DETERMINING THE CHARACTERISTICS OF SEAFOOD CONSUMERS AND THEIR PREFERENCES FOR FRESH SEAFOOD: EVIDENCE FORM HOUSE H OLD SCANNER DATA FROM FIVE US MARKETS By MATTHEW GORSTEIN A THESIS PRESENTED TO THE GRADUATE SCHOOL OF THE U NIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF SCIENC E UNIVERSITY OF FLORIDA 2014
Â© 2014 Matthew Gorstein
To my mother, father, and sister
4 ACKNOWLEDGMENTS I w ould like to thank my parents and my sister for their constant love and support. Additionally, I would like to thank my professors , most notably Dr. Larkin, for taking an interest in me , providing guidance, and being accessible throughout my academic care er. Finally, I thank the University of Florida for helping to mold me into the man that I am today.
5 TABLE OF CONTENTS page ACKNOWLEDGMENTS ................................ ................................ ................................ .. 4 LIST OF TABLES ................................ ................................ ................................ ............ 7 LIST OF FIGURES ................................ ................................ ................................ .......... 8 ABSTRACT ................................ ................................ ................................ ..................... 9 CHAPTER 1 INTRODUCTION ................................ ................................ ................................ .... 11 Goals and Objectives ................................ ................................ .............................. 13 Overview of Study ................................ ................................ ................................ ... 15 2 LITERATURE REVIEW ................................ ................................ .......................... 17 Retail level Scanner Data and Food Products ................................ ........................ 17 Household level Scanner Data and Food Products ................................ ................ 19 Separability of Seafood Products from Other Meat Pro ducts ................................ .. 20 Choice Modeling of Food Purchases ................................ ................................ ...... 21 3 THEORETICAL BACKGROUND ................................ ................................ ............ 23 4 DATA ................................ ................................ ................................ ...................... 29 Fresh Seafood Expenditures ................................ ................................ .................. 31 Demographic Comparison of Seafood Eaters and Non Seafood Eaters ................ 32 Past Purchase Behavior and Species/Product Form Preferences .......................... 34 Comparison of Census with Panelists by Market Area ................................ ........... 35 5 MODEL SPECIFICATION ................................ ................................ ....................... 51 6 EMPIRICAL RESULTS ................................ ................................ ........................... 55 Model 1: Predicting the Probability of Household Purchasing Seafood .................. 56 Model 2: Predicting the Probability of a Household Purchasing Fresh Seafood ..... 56 Odds Ratios ................................ ................................ ................................ ............ 58 7 CONCLUSIONS ................................ ................................ ................................ ..... 67 Overview of Project ................................ ................................ ................................ . 67 Discussion and Implicatio ns of Results ................................ ................................ ... 67 Recommendations to Firms and Marketers ................................ ............................ 69 Further Research ................................ ................................ ................................ .... 70
6 LIST OF REFERENCES ................................ ................................ ............................... 73 BIOGRAPHICAL SKETCH ................................ ................................ ............................ 76
7 LIST OF TABLES Table page 4 1 Fresh seafood expenditures as a percent of total seafood expenditures by all panelists in all five market areas, 2008 2010. ................................ .................... 40 4 2 Demographic comparison of seafood consuming households and non seafood consuming households. ................................ ................................ ........ 41 4 3 Distribution of annual household income amongst the panel. ............................ 42 4 4 Summary statisti cs of past purchase behavior and species/product form preference variables (n = 1,961 households). ................................ .................... 43 4 5 Top ten average annual expenditures by species by all households, 2008 2010. ................................ ................................ ................................ .................. 44 4 6 Census vs. Homescan comparison for Chicago market area. ............................ 45 4 7 Census vs. Homescan comparison for Miami market area. ................................ 46 4 8 Census vs. Homescan comparison for Houston market area. ............................ 47 4 9 Census vs. Homescan comparison for Memphis market area. ........................... 48 4 10 Census vs. Homescan comparison for New Orleans Mobile market area. ......... 49 6 1 Model 1: Predicting the probability of a household purchasin g seafood (n = 2,557 households). ................................ ................................ ............................. 62 6 2 Model 2: Predicting the probability of a household purchasing fresh, over the counter seafood (n = 1,961 households). ................................ ........................... 63
8 LIST OF FIGURES Figure page 4 1 Average household allocation of seafood expenditures for fresh seafood and frozen seafood by species (n=1,961). ................................ ................................ . 50 6 1 Effect plot for average duration between purchases. ................................ .......... 64 6 2 Effect plot for percent share of seafood expenditure dedicated to breaded process form purchas es. ................................ ................................ .................... 64 6 3 Effect plot for number of seafood items purchased. ................................ ........... 65 6 4 Change in probabilities of purchasing seafood based on stat istically significant predictor variables in model 1. ................................ ........................... 65 6 5 Change in probabilities of purchasing fresh, over the counter seafood based on statistically significant predictor variables in mod el 2. ................................ .... 66
9 Abstract of Thesis Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Master of Science DETERMINING THE CHARACTERISTICS OF SEAFOOD CONSUMERS AND THEIR PREFERENCES FOR FRESH SEAFOOD: EVIDENCE FORM HOUS E H OLD SCANNER DATA FROM FIVE US MARKETS By Matthew Gorstein December 2014 Chair: Sherry Larkin Major: Food and Resource Economics AC Nielsen Homescan data was used in this pr oject to gain insight into the traits, their preferences, and the behavioral attri butes of the consumer. The data consist of household level weekly seafood purchases by UPC over the course of three years for five major market areas in the United States: Miami, Chicago, Houston, Memphis, and New Orleans Mobile. Seafood can be arg ued as separable from other meat products, as studies concerning other forms of meat have left seafood absent form their analysis (An ders and Moeser 2008). Through logistic regression model s, using the maximum likelihood estimation approach, the probability of a household purchasing seafood, and the probability of a seafood consuming househo ld purchasing fresh seafood have be en analyzed . Household composition, educa tion, income, age, and region have statistically significant effects on the likelihood of a househo ld purchasing seafood. Additionally,
10 past purchase behavior such as average weekly seafood expenditure, coupon use, average duration between purchases, number of seafood items purchased , and household characteristics such as region, age of household head , and hous ehold composition all have statistically significant effects on the likelihood of a seafood consuming household purchasing fresh, over the counter seafood. Results from this analysis will serve to help firms create opportunities/dispel myths, aid marketers in associating demographic characteristics with seafood pur chase behavior, and gain valuable knowledge concerning the market for fresh, over the counter seafood.
11 CHAPTER 1 INTRODUCTION Since the dawn of civilization, populations have strategica lly positioned themselves near the coast of oceans, seas, and rivers, leading to dependence on the resources of the sea for nutrition, sustenance, and a way to make a living. As a result, s many cases, seafood products are inherent to the survival of particular communities. Therefore, billions of people have a large stake in the production and procurement of seafood. Determining why a consumer chooses to purchase seafood (what drives the ch oice) has many implications for this vast array of stakeholders involved in seafood procurement and seafood production. And, by extension, careful management of re sources from the sea is necessary to provide humans with a valuable and sustainable food sou rce for present and future generations . At the individual level, s eaf ood products may be sought after because they are considered to be a healthy protein alternative due to the presence of omega 3 fatty acids, which are important for normal metabolism and have been linked to improved immune function ( Daly et al. 1992). Perception alone does not drive a consumer choice, therefore t his project seeks to identify what other factors drive a consumer to choose to purchase seafood, and moreover , what factors dri ve a consumer to choose to purchase fresh seafood. It is believed that the type of consumer (i.e. their demographics) plays a large role in consumer choices (Nauman et al. 1995) , so these demographic aspects wi ll be investigated thoroughly. Despite fore casts of declining worldwide seafood supply, the sale and production of seafood products has become increasingly competitive (Wessels 2002). Moreover,
12 retail trends in the seafood market over the course of 2013 show a marked increase in the volume of fres h seafood sales, as well as the dollar value of fresh seafood sales (Frey 2014) , which have many implications for commercial fishers, seafood marketers, and others throughout the supply chain of seafood production. Additionally, investigation into the con sumer market for fresh, over the counter seafood products has been scarcely conducted. This analysis will seek to fill in this aforementioned gap in the previous literature. Information on consumer demand for seaf ood products has primarily been obtained f obtained from surveys) is critically important for determining potential demand for new for as sessing tradeoffs between existing products and product attributes because it is stores provides UPC level data on prices and quantities sold , but the data sets are expensiv e and few grants provide sufficient funds for acquisition (especially over a long time horizon) . As a result, published studies using scanner data have examined a very small set of product types (i.e., organic eggs, fluid milk, and certified organic salmon and P ollock). Most importantly, scanner data from grocers does not contain demographic information on the purchaser, which severely limits the extent to which results can be used for effective marketing campaigns. In order to market seafood products effi ciently and effective ly, information on purchasers in a market (i.e. what characteristics do the consumers of a particular product, in this case,
13 seafood products, exhibit) is needed in order to determine the most suitable marketing strate gy. Electronic scanner sin grocery stores provide one type of behavioral data on consumer purchases. Electronic scanners, are however, being used within households to track all of the products that consumers buy. These data, household level scanner dat a , are panel data set s generated from a sample of survey participants that scan universal product codes (UPCs) of all products that were purchased on each trip that they make to the grocery store over a period of time (i.e. two weeks to three months) . Each of t he participating households pr ovide their demographic information, and record their purchases in exchange for compensation; these companies (i.e. AC Nielsen) then use this data to generate market reports (i.e. projected sales) that they sell to produc ers and suppliers. this aids in associating seafood consumption preferences with demographic characteristics and in modeling consumer demand for seafood products. Such data are rich for the exa min ation of actual purchase behavior , including the choices an d tradeoffs that consumers faced, including for seafood products. Goals and Objectives T his project seeks to add to the published literature pertaining to using scanner data at the household level to make inferences on the likelihood of purchasing seafood and the likelihood of purchasing fresh, over the counter seafood. As stated previously, investigation into the retail level market for seafood products has been scarcely conducted and determining what drives the choice of a consumer when they decide to pu rchase fresh, over the counter seafood , in particular, has rarely been examined. This analysis will add to the literature by first determining the characteristics of households that buy seafood (and fresh seafood), and comparing that to the general popula tion.
