Food consumption convergence within Europe: a panel data analysis
Bolesław Borkowski*, Hanna Dudek*, Wiesław Szczesny**
*Department of Econometrics and Statistics
**Department of Computer Science
WarsawUniversity of Life Sciences
Summary
The research referred to changes in structures and profiles of consumption in enlarged European Union. Data relating to annual main food products between 1961–2003 was collected and analyzed. Great divergence in the structures of consumption was found, yet the profiles remained constant. In this article it is also investigated the convergence of food consumption per capita in EU during the 1961–2003 period by applying beta-convergence methodology.
Key words: food consumption, beta-convergence, panel data, econometric analysis,
Introduction
Within the European Union the level and profile of consumption varies depending on the country [Braun and Qaim 2000, Mergos and Mizzi 1999, Borkowski andothers 2009]. This inequity is a result of several causes, the main being: historical, cultural and environmental backgrounds, behavioral globalization, especially among young people, the country's political courseconcerning these problems and the current changes in economy and society (Miles and others 2002). General abundance and appeal of food products on the European market, price competition and also the changing preference of customers have caused significant changes in households' consumption's behavior(Aiking and others2006). The rate and direction of changes in different countries varied. Levels and profiles of consumption in those countries are still changing, each in its own time and environment. In a given country, the consumption's demand (global demand) is a multi-argument function, the most important being: population and its structure, income and regional influences. In well developed countries the relation between themand the consumption's level is weaker than in poorer countries. It is to be expected that consumption's tendencies visible in the wealthier countries will flow to less developed countries as their inhabitants needs and market choices will grow.
Convergence in food consumption patterns is important to study for severalreasons. At the most basic level, convergence in consumption would indicate a similar standard of living. Moreover, convergence in food consumption could also show that globalization is having a homogenizing impact on European cultural identity.
The paper attempts to look at a new aspect of the globalization process in Europe, and offer abroader perspective than the one economists usually adopt.
Research on convergence of food consumption is not new. Connor (1994) compared food consumption patterns in the USA and Western Europe. He found outthat European food consumption growth has been correlated with prior, but not contemporaneous, growth in the USA, so trends in the USA are good predictors of changes in Western Europe.
First research on absolute beta-convergence of food consumption, applied in this research, was Gil et al. (1995) study.In that paper, unlike in this one, Central-East European countries were excluded. Gil et al. (1995) compared the data for 15 European Union countries plus Norway in years: 1970, 1980 and 1990.They focused only on calorie intake, whereas in our study also per capita consumption of several food products was investigated. Gil et al. found beta-convergence of total per capita calorie intake and proportion of calories from such food products as: cereals, pulses, animal fats, milk and sugar.Moreover in Gil et al. (1995) research cross-section methodology was applied. This study uses models for panel data. A panel data approach is advocated and implemented for studying growth convergence.Differences in food consumption preference across countries have dimensions that are not readily measurable or observable. In the framework of cross-section regression, it is not possible to take account of such unobservable or immeasurable factors. Only a panel data approach can overcome this problem(Islam 1995).
Data
This study analyses trends in food consumption patterns in Europe over the 1961-2003 period[1]. It refers to the countries: 1) Austria, 2) Belgium-Luxembourg, 3) Bulgaria, 4) Czechoslovakia (after 1992 aggregated data from Czech Republic and Slovakia), 5) Denmark, 6) Finland, 7) France, 8) Germany, 9) Greece, 10) Hungary, 11) Ireland, 12) Italy, 13) Malta, 14) Netherlands, 15) Norway, 16) Poland, 17) Portugal, 18) Romania, 19) Spain, 20) Sweden, 21) Switzerland, 22) United Kingdom.
The analysis uses data compiled by the Food and Agriculture Organization of the United Nations (FAO). Consumption of food refers to the average annual quantity per person of the food commodity consumed[2]. The following product groups have been incorporated into the analysis: (1) cereals; (2) potatoes; (3) sugar and sweeteners; (4) pulses (5) vegetable oils; (6) vegetables; (7) fruits; (8) animal fats; (9) milk, (10), eggs; (11) meat; (12) fish and seafood (13) stimulants.
