Agribusiness Review - Vol. 8 - 2000

Paper 3
ISSN 1442-6951

Beef Consumption in Japan: What can be learnt from Sub-National Data?

Ruth Stroppiana and Paul Riethmuller
Department of Economics, The University of Queensland,
Brisbane Qld 4072, Australia. Ph: (07) 3365 6573 Fax: (07) 3365 7299
22 June 2000

Abstract

Japan consists of four large islands - Hokkaido, Honshu, Shikoku and Kyushu - and roughly 7 000 smaller islands and islets. In terms of natural terrain and climatic conditions Japan is a diverse country. There are also differences in the history, life styles and dietary habits of people living in different parts of Japan. This paper investigates the relationship between the consumption of beef and income, prices, and selected socio-economic factors in nine Japanese regions. The analysis found that consumption of beef at the regional level is influenced to differing degrees by income and by the prices of substitutes. In the heavily populated Kanto region, for example, containing the metropolises of Tokyo and Yokohama, the demand for beef was found to be not very responsive to changes in income, compared to the predominantly rural region of Hokkaido. This suggests that changes in income will have a relatively small impact on beef consumption in the Kanto region, compared to its effect on beef consumption in Hokkaido. A more general conclusion that can be drawn from the results is that programs designed to increase beef consumption in one part of Japan may need to be modified for other parts of the country if this same objective is to be achieved.

Keywords: beef consumption, elasticities, region, Japan.

1. Introduction

The purpose of this paper is to determine whether there are differences in the way economic and socio-economic factors influence beef consumption in different parts of Japan. Beef was chosen for investigation because Japan's beef consumption in aggregate is large. In 1994, for example, total demand was 1 450 000 t. while preliminary data for 1998 indicated that demand had grown to about 1 500 000 t. ( ALIC 2000 , p.2). On a per person basis, consumption is not particularly high, certainly when compared to Australia or the United States. In 1997, it was about 8 kg, compared with 6.1 kg in 1990.

Australia and the United States supplied 646 876 t. of beef to Japan in fiscal year 1998, or about 54 per cent of demand. Yet despite its importance as a market, analysts have not had great success in predicting the course of Japanese demand for beef. The Australian Bureau of Agricultural and Resource Economics (1994) , for example, estimated in late 1994 that Japanese imports of beef in calendar year 1998 would be 989 000 t. The actual level of imports for 1998 was 677 288 t. ( ALIC 2000 ). A number of factors no one could reasonably have expected at the time that the Bureau made its forecast could be used to explain the lower than expected imports. These include the slowing of the Japanese economy and a series of health scares involving beef in Britain, Europe and the United States that made Japanese consumers hesitant to buy beef.

The regional approach used in this study should add to the level of understanding of the Japanese market and help those involved in monitoring future developments in this market. A further reason for undertaking this study is to add to earlier analyses which have used regional data to investigate Japanese milk consumption ( Stroppiana, Kobayashi and Riethmuller, 1998 ) and variations in the growth rates of different food items in Japanese diets ( Riethmuller and Stroppiana, 1996 ).

2. Regions : Definitions and Differences

The Japanese Government makes use of nine broad regions covering Japan's 47 prefectures in its statistical reporting of developments in the economy. The prefectures making up these regions are analogous in broad terms to the states of the United States and Australia and Canada's provinces. The most northern of the nine regions, Hokkaido, consists of a single prefecture. All of the other regions are made up of several prefectures. As can be seen from the data in Table 1, the regions vary in area, population density and industry mix.

Table 1 Definition and Characteristics of Japan's Regions

Sources: Statistics Bureau (1996) (1996); Statistics Bureau (1998 (1998); Statistics Bureau (1999) .

Throughout Japan, the core of the diet is of rice, fish and vegetables. Ingredients used in dishes and the methods of preparing and cooking food vary across Japan, resulting in often quite distinctive regional cuisines. These differences reflect such factors as variations in climate and topography; history and culture; government policies; regional economic performance; and the availability of agricultural land. The following are examples of regional differences in food:

In preparing the popular beef-based dish sukiyaki , chefs in the Tokyo Yokohama area make use of thin slices of beef, while in Nagasaki, the beef is ground. In northern Honshu, wild boar or deer might be used Richie (1992 ). The dish was first documented in the 19 th century as the name given to a dish prepared from the meat of ducks or geese broiled on the top of a spade and basted with soy sauce ( Kodansha, 1998 , 1998);

In Osaka eels are slit along the belly, while in Tokyo eels are slit along the back. The reason for this dates back to when Tokyo was a samurai town. Splitting the eels along the belly reminds people living in Tokyo of ritual suicide ( AgExporter, 1996, p.23 ); and

Japanese noodles come in many varieties, but perhaps the most popular are the wheat-based noodles associated with Osaka and the south of Japan, and the buckwheat-based noodles associated with old Edo, or present day Tokyo, and the northern part of the country.

