)ts we run a probit regression of the contemporaneous receipt of state insurance on observable characteristics. The results are shown in the (cid:133)rst column of table 3 using the broad measure of work limitation. This table shows, reassuringly, that the main economically signi(cid:133)cant variable is disability status. Further, bene(cid:133)ts are more likely among the relatively less well-o⁄members of society, with successful claimants coming from low educated, non-home-owning households, where the claimant is more likely to be married without a working spouse. Age, gender and ethnicity have no signi(cid:133)cant e⁄ect on the likelihood of receiving bene(cid:133)ts. In the second column of table 3 we repeat this same probit estimation splitting disability status by duration and severity. We see signi(cid:133)cant di⁄erences between the coe¢ cients, with the marginal e⁄ect on the probabilityofreceivingbene(cid:133)tsincreasingfrom10%forshort-termmilddisabilitiesto almost 60% for long-term severe conditions. The more serious work limitations, both in terms of severity and duration, are much more likely to be covered by disability insurance. The results in this section inform the debate about who actually bene(cid:133)ts from disability insurance. The next sections address the question of how much these individuals bene(cid:133)t. 15

Table 3: Probit estimation for disability bene(cid:133)t receipt. Pr[Disability bene(cid:133)t] Pr[Disability bene(cid:133)t] Disabled 0.267 (0:015) (cid:3)(cid:3)(cid:3) Short-term mildly disabled 0.107 (0:015) (cid:3)(cid:3)(cid:3) Long-term mildly disabled 0.250 (0:022) (cid:3)(cid:3)(cid:3) Short-term severely disabled 0.415 (0:049) (cid:3)(cid:3)(cid:3) Long-term severely disabled 0.584 (0:037) (cid:3)(cid:3)(cid:3) Married, working spouse -0.025 -0.024 (0:004) (0:004) (cid:3)(cid:3)(cid:3) (cid:3)(cid:3)(cid:3) Single -0.005 -0.004 (0:004) (0:004) College -0.023 -0.022 (0:003) (0:003) (cid:3)(cid:3)(cid:3) (cid:3)(cid:3)(cid:3) High school -0.011 -0.011 (0:003) (0:003) (cid:3)(cid:3)(cid:3) (cid:3)(cid:3)(cid:3) Home owner -0.026 -0.026 (0:004) (0:004) (cid:3)(cid:3)(cid:3) (cid:3)(cid:3)(cid:3) Mean Value 0.087 0.087 N 5,985 5,985 Pseudo R-squared 0.23 0.25 Coe¢ cients show marginal e⁄ects. Both speci(cid:133)cations include year dummies. Other controls which are insigni(cid:133)cant: age, age2, gender, household size, race. Robust standard errors in parentheses. signi(cid:133)cant at 10%; signi(cid:133)cant at 5%; (cid:3) (cid:3)(cid:3) signi(cid:133)cant at 1%. (cid:3)(cid:3)(cid:3) 16

4 Estimation Strategy Underlyingourestimationofthebene(cid:133)tsofdisabilityinsuranceisatheoreticalframework where individuals choose consumption to smooth marginal utility over their lifecycle. When a shock to disability occurs, an individual(cid:146)s income falls, with the extent of the impact on permanent income depending on the persistence of the shock as well as its severity. This fall in income leads to lower consumption. Consumption may also fall simply because the marginal utility of consumption is lower when disabled, and this induces an intertemporal reallocation. A fall in consumption due to such non-separabilities between consumption and disability does not, in itself, constitute a welfare loss because it re(cid:135)ects smoothing of marginal utility. A fall in consumption is costly if the change in consumption re(cid:135)ects an inability to smooth marginal utility, and it is only in this case that insurance is valuable. Theseconsiderationsgiverisetothefollowingreducedformequationforconsumption: lnC = (cid:12) X +(cid:13) Z +! +" (1) it 0 it 0 it i it where Z includes indicators of disability status and interactions involving disability status; X includes a set of controls for permanent income and observable characteristics; ! represents individual heterogeneity and is interpreted as the marginal utility of wealth.7 The controls for heterogeneity in permanent income are necessary so that 7Itiswellknownthatoptimalintertemporalallocationimpliesthatthemarginalutilityofwealth, (cid:21) , should follow a unit root: t (cid:21) =(cid:21) u t t 1 t (cid:0) (cid:0) where u is a random error term. Extrapolating this through time allows us to write this stochastic t process as: t (cid:21) =(cid:21) u t 0 j (cid:0) j=1 X Hence, the marginal utility of wealth can be captured by a (cid:133)xed e⁄ect, (cid:21) , and a composite error 0 17

