THE DEMAND FOR FOOD QUALITY IN RUSSIA AND ITS LINKAGE TO OBESITY Matthias Staudigel Justus-Liebig-University of Giessen Matthias.Staudigel@agrar.uni-giessen.de 010 Seleted Poster Paper prepared for presentation at the 1 st Joint EAAE/AAEA Seminar The Eonomis of Food, Food Choie and Health Freising, Germany, September 15 17, 010 Copyright 010 by Matthias Staudigel. All rights reserved. Readers may make verbatim opies of this doument for non-ommerial purposes by any means, provided that this opyright notie appears on all suh opies.
The Demand for Food Quality in Russia and its Linkage to Obesity Abstrat This study analyses whether Russian households differ in their hoie of food quality when they differ in their number of overweight and obese members. Using survey data from the Russia Longitudinal Monitoring Survey (RLMS) for the years 1995-005, households are lassified into three weight groups. Quality elastiities of expenditures are estimated by a fixed-effets panel model regressing unit values of several food groups on expenditures and a set of household harateristis. Coeffiients for eah weight group are reeived by inluding interation terms of expenditures and weight group dummies. A set of Wald tests is applied to test for slope heterogeneity aross weight groups. Desriptive statistis reveal that obese households atually purhase larger quantities and pay less per unit for many food produts. However, estimates of the quality elastiity show low absolute values and range from -0. to 1.1 for single food groups and the null hypothesis of equal parameters for all weight groups annot be rejeted. Keywords: unit values; quality hoie; quality elastiity; obesity; Russia; RLMS; JEL-Classifiation: C; D1; I10; I18; Q18; 1 Introdution Eonomi researh has linked the global rise in overweight and obesity to tehnial progress. Industrialised food prodution, ready-to-eat meals and time saving tehnologies at home have dereased the ost of energy intake (Cutler et al. 00; Lakdawalla and Philipson 009; Philipson and Posner 00). Moreover, the relative pries of highly-proessed and energydense foods have shown a stronger derease than those of healthier foods (Gelbah et al. 007). As a onsequene, some authors suggest inreasing the osts of energy dense foods and making healthier foods heaper (e.g. Brownell et al. 009). However, there is a large body of literature, espeially in development eonomis, showing that an inrease in inome or a relative prie redution does not neessarily influene just the quantity purhased. Subramanian and Deaton (1996) state, that demand for energy will rise, not one for one, as onsumers substitute quality for quantity. Behrman and Deolalikar (1988) report, that even in very poor regions, people show onsiderable demand for higher quality. Thus, when inome inreases or pries of food deline in general, people purhase higher-
quality vegetables or uts of meat, for example. Quantity will slowly approah a saturation level and onsumers spend a larger share of rising expenditures on higher quality. Given this senario, we ould expet rather stable quantities and energy intake despite heaper foods. As only parts of the population seem to be affeted by hanging food environment and lifestyles one ould ask whether households onsisting of many overweight people make different hoies regarding food seletion than those whose members have normal weight. Given more wealth, people generally an deide whether they onsume more of the same produts or produts with higher quality. The objetive of the present paper is to test, whether this is the ase and households that differ in the weight status of their members show different quality reations when their resoures hange. Using data from the Russia Longitudinal Monitoring Survey (RLMS) the present analysis examines the ase of Russia where obesity and overweight are a serious health problem (Doak et al. 000; Huffman and Rizov 007; Jahns et al. 00). Moreover, produt variety has dramatially inreased during transition and onsumers set high value on quality but are also prie sensitive at the same time (Shmid 004; Honkanen and Frewer 009). So far, two studies have been onduted about the quality hoie of onsumers in Russia and the linkage to household resoures. Manig and Moneta (009) estimate Engel urves for quantities and quality (of alories) and onlude that Russian households inrease the quality of the foods they purhase with rising inome. However, Stillman and Thomas (008) find that a hange in resoures only leads to shifts between heap and expensive food groups (like starhes and meats) but there is little to no hange in quality within one food group. The present paper analyses the demand for quality of single food items (in terms of expensiveness) over population groups stratified by weight. Households are firstly lassified into weight groups. Then quality elastiities of expenditures on several food groups are estimated. Finally, Wald tests are applied to test for heterogeneity in the quality elastiity aross weight groups. The paper proeeds as follows. Setion gives the theoretial bakground on the demand for quality. Setion introdues the data and desribes the weight lassifiation of households. Setion 4 provides desriptive statistis on spending patterns of Russian households and presents the regression and tests results. In Setion 5 the results are disussed and onlusions are drawn.
