Microsoft Word WFP Final May 2 2011.docx

Transcript

1 Cash by any other name? vidence on lab elling E from the UK Winter Fuel Payment Working Paper 10 /11 IFS Timothy K.M. Beatty L aura Blow Thomas F. Crossley Cormac O ’ Dea

2 Cash by Any Other Name? Evidence on Labelling from the UK Winter Fuel Payment a 1 Timothy K.M. Beatty University of Minnesota and b Institute for Fiscal Studies b Institute for Fiscal Studies Laura Blow b Thomas F. Crossley Institute for Fiscal Studies c University of Cambridge and d Koç University b Institute for Fiscal Studies Cormac O’Dea May 2011 Abstract: s that the labelling of cash tr ansfers or cash-equivalents (e.g. Standard economic theory implie child benefits, food stamps) should have no e ffect on spending patterns. The empirical radict this proposition. We st udy the UK Winter Fuel Payment literature to date does not cont iting sharp eligibility cr (WFP), a cash transfer to older households. Explo iteria in a regression discontinuity design, we find robust evidence of a behavioural effect of the labelling. On average households spend 41% of the WFP on fuel. If the paym ent was treated as cash, we oximately 3% of the payment on fuel. would expect households to spend appr Keywords: labelling, benefits, expenditure JEL codes: D12, H24 a b Department of Applied Economics Institute for Fiscal Studies University of Minnesota 7 Ridgmount Street 1994 Buford Ave London St Paul, MN USA WC1E 7AE, UK 55108 c d Faculty of Economics Department of Economics University of Cambridge Koç University Sidgwick Avenue Rumelifeneri Yolu Cambridge Sariyer 34450 Istanbul CB3 9DD, UK Turkey 1 This research was made possible by a grant from the Nuffield Foundation. Thanks to Sule Alan, Mike Brewer, nts. All errors are our own. Andrew Chesher, Dominic Curran an d Valérie Lechene for helpful comme

3 1 on Labelling from the UK Winter Cash by Any Other Name? Evidence Fuel Payment Abstract: Standard economic theory implie ansfers or cash-equivalents (e.g. s that the labelling of cash tr child benefits, food stamps) should have no e ffect on spending patterns. The empirical literature to date does not cont radict this proposition. We st udy the UK Winter Fuel Payment iting sharp eligibility cr iteria in a regression (WFP), a cash transfer to older households. Explo discontinuity design, we find robust evidence of a behavioural effect of the labelling. On the WFP on fuel. If the paym ent was treated as cash, we average households spend 41% of oximately 3% of the payment on fuel. would expect households to spend appr 1. Introduction Government transfers to households and individuals are sometimes given labels indicating that they are designed to support the consumption of a partic ular good or service. For example, many countries provide transfers ren and label them a to households with child “Child Benefit”. When such transfers are made in cash there is no obligation to spend all, or even any, of the payment on its ostensive purpose. Standard economic theory implies that the label of a particular transfer should have no bearing on how that transfer is ultimately spent since all income is fungible. The recipient of a e label is expected to transfer with a suggestiv react in exactly the same way as he would ha ve reacted had he been given a transfer of bel. The receipt of an in-kind tr equivalent value with a neutral la ansfer such as food stamps is similar as long as consumers are infra-marginal – i.e. for those whom consumption of the good in question is alre ady larger than the voucher am ount. Why then do governments label doing so makes redistribu transfers? Of course, one possibility is that tion more palatable to voting taxpayers. However, anothe r intriguing possibility is that standard economic theory is mistaken on this particular point, and spending patterns can be influenced by the labelling of cash or cash-equivalent transfers. In this pa per we provide novel evidence on the behavioural effect of labelling from the UK Winter Fuel Payment (WFP). The theoretical proposition th has been challenged. For at labelling is irrelevant example, Thaler’s (1990, 1999) framework of mental accounts is one mechanism through 2 which the labelling of a tran sfer might affect its usage. There is, though, very little previous empirical evidence to support the idea that th e labelling of a transfer payment matters. Kooreman (2000) and Blow, Walker and Zhu (2010) find evidence that additional child benefit differs from other income in its e ffect on household spending patterns among child 2 In the present context, income would be labelled according to its source, and so the Winter Fuel Payment account for spending on heating. would be allocated to a mental

4 2 UK respectively. Kooreman finds some evidence benefit recipients in the Netherlands and the (i.e. child benefit is spent on child-related goods ); in contrast, Blow, of a labelling effect Walker and Zhu’s results suggest child benef it is spent disproportiona tely on adult-related 3 . Edmonds (2002) also looks at child benef it payments (in this case amongst families goods ce of a labelling effect. It is important to note that plausible in Slovakia) and finds no eviden series variation in the real value of child identification in these studies rests solely on time benefit within household type. As we explain below, the design of the Winter Fuel Payment leads naturally to a regression discontinuity de sign. Moreover, it is no t possible in two-adult households to separately iden ild benefit income from the tify a labelling effect of ch alternative explanation that th e increase in the share of tota l household income received by the mother (child benefit is almost always paid to the mother) leads the change in spending patterns. That is, it could be w ho receives the money, rather than the label, that matters. This issue of intrahousehold allocatio n seems particularly important in the case of spending on children. Among single-mother households, for wh om these intrahouse hold considerations are not relevant, Kooreman finds an effect in the direction cons istent with labelling mattering, but which is not significantly different from zero at conventional levels. Similarly, Blow, Walker and Zhu find weaker result s for single-parent households. and more recently Whitmore (2002) look Turning to in-kind transfers, Moffitt (1989) at the effect of food stamps on consumption choices and fi nd no evidence that infra-marginal equivalent cash payment. consumers treat food stamps differently than an In contrast, Abeler and Marklein (2010) have recently compared in-kind grants and (unlabelled) cash grants in small laboratory and field experiments and find evidence against the funga bility of money in 4,5 those contexts. The WFP, which we study, is a universal an nual cash transfer paid to households 6 containing an individual aged 60 or over in the qualifying w eek of the relevant year. Its 3 This does not imply parents disregard their children’s welfare. The paper finds evidence that this spending benefit which suggests that parents are altruistic and effect comes from the unanticipated variation in child insulate their children from income variation. 4 First Abeler and Marklein show in a field experiment in a restaurant that beverage vouchers increase beverage consumption by more than a general voucher towards their total bill. The difference is statistically significant and larger than what might plausibly attributed to the small number of patrons for whom the transfers might be distortionary. They then show a similar effect with notional consumption of two goods in a laboratory experiment with students. 5 There is much better evidence that labelling of transfers between levels of government has an important effect on how the transferred funds are spen t. This is called the “flypaper eff ect”. See Hines and Thaler (1995). 6 In recent years the qualif ying week has been the third full week of September. Strictly speaking the WFP is paid to households where anyone is over the female state pension age. This age was 60 for the entire period for which we have data. However, betw een April 2010 and April 2046 it is planned that eligibility will rise gradually to the age of 68.