the coe¢ cients on disability capture the e⁄ects of di⁄erences in disability on consumption rather than permanent di⁄erences across individuals. Our estimates of the e⁄ect of disability will still be a combination of the e⁄ect due to lost income and the e⁄ect due to any non-separabilities. We identify the extent of the non-separability by looking at the consumption loss for groups who we would expect to be fully insured. Due to the categorical structure of our consumption data we cannot di⁄erence out the marginal utility of wealth, and coe¢ cients can only be estimated consistently if we make a distributional assumption. To overcome this di¢ culty we implement the technique of interval regression. In e⁄ect, this is an ordered probit with the cut points (cid:133)xed. By assuming that the conditional distribution of the dependent variable is normally distributed, we can estimate our coe¢ cients using maximum likelihood. The integral in the maximum likelihood is approximated by Gaussian-quadrature. Simulation studies have shown that this is a reasonable approximation for small time dimension panels like ours, though we also perform a post estimation check on the applicability of the numerical technique used. This procedure has a number of drawbacks: (cid:133)rst, estimation is somewhat slow as quadrature methods are used to approximate the integral in the likelihood; and more importantly, with this procedure (cid:133)xed e⁄ects cannot be conditioned out of the likelihood necessitating that individual heterogeneity be assumed i.i.d.. Thus, our data restrictions force us to use random e⁄ects and a rich set of controls to condition out di⁄erences in the marginal utility of wealth across individuals. As a simple robustness test, we have run (cid:133)xed e⁄ect regressions using the mid-points of consumptionbandsasthedependentvariable. Theresultsarequalitativelyconsistent with our interval regressions.8 term (see Browning and Crossley, 2001). 8Regressions not shown, results available upon request. 18

5 Results The e⁄ect of disability on consumption is highly heterogeneous because individuals experiencedi⁄erenthealthshocksandhaveaccesstoarangeofinsurancepossibilities. Healthstatus varies widelyacross individuals, withsome beingsubject tomore minor grievances that do not persist for many periods, and others experiencing long-term severe disablement. We analyse the e⁄ect of both state insurance and self-insurance mechanisms on mitigating food expenditure loss for those experiencing disability. We consider three distinct forms of self-insurance: savings, spousal income and own labour force participation. Individuals holding a precautionary bu⁄er of assets can run down these funds during periods of poor health in order to smooth consumption; within a couple, partner income can be used to mitigate adversity in response to a work limitation; and, for some individuals a work limitation does not necessitate complete inactivity, enabling own labour income to help alleviate hardship. The e⁄ectiveness of each of these mechanisms will depend on individual circumstance, and on the severity and duration of the health shock. We analyse the value of these di⁄erent mechanisms in three stages. We begin with a broad de(cid:133)nition of disability, covering individuals who have su⁄ered both mild and severe shocks, ignoring di⁄erences in duration. Given this de(cid:133)nition, we investigate the e⁄ect of the various public and self-insurance mechanisms on consumption. For this baseline speci(cid:133)cation, we provide evidence that we have adequately controlled for heterogeneity in the marginal utility of wealth by testing that future disability status does not predict current consumption. We also provide evidence that nonseparabilities between consumption and disability are insigni(cid:133)cant. In the second stage, wedisaggregatebytheseverityanddurationofthedisabilityshockandanalyse the di⁄ering insurance mechanisms. Finally, we look at the dynamics of declines in 19

