The income elasticity of meat: a meta-analysis

Authors


Abstract

The demand for meat has been estimated by many studies utilizing various data and estimation methods. In this study, we perform a meta-analysis of the income elasticity of meat that involves regressing 3357 estimated income elasticities, collected from 393 studies, on variables that control for study characteristics. Across several meta-regression specifications, we find significant differences in income elasticities tied to the type of meat being studied, as well as a few functional forms, data aggregations, publication characteristics, and locations of demand. However, many study characteristics do not significantly influence reported income elasticities. Less concern should be given to such characteristics when choosing an income elasticity from the literature.

1. Introduction

Numerous studies estimate the demand for meat using various data and estimation methods, which several qualitative literature reviews (e.g., Kuznets 1953; Reeves and Hayman 1975; Richardson 1976; Tomek 1977; Raunikar and Huang 1987; Smallwood et al. 1989; Alston and Chalfant 1991; Moschini and Moro 1996; Griffith et al. 2001; Asche et al. 2007) suggest contribute to differences in reported outcomes. However, because qualitative literature reviews can be sensitive to the subjective decision of the reviewer to emphasize certain study attributes over others, meta-analysis is an increasingly popular method used to quantitatively survey literature. A typical meta-analysis involves regressing a parameter commonly estimated in the literature on variables that control for study characteristics. By doing so, the subjective decision of the reviewer is replaced by statistical tests, the results of which shed light on the relative statistical importance of study characteristics to influence the parameter estimate.

With respect to the demand for meat, Gallet (2010) reports the results of a meta-analysis of the price elasticity of meat. Regressing 4120 observations of the price elasticity of meat, collected from 419 studies, on a series of study characteristic variables, he finds the price elasticity is particularly sensitive to the type of meat being studied and the estimation methodology.

Yet the income elasticity also plays an important role in the literature. For example, a meat producer might use the income elasticity to gauge the growth of demand as incomes rise. Also, in light of recent attention given to the prospect of using food price policies to improve health outcomes (e.g., Kuchler et al. 2005; Chouinard et al. 2007), arguments in favor or against such policies can be bolstered by knowledge of the income elasticity. For example, consider a policymaker contemplating a tax (subsidy) on red meat (white meat) in an effort to shift consumption away from red meat towards white meat. If the income elasticity of white meat exceeds that of red meat, then in the presence of rising incomes, there is less need to use the tax and subsidy to coerce a shift in budget shares towards white meat, because consumers will do this on their own, ceteris paribus.

Accordingly, this paper complements Gallet (2010) by reporting the results of a meta-analysis of the income elasticity of meat. Specific questions addressed in this meta-analysis are the following: (i) Does the income elasticity differ across meat products? (ii) Is the income elasticity sensitive to the specification of demand? (iii) Does the type of data used to estimate meat demand influence the income elasticity? (iv) Is the income elasticity sensitive to the method used to estimate demand? (v) Do the quality of the publication outlet and the year of publication influence the income elasticity?, and (vi) Are there regional differences in the income elasticity? By answering such questions, we gain better insight into the tendencies in the literature to sway the income elasticity one way or the other.

The paper proceeds as follows. In Section 2, we discuss the data and meta-regression model, which is followed in Section 3 with a discussion of the estimation results. The paper concludes with a summary in Section 4.

2. Data and meta-regression model

2.1 Data

An initial search of the literature was conducted using EconLit, AgEcon Search, and Google Scholar, as well as several qualitative literature reviews (i.e., Kuznets 1953; Reeves and Hayman 1975; Richardson 1976; Tomek 1977; Raunikar and Huang 1987; Smallwood et al. 1989; Alston and Chalfant 1991; Moschini and Moro 1996; Griffith et al. 2001; Asche et al. 2007), to identify candidate studies that estimate the income elasticity of meat. Subsequent to surveying the reference sections of all studies identified, 393 studies (see Table 1) reporting 3357 income elasticity estimates were included in the meta-data set.1 These 3357 income elasticity estimates become observations of the dependent variable in a meta-regression model.

Table 1.   Studies included in meta-analysis
  1. Note: Complete references of the 393 studies to be posted online.

