University of Amsterdam
The price elasticity of meat demand in the Netherlands
Estimates of the price elasticity of Dutch meat demand, based on the Dutch meat consumption between 1996-2016 in order to explore the effectiveness of an environmental tax on meat consumption.Leonie Ernst, 10977007 25-6-2018
Bachelor Thesis, Economics & Business Subsection: Microeconomics
2
Statement of originality
This document is written by Leonie Tessa Ernst who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document are original and that no sources other than those mentioned in the text and its references have been used in creating it.
The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
3
Abstract
Global warming is partly caused by human consumption, of which the livestock sector is responsible for the biggest part of the greenhouse gas emissions. These emissions arise from the keeping of animals in great numbers. In order to lower these emissions, the production of meat products, and thus the meat consumption, should be abated. Policymakers have several instruments at hand in order to lower this consumption. Price-based instruments fit the reduction of this particular good best. In this research, the price elasticity of meat demand in the Netherlands is estimated, to see if the Dutch consumers are likely to reduce their consumption of meats in case of an environmental tax on meat products. The responsiveness of Dutch consumers to a change in the meat price is estimated by running Two Stage Least Squares regressions for a double-log and linear specification of Dutch meat demand. The regressions make use of data from 1996 to 2016 on disposable income, population numbers, prices of meat and fish, and lagged consumption. The resulting price elasticity is determined to be -0.880, which indicates that a 1% increase in the price of meat leads to a decrease in consumption of 0.88%. The so-called ‘meat tax’ is thus likely to be an effective measure in abating meat consumption, and consequently lowering greenhouse gas emissions.
4
Inhoud
1. Introduction ... 5 2. Literature review ... 6 2.1 Price elasticities ... 6 2.2 Demand models ... 7 2.3 Previous research ... 7 3. Method ... 10 3.1 Data ... 10 3.2 Homogeneous good ... 123.2.1 Ordinary Least Squares Regression ... 12
3.2.2 Instrumental Variables Regression ... 13
3.2.3 Two Stage Least Squares Regression ... 13
4. Results ... 15 4.1 OLS conditions ... 15 4.2 Logarithmic models... 15 4.3 Linear models... 17 4.4 Estimates of elasticities ... 18 5. Conclusion ... 20 5.1 Policy Implications ... 20 5.2 Discussion ... 21 6. References ... 24
Appendix A Testing for outliers ... 27
Appendix B Correlation matrix ... 28
Appendix C Log-log model Ramsey RESET test results ... 29
Appendix D Linear model Ramsey RESET test results ... 30
Appendix E Testing on endogeneity ... 31
Appendix F Price elasticities in the linear specification ... 32
5
1. Introduction
Global warming is caused by great amounts of greenhouse gas (GHG) emissions, partly the consequence of human consumption. Of all consumption, the consumption of meat and meat products has the largest environmental impact (Guinée et al., 2006).
One way to reduce these emissions is by abating the consumption of meat and meat products. Although many factors, such as income, population, taste, and health advertisement influence the consumers’ choice for meat (Mathijs, 2015), one factor that is stated to be crucial for meat consumption is its price. Therefore, politicians argue that the introduction of a so called ‘meat tax’ would be effective in reducing the GHG emissions in the Netherlands, as people will consume less meat due to its higher price (GroenLinks, n.d.; Partij voor de Dieren, n.d.).
According to Schmutzler & Goulder, consumption taxes can be effective if (1) the monitoring costs of emissions are high; (2) technological improvement for emission reduction is difficult; and (3) substitution possibilities between production outputs are good (1997). In their research, Wirensius, Hedenus, and Mohlin (2011) argue that all three conditions are fulfilled within the livestock sector. According to a case study in Denmark (Edjabou, & Smed, 2013), the consumption tax appears to be a cost-effective way to promote less-polluting food without negative health effects. However, such a measure will be more effective if the demand of the good on which the tax is imposed reacts strong to changes in price, measured by own-price elasticities.
The own-price elasticity of meat consumption in the Netherlands has not recently been estimated, providing little evidence for the effectiveness of an environmental tax on meat. If the price elasticity for the Dutch meat consumption differs negatively from 0, one could say that a tax on meat consumption could help decrease the demand for it. Studies have been conducted in other European countries as Denmark, Sweden, and Norway. These results range from -0.68 to -0.75, which indicate a relatively low responsiveness to price changes, as meats are considered to be inelastic. However, these results should not be generalised to the specific case of the Netherlands, as we are looking at a different population, and thus a different market.
In this research, the question what the own-price elasticity of meat demand is in the Netherlands for the most recent period is answered by estimates with Two Stage Least Squares regressions for two specifications of demand. The two models are used to check for consistency in the estimated values of the price elasticities. The concept of price elasticities will be discussed in section 2, after which the individual demand models are explained, and previous research is presented. In the third section, the data collection is introduced and the models are further specified. In section 4, the results from the regressions are presented, and lastly a conclusion, policy implications and a discussion are included in section 5.
6
2. Literature review
In this section, at first the concept of price elasticities is defined, and its specifications are explained. Secondly, the demand models are presented. Lastly, an overview of previous research on the price elasticity of meat demand is provided.
2.1 Price elasticities
(Own-)price elasticities1 of demand represent the expected percentage change in demand due
to a 1%-change in price (Pindyck, & Rubinfield, 2009, pp. 34-42). Price elasticities are useful in predicting the effectiveness of a consumption tax, as they predict consumers’ sensitivity for changing prices and the resulting change in demand under the ceteris paribus condition. A product is considered to be price elastic if the proportional change in the product’s price leads to a larger proportional change in demand. For normal goods, a negative price elasticity is assumed. Hence, meat products are price elastic if by a price increase of 1%, the demanded meat quantity decreases by more than 1%, and the elasticity has an absolute value larger than 1. Formula 1 represents the price elasticity of demand as expressed by Pindyck and Rubinfield (2009, p.49).
𝐸 = (𝑃 𝑄) (
∆𝑄 ∆𝑃) (1)
Price elasticities represent a unit-free measure of a change in demand. The average price elasticity of a certain time period is therefore a sufficient representation for that period. Short-run price elasticities concern a time span of less than one year (Pindyck, & Rubinfield, 2009, p. 40). It is assumed that it takes time for people to fully adjust their consumption patterns to price changes (Wirensius et al., 2011). Therefore the short- and long-run price elasticities can differ substantially. Food or beverage consumption changes only slowly after a price change, as consumers will begin consuming less from the price change onwards to an eventual lower level of consumption. The price elasticities in the case of meat demand are consequently considered to be larger for the long-term.
In addition, price elasticities represent a substitution and income effect (Snyder, & Nicholson, 2012, p. 140). The substitution effect represents the change in relative prices, whereas the income effect represents the change in real income and purchase power. Compensated price elasticities take only the substitution effect into account. Uncompensated price elasticities do take into account that price changes lead to a different reallocation of goods, due to the budget constraint that consumers face, and therefore show changes in demand due to both the substitution and income effects (Andreyeva, Long, & Brownell, 2010). Because of the considering of this extra effect, the uncompensated demand elasticities are often more elastic than their compensated counterparts (Gallet, 2009).
In this research, the uncompensated price elasticity is estimated, as in the case of an environmental tax on meat consumption, the price of meat will rise arbitrarily with the intention of lowering consumption. This means that the income effect is certainly relevant in identifying the changes in consumption that arise from both a change in relative prices as well as a change in purchasing power. This price elasticity is referred to as a Marshallian price elasticity, that is derived from individual Marshallian demand (Snyder, & Nicholson, 2012), which is discussed in section 2.2.