Modelling the Demand for Quality Eonomi modelling of the demand for quality is based on the work of Deaton (1988; 1997). A entral element in this ontext is the unit value whih is the ratio of expenditures divided by quantities of a ertain produt observed for eah household. Sine the observed food groups are aggregates of many heterogeneous produts hosen by eah household, unit values ontain information on the atual market pries as well as on the households quality hoie, i.e. whether households purhased more or less expensive produts. The basi assumption is that if data are olleted for households that belong to the same luster (e.g. village, site) there should be no substantial variation in market pries within eah luster for the same produt. Given this fixed prie struture, the within luster variation of unit values omputed for eah household allows to analyse the influenes of inome, expenditures and household harateristis on quantities and qualities of foods purhased. Let be a luster of households and p a vetor of market pries of individual goods within a produt group within that luster. Let the salar λ be the general prie level of this produt group (e.g. a Laspeyres index) in luster ompared to other lusters and * p be a vetor that desribes the relative prie struture between the individual goods whih is onstant within eah luster. Then we have the following relation: (1) p = λ p * Aording to (1), the atual market pries p are a funtion of the prie level λ and the relative prie struture within eah luster. A ertain produt group s quantity purhased by eah household is determined by the vetor of the quantities of all individual q food items : p * Q Q = k 0 q (), 0 where k is a vetor of ones. The expenditures for the produt group are the produt of p times q, or using equation (1): E E * (). = p q = λ p q 4
Dividing expenditures by quantity, we get the expression for the unit value V from () and (): * * i i p q p q (4) V = E / Q = λ = λ = λ i v. 0 k q q In equation (4), v is an expensiveness or quality index that indiates the relative expensiveness of a households food basket or the average ost of food items i per food group of a single household. Writing (4) in logarithms gives: V (5) ln = ln λ + ln. v Equation (5) illustrates that the observed unit value is a funtion of the market prie level within a luster and eah household s quality hoie. The prie level λ is exogenous to individual household deisions. But the quality index v is endogenous, beause it depends on inome/expenditures, pries, and household harateristis (Yu and Abler 009). Thus, quality hoie an be modelled as a funtion of total expenditures X and a vetor of household harateristis S (Beatty 007) 1 : ln v = α + β ln X + θ j S j + ε (6). j The oeffiient β in equation (6) is the quality elastiity of expenditures that shows how households hange the quality level of a produt group, when their resoures hange. Inserting (6) in (5) gives: (7) ln V = α + β ln X + θ j S j + γ D + ε. j With observable unit values on the left hand side equation (7) an now be estimated to derive values for β. Sine unit values also vary with atual market pries, the model in (7) is ideally estimated with prie data on the right hand side. Consistent estimation of non-prie parameters, however, is possible when we assume that market pries do not vary within eah luster. Therefore, (Beatty 007). γ D a set of dummies for eah luster ontrols for effets of λ 1 Subindexes for lusters are dropped. 5
Now onsider a household that has reahed its saturation level for a ertain food group. An inrease in expenditures might no longer ause an inrease in quantity but in quality (i.e. the unit value), yielding higher values for β. On the ontrary, if a ertain household still prefers to get more quantity, it might not spend that muh on quality and β would be lower. The present analysis tries to test if exatly these onsiderations apply to the ase of households with mainly normal weight and households with mainly overweight/obese members in Russia. Data.1 RLMS The Russia Longitudinal Monitoring Survey (RLMS) has been implemented to measure the impat of transition and aompanying reforms on living onditions in the Russian Federation. It