A brief analysis has shown a great dynamic in the changes in consumption levels of studied groups as well as in consumption profiles in 22 European countries (Table 1a). Below we set the share of particular consumption's groups as an average consumption of a particular group per country's inhabitant. Each country has the same weight, regardless of population.
The greatest declinewas observed within thefollowing groups:milk whole, potatoes, pulses, cereals and animal fats. The share of these groups in the general consumption has decreased by almost 50%, from 1961 to 2003. The levels of eggs, sugar and sweeteners have not changed during the studied period. Studies have shown that the milk group has been a significant part of the general consumption throughout the entire time. In an average diet the greatest increase was notedwithin the milk products, vegetables, oils and meats' groups. In the 22 European countries the consumption of these products has grown between 1961 and 2003 by an average of over 50%, with the milk group experiencing a grow of almost 90%. Vegetables, fruits, fish and seafood's consumption has increased by 20-30%. The rate of these changes depended on a country (Table 1b). The difference in consumptionstructure was calculated based on Gini's measure (see Borkowski andothers 2009).
The greatest change in consumption structure was observed in Denmark, Spain, Finland and Bulgaria, the smallest in countries such as Germany, Switzerland and United Kingdom. Table 1c presents consumptionprofiles of two countries, which experience the greatest change in consumptionstructure and a consumptionprofile of Switzerland, because it showed characteristic stability during the study period. Additionally we present the changes of consumptionstructure in Poland.
Poland characterizes itself by a large consumption of potatoes, Spain by vegetables and fruits, Denmark and Switzerland by milk products and cereals. Compared to Poles, the Swiss eat more fruits, while the Poles themselves eat a lot more potatoes than any other nationality. Milk consumption in Poland has a lesser share in the general consumption than in Denmark, Switzerland and Spain. Because of their historical and geographical backgrounds, countries differ from one another and it is extremely difficult to compare them. The table shows rather the speed of changes in consumption's structure present in each country during the studied period.
Methodology
This research examines whether convergence of food consumption per capita trends exist across European countries. Since the time of Barro’s pioneering proposals (Barro 1991), the phenomenon of economic convergence between countries has been widely and empirically studied.The term “convergence” implies dynamics, or movement toward some common outcome.
Two concepts of convergence appear in the literature of economic growth across countries or regions (Barro and Sala-i-Martin 1992). The first, may be described by the fact that poor economies tend to grow faster than rich ones, so that the poor spatial unit tends to catch up to the rich one in terms of level of per-capita income. Such a situation is always referred to as beta-convergence models. The second interpretation applies when country-wise inequality tends to reduce in time. This process is called sigma-convergence.Beta-convergence is a necessary but notsufficient condition for sigma-convergence (Sala-i-Martin 1996).
Following Sala-i-Martin’s (1996) exposition, assume that β-convergence holds for economies i = 1, ..., N.
Natural log- food consumption of the i-th economy can be approximated by
(1)
where εit – error term with finite variance , εit has mean zero,
α is assumed constant, ,
i indicates country subscript, and t time, i=1, 2,…,N, t=1, 2,…,T.
Manipulating (1) yields,
(2)
Thus, β > 0 implies a negative correlation between growth and initial log level of y.If the parameter β is significantly positive one can conclude in favor of unconditional beta-convergence[3].That is, the growth rates of consumption depend on the initial consumption levels only, and they are inversely correlated.
In the context of consumption studies, convergence is likely to be conditional on the income variable. Therefore, in addition, concept of conditional beta convergencecan be taken into account. In that case in model (1) additional explanatory variables[4]are included:
(3)
where
- a column vector of parameters1, ...k,
k – number of additional explanatory variables,
- a row vector representing the characteristics of countryiin year t.
Conditionalβ-convergence refers to the hypothesis that poor economies should grow faster per capital than rich economies, conditional on country specific characteristics.
