Fiscal-Food Policies are Likely Misinformed by Biased Price ...

Original Article

Fiscal-Food Policies are Likely Misinformed by Biased PriceElasticities from Household Surveys: Evidence from Melanesia

John Gibson* and Alessandro Romeo

Abstract

Fiscal-food policies use taxes to alter relativefood prices so as to change diets and aresuggested for reducing non-communicablediseases in the Pacific. Price elasticity esti-mates used by advocates of fiscal-food policiesare often biased and may make policy makerstoo optimistic about small taxes on unhealthyfood and drink inducing big changes in diets.The bias is illustrated using the example ofthe demand for soft drinks in a householdsurvey from the Solomon Islands, with furtherevidence from Papua New Guinea. Aboutone-third of consumer response to soft drinkprice variation in the Solomon Islands is onthe quantity margin, with two-thirds on thequality margin. If the quality response iswrongly treated as a quantity response toprice—as in most studies—the price elasticityof soft drink demand is exaggerated by a factorof two in Papua New Guinea and three in theSolomon Islands.

Key words: demand, household surveys,quality, soft drink taxes, MelanesiaJEL Classification: D12, I10

1. Introduction

Non-communicable diseases (NCDs) are agrowing problem in the Pacific, and variouspolicies are suggested to reduce them. Fiscal-food policies that use taxes and subsidies toalter food prices so as to induce a change in dietare amongst the most commonly debated mea-sures. Some discussion of such taxes, particu-larly by public health researchers, implies thatcountries in the Pacific can ‘have their cakeand eat it too’ because it is claimed that thesetaxes are beneficial from both a health and afiscal revenue standpoint. For example, in adiscussion of soft drink taxes in four Pacificcountries, Thow et al. (2010a) suggest thatshaping the tax to suit the priorities of healthand finance can facilitate uptake, and recom-mend that advocates highlight both the healthand revenue implications of these taxes so asto gain political support. Similarly, Friel et al.(2015) review all-sector public policy makingthat seeks improved health impacts andconclude that soft drink taxes in Pacific Islandcountries are justified because they contributeboth to health and to the Treasury.

This apparent ‘win–win’ situation differsfrom traditional understanding of commoditytaxes, where there are trade-offs between fiscalobjectives and reducing intake of the particularcommodity. Traditionally, excise taxes havebeen viewed as relatively efficient taxes (evenif somewhat inequitable) because they areoften imposed on goods whose demand isown-price inelastic (Carnahan 2015). Itemsthat may be efficient to tax from a revenuepoint of view are those with own-price inelastic

* Gibson: Department of Economics, University ofWaikato, Hamilton, New Zealand; Romeo: Foodand Agriculture Organization of the United Nations,Rome, Italy. Corresponding Author: Gibson, email<[email protected]>.

Received: 11 May 2017 | Accepted: 26 June 2017Asia & the Pacific Policy Studies, vol. 4, no. 3, pp. 405–416doi: 10.1002/app5.189

© 2017 The Authors. Asia and the Pacific Policy Studiespublished by JohnWiley& Sons Australia, Ltd and Crawford School of Public Policy at The Australian National University.This is an open access article under the terms of the Creative Commons Attribution-NonCommercial License, which permitsuse, distribution and reproduction in anymedium, provided the original work is properly cited and is not used for commercialpurposes.

bs_bs_banner

http://orcid.org/0000-0003-3886-6873

http://creativecommons.org/licenses/by-nc/4.0/

demand, because the pre-tax and post-tax equi-librium quantities are close together; thus,taxes on these items do not entail much distor-tion, in the sense of having smaller Harbergertriangles than for price elastic items that havea large change in the post-tax equilibriumquantity. In other words, an unhealthy food isunlikely to be efficient to tax from both a healthand a revenue standpoint, because one objec-tive needs own-price elastic demand and theother needs inelastic demand.The belief that demand for unhealthy food

and drink will respond elastically to taxes per-vades discussion of fiscal-food policies and isfound even at the highest levels. For example,the World Health Organization advocates taxeson sugar-sweetened beverages at a rate highenough to raise retail price of these drinks byat least 20 per cent, and it is argued that this willlead to proportional reductions in consumption(WHO 2016), implying that the own-price elas-ticity of quantity demand is minus one for thesedrinks. Likewise, one of the first fiscal-food pol-icy studies, by Marshall (2000), consideredtaxes on high cholesterol foods and based theanalysis on an assumed own-price elasticity ofdemand of minus one for items such as wholemilk, despite this being eight times more elasticthan existing evidence suggested (Kennedy &Offutt 2000). Perhaps unfairly, one could char-acterize some of the advocacy for fiscal-foodpolicies as the triumph of hope over experience.The goal of this article is to explain why

many elasticity estimates used in support offiscal-food policies are exaggerated. This biasmay make policy makers too optimistic abouthow small taxes on unhealthy food and drinkcan induce big changes in diets. The bias is il-lustrated using the example of the demand forsoft drinks in a household survey from theSolomon Islands, with supplementary evidencefrom Papua New Guinea. The particular choiceof commodity and countries is not because softdrink taxes are pressing policy concerns in thesecountries—in fact, across these two surveys,soft drinks account for an average of only about1/200th of the total value of household con-sumption. Instead, the commodity choice re-flects a focus in the literature on soft drinktaxes, which comprise just under one-half of