14 This analysis will add to the literature by first determining the characteristics of households that buy seafood (and fresh seafood) and comparing that to the general population. This will be accomplished using household level scanner data that wer e purchased from the Nielsen Corporation ( formerly known as ACNielsen or AC Nielsen ), which is a global market characteristics of buyers of different types of products will also be compared to the general population in each of five market areas: Chicago, Houston, New Orleans Mobile, Memphis and Miami. Using these same data, investigat ion of choice to purchase seafood in general. Since households that purchas ed seafood have product information associated with their demographic characteristics (whereas households in the data set who did not purchase seafood only have demographic s. Therefore, we first identify the household characteristics that affect the likelihood of purchasing seaf ood in general, a nd then identify the household characteristics and behaviors that drive the choice between purchasing fresh seafood sold at counters and purchasing frozen seafood in the freez er section of grocery store . In addition, the magnitude of these factors is determined and compared. Th e analysis in general seeks to explain how consumers of seafood are inherently different from consumers who d o not consume seafood, and that seafood eaters who consume fresh , over the counter seafood products are inherently different from seafood eaters who do not purchase fresh , over the counter seafood products. It is hypothesized that t hese differences exist d ue to demographic characteristics , as well as
15 behavioral characteristics and preferences of the purchaser. Predicting the likelihood of a household purchasing fresh, over the counter seafood as explained by other variables will aid in understanding and de scribing this virtually un researched group of consumers. In summary, this analysis model s the likelihood of a household purchas ing seafood; and from there, model s the likelihood of a seafood purchasing household purchasing fresh, over the counter seafood . Questions that will be addressed include but are not limited to: Do purchasers of fresh seafood purcha se seafood more frequently? Does coupon use affect the likelihood of a household purchasing fresh, over the counter seafood products? What are the ke y demographic traits associated with purchasers of fresh seafood? And are there any regional differences in the consumption of fresh, over the counter seafood? Results from this analysis will serve to help marketers gain insight into the seafood market acr oss these five major market areas and aid them in decisions and strategy . In addition, the analysis results can be used to help firms create opportunities or dispel myths when it comes the nature of the fresh, over the counter seafood product market . Over view of Study Following this introduction, the next chapter, Chapter 2, contains a review of relevant studies that have used scanner data , have assumed seafood products are separable from other protein products with respect to consumer demand, and have est imated probability based models. Chapter 3 provides a theoretical background on utility maximization and the need for linear probability models. Chapter 4 contains a description of the data used i n the analysis , including information on overall seafood e xpenditures, fresh seafood product expenditures, summary statistics of the variables
16 used in the analysis, and a city by city comparison of the households in the sample and the households in the population. In C hapter 5, an econometric model is specified to test for factors that explain households that purchase seafood and, of those, households that purchase fresh, over the counter seafood. The empirical results chapter (Chapter 6) include s descriptions of statistically significant odds ratio estimates an d their interpretations. In the final chapter (Chapter 7), the results are summarized and t he implications of the empirical results are discussed including ways to extend the analysi s and recommendations for marketers and firms that are involved in the pr oduction and sale of seafood products, and fresh seafood products in particular .
17 CHAPTER 2 LITERATURE REVIEW The literature review is organized in to four subsections. First, stu dies using point of sale (retail level) scanner data will be reviewed to sho w what types of questions can be answered with these data. Second, studies using household level scanner data are summarized to better understand what type of distinct information can be examined. Since this analysis only contains information on seafood products, how other researchers have addressed the separability of seafood products from other meat products is reviewed as well . Lastly, studies that have used (and estimated) probability based models and approaches to examine the factors that affect whe ther or not variables are correlated with a decision or outcome are reviewed. Retail level Scanner Data and F ood Products Retail level scanner data has been useful in trying to explain characte ristics of a market for seafood products . Retail scanner data h as been used to investigate the price impacts of promotional activities, the Gulf oil spill, major holidays, and product labeling hedo nic shadow pricing model to determ ine implicit price att ributes of seafood products and found that promotional activities, product labeling, product form, and time events such as Christmas, Lent, and the 2010 Gulf Deepwater Horizon Oil Spill all had statistically significant effects on the price of seafood product s . Roheim, Gardiner, and Asche 2007 also used retail level scanner data to conduct a hedonic shadow pricing model. This aforementioned study was performed in the United Kingdom and found that specie s was relevant across the suppl y chain, but that brand is only relevant at the consumer level. In addition to this, the study found that breaded and battered seafood
18 products, although technically value added, have a lower final price value when compared to smoked, fresh, or regular fro zen fish. This is due to the idea that the breaded/battered product originated from a lower qu ality in the beginning and is not of Retail level scanner data has also been used to calculate own price, cross price, and expenditure elasticities for different species of fish and shellfish, as well as provide information into the seasonality of seafood demand (Singh and Dey 2012). Singh and Dey found that all species of un breaded seafood (except flounder and lobster) had demand that was statistically significant ly affected by seasonality, that tilapia has a high overall degree of substitutability (measured by compensated cross price elasticity), and that the short run demand for shrimp and lobster is approximately unitary own price elastic and expenditure elastic, which indicates the luxurious nature of these products in the U .S. Also in 2012, C hidmi, Hanson, and Nguyen used retail scanner data to calculate the substitutability and complementarity between seafood products by regressing expenditures on weekly earnings and by regressing price on weekly earnings, CPI, PPI, gas prices, and the three month commercial paper rate. It was found that dem and for catfish, crawfish, clams, and salmon products is elastic, suggesting th at consumers are more sensitive to price changes for these products and that demand for shrimp and tilapia (which tend to be mostly imported ) is price inelastic . Although retail level scanner data provides information about the type of product that is pur chased, it is limited in that the demographic characteristics of the product purchaser remain unknown.
19 Household level Scanner Data and F ood Products The groundwork for using scanner data to model seafood demand at the household level was initiated by Chen g and Capps in 1988; these authors used at home scanner data to explain the variation of seafood expenditures with respect to a suite of demographic variables including income, occupation, age, household size, race , and region. Emphasis for the analysis wa s placed on expenditure income (Engel) relationships. It was found that household size and own price elasticity are the most important factors in de termining seafood expenditures. Additionally, household level scanner data has also been used to explain the demographic characteristics of seafood consumers by dividing people into separate groups based on race, income, and region and testing for statistically significant differences in seafood preferences amongst the groups (Davis, Yen, and Lin 2007). High in come households were found to be less responsive to price changes (exhibiting a more inelastic demand). Household level scanner data has been used to make inferences on the market for organic food products (one of the fastest growing food sectors in the United States) as well. Zhang et al. in 2009 utilized household level scanner data to investigate organic price premiums paid for fresh tomatoes and apples (two of the top organic produce sellers in the U.S.). Age, annual household income, and the presen ce of children were found to statistically significantly affect the price premiums paid for organic tomatoes, and household size, annual household income, age, and region were found to statistically significantly affect the price premiums paid for organic apples. Additionally, Dett man and Dimitri in 2009 used A C Nielsen Homescan data to comprise a demographic profile of organic vegetable consumers. These researchers found that
20 household heads with a higher level of educational attainment were statisticall y significantly more likely to purchase organic vegetables than were less educated households, and also found that African American households and older households were statistically significantly less likely to purchase organic vegetables. Separability of Seafood P roducts from Other Meat Products Although scanner data can be used to make inferences on the consumer market for all types of food (fruits, vegetables, meat, etc.), i t can be argued that meat comprises a separate category when compared to all oth er types of food bought at grocery stores due to the fact that it is considered a staple of main course meals and is high in protein. Additionally, it can be argued further that within this meat/protein category of food, seafood products comprise their ow n entity when compared to other types of meat. Using weak separability testing, it was found that when fish and other meat products were disaggregated into fresh and processed products, fish can be modeled separate from other forms of meat products (Salva nes and DeVoretz 1997). Additionally, studies concerning demand for meat have left fish and seafood absent from the analysis such as Choi and Sosin in 1990. Choi and Sosin analyzed structural changes in the demand for meat products (seafood products were not included) and found that structural change occurred in the 1970s and provides measures of the time pattern of demand shifts. Also, Anders and Moeser in 2008 used h ousehold level scanner dat a to make inferences on demand for meat in Canada (th is inclu ded beef, poultry, pork , turkey, and lamb; however seafood was left out). The absenc e of seafood products from the aforementioned analyse s concerning meat points to the separability of seafood products for the analysis that will follow.
21 Choice Modeling of Food Purchases Smith et al. (2006) utilized the 2006 Nielsen Homescan panel to investigate the probability of a household belonging to a certain organic produce consumer group (devoted, casual, or nonuser). These researchers used an ordered logit model t o quantify the impacts of socio demographic factors on the probability of a household belonging to one of the three previously mentioned organic purchaser groups and found that ethnicity, region of the country, the presence of children, education of the ho usehold head, and age of the household head are all statistically significant predictors when modeling the probability of a household being a devoted organic purchaser, a casual organic purchaser, or non purchaser of organic produce. Roheim and Johnston ( 2005) used a rank ordered logit choice model to explain preferences for eco labeled fresh, over the counter seafood. The primary emphasis was the potential tradeoff between taste and the presence of an eco label, given that multiple fresh, non processed s eafood products are available. These researchers found that consumers are unwilling to choose a less favored species (i.e., to sacrifice taste) based solely on the presence of an eco label. Given the availability of demographic variables that are availab le with household level scanner data , linking consumer choices wit h demographics is possible. For example, Martinez et al. in 2007 found that race, education, region of the county, household size, age of household head , and household composition were all statistically significant in predicting the probability of a household purchas These researchers used a logistic regression, conducted their analysis at the household level , and rojection factors to properly weight each hous ehold in the sample.