Unconditional convergence represents a process of absolute convergence in which all countries under consideration will meet at the same point or stationary state. Conditional convergence is associated with a concept of weaker convergence arising from the implications of Solow’s model which predicts convergence after taking into account the factors which determine stationary equilibrium.
This work concentrates only on unconditional convergence, causes and mechanism of convergence are beyond our main interest in this paper.
Estimation
The dynamic structure of the modelmakes the Ordinary Least Squares (OLS), the Fixed Effects (FE) and the RandomEffects estimators biased and inconsistent, since the lagged level of income is correlated with theerror term (Baltagi 2001). In order to cope with the above mentioned problems estimators basedon the General Method of Moments (GMM) are employed, which are consistent for N→∞ with fixed T.
Arellano and Bond (1991) developed so-called DIF-GMM dynamic panel data estimator that includes lags of both the dependent and independent variables as instruments. However, recent studies (Blundell and Bond, 1998; Blundell, Bond and Windmeijer, 2000) show that the estimation in first-differences has a large bias and low precision in finite samples.As a consequence, if the instruments are weakly correlated with the regressors, estimates may be biased towards 0. In the case of the convergence estimates, this turns out to overestimate the speed of convergence.In the paper so-called SYS-GMM estimator proposed by Arellano and Bover (1995) and Blundell and Bond (1998) is used.It reduces the potential biases and imprecision associated with the usual difference estimator by combining, in a system, the regression in differences with the regression in levels.
We have used the STATA v.10.0 econometric software to obtain the Arellano-Bond dynamic panel estimates of the linear models (2) and (3) described above. The consistency of the estimation depends on whether lagged values of the endogenous and exogenous variables are valid instruments in our regression. Also, this methodology assumes that there is no second-order autocorrelation in the errors, therefore a test for the previous hypotheses is needed. We have also conducted a test for autocorrelation and the Sargan test for over-identifying restrictions as derived by Arellano and Bond (1991). Failure to reject the null hypothesis in both tests gives support to our model[5].
The testing for sigma-convergence is based on thecoefficient of variation[6] of the cross-section series.
Results
Regressions are performed on apanel data set consisting of the 35countries for the 1961-2003 period. The panel data estimator used is the GMM Arellano & Bond and Arellano Bover/ Blundell Bond procedure.
Table 2 Results of the estimation of the unconditional beta-convergence model
Arrelano-Bond method / Arellano Bover/ Blundell BondmethodFood commodities / α / 1-β / p-value of AR(2) test / p-value of Sargan test / α / 1-β / p-value of AR(2) test / p-value of
Sargan test
All countries
total calories[7] / 0,5361 (0,0775) / (0,9341) (0,0096) / 0,0906 / 0,2673 / 0,5289 (0,0994) / 0,9345 (0,0123) / 0,0918 / 0,2090
share of calories from animal products / 0,2507 (0,0419) / 0,9266 (0,0124) / 0,1325 / 0,9290 / 0,1389 0,0514 / 0,9599
(0,0152) / 0,1407 / 0,5110