food expenditure post onset of the work-limiting condition. In all our analysis, we investigate the average e⁄ect of disability on consumption over the whole sample period. It is important to note that over the period in question there have been a number of changes in the state disability programme, hence our estimates average over multiple policy regimes. In section 6 we look at the e⁄ect of these policy changes. 5.1 Consumption Losses of the Disabled Table 4 shows the results of our interval regressions on the reduced form equation (1) for the broad measure of disability. This de(cid:133)nition includes all types of work limitation, incorporating mild, severe, long term and short-term disablements. In all regressions we include a rich set of controls to condition out di⁄erences in marginal utility of wealth (or permanent income) across individuals. The two columns di⁄er by whether or not we control for self-insurance mechanisms. In the (cid:133)rst column, we report the e⁄ect of a work limitation on food expenditures without controlling for any insurance mechanism. In this case, food expenditure is 2.7% lower when disabled once we have conditioned on observable characteristics. However, this number con(cid:135)ates a number of issues: (cid:133)rst, even with full insurance, the marginal utility of consumption at a given level of consumption may be di⁄erent across disabled and non-disabled individuals if there are non-separabilities between consumption and health; we return to this issue below. Second, each individual has access to vastly di⁄erent insurance mechanisms, and the (cid:133)gure for the consumption loss is averaging over these di⁄erences across individuals. In the second column, we interact disability status with dummy variables for the presence of a working spouse, positive asset holdings, and labour force participation. We also interact disability status with a dummy variable indicating individuals who 20

Table 4: The E⁄ect of Disability on Consumption Dependent variable Log food spending (1) Log food spending (2) 0:027 0:071 Disabled (cid:0) (cid:0) (0:006) (0:011) (cid:3)(cid:3)(cid:3) (cid:3)(cid:3)(cid:3) Interactions: 0:025 disabled + married with working spouse (0:014) (cid:3) 0:040 disabled + positive assets (0:014) (cid:3)(cid:3)(cid:3) 0:021 disabled + labour force participation (0:015) 0:032 disabled + no disability bene(cid:133)t (0:013) (cid:3)(cid:3) Controls: 0:032 0:032 Age (0:002) (0:002) (cid:3)(cid:3)(cid:3) (cid:3)(cid:3)(cid:3) 0:282 0:282 Age squared/1000 (cid:0) (cid:0) (0:020) (0:020) (cid:3)(cid:3)(cid:3) (cid:3)(cid:3)(cid:3) 0:005 0:005 Male (cid:0) (cid:0) (0:007) (0:007) 0:054 0:053 Married, non-working spouse (cid:0) (cid:0) (0:005) (0:005) (cid:3)(cid:3)(cid:3) (cid:3)(cid:3)(cid:3) 0:151 0:150 Single (cid:0) (cid:0) (0:005) (0:005) (cid:3)(cid:3)(cid:3) (cid:3)(cid:3)(cid:3) 0:072 0:071 College (0:009) (0:008) (cid:3)(cid:3)(cid:3) (cid:3)(cid:3)(cid:3) 0:006 0:007 High school (cid:0) (cid:0) (0:006) (0:006) 0:105 0:104 Home owner (0:005) (0:005) (cid:3)(cid:3)(cid:3) (cid:3)(cid:3)(cid:3) 0:186 0:186 Household size (0:002) (0:002) (cid:3)(cid:3)(cid:3) (cid:3)(cid:3)(cid:3) Year and regional dummies Yes Yes N 5985 Robust standard errors in parentheses. signi(cid:133)cant at 10%; signi(cid:133)cant at 5%; (cid:3) (cid:3)(cid:3) signi(cid:133)cant at 1%. (cid:3)(cid:3)(cid:3) 21