Abdulai et al. (1999)Fousekis and Pantzios (2000)Menkhaus et al. (1985)
Abdulai and Aubert (2004)Fousekis and Revell (2000, 2003, 2004, 2005)Mergos and Donatos (1989)
Abdullah (1994)Fox (1951)Miljkovic et al. (2002)
Abdullah et al. (1999)Fraser and Moosa (2002)Millan and Aldaz (2005)
Ackah and Appleton (2003)Freebairn and Rausser (1975)Miran and Akgungor (2005)
Agbola (2003)Freebairn and Gruen (1977)Mittelhammer et al. (1996)
Agbola et al. (2003)French (1952)Molina (1994)
Ahmed and Shams (1994)Fulponi (1989)Moro and Sckokai (2000)
Akbay et al. (2007)Funk et al. (1977)Morrison et al. (2003)
Alfonzo and Peterson (2006)Gao and Shonkwiler (1993)Moschini (1998, 2001)
Ali (2002)Gao et al. (1996, 1996)Moschini and Meilke (1984, 1989)
Allais and Nichele (2004, 2007)Garcia (2004)Moschini and Vissa (1993)
Alston and Chalfant (1987, 1991)Garcia et al. (2005)Moschini et al. (1994)
Alston et al. (1995, 2002)Gibson (1998)Murray (1984)
Andrikopoulos et al. (1987)Goddard and Cozzarin (1992)Mutondo and Henneberry (2007, 2007)
Angulo and Gil (2006)Golan et al. (2001)Nayga (1995)
Apaza et al. (2002)Goodwin (1992)Nayga and Capps (1994)
Armagan and Akbay (2007)Goodwin and Phaneuf (2001)Nerlove and Addison (1958)
Arzac and Wilkinson (1979)Goodwin and Sheffrin (1982)Nyankori and Miller (1982)
Asche (1996, 1997)Gould (2002)Ogunyinka and Marsh (2002, 2006)
Asche et al. (1997)Gould and Villarreal (2006)Omezzine et al. (2003)
Asche et al. (1998)Gould et al. (2002)O’Neill and Buttimer (1973)
Atkins et al. (1989)Gracia et al. (1998)Pantzios and Fousekis (1999)
Azzam et al. (2004)Greenfield (1974)Park et al. (1996)
Babula (1997)Hahn (1988, 1994)Peeters et al. (1997)
Babula and Corey (2004)Hahn et al. (2003)Peng et al. (2004)
Bacchi and Spolador (2002)Halbrendt et al. (1994)Peterson and Chen (2005)
Balcombe (2004)Hancock et al. (1984)Piggott et al. (1996)
Balcombe and Davis (1996)Hannah (1970)Piggott and Marsh (2004)
Ball and Dewbre (1989)Hanrahan (2002)Piggott et al. (2007)
Barten (1964)Hassan et al. (2001)Pitt (1983)
Beatty and LaFrance (2001)Hassan and Johnson (1979)Pope et al. (1980)
Benson et al. (2002)Hassan and Katz (1975)Price and Gislason (2001)
Bergstrom (1955)Hayes et al. (1990)Pudney (1981)
Bewley and Young (1987)Hayes et al. (1991)Purcell and Raunikar (1971)
Bhati (1987)Heien (1982)Quagrainie (2003)
Bielik and Kunova (2007)Heien and Pompelli (1988)Raper et al. (2002)
Bjorndahl et al. (1992)Heien and Wessells (1990)Reed et al. (2003)
Bjorndahl et al. (1994)Henneberry and Mutondo (2007)Reed et al. (2005)
Blackorby et al. (1978)Herrmann and Lin (1988)Regorsek and Erjavec (2007)
Blanciforti and Green (1983)Herrmann et al. (1992, 1993)Reynolds and Goddard (1991)
Blanciforti et al. (1986)Herrmann et al. (2002)Rickertsen (1996, 1997, 1998)
Boetel and Liu (2003)Hossain and Jensen (2000)Rickertsen and Vale (1996)
Boutwell and Simmons (1968)Houston and Ermita (1992)Rickertsen and Cramon-Taubadel (2000)
Boyle (1996)Hsu (2000)Rickertsen et al. (2003)
Brester (1996)Huang (1979)Roy et al. (1994)
Brester and Schroeder (1995)Huang and Raunikar (1978, 1986)Salfyurtlu et al. (1986)
Brester and Wohlgenant (1993)Huang and Rozelle (1998)Sahn (1988)
Bureau of Econ Analysis (1967)Huang and Bouis (2001)Saleh and Sisler (1977)
Burney and Akmal (1991)Huang (1994)Salvanes and DeVoretz (1997)
Burton (1992)Huang and Haidacher (1983, 1989)Sam and Zheng (2007)
Burton and Young (1992, 1996, 1997)Hudson and Vertin (1985)Sarmiento (2005)
Byrne et al. (1993)Hutasuhut et al. (2002)Sasaki (1993)
Byrne et al. (1995)Hyde and Perloff (1998)Sasaki and Fukagawa (1987)
Byron (1970, 1970)Jabarin (2005)Savadogo and Brandt (1988)
Cai et al. (1998)Jan et al. (2002)Schroeder et al. (2000)
Capps (1989)Jensen and Manrique (1998)Schroeder et al. (2001)
