Is there an animal food kuznets curve, and does it matter?

Frontuto, Vito; Felici, Tommaso; Andreoli, Vania; Bagliani, Marco Maria; Corsi, Alessandro Article Is there an animal food kuznets curve, and does it matter? Bio-based and Applied Economics (BAE) Provided in Cooperation with: Firenze University Press Suggested Citation: Frontuto, Vito; Felici, Tommaso; Andreoli, Vania; Bagliani, Marco Maria; Corsi, Alessandro (2025) : Is there an animal food kuznets curve, and does it matter?, Bio-based and Applied Economics (BAE), ISSN 2280-6172, Firenze University Press, Florence, Vol. 14, Iss. 1, pp. 19-35, https://doi.org/10.36253/bae-16172 This Version is available at: https://hdl.handle.net/10419/321808 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. https://creativecommons.org/licenses/by/4.0/ Bio-based and Applied Economics 14(1): 19-35, 2025 | e-ISSN 2280-6172 | DOI: 10.36253/bae-16172 © 2025 Author(s). Open access article published, except where otherwise noted, by Firenze University Press under CC-BY-4.0 License for content and CC0 1.0 Universal for metadata. Firenze University Press | www.fupress.com/bae Bio-based and Applied Economics BAE Citation: Frontuto, V., Felici, T., Andreoli, V., Bagliani, M. M. & Corsi, A. (2025). Is there an Animal Food Kuznets Curve, and does it matter? Bio-based and Applied Economics 14(1): 19-35. doi: 10.36253/bae-16172 Received: May 24, 2024 Accepted: September 23, 2024 Published: June 13, 2025 Data Availability Statement: All relevant data are within the paper and its Supporting Information files. Competing Interests: The Author(s) declare(s) no conflict of interest. Associated Editor: Oriana Gava Editor in Chief: Silvia Coderoni ORCID VF: 0000-0002-0627-9511 TF: 0000-0002-3779-4225 VA: 0000-0001-9163-011X MMB: 0000-0003-4918-8852 AC: 0000-0002-8519-6721 Is there an Animal Food Kuznets Curve, and does it matter? Vito Frontuto1,*, Tommaso Felici2, Vania Andreoli3, Marco Maria Bagliani1,#, Alessandro Corsi1 1 Department of Economics and Statistics, Università degli Studi di Torino, Torino 10153, Italy 2 Utrecht University School of Economics (U.S.E), Kriekenpitplein 21-22, 3584 EC Utrecht, The Netherlands 3 School of Biological Sciences, University of Western Australia,35 Stirling Hwy, Crawley, WA 6009, Australia # Sadly, Prof. Marco Maria Bagliani, passed away in July 2024 while this paper was under revision. We wish to dedicate this paper to a dear colleague and friend. *Corresponding author. E-mail address: [email protected] Abstract. Proteins from animal sources, including meat, and plant-based foods are essential for a healthy human diet. However, animal-based proteins have significantly higher environmental impacts (e.g., greenhouse gas emissions, deforestation, and water usage) and health risks (e.g., obesity, type 2 diabetes, kidney stones and cardiovascular diseases) compared to plant-based proteins. The consumption patterns of these proteins are strongly influenced by income levels. This study introduces the concept of an Animal Food Kuznets Curve by systematically analyzing the relationship between income and animal-based protein consumption. Utilizing a novel panel dataset spanning 28 years and covering 79 countries, we uncover an inverted U-shaped relationship between income and the consumption of animal-based and meat proteins. Our findings indicate that the turning points occur around 43,000-45,000 US$, corresponding to the 90th and 95th percentiles of the per capita income distribution in the sample. At these income levels, protein consumption is estimated at approximately 25 g/ day for meat and 52 g/day for animal-based proteins, as compared to recommended total protein intake of 45-56 g/day. These insights highlight the critical need for targeted policy interventions, such as taxes, nudges, and informational campaigns to promote sustainable dietary choices across all income levels. Our study provides empirical evidence for the importance of integrating economic and environmental policies to enhance global food sustainability. Keywords: protein consumption, consumption drivers, Environmental Kuznets Curve, mixed effects model, panel data. JEL Codes: Q54, Q56, C23. 1. INTRODUCTION Over the past 50 years, the global consumption of animal-based proteins, including meat, eggs, dairy, and seafood, has significantly increased in 20 Bio-based and Applied Economics 14(1): 19-35, 2025 | e-ISSN 2280-6172 | DOI: 10.36253/bae-16172 Vito Frontuto et al. both absolute and per capita terms (Bonnet et al., 2020; Marques et al., 2018; Pais et al., 2021). This growth has been mainly driven by increased meat consumption (Bonnet et al., 2020; Sans and Combris, 2015). According to OECD and FAO (2023), global per capita meat consumption has nearly doubled, rising by 87 percent from about 23 kg per person per year in 1961 to 43 kg per person per year in 2021. Similarly, other animalbased foods have seen increases, with milk consumption by 16 percent and egg consumption by 129 percent. This tendency is in accordance with the nutrition transition featuring increasing demand for animal-based foods when income rises (Popkin, 1993). However, diets rich in animal-based protein have been linked to adverse health and environmental outcomes (Tilman and Clark, 2014), while diets with a higher composition of plantbased proteins are associated with less damaging impacts (Galli and Moretti, 2024). Several studies have called for urgency in shifting protein consumption from animalbased sources to plant-based sources (Willett et al., 2019), especially in upper-middle income countries with sustained economic growth rates (Duro et al., 2020). Indeed, the increase in global meat consumption (kg/year per capita) between 