do not receive disability insurance. The coe¢ cient on disability now re(cid:135)ects the food expenditurefallsofanindividualwithoutanyself-insurance, butwhoreceivesdisability insurance. The loss for such individuals is estimated at 7.1%. For those that do not have self-insurance, and do not receive disability insurance, the food expenditure fall is actually less at 3.9%. This apparent improvement in mitigating consumption loss likely re(cid:135)ects a selection issue: disability insurance is intended to be paid to individuals who have su⁄ered the worst shocks to their health. This suggests that the screening process for disability bene(cid:133)t is partially e⁄ective. Such a selection issue creates a di¢ culty in interpreting the signi(cid:133)cantly larger fall in consumption for those on bene(cid:133)ts as evidence of insu¢ cient insurance cover. It is likely that these individuals would have fared far worse in the absence of state assistance, especially given the de(cid:133)ciency of self-insurance mechanisms. However, given that we do not observe the counterfactual of these individuals(cid:146)food expenditure levels in the absence of incapacity bene(cid:133)t, we cannot quantify the true welfare bene(cid:133)t of disability insurance. The results in the second column of table 4 highlight the bene(cid:133)ts of self-insurance. We(cid:133)ndthatthepresenceofaworkingspouseandpositiveassetholdingsmitigatethe food expenditure fall by 2.5% and 4.0%, respectively. In addition, we (cid:133)nd a positive (but statistically insigni(cid:133)cant) e⁄ect from own labour force participation. We do not look at the e⁄ects on consumption of those who receive disability insurance but who are not disabled. This is partly because our focus is on the bene(cid:133)t of disability insurancetothosewhosu⁄erfrompoorhealthconditions, andpartlybecauseofsmall sample size. Non-Separabilities The (cid:133)rst potential problem with interpreting these results is due to the possibility of non-separabilities. As discussed above, in addition to the e⁄ect through the budget constraint, health status can in(cid:135)uence the marginal utility of consumption directly: the marginal utility at a given level of consumption may be 22

di⁄erent for an individual when they are disabled, compared to when they are fully healthy. Even with full insurance, consumption may vary over di⁄erent work-limiting conditions, and this would imply that a drop in consumption across disability status may not be evidence of imperfect insurance. To try to tackle the extent of these non-separabilities, we analyse two sub-samples of households: (cid:133)rst, we look only at households with positive assets and a working spouse, where we would expect selfinsurance to be most e⁄ective; and second, we select only those households who are in the bottom ten percent of the income distribution, where we would expect state insurance to be close to complete.9 These results are shown in table 5. Table 5: Testing for Non-separabilities Dependent variable Log food spending (1) Log food spending (2) A > 0 Low income group t 0:002 0:020 Disabled (cid:0) (cid:0) (0:016) (0:015) N 1,006 630 Column (1) shows regression for subsample with positive assets and working spouse. Column (2) shows regression for the bottom decile of the income distribution. Other controls: age, age squared, household size, education, sex, marital status, home ownership, time and regional dummies. Standard errors in parentheses. Neither of the coe¢ cients reported is signi(cid:133)cant. From the (cid:133)rst column of table 5 we see that there is no signi(cid:133)cant fall in food expenditures during disability for those individuals who have positive assets holdings and a working spouse. Similarly, the insigni(cid:133)cant coe¢ cient in the second column of table 5 demonstrates that there is no clear drop in food expenditure for households in the bottom decile of the income distribution. We expect these two sets of individuals tobeclosetofullyinsured,andsothisevidenceishighlysuggestiveofnon-separability issues being unimportant. This (cid:133)nding is consistent with other work demonstrating that the presence of non-separabilities in the utility function is weak (De Nardi et al, 9This is only strictly valid if disability insurance is not subject to type I and type II errors. 23

2006). Unobserved Heterogeneity The second check on our results is over whether we have adequately controlled for individual heterogeneity. In particular, we need to ensure that our indicator of disability status is not picking up omitted characteristics of individuals. To show that this is not the case, we construct an indicator of future disability status which equals one for individuals who become disabled at some point in the future but who are currently not disabled. Table 6: Insigni(cid:133)cance of Future Disability Dependent variable Log food spending 0:030 Disabled (cid:0) (0:007)(cid:3)(cid:3)(cid:3) 0:007 Future disability (cid:0) (0:008) N 5,985 Future disability equals 1 for individuals who are currenlty not disabled, but who become disabled later in the sample. Other controls: age, age squared, household size, education, sex, marital status, home ownership, time and regional dummies. Standard errors in parentheses. signi(cid:133)cant at 10%; (cid:3) (cid:3)(cid:3) signi(cid:133)cant at 5%; signi(cid:133)cant at 1%. (cid:3)(cid:3)(cid:3) The results in table 6 show that becoming disabled at some point in the future does not have an impact on consumption in the current period, while consumption is depressed in the periods when individuals are actually disabled. As the future disability variable is not signi(cid:133)cant, this suggests that our speci(cid:133)cation adequately controls for individual heterogeneity. 5.2 The E⁄ect of Disability Severity and Duration The e⁄ect of disability on consumption and the e⁄ectiveness of di⁄erent insurance mechanisms depend on the type of health shock an individual receives and its ex- 24