the studies of fiscal-food policies that Thowet al. (2010b) review. It is also important topresent evidence from the Pacific, which is aregion where policy-makers in health- andfood-related areas have difficulty in accessingcredible evidence (Waqa et al. 2017).The source of the bias that misinforms fiscal-

food policies is that most of the demandelasticities estimated with household surveydata confuse two separate, but linked, responsesthat consumers can make as prices change. Oneresponse is to change the quantity of what isconsumed, as in the standard textbook modelof a demand curve, while the other is to changethe quality of what is consumed. This qualityresponse is largely ignored in the literature, eventhough both types of response are inherent fea-tures of household survey data, and even thoughboth responses can occur for all types of foodand drink rather than just for items subject tofiscal-food policies. For example, even a basicstaple like rice has quality-related variation,and consumer downgrading of quality inresponse to higher rice prices seems to buffernutrition of Vietnamese consumers againsteffects of price shocks (Gibson & Kim 2013);this coping would be missed if analysts onlyconsider the standard textbookmodel of quantityresponding to price with no quality response.Rather than allow for this sort of quality re-

sponse, most elasticity studies that are used toinform the advocacy for fiscal-food policieswrongly use calculations that are appropriatefor a standard, undifferentiated good, as in thetextbook model of a demand curve. This doesnot describe what household surveys provide,which is data on a mix of many differentbrands, varieties, package sizes, containertypes and so forth, as was first recognized over60 years ago by Prais and Houthakker (1955).This feature was also emphasized in a seriesof papers written thirty years ago by Nobel-prize winner Angus Deaton (1987, 1988,1989, 1990), but it remains largely ignored. Areview of the applied demand literature usinghousehold survey data shows that over 80 percent of studies ignore quality responses andwrongly attribute any adjustment on the qualitymargin to the (exaggerated) quantity responseto price (Gibson & Kim 2016).

406 Asia & the Pacific Policy Studies September 2017

© 2017 The Authors. Asia and the Pacific Policy Studiespublished by JohnWiley & Sons Australia, Ltd and Crawford School of Public Policy at The Australian National University

In the results reported below, approximatelyone-third of the consumer response to softdrink price variation in the Solomon Islandsis on the quantity margin, while almost two-thirds is on the quality margin. In Papua NewGuinea, consumer responses are split evenlybetween the quantity and quality margins.Thus, if the quality response is wrongly treatedas a quantity response to price—as it is in mostof the literature—the elasticity of soft drink de-mand with respect to own-price will be exag-gerated by a factor of two in Papua NewGuinea and three in the Solomon Islands. It isthis elasticity that is the important one from ahealth point of view, because it is only by cut-ting the volume of soft drink consumed (and atthe same time not triggering a switch into otherunhealthy food and drink) that a fiscal-foodpolicy can hope to achieve health objectives.Also, the inelastic quantity demand is unlikelyto provide support for the argument that thesetaxes may be useful for fiscal reasons, becausethe elastic response of quality to price still im-plies a loss of consumer welfare (but in termsof quality, and consumers reveal a willingnessto pay for this as well as for quantity).

The results for these two Melanesian coun-tries may understate the bias in elasticities fromhousehold surveys in other countries, becausequality variation for soft drinks in Papua NewGuinea and the Solomon Islands is fairlymodest. In settings with more quality variation,such as Mexico, the bias from ignoring con-sumer responses on the quality margin appearslarger. For example, when the own-priceelasticity of quantity demand for soft drinksin Mexico is estimated using a framework thatallows responses on the quality margin, itshrinks from around �1.2 to just �0.3(Andalón & Gibson 2017). Thus, ignoringquality responses may cause the quantity de-mand elasticity to be overstated by a factor ofabout four. In terms of health implications,such a large bias undermines the predictionby Grogger (2016) that the one peso per litretax imposed on soft drinks in Mexico fromJanuary 2014 will reduce the weight ofMexicans by up to 2 kg, because this predictionuses the exaggerated elasticities. If thecorrected elasticities are used, the expected

weight loss due to higher priced soft drinks isless than 0.5 kg, which is too small to makemuch difference to population health. Theeffect of a soft drinks tax on population healthin Melanesia would be even less, becauseconsumption is much lower than in Mexico.

2. Data Sources for Estimating PriceElasticities of Demand

There are three main sources of data availablefor estimating price elasticities of quantity de-mand for food and drink. The first is time seriesdata, which typically will be for an aggregatecategory of goods, such as all rice consumedin a country or all beer, where the original datawill be either some combination of foodbalance sheets that account for production,trade and disappearances (e.g. animal feed,seed supply), or administrative data fromexcise statistics or other statistics that trackproduction and consumption. There are twoproblems with time series data for developingcountries, including those in the Pacific; often,there are not enough observations, withperhaps only 30 years for the time series, andthe elasticities are likely to be time-varyingwith the changing structure of the economy.For example, income elasticities of demandfor rice show a trend decline in Asia, and aremuch lower in urban then in rural areas(Timmer 2014) so on-going urbanization willsee these elasticities fall further—it is doubtfulthat price elasticities will stay unchanged asincome elasticities exhibit trend decline.

