<%BANNER%>

Influence of Social Networks and Trust on Food Product Adoption

Permanent Link: http://ufdc.ufl.edu/UFE0024097/00001

Material Information

Title: Influence of Social Networks and Trust on Food Product Adoption
Physical Description: 1 online resource (76 p.)
Language: english
Creator: Mullady, Joy
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2008

Subjects

Subjects / Keywords: adoption, network, product, social
Food and Resource Economics -- Dissertations, Academic -- UF
Genre: Food and Resource Economics thesis, M.S.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: This research provides information about social networks and word-of-mouth referrals within a confined group. The research examines what influences recommendations from other network members has on others within the network to try a new product. The hypothesis of this research is that social network variables will better explain the change in a subject's willingness to try a new product. The data was collected with a web-based survey on two social groups of college-aged students. The survey is comprised of questions on a person's willingness to try a new product without previous marketing and then inquires about their relationship with other members in the same social group. Finally, the survey asks if a subject would be more willing to try a product based on a recommendation by someone in their social dynamic. Prior research on this topic lacks findings that include both anthropologic and economic results in a concurrent study; the combination of the two fields provides deeper insight into consumer behavior. The research at hand is unique in the fact that it is a new combination of two much researched fields. The research is trying to test whether there is a relationship between trust and a subject's willingness to try a new product based on word-of-mouth referral.
General Note: In the series University of Florida Digital Collections.
General Note: Includes vita.
Bibliography: Includes bibliographical references.
Source of Description: Description based on online resource; title from PDF title page.
Source of Description: This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility: by Joy Mullady.
Thesis: Thesis (M.S.)--University of Florida, 2008.
Local: Adviser: House, Lisa O.
Electronic Access: RESTRICTED TO UF STUDENTS, STAFF, FACULTY, AND ON-CAMPUS USE UNTIL 2009-12-31

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2008
System ID: UFE0024097:00001

Permanent Link: http://ufdc.ufl.edu/UFE0024097/00001

Material Information

Title: Influence of Social Networks and Trust on Food Product Adoption
Physical Description: 1 online resource (76 p.)
Language: english
Creator: Mullady, Joy
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2008

Subjects

Subjects / Keywords: adoption, network, product, social
Food and Resource Economics -- Dissertations, Academic -- UF
Genre: Food and Resource Economics thesis, M.S.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: This research provides information about social networks and word-of-mouth referrals within a confined group. The research examines what influences recommendations from other network members has on others within the network to try a new product. The hypothesis of this research is that social network variables will better explain the change in a subject's willingness to try a new product. The data was collected with a web-based survey on two social groups of college-aged students. The survey is comprised of questions on a person's willingness to try a new product without previous marketing and then inquires about their relationship with other members in the same social group. Finally, the survey asks if a subject would be more willing to try a product based on a recommendation by someone in their social dynamic. Prior research on this topic lacks findings that include both anthropologic and economic results in a concurrent study; the combination of the two fields provides deeper insight into consumer behavior. The research at hand is unique in the fact that it is a new combination of two much researched fields. The research is trying to test whether there is a relationship between trust and a subject's willingness to try a new product based on word-of-mouth referral.
General Note: In the series University of Florida Digital Collections.
General Note: Includes vita.
Bibliography: Includes bibliographical references.
Source of Description: Description based on online resource; title from PDF title page.
Source of Description: This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility: by Joy Mullady.
Thesis: Thesis (M.S.)--University of Florida, 2008.
Local: Adviser: House, Lisa O.
Electronic Access: RESTRICTED TO UF STUDENTS, STAFF, FACULTY, AND ON-CAMPUS USE UNTIL 2009-12-31

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2008
System ID: UFE0024097:00001


This item has the following downloads:


Full Text

PAGE 1

INFLUENCE OF SOCIAL NETWORKS A ND TRUST ON FOOD PRODUCT ADOPTION By JOY ELIZABETH MULLADY A THESIS PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLOR IDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF SCIENCE UNIVERSITY OF FLORIDA 2008 1