pected duration. We classify health shocks in two dimensions: severity and realised duration.10 This gives four types of disability: short-term severe, long-term severe, short-term mild and long-term mild. As with table 4, we (cid:133)rst present regressions of foodexpenditureonobservablecharacteristics, nowincludingdummiesforeachofthe possible disability shocks. We then present regressions where we interact disability status with the various insurance mechanisms. In table 4 we saw that food expenditure was 2.7% lower for those with any type of disability. The (cid:133)rst column of Table 7 shows the extent to which this number is averaging over individuals with quite di⁄erent experiences. Those with short-term andmilddisabilitiesseenosigni(cid:133)cantchangeinfoodexpenditure, whereasthosewith a mild disability of a longer duration see a signi(cid:133)cant fall of 3.3%. For the severely disabled we see even greater falls of 4.0% for short-term and 4.3% for long-term. While these numbers are signi(cid:133)cantly di⁄erent from zero, they are not signi(cid:133)cantly di⁄erentfromeachother, exceptforthelossofthosewithashort-termmilddisability. Thesedi⁄erenttypesofdisabilityhavedi⁄erentinsurancepossibilities. Thesecond column of table 7 introduces interactions for individuals with self-insurance and those who never receive disability bene(cid:133)ts. The key point is that food expenditure is about 6-8% lower for individuals with all forms of disability who receive disability insurance and have no self-insurance. This represents the consumption level supported by state insurance. The(cid:135)atnessofthebene(cid:133)tscheduleleadstonosigni(cid:133)cantdi⁄erenceinfood expenditures across the di⁄erent classi(cid:133)cations of disability. As before, the di⁄erent self-insurance mechanisms mitigate the food expenditure fall by between 2% and 4%, and we see evidence of the selection e⁄ect whereby the worst-o⁄individuals are those 10Realised duration may be a problematic characteristic to be conditioning on. It is valid only if individuals know at onset of the disability whether they have received a shock that will be of short duration or a shock of long duration. To the extent that disability is due to particular health problems, the durations of health conditions are largely predictable. An alternative would be to assume that disability shocks follow a three state (cid:133)rst-order Markov process, where the states are de(cid:133)ned by the severity of the work limitation. Individual behaviour conditioning on the realised duration and severity of the work limitation would then be identical. 25

Table 7: Consumption loss for disability disaggregated by severity and duration Dependent variable Log food spending (1) Log food spending (2) 0:009 0:065 Short-term mildly disabled (cid:0) (cid:0) (0:011) (0:016)(cid:3)(cid:3)(cid:3) 0:033 0:078 Long-term mildly disabled (cid:0) (cid:0) (0:009)(cid:3)(cid:3)(cid:3) (0:013)(cid:3)(cid:3)(cid:3) 0:040 0:080 Short-term severely disabled (cid:0) (cid:0) (0:023)(cid:3) (0:025)(cid:3)(cid:3)(cid:3) 0:043 0:061 Long-term severely disabled (cid:0) (cid:0) (0:016)(cid:3)(cid:3)(cid:3) (0:017)(cid:3)(cid:3)(cid:3) Interactions: 0:025 disability + married with working spouse (0:014) (cid:3) 0:040 disability + positive assets (0:014)(cid:3)(cid:3)(cid:3) 0:021 disability + labour force participation (0:015) 0:033 disability + no disability bene(cid:133)t (0:014)(cid:3)(cid:3) N 5,985 Controls: age, age squared, household size, education, sex, marital status, home ownership, time and regional dummies. Robust standard errors in parentheses. (cid:3) signi(cid:133)cant at 10%; signi(cid:133)cant at 5%; signi(cid:133)cant at 1%. (cid:3)(cid:3) (cid:3)(cid:3)(cid:3) 26