vegetables / 0,6627
(0,0942) / 0,8350
(0,0843) / 0,1246 / 0,8479 / 0,6842 (0,0871) / 0,8307 (0,0200) / 0,1324 / 0,8495
dairy products excluding butter / 4,0249 (0,1626) / 0,2498
(0,0127) / 0,0636 / 0,0046 / 3,3755
(0,0972) / 0,3693
(0,0168) / 0,7477 / 0,1306
fish and seafood / 2,5882 (0,1040) / 0,0187
(0,031) NS / 0,1599 / 0,9995 / 2,6150 (0,1141) / -0,0067
(0,0231)NS / 0,1982 / 0,9997
Czechoslovakia, Hungary and Poland
total calories / 1,0752
(0,3283) / 0,8678 (0,0404) / 0,4994 / 0,1864 / 0,8828 (0,2793) / 0,8914 (0,0344) / 0,4999 / 0,5269
share of calories from animalproducts / 0,1688
(0,0291) / 0,9507 (0,0078) / 0,9702 / 0,4254 / 0,2191 (0,0879) / (0,9361).(0,0256) / 0,9710 / 0,3362
cereals / 4,2636
(0,3977) / -0,0029
(0,0927) NS / 0,0904 / 0,3226 / 3,7183
(0,3583) / 0,2289
(0,0741) / 0,0911 / 0,6206
potatoes / 4.2159 (0,1390) / 0,0091 (0,0325) NS / 0,0884 / 0,3226 / 3,9879
(0,0684) / 0,0258
(0,0464)NS / 0,0860 / 0,2431
sugar and sweeteners / 2,9538 (0,3172) / 0,1888 (0,0868) / 0,1111 / 0,2880 / 2,4083
(0,0973) / 0,2350
(0,0694) / 0,1131 / 0,2735
pulses / -0, 9592 (0, 2034) / 0,0353
(0,0471) NS / 0,1000 / 0,3198 / -0,8935
(0,4571) / 0,0877
(0,0678)NS / 0,1026 / 0,4377
vegetable oils / 2,3287 (0,2210) / 0,0258
(0,0908)NS / 0,6566 / 0,4348 / 2,1908
(0,1751) / 0,0836
(0,0710)NS / 0,3913 / 0,8347
vegetables / 4,4702 (0,4056) / -0,0081
(0,0028) / 0,0917 / 0,4762 / 4,0577
(0,0893) / 0,0852 (0,0301) / 0,0968 / 0,5928
fruits / 4,5130 (0, 2134) / -0,0428
(0,0393)NS / 0,1114 / 0,3986 / 4,5037
(0,3071) / -0,0406
(0,0705)NS / 0,1038 / 0,6374
animal fats / 2,8429 (0,2278) / -0,1863
(0,0899) / 0,9272 / 0,4933 / 2,8520
(0,1794) / -0, 1901
(0,0684) / 0,9940 / 0,7721
Dairy products excluding butter / 4,1902 (0,4119) / 0,0926
(0,0886) / 0,1069 / 0,3570 / 3,5801
(0,3434) / 0,2246
(0,0736) / 0,7477 / 0,8218
eggs / 2,2230
(0,0845) / 0,0480
(0,0093) / 0,1741 / 0,4384 / 2,1143
(0,0607) / 0,0946
(0,0086) / 0,1818 / 0,1285
meat / 4,4590 (0,3810) / -0,0726
(0,0913)NS / 0,2612 / 0,5405 / 4,2528
(0,2867) / -0,0231
(0,0685)NS / 0,2942 / 0,5406
fish and seafood / 2,5073 (0,2800) / 0,0213
(0,0909)
NS / 0,7115 / 0,4378 / 2,5066 (0,2286) / -0,0216
(0,0657)
NS / 0,7147 / 0,0048
stimulants / 0,8755
(0,3131) / -0,0182
(0,0910)NS / 0,2127 / 0,4436 / 0,8609 (0,3105) / -0,0015 (0,0662)NS / 0,2227 / 0,4539
Source: Own calculation, (NS - denotes not significant at 0,05 significance level)
In above table NS-denotes not significant at 0,05 significance level, * - Yugoslavia excluded, standard errors are in parenthesis.
In the table 2 results of p-values for the Sargan test and the Arellano-Bond test for AR(2) are reported. In Sargan testnull hypothesis is that the instruments are uncorrelated with the IV residuals. If instruments were affecting through an omitted variable, then the Sargan test should reject the nullhypothesis. However, the large p-values reported in the Table 2 show that the instruments pass the test in all cases.
In the Arellano-Bond AR(2) testthe null hypothesis is that the errors in the first-difference regression exhibit no second-order serial correlation.In all reported in table instances, the p-values of AR(2) test are larger than 0,05, which indicates failure to reject the null hypotheses of second-order serial correlation of error terms.Failure to reject the null hypotheses of both tests gives support to our models.