1961 and 2021 has been driven mainly by countries with rapid economic growth such as South Korea (1,935 percent), China (1,774 percent) and Indonesia (398 percent). Several studies have demonstrated that the first global protein transition, marked by a significant increase in demand for animal-based protein over the last century, was closely linked to changes in real income (Sans and Combris, 2015). The more recent second nutrition transition, characterized by a stabilization or decline in animal-based protein consumption, particularly meat (Godfray et al., 2018; Marques et al., 2018; Vranken et al., 2014), may also be attributed to similar factors. Economic growth has initially promoted animal-based consumption and then it has slowed it down. This brought some scholars to claim the existence of an Environmental Kuznets Curve (EKC) for animal-based food consumption, which could be named Animal Food Kuznets Curve (AFKC). According to the EKC original theory, the environmental impact of economic growth increases in the first phase and subsequently declines (Grossman, 1995; Grossman and Krueger, 1991). If such a trend proved true for animal-based food consumption, it would decrease the urgency of policies aiming at curbing its consumption since income growth would automatically lead to its decline. Nevertheless, the existence of an AFKC is to be empirically verified, and its actual effect on global consumption is to be assessed. This paper aims at investigating interactions between protein consumption and income over the last 30 years. The research uniquely analyses protein intake from animal-based, meat and plant-based sources to understand the dynamics of change and the predominant factor of variation, i.e., income. While the existing literature has predominantly focused on meat consumption and its correlation with income (York and Gossard, 2004; Vranken et al., 2014), there is a noticeable gap concerning the consumption of protein from different sources. This paper aims to bridge this gap by comprehensively exploring differences in protein consumption across animal-based, meat and plant-based sources using a global panel dataset covering 28 years and 79 countries. The originality of this study is further highlighted by the application of the linear mixed effect model. This methodological advancement addresses cross-sectional dependence in errors within large panel datasets, thus enhancing the accuracy of parameter estimates compared to conventional fixed effects models. 2. NEGATIVE IMPACTS OF ANIMALBASED PROTEIN CONSUMPTION Animal-based products are an essential source of nutrients – proteins, among others – to humans. However, among protein-rich foods, those of animal-based sources produce higher greenhouse gas (GHG) emissions (Dyer and Desjardins, 2022; Errickson et al., 2021), use more land (Van Zanten et al., 2018) and water (Mekonnen and Gerbens-Leenes, 2020), cause more acidification and eutrophication (Godfray et al., 2018; Poore and Nemecek, 2018). Among animal-based foods, meat has a higher environmental damage potential than those derived from eggs, milk and seafood (de Vries and de Boer, 2010). Among meats, beef proteins have the highest impact on the environment (de Vries and de Boer, 2010; Gaillac and Marbach, 2021). There is an urgent need for transitioning to more sustainable protein sources, such as protein of vegetal sources – pulses, legumes and novel protein-rich foods (McClements and Grossmann, 2021) – which have a lower environmental impact (Mazac et al., 2022). Plantbased diets can reduce GHG emissions by 49%, land use by 76%, scarcity-weighted freshwater withdrawals by 19%, acidification by 50% and eutrophication by 49% (Poore and Nemecek, 2018). Another reason to reduce consumption of animalbased products, particularly meat, is related with the potential adverse effects of its excessive consumption on human health. A higher availability of animal-based protein consumption would benefit food-insecure countries, where fewer alternatives are available to access nutrients 21 Is there an Animal Food Kuznets Curve, and does it matter? Bio-based and Applied Economics 14(1): 19-35, 2025 | e-ISSN 2280-6172 | DOI: 10.36253/bae-16172 and micronutrients. Here, a higher animal-based protein consumption would increase food and nutritional security. By contrast, the developed world, if anything, consumes an excessive amount of proteins (Aiking and de Boer, 2020). For instance, while the Lancet Commission on healthy diets suggests that an “adequate protein intake for adults is 0.8 g/kg bodyweight, which is 56 g/ day for a 70-kg individual” (Willet et al., 2019) and the European Food Safety Authority (EFSA) sets an average requirement intake of 46 g protein per capita per day (Agostini et al., 2012), protein intake in the EU is around 82 g per day, of which 49 g from animal-based sources and 33 g from plant-based sources (Simon et al., 2024). This aspect highlights substantial inequalities of the food systems between developing and developed world, and also represents an increased risk for human health. Meat consumption contributes to global obesity (You and Henneberg, 2016), higher risks of type 2 diabetes (Malik et al., 2016), kidney stones (Asoudeh et al., 2022), cardiovascular disease mortality (Zheng et al., 2022), cancer mortality (Huang et al., 2021) in the specific, colorectal, breast and prostate cancer (Cellura et al., 2022; Gonzalez et al., 2020) and more generally all-cause mortality (Sun et al., 2021). Conversely, diets rich in plant-based proteins, such as legumes, nuts and seeds, while sufficient to achieve full protein adequacy in the developed world (Mariotti and Gardner, 2019), seem to confer protection against the