7
2.2 Demand models
Demand models are based on microeconomic theory assuming that consumers have a restrained budget and try to maximize their utility (Snyder, & Nicholson, 2012, p. 140). Therefore, the demand for a product does not only depend on its price, but also on the budget of the consumer, its disposable income, and the prices of other commodities and services. Next to that, consumers’ preferences may differ due to personal characteristics, such as age or gender.
In a simple Marshallian demand model, individual demand of 𝑥 is a function of the product’s price 𝑝𝑥 and the consumer’s income 𝑊. It shows the relationship between the price of the good and its consumed quantity (Snyder, & Nicholson, p. 142). However, by including only the price of the good examined, this is the only good assumed to exist on the market. In case of multiple goods, the prices of substitutes 𝑦 should be included in the model to identify the effects that other prices have on the demand for a certain good as well.
𝑥 = 𝐹(𝑝𝑥, 𝑝𝑦, 𝑊) (2)
In this research, both a linear demand model as well as its logarithmic transformation is used in order to estimate the own-price elasticities of meat demand. The basis of this analysis is that meat demand depends on an individual consumer’s disposable income, and the meat price. Next to that, the price of fish is included in the demand model as it is a substitute for meat.
2.3 Previous research
In the market for meat products, a market failure is present due to negative externalities. Much research has been conducted on the estimation of the price elasticities of meat demand to explore the effectiveness of an environmental tax on meat products.
To begin with the identification of the market failure, according to the literature, the livestock sector itself contributes for 18-22% to the global amount of greenhouse gas emissions (Steinfeld et al., 2006; McMichael et al., 2007). These levels of greenhouse gas emissions are not taken into account by the determination of the meat price, which leads to a negative externality. In the case of a negative externality, the private marginal utility of consumption does not correspond to the social marginal utility of it, which leads to inefficient market outcomes (Hindriks, & Myles, 2006, p. 236). This means that due to the excessive consumption of meat, the subsequently high production leads to great levels of emissions that could be abated in various ways. Next to the already mentioned environmental tax on meat products – where policymakers still have to decide whether the tax will be imposed on either consumption or production – emissions arising from the livestock sector can also be lowered by command and control instruments, such as emission rights or quota, or information provision (Lorek et al., 2008; Wirensius et al., 2011; Edjabou & Smed, 2011). However, the problems with command and control instruments are that it is difficult to measure the amount of emissions arising from production, but also complex to reduce the amounts of emissions due to technological advancements, as the emissions arise naturally from living animals (Wirensius et al., 2011). Information provision neither is an effective measure, as meat consumption is usually embedded in a culture, which means a shift in product choice will not easily occur when one has more information on a habit one already has (Olesen, 2010). Price instruments are thus the only measures that could succeed at including the social cost of meat production in the price of the goods.
8 A tax on consumption is preferred over a tax on production, as countries may lose their comparative advantages due to higher production costs, but also because of ‘carbon leakage’ that occurs if one country starts to pollute more due to the other country’s effort in abating emissions (Wirensius et al., 2011; Säll, & Gren, 2012).
In order to examine the effectiveness of these price changes as a policy instrument, price elasticities of meat demand that indicate how strongly consumers are expected to react on a price change, have been estimated in previous research2. In 1990-91, the price elasticities in Norway are
estimated to equal -0.87 for beef demand, -1.52 for pork demand, and -0.32 for chicken demand (Rickertsen, 1995). Burton, Young, and Cromb (1999) estimated the price elasticities for beef, pork, and poultry respectively at -0.763, -1.104, and -0.835 in the United Kingdom. In 2003, Rickertsen, Kristofersson, and Lothe estimated the price elasticities for meat in Denmark, Norway, and Sweden to be -0.75, -0.67, and -0.73. According to Andreyeva et al. (2010) the price elasticities for meats in the United States range from -0.68 to -0.75, and Gallet (2009) shows in a meta-analysis that the price elasticity of meat, considering homogeneous effects, is -0.851 across the globe, with a specific price elasticity for West Europe that equals -1.191. Wirensius et al. (2011) found that the price elasticities of food demand in the EU27 are approximately -1.30 for beef demand, -0.80 for pork demand, and -1.00 for poultry demand. All these estimates make use of different demand models, among which the Almost Ideal Demand System (AIDS) by Deaton & Mullbauer (1980) is often used.
The AIDS framework is useful in estimating individual demand functions so that elasticities can be derived. In the framework the budget shares of the good that is examined are linearly related to the log of total expenditure and the log of relative prices (Deaton, & Mullbauer, 1980, p. 322). The model thus examines the influences of expenditure and prices on the consumers’ purchasing behaviour. Säll, & Gren (2012) use the AIDS framework to estimate both own-price and cross-price elasticities of meat demand, where cross-price elasticities represent a change in demand of product 𝑥 as a result of a 1% price change in product 𝑦 (Pindyck, & Rubinfield, 2009, pp. 34-42). In the case of meat demand, the cross-price elasticities show the relations between different meat products that serve as substitute goods. Next to that, the AIDS framework makes use of a price index that relies on prices of the different product groups that are also included in the calculations of the budget shares. As there is an insufficient amount of data on these factors available for the case of the Netherlands, the AIDS framework is not used for estimating the price elasticities of total meat demand in this research.
Furthermore, in the Netherlands, the most consumed meat products fall under the three main categories, i.e., beef, pork, and poultry (Terluin et al., 2017). With the complete set of information, the AIDS framework could be used to estimate the cross-price elasticities between these three different meat products, but in this paper the entire food group is examined, and cross-price elasticities are beyond the scope of this research.
All in all, there is a large range of estimated price elasticities of meat demand, with the most recent estimate for the price elasticity of the Dutch meat demand originating from the orienting research by Boer, Bogers, Mangen, Van den Berg, and Bemelmans (2006), who estimated the price elasticity of meat consumption in the Netherlands to be -0.64 (p=0.000). In their research, weekly consumer data is used, which should be kept in mind as the estimate of -0.64 concerns the short-term rather than the long-term, which is considered in this research.
2 An overview of the in this section mentioned previous research on price elasticities of meat demand is
9 As this is already very different from the aforementioned price elasticity for West-Europe estimated by Gallet (2009), the results of other West-European countries such as the United Kingdom might not suffice as a comparable value for the price elasticity of Dutch meat demand. In addition, from 2011 onwards, the Dutch attitude towards daily meat consumption has shifted somewhat towards lowering meat consumption (Dagevos, Voordouw, Van der Weele, & De Bakker, 2012), which indicates that the price elasticity of meat demand might have changed – becoming more elastic – since these last estimates as well.
10
3. Method
In this paper, the price elasticity of meat demand is estimated under the main assumption of meat as a homogeneous good. The main reason for looking at meat products as a homogeneous good, is that consumers more often choose for meat products rather than for meat from a certain type of animal (Mangen, & Burrell, 2000). The whole range of meat products, i.e. beef, pork, and poultry, is considered as one good, after which the price elasticity for the entire food group is estimated. Other types of meat are for simplicity left out of this research, as they account for only 3.5% of total meat consumption, and the prices are hard to obtain3 (Terluin et al., 2017).
However, the variety in prices and meat products infer that the simplifying assumption of meat as a homogeneous good might leave out substitution effects that will occur in the case of increasing all meat prices by means of a set tax rate. If the price of for example beef rises by 1%, consumers might want to consume poultry as a substitute. Assuming a homogeneous good means that this substitution of beef by poultry will not take place, so that the exclusion of these substitutable meat products leads to lower price elasticity estimates than in the case of a heterogeneous meat market (Gallet, 2009). However, estimates in case of heterogeneous products are beyond the scope of this research, as data on meat consumption per animal type is insufficiently available.