5 3 ion to spend any of it on household fuel. The payment is unconditional - there is no obligat November or December and during most of the payment is usually made in one lump sum in period covered by our data was worth £250 to households where the oldest person is aged between 60 than 80 and £400 where the oldest pers on is aged 80 or over (these values were reduced to £200 and £300 in the UK Budget of March 2011). The shar p age cut-off for receipt eligibility (the fact th at all households where there is somebody aged 60 or older at the cut-off date qualify for the benefit, and no households where all members are younger than 60 qualify) presents an excellent ssion discontinuity design to opportunity to employ a regre assess whether there is labelling effect associated with the WFP. Relative to small laboratory or field experiments, studying the WFP has the a dvantage that the WFP is an actual transfer received by a large population. Rela tive to studies of the child benefit, the WFP offers clean identification of a labelling effect. The WFP delivers additional disposable income but eligibility for the WFP, being based on age, is easily anticipated. Thus the ad ditional disposable income may not lead to a change in spending at the onset of eligibility. To the extent that the additional disposable income that the transfer delivers does lead to an increase in total expenditure, we would in spending on fuel (because fuel is a normal expect this to be associated with an increase good) and a decrease in the fuel budget share (because fuel is a necessity), regardless of tion in fuel spending and budget share with total whether the transfer is labelled. This varia expenditure is the “income effect” of standa rd demand theory. Thus, to provide unambiguous evidence of a labelling effect, we need to be able to distinguish a la belling effect from a alysis we embed our regression discontinuity standard income effect. Therefore, in our an timate an Engel curve for fuel expenditure design within an Engel curve framework. We es allowing for flexible effects of total expend iture on the fuel budget share, and we augment this with smooth age effects on preferences and a discontinuity at age 60. This discontinuity share of total expe nditure spent on fuel, captures the effect of payment of the WFP on holding total expe . The size of this shift is informative about the proportion nditure constant of the WFP that is spent on fuel above and be yond what would be expected from the usual “income effect” (as measured by the slope of the Engel curve.) We find statistically significant and robust evidence of a substantial labelling effect. We estimate that households spend an average of 41% of the WFP on household fuel. If the payment was treated in an equivalent manner to other increases in income we would expect households to spend only about 3% of th e payment on fuel. We conduct a number of ty that this robustness and falsification tests. We carefully test – and reject – the possibili

6 4 ties between consumption and le effect arises from non-separabili isure: the effect we observe cannot be explained by retirements around age 60 altering the demand for heating fuel. We also find a statistically significant effect for bot h singles and couples, confirming that this is not an intrahousehold allocati difference in the marginal on effect. Thus this dramatic propensity to consume fuel out of the WFP is ev idence that the name of the benefit (possibly combined with the fact that it is paid in November or December) has some persuasive influence on how it is spent. Understanding the effect that labels have is important for public policy. If labelling cash or cash-equivalents influences how they are spent, then governments might use labels innovatively to try and increase consumption of particular goods or se rvices that are thought 7 Of course, if the aim of a par ticular transfer is not to increase to be under-consumed. spending on any particular good or service but rather to carry out a straightforward redistribution of resources then an operative label might actually imply a utility cost – and care should be taken in naming benefits. This paper proceeds as follows. Section 2 give s a brief introduction to the data that we use (the Living Costs and Food Survey). Sectio n 3 outlines the empirical framework that we thods. Section 4 presents our apply to identify the labelling effects, and our estimation me estimates of the magnitude of the la belling effect. Section 5 concludes. 2. Data 8 The Living Costs and Food Survey (LCFS) is the primary source of household-level is a nationally representative a expenditure data in the UK. It nnual survey with a sample size of approximately 6,000 households. Surveys ar e conducted throughout the year. The survey consists of an interview and an expenditure diary. Each respondent is asked to keep a diary for a two-week period in which they record ev ery purchase that they make. In addition, an expenditure questionnaire asks them to record recen t purchases of more infrequently-bought items. The combination of the diary and que e construction of a stionnaire allows th comprehensive measure of household expenditu re. In the case of fuel spending, some information comes from the questionnaire (for example last payment of electricity on account) and some from the diary (for example slot meter payments). Total spending on fuel includes gas and electricity payments, and the purchase of coal, coke and bottled gas for 7 Because labels do not impose constraints, this would be very much in the spirit of Thaler and Sunstein’s (2008) “paternalistic libertarianism”. 8 The LCFS was known as the Expenditure and Food Survey (EFS) between 2001 and 2007 and previous to that was known as the Family Expenditure Survey (FES).