in receipt of incapacity bene(cid:133)t. As discussed in section 2, incapacity bene(cid:133)t becomes more generous as duration increases, and it is targeted at more severe conditions. This is evident in our regressions, as we (cid:133)nd that the long term severely disabled have higher food expenditure than the short term severely disabled, once we condition on receiving disability insurance. This may re(cid:135)ect the more generous bene(cid:133)ts being paid out to longer duration claimants. In addition, these long-term severely disabled have higher food expenditure than the long-term mildly disabled, possibly re(cid:135)ecting the targeting of bene(cid:133)ts towards more acute health shocks. However, the di⁄erences between coe¢ cients are only marginally signi(cid:133)cant. 5.3 Dynamics of Consumption Loss after Disability Onset Up to this point, our regressions have shown estimates for responses averaged over time for each individual in our sample. We now disaggregate the responses into three time categories to capture the dynamics of consumption changes associated with disability. The time categories we use are: the period of disability onset; three years after onset; and more than three years after onset. This approach is similar in nature to that followed by Meyer and Mok (2006) and Stephens (2001).11 Weshowthedynamicsofdeclinesinfoodexpenditureforindividualswithdi⁄erent severities of disability shock. In table 8, we report the regression without controlling for self-insurance in column (1), and with controls for self-insurance in column (2). As before, in both regressions we include a number of variables on observable characteristics. Those with a mild disability experience no fall in food expenditure at onset, with food expenditure 2-3% lower for subsequent periods. The severely disabled do see a 11Better panel data sources in the US allow these authors to analyse years by year dynamics, we do not have su¢ cient data to do this. 27

Table 8: The dynamics of consumption loss Dependent variable Log food spending (1) Log food spending (2) Mild disability: 0:018 0:070 -onset (cid:0) (cid:0) (0:011) (0:015)(cid:3)(cid:3)(cid:3) 0:021 0:070 -3 years after onset (cid:0) (cid:0) (0:009)(cid:3)(cid:3) (0:014)(cid:3)(cid:3)(cid:3) 0:032 0:081 -more than 3 years after onset (cid:0) (cid:0) (0:011)(cid:3)(cid:3)(cid:3) (0:015)(cid:3)(cid:3)(cid:3) Severely disabled: 0:041 0:071 -onset (cid:0) (cid:0) (0:021)(cid:3) (0:022)(cid:3)(cid:3)(cid:3) 0:053 0:079 -3 years after onset (cid:0) (cid:0) (0:017) (0:018) (cid:3)(cid:3)(cid:3) (cid:3)(cid:3)(cid:3) 0:025 0:048 -more than 3 years after onset (cid:0) (cid:0) (0:019) (0:020) (cid:3)(cid:3) Interactions: 0:025 -married with working spouse (0:014) (cid:3) 0:040 -positive assets (0:014) (cid:3)(cid:3)(cid:3) 0:021 -labour force participation (0:015) 0:034 -no disability bene(cid:133)t (0:014) (cid:3)(cid:3) N 5,985 5,985 Column (1) shows baseline results; column (2) includes controls for self-insurance. Other controls: age, age squared, household size, education, sex, marital status, home ownership, time and regional dummies. Robust standard errors in parentheses. signi(cid:133)cant at 10%; signi(cid:133)cant at 5%; signi(cid:133)cant at 1%. (cid:3) (cid:3)(cid:3) (cid:3)(cid:3)(cid:3) 28

fall in food expenditure at onset, this persists for three years after onset and then recovers. Meyer and Mok (2006) (cid:133)nd similar results for the US.12 They (cid:133)nd that individuals su⁄ering severe work-limiting conditions su⁄er a 6% food expenditure fall at onset. This worsens to a 15% fall and then to an 18% fall in years two and three respectively, with later years seeing a slight improvement in consumption. For those experiencing a mild condition these authors also (cid:133)nd no signi(cid:133)cant fall at onset. However, in contrast to our results, they (cid:133)nd little evidence of consumption falls post onset, possibly re(cid:135)ecting their choice not to exclude one-o⁄disablements as we do. As with the previous regressions, these e⁄ects do not take account of the availability of di⁄erent insurance mechanisms, and in the second column we report the results once we have controlled self-insurance. We (cid:133)nd that food expenditure is now 5-8%lower forboth disability types over all periods. Those individuals who only have the state insurance for support have a large and persistent fall in food expenditure, highlighting the imperfect insurance o⁄ered by disability bene(cid:133)t. 6 The Reform of 1995 In 1995, the UK government reformed incapacity bene(cid:133)t to reduce the generosity of bene(cid:133)ts. For example, Bell and Smith (2004) document how the generosity of incapacity bene(cid:133)t compared to unemployment bene(cid:133)t increased for all age groups in the 1980s and especially in the early 1990s, but that the generosity of incapacity bene(cid:133)t was scaled back substantially in 1995. This occurred primarily through reducing the pension entitlement of those on incapacity bene(cid:133)t. This scaling back of generosity, and the increased work-related requirements introduced by the Labour government in the (cid:147)Pathways to work(cid:148)programme, reduced the insurance provided by incapacity (invalidity) bene(cid:133)t. In this section, we test the extent to which this reduction in 12See results on food consumption in Table 12 (p.84). 29