omprises a series of repeated ross-setion surveys that ollet detailed data on, for example, individual health and nutrition, expenditures, assets and soiodemographi harateristis of households as well as ommunity level food pries and infrastruture. In order to get a nationally representative sample, the RLMS was designed as a stratified threestep luster sample. Households were the target units, defined as a group of people dwelling together and sharing a ommon budget (Zohoori et al. 1998). Additional to the (weighted) ross-setions that are nationally representative, there is a longitudinal omponent that allows a panel to be reated that onsists of those households that have been interviewed in two or more onseutive rounds. These longitudinal data show what has happened to households and individuals with given harateristis over time. The present analysis uses household-level data from the nine Phase II rounds 6 to 14 overing the years 1995 to 005. This panel omprises a total of 8,951 responding households. Of these, 6,48 have been interviewed in at least two rounds. Observations with negative inome and expenditures were exluded. Also those households that live in rural areas are exluded, as the onsumption and shopping behaviour of farming households might signifiantly differ from non-farming households beause they rely on home produed goods in their usual diet. After purging missing and implausible values the analytial sample inludes 4,841 responding households and 4,5 household-year observations. About twenty-two perent of the households responded in two waves, 17% in three, 1% in four, 9% in five, 8% in six, 7% in seven, 7% in eight, and 18% in all nine waves. For more details on the design of the RLMS see e.g. Heeringa (1997); Swafford and Kosolapov (00); Jahns et al. (00). Altogether Phase II inludes waves 5 to 17 that were onduted in the last quarter of 1994, 1995, 1996, 1998 and from 000 to 008, respetively. In 1997 and 1999, no surveys were onduted. 6
Finally, the RLMS ontains post-stratifiation weights for unbiased (e.g. nationally representative) estimation of desriptive statistis for ross-setions. However, the present analysis does not use sample weights as it is longitudinal and inludes follow-up households from the non-ross-setional part who have sample weights zero (RLMS, 010). 4. Classifiation of Households Expenditures, quantities purhased, and thus, unit values, are only observed at the household level. Therefore, we need to ategorise households aording to their members weight status in order to test whether households that differ in the number of overweight and obese members also differ in their behaviour. Following an approah of Doak et al. (000) the present analysis divides households into three weight ategories: normal, overweight and obese. First, eah adult member was ategorized using the BMI ut-offs BMI<5 (normal), 5<BMI<0 (overweight) and BMI>0 (obese). For the members of age to 18 the age adjusted perentile equivalents published by Cole (000) were used for lassifiation. In the next step households were lassified as follows: Obese: any household with an obese member; Overweight: any household with an overweight person and no obese person Normal: neither obese nor overweight household members. After this lassifiation, 5% of all household-year observations show neither obese nor overweight persons. Obese and overweight household observations aount for about 7% eah. Table 1 depits transition probabilities between household weight groups for onseutive rounds. In 85.8% of all ases, a household lassified as obese in one round will also be obese in the next round. The same interpretation holds true for overweight with 74.8% and normal with 76.7%. These figures indiate that the groups are relatively stable. 4 Heeringa (1997) points out that there is onsiderable debate over the value of using weights in multivariate analysis. Some statistiians argue that using weights is not neessary if the fixed effets that explain the variation in weights are inluded in the model. In RLMS data, the household harateristis that explain the greatest variation in weights are the geographi region and the urban/rural harater of the ivil division in whih the dwelling is loated. Variation in individual weights will reflet the geographi effets for households as well as differentials due to post-stratifiation of the sample by major geographi regions, age, and sex. 7