Table 2includes only results for models which satisfy theno second-order serial correlationrequirement.
Other results were omitted because of lack of space. For models in which we discovered the lack of autocorrelation of the second order hypothesis, we tried to use a two-step procedure incorporating AR(2). However, the results were in general not promising. In that case either AR(2) still existed or we gathered different results of the convergence problem when we used Arellano & Bond and Arellano & Bover/Blundell & Bond procedure.
The positive and significant parameter β confirms the presence of unconditional beta-convergence.The speed of convergence can be computed as: s =−ln(1-β).
In the end it was only shown that fish and seafood have no coverage. Other products that did not yield unambiguous results were categorized in different subgroups of countries and periods.We studied Czechoslovakia, Hungary and Poland – three, now post soviet, countries with a similar geographical position. Results gathered for these countries permit in a majority of cases an unambiguous identification – for cereals, sugar and sweeteners, vegetable, dairy products excluding butter, animal fats, eggs we have observed beta-convergence, while for potatoes, pulses, vegetable oils, meat, fruits, fish and seafood, stimulants – a lack of beta-convergence.
Afterwards, we searched for sigma-converge in the group of all countries in which beta-convergence appeared. Figure 1 presentsthe trends of coefficients of variation.
Figure 1. Trends of coefficients of variation
Source: Own calculation
Based on the chart it is plausible to believe thatshare of calories from animal products and dairy products excluding butter (except forthe last four years) has shown sigma convergence. It is confirmed by a negative parameter of trend in the following models:
R2=0,97
R2=0,96
where: as - coefficients of variationof share of calories from animal products,
dp - coefficients of variationof dairy products excluding butter,
standard errors are in parenthesis.
Sigma-convergence of share of calories from animal products can be explained by a progressing alongside economical evolution phenomenon of growing share of highly-processed products, especially animal ones. For vegetables the situation is not so clear. For total calories we can believe that sigma-convergence does not appear. It can be explained by a variety of diets in the studied period, each depending on geographical position, consumer’s preferences and abiding to rules of a healthy lifestyle.
Concluding remarks
Studies in European countries have shown a significant divergence in consumptionstructure and stability of its profile during the period of 1961 – 2003. Divergence in consumptionstructure was caused mainly by climatic and cultural backgrounds. Nevertheless, it was shown that culinary preferences remained the same throughout the investigated period.
This paper tests for convergence offood consumption among European countries during the period 1961–2003.As is conventionally donein unconditional beta-convergence, we regress growth rate of food consumption per capita on initial level of food consumption per capita.
The following major findings can be drawn from results of models estimated for all considered countries:
- unconditional beta-convergence is confirmed for:total calories, share of calories from animal products, vegetables, milk,
- speed of convergence for calories ranges from about 4% for proportion of total calories derived from animal products and 7% for total calories to 17% for vegetable and near 100% for dairy products[8],
- for share of calories from animal productsand dairy products excluding butter,statisticallysignificant sigma-divergence is found,
- pulses,fish and seafood don’t converge unconditionally,
- we suspect remaining food commodities to not converge, but for these items obtained results are ambiguous.
Furthermore a group of three countries was considered: Czechoslovakia, Hungary and Poland. Beta-convergence took place in cereal, sugar and sweeteners, vegetable, dairy products excluding butter, eggs. It was shown that if a convergence existed in these countries then it possessed a higher rate than in all other countries.
Additional analysis is called for on the convergence of food consumption patterns, but these initial findings are encouraging.
Generally, diets do not converge, except for some products. This result must be taken into consideration when attempting to plan food policies for European countries or when reorienting production in these countries.
Total calorie intake was increasing. In some countries thisphenomenon, and increasingly sedentary lifestyles, are leading to rising obesity levels.
One reason that food consumption patterns should not be expected completely to converge among countries even if socio-economic and demographic factors do, is that culture is an important influence on behavior and cultural diversity has proved resistant to the pressures from foreign travel and global media.