In this section, the data used for this research is presented and the origin is specified. Then, the model in case of a homogeneous good is clarified.
3.1 Data
The estimation of price elasticities of meat demand in the case of a homogeneous good is based on annual consumer price indexes of meat that are collected from Statistics Netherlands for the period 1996-2016. The consumer price indexes have base year 2015. For the estimation of the price elasticities it is however necessary to have prices and not indexes. As this data is not readily available, the prices are calculated by using the prices of base year 2015, as given in the report by Bruyn, Warringa & Odegard (2018). Bruyn et al. published the prices and consumption of the separate categories of beef, pork, and poultry. The weighted average of these separate prices serves as the average meat price in 2015, and together with the CPIs provides the annual average meat price for the desired time period.
The annual volume mutations of consumption for the period of 1996-2012 are also collected from Statistics Netherlands. However, only the percentage change in consumption compared to the previous year is available, so the yearly consumption for the years 2012-2016 that is determined by Terluin et al. (2017) is used to determine the yearly consumption per capita for the period of which the volume mutations are available as well.
As a control variable, the annual growth of Dutch disposable income per capita is obtained from the database of the OECD for the years 1996-2016. Again, a particular amount is needed for the precise determination of disposable income per capita in the country. In the national accounts report 2016 by the CBS, the disposable national income (net) per capita in 2015 is set at €32,883. This in combination with the annual growth numbers provided by the OECD leads to national disposable income per capita for the years 1996-2016.
Next to that, the Dutch population itself serves as a control variable. As the amount of people within the country influences the production and consumption of every product, it is necessary to
3 Other types of meat are lamb, mutton, goat meat, horsemeat and veal. According to the consumption patterns of 2016 as
reported by Terluin et al. (2017), this consumption accounts for 3.51% of total meat consumption and is thus considered negligible.
11 include the population numbers of the Netherlands. This data is retrieved from Statistics Netherlands.
Moreover, the price of fish is included in the demand model, since fish is considered to be a substitute for meat consumption. The price of fish is thus influencing the demand for meat. The consumer price of fish is derived from the EU Fish Market report – 2017 edition by EUMOFA. In this report, the value and volume of fish consumption is reported. Dividing the value by the volume numbers for the year 2015, provides the price of fish for base year 2015. The annual prices for the entire period are then calculated the same way as the annual price of meat, but now by making use of the CPIs for fish products, published by Statistics Netherlands.
Other control variables considered to include in this model are personal characteristics such as age, gender, socio-economic class, and living area, but also habit and meat advertisement. Habit can be included in the model by adding a variable for the consumption in the previous period that serves as a representation of the anchoring point of consumption decisions. In addition, in their multilevel study in the Netherlands, Giskes, Turell, Lenthe, and Brug (2005) conclude that the area in which consumers live is not of great importance for someone’s dietary intake decisions. Next to that, the variables of age, gender, and socio-economic class are identified to be influencing the consumption choices (Smed, 2002; Angulo et al., 2003), but cannot be determined as only aggregate consumption, and not consumption per separate characteristic group, is available. Lastly, data on meat advertisement is not available. More specific data on Dutch meat consumption is strongly advised for further research.
Next to the control variables, instrumental variables are also needed in order to run the regressions. Instrumental variables that will be tested on usefulness are average temperature and subsidies on meat production per kilogram. The average temperature in the Netherlands is retrieved from Statistics Netherlands. The subsidies per kilogram of animal products are calculated by using the subsidies in million euros per product retrieved from the Eurostat database for the years 1996-2016, and dividing these amounts by the production of meat in kg, retrieved from the FAOSTAT database for that same period.
Important to note is that annual data is used instead of monthly consumer data. The estimated price elasticities are therefore interpreted as long-term elasticities rather than short-term elasticities.
Table 1: Data summary
Variable Formula form Entity
Meat consumption 𝑄 kg/capita
Price of meat 𝑃𝑚 €/kg
Disposable income 𝑊 €/capita
Population 𝐼 1 Price of fish 𝑃𝐹 €/kg Temperature 𝑇 ◦C Subsidies 𝑆 €/kg Subsidies 𝟐 𝑆2 (€/kg)^2 Lagged consumption 𝑄𝑖−1 kg/capita
12
3.2 Homogeneous good
The price elasticity of a homogenous good can be determined by means of a linear model as well as a logarithmic transformation of that model. In the linear model, it is assumed that there exists a linear relationship between the dependent and explanatory variables, whereas the logarithms allow for a nonlinear relation. In the log-log model, the natural logarithm of meat demand 𝑄𝑖 depends on a
constant 𝛼0, the log of the price of meat 𝑃𝑖, the log of disposable income per capita 𝑊𝑖, the log of the
population 𝐼𝑖, the log of the annual price of fish 𝑃𝑖𝐹, and the log of lagged consumption 𝑄𝑖−1.
ln 𝑄𝑖 = 𝛼0+ 𝛼1ln 𝑃𝑖+ 𝛼2ln 𝑊𝑖+ 𝛼3ln 𝐼𝑖+ 𝛼4ln 𝑃𝑖𝐹+ 𝛼5ln 𝑄𝑖−1+ 𝑢𝑖, 𝑖 = 1, … , 𝑛 (3)
In this model, the elasticity is specified as 𝛼1, and represents a constant.
However, the price elasticity of meat demand does not necessarily have to be a constant. Therefore, it is useful to also look at a linear model, in order to look for different outcomes for the price elasticity of meat demand in the Netherlands. Hence, the second model that will be addressed during this research has the meat demand 𝑄𝑖 depending on the meat price 𝑃𝑖 , disposable income per
capita 𝑊𝑖, the population 𝐼𝑖, the price of fish 𝑃𝑖𝐹, and lagged consumption 𝑄𝑖−1.
𝑄𝑖 = 𝛽0+ 𝛽1𝑃𝑖+ 𝛽2𝑊𝑖 + 𝛽3𝐼𝑖+ 𝛽4𝑃𝑖𝐹+ 𝛽5𝑄𝑖−1+ 𝑢𝑖∗, 𝑖 = 1, … , 𝑛 (4)
The price elasticity for year 𝑖 is determined as follows: 𝜀𝑖 =
𝛽1𝑃𝑖
𝛽0+ 𝛽1𝑃𝑖+ 𝛽2𝑊𝑖 + 𝛽3𝐼𝑖+ 𝛽4𝑃𝑖𝐹+ 𝛽5𝑄𝑖−1
(5)
3.2.1 Ordinary Least Squares Regression
Estimating price elasticities can be done by running Ordinary Least Squares regressions for both models. To run an OLS-regression, the following assumptions must hold (Stock & Watson, 2015, p. 247):
1. The conditional distribution of the error term 𝑢𝑖 equals 0 given the explanatory variables,
i.e., 𝐸(𝑢𝑖|𝑃𝑖, 𝑊𝑖, 𝐼𝑖, 𝑃𝑖𝐹, 𝑄𝑖−1) = 0;
2. The data (𝑃𝑖, 𝑊𝑖, 𝐼𝑖, 𝑃𝑖𝐹; 𝑄𝑖 ) is randomly collected and thus independently and identically
distributed;
3. Large outliers are unlikely;
4. There is no perfect multicollinearity.
In the case of meat consumption price is an explanatory variable of demand. However, Marshallian demand equals supply in equilibrium, which makes the price of meat an endogenous factor (Rickertsen, 1995). Due to this endogeneity, the first OLS-assumption is violated, leading to biased estimates. In order to overcome the arising endogeneity bias, we resort to Two Stage Least Squares (TSLS) regression, for which instruments are needed.
