Market Compettveness and Demographc Profles of Dary Alternatve Beverages n the Unted States: The Case of Soymlk Senarath Dharmasena* Oral Capps, Jr.* *Agrbusness, Food and Consumer Economcs Research Center (AFCERC) Texas A&M Unversty Emals: sdharmasena@tamu.edu ocapps@tamu.edu Phone: (979) 845-8491 Selected Paper, Southern Agrcultural Economcs Assocaton (SAEA) Annual Meetng, Brmngham, Alabama, February 4-7, 2012 Copyrght 2012 by Senarath Dharmasena and Oral Capps, Jr. All rghts reserved. Readers may make verbatm coped of ths document for non-commercal purposes by any means, provded that ths copyrght notce appears on all such copes.
Market Compettveness and Demographc Profles of Dary Alternatve Beverages n the Unted States: The Case of Soymlk Abstract Data from U.S. households for year 2008 were used n examnng market compettveness of soymlk usng tobt procedure. Uncondtonal own- and crossprce elastctes are larger than ther condtonal counterparts. Income, age, employment status, educaton level, race, ethncty, regon and presence of chldren are sgnfcant drvers affectng the demand for soymlk. Key Words: Soymlk, whte mlk, flavored mlk, Nelsen HomeScan data, tobt procedure JEL Classfcaton: D11, D12 1
Market Compettveness and Demographc Profles of Dary Alternatve Beverages n the Unted States: The Case of Soymlk Background: Currently, calcum fortfed soy based dary alternatve beverages are enterng the market on one hand to compete wth whte and flavored mlk n the marketplace and on the other provdng consumers an alternatve calcum fortfed beverage, specfcally for those who are havng trouble consumng dary based products (due to lactose ntolerance and other health concerns). To strengthen the poston of calcum fortfed soy-based beverages n the U.S. market, the new food gudelnes developed under the ChooseMyPlate of Unted States Department of Agrculture, placed calcum fortfed soy beverages n the Dary Group whch s ntroduced as a sde dsh (USDA, 2011). Ths rased eyebrows of dary producers and dary marketers n the Unted States and t s of best nterest for them to know the compettveness of calcum fortfed dary alternatve beverages, n partcular, soy beverages n the U.S. dary marketplace. Accordng to Beverage Marketng Corporaton (2010), soymlk has been one of the fastest growng categores n the general beverage marketplace and has had a much hgher growth rate than the dary mlk segment over the last decade. Growth n soymlk has been attrbuted to mproved health-related clams and consumer perceptons, flurry of soymlk brands, appealng and convenent packagng and multtude on flavors avalable. Soy beverage retal sales topped to $1.7 bllon n 2008 and contnue to grow addng flavored soymlk products such as chocolate, vanlla and strawberres, hence drectly competng wth flavored dary mlk (Beverage Marketng Corporaton, 2010). As far as brand specfc nformaton 2
s concerned, Slk soymlk brand has the hghest market share (62%), followed by Rce Dream (6%), 8 th Contnent (6%), Lfeway (2%), and Odwalla (1%) (Beverage Marketng Corporaton, 2010). Gven ths backdrop, knowledge of prce senstvty, substtutes and complements and demographc profles wth respect to consumpton of soymlk s mportant for manufacturers, retalers and advertsers of soymlk and dary mlk from a compettve ntellgence perspectve as well as from a strategc decsonmakng perspectve. We dd not fnd any past study pertanng to demand for soymlk n the extant lterature. Therefore, to the best our knowledge, our study s the frst to examne the market compettveness and demographc factors determnng U.S. demand for soymlk. A thorough and a complete analyss of demand for soymlk s mportant due to ncreasng growth n consumpton n recent tmes as an alternatve beverage to dary based mlk n the Unted States and to the lack of nformaton n the lterature. In ths lght, specfc objectves are: (1) to determne the condtonal factors affectng the volume of purchase of soymlk; (2) to determne the uncondtonal factors affectng the volume of purchase of soymlk; (3) to determne the condtonal and uncondtonal own- and cross-prce demand elastctes of soymlk and ts compettors; (4) to determne retal level prcng strateges for soymlk n compettve marketplace. Data and Methodology Household purchases of soymlk, whte mlk and flavored mlk (expendture and quantty) and soco-economc-demographc characterstcs are generated for each household n the Nelsen Homescan Panel for calendar year 2008 (total of 3