Table 1: Transition Probabilities between household weight groups Obese Overweight Normal Total n 6,196 851 178 7,5 Obese % 85.8% 11.8%.5% 100.0% n 1,079 5,50 779 7,61 Overweight % 14.7% 74.8% 10.6% 100.0% n 19 990,679 4,798 Normal %.7% 0.6% 76.7% 100.0% n 7,404 7,44 4,66 19,84 Total % 8.% 7.9%.9% 100.0% Soure: RLMS, 1995-005. Table shows orrelation oeffiients between the group indiators obese and overweight and the share of obese and overweight persons in the household. As the oeffiient of.8 for obese is quite high, this group seems to reflet the number of obese very well. The orrelation between the indiator overweight and the share of overweight people is somewhat lower but still over.70. This is possibly aused by the fat that many households in the obese group have also a high share of overweight people. Table : Correlation Weight Indiators and shares of overweight and obese family members Obese Overweight Share obese Share overweight Share obese 0.856* -0.4958* 1 Share overweight -0.79* 0.715* -0.07* 1 * Signifiant at the 1%-level. Soure: RLMS, 1995-005. 4 Results 4.1 Desriptive Statistis To provide a ontext for the regression results in Setion 4., this setion shows a series of desriptive statistis on food onsumption and spending patterns aross household weight groups. Sine body weight and energy intake likely are a positive funtion of inome in Russia (Jahns et al. 00) statistis are further stratified by inome in order to ompare the behaviour of weight groups at the same level. Starting with Table we get a first important result on spending patterns. 5 The first row shows that normal households have higher per-apita inomes (PCY) than obese households for all inome levels. The average inome growth from the medium to the highest tertile is 5 All monetary values here and later are expressed in onstant rubles (100 = 005). 8
also onsiderably larger than from the lowest to the medium level. The total per-apita expenditures (PCE) in the seond row follow the same pattern. They are higher for normal households than for obese households. At the high-inome level, for example, normal households spend 7,470 rubles per apita per week ompared to 6,8 rubles per apita per week for obese households. However, when we look at the total food expenditures per-apita (PCEF), the piture has ompletely hanged. Obese households spend more on food than normal households in the lowest (1,406 rubles vs. 1,91 rubles), medium (1,97 rubles vs. 1,810 rubles), as well as the highest inome tertile (,68 rubles vs.,58 rubles). Translated into budget shares, obese households spent 5.% of their total expenditures on food ompared to 51.6% and 51.0% for overweight and normal households, respetively. These fats point to the great importane that is attahed to food by households with more obese members. Table : Inome, total expenditures and food expenditures per apita by inome tertile and weight ategory Inome Normal Overweight Obese Low,58,10,7 PCY Medium,1,178,17 High 6,410 7,01 6,190 Low,869,861,859 PCE Medium,95,87,87 High 7,470 7,451 6,8 Low 1,91 1,5 1,406 PCEF Medium 1,810 1,855 1,97 High,58,607,68 Soure: RLMS, 1995-005. Budget shares of single food groups reveal further preferenes of different household weight groups. Most of all, obese households devote a larger share of their budget to the group of meats (reahing 14.1% vs. 1.0% and 11.8% for overweight and normal households, respetively). Moreover, obese households have higher budget shares of potatoes (.1% vs. 1.5% and 1.%) and fish. Surprisingly, this is also true for vegetables whose share amounts to.7% for obese households and.1% for normal households. On the ontrary, obese households show substantially lower budget shares of eating out (.6% vs. 4.5% for normal households) as well as of alohol (1.6% vs..5% for normal households). Table 4 depits quantities onsumed per apita and per week by household weight groups and inome tertiles for seleted produt groups. These figures stress the differenes between weight groups. Obese households onsume about 1.5kg to.0kg more potatoes and about 0.5-9