Compared to other high-income countries, Japanese people are not large consumers of meat in general and beef in particular. For much of its documented history, the Buddist code of non-violence towards four-legged animals maintained a ban on the consumption of meat from cows, horses and monkeys. However, meat was still eaten despite this ban. For example, boar and deer were eaten as medicinal preparation to ensure long life. Deer were known as yama kujira (mountain whale). Richie (1992, p.49) points out that "since the whale was thought to be a fish and since Buddist prophets said little about consuming them, one might safely consume these creatures".

Figure 1 contains data on the consumption of beef for each of the regions in 1974 and 1994. It is apparent from this figure that in 1974, there were large differences in per person beef consumption amongst the nine regions. For example, average daily consumption of beef in Tohoku and Kinki were 3.5 and 27.7 grams respectively. By 1994, per person beef consumption in Tohoku had increased by almost 15 grams to 18.4 grams per day, while in Kinki average daily consumption had increased less than 1 gram to 28.6 grams.

Figure 1
Source: Ministry of Health and Welfare (1974 and 1994) National Nutrition Survey

3. Data , Model and Method

It can be shown that under certain assumptions, the demand for a commodity depends upon income, the price of the commodity and the prices of substitutes and complements ( Intriligator, 1971 ). In empirical work, other factors such as expenditure on advertising, government intervention and socio-economic change are often included in demand models since these are thought to bring about shifts in demand.

Knowledge about the extent to which socio-economic characteristics affect food consumption is important because if the decline in the consumption of a particular food has primarily resulted from the effect of prices, then producers of that food should focus on reducing production and marketing costs to maintain or expand its market. However, if the shifts in consumption are the result of changes related to health concerns and lifestyle, then the product may need to be altered to take account of this.

A number of studies of Japanese food consumption have investigated the role of socio-economic characteristics in influencing food consumption. For example, Morishima, Aita and Nakagawa (1992) suggest that age is the most important socio-economic factor affecting food consumption in Japan. Saski (1993) found that the average number of people in the household in Japan was an important variable in his analysis of a range of food items.

Other researchers, for example Tokoyama and Egaitsu (1994) and Tokoyama (1995) , attribute a significant proportion of the changes in consumption of a large number of food items that they studied, to factors such as convenience and health concerns. Riethmuller, Smith Morison, Nagano, Kobayashi, Koizumi, Jussaume (1995) present survey data highlighting the importance of food safety as a factor in influencing meat purchasing decisions, while the importance of social and religious factors is discussed by Longworth (1983) and by Field (1986) .

Most of what has been written on changes in food consumption in Japan has made use of national level data (see, for example, Higuchi 1991 and Kamiya 1997 ). Relatively little analysis has been carried out at the regional level. A question that is worth considering is whether the demand for a food such as beef, responds to economic and socio-economic factors the same way in different parts of Japan. To help resolve this, per person consumption of beef was examined with the same set of explanatory variables for each of the nine regions listed in Table 1.

3.1 Factors Included in the Analysis and Data Sources

The data on beef consumption used in this study as the dependent variable came from Kokumin Eiyo Chosa, the National Nutrition Survey. This is a cross sectional survey conducted annually by the Ministry of Health and Welfare (1994). The survey involves approximately 6000 households in all parts of Japan, and covers all food and beverages taken in three "normal" days in November. The sample period is 1974 to 1994.

Per person expenditure in each region - used as a proxy for income - was taken from the Annual Report on the Family Income and Expenditure Survey ( Statistical Bureau , 1996) and then deflated by the CPI (1990=100). The new variable was used as a measure of per person real income in each of the nine regions. Per person real income in Japan increased over the period 1974 to 1992. From 1992 to 1994, however, growth tapered off, almost becoming zero in the recession of the early 1990s. There was some variation in the growth rate and real income amongst the regions. For example, throughout the study period, people living in Kanto - Japan's manufacturing and commercial centre - had noticeably higher average real incomes. However, amongst the remaining eight regions, there was little variation in per person average real income. Fujita and Tabuchi (1997) provide a detailed review of post-war regional growth in Japan. They found there were substantial interregional differences in income until 1970. Since then, differences have decreased due to factor migration within Japan, the relocation of industries off shore, and changes in Japan's industry mix towards technology and service intensive industries. Income in each of the regions was expected to be positively associated with consumption since beef is unlikely to be an inferior good.