Capps and Havlicek (1984, 1987)Jiang and Davis (2007)Schroeter and Foster (2004)
Capps and Pearson (1986)Johnson et al. (1998)Schroeter (1988)
Capps and Schmitz (1991)Johnson (1978)Schultz (1935)
Capps et al. (1994)Jones and Yen (2000)Shahid and Gempesaw (2002)
Cashin (1991)Jung and Koo (2000, 2002)Shonkwiler and Taylor (1984)
Chalfant (1987)Kaabia et al. (2001)Soe et al. (1994)
Chalfant et al. (1991)Kaabia and Gil (2001)Soshnin et al. (1999)
Chang (1977, 1980)Karagiannis and Velentzas (1997)Steen and Salvanes (1999)
Chang and Green (1989, 1992)Karagiannis et al. (1996, 2000)Stone (1951)
Chavas (1983)Kastens and Brester (1996)Stroppiana and Riethmuller (2000)
Chen (1996)Katchova and Chern (2004)Su and Yen (1996)
Chen and Veeman (1991)Keller and Driel (1985)Sulgham and Zapata (2006)
Cheney et al. (2001)Kennes (1983)Taljaard et al. (2004, 2006)
Cheng and Capps (1988)Kim and Gould (1998)Talukder (1993)
Chern et al. (2003)Kinnucan and Thomas (1997)Tambi (1996, 1998)
Chesher and Rees (1987)Kinnucan et al. (1997)Teisl et al. (2002)
Choi and Sosin (1990)Kinnucan and Miao (1999)Teklu and Johnson (1988)
Christensen and Manser (1977)Klonaris (2001)Thompson (2004)
Chung (1994)Klonaris and Hallam (2003)Throsby (1974)
Coulibaly and Brorsen (1999)Kokoski (1986)Thurman (1986, 1987, 1989)
Court (1967)Kouka (1995)Tintner (1950, 1952)
Cowan and Herlihy (1982)Kounker (1977)Tomek and Cochrane (1962)
Cramer (1973)Kreinin (1973)Tonsor and Marsh (2007)
Cranfield and Goddard (1995)Kulshreshtha (1979)Traesupap et al. (1999)
Crutchfield (1985, 1985)Kulshreshtha and Wilson (1972)Trierweiler and Hassler (1971)
Davis et al. (2004)Kusumastanto and Jolly (1997)Tryfos and Tryphonopoulos (1973)
Davis et al. (2007)Ladd and Tedford (1959)Tsoa et al. (1982)
DeVoretz (1982)Lambert et al. (2006)Unnevhr and Khoju (1991)
DeVoretz and Salvanes (1993)Lanfranco et al. (2002)Vale (1996)
Dey (2000)Langemeier and Thompson (1967)Van Der Meulen (1961)
Dey and Garcia (2007)Lazaridis (2003)Veeman et al. (2004)
Dhehibi and Laajimi (2004)Le et al. (1998)Verbeke and Ward (2001)
Dhehibi et al. (2005)Lechene (2000)Vere and Griffith (1988)
Doll (1972)Lee et al. (1992)Vickner et al. (2006)
Dong et al. (1998)Lee and Seaver (1971)Wahby (1952)
Dong et al. (2004)Lerdau (1954)Wahl and Hayes (1990)
Dong and Fuller (2004, 2006)Leuthold and Nwagbo (1977)Wahl et al. (1991)
Dono and Thompson (2002)Lin et al. (1989)Wang et al. (1998)
Duffy (1999)Liu and Chern (2004)Wellman (1992)
Duffy and Goddard (1995)Liu and Sun (2005)Wessells and Wilen (1993, 1994)
Durbin (1953)Ma et al. (2004)Wessells et al. (1995)
Eales (1996)Main et al. (1976)Wilkie and Godoy (2001)
Eales and Unnevehr (1988, 1993)Mainland (1998)Wilkie et al. (2005)
Eales et al. (1997)Maki (1957)Wilson and Marsh (2005)
Eales et al. (1998)Manrique and Jensen (2001)Wohlgenant (1985, 1986, 1989)
Eales and Wessells (1999)Manser (1976)Wohlgenant and Hahn (1982)
Edgerton (1996, 1997)Marceau (1967)Working (1952)
Effiong and Njoku (2001)Marsh et al. (2004)Wu et al. (1995)
Fabiosa (2000)Martin (1967)Xi et al. (2003, 2004)
Fabiosa and Ukhova (2000)Martin and Porter (1985)Xu and Veeman (1996)
Fan and Chern (1997)Mazany et al. (1996)Yanagida and Tyson (1984)
Fan et al. (1994)Mazzocchi (2003, 2006)Yandle (1968)
Fan et al. (1995)Mazzocchi et al. (2004)Yang and Koo (1994)
Fanelli and Mazzocchi (2002)Mazzocchi and Lobb (2005)Yeboah and Maynard (2004)
Fayyad et al. (1995)Mazzocchi et al. (2006)Yen and Huang (1996, 2002)
Felixson et al. (1987)Mbala (1992)Yen et al. (2003)
Fidan (2005)McGuirk et al. (1995)Yen et al. (2004)
Fisher (1979)McNulty and Huffman (1992)Zhuang and Abbott (2007)
Flake and Patterson (1999)Mdafri and Brorsen (1993)Zidack et al. (1992, 1993)
Fofana and Clayton (2003)Meinken et al. (1956)Zwick (1957)