61,440 households). Only 7,729 households purchased soymlk, whle 58,268 households purchased whte mlk. Flavored mlk was purchased by 16,468 households. Quantty data are standardzed n terms of lqud ounces and expendture data are expressed n terms of dollars. Then takng the rato of expendture to volume, we generate unt values (prces n dollars per ounce) for each beverage category. Factors hypotheszed to affect the volume of soymlk purchased are: prce of soymlk, prce of competng beverages such as regular whte mlk and flavored mlk; age, gender, employment and educaton status of the household head; regon; race; Hspanc orgn; age and presence of chldren, ncome of the household. A common characterstc n mcro level data (data gathered at consumer level such as at the ndvdual or household level) s a stuaton where some consumers do not purchase some tems durng the samplng perod and presence of them n the sample creates a zero consumpton level for that data perod. The data used n ths study are gathered at household level and due to that t suffers from zero consumpton data, hence zero expendture. As such we face a censored sample of data. Applcaton of ordnary least squares (OLS) to estmate a regresson wth a lmted dependent varable (such as n a censored sample lke ours) usually gve rse to based estmates, even asymptotcally (Kennedy, 2003). Removng all observatons pertanng to zero purchases and estmatng regresson functons only for non-zero purchases too creates a bas n the estmates (Kennedy, 2003). Ths phenomenon also s known as sample selecton bas. Tobn (1958) and Heckman (1979) suggested alternatve models to deal wth sample selecton bas n estmatng regresson models n the presence of censored data. In ths paper,we 4
center attenton on Tobn (1958) model to glean both condtonal and uncondtonal demand estmates pertanng to soymlk. Heckman (1979) model only wll be able to speak to condtonal demand estmates, although n the frst stage probt analyss wll provde nformaton on consumer s probablty to purchase or not to purchase the product. Also, we use the decomposton of the beta coeffcent estmates of tobt model suggested by McDonald and Mofftt (1980) to shed lght on changes n probablty of beng above the lmt (lmt beng zero n ths paper) and changes n the value of the dependent varable f t s already above the lmt. Ths s the McDonald and Mofftt decomposton assocated wth tobt parameter estmates. For all those transactons assocated wth zero quanttes and hence zero expendtures, we do not observe any unt value or prce. However, snce we are expectng to use prce of each beverage category as explanatory varables n the tobt model, we have to mpute prce for those observatons where no prce s observed. Prce mputaton s done usng an auxlary regresson, where observed prces for each beverage are regressed on household ncome, household sze and regon where the household s located. These varables are used extensvely n the prce mputaton lterature as good nstruments n mputng prces. Once the prces for each beverage concerned (soymlk, whte mlk, and flavored mlk) are mputed, we use them and aforementoned explanatory varables to estmate the tobt model pertanng to soymlk consumpton. Table 1 shows dfferent categores of explanatory varables used n ths study along wth base categores for dummy varables. 5
The Tobt Model follows: The stochastc model underlyng the tobt model can be expressed as (1) =, >0 0, 0 where =1,2,3,..,, the number of observatons. s the censored dependent varable; s the vector of explanatory varables; s the vector of unknown parameters to be estmated; =0 and ~(0, ). The uncondtonal expected value for s expressed n equaton (2) and the correspondng condtonal expected value for s shown n equaton (3), where the normalzed ndex value z s shown as =. Also, () s the cumulatve dstrbuton functon (CDF) assocated wth z and () s the correspondng probablty densty functon (pdf). (2) ()=()() (3) ( )= () () The uncondtonal margnal effect s represented by, (4) () =() The condtonal margnal effete s shown by, (5) ( ) =(1 () () () () ) 6
Emprcal Estmaton We tred several functonal forms such as lner, quadratc and lnear-log to fnd that Lnear-Log model (we used logged prce varables n the model) outperforms other functonal forms as far as the model ft, sgnfcance of varables and loss matrces such as AIC and Schwarz crtera are concerned. The tobt model for soymlk can be represented as follows, (6) ( Q _ Soy _ Mlk) = β β log PRICE _ SOY β log PRICE _ SOY β log PRICE _ WMILK 3 β AGEHH 3034 7 β AGEHHGT 64 11 β EDUHHU 15 β WEST 19 β OTHER 22 25 27 29 β MHONLY 31 β 20 β EDUHHPC BLACK 23 β β AGEHH 3544 β EMPHHPT β AGEPCLT 6 _ 6 _12ONLY 16 32 1 8 12 β AGEPC6 _12ONLY 2 FHONLY β PRICE _ FMILK β 5 21 β HISP _ YES 26 β MIDWEST ASIAN β AGEPC6 _12AND13 _17ONLY 24 28 β INCOME 33 β EMPHHFT β AGEHH 4554 17 β AGEPC13 _17ONLY 13 30 9 2 β AGEHH 2529 6 β EDUHHHS β SOUTH β AGEPCLT 6 _ ONLY β AGEPCLT 6 _13 _17ONLY 18 14 10 β AGEHH 5564 β AGEPCLT 6 _ 6 _12AND13 _17 As such, we wll calculate both condtonal and uncondtonal margnal effects assocated wth each explanatory varable. The level of sgnfcance we wll be usng n ths study s 0.05. We further conduct an F-test for demographc varable categores to fnd statstcally sgnfcant demographcs. The equatons for uncondtonal and condtonal margnal effects for the Lnear-Log model and correspondng uncondtonal and condtonal own- and cross-prce elastcty estmates are explaned below. The uncondtonal margnal effect for the Lnear-Log model s as follows, (7) () = () 7