Annual retail price data for beef in each of the nine regions were taken from the Statistics Bureau (1999) . These data are based on a retail price survey conducted in 177 cities, towns and villages in Japan. Prices are normal retail prices, excluding sale prices. Each retail price series was deflated by the CPI to obtain real prices. Due to the limited degrees of freedom, only one substitute was selected to include in the analysis. Some judgment was needed in choosing a close substitute. Whilst pork and chicken are generally considered to be substitutes for beef, previous research has found relatively low cross-price elasticities of demand for beef with respect to these foods in Japan. Therefore, the real retail price of fish was used as the substitute for beef. Over the period of the study, beef and fish prices steadily increased along with the general price level. However, the rate of increase slowed in the latter half of the study period. There was substantial variation in the real retail prices of beef and fish amongst the regions between 1974 and 1994. It was not obvious why these differences might have arisen. A negative relationship between the consumption of beef and the price of beef was expected while a positive relationship between consumption of beef and the price of fish would tend to support the argument that beef and fish are substitutes.

Since the mid-1950s, there has been a decrease in the average number of people living in Japanese households. There are a number of reasons for this. These include an increase in the number of single person households; a growth in the number of single parent households; a decrease in the average number of children per family; and a decline in the incidence (except in rural areas) of elderly parents living with their children and grandchildren. In 1995, about 11 million of Japan's 40 million households were single person households ( Asahi Shimbun , 1998, p.58). The increase in the number of single person households has been very high amongst the 18 to 25 years old age group.

This is probably because many in this age group are university students who leave their home town to attend university. It is common practice for these students to live by themselves in single room apartments. In addition, the growth in the number of single person households partly reflects the common practice of Japanese salarymen living in another city alone, while other family members remains in the family home.

The growth in the number of single person households is one of the main factors behind the shift towards the consumption of convenience and highly processed foods. Although a substantial portion of beef (about 50 per cent in recent years) is consumed outside the home, a substantial quantity is still cooked in the home. Since beef takes time to prepare, single person households are unlikely to go to the trouble of preparing a beef meal for themselves. This could result is a positive relationship between beef consumption and household size.

On the other hand, households with many members may not be able to afford beef (eaten in the home or in restaurants) and this could result in a negative relationship between consumption and household size. Data on the average number of people in the household in each of the nine regions over the period 1974 to 1994 were taken from the Statistical Bureau (1996) 1 . As just explained, the relation between beef consumption and household size is unclear.

Japan's population density of around 339 people per square kilometre places it amongst the most densely populated countries in the world. Population density within Japan varies widely from the crowded metropolises that comprise Kanto and Kinki to the agricultural areas of Hokkaido and Kyushu. Regions with a low population density are likely to be more agriculturally oriented and – it might be argued - are likely to have greater per person consumption of traditional foods than the more densely populated regions. If this is true, then a positive relationship between population density and consumption of beef (a non-traditional food) may be found. Population density per square kilometre for each region over the period 1974 to 1994 was calculated using population and land area data taken from the Statistics Bureau (1999 ). In this study, it is used as a measure of the agricultural orientation of a region. 2

The final variable included in the analysis was a time trend. This was intended to capture the effect of other factors that have been increasing with time, but which are difficult to measure quantitatively. These include overseas travel and westernisation. It was measured the same way for all regions. A positive relationship between time and consumption of beef in each of the regions is expected.

3.2 The Model and Method

In the analysis, a single equation approach was taken since developing a system of equations was outside the scope of the study. Moreover, since the purpose of this research was to investigate whether there were differences in the way beef consumption in the nine regions of Japan responded to the economic and social factors outlined in the previous section, a single equation approach was considered able to meet this objective.

Economic theory provides some guidance as to which variables should be included in any analysis of food consumption. However, it does not provide guidance as to which is the most appropriate functional form for the estimating equation. Since there was no a priori reason to favour one functional form over another, the Box-Cox transformation was used to allow the data to choose the most appropriate form of the model. To test the null hypothesis that a specific functional form is not significantly different from that of the maximum likelihood functional form, the likelihood ratio test was used. For most of the estimated equations, the double-log specification provided a reasonably close approximation to the best fitting non-linear model. Therefore, the results from the double-log model will be the ones discussed. The double-log functional form has an added advantage in that the estimated regression coefficients can be interpreted directly as elasticities.