Similar to other meta-analyses of the income elasticity (e.g., Espey 1998; Dalhuisen et al. 2003; Gallet and List 2003; Gallet 2007), in addition to the reported income elasticity estimates, several characteristics of the 393 meat demand studies were noted. First, it is common to estimate the income elasticity for a variety of meats, including beef, pork, lamb, poultry, fish, and a composite category consisting of several meats. Second, concerning the specification of demand, in addition to the commonly adopted linear and double-log functional forms, many studies estimate the demand for meat using theoretically consistent functional forms, such as the linear-approximate almost ideal demand system (AIDS-Linear), which relies on a price index to linearize Deaton and Muellbauer’s (1980) AIDS specification, the traditional nonlinear AIDS form (AIDS-Nonlinear), the quadratic AIDS form (AIDS-Quadratic) of Banks et al. (1997), and the generalized AIDS form (AIDS-General) of Bollino (1990). Studies have also estimated the demand for meat using a variety of other functional forms (i.e., semi-log, Rotterdam, CBS, translog, S-Branch, Box–Cox, the generalized addilog, and the quadratic expenditure forms).

Third, continuing with demand specification, several income elasticity estimates come from specifications that include other meats as substitutes. Also, some studies estimate dynamic specifications of demand by including lag terms on the right side of the demand equation, while others estimate a two-step model, in which meat demand is modeled as (i) the choice of whether or not to consume meat followed by (ii) the decision of how much to consume.