incidence of cancers (Gonzalez et al., 2020) and to reduce global mortality (Springmann et al., 2016). Increasing the share of plant-based proteins will provide significant health and environmental co-benefits (Bonnet et al., 2020; Stylianou et al., 2021). This study aims to assess the relationships between income and different protein sources to highlight potential differences that can be useful to understand the impact of policies. The paper will discuss the relationship between food consumption and income using existing literature, which, however, rarely took into consideration protein sources other than meat, and explains the theory behind the model in Section 3. We will then outline the data and the econometric strategy we chose to apply to describe this relationship in Section 4. The results of the estimated models are presented in Section 5 and their implications are discussed in Section 6 and 7. 3. THE RELATIONSHIP BETWEEN ANIMAL-BASED PROTEIN CONSUMPTION AND INCOME Rising real Gross Domestic Product (GDP at constant prices) over the last century has been identified as the root-cause of a global nutrition transition. The transition encompasses a shift towards animal-based sourced proteins in general (Gerbens-Leenes et al., 2010; Sans and Combris, 2015) and proteins from meat in particular (Milford et al., 2019; York and Gossard, 2004). As income increases, consumers tend to shift their dietary preferences toward more resource-demanding foods (Tilman and Clark, 2014). This transition is taking place at different stages and paces worldwide (GerbensLeenes et al., 2010). Consumption of animal-based foods is much higher in developed countries than in developing and least developed countries. However, the upward trend is more pronounced in developing countries (Henchion and Zimmermann, 2021), where the rise above the poverty line occurs at a faster pace than it did in developed countries (Sans and Combris, 2015; Drewnowski and Poulain, 2018). Meanwhile, in higher income countries a “second nutrition transition” seems to occur (Pais et al., 2021; Vranken et al., 2014). In these countries, the consumption of animal-based proteins, especially from meat, seems to stagnate or decline when reaching a high level of income. Vranken et al. (2014) and Cole and McCoskey (2017) have therefore found evidence of an inverted U-shape relationship between meat consumption and income, indicating that the consumption of unsustainable proteins could reach a maximum and then decline. Therefore, these studies suggested that meat protein consumption follows an EKC. Arguably, the reasons for an AFKC differ from those of the EKC. The latter is justified by the increasing environmental impact of the shift from an agricultural to an industrial economy, followed by a decreasing impact due to resources-saving technological progress and increasing environmental awareness. In the case of the AFKC, the same reasons do not apply, and the determinants have to be ascribed to the factors mentioned above. The reasons behind this decline can be attributed to several factors: i) increasing awareness of the health risks associated with high meat consumption, ii) concerns about the environmental impact of meat production, including greenhouse gas emissions, deforestation, and water usage, iii) growing awareness of animal welfare, iv) the rising availability and popularity of plant-based meat alternatives, v) the spread of popular dietary trends, such as vegetarianism, veganism, and flexitarianism. We therefore present a model of animal-based food consumption that incorporates the above reasons for an inverted U-shaped consumption-income pattern for animal-based food (AF) consumption. The theoretical model sheds light on past trends in AF consumption and the reasons that render possible an AFKC. Nevertheless, while the model may justify the existence of an AFKC, it does not predict it unequivocally. 22 Bio-based and Applied Economics 14(1): 19-35, 2025 | e-ISSN 2280-6172 | DOI: 10.36253/bae-16172 Vito Frontuto et al. The model (for a formal presentation see Appendix 1) assumes that utility from AF consumption has two components. The first one directly stems from its consumption per se, due to its taste and appetite value. Utility is therefore a positive function of animal-based food consumption, so that its marginal utility is positive, but decreasing, due to increasing satiation: additional AF consumption provides less and less additional utility. The second component is the nutritional and health one. According to the nutritional literature, consumption of animal-based proteins has initially a positive effect on nutrition and health (receding from famine, mortality declines, see e.g., Mathijs, 2015) but, at higher levels, it brings several adverse health effects (e.g., cardiovascular risks, obesity-related issues). Hence, if consumers are aware of and care about the negative impacts of high animal-based food consumption on health, this component of utility has an inverted U-shape. In addition, as mentioned above, concern for animal welfare and for the environment can be reasons for a lower utility associated with large animal-based food consumption (Frank, 2008). In this model, for simplicity we include these effects in the health one. The model assumes that a consumer maximizes his/ her utility subject to a budget constraint. The equilibrium condition states that the marginal utility from AF consumption per se, plus the marginal utility stemming from the variation in nutrition-health due to the effect on consumers’ health of an additional AF consumption, equals the additional utility that could be drawn from other goods that could be purchased with the animal-based food