The second source of data is very finelydisaggregated at the Universal Product Code(UPC) level. These are barcodes for trackingitems within stores and for scanning at check-out and allow up to 100 billion different combi-nations and are found on all packaged food anddrink products and increasingly on fresh pro-duce as well. With bar-coded data, each indi-vidual specification is separately distinguished(e.g. a 600-ml bottle of Coke has a differentUPC than a 600-ml bottle of Fanta or a 1-litrebottle of Coke). The UPC-level data are typi-cally available frommarket research firms suchas AC Nielsen, and enable detailed and precisemeasures of sales for defined geographic

407Gibson & Romeo: Fiscal-Food Policies


markets. The UPC data may be combined withconsumer level data, where a consumer panelis given barcode scanners so that all of theirfood and drink purchases are recorded in theirhome, and can be matched to the prices ofthose same items in local stores. These UPCdata are mostly used in developed countries,where they show that previous estimates ofprice elasticities of demand based on lessdetailed data may be biased (e.g. Ruhm et al.2012 show this for alcohol demand in theUnited States). While there are some marketresearch consumer panels in middle incomecountries like Mexico, and these have beenused for studying fiscal-food policies (e.g.Aguilar et al. 2016), the linkage to UPC leveldata from stores appears to be restricted todeveloped countries.The third source of data, and the most

widely used one in developing countries, ishousehold surveys. These surveys cover awide range of approaches to measuring the to-tal consumption expenditures of householdsand include surveys described as householdbudget surveys, household income and expen-diture surveys, living standards surveys andothers. Most countries field surveys of this typeevery three to five years, making them a widelyavailable data source. Moreover, price varia-tion is potentially available for these surveysbecause prices differ over space in developingcountries due to the weak infrastructure andthe lack of national retail chains that mightadopt a single-price strategy for a particularproduct. These surveys also allow distribu-tional effects to be highlighted (e.g. Hasan2017) because they have sufficient observa-tions to disaggregate analyses by income orexpenditure quintiles, by ethnic or castegroups, and so forth.Although there is a wide variety of house-

hold survey approaches, a common feature ofall types of these surveys is that the data foreach category of consumption comprise sev-eral different types, varieties or packages ofparticular goods, each sold at a different price.For example, the Papua New Guinea House-hold Survey used here is a living standardssurvey where an individual respondentrecalled the purchases, gifts, production and

consumption of 36 categories of food and drink(out of about 100 categories across all types ofconsumption) for the whole household over atwo-week period. One of these 36 categoriesof food and drink was ‘soft drink’, and so thiscovers a multitude of different brands, pack-ages (cans or bottles) and sizes, each sold at adifferent price per litre. A different approachis used by the Solomon Islands Household In-come and Expenditure Survey, where a diarywas kept by respondents, with each transactionallocated to a nine-digit COICOP (Classifica-tion of Individual Consumption by Purpose)code, where there were different codes usedfor juice, for cordial and for each of the majorbrands of fizzy drink (e.g. Coke, Fanta, Szeba,etc). Although the Solomon Islands survey hasmore detail, even within a brand like Coke,there are cans and bottles of different sizes(which would each have a different UPC code)so restricting to, say, COICOP code012207201 (for Coke products) would stillsee analyses based on a mix of differentproducts. The same is true across all householdsurveys, and this matters because in moststudies the elasticity estimates are incorrectlycalculated from a framework that is onlyappropriate for standardized products forwhich there is no mix of different qualities.

3. Sources of Bias in Quantity DemandElasticities from Household Survey Data

The typical elasticity study with household sur-vey data uses budget share models, such as theAlmost Ideal Demand System (Deaton &Muellbauer 1980), or the Linear-Approximateversion (LA-AIDS), or the Quadratic AlmostIdeal Demand System (QUAIDS) of Bankset al. (1997). In these models, the dependentvariable is wGi, the share of the budget devotedto food groupG by household i. Budget sharesare usually modeled as varying with indicatorsof the household’s income, such as a quadraticin the log of per capita expenditures, ln xi, withthe log of the price index for foods in group H(where θGG is for the own-price effect and θGHis for the cross-price effect), with conditioningvariables, zi, and with random noise, u:



wGi ¼ αG þ β1G ln xi þ β2G ln xi½ �2

þ ∑N

H¼1θGH ln pH þ γG�zi þ uGi (1)

The usual calculation of the elasticity ofquantity with respect to price is based on:

εGH ¼ θGH=wGð Þ � δGH (2)

where δGH equals 1 if G = H, and 0 otherwise,and equation 2 is typically evaluated at themeanbudget share. Thus, the own-price elasticity ofquantity demand is treated as depending on therate that budget shares vary as own-price varies,which is shown by θGG, and has a default valueof�1 (e.g. if θGG = 0). Mhurchu et al. (2013) isan example of this approach in a fiscal-foodpolicy study, using 24 food and drink groupsin the Household Economic Survey of NewZealand, with price data coming from the FoodPrice Index for each month and for six regions.