22 Much like purchasing , a consumer must make a similar choice when it comes to purchasing and consum ing fresh, over the counter seafood versus purchasing seafood products in the freezer section of the grocery store and demographic factors (Nauman et al. 1995). Nauman et al. used a consumer survey to gather information regarding the consumer choice to purchase fresh hybrid striped ba ss, trout, and salmon. Through logistic regression analysis, it was found that knowledge that the product was farm raised along with the perception that seafood is a healthy alternative to other and statistically significant ly affect the consumer decision to purchase fresh seafo od. For the analyse s that follow , experience and perception manifest themselves in demographic traits to drive the choice between purchasing fresh seafood sold at counters and purchasing frozen seafood in the freezer section of the grocery store .
23 CHAPTER 3 THEORETICAL BACKGROUND In economics, i t is assumed that consumers make choices to maximize their from making choices and that utility is correlated with both observable and unobservable factors. For example, consider that an individual derives utility from eating seafood, which can be represented as: U S i = B S X i + S i (3 1) Where U s characteristics (such as age, education, income, number of children, city, etc.), B is a vector of associated par ameters (i.e., marginal effects of each variable on utility from seafood), and certain other individuals may derive utility from avoiding seafood: U NS i = B NS X i + NS i (3 2) Where here , U NS is the utility derived from avoiding seafood products, and is an their decision based on the option yielding the highest utility, as assumed by economic theory, th at (s)he will buy seafood if and only if U S i > U NS i , which can be expressed as if : B S X i + S i > B NS X i + NS i (3 3) O r, by rearranging to isolate the error terms on the left hand side, as : S i NS i > B NS X i B S X i (3 4) Finally, representing the unobserved tastes: i > B NS X i B S X i (3 5)
24 In other words, this latter expression (generically: i > BX i ) is modeling a consumer that chooses to buy seafood. If we consider this problem as a choice, then we can define a binary variable as follows: Y i = 1 if and only if i > BX i and Y i = 0 if and only if i < BX i . Considering that the dependent variable in question ( Y i ) is a choice that has a binary distribution (yes, the household purchased seafood; or no they di d not), and this distribution is bounded by zero and one, it wou ld be possible to use a linear probability model in order to predict the probability of that dependent variable occurring. Coupling this fact with utility maximization theory, the following e quation can be obtained: Pr(Y=1) = Pr(U S >U NS ) ( 3 6 ) Where Y is a binary dependent variable (the choice to purchase seafood , U S is the utility of purchasing seafood, and U NS is the utility of n ot purchasing seafood . Predicting the probability that Y occurs is the same as predicting the probability that the utility of purchasing seafood is greater than the utility of not purchasing seafood . Similar to Equation 3 6 , the following can be assumed: Pr(Y=0 ) = Pr(U NS >U S ) ( 3 7 ) Equation 3 7 says that predicting the probability that Y does not occur is the same as predicting the probability that the utility of not purchasing seafood is grea ter than the utility of purchasing seafood. Linear probability models can be estimated with the OLS method, however in using the OLS method to estimate a linear probability model, the error term, by definition, must be heteroskedastic. Heteroskedasicity m eans that the error terms in an econometric model are not constant and random , that is, the error term varies from
25 observation to observation in some sort of predictable fashion ( i.e. the error terms are dependent upon the observations in some way) . This problem can lead to biased standard error estimates of explanatory variables and in turn lead to invalid conclusions. To illustrate how using the OLS method in a linear probability model leads to heteroskedasicity, the following simple OLS linear probabil ity model will be examined: where Fresh ( 3 8 ) Where Y is a binary dependent variable equaling 1 if a consumer purchased fresh pa rameters, x is a vector of explanatory variables, and is an error term. By assumpt ion, the error term in E quation 3 3 must be heteroskedastic because Y can only take on two values (zero and one). Therefore, given x, can only take on two values as wel l. Using alg ebraic manipulation of E quation 3 8 : ( 3 9 ) W hen Y takes on the value of one, then ( 3 10 ) W hich given that E( of zero: ( 3 11 ) W hich given that E( )=0, occurs with probability 1 r term can only take on these two values in a linear probability model with a binary dependent variable. Therefore: Var( )=E( 2 )=(1 2 2 (1 ( 3 12 ) 2 ] ( 3 13 )
26 Var( )= (1 2 x 2 2 x 2 ] ( 3 14 ) Var( )= (1 ( 3 15 ) Var( )= E(y|x) Â· E((1 y)|x) ( 3 16 ) E quation 3 16 shows the variance of the error term is not constant, and is heteroskedastic. T he variance of the error term will be h igher when the values of x are such that the predicted probability of the outc ome is cl ose to 0 .5, and lowest when the predicted probability of the outcome is close to 0 or 1. Considering that it is necessary that the model does not exhibit heteroskedasicity, a different method from OLS must be employ ed. Additionally, some predictions usin g the OLS method will fall outside these bounds of zero and one (which are the only two possible outcomes of this dependent variable). To keep these predictions inside these zero and one bounds, a generalized linear model must be used, and therefore one m ust choose between a probit or a logit link function (i.e. the distribution of . The generalized linear model uses the maximum likelihood estimation approach to calculate parameter estimates, and the predicted probability of the dependent variable occurring is always greater than or equal to zero and less than or equal to one. T his is the type of estimation that is preferred for the econometric model specified in this analysis. The probit link function assumes that the error terms in the model take the form of the standard normal distribution, whereas the logit link function ass umes that the error terms in the model take the form of a logistic distribution. The response curves for logit and probit are very similar, and for a large number of observations, the results from each method will be very simil ar as well. Hahn and Soyer ( 2007 ) state that the
27 the choice of the link function is largel ( 1997 ) concludes his it seems to not mak e much 2001 ) , perhaps most succinctly , link is found elsewhere as well ( e.g. Long 1997, Davidson and MacKinnon 1993, Hardin and Hilbe 2001). Referring back to E q uation 3 8 , is the structural component, and is the random component. C ontinuing from this, the most appropriate link function (either logit or probit) must be specified. T he mathematically the same as the response distribution's canonical location parameter; so the link function that does equate them is known as the canonical link f unction. The advantage of using the canonical link fu nction exists . The canonical link for the binomial distribution is the logit link . Therefore, the analysis that follows will use the logit link function and a logistic regression model will be specified. Past literature that used H omescan data to model the likelihood of a binary dependent variable occurring utilized the logit method as well (Martinez et al. 2007 , Nauman et al. 1995 ). A nd the logit method is modeling the logged odds of a dependent varia ble occu rring; t his provides ease of interpretation since t he in dependent variable effects are interpretable as odds ratios . As further evidence, a study was performed to compare OLS and logistic regression in terms of their underlying assumptions as well as the r esults obtained on common data sets (Pullman and Leitner 2003) . In both methods, the dependent variable had a b inary distribution of whether a student dropped out of school or if they
28 attended a private college. These researchers found that the predicted values using OLS regression and using logistic regression were highly correlated, and they concluded that both models can be used to test relati onships with a binary criterion. H owever, the researchers specified that logistic regression is superior to OLS at predicting the probability of a binary dependent variable occurring, and should be the model of choice for that application because the l ogistic regression yielded more accurate predictions of dependent variable probabilities as measured by the average squared differences between the observed and predicted probabilities. Based on this the oretical analysis, and that the dependent variable in question for this analysis is a binary response variable, the maximum likelihood estimation approach will be uti l ized with a logit link function to facilitate the interpretation of the coefficient estimates.