Moreover, as the data used for this research concerns a time series, the data is not independently and identically distributed. Newey-West estimators are used to overcome serial autocorrelation and heteroscedasticity in the error terms.
The third OLS-assumption is tested by plotting the data, to see whether any outliers exist. Lastly, the fourth assumption can be validated by checking for perfect correlation between all variables.
13
3.2.2 Instrumental Variables Regression
Instrumental Variables (IV) estimation is in this research used to correct for endogeneity. Adding an endogenous variable to a regression leads to inconsistent results, since changes in the explanatory variable are partly explained by the residuals of the regression. The mean of the residuals is assumed to be zero in an Ordinary Least Squares regression in order to obtain unbiased estimates. The rationale behind IV-regression is that an instrument is chosen so that the changes in the explanatory variable X are explained by changes in the instrumental variable rather than the error term.
The conditions for Instrumental Variables (IV) estimation are that: (1) the instrument 𝑍𝑖 is
relevant [𝑐𝑜𝑟𝑟(𝑍𝑖, 𝑃𝑖) ≠ 0], and (2) the instrument is exogenous [𝑐𝑜𝑟𝑟(𝑍𝑖, 𝑢𝑖) = 0] (Stock & Watson,
2015, pp. 470-474). Hence, the instrument should be influencing supply and prices, but not directly the demand itself. Instruments can be determined by use of economic theory and by testing the instruments on the aforementioned assumptions. In this research, the instruments are tested on relevance in the first stage of the Two Stage Least Squares regression.
In the case of meat, certain factors influence meat production. One of these factors is the average temperature in a country, as low temperature leads directly to a decrease in animal production (Warwick, 1976, p. 187). Since this reduction in production leads to a decrease in supply, and through this decrease to a higher price, the change in price is explained by an exogenous factor rather than demand itself. Therefore, the average annual temperature in the Netherlands is used as an instrument. The correlation between the price of meat and temperature is 0.9705.
Another instrumental variable used is the amount of subsidies on meat production. As agriculture is one of the sectors that is partly subsidized by the European Union (European Commission, 2018), the prices are artificially kept at a lower level. As this subsidy is not implemented because of the Dutch demand, but solely as a protection for the Dutch farmers to compete on the world market, the subsidy is exogenous, but affecting the price. Thus, the amount of subsidies on meat production is influencing the meat supply and prices, and satisfies both criteria for valid instruments. However, the correlation between the subsidies and meat prices is much lower (-0.2770) than the correlation between the price and temperature.
In addition, sales taxes could be considered to be a valid instrument. However, the sales tax on meat in the Netherlands has been equal for the entire period examined, and thus does not correct for any price changes. The tax percentage has been set at 6% in October 1986, and will be increased to 9% from January 2019 onwards (Tweede Kamer, 2017).
3.2.3 Two Stage Least Squares Regression
The IV regression is executed by means of the Two Stage Least Squares (TSLS) model, that consists of two parts. In the first stage, an OLS-regression is run with the endogenous variable X depending on the exogenous control variables of the total regression and the instruments chosen before. The intuition behind this, is that the endogenous variable is decomposed into a part explained by the instrument, and a part explained by the error term (Stock, & Watson, 2012, p. 472). By decomposing X into these two parts, the part correlated with the error term is left out of the regression model.
If the model is significant, the instruments and control variables do a good job in explaining the changes in the values of the endogenous variable. The fitted values of X are then used for the second stage, in which the Y, in this research meat demand, depends on the fitted values of X, and the same control variables used for the first stage. In this research, the endogenous variable is the price of
14 meat, and the control variables are assumed to be exogenous.
The first stage thus consists of regressing disposable income per capita, population, lagged consumption, the price of fish, and alternately temperature, subsidies per kilogram, and both instruments on the price of meat.
In the second stage, the fitted values of the first stage regression are used to run the regression with the predicted values of (the natural logarithm of) the price of meat. In this way, the endogeneity bias is overcome and the estimates are consistent. The model looks the following for each year 𝑖 for the logarithmic transformation of the model, and the linear model respectively:
Double-log model Stage 1 ln 𝑝𝑖 ̂ = 𝜑0+ 𝜑1ln 𝑧𝑖+ 𝜑2ln𝑤𝑖+ 𝜑3ln 𝑖𝑖+ 𝜑4ln 𝑝𝑖𝐹+ 𝜑5ln 𝑄𝑖−1+ 𝜉𝑖 (6) Stage 2 ln 𝑞𝑖 = 𝛼0+ 𝛼1ln 𝑝̂ + 𝛼𝑖 2ln 𝑤𝑖+ 𝛼3ln 𝐼𝑖+ 𝛼4ln 𝑝𝑖𝐹+ 𝛼5ln 𝑞𝑖−1+ 𝑢𝑖 (7) Linear model Stage 1 𝑝̂ = 𝜋𝑖 0+ 𝜋1𝑧𝑖+ 𝜋2𝑤𝑖+ 𝜋3𝑖𝑖+ 𝜋4𝑝𝑖𝐹+ 𝜋5𝑄𝑖−1+ 𝜐𝑖 (8) Stage 2 𝑞𝑖 = 𝛽0+ 𝛽1𝑝̂ + 𝛽𝑖 2𝑤𝑖+𝛽3𝑖𝑖+ 𝛽4𝑝𝑖𝐹+ 𝛽5𝑞𝑖−1+ 𝑢𝑖∗ (9)
After the first stage, the regression is tested on omitted variable bias, and serial correlation. This is done by use of the Ramsey RESET test, and Durbin-Watson test respectively. Also, the coefficients of the instruments used are tested on significance by their t-values. The regressions are executed with robust standard errors to correct for heteroscedasticity.
Subsequently, the second stage regression is tested on omitted variable bias, and serial correlation likewise. In addition, the entire TSLS regression is tested on endogeneity by use of the Durbin-Wu-Hausman test. Furthermore, the TSLS regression that relies on more than one instrument is tested on overidentification by use of the Sargan’s J-test. Lastly, the Newey-West estimator using two lags is used to correct for heteroscedasticity and autocorrelation.
15
4. Results
This section starts with testing the conditions for the regression models. Secondly, the logarithmic regressions will be analysed, after which the same is done for the linear models. If the instruments from the first stage prove to be valid instruments, the results of the second stage are presented and analysed. Important to note is that also the Newey-West estimator for two lags4 is used to correct for serial correlation, which changes the variances of the estimated coefficients. The model with the highest goodness of fit, measured by the R-squared statistic is chosen for the determination of the estimate of the price elasticity of meat demand in the Netherlands. Lastly, the price elasticities of demand are estimated and presented.
4.1 OLS conditions
The demand models presented need to fulfil the four assumptions explained in the previous section in order to lead to reliable estimates. As noted before, due to the endogeneity of the explanatory variable 𝑃𝑖 the outcomes of the OLS-regression have an endogeneity bias. The
instrumental variable regression is used to overcome this bias. Appendix E shows an overview of the results of the robust Durbin-Wu-Hausman scores that test for endogeneity. The null hypothesis that the variable are exogenous cannot be rejected, and the variables can be considered to be exogenous.
The second OLS-assumption concerns the independent and identical distribution of the data. As this research concern a time-series, this assumption does not hold. Due to this time series, the error terms might show a pattern that leads to inconsistent estimates. To solve this problem of serial correlation, the Newey-West standard errors are used.