Equation 2 is the correct price elasticity for-mula only if the item is standardized and noquality variation is possible, or else is an aggre-gate good (as in a time series study) with thesame quality composition over time. Under ei-ther of those two conditions, the numerator ofthe budget share is simply the product of priceand quantity for good G. Thus, when the bud-get share is differentiated with respect to price(which is what the regression in equation 1

does), the only possible thing that can be identi-fied is the response of quantity to price.

However, household survey data are not fora standardized good (like a single UPC), andinstead a food group is comprised of many dif-ferent goods, each of different quality (asshown by a different price per unit weight).For example, Figure 1 shows some of the rangein price per litre for fizzy soft drinks in theWaterfront-Foodworld supermarket in PortMoresby (as observed in December, 2016).Just within this store, there were 33 differentspecifications of fizzy drink, including localbrands like Gold-Spot and Gogo Cola,imported brands like L&P and San Pellegrino,and locally bottled global brands likeCoke andPepsi. Across all of these different specifica-tions, there was a 5:1 range in price fromdearest to cheapest, and even if San Pellegrinois excluded there was a 4:1 ratio.1 Some of thisvariation is brand related, and some is relatedto presentation (size and type of container);for example, within the Coke brand, there is a2:1 variation in price per litre depending oncontainer size and type.

1. A similar range was seen at other supermarkets. Thequality range is limited in the countries studied here (andso bias in the elasticities may be smaller than elsewhere);for example, the supermarket nearest the lead author has160 different specifications with a 15:1 ratio of dearest tocheapest brand and presentation of fizzy drink.

Figure 1 Example of the Quality Ladder for Fizzy Soft Drinks,Waterfront-Foodworld Supermarket, Port MoresbyDecember 2016



When equations 1 and 2 are applied tohousehold survey data, there is noway to knowif budget shares are lower because the con-sumer buys a cheaper brand (e.g. a can ofGold-Spot at K5.45 per litre rather than Cokeat K6.06 per litre) or because they buy lessquantity. Thus, when the budget share is differ-entiated with respect to price, there is no way toknow whether it is a quantity response, a qual-ity response, or some hybrid of the two beingidentified. This lack of identification can beshown by noting that the unit value, vGi fromthe ratio of group expenditure to group quan-tity (QGi) shows where on the quality ladderthe consumer locates, because it shows expen-siveness. Thus, total expenditure on the groupcan be written as vGiQGi, and any response ofthe budget share to price will, potentially, in-volve both quantity adjustment (a change inQGi) and quality adjustment (a change in vGi ).Anotherway tomake this point is to note that

within a single store there is a 5:1 ratio of pricesdue to different brands and presentation (as inFigure 1) that is not controlled for in the analy-sis using equations 1 and 2.Meanwhile, the lesssubstantial price variation over space (in thePNG survey the soft drink price index has a2:1 ratio between the dearest and cheapestareas) is controlled for. Thus, the typical analy-sis attributes all the differences in budget sharesto the inter-area price differences while leavinguncontrolled the much larger within-store dif-ferences that reflect quality variation. This re-search design seems incomplete and subject tobias, because it would be expected that in moreexpensive areas (or periods) consumers wouldslide down the quality ladder as one way tocope with the higher prices.This same weakness in the analysis also af-

fects another commonly used approach, wherebudget shares are regressed on unit valuesrather than on prices:

wGi ¼ α�G þ β�1G ln xi þ β�2G ln xi½ �2

þ ∑N

H¼1θ�GH ln

EHi

QHi

� �þ γ�G�zi þ u�Gi

(3)

where the coefficients are given asterisks todistinguish them from equation 1. An example

of this approach is Colchero et al. (2015) whoreport price elasticities for soft drinks inMexico using the unit values from the house-hold survey as a proxy for prices. Elasticitiesderived from equation 3 will not be the sameas those from equation 1, even with the sameequation 2 elasticity formula used, becausethe unit values introduce measurement errors(e.g. due to respondents misreporting expendi-ture or quantity) and because unit values are apoor proxy for price because the quality mixchanges the further one moves in time or spacefrom the point of production due to the effect ofshipping and storage costs (Gibson 2016).Nevertheless, the same lack of identificationoccurs as when using prices because changesin budget shares could be due to adjustmentsthat consumers make on either the quality mar-gin or the quantity margin, and equation 3gives no way to know which.The only way to overcome this identification

problem is to model consumer behavior usingtwo equations; one for each margin of adjust-ment—quantity and quality. Based on Deaton(1990) and McKelvey (2011) but with a qua-dratic in the log of per capita expenditures (soequivalent to a QUAIDS), the appropriateframework is:

wGi ¼ α0G þ β01G ln xi þ β02G ln xi½ �2

þ ∑N

H¼1θGH ln pH þ γ0G�zi þ u0Gi (4)

with quality choice indicated by household i’sunit value for group G, vGi (conditional onprice):

vGi ¼ α1G þ β11G ln xi þ β12G ln xi½ �2

þ ∑N

H¼1ψGH ln pH þ γ1G�zi þ u1Gi (5)