Fourth, we also note several characteristics of the data and estimation methods used by the 393 meat demand studies. Specifically, in addition to cross-sectional, time-series, and panel data, studies utilize data that are temporally aggregated to the annual, quarterly, and less than quarterly (i.e., monthly and weekly) levels, as well as spatially aggregated to the multiple countries, country, region of country (i.e., multiple states or provinces), state or province, city, firm, and individual consumer levels. Also, in addition to ordinary least squares (OLS), studies have estimated meat demand using two-stage least squares (2SLS), three-stage least squares (3SLS), full information maximum likelihood (FIML), single-equation maximum likelihood (MLE), seemingly unrelated regression (SUR), generalized method of moments (GMM), generalized least squares (GLS), and although sparingly, the minimum distance estimator and maximum entropy.

Fifth, information on the publication outlet in which each of the 393 studies appeared was also collected. In particular, we note the year in which the study was published, as well as whether or not the study was published in a premium journal, such as a top-36 economics journal (as identified by Scott and Mitias (1996)) or the American Journal of Agricultural Economics (AJAE), and whether or not the study was published in a book.

Lastly, the demand for meat has been estimated throughout the world. Accordingly, using the Nations Online Project, we note the location of demand across 13 different regions (i.e., Australia, North America, South America, North Europe, West Europe, South Europe, East Europe, East Asia, South East Asia, South Central Asia, Middle East, South Africa, and other parts of Africa).2

2.2 Meta-regression model

Observations of the income elasticity of meat collected from the literature serve as the dependent variable in a series of meta-regressions. Specifically, as studies typically report multiple income elasticity estimates, similar to other meta-analyses (e.g., Rosenberger and Loomis 2000; Gallet and List 2003; Johnston et al. 2006; Gallet 2010), we consider the following unbalanced panel data meta-regression model:

image(1)

where Eij is the study i’s jth income elasticity estimate, αi is the ‘random researcher’ effect, which controls for unobserved study-specific effects that might influence the income elasticity, β is the vector of coefficients, and Xij accounts for the study characteristics mentioned previously. Specifically, included in Xij are the year the study was published, as well as a series of dummy variables controlling for each of the study characteristics mentioned (i.e., variable equals 1 if the respective study characteristic holds, 0 if not).3 Finally, eij is an iid error term with zero mean and variance inline image.

There are several issues concerning the estimation of Equation (1) that need to be addressed. First, to avoid perfect multicollinearity, dummy variables for several of the study characteristics must be dropped from the meta-regressions. These variables comprise the baseline upon which results are compared.4 Second, because many of the study characteristics in Xij do not vary within studies, this prevents using a fixed effects estimator. Instead, in addition to using OLS as a point of comparison, we estimate Equation (1) using a random effects estimator. Third, White’s (1980) test rejected the null of no heteroskedasticity in each meta-regression, and so similar to other meta-analyses of the income elasticity (e.g., Espey 1998; Dalhuisen et al. 2003; Gallet 2007), heteroskedasticity-consistent standard errors are used to construct t-statistics. Fourth, we explore the impact of different meta-regression specifications by comparing the results with all study characteristics included as regressors (labeled the full model) to those that exclude study characteristics that are jointly insignificant in the full model (labeled the restricted model). Fifth, across all 3357 observations, the mean income elasticity equals 0.90, and so a positive (negative) meta-regression coefficient is interpreted as that particular study characteristic inflating (deflating) the income elasticity.

3. Estimation results

Table 2 presents the results for the full and restricted models. Across nine major categories of variables, each restricted model was determined by eliminating those categories for which the corresponding coefficients were jointly insignificant in the full model. Accordingly, based on the F-test values at the bottom of Table 2, the restricted model corresponding to the OLS meta-regression eliminates the variables controlling for the nature of data and temporal aggregation, while the restricted model corresponding to the random effects meta-regression eliminates the variables controlling for specification issues, nature of data, and spatial aggregation. As provided at the bottom of Table 2, LaGrange multiplier tests reject the null hypothesis of homogeneous researcher effects, thus favoring the random effect results over the OLS results. Nonetheless, a perusal of the coefficients in Table 2 indicates similarities in their sign and significance across the meta-regressions, and so rather than discussing the results of each meta-regression separately, we focus on the pattern of the coefficients across all four meta-regressions.