where s the average prce of all observatons (uncondtonal prce) consdered. The condtonal margnal effect for the Lnear-Log model s as follows, (8) ( ) = (1 () () () () ) Where, s the average prce of non-censored sample (condtonal prce). The uncondtonal own- and cross-prce demand elastctes are represented by equatons (9) and (10) respectvely. (9) = (10) = () () The condtonal own- and cross-prce demand elastctes are represented by equatons (11) and (12) respectvely, (11) = (12) = (1 () (1 () () () () ) () () () ) Results and Dscusson Our analyss reveals that market penetraton for soymlk, whte mlk and flavored mlk s 12.6%, 94.8% and 26.8% respectvely. The uncondtonal own-prce elastcty of demand for soymlk s estmated to be -1.98 and ts condtonal counterpart s -0.41. It must be noted when the whole sample of observatons are concerned, soymlk shows an elastc demand vs-à-vs an nelastc demand for the truncated sample wth those who actually purchase soymlk. In other words, the sample of consumers who actually bought soymlk s not very prce senstve, whereas the pooled sample (wth those who bought and dd not buy) would move 8
away from soymlk to ts closest substtute n the event of prce ncrease (more prce senstve). The uncondtonal cross-prce elastcty of demand wth respect to soymlk and whte mlk s 1.26 and the condtonal counterpart assocated wth ths crossprce elastcty s estmated to be 0.25. Smlar trend s observed wth the crossprce elastcty assocated wth flavored mlk, where the uncondtonal cross-prce elastcty wth respect to flavored mlk s 0.16 and the condtonal cross-prce elastcty s estmated to be 0.03. In all, both whte mlk and flavored mlk are substtutes n consumpton for soymlk. Income, age of the household head, employment status of household head, educaton level of household head, race, ethncty, regon and presence of chldren n the household are found to be mportant n affectng the demand for soymlk. References: Beverage Marketng Corporaton, (2010), Soy Beverages n the U.S. Heckman, J. J., (1979). Sample Selecton Bas as a Specfcaton Error. Econometrca 47: 153-161 Kennedy, P. (2003). Lmted Dependent Varables. A Gude to Econometrcs, MIT Press. McDonald, J. F., and R. A. Mofftt, (1980), The Uses of Tobt Analyss. The Revew of Economcs and Statstcs, 62(2): 318-321 Tobn, J., (1958), Estmaton of Relatonshps for Lmted Dependent Varables. Econometrca, 26(1): 24-36 USDA, (2011). What foods are ncluded n the Dary Group?. ChooseMyPlate, (nternet access on September 9, 2011: http://www.choosemyplate.gov/foodgroups/dary.html) 9
Table 1 Descrpton of the Rght-Hand Sde Varables Used n the Econometrc Analyss Varable Explanaton PRICE AGEHHLT25 AGEHH2529 AGEHH3034 AGEHH3544 AGEHH4554 AGEHH5564 AGEHHGT64 EMPHHNFP EMPHHPT EMPHHFT EDUHHLTHS EDUHHHS EDUHHU EDUHHPC EAST MIDWEST SOUTH WEST WHITE BLACK ASIAN RACE_OTHER HISP_NO HISP_YES Prce of Drnkable yogurt Age of Household Head less than 25 years (Base category) Age of Household Head between 25-29 years Age of household Head between 30-34 years Age of household Head between 35-44 years Age of household Head between 45-54 years Age of household Head between 55-64 years Age of household Head greater than 64 years Household Head not employed for full pay (Base category) Household Head Part-tme Employed household Head Full-tme Employed Educaton of Household Head: Less than hgh school (Base category) Educaton of Household Head: Hgh school only Educaton of Household Head: Undergraduate only Educaton of Household Head: Some post-college Regon: East (Base category) Regon: Central (Mdwest) Regon South Regon West Race Whte (Base category) Race Black Race Orental Race Other (non-black, non-whte, non-orental) Non-Hspanc Ethncty (Base category) Hspanc Ethncty
Table 1 Contnued. Varable NPCLT_18 AGEPCLT6_ONLY AGEPC6_12ONLY AGEPC13_17ONLY AGEPCLT6_6_12ONLY AGEPCLT6_13_17ONLY No Chld less than 18 years (Base category) Explanaton Age and Presence of Chldren less than 6-years Age and Presence of Chldren between 6-12 years Age and Presence of Chldren between 13-17 years Age and Presence of Chldren less than 6 and 6-12 years Age and Presence of Chldren less than 6 and 13-17 years AGEPC6_12AND13_17ONLY Age and Presence of Chldren between 6-12 and 13-17 years AGEPCLT6_6_12AND13_17 FHMH MHONLY FHONLY Age and Presence of Chldren less than 6, 6-12 and 13-17 years Household Head both Male and Female (Base category) Household Head Male only Household Head Female only 11