7 5 some electricity and gas use may central heating. Clearly have been for cooking, lighting etc and not heating, but it is not possible to separate this out. In addition to these measures, the LCFS records detailed income, demographic and socio-economic information on respondent households. from the years 2000 through 2008. The nominal In our main analysis, we pool data able over this period, with the main rate (paid at age 60) value of the WFP was fairly st varying between £200 and £250 per year. In some an alysis (to be descri bed below) we also use a second tranche of data covering the years 1988 through 1996. These data predate the introduction of the WFP in 1997. We do not us e data from the years 1997 through 1999. In this period the WFP existed, but was much less generous than it is currently. The sample that we use is comprised of single men and couples without children in which the male member of the couple is older. We exclude all households in which the oldest member of the household is less than 45 y ears old. We exclude si ngle women and couple households in which the oldest member is a woman because for such households, eligibility for the WPF occurs at the same time as the wo man becomes eligible for the state pension. Table 1 presents summary statistics for this sa mple divided between eligible households and households in which the oldest member is below the age cut-off. [TABLE 1 ABOUT HERE] For both eligible and ineligible households , we present summary statistics for the entire subsample, and for the poorest quartil e of households (as de termined by household relative to the average, poorer total expenditure). Note that households spend less on fuel absolutely, but spend a larger share of th eir budget on fuel. Fuel is a normal good, and a necessity. These facts are well known, but they play an important role in our empirical design, which we turn to next. 3. Empirical Framework and Estimation Households where the eldest member turns 60 before the qualifying week are eligible for the WFP and households where the eldest member turns 60 after the qualifying week are not. This sharp eligib ility criterion suggests estimating the effects of the WFP using a regression discontinuity design (RDD). Take up of the WFP is very high, and so the sharp 9 eligibility criterion can be considered a sharp receipt criterion. 9 The rate of take-up was above 90% in each year since 2003 - the first year our data allows us to estimate it.

8 6 straightforward: households immediately The intuition behind an RDD approach is holds immediately above the cut-off would below the cut-off provide evidence on how house have behaved had they not rece ived the transfer. The identifying assumption is that, in the absence of the transfer, expenditures vary co ntinuously with the forcing variable, age, erences and budgets evol ve smoothly with age. implying that, for the sample we consider, pref thus attributable to the aver age effect of the WFP (at age Any discrete change at age 60 is 10 Age has previously been used as the exogenous forcing variable in regression 60). discontinuity designs. See for example: Edmonds et al. (2005), Card et al. (2008), Carpenter and Dobkin (2009) and Lee and McCrary (2009). Testing For Labelling Effects in an Engel Curve Framework Receipt of WFP might lead ending simply because of a to an increase in fuel sp standard income effect. In our analysis we ne ed to distinguish a labelling effect from an income effect and to assess whether the WFP is allocated differently to how an unlabelled transfer would be allocated. Th erefore, we embed a regression discontinuity design within an Engel curve framework. If households on eith er side of the elig ibility criteria spend significantly different shares of expenditure on fuel, holding total expen diture constant , this of a labelling effect. would be direct evidence In standard demand analysis, Engel cu rves measure the re lationship between nditure as total expe household spending on a good and total household expe nditure increases. A common empirical specification of Engel curves relates budget shares to the logarithm of mal good so as the level of tota l expenditure rises we would total expenditure. Fuel is a nor expect fuel expenditure to rise. y, we would expect it to rise Because fuel is also a necessit less quickly than total expend iture, and so the budge t share should fall. These are standard income effects. Thus, an increase in fuel spe nding, or a decrease in the fuel budget share, with receipt of the WFP might si mply represent a standard move along the Engel curve – i.e. an income effect; this is illustrated by the m B in Figure 1, where ove from point A to point the Engel curve is presented in share form. In contrast, if there is a labe lling effect, when a household receives a labelled transfer, they will shift off this Engel cu rve, as illustrated in Figure 1 by the move from point B to point C. 10 In principle we could also search for an effect at age 80, at which point the WFP becomes more generous. However, in the LCF age has been topcoded at 80 since 2002 which means that we are unable to implement the RDD around age 80.

9 7 [FIGURE 1 ABOUT HERE] To test for a labelling effect, while allowing for standard income effects, we estimate a function of total expenditure. We allow Engel curves which relate budget shares to preferences to evolve continuous ly with the forcing variable, age of the oldest household , by including polynomials in age. We augm ent this empirical specification with a A member, i , for WFP eligibility. This variable captu res any discontinuity in the way that dummy, D i budget shares vary with age, conditional on total expenditure (and other covariates). We labelling the transfer. E ligibility is related to attribute any such discontinuity to the effect of 11 ሾ ሿ ሾ ሿ 1 ൌ1 As per Lee and Lemieux ܣ . ൒60 is the indicator function. where age by ܦ ௜ ௜ 2 A 60 − (2010), we interact and − 60 A with program eligibility to allow the slope and () () i i curvature of the regression line to differ on either side of the eligibility cut-off. Finally, we include a number of covariates, Z , to increase the precision of the regression discontinuity i estimator and to capture variation in relative prices. In all specifications, these include d area/year interactions. In several specifications we also household size, month, area, year an include employment (of head and, where releva nt, spouse), housing tenure, number of rooms and education controls. on discontinuity Engel curve specification, Hence, in complete form, our regressi using quadratic terms in age, can be written: 22 − + ⋅ − + ⋅ − − + 60 60 60 60 ατ =+ + β wDA A DADA β β β ()() () () ki i i i i i i i 4 3 12 TT Z e δγ + + +⋅ fX () iii (1) e is an independent (and possibly heteroskedastic) disturbance term and, where the dependent variable is the budget share of good k , and E = ==− = lim [ | 60, , ] lim [ | 60, , ] τ Z zX x = == wA Z zX x EwA provides a local estimate of the kk AA 60 ↓↑ 60 holding total expenditure constant. We estimate effect of the WFP on budget shares at age 60, this model (and all subsequent models unless ot herwise stated) using least squares and report robust standard errors. 11 ibility reference week has been in September. Because the LCF collects Note that in recent years the elig information on age at the time of interview, there is some risk of misclassifying households interviewed in October through December as being elig ible, when they were not . To this end, we follo w Lee and Card (2010) and adjust the discontinuity to reflect the probability that that the oldest member of the household was 60 in the previous September and were thus eligible to receive the winter fuel payment. In pr actice, households in which the oldest member is 60 and are observed in October receive a weight of 11/12, if they are observed in November they are assigned a weight of 10/12, and so on. Every household with a person aged 61 and above simply has a weight of 1.