0.9kg vegetables per apita and per week than normal households. The differenes in meat onsumption get larger when inome inreases. At the low-inome level obese households onsume 0.1kg more than normal households per apita and per week. At the high-inome level, the differene amounts to 0.kg. On the ontrary, normal households onsumption of tobao and alohol is onsiderably higher for all inome levels. The differenes for all other food groups do not reveal different diet patterns. Hene, obese households do not follow the stereotype of just onsuming produts high in fat and sugar. They do purhase more meat and potatoes, but they also onsume more vegetables and bread and even less alohol and tobao than normal-weight households. Thus, the problem seems not to be that they tend to onsume a different (less healthy) diet mix ompared to normal-weight households, but that they onsume larger quantities. Table 4: Quantities per apita by weight and inome tertile for seleted food groups. Potatoes Vegetables Meats Alohol Tobao Soure: RLMS, 1995-005. Weight Category Tertile Normal Overweight Obese Low 1.48 1.7.01 Medium 1.64.60.76 High 1.48 1.64.7 Low 1.06 1.6 1.8 Medium 1.75.0.70 High 1.6 1.86.15 Low 0.9 0.96 1.0 Medium 1.4 1.0 1.46 High 1.45 1.64 1.77 Low 0.6 0.4 0.16 Medium 0. 0. 0.18 High 0.5 0.9 0.40 Low 1.4 1.17 1.06 Medium 1.46 1.04 1.1 High 1.81 1.47 1.60 The question is now whether obese households realise higher quantities by spending onsiderably less per unit purhased. The figures for unit values of seleted produt groups in Table 6 indiate, that obese households atually buy lower quality (i.e. less expensive) goods. The most impressive differenes an be observed within the meat group, where normal households spend about 10 rubles per kg and obese households only about 85 rubles/kg. Cereals, vegetables, fruits, as well as sugar & onfetionery show the same pattern, although with lower magnitudes. Obese households pay also less for alohol but more for tobao. 10
This supports the hypothesis that households with one or more obese members purhase less expensive produts. Only in the lowest inome tertile, the differenes are not very large, and some unit values are higher. This ould arise from laking possibilities to purhase even heaper foods at this level. On the other hand, higher-inome households might be able to hoose from a greater variety of produts with different pries. Table 5: Unit Values by weight and inome tertile for seleted food groups. Cereals Vegetables Fruits Meats Sugar & Confetionery Alohol Tobao Soure: RLMS, 1995-005. Weight Category Tertile Normal Overweight Obese Low 0.4 1.0 19.9 Medium 1.4 1.1 0. High 5. 4..8 Low 16. 14.7 1.6 Medium 16.9 15.9 14.5 High.1 0.8 0.0 Low. 0.6 0. Medium.6 0.6 9.9 High 40. 9.0 5.8 Low 10.4 90. 84. Medium 10.7 9. 86. High 10.8 90.1 85.1 Low 51.7 5. 5.9 Medium 61.8 60.5 57.0 High 77.4 76.7 7.6 Low.8 6.4.6 Medium 5.0 7.8.8 High 61.0 60.0 54.9 Low 7.9 7.6 7.6 Medium 75.8 8.9 8.7 High 84.5 9.1 91. 4. Estimation Strategy The empirial analysis is based on equation (7) in Setion. The objetive is to test whether normal, overweight and obese households show a different demand for quality, i.e. whether they are heterogeneous in β and β1 β β. Thus, following Gould (00), we rewrite: (8) V = ln ht β X ht + β X ht G + β X ht G + θ j S jht + γ D + 1 ln (ln ) (ln ) ξ yty + ε ht j y with, β = β and β = β ; h indexes households, t time, and lusters. G and G β1 β1 are dummies for overweight and obese household groups, respetively. S jht is a set of 11