Superscripts 0 and 1 distinguish between pa-rameters on the same variables in each equation.Differentiating the logarithm of equation 4

with respect to price gives:

∂ lnwG=∂ ln pH ¼ θGH=wG ¼ εGH þ ψGH

(6)

where εGH is the elasticity of quantity demandwith respect to the price of H, which is the



parameter of interest for considering fiscal-food policies, and ψGH is the elasticity of theunit value with respect to the price of H. Theown-price elasticity of quality is ψGG�1. Byrearranging equation 6, it becomes clear whyone needs equation 5, for the household’schoice of quality amongst the items withingroup G:

εGH ¼ θGH=wGð Þ � ψGH (7)

The importance of equation 7 is that it showsthat without knowing how quality responds toprices, which can be derived from the ψGH

term, it is impossible to identify the elasticityεGH that shows how quantity responds toprices.

Thus, one needs data on budget shares, onprices and on an indicator of quality, such asthe unit value, to correctly estimate the elastic-ity of quantity demand. Few surveys have theneeded data. The method based on equations 4,5 and 7 is called the ‘unrestricted method’(McKelvey 2011) because no restriction isput on how the household’s choice of qualityresponds to price variation. In contrast, the‘standard price method’ from equations 1 and2 ignores any within-group quality substitutionand is only suited to standardized goods wheresuch substitution is impossible. Likewise, the‘standard unit value method’ based on equa-tion 3 also ignores any quality substitution inresponse to price variation. In Section 5, own-price elasticities of quantity demand for softdrinks that come from each of the threemethods described here are reported, basedon the data discussed in the next section.

4. Data Description

While the Solomon Islands survey has data onseveral types of soft drinks, based on the nine-digit COICOP codes, the price data are not asfinely detailed. The only soft drink prices ob-tained were for a 300-ml bottle of Szeba anda 280-ml can ofCoke (data for the second spec-ification were less frequently available), andthese were for less than half of the areas withhousehold survey data. While the price of a

single, representative, specification can proxyfor the spatial price index for a related groupof foods, the analysis then has to be at grouplevel rather than for individual nine-digitCOICOP codes. The same applies to the PapuaNew Guinea (PNG) survey, which only hasdata available for the group described as ‘softdrinks’ when survey respondents recalledspending and consumption. In the PNG sur-vey, the price index is based on the averageof the prices for a can of Pepsi and a bottle ofCoke, which was obtained from the nearesttrade-store to each Census Unit in the sample.

The descriptive statistics for the budgetshares, unit values and prices are reported inTable 1. The values from the SolomonIslands survey are reported for the full sampleand the sub-sample that was in the areas wherethe price survey was carried out. The averagebudget share for soft drink is the same in thefull and sub-sample, at 0.2 per cent but the(log) unit value is slightly higher in the fullsample (3.11 versus 2.89), suggesting that itmay have been in the less densely populated,and more remote and expensive areas thatsurveyors did not do a price survey. Thecomments on the price survey sheet for theenumeration areas (EAs) with no data collectedbear this out, with reasons like: ‘this canteendepends on trading vessel for cargo’ and ‘thereis no main market in this EA. Only villagestreet stalls, irregular-not every day’

In the PNG survey, the prices were morewidely observed because if teams came backfrom the field with no price data, an individualwas sent back to gather prices from the neareststore and market.2 In rural areas, these maysometimes be an hour walk or more away fromthe households in the survey (who also may bein scattered locations) so it does take strict dis-cipline on survey teams to gather prices inMelanesia, particularly if the survey has a(misplaced) focus on collecting nominal

2. Full details on the logistics and cost of the PNG pricesurvey are in Gibson and Rozelle (2005). Symptomatic ofthe neglect of price data, the 2009–10 HIES in PNG had in-terviewers live in villages for up to 3 weeks to complete thefortnight expenditure diary, but no price survey was carriedout despite the survey teams needing to visit stores andmarkets to purchase their own supplies.



expenditures (which are hard to interpret in theabsences of prices). The average budget sharefor soft drink in the PNG survey was 0.9 percent. The prices and unit values had a similarmean but the unit values are more variable (be-cause they differ between households ratherthan just between EAs).3

The price surveys are limited in commoditycoverage, with only 12 items covering foodand non-food in the survey for the SolomonIslands and 11 store-bought food items in thePNG survey, so data on possible substitutesand complements for including cross-prices inthe budget sharemodels are lacking.Moreover,one possible substitute—beer—has even lesscoverage in the price survey, with beer pricesreported in just 46 per cent of the EAs with softdrink prices for the Solomon Islands survey and60 per cent for PNG, reflecting the restrictionson outlets that can sell beer. Therefore, themodels only consider own-price elasticities,which is a limitation that should be kept inmind, although the evidence from elsewhereis that the bias in estimates of the own-priceelasticity of quantity demand due to ignoringresponses on the quality margin is largely thesame whether or not cross-prices are in themodels (Gibson & Kim 2016; Andalón &Gibson 2017).