Using 'ln' to denote the natural logarithm, the general model estimated was as follows:

ln Q_jt = a _0j + a _1j ln Y _jt + a_2j ln P_jt + a_3j ln S_jt + a_4j ln H_jt + a_5j ln D_jt + a_6j T_t + e_jt

Where Q _jt = per person consumption of beef in region j in year t

Y _jt = real per person expenditure in region j in year t

P _jt = real retail price of beef in region j in year t

S _jt = real retail price of fish in region j in year t

H _jt = average number of people per household in region j in year t

D _jt = population density in region j in year t

T _t = time in year t

j = region

t = year, 1974 to 1994

a _ij = parameters to be estimated, for i = 1 to 6

e _jt = error term

In estimating this model, the homogeneity condition - this restricts the sum of the price and income elasticities to zero - was imposed and tested at the 5 per cent level of significance using an F test. A high degree of multicollinearity among the explanatory variables was found in all the estimated equations. In a number of instances the pair-wise correlation coefficients between any two regressors was in excess of 0.8. Therefore, ridge regression was used in order to obtain "improved estimates" of the coefficients on the explanatory variables. Ridge regression involves adding a constant k (0< k <1) to the variances of the explanatory variables.

Unlike OLS, the estimated coefficients from ridge regression are biased but generally have a lower variance. By accepting some bias in the estimators, using ridge regression reduces the variance and an overall reduction in the mean square error (MSE) when the degree of multicollinearity is high. A difficulty with using ridge regression is deciding on the value of k . The larger the value of k , the larger the bias but the smaller the variance. 3

Thus, there is a trade off between bias and variance. Since it is undesirable to introduce too much bias, the smallest possible value of k should therefore be selected. The value of k for each model was chosen using a ridge trace. 4 The guidelines suggested by Hoerl and Kennard (1970) were used in order to select the "best" minimum value of k . Simply stated these guidelines are when: the system stabilises and has the general character of an orthogonal system; the coefficients have reasonable absolute values; the coefficients change to the proper signs; and the residual sum of squares (RSS) is not inflated to an "unreasonable" value.

Each equation was tested for first-order autocorrelation. This was done by the calculation of the Durbin-Watson statistic as well as examining the residual plot of each estimated equation to determine the size and pattern of the residuals. Because of the limited number of observations the estimated equations were not tested for higher order autocorrelation.

4. The Results

The regression results presented in Table 2 reveal that all of the estimated equations for Japan had relatively high R² statistics. This indicates that the variables included in each of the estimated equations explain a large proportion of the variation in per person consumption of beef in the regions. As can be seen in Table 2, the Durbin-Watson statistics for each of the estimated equations were close to two. Therefore, no corrections for first-order autocorrelation was made.

Table 2 Coefficients for Regional Demand for Beef

Notes: Three, two and one asterisks indicate significance of estimated coefficients using a two tailed test at the 0.1 per cent; 1 per cent; and 5 per cent level of significance, respectively.

All the income coefficients were significant at the 5 per cent level or better, positive and less than one - an indication that beef is a normal good. However, the estimated elasticities varied widely among the nine regions, from 0.70 in Hokkaido to 0.17 in Kanto. Interestingly, the regions in the northern part of Japan, Hokkaido and Tohoku, have the largest values for the income elasticities. This suggests that consumption in these regions is more vulnerable to the effect of economic downturn than other parts of Japan (and vice versa ). On the other hand, Kanto and Chugoku had low values for the income elasticity, indicating that the quantity of beef consumed is relatively immune to change in income in these regions. Therefore, it may be that people living in these regions will demand better quality beef rather than greater quantities of beef in the future. ABARE (1994) estimated that a 1 per cent change in consumer expenditure (a proxy for income) would change aggregate beef demand in Japan by 0.28 per cent.

The coefficients on the price of beef were significant in only five regions and were negative and less than one, indicating that beef is relatively price inelastic. The price inelasticity of beef indicates that further reductions in consumer prices, due for example to reduced border restrictions or an appreciation of the yen, may have little effect on consumption. There was, however, some variation in the elasticities obtained, implying that Japanese consumers living in different regions react differently to changes in the price of beef. The lack of statistical significance of the coefficient on beef price in Hokkaido, Tohoku, Chugoku and Shikoku indicates that the beef price plays little part in influencing consumption. It might be that health concerns and product safety may be the important factors for consumers in these regions rather than price.