household harateristis, namely household size, the household head s eduation, age, and gender, D is a set of luster dummies that ontrol for spatial prie variation, T is as a set of year dummies that ontrol for maro effets in the ourse of time. Unobservable household harateristis like preferenes or abilities or other ommunity level fators than food pries like availability of shops, infrastruture or tradition and eating habits are likely to influene unit values. These are potential soures of bias in eonometri analyses when orrelated with exogenous variables that result in onfounding bias when they are not expliitly ontrolled for. Hene, the error term beomesε ht = e + u h ht. Hausman tests indiated that the y regressors are orrelated with individual-speifi error terms e h, so fixed-effets is the appropriate model. One onern with the model is that X might possibly be endogenous to V. Therefore, the model was additionally estimated by an instrumental-variable regression using inome as instrument for X (Beatty 007). However, a Hausman test showed no differenes in parameter estimates for the initial and the IV regression, indiating that X an be treated as exogenous. β β Having reeived oeffiients for β 1,, and, a set of Wald tests is performed that test whether the quality elastiity is signifiantly different between weight groups aording to the following sheme: β 1 1) Test of = 0, H 0 : β = β H 1 : β β1 ) Test of β = 0, H 0 : β = β1 H 1 : β β1 ) Test of β = β, H 0 : β = β H 1 : β β 4. Regression Results Model (1) in Table 6 presents the oeffiients of ln X from the regressions for several produt groups and additionally for the unit value of energy omputed as per-apita food expenditures divided by energy intake per apita. These regressions were performed over the full sample without dummies and serve as a basis for the further analysis. The results onfirm the findings by Manig and Moneta (009) and Stillman and Thomas (008). Russian onsumers reat to resoure hanges by shifting onsumption between rather than within produt groups. When total per-apita expenditures inrease by 10%, the per-unit ost of energy inreases by 6%. Most of the oeffiients for single produt groups are highly signifiant, but their values range 1
Table 6: Fixed-effets regression estimates of the quality elastiity of expenditures for several produt groups. Model (1) Model () inluding group dummies δ lnv / δ ln X δ lnv / δ ln X δ lnv / δ ln X G δ lnv / δ ln X G Energy 0.60 (0.0111) *** 0.618 (0.011) *** 0.00 (0.0018) 0.00 (0.00) Meat 0.01 (0.0045) *** 0.00 (0.0047) *** 0.000 (0.0009) 0.001 (0.001) Bread 0.008 (0.005) ** 0.008 (0.005) ** 0.000 (0.0009) 0.000 (0.0011) Cereals -0.017 (0.0067) ** -0.016 (0.0068) ** 0.000 (0.0014) -0.00 (0.0018) Potatoes 0.009 (0.0104) 0.011 (0.0108) -0.001 (0.00) -0.00 (0.009) Vegetables 0.061 (0.0116) *** 0.06 (0.0119) *** -0.00 (0.009) -0.001 (0.005) Fruits 0.04 (0.019) *** 0.06 (0.01) *** -0.00 (0.007) -0.00 (0.004) Milk 0.046 (0.0070) *** 0.046 (0.0071) *** 0.001 (0.0015) 0.000 (0.0018) Dairy 0.04 (0.0086) *** 0.040 (0.0088) *** 0.00 (0.0019) 0.004 (0.00) Vegetable fats 0.010 (0.0057) * 0.01 (0.0058) * -0.00 (0.001) -0.00 (0.0016) * Sugar & Confetionery 0.10 (0.0105) *** 0.101 (0.0107) *** -0.001 (0.001) 0.00 (0.007) Fish 0.11 (0.06) *** 0.11 (0.040) *** -0.00 (0.0051) -0.001 (0.006) Coffee & Tea 0.100 (0.015) *** 0.098 (0.015) *** 0.004 (0.00) 0.001 (0.0041) Beverages 0.017 (0.0188) 0.0 (0.019) -0.00 (0.0040) -0.008 (0.005) Alohol 0.091 (0.0196) *** 0.094 (0.0199) *** -0.001 (0.0041) -0.005 (0.0051) Tobao 0.087 (0.009) *** 0.088 (0.009) *** -0.001 (0.001) 0.000 (0.007) * Signifiant at the 10% level. ** Signifiant at the 5% level. *** Signifiant at the 1% level. Note: Heterosedastiity-robust Huber/White standard errors are reported in parentheses. All regressions ontrol for household size, the household head s eduation, age, and gender as well as year fixed effets.
from -.0 for ereals to.11 for fish. Hene, the quality elastiity of expenditures is rather low. For example, a 10% inrease in total expenditure would inrease the average quality purhased expressed in unit values by 0.% for meat, 0.6% for vegetables, 0.5% for milk, and 1.1% for fish. These magnitudes are in line with results from Stillman and Thomas (008) for the same dataset who report values from -0.05 for meat to 0.15 for all fruits and vegetables. However, they used per-apita expenditures as variable and performed the analysis over individuals rather than households. Model () in Table 6 presents these regression oeffiients again but now inludes interations of ln X with dummies for overweight and obese households. The signs of these interation terms do not show a learly negative or positive tendeny and their absolute values range from 0.01-0.001. This already indiates that the quality reations do not differ onsiderably aross weight groups. The related Wald tests for slope heterogeneity in Table B (Appendix) substantiates that there is no signifiant differene. Only vegetable fats and sugar & onfetionery show signifiant tests, but still the differenes are rather small. Thus, the present analysis does not provide any evidene for a different demand for quality of households with more overweight or obese people. Table 7: Wald tests for slope heterogeneity aross weight groups. Hypotheses β = β 1 β = β1 β = β Energy 0.99 (0.0) 1.49 (0.) 