price, i.e., the marginal opportunity cost of AF. The marginal utility of AF consumption per se decreases when AF increases, and reaches a lower bound at zero for satiation, when further consumption provides no additional utility. The marginal utility from nutritional-health benefits also decreases with AF consumption and remains positive as long as the marginal health benefit is positive, then it becomes negative. When the marginal health benefits, at high consumption levels, become negative, they may determine a decrease in overall utility if disutility from health damages prevails over utility due to taste. In this case, an inverted U-shape of the income-consumption relationship results. The model implies that a decrease in the AF price relative to all other prices (i.e., a decrease in real AF price) leads to higher AF consumption. This explains what actually happened in the past (FAO, 2009) when the relative price of AF declined with reference to other food prices. The crucial question for the existence of an AFKC is nevertheless the shape of the relationship between income and AF consumption. Among necessities, animal-based food is more expensive than plant-based food. At low-income levels, a higher income allows a shift from cheap staple food to animal-based food, as empirically observed in all countries in the initial stages of development and as a general trend in the recent decades (Sans and Combris, 2015; Delgado et al., 2009; among others). However, the model cannot unambiguously predict a priori whether a further income growth leads to an increase or decrease of AF consumption, because the resulting equilibrium will depend on how the marginal utilities of AF of other consumptions and of nutritionhealth react to income, and on their interrelationships. The model allows for the existence of an AFKC, but does not imply its necessity. The form of the incomeAF consumption relationship has therefore to be determined empirically. Plant-based protein consumption also increases with income at the initial stages of development. However, its increase is presumably slower than the one of animal-based proteins, since income growth allows consumption of the more expensive animal-based proteins, so that in the diet the share of animal-based proteins grows. If consumption of animal-based proteins declines at high income levels, it is possible that plant-based protein consumption will increase as a substitute. The relationship between plant-based protein consumption and income must also be determined empirically. 4. MATERIALS AND METHODS 4.1. Variables and data We employ a balanced panel dataset covering 79 countries from 1991 to 2018 (Table A.1 in Appendix 2). We draw on data from the Food and Agriculture Organization (FAO) New Balance Sheets (NBSs; FAO, 2021), where food supply quantities are used as proxies for consumption (Cole and McCoskey, 2017; You and Henneberg, 2016). These quantities are measured in grams per capita per day and reflect food reaching consumers, with the caveat that actual consumption may be lower due to waste and spoilage during preparation. The study classifies protein consumption into three types: “meat protein” from poultry, pork, goat, mutton and bovine; “animal-based protein” encompassing all animal products including dairy and eggs; and “plant-based protein” derived from cereals, vegetables, fruits, beans, nuts, seeds, roots and spices. We explore potential determinants of protein consumption across three principal dimensions: economic, socio-cultural, and land use. In the economic dimen- 23 Is there an Animal Food Kuznets Curve, and does it matter? Bio-based and Applied Economics 14(1): 19-35, 2025 | e-ISSN 2280-6172 | DOI: 10.36253/bae-16172 sion, the primary focus is income expressed by GDP per capita (p.c.) at chained Purchasing power parities (PPPs) measured in million constant 2017 US$. The data are collected from the Penn World Table (Feenstra et al., 2015), a set of national-accounts data to measure real GDP across countries and over time. In the presence of an inverted U-shape, i.e. Kuznets curve, we expect positive estimated coefficients for the linear terms and negative coefficient for the quadratic terms. In addition to income p.c., we recognize the substantial influence of food prices on protein consumption patterns. To capture this influence, we build national price indexes using data from FAOSTAT (FAO, 2022b). Specifically, we select the price of the most consumed item within each of the three protein sources (meat, animal-based and plant-based) for every country and year under study and build an index using the first year of the time series (1991) as base year. This approach aims to quantify how variations in food prices across different protein sources impact dietary choices and consumption behaviors globally. Indeed, our hypothesis is to observe a negative coefficient for the price index meaning that an increase in prices determines a reduction in protein consumption. In addition to own price for each protein source, we have also tested relative prices. In fact, as suggested by FAO (2009) over the last 50 years there has been a decline in the prices of livestock products relative to those of other products, making consumption of animalbased and plant-based foods more affordable than meat even without rising income. A third economic variable used in our empirical application is the trade openness, built as the ratio of imports and exports over national GDP. Our hypothesis is to observe a positive effect of trade on the three proteins consumption due to the likely larger availability of different products and thus protein sources. Beyond economic factors, social and cultural influences could also shape