29 CHAPTER 4 DATA Th e data used in this analysis were obtained from the Nielsen Corporation (formerly AC Nielsen ), a global marketing firm. One of their products ( the Homescan data) includes information on retail purchases made by panel members over time Homescan Panel. Each time that a participating househo ld purchases groceries (including seafood products ), they scan all of their purchases from that particular t rip to the grocery store and enter quantities and purchase price from the receipts in their home. These data collectively comprise a national panel dataset of purchases by representative households. The households in T he Nielsen Corporation database (i.e. Homescan data) are recruited to represent a stratified random sample, selected on geographic, as well as demographic targets. This method ensures that the sample matches the demographic profile of consumers according to the United States Census . T he prima ry sampling unit is the household and AC Nielsen specifies that there was no intentional clustering. All of the summary statistics and all empirical analysis in this project utilize the household level projection factor s , which are sample weights calculate d for e ach household in the panel to reflect the demographic distribution o f the US population . The H omescan data used in this project encompasses seven AC Nielsen seafood modules: 2683 (un breaded frozen fish), 2682 (un breaded frozen shrimp), 2643 (bread ed frozen shrimp), 2607 (breaded frozen fis h), 2679 (breaded frozen shellfish), 2686 (un breaded frozen shellfish), and 0750 (RFCD fish shellfish; which is the fresh seafood sold at counters). Module 0750 is one focus of this project. This particular m odule represents all seafood products sold over the counter at grocery
30 stores; however the limitation of this module is that the specie s of the product is not k nown; It is only known that it Therefore, spec ie s information is only available for the frozen seafood product purchases that are not sold over the counter . The data were obtained for five major market areas included in the AC Nielsen Homescan panel: Miami, Chicago, Houston, Memphis, and New Orleans M obile, and over a three year period beginning on December 30, 2007 and ending on December 25, 2010 . The data consist of weekly purchases by UPC and household and corresponding household information. E ach participating household is represented by a Panelist ID number and several demographic characteristics at the household level such as income , presence of children, household siz e, household composition , and market area are provided . In addition, detailed information is available on each the head of each ho usehold including age, education, race, occupation and employment status. The raw data were provided in s eparat e sets of data: one data set had panelist IDs associated with purchase and product informat ion (by module), and one data set had p anelist IDs ass ociated with the demographic information . Therefore, the first step was to merge the two data sets and have a single Panelist ID associated with each product purchased , as well as associated with the demograph ic information of the panelist. The merge revea led that some seafood purchases were not associated with a valid p anelist ID, and that some p included in the data set on the seafood purchases. In all, there are 2,557 different
31 households in the data set that have demographic information associated with panelist IDs, and 1,961 households out of the 2,557 households with demographic information purchased seafood at least once over the three year duration that the data encompass es . The remainder of thi s chapter includes detailed discussions of the data on the following four topics: (1) statistics on fresh seafood expenditures by panelists, (2) a comparison of seafood eating households to those that did not purchase seafood, (3) an examination of past se afood purchases by species and product form, and (4) a comparison of panelists by market area and compared to the most recent U.S. Census. This information is intended to help identify factors that are hypothesized to be correlated with seafood consumption and, of those that have purchased seafood, those households that buy fresh seafood. Fresh Seafood Expenditures The first step in analyzing purchases of fresh, over the counter seafood products was to investigate the expenditures on fresh, over the counter seafood as a proportion of t he expenditures on seafood as a whole ( Table 4 1) . All monetary f igures reported in Table 4 1 and from this point forward are adjusted with the seasonally adjusted consumer price index and expressed in terms , using Dece mber 2010 dollars . As one can in fer, the share of fresh product expenditures as a fraction of total seafood expenditures is slight ly increasing over the three years that the data encompass es . In 2008, fresh, over the counter seafood products represent ed 20 .1 % of total expenditures on seafood by panelists in the data set. This figure increases to 25 .4% in 2009, and then to 26 .5 % in 2010. This trend of more consumer dollars being allocated toward fresh seafood products when compared to other seafood produ cts has many
32 implications for commercial fishers, seafood distributors, and grocery stores throughout the supply chain. However, considering that this trend is only slight at best and was consistent cross all market areas, all of the empirical analysis wa s taken at the household level (as done by Martinez in 2007 ) and the three years of purchase data were pooled . Demographic Comparison of Seafood Eaters and Non Seafood Eaters A demographic profile of the panel as whole, as well as a comparison between pane lists that purchased seafood and panelists that did not purchase seafood is displayed in Table 4 2 . All of the figures represented in Table 4 2 utilize AC Nielsen household projection factors. Some interesting result s were uncovered when the demograp hic summary statistics were examined. T here was a very high proportion of households in which the head was over 65 years old. 40 % of the seafood eating, and 50% of the non seafood eating households in the panel had a household head that was 65 years or older . In terms of occupation, nearly the entire sample (95%) was comprised of individuals. T he only other response for this is likely that this study will not be able to test for a correlation betw een occupation and probability of eating (buying) seafood . A lso , the panel shows a lack of racial diversity due to the high proportion of white/Caucasian households (73% of the panel, 72% of seafood eating households, and 75% of non seafood eating househo lds) when compared to other races. Annual household i ncome seems to be fairly evenly spread across the categories. This variable was coded as an ordinal variable and the categories were as follows: under $5,000, $5,000 $7,999, $8,000 $9,999, $10,000 $11,9 99, $12,000 $14,999, $15,000 $19,999, $20,000 $24,999, $25,000 $29,999, $30,000 $34,999, $35,000 $39,999, $40,000 $44,999,
33 $45,000 $49,999, $50,000 $59,999, $60,000 $69,999, $70,000 $99,999, and $100,000 & over . T he categories for in come were truncated ( T able 4 2 ) to facilitate comparison , however, to provide a continuous representation of this income variable, the midpoint of the range of each income category was calculated ( except for the highest income , which was simply assumed to equal the lower bound of the category). The mean income midpoint for the entire panel is $40, 96 0 annu ally, but the mean income midpoint for seafood eating households is $42, 15 0 versus just $36, 82 0 for non seafood eating households . Finally, the income distribution of the panel and the corresponding midpoint for each income category is presented in Table 4 3. Married households represent 37% of the panel like for income, seafood eating households in the panel have a higher share of married households compared to non seafood eati ng households (40% versus 25%). Delving more into household composition, a larger share of seafood eating households have children in the household (22%) than do the non seafood eating households (17%). There is also a very hig h rate of educational attai nment amongst the panel, with 94% of the household heads completing high school, 55% completing some form of college, and 24% completing college. Additionally, w hen examining household size, it was discovered that seafood eating households had a higher av erage household size (2.34) than did non seafood eating households (2.04). Because of these aforementioned findings , especially the noticeable oddities when examining the high proportion of households in which the head is over 65 years old, the high propo rtion of white households, and the high proportion of married households; the demographic profile of the panelists in the
34 data set for each market area was compared to the United States Census figures from 2010 in the final section of this chapter. Past Pu rchase Behavior and Species/Product Form Preference s Along with demographics of the household, it is hypothesized that the choice to purchase fresh, over the counter seafood is also driven by past purchase behavior as well as species/product form preferenc es of the consumer. The summary statistics for these va riables are displayed in Table 4 4 . An important thing to note about this table is that it specifically represents households in the panel that purchased seafood (the 596 households that did not purc hase seafood are not represented in this table because they do not have purchases associated with them). The variable response variable and represents the households that made at least one fresh, over the counter seafood purchase over the three year period. The mean of this variable is 0.33, which indicates that one third of all seafood purchasing households purchased fresh, over the counter seafood products. The average duration between purchases is 11.27 w eeks, and the average numbe r of seafood items purchased made by a household over the three year span is 13.1 8. The average weekly expenditure (mean of $14.66) for a household was calculated by taking the average of all of the weekly seafood expenditures of that household, but only in weeks in which seafood was purchased (to zero). Whether a household used coupons or not was examined as well. 20% of the seafood consuming households utilized a coupo n for their seafood purchases at least once over the three year period. The next set of variables reflect the share of a frozen seafood dedicated to the following species: shrimp,
35 tilapia, salmon, catfish, and cod. Each of thes maximum of 1.00, which indicates that the re are households that dedicate their entire frozen seafood budget toward one parti cular species. T hese five species were chosen is because they represent the species that househo lds in the data set spent the greatest dollar amount (on average) of thei r seafood budget on ( Table 4 5 ). When considering all types of seafood (fresh and frozen), the average household in the data set allocates 31.83% of their seafood budget (in dollars) to frozen shrimp, 8.78% of their seafood budget to frozen tilapia, 7.89% of their seafood budget to frozen salmon, 1.76% to frozen catfish, 1.70% to frozen cod, and 18.37% to fresh over the counter seafood products. The remaining 29.68% of the average ho allocated to other specie s of frozen seafood ( Figure 4 1). process form. These dedicated to t he following: frozen filleted purchases, breaded process form purchases, and generic (store brand purchases) . These figures were evaluated to gain more insight to household preferences for certain process forms and as well as household preferences for sto re brand seafood products (which tend to be more affordable than branded products). These aforementioned variables all had maximums equal to 1 (100%) as well, indicating that some households dedicate their entire seafood budget to a particular process for m of seafood. Comparison of Census with Panelists by Market Area The 2010 National Census was utili zed for this comparison. All Census figures were analyzed at the city level and were obtained from the Bureau of Labor Statistics . The H omescan panelists i n each city we r e compared their corresponding Census
36 figures for that city ( Tables 4 6 through 4 10 ) and were weighted with each corresponding AC Nie lsen projection factor . The process for this Census comparison operated as follows: first, the Census figures for each market area were collected and comp ared to th e H omescan panelists in that particular market area. From there, the Census figures were compared to seafood eaters from the matching market area, and subsequently compared to consumers of fresh, over the counter seafood products from the matching market area. Some consistent patterns exist across all five of the market areas. For one, the proportion of household heads in the panel that are 65 years and older is much greater than as re ported by the Census . However when compa ring seafood eaters in each market area to all of the panel ists from its matching market area , the proportion of househol d heads over 65 years of age is less for seafood eaters , and even less for purchasers of fresh , over the counter seafood products. Even though the proportion of household heads over 65 years old is less for consumers of fresh seafood than it is fo r the entire panel, the figure is still much greater than the Census reported figures in all five mark et areas . Addit ionally, all of the market areas except for New Orleans Mobile exhibit a lower proportion of households with children when compared to the Census. For seafood eaters, the proportion of households with ch ildren tends to be larg er for each ma rket when compared to the all of the panel ists in the matching market area. Furthermore, except for New Orleans Mobile, the proportion of house holds with children is even larg er for households that consume fresh, over the counter seafood products when com pared to seafood eaters as a whole from the matching market area, and also when compared to the corresponding Census figures for that particular city . Continuing
37 with household composition, married households tend to be more prevalent amongst the panel wh en compared to the corresponding Census figures (except for Houston) . The proportion of married households is larg er when examining seafood consuming households, and even larg er when examining households that consume fresh, over the counter seafood produc ts in all five market areas over the time horizon examined . For education, there i s a higher level of college completion reflected by the 2010 Census when compared to the panel . For consumers of fresh, over the counter seafood, the proportion of household s in the panel that completed college is even larger than the figures that represent the entire panel (except for Houston) , perhaps indicating that households who consume fresh, over the counter sea food tend to be more educated. H ousehold sizes in the pane l as a whole are smaller , on average, than as reported by the Census . However , the mean household size for seaf ood consuming households is larg er than the entire panel in e ach market area, a nd the me a n household size is even larg er than that when examinin g consumers of fresh, over the counter seafood in each market ar ea as well. Along with the finding that the mean household sizes of fresh seafood consuming households wer e larg er than household siz es in the panel as a whole, the mean household sizes were also larg er than the Census reported figures for every market area except for Miami . This could imply a positive relationship between household size and the likelihood of purchasing fresh, over the counter seafood products. Pe rhaps most striking is the la ck of racial diversity amongst the panel when compared to the Census figures. There is a much lower proportion of minorities in each
38 of the market areas in the panel than as reported by the corresponding 2010 US Census figures for each of the cities. How ever, there is no t much of a discernable trend for racial composition as seafood consuming households and fresh seafood consuming households are compared to the panel as a whole. As it pertains to median annual household income, some of the market areas r eported a higher median annual income in the Census when compared to panelists in the market area (Memphis, Chicago, Houston, New Orleans Mobile) and Miami panelists were discovered to have a higher median annual household income than as reported by the 20 10 Census. Additionally, fresh seafood consuming households in Chicago, Houston, Memphis, and Miami had a higher median annual household income when compared to all of the panelists in each of their corresponding market area s . Summarizing, although AC Nie lsen projection factors were used in this census comparison to properly weight each household, households in the panel still exhibit some key differences when compared to the population, especially as it concerns age, race, and household composition. T he AC Nielsen household projection factors will be used in the models as well to properly weight each household and help correct for an y disparities between the sample and the population . When using data from a sample , the sum of weights in a particular subg roup of the sample is used to estimate the population count for that particular subgroup. Each sample household is to represent other households in the entire population, and to indicate how many households are represented , a weight is used . In the case of the H omescan data, the weights used are the AC Nielsen projection factors .