insurance has led to greater consumption losses among the disabled. To do this, we split the sample by the date of onset of disability into those where onset occurred in 1995 or earlier and those where it occurred after 1995. We then reproduce in table 9 the consumption regressions reported in table 4 above. After 1995, disability was associatedwithconsumptionbeing7%lower, whereasintheearlierperiodthedecline was smaller and not statistically signi(cid:133)cant. Table 9: Consumption Loss before and after 1995 Dependent variable Log food spending Log food spending (pre-1995) (post-1995) 0:039 0:069 Disabled (cid:0) (cid:0) (0:029) (0:019) (cid:3)(cid:3)(cid:3) Interactions: 0:024 0:042 disabled + married with working spouse (cid:0) (0:026) (0:021) (cid:3)(cid:3) 0:081 0:031 disabled + positive assets (0:027) (0:022) (cid:3)(cid:3)(cid:3) 0:003 0:017 disabled + labour force participation (0:028) (0:022) 0:005 0:033 disabled + no disability bene(cid:133)t (0:030) (0:023) N 5,059 5,384 Controls: age, age squared, household size, education, sex, marital status, home ownership, time and regional dummies. Robust standard errors in parentheses. signi(cid:133)cant at 10%; signi(cid:133)cant at 5%; signi(cid:133)cant at 1%. (cid:3) (cid:3)(cid:3) (cid:3)(cid:3)(cid:3) 7 Conclusions The aim of this paper was to analyse consumption losses due to disability, and to explorehowe⁄ectivelyalternativeinsurancemechanismsmitigatesuchadversity. Our mainconclusionis that individuals receivingdisabilityinsurance, without anyformof self-insurance, have food consumption which is 7% lower than those without a work limitation. Since we are analysing food expenditures, which we would a priori expect 30

to respond less than other expenditure to income shocks, this suggests that insurance against a work-limiting health shock is fairly incomplete. We consider various forms of self-insurance, including savings, a working partner and own work. Each of these mitigates the food expenditure loss by between 2% and 4%. When we compare the group receiving disability bene(cid:133)t to those not receiving it, we (cid:133)nd that those receiving disability insurance have lower food expenditure. This suggests that individuals receiving disability insurance experience more severe work-limiting conditions than those not receiving bene(cid:133)ts, providing some support for the e⁄ectiveness of the screening mechanism onto state disability support. This is further supported by our (cid:133)nding that the receipt of disability insurance is correlated the severity of the condition. This result also implies a selection issue in interpreting our results as providing evidence of the bene(cid:133)t of disability insurance and, in particular, we do not know how far consumption would have fallen for those in receipt of incapacity bene(cid:133)t had they not been receiving the bene(cid:133)t. In our regressions, we have concentrated on the average e⁄ect of a health-limiting condition on food expenditures over time. However, during the period we analyse there have been a number of sizeable shifts in policy, particularly in 1995 and after Labour came to power. We present evidence that the reductions in generosity associated with these policy changes led to less insurance against consumption losses. References Anyadike-Danes M, and McVicar D, 2008, (cid:147)Has the boom in Incapacity Bene(cid:133)t claimant numbers passed its peak?(cid:148)Fiscal Studies, 29(4), 415-434 Banks J, Blundell R, Old(cid:133)eld Z, and Smith J, 2007.(cid:147)Housing Price Volatility and Downsizing in Later Life,(cid:148)NBER Working Papers 13496 31