The third OLS-assumption holds, as there are no outliers in the sample. Appendix A shows the overview of scatterplots of the variables used for the regressions. However, the subsidies per kilogram show a quadratic relationship over the entire period. Therefore, the variable 𝑠𝑢𝑏𝑘𝑔2 is added to the
first stage of both regression models.
The fourth assumption of no perfect multicollinearity also holds. Even though some variable are highly correlated, no variables are perfectly correlated, which means there is no perfect multicollinearity. The correlation matrix is included in appendix B.
4.2 Logarithmic models
The logarithmic variables represent the relative amounts of each variable. In the first stage, the logarithmic transformations of the instrumental variables and control variables are used separately and in combination to find the best instrument for the TSLS regression. The results of the first stage regression are presented in table 2.
The models that include both the instruments temperature and subsidies per kilogram show no significant results at a significance level of 5%. The reason for this might be that the subsidies provided to farmers are determined on the base of the amount of meat produced (European Commission, 2018). As a lower temperature leads to lower levels of meat production and consequently to less subsidies, this may lead to ambiguous changes in the price. It might be the case that temperature is correlated with subsidies, which would explain why adding both instruments to the regression does not lead to better estimates.
4 Two lags are used to correct for serial autocorrelation, since one lag is already included in the model in the form of lagged
16 Moreover, the Durbin Watson test statistics for the first two models show no sign of autocorrelation and are 1.820 for the first model, and 1.781 for the second model. The Ramsey RESET tests reveal that there is not enough evidence for omitted variable bias. These results are summarized in appendix C.
Table 2: First stage regressions of the log-log model
(1) (2) (3) (4) (5)
ln(price) ln(price) ln(price) ln(price) ln(price) ln(population) -1.997 1.721** 1.663 -1.597 -1.885 (1.534) (0.583) (0.911) (1.672) (1.566) ln(disposable -0.401 -0.282 -0.293 -0.357 -0.308 income) (0.192) (0.210) (0.241) (0.195) (0.216) ln(lagged -0.381* -0.209 -0.209 -0.351* -0.375* consumption) (0.157) (0.143) (0.149) (0.159) (0.150) ln(price of fish) 0.518** 0.320* 0.333 0.503* 0.468* (0.173) (0.121) (0.189) (0.175) (0.205) ln(temperature) 3.853** 3.246* 3.837* (1.136) (1.385) (1.359) ln(subsidies) -0.003* -0.003 -0.001 -0.000 (0.001) (0.003) (0.001) (0.003) 𝐥𝐧(𝐬𝐮𝐛𝐬𝐢𝐝𝐢𝐞𝐬)𝟐 -0.000 0.001 (0.001) (0.001) Constant 30.68 -23.51* -22.46 24.91 28.00 (24.02) (10.12) (16.17) (25.91) (24.93) N 21 21 21 21 21 R2 0.978 0.976 0.976 0.979 0.980 adj. R2 0.971 0.968 0.965 0.970 0.969 rmse 0.0152 0.0161 0.0166 0.0154 0.0157 Standard errors in parentheses
* p < 0.05, ** p < 0.01, *** p < 0.001
For the second stage, the instruments log of temperature and log of subsidies per kilogram are used separately to see which regression yields the best results.
In table 3 the results of the second stage and Newey-West estimators are presented for regression 1 with the logarithmic instrument temperature, and regression 2 with the logarithmic instrument subsidies per kilogram. In both specifications of the model, the parameters for the price of meat have the expected negative sign. However, only in the first regression the parameter for the price of meat differs significantly from zero (p=0.028), considering the Newey West estimator for the error terms. The J-test on overidentification did not need to be executed, as in both regressions only one instrument was used.
In the double-log model, it is found that the model which includes the log of the temperature in the regression as an instruments brings about the best result. Next to that, there is no evidence for omitted variable bias5 nor autocorrelation6, which means that this estimation appears to be rather
consistent in estimating the price elasticity of meat demand, that in this case equals -0.882.
5 See Appendix E.
6 The Durbin Watson test indicates no autocorrelation, and is 2.275 in the model with the log of the
17
Table 3: Second stage regressions and Newey West corrections
(1) (1) (2) (2)
ln(meat consumption) Newey West ln(meat consumption) Newey West
ln(predicted -0.882 -0.882* -0.769 -0.769 price of meat) (0.416) (0.361) (0.792) (0.611) ln(population) 1.643 1.643 1.405 1.405 (0.990) (0.933) (1.672) (1.417) ln(disposable 0.00126 0.00126 0.0391 0.0391 income) (0.163) (0.154) (0.291) (0.222) ln(lagged 0.583*** 0.583*** 0.608* 0.608** consumption) (0.131) (0.124) (0.229) (0.179) ln(price of fish) 0.212 0.212* 0.180 0.180 (0.125) (0.0868) (0.234) (0.173) constant -24.22 -24.22 -20.92 -20.92 (15.58) (15.29) (24.23) (21.47) N 21 21 21 R2 0.779 0.749 adj. R2 0.706 0.665 rmse 0.0137 0.0146
Standard errors in parentheses
* p < 0.05, ** p < 0.01, *** p < 0.001
4.3 Linear models
For the linear models, we start again with the first stage of the TSLS regression to identify in which way the instruments are best at estimating the changes in the price of meat. The results of the first stage are shown in table 4. Similar to the first stage in the logarithmic transformation of this model, the combinations of instruments in explaining the changes in the price of meat do not lead to significant results, but the instruments do show significant results separately.
Table 4: First stage regressions of the linear model
(1) (2) (3) (4) (5)
price of meat price of meat price of meat price of meat price of meat population -0.000* 0.000 0.000 -0.000 -0.000 (0.000) (0.000) (0.000) (0.000) (0.000) disposable -0.000* -0.000* -0.000* -0.000* -0.000 income (0.000) (0.000) (0.000) (0.000) (0.000) lagged -0.041** -0.025 -0.026 -0.0407* -0.037* consumption (0.012) (0.012) (0.016) (0.0154) (0.015) price of fish 0.429*** 0.368** 0.366** 0.429** 0.498*** (0.105) (0.098) (0.100) (0.109) (0.107) temperature 2.704*** 2.588 4.106 (0.646) (1.638) (3.203) subsidies -0.013* -0.011 -0.001 -0.037 (0.005) (0.032) (0.0108) (0.045) subsidies^2 -0.000 0.003 (0.002) (0.004) Constant -0.746 2.902 2.762 -0.450 0.364 (4.729) (4.861) (5.058) (6.395) (6.728) N 21 21 21 21 21 R2 0.982 0.980 0.980 0.982 0.983 adj. R2 0.976 0.973 0.971 0.975 0.975 rmse 0.103 0.110 0.113 0.106 0.106
Standard errors in parentheses
18 The first regressions 1 and 2 show that the instruments temperature and subsidies per
kilogram are significant, and explain the variance in the price of meat quite well, as measured by their
high R-squared (0.982 and 0.980 respectively).
It is however remarkable that the inclusion of the variable subsidies per kilogram squared does not lead to significant outcomes, since the scatterplot in appendix A shows an apparent quadratic relation. The reason for this might be that the correlation between the price of meat and the subsidies per kilogram is too low for significant results, and the relevance of the instrument can be disputed. In the second stage of the TSLS regression, of which the results are presented in table 5, the coefficients for the price of meat differ quite a lot per instrument used. Considering the fact that the relevance of the subsidies per kilogram can be doubted, and the R-squared of the model that relies on temperature as an instrument is substantially higher, the results of the latter are used for the final estimation of the price elasticities of meat demand. In appendix E, it is shown that there is in both linear models no evidence for an omitted variable bias, which indicates the estimates are reliable. Next to that, both models do not show autocorrelation according to the Durbin Watson statistics7. Again, no J-test was
necessary.