39 Additionally, based on the above results obtained through the market by market census comparison, it can be inferred that these market areas are, for the most part, exhi biting differences from the population in a very similar fashion. Therefore, rather than specifying a separate model for each market area, each of the market areas in the data set will be all be included in the same model, while using the Chicago market a rea as the base market. Chicago is chosen as the base because it is the most represented market area in the data and also allows for direct interpretation of each of the four southeast US markets.
40 Table 4 1. Fresh seafood expenditures as a percent of t otal seafood expenditures by all panelists in all five market areas , 2008 2010. 2008 2009 2010 Total Fresh Seafood Expenditures $14,400 $18,828 $23,733 Total Seafood Expenditures $71,798 $74,084 $89,653 Percent of Total 20.06% 25.41% 26.47% Note : Exp ressed in real dollars (base=December 2010)
41 Table 4 2 . Demographic compariso n of seafood consuming households and non seafood consuming households . Variable All households (n=2,557) Sea food Eaters (n=1,961) Non seafood eaters (n=596) Edu cation of Household H ead (%) : High school 0.9 4 0 .94 0.96 College 0.24 0.23 0.27 Some college 0.55 0.54 0.58 Graduate School 0.07 0.07 0.08 Household Composition (%) : Married 0.37 0.40 0.25 Female head only 0.37 0.36 0.39 Male head only 0.26 0.24 0.36 Presence of Children (%) 0.21 0.22 0.17 Annual Household I ncome (%): $0 $29,999 0.35 0.33 0.42 $30,000 $59,999 0.37 0.38 0.33 $60,000 $99,999 0.20 0.20 0.19 $100,000+ 0.08 0.09 0.06 Income (1,000) 40.96 42.15 36.81 Race/Ethnicity (%): White 0.73 0.72 0.75 Black/African American 0.21 0.22 0.18 Asian 0.01 0.01 0.01 Hispanic 0.13 0.14 0.10 "Other" race 0.06 0.05 0.07 Age (% that is over 65) 0.42 0.40 0.50 Household size (numbe r) 2.27 2.34 2.04 Occupation (% retired) 0.95 0.95 0.96 Market Area (%): Chicago 0.29 0.29 0.33 Miami 0.27 0.29 0.19 Houston 0.18 0.16 0.24 Memphis 0.09 0.09 0.09 New Orleans Mobile 0.17 0.17 0.15 N ote: All figures representing panelists utilize AC Nielsen household projection factors.
42 Table 4 3. Distribution of annual household income amongst the panel. Income Category Percent of Total Midpoint (in thousands) Under $5,00 0 1.84% 2.5 $5,000 $7,999 1.68% 6.5 $8,000 $9,999 1.84% 8.5 $10,000 $11,999 1.88% 10.5 $12,000 $14,999 4.38% 13 $15,000 $19,999 7.39% 17.5 $20,000 $24,999 8.02% 22.5 $25,000 $29,999 8.10% 27.5 $30,000 $34,999 8.10% 32.5 $35,000 $39,999 6.84% 37.5 $40,000 $44,999 5.71% 42.5 $45,000 $49,999 6.26% 47.5 $50,000 $59,999 9.66% 55 $60,000 $69,999 7.08% 65 $70,000 $99,999 12.83% 85 $100,000 & Over 8.41% 100 N ote: All figures representing panelists utilize AC Nielsen household projection factors.
43 Tabl e 4 4. Summary statistics of past purchase behavior and species/product form preference variables (n = 1,961 households) . Variable Mean Standard Deviation Minimum Maximum FRESH (1 if panelist purchased fresh, over the counter seafoo d ; 0 otherwise ) 0.33 0.47 0 1 Average duration between purchases (weeks) 11.27 13.06 0 146 Number of seafood items purchase d 13.18 17.89 1 205 Average weekly seafood expenditure ($) $14.66 $29.79 $0.99 $1,044.61 Coupon use (1 if panelist used a coupon at least once, 0 otherwise) 0. 2 0 0.40 0 1 Percent share of seafood expenditure dedicated to: Frozen fillet purchases 0.30 0.35 0.00 1.00 Breaded process form purchases 0.27 0.35 0.00 1.00 Generic/store brand purchases 0.32 0.36 0.00 1.00 Percent share of frozen seafood expenditure dedicated to: Shrimp 0.37 0.39 0.00 1.00 Salmon 0.09 0.23 0.00 1.00 Tilapia 0.10 0.23 0.00 1.00 Catfish 0.02 0.11 0.00 1.00 Cod 0.02 0.10 0.00 1.00
44 Table 4 5. Top ten avera ge annual expenditures by species by all households , 2008 2010. Specie s Average Yearly Expenditure RFCD fish shellfish $19,373.57 Shrimp $16,512.61 Salmon $4,784.28 Tilapia $4,223.61 Catfish $1,132.66 Cod $1,047.88 Scallop $943.42 Crawfish $874.92 Flounder $852.10 Orange roughy $673.92
45 Table 4 6. Census vs. H omescan comparison for Chicago market area. Variable 2010 Census Homescan Panelists (n=831) Seafood consuming panelists (n=644) Fresh seafood consuming panelists ( n= 229) Age (percent of adult population 65 years or over) 13.9% 43.2 % 41.7 % 23.8 % Median annual household i ncome $47,408 $35,000 $39,999 $40,000 $44,999 $45,000 $49 ,999 Educ ation (percent of adult population with a bachelor's degree) 33.6% 27.4% 27.9% 29.0% Household c omposi tion ( percent of married households) 32.7% 36.1 % 40.6 % 51.6 % Race (percent of adult population that is white) 45.0 % 71.9 % 71.0 % 66.3 % Household s ize (mean number of persons) 2.57 2. 25 2 .33 2.61 Percent of h ouseholds with c hildren 25.2% 20.4 % 20.6 % 24.8 % N ote: All figures representing panelists utilize AC Nielsen household projection factors.
46 Table 4 7 . Census vs. H omescan comparison for Miami market area . Variable 2010 Census Homescan Panelists (n=662) Seafood consuming panelists (n=532) Fresh sea food consuming panelists (n=173) Age (percent of adult population 65 years or over) 20.2% 45.4 % 43.5 % 27.4 % Median annual household i ncome $29,762 $30 ,000 $3 4 ,999 $ 30,000 $3 4,999 $35,000 $39 ,999 Education (percent of adult population with a bachelor's d egree ) 22.9% 25.5% 24.7% 36.2% Household composition (percent of married households) 31.3% 38.0 % 40.7 % 44.7 % Race (percent of adult population that is white) 72.6% 81.5 % 81.4 % 82.9 % Household s ize (mean number of persons) 2.6 2.16 2.24 2.56 Percent of h ouseholds with c hildren 22.7% 17.0 % 18.4 % 31.7 % N ote: All figures representing panelists utilize AC Nielsen household projection factors.
47 Table 4 8. Census vs. H omescan comparison for Houston market area . Variable 2010 Census Homescan Panelists (n =501) Seafood consuming panelists (n=356) Fresh seafood consuming panelists (n=89) Age (percent of adult population 65 years or over) 12.7% 41.9 % 39.1 % 25.3 % Median annual household i ncome $44,648 $30 ,000 $3 4 ,999 $ 30,000 $3 4,999 $35,000 $3 9,999 Educatio n (percent of adult population with a bachelor's degree ) 2 8.7% 22.9% 22.3% 21.2% Household composition (percent of married households) 39.0% 37.5 % 42.0 % 53.0 % Race (percent of adult population that is white) 50.5% 67.0 % 67.4 % 55.4 % Household s ize (mean number of persons) 2.69 2.52 2.63 3.18 Percent of households with c hildren 30.2% 24.4 % 24.5 % 36.9 % N ote: All figures representing panelists utilize AC Nielsen household projection factors.
48 Table 4 9 . Census vs. H omescan comparison for Memphis marke t area . Variable 2010 Census Homescan Panelists (n=239) Seafood consuming panelists (n=179) Fresh seafood consuming panelists (n=56) Age (percent of adult population 65 years or over) 14.6% 34.4 % 29.7 % 22.6 % Median annual household i ncome $36,817 $20,000 $2 4,999 $2 0,000 $24 ,999 $25,000 $2 9,999 Education (percent of adult population with a bachelor's degree ) 23.4% 10.1% 9.4% 13.0% Household composition (percent of married households) 29.7% 32.7 % 34.3 % 48.2 % Race (percent of adult population that is whit e) 29.4% 71.7 % 72.2 % 68.7 % Household s ize (mean number of persons) 2.59 2.16 2.26 2.62 Percent of households with c hildren 27.8% 21.3 % 27.0 % 36.7 % N ote: All figures representing panelists utilize AC Nielsen household projection factors.
49 Table 4 10 . Census vs. H omescan comparison for New Orleans Mobile market area . Variable 2010 Census (New Orleans) Homescan Panelists (n=324) Seafood consuming panelists (n=250) Fresh seafood consuming panelists (n=103) Age (percent of adult population 65 years or over) 14.9% 37.6 % 35.2 % 26.4 % Median annual h ousehold i ncome $36,691 $25,000 $29 ,999 $25,000 $29 ,999 $25,000 $2 9,999 Education (percent of adult population with a bachelor's degree ) 33.0% 24.0% 19.6% 24.5% Household composition (percent of married house holds) 27.5% 36.6 % 38.5 % 39.5 % Race (percent of adult population that is white) 33.0% 66.1 % 62.1 % 62.4 % Household s ize (mean number of persons) 2.29 2.28 2.29 2.34 Percent of households with c hildren 22.7% 22.8 % 23.7 % 22.3 % N ote: All figures representi ng panelists utilize AC Nielsen household projection factors.