5. Results

The estimates of the own-price elasticity ofquantity demand for soft drink that come from

the three different methods discussed inSection 3 are reported in Table 2 for theSolomon Islands and in Table 3 for PapuaNew Guinea. The price survey for theSolomon Islands covered less than half of theareas, so when this is matched to the householdsurvey, it reduces sample size from about 4300to about 1700. It is partly for this reason thatthe Papua New Guinea results are also re-ported, because the inferences for theSolomon Islands are weakened by having onlya partial price survey. To ensure that the resultsare robust to outliers, any observations withprices or unit values more than five standarddeviations from the mean are trimmed.Five equations are estimated to get the elas-

ticities reported in Table 2 and four are used forTable 3 and these equations have up to 15 con-trol variables other than prices or unit values; aquadratic in log per capita real expenditures;household size, three demographic ratios, upto six attributes of the household head (age,gender, education, ethnicity, migrant statusand main livelihood), an area characteristic (ur-banity) and three regional fixed effects.4 Thefull results for each equation are reported inAppendix Tables 1 and 2.When responses on the quality margin are

ignored, due to using either the standard unitvalue method, or the standard price method,the own-price elasticity of quantity demandfor soft drink is estimated to be from �1.02to �1.20 for the Solomon Islands (Table 2).The results for the reduced sample that matches

3. The correlation between prices and unit values was 0.25in PNG and 0.29 in the Solomon Islands, which furthershows that unit values are a fairly imperfect proxy forprices.

4. Area characteristics and fixed effects are included toprovide a short-run interpretation for the elasticities; theseare more appropriate than long-run ones for consideringprice reforms.

Table 1 Summary Statistics for Soft Drink Budget Shares, Prices and Unit Values

Solomon Islands 2012–13 HouseholdIncome and Expenditure Survey

Soft drinkbudget share

Log of unit valuefor soft drink

Log of price indexfor soft drink

Full sample 0.002 (0.005) 3.113 (0.904) n.a.Sub-sample with price survey 0.002 (0.005) 2.892 (0.769) 2.920 (0.141)

1996 PNG Household Survey 0.009 (0.018) 5.639 (0.202) 5.649 (0.142)

Notes: Unit values and prices are in the currency of each country at the time of the survey. The price survey for the SolomonIslands was only carried out for a sub-sample (N = 1730 households in these areas out of a total ofN = 4357 households in thefull survey). The PNG survey has N = 1025 observations. Standard deviations in ().



to the price survey (N = 1730 households) aresimilar to the full sample (N = 4357) accordingto the standard unit value method, and so theneed to rely on only a sub-sample should notbias the remaining comparisons. While thepoint estimates for the own-price elasticitiesare similar, the apparent precision of the esti-mates is much greater with the standard unitvalue method, but this is somewhat mislead-ing; unit values vary across households whoface the same prices (e.g. if a consumer buysa different bundle within the items covered bygroup G than that bought by their neighbourwho shops at the same store) and so unit valuesgive the impression of more price variabilitythan their truly is, artificially improving theprecision of the elasticity estimates.

When the unrestricted method is used, witha budget share equation and a unit value equa-tion so as to study consumer responses on twomargins, the own-price elasticity of quantitydemand for soft drink is only �0.46, whilethe price elasticity of quality is �0.74. Effec-tively, the elasticity from the standard pricemethod (�1.20), which shows how the budgetshare changes as prices vary, is beingdecomposed into two parts—changes in the

budget share from quality change and changesin the budget share from quantity change.Thus, approximately one-third of the consumerresponse to soft drink price variation in theSolomon Islands is on the quantity margin,while almost two-thirds is on the quality mar-gin. If researchers ignore the second marginof adjustment, by wrongly using a single-equation framework that attributes all of theadjustment to changes in quantity, they willoverstate the response of quantity to own-priceby a factor of up to three (based on results forthe standard price method). This exaggerationis likely to make policy makers too optimisticabout how small taxes on unhealthy items cancause big changes in diets.

Because the results for the Solomon Islandsrely on only a partial price survey, it is helpfulto corroborate them. While the PNG House-hold Survey is 20 years old and the structureof demand likely has changed since it wasfielded, few other household surveys in the re-gion have a linked price survey. This failure ofstatistical practice in low-income countries isin the very places with spatially varying pricesand where nominal data are a poor guide to realwelfare levels (Gibson 2013). In the PNG

Table 2 Own-Price Elasticity of Quantity Demand for Sugar-Sweetened Soft Drinks, Applying Various Methods tothe 2012–13 Solomon Islands Household Income and Expenditure Survey

Elasticity Standard error 95% confidence interval

Own-price elasticity of quantityStandard unit value method (N = 4357) �1.05 (0.04) �1.12 to�0.98Standard unit value method (N = 1730) �1.02 (0.08) �1.17 to�0.87Standard price method �1.20 (0.50) �2.18 to�0.22Unrestricted method �0.46 (0.41) �1.27 to 0.35Own-price elasticity of quality �0.74 (0.14) �1.02 to�0.46

Notes: Based on regressions reported in Appendix Table 1. Standard errors are cluster adjusted.