The real retail price of fish was included as a substitute for beef because they are both sources of animal protein and because fish has traditionally been an important part of the Japanese diet. However, the coefficients were only significant in four regions and in one case - Shikoku - the coefficient had a negative sign indicating, somewhat implausibly, that beef and fish are complements. Negative coefficients were also obtained for Hokuriku and Chugoku, but in these cases, the coefficient was not statistically different from zero. These results lend support to the view that the cooking methods and the way fish and beef are eaten is sufficiently different that they are not to be regarded as substitutes ( Kodansha, 1998, pp.716-728 ). Given the dominance of seafood in the dietary culture, it is hardly surprising therefore that the econometric results reported here show fish price to be not overly important in influencing beef consumption.

The coefficients for household size for each region were found to be negative and to vary widely among the nine regions, from -2.26 in Tohoku to -0.71 in Kyushu. In all regions but Kanto, the coefficients were significant. The negative coefficients suggest that as household size increases, per person consumption of beef falls. One possibility is that smaller households have higher levels of discretionary income, due to smaller fixed costs such as education and housing. This higher level of discretionary income would enable households with fewer members to buy more beef. In addition, young single people, who represent many of the single person households, would perhaps be more likely than families to eat hamburgers or to dine in Japanese fast food restaurants.

The coefficients for population density were highly significant in six of the nine regions. In the less densely populated regions - Hokkaido, Tohoku and Kyushu, which all happen to be major beef producing regions - the coefficients were not significant. A possible explanation for these results is that people in the heavily populated regions are less likely to have contact with Japanese farmers than people in rural areas, and hence would be more likely to eat non-traditional products such as beef. These results may also be picking up an income effect, since incomes in the more heavily populated areas tend to be higher than incomes in agricultural regions.

The time trend was significant in all regions with its positive value indicating that per person consumption of beef has increased with time. Technology, government policy, overseas travel, changes in the number and types of restaurants and westernisation of tastes are factors that have changed over time, and so they are likely to be some of the influences being picked up by the time variable.

5. Concluding Comments

The purpose of this paper has been to investigate the consumption of beef across different regions of Japan. To do this, regional beef consumption data were examined over the period 1974 to 1994. Since income and price elasticities of demand for beef appear to be different across regions, Japan's beef market should not be treated as a homogenous market by foreign beef suppliers.

The relatively low income elasticity estimates in a number of regions suggest that these regions are mature markets and that beef consumption levels may be nearing saturation levels. The different price elasticity estimates show that different pricing strategies could be used in different parts of Japan, although arbitrage would limit the size of any price difference. The lack of responsiveness of consumption to beef prices in four of the regions indicates that factors other than beef price drives demand. Food safety and health are factors that spring to mind. The finding that consumption of beef in Japan at the regional level is associated with household size and population density is important, not just because these variables have been shown to be significant, but because it illustrates that specific regional factors need to be understood in any marketing program for beef.

It should be noted that the analysis presented in this paper could be extended. For example, no account was taken of differences in beef quality. Whether the beef consumed in different regions is primarily the better quality Wagyu or lower quality dairy is just as important for an understanding of developments in the Japanese diet, as changes in the quantity of beef being consumed. There are many other differences between regions that could influence beef consumption, but were not included. Distribution systems differ between prefectures. In 1994, for example, 26 per cent of the 48 405 convenience stores in Japan were in just four prefectures - Saitama, Chiba, Tokyo and Kanagawa.

There were also differences in the number of large scale stores per 100 000 people. In Tokyo in 1991, there were almost 11, but in Nagasaki (in Kyushu), there were only 5.6. The percentage of households with members 65 years and older also is another factor where there are differences. In Tokyo, the percentage of households with a member 65 years and older was 22.4 per cent, but in Yamagata (in Tohoku), it was 47 per cent in 1995 ( Statistics Bureau, 1998 ). This factor is potentially of importance because older people are more likely than young people to want to maintain the traditional Japanese fish and rice based diet.

References

Australian Bureau of Agricultural and Resource Economics [ ABARE] (1994) North Asian Markets for Australian Beef. ABARE Research Report . 94.10, AGPS, Canberra.

AgExporter (1996) U.S. Seafoods Exporters: Fish for New Asian Markets. AgExporter . 8 (2), 22-23.

ALIC [Agriculture and Livestock Industries Corporation] (2000) Monthly Statistics. May, Tokyo.

Asahi Shimbun (1998) Japan Almanac 1998. Asahi Shimbun Publishing Company, Tokyo.