10 8 ling effect on the budget share, if any, is This specification imposes that the label 12 We will test this specification below, and in independent of the level of total expenditure. the appendix, we lay out a more general specification which nests equation (1). f ) X ( In results presented below, we specify to be a quadratic function of the natural 13 logarithm of total expenditure, but results are robust to more flexible specifications. Note that the total expenditure variables are also interacted with year dummies; within the constraints imposed by theory, we want to allow the form of the Engel curves we estimate to be quite general and so we allow the slope (as well as the intercept) of the Engel curve to change as relative prices change. This is importa nt to ensure that the discontinuity effect we estimate is not picking up changes in the shape of the Engel curve over time that we have not allowed for. We now turn to possible threats to the vali dity of this research design and how we deal with them. Measurement error One possible concern is that measurement error in household expenditure could bias our estimate of the effect of WFP. In gene ral, measurement error in one variable can a simple example with classical potentially bias the estimate of all regression coefficients. In measurement error where the only regressors are log expenditure and WFP receipt, the bias on the WFP coefficient would have the same sign tween log expenditure as the relationship be the bias would actually and the fuel share, which is negative, and so be downwards (against finding a labelling effect). However, we cannot be sure that this w ould be the case in our as a check, we follow more complicated specification. Therefore, standard practice in demand analysis and instrument tota l expenditure with household income. Employment Effects From 1988 onwards individuals aged 60 or over have been entitled to a benefit, the name and exact details of which have changed, but which is essentially a pensioner minimum income guarantee (i.e. a minimum income guarant seek work). From ee without obligation to 1988 to 1999 this was called Pensioner Income Support, from 1999 to 2003 it was known as the Minimum Income Guarantee, and in 2003 this was replaced with Pension Credit. For the rest of this paper we will refer to this benefit as the Minimum Income Guarantee (MIG). 12 Of course, this specification implies that the effect, in any, on pounds of fuel expenditure varies with the level of total expenditure. 13 Engel curves relating budget shares to a quadratic function of the natural logarithm of total expenditure are the basis of the well known Quadratic Almost ideal Demand System (QuAIDS) of Banks, Blundell and Lewbel (1997).

11 9 where age 60 brings only eligibility for WFP; Therefore, note that we do not have a period from 1988-1996 we have the MIG alone and fr om 1997-2008 we have the MIG plus WFP. Whilst we would not expect the MIG to have a labelling effect, it might have a labour market participation effect, and, if consumption is not separable from leisure, this in turn will have an effect on spending patterns. Specifical ly, when a working individual turns 60, they become entitled to the MIG and they might pref er stopping work and receiving the MIG to carrying on in employment. But dropping out of the labour market might influence spending y might heat their home more and now at home for more of the da patterns; someone who is therefore have higher fuel spending. It might be that controlling fo r observable labour market stat us is enough to deal with this issue, and among our specification tests we include employment and self-employment of household and (where there is one) the dummies and hours of work for both the head ebo period allows an additional check on spouse. However, using 1988-1996 as a plac market effect of th e MIG. Estimating an whether our results are contaminated by the labour RDD on a pre-program period as a falsification test is normally good practice (see, for example Lemieux and Milligan (2008)), but here it is particularly important because the potential confounding of the WFP effect by the MIG. We proceed by pooling data from the period when only the MIG was paid (1988- ) with the period in which both th e MIG and the WFP were paid (2000- ܶ 1996, denoted ଵ ܶ 2008, denoted M, our Engel curve specification ). Denoting eligibility for the MIG by ଶ becomes: 22 − +⋅ +⋅ − − − + =+ + + β β 60 60 60 60 β β ατ λ Age w D M Age Age D Age D ()()()() k 4 3 12 2 TT 60 60 + + M Age Age f X Z e − + ⋅ M +⋅ − + ⋅ ββδγ ()()() 56 Note that here the MIG eligibility dummy member of the household M is one if the oldest was over 60 in the reference week, while the WFP eligibility dummy is now equal to one only if the oldest member of the house hold was over 60 in the reference week and the observation is drawn from period ܶ ion between age and period). (that is, it is an interact ଶ The coefficient on the MIG eligibility dum my measures any discontinuity in the way expenditure patterns vary with age in the period prior to the introduction of the WFP. Thus a significant effect would falsify the assumption that preferen ces evolve continously with age. The coefficient on the WFP eligibility dummy,

12 10 | τ x Z X lim [ | 60, , ] lim [ z 60, , ] E w age = z x Z X = == = ==− E w age kk } { ↓↑ 60 60 age age T 2 [ | 60, , ] lim [ | 60, , − = == = ==− Ew age X xZ z Ew age X xZ z lim ] kk {} ↓↑ 60 60 age age T 1 age effect of the WFP is our “differenced-RDD” estimate of the aver on budget shares at age 60, net of any labour market effect at age 60. Analysis by sub-group in equation (2) measures the average effect of the The discontinuity captured by τ n does not allow the e WFP at age 60; that is, our base specificatio ffect to vary by any household characteristics. Rather than imposing any additional structure, we investigate this further by splitting our sample according to some characteristics and testing for equality of which we split our sample are income the WFP effect across groups. The variables on our sample the latter means between single quartile, season and household structure (within men and couple households). Additional Robustness Checks gns can be sensitive to the choice of the range of the Regression discontinuity desi the age of the oldest household member. In forcing variable included in the regression, here ds located immediately on either side of the principle, one would like to compare househol potential discontinuity, but in practice sample size considerat ions prevent this. Our basic specification uses a window of fifteen years on eith er side of the discontinuity (45-75). As a robustness check we re-estimate with a win dow of ten years on either side of the discontinuity (50-70). Finally, we conduct a further falsification te st. We rerun our main analysis but with 14 rather than 60. Under the maintained assumptions of the regression cut-offs at 55 and 66 discontinuity design we should no or shares) at these age cut- t find discontinuities (in levels offs. 4. Results Testing For Labelling Effects in an Engel Curve Framework Table 2 shows the results of our Engel cu rve estimation. The first column of the Table, specification 1, gives our baseline results. We find a positive, statistically significant discontinuity effect for the fuel share a nd no significant effect for any other good. We interpret this effect on the fuel share, holding total expenditure constant, as a labelling effect. 14 Note we use 66 rather than 65 as 65 was the state pe nsion age for men during the period for which we have data.