Table 5: Second stage regressions and Newey West corrections
(1) (1) (2) (2)
Meat consumption Newey West Meat consumption Newey West
price of meat -7.769* -7.769* -12.02* -12.02** (3.746) (2.869) (5.412) (3.124) pop 0.000 0.000 0.000* 0.000* (0.000) (0.000) (0.000) (0.000) disinccap 0.000 0.000 -0.000 -0.000 (0.000) (0.000) (0.001) (0.001) lagcons 0.578*** 0.578*** 0.468** 0.468** (0.139) (0.128) (0.156) (0.118) fishprice 1.504 1.504 2.650 2.650** (0.961) (0.801) (1.685) (0.858) _cons -23.30 -23.30 -35.35 -35.35 (27.21) (40.93) (39.50) (42.65) N 21 21 21 21 R2 0.803 0.687 adj. R2 0.737 0.583 rmse 1.027 1.040
Standard errors in parentheses
* p < 0.05, ** p < 0.01, *** p < 0.001
4.4 Estimates of elasticities
The resulting estimated parameters lead to different price elasticities. As explained before, the logarithmic model estimates a constant price elasticity of meat demand whereas the linear model estimates different price elasticities per period. In the logarithmic model, looking at the results from the TSLS regressions, model 1 has the highest goodness of fit as measured by the 𝑅2. This model estimates the relation between the price and individual meat demand in the Netherlands best. The price elasticity of meat demand in this logarithmic specification is a constant and equals -0.882 for a significance level of 5%. The model that makes use of subsidies per kilogram as an instrumental variable has no significant coefficient for the price of meat, meaning that in this specification the coefficient of the price of meat does not significantly differ from zero. As argued before, the relevance
7 Durbin Watson statistics for the first stage equal 1.991 for model 1, and 1.889 for model 2. For the second
19 of this instrument might be disputed, and the constant price elasticity of meat demand in Netherlands is estimated to be -0.882.
As mentioned in the previous subsection, for the final estimation of the price elasticities of meat demand in the linear specification, only the outcomes of the regression with temperature as a single instrument are used. The estimates for the annual price elasticities can be obtained by using the estimated coefficients from table 5 and formula 5 that combined lead to formula 10.
𝜀𝑖=
−7.769𝑃𝑖
−23.3 − 7.769𝑃𝑖+ 0.0000061𝑊𝑖+ 0.00000592𝐼𝑖+ 1.504𝑃𝑖𝐹+ 0.578𝑄𝑖−1
(10)
The results are summarized in table 7, and in table 8 the price elasticities for the last five years of the examined time period are shown8.
As the price elasticities in this specification are no constants, it fulfils to say that if we look at most recent estimates, the average price elasticity of meat demand equals -0.880, which corresponds with the constant price elasticity estimated with the logarithmic transformation of the model.
Table 7: Summary of the estimated price elasticities Obs Mean Std. Dev. Min Max Elasticities 21 -0.7858247 0.0658788 -0.8966517 -0.6894183
Table 8: Estimated price elasticities 2012-2016 Estimated annual price elasticity 2012 -0.8434 2013 -0.8967 2014 -0.8963 2015 -0.8811 2016 -0.8802
20
5. Conclusion
In this research the price elasticity of meat demand in the Netherlands is estimated based on the Dutch meat consumption between 1996 and 2016. The main objective of this research was to identify the responsiveness of Dutch meat consumption to a change in the price of meat, as that would implicitly prove the effectiveness of an environmental tax on meat consumption in order to abate the greenhouse gas emissions that are the result of meat production. For this research, two specifications of demand have been used to estimate the own-price elasticities.
The estimate for the price elasticity resulting from the double-log specification equals -0.882, whereas the average price elasticity for the entire period 1996-2016 equals -0.786 according to the linear specification. Both estimates are of the expected negative sign. Even though the historical data of the entire period is used for the estimation of the annual price elasticities, the average price elasticity of the last five years is a better representation of the current price elasticity than the average price elasticity of the entire period, as the estimated price elasticities are rising over time9. Looking at
the most recent estimates of these elasticities, the average price elasticity for the last five years of the examined period is -0.880. The results of both models are in this case consistent and do not differ much. Still, the linear model provides more reliable results, as the price elasticity is allowed to differ over time, and is not considered to be a constant as it is in the double-log specification. If we look at for example the price elasticity of the year 2000, which equals -0.701, the difference between the constant price elasticity resulting from the double-log specification and the linear estimate is quite large and inconsistent. Furthermore, it is sensible that the price elasticity might change over time, as with the growing awareness about health risks and environmental concerns, consumers might take less utility from meat consumption for any given price. Consequently, the Dutch responsiveness to changes in meat prices might be increasing, leading to this rising trend in price elasticities identified by the linear specification.
In this section, policy implications arising from this research are discussed, after which a discussion is provided on the limitations of this research.
5.1 Policy Implications
The estimated price elasticity of meat demand in the Netherlands of -0.880 indicates that meat is an inelastic good, and a price change of 1% leads to a 0.88% decrease in demand. This implies that an environmental tax on meat consumption would indeed be an effective measure in lowering the consumption somewhat, but that the price should always increase more than the desired proportional decrease in consumption.
The in section 2 identified negative externality that arises from meat production can be solved by either command and control measures, information provision or price instruments. As the first three appear not to be effective in the case of meat products, policymakers should directly resort to price instruments. However, as this research suggests, price instruments will have only a slight effect. As the price elasticities present consumers’ responsiveness under the ceteris paribus condition of all other things being equal, the behaviour of people may change if ‘all other things’ change as well. This mostly concerns the prices of substitutes, which are not only fish products, but also products such as plant-based meat substitutes, made of soy and tofu. If through the prices of these products also the relative price of meat increases substantially, consumers might try environmental-friendly products more often, and become more likely to include them in their dietary patterns in the longer term.
21 Depending on the cross-price elasticities, an environmental tax on meat products would therefore work best alongside a decrease in the prices of its meatless counterparts. Two beef burgers in the supermarket are for instance priced €2.09, while two vegetarian burgers cost €2.5910. This illustrates
that, assuming that consumers are indifferent between meat or no meat and the only difference is the price, consumers will most likely choose the beef burgers instead of the meatless variant. In order for customers to become indifferent between these two options, the price of the meat burgers should rise by 23.92%, which is a quite radical increase.
Consequently, it might be helpful to also make use of information provision, as through this, awareness of the externality increases, but consumers are also informed on the alternatives that are at hand. Especially in the case of meat products in the Netherlands, where meat is consumed on a daily basis (Van Rossum et al., 2016; Dagevos, 2016) the consumption pattern is most likely to change once consumers are nudged into the direction of less polluting foods by both information gathering and a change in price. Most ideally, only information provision would suffice in order to let consumers themselves decide to lower meat consumption out of intrinsic concerns about the environment. The main argument against the use of information provision is that meat consumption is deeply embedded in a culture, and therefore difficult to change. On the other hand, price instruments are steering measures through which policymakers limit customers’ freedom of choice. It is debatable whether it is desirable to let policymakers intervene in the dietary intake decisions of their citizens. Though, if information provision does not suffice, and the market is inefficient with prices not being adjusted for the environmental costs, it is the task of policymakers to in this case take the negative externalities of meat production and consumption into account.