protein consumption patterns. We integrate several key variables to explore these dimensions. First, the religious beliefs were incorporated by using the percentage of population adhering to Islam as a proxy to understand dietary restrictions that may affect consumption preferences, for example by reducing meat consumption and increasing plant-based protein intakes. Second, we integrate the percentage of women participating in the labour force as an indicator of evolving food preparation practices. Third, the percentage of adults with tertiary education levels is used to capture the influence of educational attainment on dietary preferences and awareness of nutritional choices, potentially affecting protein intake patterns. We hypothesize that more educated people tend to prefer diets with more plant-based food for both health and environmental concerns. However, we are aware that education is strongly correlated with income levels. Finally, to further explore other contextual conditions likely influencing protein consumption, the study includes two proxies of land use: the harvested area per capita as a measure for the relevance of the agricultural sector for self-provision of proteins and the percentage of the population living in urban areas. These variables are used to examine the impact of urbanization on dietary habits and access to diverse food options, including protein sources. Table 1 provides a comprehensive list of variables used in the study, and their descriptive statistics and sources. Unlike the typical practice in Environmental Kuznets Curve (EKC) literature, the study uses variables in their original levels instead of logarithmic transformations, aligning with findings by Hasanov et al. (2021).1 4.2. Econometric strategy Since the mid-1950s, scholars testing the Kuznets Curve (KC) hypothesis on various environmental and non-environmental indicators have primarily used crosssectional data and longitudinal data with fixed effects estimators (e.g., Vranken et al. (2014) for consumption of meat protein). However, traditional panel data estimators assume cross-sectional independence, basing the models on homogeneous coefficients and yielding inconsistent estimated parameters (Heck and Thomas, 2020). Indeed, cross-sectional units may exhibit shared characteristics, such as spatial effects, omitted common factors, or socioeconomic networks interaction leading to cross-sectional dependence, calling for estimators that account for intercepts and slopes heterogeneity. The literature on heterogeneous panels has evolved along two main strands: i) the application of mean group (MG) estimators (Pesaran and Smith, 1995) and subsequent modifications (Augmented MG and Common Correlated Effects MG; Teal and Eberhardt, 2010), ii) the application of multilevel or mixed effect models to panel data (McCulloch et al., 2001). The key distinction between panel models (such as MG estimators) and mixed effects models lies in the treatment of the independent variables. In mixed effects 1 Hasanov (2021) argues that in non-linear logarithmic Environmental Kuznets Curves (EKC), the signs of estimated coefficients and the statistical significance of lower-order polynomial terms can vary arbitrarily based on the units of measurement chosen for the independent variables. Consequently, Hasanov suggests that researchers should first study the EKC in levels considering the potential issues with the logarithmic specification. 24 Bio-based and Applied Economics 14(1): 19-35, 2025 | e-ISSN 2280-6172 | DOI: 10.36253/bae-16172 Vito Frontuto et al. models, independent variables are treated as non-random variables, whereas in panel data models, they are always assumed to be random. Another significant difference is in estimating the average effects (invariant between individuals) and individual (or random) effects. In the case of MG estimators, individual-specific ordinary least-squares (OLS) regressions are estimated then the individual-specific parameters are averaged across the panel to determine an overall effect. In the case of mixed effect models, the estimated parameters are the common effect with the random effects representing individual deviations from this average, inferred from estimated variances and covariances (Dinda, 2004). A meta-analysis conducted by Saqib and Benhmad (2021) on more than five hundred studies concluded that the econometric strategy does not significantly impact the test of the EKC hypothesis. However, they highlighted the greater reliability of longitudinal data and the robustness of methods that deal with heterogeneous panels such as MG estimators and mixed effect models. In this paper, we employ a mixed effect model because of our focus on the variation in regression coefficients rather than a global behaviour as an average of country-specific dynamics. Country-specific estimates, limited by the income ranges, cannot properly identify the curvature of a general function. To account for intercept and slopes heterogeneity in parameters the unknown parameters are decomposed in a fixed term γ (constant across countries) and a random term δ (specific for each country). Thus, the relationship between protein consumption per capita (animal-, plant-based and meat) and GDP per capita is modelled as: (Proteins/P)it = (γs0 + δsi0) + (γs1 + δsi1) (GDP/P)sit + (γs2 + δsi2) (GDP/P)2sit + ∑Jj=3 βsijXsitj + εit (1) where s= a,m,p identifies the protein source (animal-, plant-based and meat), i=1,…,N indicates the countries, t=1,…,T the time periods, GDP is defined as above and P is population, Xj the j-covariates. Note that represents the potential non-linear effect of GDP per capita on proteins consumption and it is used in the Kuznets framework