50 Figure 4 1. Average household allocation of seafood expenditures for fresh seafood and frozen seafood by species (n=1,961) . 31.83% 8.78% 7.89% 1.76% 1.70% 29.68% 18.37% Frozen shrimp Frozen tilapia Frozen salmon Frozen catfish Frozen cod Frozen other species Fresh, over the counter
51 CHAPTER 5 MODEL SPECIFICATION This analysis is concerned wi th predicting the probability of a household purchasing seafood and then predicting the probability of a seafood consuming household purchasing fresh, over the counter seafood. Therefore, for both of these models, the dependent variable will take the form of a Bernoulli distribution (only two viable outcomes). When considering the probability of purchasing seafood the outcome the probability of a household purchasing fresh, over the counter seafood, the outcome of the dependent variable can es are of interest here, teo probabilistic stati stical classification model were specified. Before modeling the probability of a seafood purchasing household purchasing fresh, over the counter seafood, the probability of a household in the pa nel purchasing seafood will be analyzed. As stated previously, 1,961 out of the 2,557 households (76.69%) in the data set purchased seafood at least once over the three year period that the data encompass ed . The dependent variable for this first model (denoted as is a binary response variable coded as 1 if the household made at least one seafood purchase, and zero if the household did not purchase seafood . For this analysis, it is hypothesized that the probability of a household purchasing seafood is a func tion of household demographics: (5 1 ) Where D is a vector of demographic variables. Equation 5 1 can be re written as:
52 ( 5 2) Where x is a vector of demographic variables, is a vector of unobserved logistic regression parameters, is a constant, and is an error term. The probability of a household purchasing seafoo d s ) can be re written, and then simplified and linear ized to form a viable logistic regression equation in the following fashion : ( 5 3) Then, by taking th e reciprocal of each side of E quation 5 3 : ( 5 4) And then s ubtra cting one from each side of E quation 5 4 yields: ( 5 5) When the rec iprocal of each side is taken again, it yields: ( 5 6) Finally, by taking the natur al logarithm of each side of E quation 5 6 , the logit transformation is complete: Logged odds of SEAFOOD = ln = Logit(SEAFOOD ) = ( 5 7) Equation 5 7 is now linearized on the right side of the equation and represents the logistic regression model that will be used to predict the probability of household purchas ing seafood . The results from this model are illustrated in T able 6 1 . Similarly, a maximum likelihood mode l with for a binary dependent variable ad assuming the logit link function was be constructed to model the probability of a seafood eating household purchasing fresh, over the counter seafood products. The analysis is again conducted at the household level, however since this model is only concerned
53 with seafood purchasing households, the number of observations used is 1,961. The reason why only sea food purchasing households are investigated when predicting the probability of a household purchasing fresh, over the counter seafood is because this allows for purchase behavior and specie s /product for m pre f erence s to enter the model. Since 596 household s in the data set did not purchase seafood, they would be represented as missing values in the model because there is no purchase information associated with the household . Therefore, these 596 households that did not purchase seafood at all over the thre e years are excluded from the second model. It is hypothesized that the probability of purchasing fresh, over the counter seafood for a seafood purchasing household is a function of demographics, past purchase behavior, and species/product form preferences . This can be represented by the following general equation: Pr(FRESH = 1) = f(D,B,S) ( 5 8) least one fresh, over the counter seafoo d purchase over the three year period , and 0 if not , D is a vector of demographic variables as before , B is a vector of past purchase behavior variables, and S is a vector of species/product form preference variables. Equation 5 8 can be re written as: Pr (FRESH = 1) = = + 1 D + 2 B + 3 S + ( 5 9) is an error term, and 1 , 2 , and 3 are vectors of unobserved logistic regression parameters corresponding to demographics, past purchase behavior, and specie s /pr oduct form preferences respectively. Using the same approach that is used in Equations 5 3 through 5 6 , the following logit equation was obtained:
54 Logged odds of FRESH = ln = Logit(FRESH) = (5 1 0) Equ ation 5 1 0 is now linearized on the right side of the equation and represents the logistic regression model that will be used to predict the probability of a seafood purchasing household purchasing fresh, over the counter seafood products. The results fro m this mod el are illustrated in T able 6 2 .
55 CHAPTER 6 EMPIRICAL RESULTS The LOGIT procedure in SAS was first used to estimate model parameters, and AC Nielsen projection factors were used to properly weight each household. However, using the WEIGHT state ment in a LOGIT procedure in SAS can distort standard error estimates, which can lead to invalid conclusions. To alleviate this problem, the SURVEYLOGIT procedure in SAS was used, because using the WEIGHT statement in this procedure does not distort stand ard error estimates. The following results of this analysis reflect the results of the SURVEYLOGIT procedure. Many of the demographic variables had to be coded as dummy variables to rem ain quantitative in nature. Therefore, base categories for these dumm y variables had to be established. For household composition, the base category is m arried (co headed) households. F or race, the base category i s white households. F or market area, the base category is Chicago households (Chicago is the most represented m arket area in the panel , and is the only market area outside the southeastern United States ). For the presence of children, the base category is households that do not have any childre n under eighteen years of age. F or age, the base category is households in which the head is under 65 years of age. F or education, the base category is households in which the he ad has not completed college. For h ousehold income , the midpoint of each income category was ca lculated and used in the models; this follows p reviou s literature (Martinez 2007) that utilized H omescan data and a logit model (Martinez 2007) . This income coding technique also provides f or ease of interpretation since the income a one thousand dollar increase in annual hou sehold income leads to some incremental probability increase/decrease in the probability of a
56 Finally, h ousehold size was coded as a continuous variable . Model 1 : Predicting the Probability of Household Purchasing Seafood Eq uation 5 7 was used to predict the probability of a household in the panel purchasing seafood (model 1) . Households headed by just a female or just a male had statistically significant negative parameter estimates; this implies that married (co he aded) ho useholds a re more likely to purchase seafood when compared to single headed homes . Annual household income had a statistically significant positive parameter estimate, implying that as income increases, the likelihood of purchasing seafood increase s as we ll. Age and education had statistically significant negative parameter estimates, which implies that households with older heads a re less likely to purchase seafood and that households with more educated heads a re less likely to purchase seafood. As it p ertains to market area, Miami households a re statistically significant ly more likely to purchase seafood than their Chicago household counterparts. Model 2 : Predicting the Probability of a Household Purchasing Fresh Seafood Equation 5 10 was used to predic t the probability of a seafood purchasing household purchasing fresh, over the counter seafood products (model 2) . Households headed by just a female are statistically significant ly less likely to purchase fresh, over the counter seafood than married (co headed) households. Additionally, the parameter estimate for age is statistically significant and negative (similar to model 1), implying that older household s are less likely to purchase fresh, over the counter seafood products. When examining m arket ar ea, households in Miami and New Orleans Mobile a re statistically significant ly less likely to purchase fresh, over the counter seafood than were panelists in Chicago. The behavioral variables and species/product form
57 preference variables that were enter ed into the model for model 2 a re all statistically significant except for percent share of frozen seafood expenditure dedicated to cod and dedicated to catfish. As number of seafood items purchased increase s , so does the likelihood of a house hold purchasing fres h, over the counter seafood. Additionally, if a household used a coupon at least once on a seafood purchase, they are statistically significantly more likely to purchase fresh, over the cou nter seafood products as well. Increases in al l of the following variables lead to a decrease in the likelihood of a household purchasing fresh, over the counter seafood products: average weekly seafood expenditure, average dura tion between purchase s , the percent share of seafood expenditure dedicated to generic/store brand p urchases, fro zen fillet purchases, and breaded process form purchases, and the percent share of frozen seafood expenditure dedicated to frozen shrimp, tilapia, and salmon . To further examine the relationship between some of the sta tistically significant purchase behavior variables and the probability of a household purchasing fresh, over the counter seafood pro ducts, effect plots were used. These effect plots were generated in SAS to explain the relationship between an independent variable and its effect on the predicted probability of a household purchasing fr esh, over the counter seafood while holding all other independent variables in the models constant at their mean values. When examining the effect plot for average duration between purchases (Figure 6 1), it is evident that as the average duration between purchases likelihood of purchasing fresh, over the counter seafood decrease d . Therefore, those households who purchase seafood mor e frequently are more likely to purchase fresh, over the counter seafood products. When
58 breaded process form purchases, 2) as a household increased the share of th eir seafood budget allocated to breaded seafood, they wer e less likely to make the choice to go to the fresh counter at the grocery store and purchase fresh seafood. (Figu r e 6 3), as a household purchases more seafood items in general, they were more likely to go to the fresh counter at the grocery store and purchase fresh seafood (Figur e 6 1) in some respects because both reflect the fact that household s that purchase seafood more often are more likely to purchase fresh, over the counter seafood products. Odds Ratios The logit parameter estimates in the previous discussion have little di rect meaning, however they are used to calculate the incremental probability increase/decrease that an independent variable leads to as it pertains to the probability of a dependent variable occurring. Direct interpretation of the independent variables ca n be obtained through the calculation of odds ratios based on the parameter estimates of the independent variables in question. For example, in T able 6 2 , the odds seafood items purchased . This can be interpreted a seafood purchase, they are 1.041 times more likely to purchase fresh, over the counter seafood products (their probability of purchasing fresh , over the counter seafood increases by 4.1 %) . Graphi cal representations of these odds ratio esti m a tes are depicted in F igure 6 4 and in F igure 6 5 (corresponding to model 1 and model 2, respectively).