13 11 ng suggest a negative effect; the budget The point estimates for food and clothi e effect on fuel spending must be offset by constraint of course implies that the positiv reductions elsewhere. variables for education, employment and In column (2) we add additional control and in column (3) we vary the age window housing tenure and number of rooms in the home, used in estimation. The positive effect on the fu el share is robust across these specifications. The negative effect on the food and clothing shar es become statistically significant at the 10% and 5% level, respectively, when we narrow the age window. [TABLE 2 ABOUT HERE] In Table 3 we report the results of additional specification checks for the fuel share. In column (1) we instrument for total expenditu re with household income to account for the possibility of measurement error in total ex penditure. This has almost no impact on the estimated labelling effect. In column (2) we report the results of es timating our “differenced-RDD” specification on pooled data from 1988-1996 and 2000-2008. This is therefore the average effect of the of any employment WFP on budget shares at age 60, conditional on total expenditure net effect at age 60. Note that the estimate here is larger than our baseli ne estimate, and though significant at the 5% level. Th less precisely estimated, is still is suggests that the labelling od is not an empl oyment effect. effect that we find in the 2000-2008 peri [TABLE 3 ABOUT HERE] Our basic specification imposes that the labe lling effect on budget sh ares, if present, is unrelated to the level of total expenditure and to any other variable. In Table 4 we report the results of relaxing this assumption and allo wing the effect to vary by quartile of total expenditure, by season, and by household type. Mo stly the coefficients are not precisely estimated, which is to be expected given the now much smaller sample sizes. In none of the three divisions can we reject the null that the coefficients are the same across the groups. Features to note are that th e point estimates in column (1) suggest that the effect on shares is larger for poorer households. This does not mean, though, that the absolute labelling ) varies this much; a larger share shift at lower total effect (on pounds of expenditure

14 12 ng effect as a smaller shift at higher total expenditure could translate into a similar spendi expenditure. We will elaborate on this in the discussion below. The WFP differs from child benefit in that there is no compelling reason to believe that its effect on spending patterns works thr ough the intra-household distribution of income receipt. First, as note d above, there is reason to think that the intra- household distribution of income receipt is particularly important in the case of spending on children. Second, in the always older. Thus at the eligibility threshold sample of couples we study the male member is for WFP, only the male is eligible and when only one member of a household is eligible for 15 This means that, when implemented on our WFP, the transfer is paid to that member. r, our regression discont sample of couples in which the husband is olde inuity design studies the effect of a labelled transfer to husbands . In the birth cohorts we study husbands were the primary earners and it is implausible that this gnificant effect on the £250 transfer had a si influence those husbands had over household spe nding patterns. Despite these considerations, it is reassuring to see, in column (3), that the la belling effect is still si gnificant when we split ly more so for single men our sample into single men a nd couples, and indeed marginal despite the much smaller size of this group relativ e to couples. This confirms that the effect , instead, an in we find is indeed a labelling effect and not tra-household effect. The point estimate for single men is larger than for couples (although, as st ated, not significantly e total expenditure of this sample of single different from each other) but, again, the averag men is much lower than the couples sample. . [TABLE 4 ABOUT HERE] Table 5 presents the results of our falsific ation tests. Columns (1) and (2) of Table 5 report tests for discontinuities in the relationship between age and fuel budget share at ages 55 and 66. Column (3) is the complement to colu mn (2) of Table 3. Here we report estimates of a discontinuity at age 60 in the period before the WFP wa s introduced (1988-1996). In all three cases, we find no effect. [TABLE 5 ABOUT HERE] 15 Where both members of a couple are eligible for the WFP half of the amount is paid to each member. However, for our sample, this is not relevant in at th e eligibility threshold, because it is the husband that qualifies initially.

15 13 WFP eligibility on the budget share of fuel, To summarise, we find a positive effect of age in a continuous erences to evolve with conditional on total expenditure and allowing pref ongly statistically significant and robust across alternative fashion. The effect is str specifications. Because of the ve ry high take-up of this transfer among eligible households, the effect of eligibility is for all intents and pu rposes also the effect of receipt. We attribute this effect to the labelling of this transfer. A series of falsification tests failed to contradict our , we find no evidence of a confounding of the identifying assumptions, and in particular labelling effect with empl oyment effects around age 60. Discussion We can translate the magnitudes in the table into spending changes as follows. Ignoring other covariates for simplicity, if ݔ ௞ ሺ ሻ ൌ ݂ൌ ݔ ݓ ௞ ܺ then ݓ߲ ݔ߲ ௞ ௞ ൌ ݓ൅ܺ ௞ ߲ܺ ߲ܺ so if households rece ive a transfer of ݌݂ݓ then the slide along the Engel curve starting from ܺ total budget Figure 1) is approximately (the move from A to B in ݓ߲ ௞ ݓ൅ܺ ݌݂ݓ ൰ ൬ ௞ ߲ܺ ngel curve measured in percentage points of and if our estimate of the movement off the E ߬ budget share (the move from B to C in Figure 1) is the labelling effect , then the estimate of measured in pounds of expenditure is approximately. ሺ ሻ ݌݂ݓ൅ܺ ߬ ሺ2ሻ With the results from, say, specification 2 in Table 2 our estimate of the slide along the Engel curve for someone with the aver of 0.0613 and total budget age fuel share in 2008 g a transfer of £250 a year of around £308 per week receivin (so just under £5 a week) is £0.128 with a standard error of 0.010 and a 95% confidence interval around this point estimate of £0.108 to £0.148. Our estimate of the la belling effect is £1.818 with a standard error of 0.623 and 95% confidence interval of £0.600 to £3.037. In other words, if there was no labelling effect an average household would spend around 3% of a small transfer on fuel. We estimate an additional labelling effect of 38% (with a confidence in terval of 12% to 63%) so that the overall marginal propensity to sp end on fuel associated with the WFP is around 41%.