David, C . and Huang, J. (1996) Political Economy of Rice Protection in Asia. Economic Development and Cultural Change . 44 (3), 463-83.

Field, H . (1986) Some Issues Related to Wagyu Beef Supply and Beef Policy in Japan. Unpublished M.Agr.Sc. thesis. LaTrobe University, Melbourne.

Higuchi, T . (1991). Japanese Dietary Habits and Food Consumption in Committee for the Japanese Agricultural Session (ed), Agriculture and Agricultural Policy in Japan, XXI IAAE Conference, Tokyo, pp87-104.

Hoerl, A . and Kennard, W. (1970) Ridge Regression Biased estimation for Non-orthogonal Problems. Technometrics . 12 (1), 55-67.

Intriligator, M . (1971) Mathematical Optimisation and Economic Theory. New Jersey: Prentice-Hall, Inc. Chapter 7.

Japan Meat Conference (1998) The Meat Statistics in Japan. Sep.1, Tokyo.

Judge, G. , Griffiths, W., Hill, R. and Lee, T. (1980). The Theory and Practice of Econometrics. John Wiley and Sons, New York.

Kamiya, M. (1997) . Structural Change in Japan's Food System, Food and Agricultural Policy Research Center, Tokyo.

Kodansha (1998) The Kodansha Bilingual Encyclopaedia of Japan. Kodansha International Pty Ltd, Tokyo.

Longworth, J . (1983) Beef in Japan. University of Queensland Press, Brisbane.

Morishima, M., Aita , Y. and Nakagawa, M. (1992) Food Consumption in Japan, in L. Tweeten; C. Dishon; W. Chern: N. Imamura and M. Morishima, eds. Japanese and American Agriculture: Tradition and Progression in Conflict. Westview Press Inc., Colorado.

Ministry of Health and Welfare (1994) Kokumin Eiya Chosa (National Nutrition Survey). Ministry of Health and Welfare, Tokyo ( In Japanese ) (Including Previous Years).

Richie, D . (1992) A Taste of Japan. Kodansha International, Tokyo.

Riethmuller, P., Smith , D., Morison, J., Nagano, M., Kobayashi, S., Koizumi, S. and Jussaume, R. (1995) Consumer Attitudes Towards Meat Consumption: an Asian Pacific Perspective. Paper presented to the Annual Conference,

Economics Society of Australia. University of Adelaide, Adelaide, 26 September.

Riethmuller, P . and Stroppiana, R. (1996) Regional Variations in Diets in Japan. Pacific Economic Papers. 257 , Australia Japan Research Centre, ANU, Canberra.

Sasaki, K . (1993) The Structure of Food Demand in Japan: An Application of the Rotterdam System. Agribusiness . 9 (5) :425-439.

Statistical Bureau (1996 ) Annual Report on the Family Income and Expenditure Survey. Management and Coordination Agency, Tokyo (Including Previous Years).

Statistics Bureau (1998) Social Indicators by Prefecture. Management and Coordination Agency, Tokyo

Statistics Bureau (1999) Japan Statistical Yearbook. Management and Coordination Agency, Tokyo (Including Previous Years).

Stroppiana, R ., Kobayashi, K. and Riethmuller, P. (1998) Regional Differences in the Japanese Diet: The Case of Drinking Milk. Economic Analysis and Policy . 28 (1), 85-102.

Teklu, T .; Hassan, Z.; Johnson, S. and Stonehouse, P. (1992) "Empirical Demand Systems", in S. Johnson; P. Stonehouse and Z. Hassan (Eds.) Market Demand for Dairy Products. Ames: Iowa State University Press, pp.46-64.

Tokoyama, H . (1995) The Relationship Between Trends and Income Elasticities in Food Consumption Structure: An Attempt to Quantify Define ‘Maturity' in Food Consumption. Japanese Journal of Rural Economics . 67 (1) ( In Japanese ).

Tokoyama, H. and Egaitsu, F. (1994) Major Categories of Changes in Food Consumption Patterns in Japan 1963-91. Oxford Agrarian Studies . 22 (2), 191-202.

Footnotes

1 - Other studies modelling food consumption in Japan have included a variable for the number of people in the household as a proxy for convenience. For example, see: Sasaki (1993).

2 - David and Huang (1996) used population density as a proxy for comparative advantage in rice production in their study of government intervention in the rice industry.

3 - If k = 0 the ridge estimators reduce to OLS estimates.

4 - The ridge trace is a two-dimensional plot of values of ridge coefficients for a number of values of k .