Table 2.   Meta-regression results
CategoryVariableFull modelRestricted model
OLSRandom effectsOLSRandom effects
  1. Note: t-statistics (in absolute value) provided in parentheses. Levels of significance: *10%, **5%, and ***1%. †F-tests of the joint significance of coefficients associated with respective category. For example, F (product) refers to an F-test of the significance of the five coefficients of the meat product dummy variables. SUR, seemingly unrelated regression; OLS, ordinary least squares; MLE, single-equation maximum likelihood; GMM, generalized method of moments; GLS, generalized least squares; FIML, full information maximum likelihood; 2SLS, two-stage least squares; 3SLS, three-stage least squares.

ProductBeef0.0080.02850.0090.032
(0.202)(0.651)(0.228)(0.724)
Pork−0.175***−0.174***−0.172***−0.171***
(4.773)(4.117)(4.454)(3.990)
Lamb−0.146*−0.227***−0.140*−0.222**
(1.686)(2.684)(1.645)(2.491)
Poultry−0.125**−0.156***−0.123**−0.148***
(2.438)(2.961)(2.536)(2.599)
Fish−0.005−0.0750.015−0.070
(0.068)(1.141)(0.263)(0.989)
Functional formDouble-Log0.0810.0310.0780.017
(0.710)(0.196)(0.703)(0.111)
Semi-Log−0.190**−0.221*−0.210***−0.222**
(2.251)(1.930)(2.761)(1.974)
AIDS-Nonlinear−0.067−0.034−0.066−0.049
(0.659)(0.229)(0.655)(0.336)
AIDS-Linear−0.025−0.150−0.027−0.153
(0.296)(1.118)(0.322)(1.170)
AIDS-Quadratic0.005−0.0400.009−0.112
(0.047)(0.247)(0.089)(0.631)
AIDS-General0.023−0.0610.021−0.064
(0.159)(0.146)(0.137)(0.154)
Rotterdam−0.029−0.098−0.035−0.123
(0.320)(0.803)(0.381)(0.773)
CBS−0.310***−0.329*−0.317***−0.373**
(2.824)(1.944)(2.975)(1.971)
Translog0.335***0.1310.318***0.110
(3.285)(0.633)(3.189)(0.573)
S-Branch0.615***0.405*0.620***0.287
(5.935)(1.820)(6.219)(1.325)
Box–Cox−0.0530.003−0.060−0.014
(0.442)(0.031)(0.505)(0.138)
Other form0.010−0.159−0.006−0.194
(0.125)(1.183)(0.071)(1.476)
Specification issuesSubstitute meats−0.077**−0.006−0.091*** 
(2.554)(0.081)(3.035) 
Two-step0.021−0.042−0.006 
(0.517)(0.593)(0.160) 
Dynamic0.032−0.0610.045 
(0.641)(0.774)(1.010) 
Nature of dataTime-series−0.214−0.104  
(1.397)(0.356)  
Cross-sectional−0.0310.038  
(0.429)(0.615)  
Temporal aggregationQuarterly0.0430.209*** 0.262***
(0.876)(3.089) (4.880)
Less than quarterly0.169*0.239** 0.338***
(1.710)(2.032) (3.088)
Spatial aggregationMultiple countries0.4840.3070.329 
(1.248)(0.461)(1.249) 
Country0.417**0.3030.244*** 
(2.425)(0.865)(4.165) 
Region of country0.952***1.802**0.839*** 
(3.042)(2.344)(2.734) 
State/province0.143−0.0210.122** 
(1.449)(0.114)(2.012) 
City0.3000.4190.284 
(1.368)(1.178)(1.339) 
Firm0.527***0.4150.492*** 
(2.588)(1.104)(3.631) 
Estimation method2SLS0.744***0.356**0.753***0.382***
(4.694)(2.450)(4.761)(2.974)
3SLS0.247***0.0080.253***0.110
(3.419)(0.064)(3.456)(0.839)
FIML0.202***0.0620.182***0.091
(6.519)(1.065)(6.063)(1.618)
MLE−0.025−0.045−0.036−0.041
(0.834)(1.131)(1.209)(0.918)
SUR0.114***−0.0440.102**−0.013
(2.759)(0.751)(2.289)(0.197)
GMM0.101−0.1080.0390.039
(0.958)(0.487)(0.344)(0.162)
GLS0.032−0.1000.068−0.082
(0.266)(0.753)(0.575)(0.507)
Other method−0.201*−0.231−0.247**−0.175
(1.710)(1.213)(2.198)(0.882)
PublicationTop-36 journal−0.091**0.014−0.068*−0.025
(2.362)(0.199)(1.766)(0.367)
AJAE−0.0060.026−0.0220.005
(0.199)(0.429)(0.858)(0.093)
Book0.182***0.196*0.148***0.267**
(3.389)(1.677)(3.294)(2.565)
Year published0.009***0.011***0.010***0.008***
(4.807)(3.648)(7.013)(2.889)
RegionAustralia−0.416***−0.480**−0.421***−0.553***
(3.184)(2.286)(4.030)(3.823)
North America−0.100−0.117−0.109−0.136
(0.853)(0.581)(1.036)(0.913)
South America−0.287−0.425−0.336*−0.511**
(1.544)(1.446)(1.881)(2.319)
North Europe−0.143−0.010−0.154*−0.051
(1.412)(0.045)(1.699)(0.285)
West Europe0.1460.1840.1180.196
(1.304)(0.800)(1.239)(1.214)
South Europe0.1210.1870.1120.139
(1.105)(0.897)(1.175)(1.237)
East Europe−0.040−0.059−0.045−0.158
(0.240)(0.280)(0.305)(1.624)
East Asia0.0620.0770.0520.058
(0.619)(0.407)(0.592)(0.323)
South East Asia0.1680.1050.1280.059
(1.283)(0.430)(0.980)(0.309)
South Central Asia0.361***0.356*0.349***0.267
(2.960)(1.770)(2.909)(1.514)
Middle East0.558***0.864***0.520***0.929**
(6.082)(3.457)(5.451)(2.039)
South Africa−0.185−0.133−0.234−0.138
(0.930)(0.433)(1.153)(0.499)
Other Africa−0.239*−0.117−0.283**−0.165
(1.735)(0.510)(2.229)(0.800)
F (Product)† 9.84510.08010.0559.736
F (Functional Form) 57.3678.16458.8026.021
F (Specification Issues) 2.4690.2334.155
F (Nature of Data) 0.9860.493
F (Temporal Aggregation) 1.6065.01115.734
F (Spatial Aggregation) 3.8311.31211.072
F (Estimation Method) 14.0182.97415.4042.965
F (Publication) 48.7344.80957.1734.745
F (Region) 65.36320.32573.19411.954
Adjusted R2 0.120.12
χ2 (1 df) 334.92448.56
N 3357335733573357