16 14 ling effect depends on the estimated size of Equation (2) shows that the absolute label the discontinuity and on total household expenditure. Therefore, the different sh ifts estimated by expenditure quartile translate into relatively similar point estimates of additional labelling effects of £1.857, £1.410, £1.475 and £1.446 respectively (although we state again that a test of equality of the WFP coefficient or of the ab solute labelling effect is not rejected). 5. Conclusion This paper asks whether labelling an unconditi onal cash transfer has any effect on the way in which recipients spend it. In other words, does calling the £250 that most elderly households in the UK receive in November / December a “Winter Fuel” payment make any difference? Sharp differences in the eligibility requirements allow us to use a regression nditure changes on receipt of the benefit. We discontinuity design to examine how fuel expe find a substantial and robust labelling effect . Our estimate of the (average) marginal unlabelled income is approximately 3%. On propensity to spend on household fuel out of average, we find recipient households exhibit an additional marginal propensity to spend on 12% and 63%, and so the combined effect is household fuel out of the WFP of between about between 15% and 66%. The interpretation of this is straightforward: if households are given an unconditional and neutrally-nam ed cash transfer of £100 they would be expected to spend approximately £3 on household fuel. If they ar e given an unconditional cash transfer called the Winter Fuel Payment in the middle of wint er we estimate that th ey will spend between £15 and £66 on fuel (our point estimate is £41). Ov erall, our evidence implies that the label of this particular transfer ha s a critical impact on the behavi oural response displayed by those who receive it.

17 15 References Abler, Johannes and Marklien, Felix, 2010. “Fungability, Labels and Consumption.” University of Nottingham, Working Paper. Banks, James, Blundell, Richard and Arthur Lewbel, 1997. “Quadratic Engel Curves and The Review of Economics and Statistics , vol. 79(4), pages 527- Consumer Demand,’’ 539. Blow, Laura, Walker, Ian and Zhu, Yu, 2010, “Who Benefits from Child Benefit?” Economic , no. doi: 10.1111/j.1465-7295.2010.00348.x. Inquiry Card, David, Dobkin Carlos and Maestas, Nico le, 2008. “The Impact of Nearly Universal Insurance Coverage on Health Ca re: Evidence from Medicare,” American Economic Review , vol. 98(5), pages 2242–58. Carpenter, Christopher & Dobkin, Carlos, 2009. “The Effect of Alcohol Consumption on Mortality: Regression Discontinuity Evid ence from the Minimum Drinking Age,” American Economic Journal: Applied Economics , vol. 1(1), pages 164-182. ffect for child benefits: evidence from a Edmonds, Eric, 2002."Reconsidering the labeling e transition economy," Economics Letters 3), pages 303-309. , Elsevier, vol. 76( r, 2005. “Rearranging the Family? Household Edmonds, Eric V., K Mammen and D. Mille The Journal of Human Composition Responses to Large Pension Receipts,” vol. 40(1), pages 186-207. Resources, ., 1995, “Anomalies: The Flypaper Effect”, Journal of Hines, James R., and Thaler, Richard H Economics Perspectives , vol. 9(4), pages 217-226, Fall. Kooreman Peter, 2000."The Labeling Eff ect of a Child Benefit System," American Economic Review , vol. 90(3), pages 571-583. Lee, David S. and Card, David, 2008, “Regress ion Discontinuity with Specification Error,” Journal of Econometrics , vol. 142(2), pages 655-674. Lee, David S. and Lemieux, Thomas, 2010. “Regression Discontinuity Designs in Economics,” Journal of Economic Literature . vol. 48(2) , pages 281-355. Lee, David S. and McCrary, Ju stin, 2009. “The Deterrence Effect of Prison: Dynamic Theory and Evidence”, Working Paper 550, Princet on University, Department of Economics, Industrial Relations Section.

18 16 Lemieux, Thomas and Milligan, Kevin, 2008. “In centive effects of so cial assistance: A regression discontinuity approach,” Journal of Econometrics , vol. 142(2) pages 807- 828. Moffitt, Robert, 1989. "Estimating the Value of an In-Kind Tr ansfer: The Case of Food Stamps," Econometrica , vol. 57(2), pages 385-409. udge: Improving Decisions About Health, Thaler, Richard and Sunstein, Cass 2008. N Wealth, and Happiness. New Haven, CT: Yale University Press. Thaler, Richard H., 1990. “Saving, fungibility and mental accounts”, Journal of Economic Perspectives , vol. 4, pages 193-205. Thaler, Richard H., 1999. “Mental accounting matters”, Journal of Behavioral Decision Making , vol. 12(3), pages 183–206. Whitmore, Diane, 2002."What Are Food Stamps Worth?," Working Papers 468, Princeton University, Department of Economi cs, Industrial Relations Section.