0.1 (0.576) Meat 0.04 (0.848) 0.0 (0.656) 0.17 (0.679) Bread 0. (0.641) 0.06 (0.801) 0.74 (0.90) Cereals 0.05 (0.8) 1. (0.49) 1.76 (0.184) Potatoes 0.7 (0.601) 0.99 (0.1) 0.66 (0.417) Vegetables 0.7 (0.9) 0.10 (0.748) 0.41 (0.5) Fruits 0.94 (0.) 0.6 (0.550) 0.05 (0.817) Milk 0.45 (0.504) 0.07 (0.789) 1.0 (0.55) Dairy 1.44 (0.1).66 (0.10) 0.9 (0.7) Vegetable fats 1.48 (0.4).00 (0.084)* 1.46 (0.8) Sugar & Confetionery 0.6 (0.611) 1.8 (0.40) 4.71 (0.00)** Fish 0.1 (0.74) 0.0 (0.87) 0.04 (0.840) Coffee & Tea 1.67 (0.196) 0.0 (0.871) 1.6 (0.61) Beverages 0.0 (0.58). (0.17).06 (0.151) Alohol 0.07 (0.790) 0.86 (0.5) 0.89 (0.45) Tobao 0.49 (0.484) 0.0 (0.886) 0.9 (0.589) Note: F-statistis are reported and p-values are reported in parentheses. To assess whether there are larger differenes at a ertain inome level, additional regressions were onduted stratified by inome, but again, no signifiant differenes ould be found. 14
Another reason that weight groups were found to reat rather similar ould be, that they have not been divided sharply enough. Therefore, a further regression inludes only those households that stay in one weight group over all nine rounds. This should ensure that there is atually a lear ut lassifiation. However, also here the tests for heterogeneous parameters were insignifiant. 5 Disussion and Conlusion The present analysis sought to identify whether households that differ in the number of overweight/obese members also differ in their hoie of food quality. This investigation was based on a theoretial model by Deaton (1988; 1997) that identifies the demand for quality by assessing the impat of expenditures on unit values. Using data from the Russia Longitudinal Monitoring Survey, households were lassified into normal, overweight and obese and different spending patterns were desribed. Fixed-effets hedoni regressions on unit values for several food groups were performed that inluded interation terms for expenditures and weight ategories. A series of Wald tests was then applied to test for differenes in the quality elastiities of expenditure aross weight groups. Desriptive statistis revealed that obese households atually spend more on foods, onsume larger quantities and show lower per-unit osts for the majority of food groups. Espeially meat showed onsiderable differenes in quantities and unit values, indiating that obese households might trade off quality for quantity. However, the estimates of the hedoni regressions do not support this view. Quality elastiities ranging from -0. to 1.1 for single food groups indiate that the quality reation due to resoure hanges is low within food groups. For the unit value of energy the oeffiient is onsiderably higher, with about 0.6. These results are in line with previous work from Stillman and Thomas (008) as well as from Manig and Moneta (009): Russian households shift between but not within food groups, when their inome/expenditures hange. Finally, the Wald tests on slope heterogeneity deteted no signifiant differenes in quality elastiities aross household weight groups. Thus, normal and obese households do not differ signifiantly in their demand for quality. What are possible reasons for these results? Firstly, the present analysis only examines the hanges in quality when expenditures hange. Given the fat that obese households spend more on food but less in total, we might onlude that they do not atually have to deide whether to buy more quality or to purhase more quantity. They ould inrease their quality 15