to check the inverted U-shaped curvature of the relation. This model has been estimated using maximum likelihood estimators for the three sources of protein (Rabe-Hesketh and Skrondal, 2008) and likelihood-ratio tests have been employed to compare different models and to validate the use of random coefficients. Moreover, the models are first estimated with an unstructured random-effects covariance matrix, which allows for distinct variances and covariances between all randomTable 1. List of variables with descriptive statistics. Variable Description Mean Std. Min Max Source Dependent variables MeatProt Per capita Meat-based Protein consumption (g/day) 16.4 11.1 1.2 46.9 FAO (2021) AnimalProt Per capita Animal-based Protein consumption (g/day) 35.2 21.4 3.2 79.7 FAO (2021) PlantProt Per capita Plant-based Protein consumption (g/day) 44.0 10.1 22.9 82.7 FAO (2021) Independent variables GDPPc Per capita expenditure-side real GDP at chained PPPs (000 US$) 16.6 16.3 0.4 90.3 Penn world table (Feenstra et al., 2015) GDPPc2Squared GDPPc 541.7 880.9 0.2 8154.2 Penn world table (Feenstra et al., 2015) Price Index Animal-based products 0.99 0.16 -0.88 2.65 FAO (2022b) Meat products 0.98 0.17 -0.45 2.50 Plant-based products 1.11 0.52 -2.73 4.67 Trade (Imports+exports) / GDP (%) 68.7 34 13.8 227.4 World Bank (2022a) Education Share of post-secondary education (%) 10.2 8.9 0.15 48.3 World Bank (2022b) PerMus Share of Muslims over population (%) 23.4 35.7 0 99.8 ARDA (2022) PerFemWork Share of female employment (%) 40.2 9.3 10.7 56 World Bank (2022c) Urbanization Share of people living in urban areas (%) 58.6 22.2 5.5 95.3 World Bank (2022d) HarvArea Harvested area/population (per capita ha) 0.172 0.2 0 1.4 FAO (2022a) N. obs. 2212 N. groups 79 Sources: FAO, Penn World Table, World Bank, ARDA and own calculation. 25 Is there an Animal Food Kuznets Curve, and does it matter? Bio-based and Applied Economics 14(1): 19-35, 2025 | e-ISSN 2280-6172 | DOI: 10.36253/bae-16172 effects covariates. However, inconsistent estimations for the plant-based protein model necessitated an identity covariance structure, assuming equal variances. According to the literature on testing the nature of the time-series to select the appropriate panel estimator (Perman and Stern, 2003; Eberhardt, 2012), the model of equation [1] was tested relative to: i) cross-sectional dependence; ii) presence of unit roots (i.e., stationarity); iii) long-run relationship (i.e., cointegration). To select the appropriate test for investigating unit roots, we initially checked the cross-sectional dependence of the series using the Pesaran test (Pesaran, 2021) under the null hypothesis of cross-sectional independence. Most variables exhibited cross-sectional dependence (except for trade) (see Table A.2 in Appendix 3). Subsequently, we tested the stationarity of the series by implementing the modified pCADF test (Costantini and Lupi, 2013) which consider cross-sectional dependence under the null hypothesis of non-stationarity. The results suggested that the null hypothesis of non-stationarity can be rejected only after transforming the series in their first differences except for the urbanization rate and the education (Table A.3 in Appendix 3). Then, we checked the cointegration assumption to prevent the regression from providing biased statistical evidence of the relationship among variables. Cointegration was investigated through various tests, including the Phillips-Perron, the Modified Phillips-Perron, the Augmented Dickey-Fuller tests (Pedroni, 1999; Pedroni, 2004) and the so-called Westerlund test (Westerlund, 2005) by assuming the presence of cross-sectional dependences (Table A.4 in Appendix 3). The rejection of the null hypothesis of all these tests indicates that our models are cointegrated. The findings support the selection of the mixed effect model as appropriate to estimate heterogenous coefficients for intercepts and slopes. 5. RESULTS Table 2 presents the estimated coefficients of the model of meat animal-based and plant-based protein consumption which exhibit overall significance. Likelihood-ratio tests have been applied to compare different models and different covariates. Education and urbanization rate turned out to be non-stationary even when transformed in their first differences and were therefore not used to avoid spurious estimated coefficients (see Table A.3 in the Appendix 3). Prices for the three protein sources are not statistically significant and hence are not included in our preferred specification in Table 2 (see Table A.5 for the estimated coefficients of model including prices).2 Table 2 also reports the estimated standard deviations for the intercept, the GDP per capita and the GDP2 per capita coefficients. All of these standard deviations are statistically significant, indicating the intercept and slopes heterogeneity and thus supporting the use of the mixed effects model. The most important determinant of meat protein consumption is per capita income, with both its estimates of the linear and the quadratic term highly significant. The estimates indicate that a thousand dollar increase in per capita income induces a 0.725 g/ day increase in meat protein consumption. Notably, the negative sign of the squared term suggests that meat protein consumption does increase with income, but at a decreasing pace. Among the variables aside from income, the Trade and the percentage of Muslims are significant. Specifically, every additional percentage point in the ratio of imports plus exports over GDP implies a 0.008 g/day increase in the average meat protein consumption. A percentage point increase in the share of Muslims over the population translates into a 0.11 g/day