59 Referring to F igure 6 4 ha s the largest probability effect (in terms of absolute valu e). This show s that panelists in Miami are 73 . 3 % more likely to purchase seafood than were panelists in Chicago . The only other positive and statistically significant household income increase d by $ 1,000, a household was 0. 6 % more likely to purchase seafood. When household composition is examined, female only headed households an d male headed only households were statistically significant ly less likely to purchase he e only headed households we re 30.1 % less likely to purchase seafood than married households and mal e only headed households we re 50.0 % less likely to purchase seafood than were married households. When considering age, households i n which the he ad was 65 years or older we re 33 . 5 % less likely to purchase seafood than were household s in indepen dent variable was surprising; c onsidering that income had a statistic ally significant positive relationship with the probability of a household purchasing seafood, and that income and education tend to be positively correlated, it came as a surprise that households in which the he ad had completed college we re 35 . 5 % less lik ely to purchase seafood than were households in which the head did not complete college. Referring to F igure 6 5 , that wa s allocated toward shrimp, tilapia, and s almon all had very large probability effects (in terms of absolute value). This means that as a household increases the share of their frozen seafood budget on these particular specie s of seafood (at the margin), they we re statistically significant ly less
60 likely to purchase fresh, over the counter seafood products. Additiona lly, as a household spent more on breaded seafood products and frozen filleted seafood products (at the margin), they we re statistically significant ly less likely to purchase fresh ove r the counter seafood. This implies that as households spen t more on processed seafood , they we re less likely to make the choice to go to the counter at the grocery store and purchase fresh seafood. When examining the purchase behavior of the households , as a household spent one extra dollar, on average, on seafood in a week in which the y purchase seafood, they we re 3.2 % l e ss likely to purchase fresh, over the counter seafood. Therefore, as a household spends more mon ey on seafood in general, they we re s tatistically significant ly less likely to make the choice to go to the counter and buy fresh seafood , which is surprising considering that fresh seafood products have been considered luxury goods . Adding to this, households that purchase seafood more freq uently we re statistically significant ly more likely to purchase fresh, purchases inc reased by one week, they we re 4.2 % less likely to purchase fresh, over the counter seaf seafood items purchased incr ease d at the margin, they we re 4.1 % more likely to purchase fresh, over the coun ter seafood. Due to these findings , marketers and distributors of fresh seafood should target areas in which con sumers tend to buy more seafood in general and buy seafood more frequently when compared to other possible markets for fresh seafood products. The use of coupons plays a large role in predicting the probability of a household purchasing fresh, over the cou nter seafood. Households that used coupons on seafood (fresh or frozen) we re 118.3 % more likely to purchase fresh, over the counter seafood.
61 Therefore, the use of coupons is a very strong predictor in trying to model the likelihood of a household purchas ing fresh, over the counter seafood products. Additionally, as a household increase d their share of generic/store brand purchases by 1%, they we re 86 .6 % less likely to purchase fresh, over the counter seafood. When examining the demog raphic effects in Mod el 2, four variables are statistically significant and negative . Households in which the head is 65 years or older are 60.3 % less likely to purchase fresh, over the counter seafood than were households in which the head is under 65 years of age. Also, fe male only headed households a re 47.9 % less likely to purchase fresh, over the counter seafood than were married households. And fin ally, Miami households a re 4 3.8 % less likely and New O rleans Mobile households a re 45.9 % less likely to purchase fresh, over the counter seafood than were households in Chicago.
62 Table 6 1 . Model 1: Predicting the probability of a household purchasing seafood (n = 2,557 households). Variable Mean Parameter Estimate Odds Ratio SEAFOOD (1 if yes) 0.77 Depende n t Variable Ed ucation ( 1 if completed college) 0.24 0.438** 0.645 Female head only 0.37 0.359* 0.699 Male head only 0.26 0.693*** 0.5 00 Income ( thousands) 40.96 0.006** 1.006 Age (1 if over 65) 0 .42 0.407 *** 0.665 Household size 2.27 0.079 1.082 Presence of children (1 if yes) 0.21 0.206 0.814 Black/African American (1 if yes) 0.21 0.226 1.253 Asian (1 if yes) 0.01 0.224 1.251 "Other" race (1 if yes) 0.06 0.488 0.614 Hispanic (1 if yes) 0.13 0.429 1.535 Miami (1 if yes) 0.27 0.550*** 1.733 Houston (1 if yes) 0.18 0.286 0.751 Memphis (1 if yes) 0.09 0.180 1.197 New Orleans Mobile (1 if yes) 0.17 0.291 1.338 Note: * significant at 10% level, ** significant at 5% level, *** significant at 1% level
63 Table 6 2 . Model 2: Predicting the prob ability of a household purchasing fresh, over the counter seafood (n = 1,961 households). Variable Mean Parameter Estimate Odds Ratio FRESH (1 if yes) 0.33 Dependent Variable Education ( 1 if completed college) 0.23 0.10 7 1.112 Female head only 0.36 0.651 ** 0.521 Male head only 0.24 0.225 0.799 Income ( thousands) 42.15 0.005 0.995 Age ( 1 if over 65) 0.40 0.924 *** 0.39 7 Household size 2.34 0.03 1 1.03 1 Presence of childre n (1 if yes) 0.22 0.280 1.323 Black/African American (1 if yes) 0.22 0.105 1.111 Asian (1 if yes) 0.01 0.760 0.468 "Other" race (1 if yes) 0.05 0.51 2 0.599 Hispanic (1 if yes) 0.14 0.04 1 0.960 Miami (1 if yes) 0.29 0.577 ** 0.56 2 Houston (1 if yes) 0.16 0.43 1 0.650 Memphis (1 if yes) 0.09 0.022 0.978 New Orleans Mobile (1 if yes) 0.17 0.61 5 ** 0.541 Average duration between purchases (wee ks) 11.27 0.043 *** 0.958 Average weekly seafood expenditure (dollars) 14.66 0.033 *** 0.96 8 Number of seafood items purchased 13.18 0.041 *** 1.04 1 Coupon use (1 if yes) 0. 2 0 0.781* ** 2.183 Percent share of seafood expenditure dedicate d to: Generic/store brand purchases 0.32 2.012 *** 0.134 Frozen fillet purchases 0.30 2.646 *** 0.071 Breaded process form purchases 0.27 4.657 *** 0.009 Percent share of frozen seafood expenditure dedicated to: Shrimp 0.37 1.526 *** 0.2 17 Salmon 0.09 2.252 *** 0.105 Tilapia 0.10 2.017 *** 0.133 Catfish 0.02 1.114 0.328 Cod 0.02 0.745 0.475 Note: *=significant at 10% level, **=significant at 5% level, ***=significant at 1% level
64 Figure 6 1 . Effect plot for average duration between purchases. Figure 6 2. Effect plot for percent share of seafood expenditure dedicated to breaded process form purchases .
65 Figure 6 3. Effect plot for number of seafood items purchased. Figure 6 4 . Change in probabilities of purchasing seafood based on statistically significant predictor variables in model 1. -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 Male head only (compared to married households) Education (completed college) Age (over 65 compared to under 65) Female head only (compared to married households) Income (in thousands) Miami (compared to Chicago households) Change in Probaility Variable
66 Figure 6 5. Change in probabilities of purchasing fresh, over the counter seafood based on statistically significant predictor variables in model 2. -1 -0.5 0 0.5 1 Share of breaded process form purchases Share of frozen fillet purchases Share of salmon purchases Share of tilapia purchases Share of generic/store brand purchases Share of shrimp purchases Age (over 65 compaed to under 65) Female head only (compared to maried households) New Orleans-Mobile (compared to Chicago households) Miami (compared to Chicago households) Average duration between purchases (weeks) Average weekly seafood expenditure (dollars) Number of seafood items purchased Coupon use (compared to households who didn't use a coupon) Change in probability Variable
67 CHAPTER 7 CONCLUSIONS Overview of Project The goal of this analysis was to correlate the probability of a household purc hasing seafood with household demogra phics, and from there, to correalte the probability of a seafood consuming household purchasin g fresh, over the counter seafood products with household demographics, household purchase behavior, and household species/product for preferences. To that end, weekly AC Neilson Homescan data covering a three year period was utilized to model these choic es based on utility maximization theory and accounting for dichotomous response variables. In addition to the modeling results, the U.S. Census data were compared with the characteristics of the Homescan panelists (both those that purchased seafood and tho se that did not) for each of five major U.S. market areas. Finally, results are compared and discussed in relation to results from previous studies. Discussion and Implications of Results Results from this analysis suggest that seafood consuming household s are demographically different from households that do not consume seafood. Households that were more l ikely to purchase seafood tend to be wealthier ( perhaps indicating the luxury nature of seafood products ) and have younger household heads . However, h ouseholds in which the head completed college were overall less likely to purchase seafood. This is perhaps a possible indication of a growing shift away from seafood consumption amongst more educated consumers due to negative perceptions of overseas seaf ood farming , concerns for the health of wild stocks, and environmental impacts of the 2010 Dee pwater Horizon Gulf Oil Spill.