by the same amount as normal households do but at the same time inrease their quantity even more. Moreover, as desriptive statistis reveal, there already seems to be an initial differene in unit values. Hene, rather than omparing expenditure effets on quality aross weight groups a diret assessment of the linkage between unit values and obesity/body weight would be more insightful. However, suh a relation would be simultaneously determined and the analysis would suffer from possible endogeneity problems. 6 Referenes Beatty, T.K.M., 007. The Sope of the Unit Value Problem. In: Consumer and Market Demand Agriultural Poliy Researh Network, Projet Report Number CMD-07-06. Department of Rural Eonomy, University of Alberta, Edmonton, Canada. Behrman, J.R., Deolalikar, A.B., 1987. Will Developing Country Nutrition Improve with Inome? A Case Study for Rural South India. Journal of Politial Eonomy, 95(5), 49-507. Brownell, K.D., Farley, T., Willett, W.C., Popkin, B.M., Chaloupka, F.J., Thompson, J.W., Ludwig, D.S., 009. The publi health and eonomi benefits of taxing sugar-sweetened beverages. New England Journal of Mediine, 61, 1599-1605. Cole, T. J, Bellizzi, M.C., Flegal K.M., Dietz, W. H., 000. Establishing a standard definition for hild overweight and obesity worldwide: international survey. British Medial Journal 0, 140-14. Deaton, A., 1988. Quality, Quantity, and Spatial Variation of Prie. Amerian Eonomi Review 78(), 418-40. Deaton, A.,1997. The Analysis of Household Surveys. A Miroeonometri Approah to Development Poliy. The World Bank: Washington, D.C. Doak, C.M., Adair, L.S., Monteiro, C., Popkin, B.M., 000. Overweight and underweight oexist within households in Brazil, China and Russia. Journal of Nutrition 10, 965-971. Gelbah, J.B., Klik, J., Stratman, T., 007. Cheap Donuts and Expensive Brooli: The Effet of Relative Pries on Obesity. Florida State University, College of Law, Publi Law Researh Paper No. 61. Gould, W. (00), Chow tests. STATA FAQs. Available online: http://www.stata.om/support/faqs/stat/how.html. Aess: August 17, 010. Heeringa, S.G., 1997. Russia Longitudinal Monitoring Survey (RLMS). Sample Attrition, Replenishment, and Weighting in Rounds V-VII, available at http://www.p.un.edu/projets/rlms/projet/samprep.pdf. Aess: July 15, 008. Honkanen, P., Frewer, L., 009. Russian Consumers motives for food hoie. Appetite 5, 6-71. Huffman, S.K., Rizov, M., 007. Determinants of Obesity in Transition Eonomies: The Case of Russia. Eonomis and Human Biology 5, 79-91. Jahns, L., Baturin, A., Popkin, B.M., 00. Obesity, diet, and poverty: trends in the Russian transition to market eonomy. European Journal of Clinial Nutrition 57, 195-10. Lakdawalla, D., Philipson, T.J., 009. The Growth of Obesity and Tehnologial Change. Eonomis and Human Biology 7, 8-9. Manig, M., Moneta, A., 009. More or better? Measuring Quality versus Quantity in Food Consumption. Working Paper 009/17, Laboratory of Eonomis and Management, Sant Anna Shool of Advaned Studies. 16
Philipson, T.J., Posner, R.A., 00. The Long-Run Growth in Obesity as a Funtion of Tehnologial Change. Perspetives in Biology and Mediine 46, 87-107. RLMS (Russia Longitudinal Monitoring Survey), 010. Sampling Overview, available at http://www.p.un.edu/projets/rlms/projet/sampling.html. Aess: May 19, 010. Shmid, S., 004. Der russishe Konsument. Lebenswelt, Konsumverhalten, Markenwahrnehmung. Münster et al.: OWC Verlag für Außenwirtshaft. Swafford, M.S., Kosolapov, M.S., 00. Sample Design, Response Rates, and Weights in the Russia Longitudinal Monitoring Survey, Rounds 5 to 10: An Abbreviated Desription. Stillman, S., Thomas, D., 008. Nutritional status during an eonomi risis: Evidene from Russia. The Eonomi Journal 118, 185-1417. Subramanian, S. and A. Deaton (1996): The Demand for Food and Calories. Journal of Politial Eonomy 104(1), 1-16. Yu, X., Abler, D., 009. The Demand for Food Quality in Rural China. Amerian Journal of Agriultural Eonomis 91(1), 57-69 Zohoori, N., Mroz, T.A., Popkin, B.M., Glinskaya, E., Lokshin, M., Manini, D., Kozyreva, P., Kosolapov, M., Swafford, M., 1998. Monitoring the Eonomi Transition in the Russian federation and its Impliations for the Demographi Crisis the Russian Longitudinal Monitoring Survey. World Development 6, 1977-199. 17