Table 3 Own-Price Elasticity of Quantity Demand for Soft Drinks, Applying Various Methods to the 1996 PapuaNew Guinea Household Survey

Elasticity Standard error 95% confidence interval

Own-price elasticity of quantityStandard unit value method �0.97 (0.26) �1.48 to�0.47Standard price method �1.78 (0.47) �2.72 to�0.83Unrestricted method �0.86 (0.33) �1.51 to�0.21Own-price elasticity of quality �0.86 (0.03) �0.93 to�0.80

Notes: Based on regressions reported in Appendix Table 2. N = 1025. Standard errors are cluster adjusted.



results, the own-price elasticity of quantity de-mand for soft drinks is�0.97 with the standardunit value method, and it is �1.78 with thestandard price method. This gap between theresults for the two standard methods showsthe possibility of equations 1 and 3 differingeven when the same elasticity formula in equa-tion 2 is used, partly due to potential measure-ment error in unit values.5

The remainder of Table 3 shows that theown-price elasticity from the standard pricemethod splits evenly into an unrestricted elas-ticity of quantity of �0.86 and an own-priceelasticity of quality of �0.86. In keeping withthe results for the Solomon Islands, the elastic-ity of quality is more precisely estimated thanis the elasticity of quantity. The even split be-tween the two elasticities means that the varia-tion in soft drink budget shares in the PNGsurvey in response to spatial price variation ismade up on two equal responses—a cut inquantity and a cut in quality (e.g. sliding downquality ladders like those shown in Figure 1).Therefore, if the standard price method wasused it would result in the estimated own-priceelasticity of quantity demand for soft drink inPNG being overstated by a factor of two. Thisbias would be likely to cause policy makers tobe too optimistic about the quantity reductionand consequent health benefits that might re-sult from any tax on soft drinks.

6. Conclusions and Implications

Taxes on drinks with added sugar are beingdebated in several countries. These are an ex-ample of fiscal-food policies, which are moti-vated by concerns about non-communicablediseases like diabetes and obesity. Advocatesfor these taxes assume that quantity demandis fairly responsive to price. At the same time,some advocacy in the Pacific for these taxeshas suggested that they also will be good forthe Treasury, despite the usual trade-off be-tween fiscal efficiency and reducing the

intake of an unhealthy item. Some of the opti-mism about these taxes is likely to bemisplaced and may reflect the exaggeratedprice elasticities that are coming from house-hold survey data when researchers wronglyuse the elasticity framework for a standard, un-differentiated, good.This article provides an example of this bias

for two Melanesian countries. The resultssuggest that about one-third of the consumerresponse to soft drink price variation in theSolomon Islands survey is on the quantity mar-gin, while almost two-thirds is on the qualitymargin. In the Papua New Guinea survey,consumer responses are split evenly betweenthe quantity and quality margins. As a conse-quence of these dual responses by consumers,if an analyst used the standard methods(whether with prices or with unit values onthe right-hand side of the budget shareequation), they would overstate the own-priceelasticity of quantity demand for soft drink bya factor of between two and three.The sort of biases illustrated here are

inherent in elasticity estimates from householdsurvey data when responses on the qualitymargin are ignored (Gibson & Kim 2016).Because the range of soft drink qualities foundin stores in these two countries is more limitedthan in many countries, it is possible that thebias in the elasticity estimates would be largerelsewhere in places where consumers have awider range of qualities to choose over. Theempirical study of these biases requires house-hold survey data with budget shares, with unitvalues and with market prices, and thiscombination of data is surprisingly rare. Arecommendation for agencies interested infiscal-food policy in the Pacific is to lendsupport to survey efforts that will create morecomprehensive databases for studying theentirety of consumer responses, rather than letdata limitations restrict analysts to using theinappropriate textbook model where onlyquantity adjusts as prices vary. In the absenceof these comprehensive databases, and of morereliable elasticity estimates, policy makersshould be cautious about the evidenceproduced by advocates in support of fiscal-food policies.

5. Previous analysis with the PNG survey also shows thatresults (for poverty) differ substantially when using the unitvalues compared to using the price survey (Gibson &Rozelle 2005).



ACKNOWLEDGEMENTS

We thank Bonggeun Kim for assistance andthe Marsden Fund and the Food and Agricul-ture Organization for financial support. Theviews expressed are those of the authors anddo not necessarily reflect the views or policiesof FAO while errors are sole responsibility ofthe authors.

References

AguilarA,GutierrezE,SeiraE (2016)Taxing toReduce Obesity. Mimeo downloaded fromhttp://www.enriqueseira.com/uploads/3/1/5/9/31599787/taxing_obesity_submited_aer.pdf on 6 Sept 2016.

Andalón M, Gibson J (2017) The ‘Soda Tax’ isUnlikely to Make Mexicans Lighter: NewEvidence on Biases in Elasticities of De-mand for Soda (Working Paper No.10765). Institute for the Study of Labor(IZA).

Banks J, Blundell R, Lewbel A (1997) Qua-dratic Engel Curves and ConsumerDemand. Review of Economics and Statis-tics 79(4), 527–39.

Carnahan M (2015) Taxation Challenges inDeveloping Countries. Asia & the PacificPolicy Studies 2(1), 169–82.