In conclusion, according to this research, meat consumption in the Netherlands is slightly inelastic, and a change in its price will lead to a decrease in consumption that is of a smaller proportion than the price change itself. This will in turn lead to a lower level of meat consumption, once consumers get used to eating less meat on a daily basis. By adding the negative externalities of meat production to the consumer prices by means of an environmental tax, the market failure that arose due to the relatively low price of meat is solved and gas emissions decrease. This can go hand in hand with information provision, but is most likely to be – though only slightly – effective on its own as well.
5.2 Discussion
Even though the data for this research fulfilled the criteria for the regressions and led to significant estimates of the price elasticity of Dutch meat demand, there are some limitations to this research and its outcomes.
Demand models and variables
Firstly, to the chosen demand models, i.e. linear and double-log, control variables and instrumental variables have been added. Although disposable income per capita, population, lagged consumption, and the price of fish are all relevant in this model, it has to be noted that the price of fish, just as the price of meat, might be an endogenous variable. If this were the case, the estimates could be slightly inconsistent. However, as the endogeneity tests executed in Stata did not reveal an endogeneity bias, the estimates have been conducted without an instrument for the price of fish. This is however recommended for further research.
10 Prices come from the home-brand products of a big supermarket in the Netherlands, as published on their website.
22 Concerning instruments, in this research two instruments are used. The variable temperature was rather useful for the TSLS-regressions, whereas the variable subsidies per kilogram turned out not to be. The inclusion of instruments is however necessary to overcome the endogeneity bias arising from the simultaneous demand and supply models, so looking into the possibility of adding more instruments to the model is highly recommended. Regardless of the difficulty of finding new instruments, the sales tax rise that is due in January 201911 might be considered to add to the
specification in a few years’ time. For now, the focus should lie on instruments on the production side, such as the costs of animal nutrition of the amount of land used for keeping cattle, poultry, and pigs. Next to that, a few more control variables would be useful additions to the regression models. As mentioned in section 3.1, demographic factors such as age, gender, and socio-economic class are influencing a consumer’s choice for a certain meat product. In addition, meat and health advertisements might influence consumer behaviour as well, but no data is available. Education level is another factor that could possibly influence the environmental concerns and thus lead to a lower level of meat consumption. Therefore, more detailed information of meat consumption is desired for further research. Additionally, prices of not only fish, but also of other meat substitutes, such as vegetarian burgers made of soy or tofu or peas, could be included in the specification.
Furthermore, this research makes use of a small number of observations. Only 21 years are included in the research, which was partly due to a lack of available data, but mostly due to the decision to base the estimates on yearly consumption rather than monthly consumption patterns. The annual numbers are reliable in predicting long-term elasticities, as they are derived from larger samples and concern aggregate data. However, the difference between monthly and annual price elasticities is worth investigating as well.
Homogeneous versus heterogeneous goods
The most important assumption in this research is that meat products are considered to be homogeneous. Even though it might be the case that consumers are more likely to choose between meat products rather than type of animal products, there is a wide range of both meat prices and products. Accordingly, meat can be considered to be a heterogeneous good. In this case, the different meat products serve as substitutes for each other, and different price elasticities should be estimated. Several research has been conducted on the differences between the types of meat products. Gallet (2009) for example found that poultry worldwide has a rather low price elasticity, whereas beef is among the meat products with the highest price elasticity. Consequently, a meat tax, that would increase all meat products to the same proportion, might result in substitution of one meat product by another sort of meat. If beef products are for instance substituted by poultry products, this might lower the greenhouse gas emissions a bit, but the entire effect might still be relatively large.
In addition, these different kind of animals are also responsible for different proportions of greenhouse gas emissions. According to Fiala (2008), one kilogram of beef causes 14.8 kilogram of CO2, whereas this is only 1.1 kg for chicken, and 3.8 kg for pork. In the design of a meat tax, such differences should be taken into consideration alongside the substitution effects that can be derived from the cross-price elasticities of meat demand.
Further considerations
This research focuses on the consumption of meat products solely. As argued in the first two sections, meat production leads to higher emissions. Regardless of how the consumption of meat
23 leads to higher meat production and thus higher emissions, the consumption of dairy products has a similar carbon footprint. Of course less cattle is needed for the production of dairy products, as one animal can produce dairy products for several years, but meat products only once. Still, policymakers should take the dairy products into consideration if they want to lower the greenhouse gas emissions by adjusting the price to the social cost.
Moreover, efforts of the Dutch policymakers in including the environmental costs of meat production into its price, might be negligible on a global scale, if these efforts are not shared by other countries as well. Next to that, much of the Dutch meat production is exported to other countries, meaning that even if the Dutch consumers start behaving differently, the greenhouse gas emissions in the Netherlands would still be of great magnitude due to meat production. In this case, control and command measures might again be considered to not only lower consumption, but also production in
the country.
Lastly, it is debatable whether it is desirable for policymakers to interfere in the dietary choices of their people. In the Netherlands, there already exist consumption taxes on alcohol, tobacco, and mineral oils, such as petroleum and diesel (Belastingdienst, n.d.). These taxes are inferred in order to increase the Dutch people’s welfare, which is a form of paternalism. It is the task of policymakers to steer people in the direction that increases people’s own welfare without limiting people in their choices, which is called libertarian paternalism (Thaler, & Sunstein, 2003). A consumption tax on meat products is a limitation of the free choice of people, even when the intention of policymakers is to increase their people’s welfare.
24
6. References
Andreyeva, T., Long, M. W., & Brownell, K. D. (2010). The impact of food prices on consumption: a systematic review of research on the price elasticity of demand for food. American journal of
public health, 100(2), 216-222.
Angulo, A. M., Gil, J. M., Gracia, A. Z. U. C. E. N. A., & Kaabia, M. B. (2003). The impact of nutrient intake on food demand in Spain. Health, Nutrition and Food Demand, 153-172.
Burton, M., Young, T., & Cromb, R. (1999). Meat consumers’ long-term response to perceived risks associated with BSE in Great Britain. Cahiers d'Economie et de Sociologie Rurales, 50, 7-19. Boer, J. M. A., Bogers, R. P., Mangen, M. J. J., van den Berg, S. W., & Bemelmans, W. J. E. (2006). Het
mogelijk effect van prijsbeleid op de voedselconsumptie: een oriënterend onderzoek. RIVM
rapport 350020003.
Belastingdienst. (n.d.). Accijns en verbruiksbelasting. Retrieved June 23, 2018, from https://www.belastingdienst.nl/wps/wcm/connect/bldcontentnl/belastingdienst/douane_voo r_bedrijven/accijns_en_verbruiksbelasting/.
De Bruyn, S., Warringa, G., & Odegard, I. (2018). De echte prijs van vlees. Retrieved from https://www.ce.nl/publicaties/2091/de-echte-prijs-van-vlees
Dagevos, H. (2016). Exploring flexitarianism: Meat reduction in a meat-centred food culture. In Impact
of Meat Consumption on Health and Environmental Sustainability (pp. 233-243). IGI Global.
Dagevos, H., Voordouw, J., van der Weele, C. N., & de Bakker, E. (2012). Vlees vooralsnog
vanzelfsprekend: consumenten over vlees eten en vleesminderen. LEI, onderdeel van
Wageningen UR.
Deaton, A., & Muellbauer, J. (1980). An almost ideal demand system. The American economic review,
70(3), 312-326.
Edjabou, L. D., & Smed, S. (2013). The effect of using consumption taxes on foods to promote climate friendly diets–The case of Denmark. Food Policy, 39, 84-96.