68 Moreover, there are some key differences among seafood consuming households as it pertains to the consumption of fresh, over the co unter seafood products. As stated previously, households with older heads we re less likely to purchase fresh, over the counter seafood , as we re female only headed households when compared to married households . Households in Miami and New Orleans Mobile households we re less likely to purchase fresh, over the counter seafood as well (when compared to households in Chicago) . Coinciding with previous literature, households who purchase more seafood (higher number of seafood items purchased ) and more frequent ly (smaller avera ge duration between purchases) we re more likely to purchase fresh, over the counter seafood (Nauman et al. 1995). A dditionally, households that devote d more time to using coupons for seafood we re more lik ely to choose to buy fresh seafood from the counter. Also as expected, as a household devoted more of their seafood budget toward frozen shrimp, tilapia, or salmon, their likelihood of purchas ing fresh, o ver the counter seafood declined . Similarly, as a household purchased more frozen fi lleted, breaded, or generic/store branded seafood products, they were less likely to choose to go to the counter for fresh fish and seafood. Households with higher average weekly seafood expenditure values were also less likely to purchase fresh, over the counter seafood, which came as a surprise. These key demographic, behavioral, and preferential di fferences amongst households that did/did not purchase fresh, over the counter seafood allow for firms and marketers to better segment the m arket for fresh se afood and create more opportunities in relatively advantageous markets for their products. By analyzing the statistically significant factors related to the lik elihood o f purchasing fresh, over the counter seafood
69 and having a demographic profile of a mar ket in question, emerging and future demand can be investigated and calculated. Recommendations to Firms and Marketers Results from this analysis have several implication s for firms involved in the production of seafood, marketers of seafood products, and other vested interests throughout the supply chain of seafood production. Targeting areas in which the population has a high proportion of married households would be relatively ad vantageous for seafood sales, because model 1 show ed that that single heade d households are statistically significantly less likely to purchase seafood than are married households. Additionally, areas in which the population tends to be older in age would not be a wise target market for seafood producers and seafood marketers be cause model 1 suggests that households in which the head is 65 years or older are statistically significantly less likely to purchase seafood. Moreover, targeting high income areas is wise because as income increase, so too does the likelihood of purchasi ng seafood. One of the surprising findings from model 1 was that college educated households were less likely to purchase seafood, leading to the conclusion that more educated market areas would not be the wisest choice when trying to market/sell seafood product s effectively ; perhaps because they have heard more about the environmental concerns associated with fishing wild stocks in the ocean or culturing seafood near shore . Also as far as region is concerned , Miami households are statistically significan tly more likely to purchase seafood than are Chicago households. Model 2 ( the likelihood of a seafood consuming household purchasing fresh, ov er the counter seafood products) revealed t hat markets with a high proportion of older residents would not be an a dvantageous place for the sale/marketing of fresh, over the
70 counter seafood products. Addi tionally, Miami households are statistically significantly less likely than Chicago households to purchase fresh, over the counter seafood, and the same is true for households in New Orleans Mobile. Purchase behaviors and species/product form preferences explained a lot regarding the likelihood of a household purchasing fresh, over the counter seafood products. For example, c oupon using households were statistically significantly more likely to purchase fresh seafood when compared to households who do not utilize coupons; as such to increase purchases of fresh seafood, it would be wise to introduce coupons for seafood in general into a market. Also as expected, thos e who purchase more seafood in general and those who purchase seafood more frequently are more likely to purchase fresh seafood products; therefore if consumers in a particular area exhibit characteristics of frequent and numerous seafood purchases, these consumer s would be a wise target for producers of fresh seafood products. Market research can also be completed to identify areas with consumers who purchase a large share of seafood products that have undergone some form processing (filleting, breading, etc.); t hese consumers are less likely to purchase fresh, over the counter seafood products and it would be more effective for producers and marketers of fresh seafood products to avoid markets in which these type of consumers are prevalent. Further Resear ch The analysis in this project can be ex tended in several ways. First , there is the possibility of a n alternative estimation procedure, such as a two sta ge estimation approach (i.e. the double hurdle model). The motivation underlying the hurdle formulat ions is that a binomial probability model governs the binary ou tcome of whether a count variable has a zero or a positive realization. If the realization is positive, the
71 a truncated at zero count data model (Mullahy 1986). As a result, there could be some efficiency gains by modeling both household decisions simultaneously. The characteristics of the AC Nielsen Homescan panel also should be considered in more detail. In terms of the time horizon, the data encompass a time period that includes purchases from before and after the 2010 Gulf Deepwater Horizon Oil Spill. This has several implications such as the possible decline in fresh, over the counter seafood products (es pecially in the southeastern United States) after the oil spill and to the effect of negative perceptions concerning the dangers of consuming households that may have spent sho rter or longer times in the panel could also affect the results that relate to many of the variables (since AC Nielson aims for a balanced panel, this would only be an issue with households at the beginning and end of the time series covered). In terms of representativeness, t here are some stark differences between the panel and the Census profiles of the market areas in the data set (Tables 4 6 through 4 10) . The panel was overwhelmingly retired and older, leading to the conclusion that those who volunte er to participate in the Homescan panel tend to have more extra time on their hands to complete the task of scanning every purchase when they get home from the grocery store. Additionally, it seems as if minorities are harder to reach for participation in the Homescan panel because their share is less in the sample than as reported by the Census . That said, t he projection factors (post stratification weights) associated with each household in the panel are calculated precisely to correct for any over/unde r sampling of a particular population subgroup ; as
72 such, the characteristics of the sample are adequate for modeling the choices examined in this paper. Finally, since the data contain information on price and quantity, demand functions can be estimated us ing the Almost Ideal Demand System model. Using this method, the price fluctuations of seafood products and subsequent consumer responses can be used to calculate own price, cross p rice, and income elasticities. The data set is very rich in information a nd analysis can go in several directions. Investigation into the nature of the market for fresh, over the counter seafood products is an exciting and under researched topic, and there is more work that can be done out side the scope of this analysis, parti cularly within each market and for particualr species.
73 LIST OF REFERENCES Value 12th Congress of the European Association of Agric ultural Economists . August 2008. Ghent, Belgium. American Journal of A gricultural Economics 70(3): 533 542. Chidmi, B., T. Hanson, and G. Products at the US National Retail level Marine Resource Economics 27(4): 359 70. American Journal of Agricultural Economics 72(1): 227 236. Agricultural and Resource Economics Review 23(2): 125 39. Daly J . , M. Lieberman , J. Goldfine , J. Shou , F. Weintraub , E. Rosato , and P. Lavin. 3 Fatty A cids in Patients after Operation: Immunologic, Metabolic, and Clinical O utcome. Department of Surgery, University of Pennsylvania School of Medicine. Philadelphia 19104. Surgery 112(1):56 67. Davidson, R. and J. G. MacKinnon . 1993 . Estimation and Inference in Econometrics . New York: Oxford. American Agricultural Economics Associati on 2007 Annual Meeting. August 2007. Portland, Oregon. Dettman, Rachel. 2009 Journal of Food Products Marketing 16(1) 79 91. ng Fruit and Vegetable Consumption: Are of Agriculture Economic Research Service. Economic Research Report Number 96. f Nielsen Homescan Economic Research Report Number 69. Nielsen Perishables Group. January 2014 publication.
74 Gill, J. 2001 . Generali zed Linear Models: A Unified Approach . Thousand Oaks, CA: Sage . p. 33. Greene, W. H. 1997 . Econometric Analysis (3rd edition ). Upper Saddle River, NJ: Prentice Hall. p. 875. Hahn, E. and R. Soyer. 2007. Probit and Logit Models: Differences in the Mul tivariate Realm. Accessed 11/18/14. Hardin, J. and J. Hilbe . 2001 . Generalized Linear Models and Extensions . College Station, TX: Stata Press. Homescan Data and Complex Survey Design Techniques to Analyze Convenience F American Agricul tural Economics Association 2005 Annual Meeting. July 2005. Providence, Rhode Island. Jang, D., J. Garnett, F. Piortroski, and W. Owens. 1999. Empirical Investigation of Variance Estimators Based o n Various Design Option s When a Ra king Method i s I mplemented Mathematica Policy Research Inc. December 1999 Mathematica Policy Research Report : 796 799. Retail Promotion Strategies: He donic Analysis of Weekly Scanner Data in the Southern Agricultural Economics Association 2012 Annual Meeting. February 2012. Birmingham, Alabama. Long, J. S. 1997 . Regression Models for Categorical and Limited Dependent Variables. Thousand Oaks, CA: Sage. Affecting Deand for Branded Beef: Applying a Logit Model to 2004 Neilson Home Journal of Business and Economics Research. 5(11): 95 102. Mullahy, J. 1 986. Specification and testing of some modified count data models. Journal of Econometrics 3: 341 365. Homescan ERS Data Quality Study Design Contract 53 3K06 06 1301. February 2007. Nahab, F., A. Le, S. Judd, M.R. Frankel, J. Ard, P.K. Newby, and V.J. Howard. 2011. Neurology 76(2): 154 158.
75 Fre Marine Resource Economics. 10(2): 117 142. The Ohio Journal of Science. 103(5): 118 125. Roheim, C., and R American Agricul tural Economics Association 2005 Annual Meeting. July 2005. Providence, Rhode Island. Roheim, C., L. Gardiner, and F. Asche. 2 Marine Resource Economics 22(3): 239 53. Marine Resource Economics 12: 37 55. Journal of Agricultural and Applied Economics 41(3): 731 744. Singh, K., M. Dey, and P. Surat Unbreaded Frozen Seafood in the United States Using Store Level Scanner Marine Resource Economics 27(4): 371 87. Singh, K., M. Dey, and P. Surathkal. Seasonal and Spatial Variations in Demand for and Elasticities of Fish Products in the United States: An Analysis Based on Market Level Scanner Dat Canadian Journal of Agricultural Economics . 62: 343 363. Home Seafood Co nsumption in Kentucky: A Double Hurdle Model App Marine Resource Economics. 17: 153 162. Fresh Tomatoes and Apples by U.S. Households: Evidence from Nielsen Journal of Food Distribution Research 40(3):105 114. Understanding Differences in Self Reported Expenditur es between House hold Scanner Data and Diary Survey Data: A Comparison of Homescan and Consumer Expenditure Survey . Applied Economcs and Perspectives Policy. 31(3): 470 492.
76 BIOGRAPHICAL SKETCH Matthew Gorstein earned his Bachelor of Science in economics from University o in the spring of 2013 . He obtained his Master of Science in food and resource economics from University of Besides devoting time to coursework, Matthew worked as a graduate research assistant for the D epartment of Food and Resource Economics and as a sales/service associate at a local furniture store. He was also Student Or ganization, serving as Vice President of Communications for the 2014 calendar year. Matthew was awarded the Aylesworth Scholarship from the Aylesworth Foundation for the Advancement of Marine Science in February of 2014 for his research into the demand f or fresh seafood products . His research has been presented at conferences and workshops such as the International Institute of Fisheries Economics and Trade in July 2014 and the Gulf State Marine Fisheries Commission Fisheries Economics Workshop in March 2014.