There are several noteworthy results concerning the individual coefficients. First, compared to the baseline composite meat category, the income elasticity is significantly lower for pork, lamb, and poultry.5 Second, concerning the specification of meat demand, although the income elasticity tends to be deflated (inflated) when a semi-log or CBS (translog or S-branch) functional form is adopted, for the majority of the functional forms the meta-regression coefficients are insignificantly different from zero. Consequently, compared to the linear baseline form, theoretically consistent functional forms, such as the various AIDS forms and the Rotterdam form, have little statistical influence on the estimated income elasticity.6 Also, with the exception of including substitute meats in the OLS meta-regressions, specification issues matter little in determining the income elasticity.

Third, given that many of the coefficients associated with data issues are jointly insignificant, data issues overall appear to have little influence on the income elasticity. Yet there are a number of individually significant coefficients associated with temporal and spatial aggregation of data that do affect the income elasticity. In particular, compared to the baseline use of annual data from individual consumers, the use of quarterly and less than quarterly data, as well as data aggregated to the country, region of country, and firm-level tend to inflate the income elasticity.7

Fourth, there is a noticeable difference between the OLS and random effects results concerning the influence of estimation methods on the income elasticity of meat. In particular, compared to the baseline OLS estimator, the use of 2SLS, 3SLS, FIML, and SUR inflates the income elasticity in the OLS meta-regressions, while the use of other methods (i.e., minimum distance and maximum entropy) deflates the income elasticity. With the exception of 2SLS, though, each of the estimation methods fails to significantly affect the income elasticity in the random effects meta-regressions.