Colchero M, Salgado J, Unar-Munguía M,Hernández-Avila M, Rivera-Dommarco J(2015) Price Elasticity of the Demand forSugar Sweetened Beverages and SoftDrinks in Mexico. Economics and HumanBiology 19(1), 129–37.

Deaton A (1987) Estimation of Own- andCross-Price Elasticities from HouseholdSurvey Data. Journal of Econometrics36(1), 7–30.

Deaton A (1988) Quality, Quantity, and Spa-tial Variation of Price. American EconomicReview 78(3), 418–30.

Deaton A (1989) Household Survey Dataand Pricing Policies in DevelopingCountries. World Bank Economic Review3(3), 183–210.

Deaton A (1990) Price Elasticities from SurveyData: Extensions and Indonesian Results.Journal of Econometrics 44(3), 281–309.

Deaton A, Muellbauer J (1980) An AlmostIdeal Demand System. American EconomicReview 70(3), 312–26.

Friel S, Harris P, Simpson S, Bhushan A, BaerB (2015) Health in All Policies Approaches:Pearls from theWestern Pacific Region.Asia& the Pacific Policy Studies 2(2), 324–37.

Gibson J (2013) The Crisis in Food Price Data.Global Food Security 2(2), 97–103.

Gibson J (2016) Poverty Measurement: WeKnow Less than Policy Makers Realize. Asia& the Pacific Policy Studies 3(3), 430–42.

Gibson J, Kim B (2013) Quality, Quantity, andNutritional Impacts of Rice Price Changes inVietnam.World Development 43(1), 329–40.

Gibson J, Kim B (2016) Quality, Quantity andSpatial Variation of Price: Back to the Bog.Working Paper 10/16, Department of Eco-nomics, University of Waikato.

Gibson J, Rozelle S (2005) Prices and UnitValues in Poverty Measurement and TaxReform Analysis. World Bank EconomicReview 19(1), 69–97.

Grogger J (2016). Soda Taxes and the Prices ofSodas and Other Drinks: Evidence fromMexico. Discussion Paper No. 9682, Insti-tute for the Study of Labor (IZA).

Hasan S (2017) The Distributional Effect of aLarge Rice Price Increase on Welfare andPoverty in Bangladesh. Australian Journalof Agricultural and Resource Economics61(1), 154–71.

Kennedy E, Offutt S (2000) Commentary: Al-ternative nutrition outcomes using a fiscalfood policy. British Medical Journal320(7230), 304–5.

Marshall T (2000) Exploring a Fiscal FoodPolicy: The Case of Diet and IschaemicHeart Disease. British Medical Journal320(7230), 301–4.

McKelvey C (2011) Price, Unit Value andQuality Demanded. Journal of DevelopmentEconomics 95(1), 157–69.

Mhurchu C, Eyles H, Schilling C, et al. (2013)Food Prices and Consumer Demand: Differ-ences Across Income Levels and EthnicGroups. PLoS One 8(10), 1–2.

Prais S, Houthakker H (1955) The Analysis ofFamily Budgets. Cambridge UniversityPress, New York.



http://www.enriqueseira.com/uploads/3/1/5/9/31599787/taxing_obesity_submited_aer.pdf



Ruhm C, Jones A, McGeary K, et al. (2012)What US Data Should be Used to Measurethe Price Elasticity of Demand for Alco-hol? Journal of Health Economics 31(6),851–62.

Thow A, Quested C, Juventin L, Kun R, KhanN, Swinburn B (2010a) Taxing Soft Drinksin the Pacific: Implementation Lessons forImproving Health. Health Promotion Inter-national 26(1), 55–64.

Thow A, Jan S, Leeder S, Swinburn B (2010b)The Effect of Fiscal Policy on Diet, Obesityand Chronic Disease: A Systematic Review.Bulletin of the World Health Organization88(8), 609–14.

Timmer CP (2014) Food Security in Asia andthe Pacific: The Rapidly Changing Role ofRice. Asia & the Pacific Policy Studies1(1), 73–90.

Waqa G, Bell C, Snowdon W, Moodie M(2017) Factors Affecting Evidence-Use in

Food Policy-Making Processes in Healthand Agriculture in Fiji. BMC Public Health17(1), 51.

World Health Organization [WHO]. 2016. Fis-cal Policies for Diet and Prevention of Non-Communicable Diseases Technical MeetingReport, 5-6 May, Geneva, Switzerland.http://apps.who.int/iris/bitstream/10665/250131/1/9789241511247-eng.pdf

SUPPORTING INFORMATION

Additional Supporting Information may befound online in the supporting informationtab for this article.

Appendix Table 1 Regression Results for Sol-omon Islands 2012–13 Household Income andExpenditure Survey.Appendix Table 2 Regression Results for Pa-pua New Guinea Household Survey.



http://apps.who.int/iris/bitstream/10665/250131/1/9789241511247-eng.pdf

http://apps.who.int/iris/bitstream/10665/250131/1/9789241511247-eng.pdf

Fiscal-Food Policies are Likely Misinformed by Biased Price ...

Documents

Transcript of Fiscal-Food Policies are Likely Misinformed by Biased Price ...