European Commission. (2017). The EU Fish Market - 2017 Edition. Retrieved June 23, 2018, from https://www.eumofa.eu/nl/the-eu-fish-market. ‘
European Commission. (2018). Agricultural Policy Perspectives. Retrieved June 23, 2018, from https://ec.europa.eu/agriculture/sites/agriculture/files/statistics/factsheets/pdf/eu_en.pdf.
Fiala, N. (2008). Meeting the demand: An estimation of potential future greenhouse gas emissions from meat production. Ecological Economics, 67(3), 412-419.
25 Gallet, C. A. (2010). Meat meets meta: a quantitative review of the price elasticity of meat.
American Journal of Agricultural Economics, 92(1), 258-272.
Giskes, K., Turrell, G., Van Lenthe, F. J., Brug, J., & Mackenbach, J. P. (2006). A multilevel study of socio-economic inequalities in food choice behaviour and dietary intake among the Dutch population: the GLOBE study. Public health nutrition, 9(1), 75-83.
GroenLinks. (n.d.). Vleestax. Retrieved April 23, 2018, from
https://groenlinks.nl/standpunten/vleestax
Guinée, J., Heijungs, R., De Koning, A., Van, L., Geerken, T., Van Holderbeke, M., & Delgado, L. (2006). Environmental Impact of Products (EIPRO) Analysis of the life cycle environmental impacts related to the final consumption of the EU25.
Hindriks, J., & Myles, G. D. (2006). Intermediate public economics (Vol. 2). Cambridge: MIT press. Lorek, S., Spangenberg, J., Oman, I., (2008). Sustainable Consumption Policies Effectiveness Evaluation
(SCOPE2): Conclusion. Sustainable Europe Research Institute (SERI), Overath/Vienna.
Mangen, M. J., & Burrell, A. M. (2000). Decomposing preference shifts for meat and fish in the Netherlands. Journal of Agricultural Economics, 52(2), 16-28.
Mathijs, E. (2015). Exploring future patterns of meat consumption. Meat science, 109, 112-116. McMichael, A. J., Powles, J. W., Butler, C. D., & Uauy, R. (2007). Food, livestock production, energy,
climate change, and health. The lancet, 370(9594), 1253-1263.
Olesen, J., (2010). Fødevarernes Andel af Klimabelastningen. In: Vores Mad og Det Globale Klima – Etik
til en Varmere Klode. The Ethical Counsil, 2010.
Rickertsen, K. (1996). Structural change and the demand for meat and fish in Norway. European review of agricultural economics, 23(3), 316-330.
Rickertsen, K., Kristofersson, D., & Lothe, S. (2003). Effects of health information on Nordic meat and fish demand. Empirical Economics, 28(2), 249-273.
Säll, S., & Gren, I. M. (2012). Green consumption taxes on meat in Sweden (No. 2012: 10).
Schmutzler, A., & Goulder, L. H. (1997). The choice between emission taxes and output taxes under imperfect monitoring. Journal of Environmental Economics and Management, 32(1), 51-64. Steinfeld, H., Gerber, P., Wassenaar, T. D., Castel, V., Rosales, M., Rosales, M., & de Haan, C.
(2006). Livestock's long shadow: environmental issues and options. Food & Agriculture
26 Stehfest E., Bouwman L., Vuuren, D.P. van, Elzen, M.G.J. den, Eickhout, B., Kabat, P. (2009). Climate
benefits of changing diet. Climatic Change 95(1-2), 83-102.
Nicholson, W., & Snyder, C. M. (2012). Microeconomic Theory: Basic Principles and Extensions. Cengage Learning.
Partij voor de Dieren. (n.d.). BTW-verhoging voor vlees / vleestaks. Retrieved June 23, 2018, from https://www.partijvoordedieren.nl/items/btw-verhoging-voor-vlees-vleestaks.
Pindyck, R.S., & Rubinfeld, D.L. (2009). The Basics of Supply and Demand. Microeconomics (7th ed., pp. 21-61). New Jersey, United States: Pearson Education International.
Terluin, I. J., Dagevos, J. C., Verhoog, A. D., & Wijsman, J. C. G. (2016). Vleesconsumptie per hoofd van
de bevolking in Nederland, 2005-2015. Wageningen Economic Research.
Terluin, I. J., Verhoog, A. D., Dagevos, J. C., Horne, P. van, & Hoste, R. (2017). Vleesconsumptie per
hoofd van de bevolking in Nederland, 2005-2016. Wageningen Economic Research.
Thaler, R. H., & Sunstein, C. R. (2003). Libertarian paternalism. American economic review, 93(2), 175 179.
Tweede Kamer. (2017). Vertrouwen in de toekomst. Regeerakkoord 2017-2021. VVD, CDA, D66 en
ChristenUnie. Vergaderjaar, 2018, 34700.
Van Rossum, C. T. M., Buurma-Rethans, E. J. M., Vennemann, F. B. C., Beukers, M., Brants, H. A., De Boer, E. J., & Ocké, M. C. (2016). The diet of the Dutch: Results of the first two years of the Dutch National Food Consumption Survey 2012-2016.
Warwick, E. J. (1976). Climatic factors affecting red meat production. International journal of
biometeorology, 20(1), 186-192.
Wirsenius, S., Hedenus, F., & Mohlin, K. (2011). Greenhouse gas taxes on animal food products: rationale, tax scheme and climate mitigation effects. Climatic change, 108(1-2), 159-184.
27
Appendix A
Testing for outliersTesting on outliers by plotting all of the data points.
7 2 7 4 7 6 7 8 8 0 M e a t C o n s 1995 2000 2005 2010 2015 Year 6 .5 7 7 .5 8 8 .5 P ri c e m e a t 1995 2000 2005 2010 2015 Year 1 .6 e + 0 7 1 .6 e + 0 7 1 .7 e + 0 7 1 .7 e + 0 7 P o p 1995 2000 2005 2010 2015 Year 2 9 0 0 0 3 0 0 0 0 3 1 0 0 0 3 2 0 0 0 3 3 0 0 0 3 4 0 0 0 D is In c C a p 1995 2000 2005 2010 2015 Year 8 1 0 1 2 1 4 F is h P ri c e 1995 2000 2005 2010 2015 Year 1 0 1 0 .2 1 0 .4 1 0 .6 1 0 .8 T e m p 1995 2000 2005 2010 2015 Year 0 5 1 0 1 5 S u b k g 1995 2000 2005 2010 2015 Year
Appendix B
Correlation matrixIn the correlation matrix it is shown that there exists no perfect multicollinearity between the variables used in the regressions.
𝑸𝒊 𝑷𝒊 𝑸𝒊−𝟏 𝑰𝒊 𝑾𝒊 𝑷𝒊𝑭 𝑻𝒊 𝑺𝒊 𝑺𝒊𝟐 Meat consumption 𝑸𝒊 1.0000 Price of meat 𝑷𝒊 0.3511 1.0000 Lagged consumption 𝑸𝒊−𝟏 0.8317 0.5087 1.0000 Population 𝑰𝒊 0.4532 0.9800 0.5621 1.0000 Disposable income 𝑾𝒊 0.5394 0.8892 0.5792 0.9139 1.0000 Price of fish 𝑷𝒊𝑭 0.4806 0.9755 0.5862 0.9811 0.9443 1.0000 Temperature 𝑻𝒊 0.4416 0.9705 0.5763 0.9915 0.8879 0.9570 1.0000
Subsidies per kilogram 𝑺𝒊 0.2131 -0.2770 -0.0341 -0.2716 -0.1109 -0.1553 -0.3719 1.0000
(Subsidies per kg)^2 𝑺𝒊𝟐 0.2064 -0.1236 -0.0536 -0.1158 -0.0001 -0.0162 -0.2272 0.9616 1.0000