Fifth, similar to Gallet (2010), we find certain publication characteristics influence the income elasticity of meat. Specifically, across all four meta-regressions, not only is the income elasticity higher when published in a book, but more recent studies report higher income elasticities compared to older studies.8 Nonetheless, publishing in the AJAE or a top-36 economics journal (with the exception of the OLS results) does not appreciably influence the reported income elasticity.

Lastly, although the coefficients of many of the region dummy variables are insignificantly different from zero, which suggests the income elasticity differs little across locations, there are a few notable regions. In particular, across the majority of meta-regressions, we find the income elasticity is lower in Australia and higher in South Central Asia and the Middle East, which is consistent with the preferences for meat differing in these regions.

4. Concluding comments

Based on the meta-regression results, we find several patterns concerning estimates of the income elasticity of meat in the literature. For instance, the income elasticities of lamb, pork, and poultry tend to be lower than those of other meats. Furthermore, the income elasticity is sensitive to a few functional forms, data aggregation, publication, and regional characteristics. Nonetheless, it is interesting that a number of factors commonly employed in the literature (e.g., AIDS and Rotterdam functional forms, other specification issues, whether or not time-series or cross-section data is used, and many estimation methods) do not significantly affect the reported income elasticity; and so less concern needs to be given to such factors when choosing an income elasticity from the literature.

Having a more clear understanding of tendencies in the literature to sway the income elasticity one way or the other is beneficial to policymakers and academics alike. For instance, based on our results, increasing income will shift a greater (lesser) budget share towards beef and fish (lamb, pork, and poultry). Not only is this of interest to those teaching courses in consumer theory, but such a finding suggests that policymakers wishing to alter meat consumption (e.g., shift consumption away from certain meats towards others) should develop policies that are meat specific. Furthermore, our results suggest avenues for future research to uncover why such tendencies are observed in the literature.

Footnotes

  • 1

     Initially, 3363 income elasticity estimates were retrieved from the 393 studies. However, six observations were two or more standard deviations from the mean, and so similar to that stated by Gallet (2010), these observations were dropped to reduce the influence of outliers. Nonetheless, the results presented in Section 3 change little whether or not these six outliers are included.

  • 2

     See Gallet (2010) for the frequency of each study characteristic. For example, in the literature, it is most common to adopt the AIDS-Linear specification of meat demand, which is estimated with country-level time-series data using SUR.

  • 3

     Because they are adopted infrequently in the literature, the generalized addilog and quadratic expenditure functional forms are collectively accounted for by the dummy variable labeled ‘Other Form’, while the minimum distance and maximum entropy estimators are collectively accounted for by the dummy variable labeled ‘Other Method’.

  • 4

     For instance, similar to Gallet (2010), the dummy variable corresponding to the composite meat category is dropped from each meta-regression, and so results are interpreted relative to this baseline meat. The baseline further corresponds to one obtained from a linear version of meat demand (absent substitute meats, dynamic considerations, and a two-step treatment) that is estimated with panel data (aggregated to the annual individual consumer level) using OLS. Also, the baseline income elasticity is not published in a top-36 economics journal, the AJAE, or a book. Finally, the baseline income elasticity is not specific to a particular region of the world.

  • 5

     To put these differences into perspective, using the random effects results for the full model, similar to that followed by Gallet (2010), the predicted income elasticities for each meat are calculated at the mean of each study characteristic (with the exception of the dummy variables corresponding to each other meat, which are set to zero). At these values, the rank order of income elasticities (provided in parentheses) are as follows: beef (1.00), composite meat (0.97), fish (0.90), poultry (0.82), pork (0.80), and lamb (0.74). Hence, the income elasticity of lamb is nearly 25 per cent lower than that of beef, ceteris paribus.

  • 6

     Although we might expect theory-based functional forms to yield estimates closer to the true demand and thus contribute to differences in income elasticity estimates across functional forms, the meta-regression results do not provide appreciable evidence of this.

  • 7

     Such results are consistent with a number of studies (i.e., Blundell et al. 1993; Denton and Mountain 2001) that find evidence of aggregation bias in the estimation demand.

  • 8

     This positive trend in the income elasticity could be the result of (i) changes in consumer preferences over time or (ii) later studies extending the results of earlier studies, thereby refining income elasticity estimates.

Ancillary