decrease in the average meat protein consumption, ceteris paribus. These results are consistent with Andreoli et al. (2021) and Milford et al. (2019). Female participation, however, does not show statistical significance, as in Milford et al. (2014). The positive sign of the GDP parameter and the negative sign of the GDP2 parameter, both significant, suggest the existence of an inverted U-shaped relationship, thus supporting the existence of an AFKC where meat protein consumption increases with per capita income up to a maximum before decreasing. A crucial point for assessing the policy implications of the AFKC is nevertheless determining the level of the turning point. This can be calculated as where is the estimated parameter of per capita income and the estimated parameter of its square.3 This simple calculation results in a turning point of 42,923 US$,4 located between the 90th and the 95th percentiles of the per capita income distribution in the whole sample, and above the 80th percentile of the income distribution in the last year of the panel (2018). It could be argued that the turning point should also be estimated consid2 We used version 18 of STATA for Windows to carry out the analysis of the data in this paper. The mixed command has been used to estimate the mixed models presented in Table 2. 3 The formula for the maximum income in the estimated second-degree equation is obtained by setting the derivative of the equation to zero and solving for the income variable. 4 To present a more concise table of results, the coefficients have been rounded to three decimal places. Consequently, the turning point value derived from rounded coefficients differs from the one presented in the text, which uses estimated coefficients to six decimal places. 26 Bio-based and Applied Economics 14(1): 19-35, 2025 | e-ISSN 2280-6172 | DOI: 10.36253/bae-16172 Vito Frontuto et al. ering the variation of the estimated parameters. Unfortunately, the turning point results from the ratio of two normal random variables, which results in a Cauchy distribution, whose expected value and variance are undefined. However, its mode and median are defined, and the distribution is symmetrical. We therefore perform a Monte Carlo simulation of the median turning point. We randomly draw couples of and parameters from a bivariate normal distribution, calculate the turning point, repeated for 1000 draws, and individuate the median turning point of these simulations. By repeating the procedure 10,000 times we obtain an empirical distribution of the medians, from which we calculate their mean and standard deviation. The result of 42,891 US$ is sensibly similar to the simple calculation from the estimated parameters. The standard deviation is relatively modest, 318 US$, and the range went from a minimum of 41,867 to a maximum of 44,159 US$. The minimum value is around the 90th percentile. Comparisons with previous studies reveal similar turning point estimates, i.e. 46,000-66,000 constant 2017 International US$ p.c. (Andreoli et al., 2021); 36,37549,848 constant 2005 US$ p.c. (Cole et al., 2013); 35,00053,000 constant 2005 international US$ p.c. (Vranken et al., 2014), indicating consistency across analyses. However, employing mixed effects models alongside Monte Carlo simulation produces more efficient estimates with reduced variability. Furthermore, the results allow us to predict the meat protein consumption corresponding to the turning point, by using the estimated parameters and the per capita income of the turning point and setting the other variables at their mean. To appreciate the variation of the prediction, we also calculate the predicted consumption when the other variables are taken at the minimum and maximum of their observed values,5 and when they are taken at their mean plus/minus their standard deviation. Table 3 presents the results. The calculated meat protein consumption at the income turning point and the mean of the other variables is 24.61 g/day, slightly below the 75th percentile. The maximum value (29.57 g/day) is between the 85th 5 When calculating the maximum and minimum consumption, variables with a negative parameter were taken as positive, so to identify the maximum possible range. Table 2. Results of the models of protein consumption. Indep. variables Dependent variables Meat Protein Animal-based Protein Plant-based Protein Coef. Std.Err. Coef. Std.Err. Coef. Std.Err. GDPPc 0.725*** 0.135 1.255*** 0.206 1.506*** 0.419 GDPPc2-0.008*** 0.003 -0.015*** 0.004 -0.003 0.002 Trade 0.008*** 0.003 0.019*** 0.005 0.032*** 0.005 HarvArea 1.097 1.317 1.930 1.863 -15.479*** 1.739 PerMus -0.105*** 0.0281 -0.177*** 0.484 0.212*** 0.421 PerFemWork -0.008 0.033 0.058 0.048 0.113** 0.046 Constant 11.092*** 2.191 25.517*** 3.689 30.675*** 2.613 sd(GDPPc) 1.086 0.124 1.686 0.161 3.644 0.361 sd(GDPPc2) 0.021 0.006 0.030 0.004 0.005 0.002 sd(Constant) 13.364 1.414 25.170 2.378 12.892 1.120 sd(Residual) 1.541 0.028 2.167 0.035 2.093 0.034 N. obs. 2212 2212 2212 N. groups 79 79 79 Wald Chisq(6) 78.33*** 108.10*** 194.50*** Log likelihood -4533.46 -5302.99 -5300.83 *, **, *** indicate that statistics are significant at the 10%, 5% and 1% level of significance respectively. Table 3. Predicted protein consumption at the turning points. Meat Protein Animal-based Protein Mean 24.61 51.62 Min 16.16 35.02 Max 29.57 61.96 Mean-SD 20.45 43.79 Mean+SD 28.76 59.45 33 Is there an Animal Food Kuznets Curve, and does it matter? Bio-based and Applied Economics 14(1): 19-35, 2025 | e-ISSN 2280-6172 | DOI: 10.36253/bae-16172 APPENDIX 1 In formal terms, the model of AF consumption assumes the consumer maximizes his/her utility subject to a budget constraint: Max U[a, H(a), C] [A1] s.t.: C + pa = I where a is animal-based protein consumption, H indicates health-nutrition components of utility as a function of animal food consumption, C is expenditure for all other consumption goods, I is income, pa is the price of a and the price of C is taken as numeraire. The usual general assumptions hold: U’a > 0, U’’a < 0; U’C > 0, U’’C < 0; U’H > 0, U’’H < 0. To represent the U-shape of nutritional-health benefits, it is assumed that H’a ≥ 0 for a ≤ a , H’a < 0 for a > a where a is the animal-based protein consumption yielding the maximum nutrition-health benefit; H’’a is assumed < 0. The first order conditions (FOCs) is: U’a + U’HH’a = paU’C [A2] Equation [A2] simply states that, at equilibrium, the marginal utility from consumption of AF (the first leftside term), plus the marginal utility from the nutritional-health benefits from its consumption (the second leftside term) is equal to the additional utility that could be drawn from other goods that could be purchased with the animal food price, i.e., the marginal opportunity cost of AF (the right-side term). The effect of income on AF consumption can be computed as the derivative of a with respect to I in eqn. [A2]. The result is nevertheless a complex function of the second direct and cross derivatives of a, H, C, and its sign cannot be unambiguously determined, it can be positive or negative. APPENDIX 2 Table A.1. List of countries analysed. Country ISO CODE Algeria DZA Argentina ARG Australia AUS Austria AUT Bangladesh BGD Bolivia BOL Brazil BRA Country ISO CODE Belize BLZ Cameroon CMR Canada CAN Cabo Verde CPV Sri Lanka LKA Chile CHL China CHN Colombia COL Congo COG Cyprus CYP Denmark DNK Dominican Republic DOM Ecuador ECU Egypt EGY El Salvador SLV Finland FIN France FRA Gambia GMB Germany DEU Ghana GHA Greece GRC Guinea GIN Honduras HND Hungary HUN India IND Indonesia IDN Iran (Islamic Republic of) IRN Ireland IRL Israel ISR Italy ITA Côte d’Ivoire CIV Japan JPN Jordan JOR Kenya KEN Cambodia KHM Republic of Korea KOR Lao People’s Democratic Republic LAO Lebanon LBN Madagascar MDG Malaysia MYS Mali MLI Mauritius MUS Mexico MEX Morocco MAR Mozambique MOZ Namibia NAM Nepal NPL Netherlands NLD New Zealand NZL Nicaragua NIC Niger NER Nigeria NGA 34 Bio-based and Applied Economics 14(1): 19-35, 2025 | e-ISSN 2280-6172 | DOI: 10.36253/bae-16172 Vito Frontuto et al. Country ISO CODE Norway NOR Pakistan PAK Panama PAN Paraguay PRY Peru PER Philippines PHL Poland POL Portugal PRT Rwanda RWA Saudi Arabia SAU South Africa ZAF Spain ESP Sweden SWE Switzerland CHE Togo TGO Turkey TUR United Kingdom GBR United States of America USA Burkina Faso BFA Uruguay URY APPENDIX 3 Table A.2. Test of cross-sectional dependence of variables. Variable#Pesaran test AnimalProt 2.678*** MeatProt 2.398** PlantProt 2.798*** GDPPc 9.247*** GDPPc216.342*** Trade 0.158 HarvArea 5.03*** PerFemWork 3.804*** Urbanization 1.948* Education 85.837*** Animal-based Price Index 12.883*** Meat Price Index 6.536*** Plant-based Price Index 24.181*** #The percentage of Muslim (PerMus) has not been tested because time invariant. *, **, *** stand for the significance level of 10%, 5% and 1% respectively. The null hypothesis is the absence of cross-sectional dependence. Table A.3. Unit root test on variables and their first difference. Variable pCADF test Variable pCADF test AnimalProt 2.965 ∆AnimalProt -8.341*** MeatProt 1.408 ∆MeatProt -10.136*** PlantProt 3.616 ∆PlantProt -19.977*** GDPPc 5.426 ∆GDPPc -3.426*** GDPPc26.295 ∆GDPPc2-1.592* Trade 4.052 ∆Trade -7.158** HarvArea 2.232 ∆HarvArea -7.700*** PerFemWork 1.631 ∆PerFemWork -2.565*** Urbanization 4.302 ∆Urbanization 6.272 Education 11.310 ∆Education 3.016 Animal-based Price Index -12.225*** ∆Animal-based Price Index -20.497*** Meat Price Index -12.825*** ∆Meat Price Index -5.473*** Plant-based Price Index -16.698*** ∆Plant-based Price Index -9.583*** *, **, *** stand for the significance level of 10%, 5% and 1% respectively. The null hypothesis is non-stationarity. Table A.4. Cointegration test assuming cross-sectional dependence. Test name AnimalProt MeatProt PlantProt Pedroni Modified Phillips–Perron t 4.590*** 3.697*** 1.88** Phillips–Perron t -7.509*** -8.279*** -9.710*** Augmented Dickey–Fuller t -8.99*** -10.195*** -10.324*** Westerlund Variance ratio -2.579 *** -2.751*** -1.6434* *, **, *** indicate that statistics are significant at the 10%, 5% and 1% level of significance respectively. The null hypothesis is no-cointegration. 35 Is there an Animal Food Kuznets Curve, and does it matter? Bio-based and Applied Economics 14(1): 19-35, 2025 | e-ISSN 2280-6172 | DOI: 10.36253/bae-16172 Table A.5. Results of the full models of protein consumption. Indep. variables Dependent variables Animal-based Protein Meat Protein Plant-based Protein Coef. Std.Err. Coef. Std.Err. Coef. Std.Err. GDPPc 1.256*** 0.206 0.725*** 0.135 1.505*** 0.419 GDPPc2-0.015*** 0.004 -0.008** 0.003 -0.002 0.001 Trade 0.018*** 0.005 0.008** 0.003 0.032*** 0.004 HarvArea 1.190 1.862 1.087 1.317 -15.486*** 1.738 PerMus -0.177*** 0.048 -0.105*** 0.028 0.211*** 0.421 PerFemWork 0.057 0.047 -0.008 0.033 0.111** 0.046 Price Index (animal-based) 0.304 0.289 Price Index (meat) 0.065 0.200 Price Index (plant-based) -0.184 0.148 Constant 25.245*** 3.697 11.033*** 2.199 30.910*** 2.619*** sd(GDPPc) 1.686 0.161 1.087 0.125 3.644 0.361 sd(GDPPc2) 0.029 0.003 0.021 0.006 0.005 0.002 sd(Constant) 25.17 2.378 13.369 1.415 12.894 1.119 sd(Residual) 2.167 0.035 1.541 0.028 2.092 0.033 N. obs. 2212 2212 2212 N. groups 79 79 79 Wald Chisq(6) 109.25*** 78.31*** 196.14*** Log likelihood -5302.43 -4533.41 -5300.07 *, **, *** indicate that statistics are significant at the 10%, 5% and 1% level of significance respectively.