19 17 Appendix This specification of equation (2) imposes that the labelling effect, if any, measured in share nditure. A more general formulation which nests form, is unrelated to the level of total expe ates for the moment, write the budget share of equation (2) is as follows. Ignoring other covari k , good w , as k π , =+ + wfXgAX hA () () () i i i ki i i π is the WFP measured in pounds and where , gAX is some function of age and total () i ii expenditure. The null hypothesis of no labelling effect corresponds to . Taking ,0 = gAX () ii = π 0 a (first order) Tayl or approximation of we obtain around π , fX gAX + () () i iiii fX ∂ () i ,, ππ  fX gAX ++ fX gAX () () () () iii i iii i X ∂ i , γπ fX AX ≡+ ) () ( iiii then we can approximate Noting that we can always write + =− D h A Dh A hA 1 )() ()( () iiiii 12 the more general model above by: γπ ⎡⎤ +− + =+ hA , h A hA AX w fX D ( ) ) () () () ( i i ki 21 1 i i i i i i ⎣⎦ X We do not have sufficient data to estimate properly how and AX γ might vary with , () i ii π so, as in addition there is very little variation in , we estimate an average effect, replacing i AX is some constant. The only ge , λ neral thing we are prepared γ π with where γλ A () () i ii i 60 hA is that to assume about and hence the only age at which we can = 60 hh () () () i 12 (this separately identify −= hA hA A γλ from 0 is at age 60 where − hA hA () () () () () ii 21 ii i 21 is basically a restatement of the assumptions underlying the regr ession discontinuity design as applied to our particular case.)

20 18 Figures and Tables Figure 1: Engel Curve with Income Effect and Labelling Effect

21 19 Table 1. Descriptive Statistics – weekly means (£ and shares) Ages 45-60 WFP Eligible All Poorest All Poorest Quartile Quartile Income 531.35 199.63 405.25 244.95 434.59 124.47 362.42 151.62 Total expenditure Fuel 18.37 11.96 18.79 13.96 Food 44.14 24.74 47.23 34.32 Clothing 13.37 2.01 11.95 3.33 Leisure Goods 14.04 3.67 13.09 5.32 0.081 Fuel Share 0.046 0.084 0.055 Food Share 0.210 0.162 0.232 0.128 Clothing Share 0.033 0.018 0.036 0.025 Leisure Goods Share 0.039 0.037 0.044 0.042 Sample Size 4423 760 6326 1746 Data: Living Costs and Food Survey (LC FS), 2000-2008. Single men and couples without children in which the male is older. The LCFS was known as the Expenditure and Food Survey (EFS) between 2001 and 2007 and previ ous to that was known as the Family Expenditure Survey (FES).The poorest quart ile is defined by total expenditure.

22 20 Table 2. RDD estimates. Effects of WFP on budget Shares (conditional on total expenditure) (3) (2) Shares (1) OLS OLS OLS Fuel 0.0057** 0.0058** 0.0062* (0.0020) (0.0020) (0.0025) Food -0.0034 -0.0032 -0.0103* (0.0038) (0.0038) (0.0048) Clothing -0.0035 -0.0039 -0.0074† (0.0032) (0.0032) (0.0040) 0.0032 0.0032 0.0057 Leisure Goods (0.0031) (0.0031) (0.0040) Age Window 45-75 50-70 45-75 Data Period 2000-2008 2000-2008 2000-2008 Additional Controls Y Y Notes: ng controls: (the natural logarithm of) 1. The base specification includes the followi total expenditure and its square; year dummies, region dummies and their year dummies and the total expenditure interactions; interactions between the tural logarithm of) household size. The variables; month dummies; and (the na additional controls are employment (of th e head, and where relevant, the spouse), housing tenure, number of rooms and education controls. 2. The age window pertains to the oldest person in the household. 3. Robust standard errors ar e given in parentheses 4. † = significant at 10% leve l, * = significant at 5% leve l, ** = significant at 1% level, *** = signifi cant at 0.1% level

23 21 Table 3: Further Specification Checks Effects of WFP on Fuel Budget Share (Conditional on Total Expenditure) (2) (1) OLS IV Expenditure Quartile: All 0.0056** 0.0066* (0.0020) (0.0031) Age Window 45-75 45-75 Data Period 2000-2008 2000-2008 and 1988-1996 Additional Controls Y Y MIG Y Notes: 1. The base specification includes the followi ng controls: (the natural logarithm of) dummies, region dummies and their total expenditure and its square; year interactions; interactions between the year dummies and the total expenditure variables; month dummies; and (the na tural logarithm of) household size. The additional controls are employment (of th e head, and where relevant, the spouse), housing tenure, number of rooms and education controls. 2. The age window pertains to the oldest person in the household. 3. Robust standard errors ar e given in parentheses. 4. † = significant at 10% leve l, * = significant at 5% leve l, ** = significant at 1% level, *** = signifi cant at 0.1% level

24 22 Table 4. RDD estimates for different sub-groups Effects of WFP on budget Shares (conditional on total expenditure) (3) (2) (1) Household Type: Season: Expenditure Quartile: t s 0.0135† Single men 0.0105* 1 Winter 0.0061 (0.0076) (0.0046) (0.0052) n d 0.0054 Spring 0.0068 Couple 0.0037† 2 (0.0035) (0.0045) (0.0019) d r Summer 0.0080* 3 0.0037 (0.0028) (0.0040) t h 0.0020 Autumn 0.0038 4 (0.0023) (0.0037) F(1,10433) F(3,10129) F(3,10165) F-test of Equality = 0.21 = 1.55 = 0.81 (0.49) (0.21) (0.89) (p-value) 45-75 45-75 45-75 Age Window 2000-2008 Data Period 2000-2008 2000-2008 Additional Controls Y Y Y Notes: ng controls: (the natural logarithm of) 1. The base specification includes the followi total expenditure and its square; year dummies, region dummies and their interactions; interactions between the year dummies and the total expenditure variables; month dummies; and (the na tural logarithm of) household size. The additional controls are employment (of th e head, and where relevant, the spouse), housing tenure, number of rooms and education controls. 2. The age window pertains to the oldest person in the household. 3. Robust standard errors ar e given in parentheses. 4. † = significant at 10% leve l, * = significant at 5% leve l, ** = significant at 1% level, *** = signifi cant at 0.1% level