Banks J, Kapteyn A, Smith J, and van Soest A, 2005.(cid:147)Work disability is a pain in the *****, especially in England, the Netherlands, and the United States,(cid:148)NBER Working Papers 11558 Beatty C, Fothergill S, and Macmillan R, 2000, (cid:147)A theory of employment, unemployment and sickness,(cid:148)Regional Studies, 34, 617-30 Bell B and Smith J, 2004, (cid:147)Health, disability insurance and labour force participation,(cid:148)Bank of England Working Paper, 218 Benitez-Silva H, Buchinsky M, and Rust J, 2004, (cid:147)How Large are the Classi(cid:133)cation Errors in the Social Security Disability Award Process?,(cid:148)NBER Working Paper 10219 Benitez-Silva H, Buchinsky M, Chan H M, Rust J, and Cheidvasser S, 1999, (cid:147)An empirical analysis of the social security Application, Appeal and Award Process,(cid:148) Labor Economics, 6, 147-78 BerthoudR,1998,DisabilityBene(cid:133)ts: AReviewoftheIssuesandOptionsforReform, York: York Publishing Services Bound J, 1989, (cid:147)The Health and Earnings of Rejected Disability Insurance Applicants,(cid:148)American Economic Review, 79, 482-503 Bound J, and Burkhauser R V, 1999, (cid:147)Economic analysis of transfer programs targeted on people with disabilities,(cid:148)in Ashenfelter O, and Card D, Handbook of Labor Economics, Volume 3 (eds.), Elsevier Bound J, Stinebrickner T R, and Waidmann T, 2006 (cid:147)Health, economic resources and the work decisions of older men,(cid:148)Working Paper, University of Michigan and NBER 32

BrowningM,andCrossleyTF,2008,(cid:147)Shocks,stocksandsocks: smoothingconsumption over a temporary income loss,(cid:148)Journal of the European Economic Association (forthcoming) BrowningM,andCrossleyTF,2001, (cid:147)Unemploymentbene(cid:133)tlevelsandconsumption changes,(cid:148)Journal of Public Economics, 80(1), 1-23 Burchardt T, 2000, (cid:147)The dynamics of being disabled,(cid:148)CASE paper 36 DeNardiM,FrenchE,BaileyJonesJ,2006,(cid:147)Di⁄erentialmortality,uncertainmedical expenses, and the saving of elderly singles,(cid:148)Working Paper Series WP-05-13, NBER Working Papers 12554 Disney R, and Webb S, 1991, (cid:147)Why are there so many long term sick in Britain?,(cid:148) Economic Journal, 101, 252-62 DWP, 2007a, (cid:147)Bene(cid:133)t expenditure tables,(cid:148)available at http://www.dwp.gov.uk DWP, 2007b, (cid:147)Bene(cid:133)t claimants tables,(cid:148)available at http://www.dwp.gov.uk Hall G, and Rust J, 1999, (cid:147)Econometric methods for endogenously sampled time series: the case of commodity price speculation in the steel market,(cid:148)Unpublished Manuscript Huddleston T, 2000, (cid:147)Explaining the growth in the number of people claiming Incapacity Bene(cid:133)ts,(cid:148)mimeo, Department for Work and Pensions, London MachinS,2003, (cid:147)Wageinequalitysince1975,(cid:148)inDickensR,GreggP,andWadsworth J, (eds.) The Labour Market under New Labour, Basingstoke: Palgrave Macmillan McVicar D, 2008, (cid:147)Why have UK disability bene(cid:133)t rolls grown so much?,(cid:148)Journal of Economic Surveys, 22(1), 114-139 33

Meyer B D, and Mok W, 2006, (cid:147)Disability, earnings, income and consumption,(cid:148) Harris school working paper 06.10 Parsons D O, 1980a, (cid:147)The Decline of Male Labor Force Participation,(cid:148)Journal of Political Economy, 88, 117-34 Parsons D O, 1980b, (cid:147)Racial Trends in Male Labor Force Participation,(cid:148)American Eccnomic Review, 70, 911-20 Stephens M, 2001, (cid:147)The long-run consumption e⁄ects of earnings shocks,(cid:148)Review and Economics and Statistics, 83(1), 28-36 34