25 23 Table 5. Falsification Tests. Effects on Fuel Budget Share (Conditional on Total Expenditure) (3) (2) Shares (1) OLS OLS OLS Discontinuity at Prior to Policy Discontinuity at 5 66 55 Introduction Fuel 0.0029 0.0000 -0.0016 (0.0024) (0.0022) (0.0023) 6 6 45-75 Age Window 45-75 45-75 Data Period 2000-2008 2000-2008 1988-1996 Additional Controls Y Y Y Notes: 1. The base specification includes the followi ng controls: (the natural logarithm of) dummies, region dummies and their total expenditure and its square; year year dummies and the total expenditure interactions; interactions between the variables; month dummies; and (the na tural logarithm of) household size. The e head, and where relevant, the spouse), additional controls are employment (of th and education controls. housing tenure, number of rooms The age window pertains to the oldest person in the household. 2. 3. Robust standard errors ar e given in parentheses. 4. † = significant at 10% leve l, * = significant at 5% leve l, ** = significant at 1% level, *** = signifi cant at 0.1% level 5. To avoid issues around the male reti rement age of 65 we chose 66, although results for age 65 are similar 6. Rebalancing the sample (for example changing the age window around 55 to be 40-70) also yields insignificant results.

Related documents

10749

10749

Cash transfers: what does the evidence say? A rigorous review of programme impact and of the role of design and implementation features Francesca Bastagli, Jessica Hagen-Zanker, Luke Harman, Valentina...

More info »
Government Finance Statistics Manual 2014

Government Finance Statistics Manual 2014

MANUAL GOVERNMENT FINANCE STATISTICS MANUAL 2014 2014 INTERNATIONAL MONETARY FUND

More info »
final report final web

final report final web

ACCESS TO CASH REVIEW Final Report March 2019

More info »
Applying IFRS 17

Applying IFRS 17

Applying IFRS 17 A closer look at the new Insurance Contracts Standard May 2018 1

More info »
Maternity and paternity at work – Law and practice across the world

Maternity and paternity at work – Law and practice across the world

Law and practice across the world Maternity and paternity Law and at work practice across the world Maternity and paternity at work –

More info »
Building Excellent Schools Today (BEST) Annual Report

Building Excellent Schools Today (BEST) Annual Report

Building Excellent Schools Today (BEST) Annual Report Submitted to: Senate Education Committee Senate Finance Committee House Education Committee House Finance Committee Capital Development Committee ...

More info »
PM VOL VIII.Doc

PM VOL VIII.Doc

POSTS AND TELEGRAPHS MANUAL VOLUME VIII POST OFFICE AND RAILWAY MAIL SERVICE SUPERVISING OFFICERS THIRD PRINT (REPRINT) ST ND LIST OF CORRECTIONS DATED 1 - 7 - 1976 TO 2 (1 LIST OF CORRECTIONS DATED 1...

More info »
2018 Investment Company Fact Book

2018 Investment Company Fact Book

2018 Investment Company Fact Book A Review of Trends and Activities in the Investment Company Industry 58th edition www.icifactbook.org

More info »
Fannie Mae 2017 Form 10 K

Fannie Mae 2017 Form 10 K

UNITED ST ATES SECURITIES AND EXCHANGE COMMISSION Washington, D.C. 20549 Form 10-K O SECTION 13 OR 15(d) ANNUAL REPORT PURSUANT T ACT OF 1934 OF THE SECURITIES EXCHANGE For the fiscal year ended Decem...

More info »
The use of cash by households in the euro area

The use of cash by households in the euro area

Occasional Paper Series Lola Hernández Henk Esselink, The use of cash by households in the euro area 2017 / November 201 No : Disclaimer This paper should not be reported as representing the views of ...

More info »
Reply Annual Report 2017

Reply Annual Report 2017

reply annual financial report 2017 REPLY ANNUAL FINANCIAL REPORT 2017

More info »
europolcik (1)

europolcik (1)

FINANCIAL INTELLIGENCE GROUP WHY IS CASH L KING? STIL Eisenhowerlaan 73 A STRATEGIC REPORT ON THE USE 2517 KK The Hague OF CASH BY CRIMINAL GROUPS The Netherlands AS A FACILITATOR FOR MONEY LAUNDERING...

More info »
Ademaetal2011OECD

Ademaetal2011OECD

Please cite this paper as: Adema, W., P. Fron and M. Ladaique (2011), “Is the European Welfare State Really More Expensive?: Indicators on Social Spending, 1980-2012; and a Manual to the OECD Social E...

More info »
SAP S/4HANA 1809   Feature Scope Description

SAP S/4HANA 1809 Feature Scope Description

PUBLIC Document Version: 1.2 – 2019-01-21 SAP S/4HANA 1809 - Feature Scope Description company. All rights reserved. affiliate THE BEST RUN 2019 SAP SE or an SAP ©

More info »
Good Mining   Illustrative Financial Statements 2018

Good Mining Illustrative Financial Statements 2018

Good Mining (International) Limited Illustrative financial statements for the year ended 31 December 2018 International GAAP ®

More info »
Microsoft Word    02 APB VFR Valuation Advisory 2 Valuation of Customer Related Assets 061516 Final

Microsoft Word 02 APB VFR Valuation Advisory 2 Valuation of Customer Related Assets 061516 Final

JUNE 2016 APPRAISAL PRACTICES BOARD VFR VALUATION ADVISORY 2: THE VALUATION OF CUSTOMER-RELATED ASSETS

More info »
Untitled

Untitled

"INSURANCE COMPANY OF 1921, THE" LAW 40 Act of May. P.L. 682, No. 284 Cl. 17, 1921, AN ACT to insurance; and consolidating amending, Relating revising, for incorporation of insurance providing the the...

More info »
Operating Circular No. 3 Collection of Cash Items and Returned Checks

Operating Circular No. 3 Collection of Cash Items and Returned Checks

Federal Reserve Banks Operating Circular No. 3 COLLECTION OF CASH ITEMS AND RETURNED CHECKS Effective January 1, 2019

More info »
Technical Line: Lessee model comes together as leases project progresses

Technical Line: Lessee model comes together as leases project progresses

Applying IFRS IFRS 15 Revenue from Contracts with Customers A closer look at IFRS 15, the revenue recognition standard (Updated October 2018)

More info »