PAGE 2

2008 Joy Elizabeth Mullady 2

PAGE 3

To Kathryn 3

PAGE 4

ACKNOWLEDGMENTS I thank my family for all their support. I tha nk both my parents for their patience in my 6year commitment. My sister was a great lifesave r in crisis times, I thank her for always being there and able to spell-check th ings at late hours. I thank Ja mes for keeping my math skills sharp. William was there to keep me sane at family gatherings through all his comedy insights. Without a friend like Kathryn I do not think I would be here getting my degree. I thank her for always being around in times of need to help me with homework, life, and relationships. Steven was there in the good times and bad, always reminding me of the future and what I have to look forward to. A few professors come to mind whom I would like to thank. I thank Dr. Gary Fairchild, for believing in my potential and showing me how one countrys actions affect the rest of the world. I thank the Drs. House for helping me explore all the different as pects of economics and helping me understand that anthropology can play a role virtually anywhere. I thank Dr. Carrion-Flores for all the support. 4

PAGE 5

TABLE OF CONTENTS page ACKNOWLEDGMENTS ............................................................................................................... 4LIST OF TABLES ...........................................................................................................................7LIST OF FIGURES .........................................................................................................................8ABSTRACT ...................................................................................................................... ...............9 CHAPTER 1 DIFFUSION OF INNOVATION ...........................................................................................10Introduction .................................................................................................................. ...........10Researchable Problem ............................................................................................................132 LITERATURE REVIEW .......................................................................................................153 THEORY ................................................................................................................................20Product Adoption ....................................................................................................................20Consumer Decision Behavior .................................................................................................21Word Of Mouth Referral ........................................................................................................22Social Networks ............................................................................................................... .......234 DATA ........................................................................................................................ .............275 METHODS ..................................................................................................................... ........37Social Network ................................................................................................................ .......37Empirical Models .............................................................................................................. ......39Ordered Probit .................................................................................................................40Selectivity Bias .............................................................................................................. ..41Limitations ................................................................................................................... ....426 EMPIRICAL RESULTS ........................................................................................................45Social Network Model .......................................................................................................... ..45Econometric Results ........................................................................................................... ....46Candy Bar Equation Results ............................................................................................49Sandwich Equation Results .............................................................................................51Difference in Restaurant ..................................................................................................527 CONCLUSIONS ................................................................................................................. ...67APPENDIX: WEB SURVEY ........................................................................................................72 5

PAGE 6

LIST OF REFERENCES ...............................................................................................................74BIOGRAPHICAL SKETCH .........................................................................................................76 6

PAGE 7

LIST OF TABLES Table page 4-1 Demographic profile of groups ..........................................................................................314-2 Knowledge level of Group 1 ..............................................................................................324-3 Knowledge level of Group 2 ..............................................................................................334-4 Socially interacted within last 30 days, Group 1 ...............................................................334-5 Socially interacted within last 30 days, Group 2 ...............................................................344-6 Trust between group members, Group 1............................................................................344-7 Trust between group members, Group 2............................................................................356-1 Model prediction .......................................................................................................... ......596-2 Difference candy regression results, Group 1 ....................................................................606-3 Difference candy regression results, Group 2 ....................................................................616-4 Difference sandwich regr ession results, Group 1 ..............................................................626-5 Difference sandwich regr ession results, Group 2 ..............................................................636-6 Difference restaurant regression results, Group 1 ..............................................................646-7 Difference restaurant regression results, Group 2 ..............................................................656-8 Significant variables in regression models ........................................................................666-9 Wald Test ................................................................................................................. ..........66 7

PAGE 8

LIST OF FIGURES Figure page 3-1 Diffusion of Innovation curve ............................................................................................2 63-2 S-shaped curve representing product adoption ..................................................................264-1 Willingness to try, Group 1 ............................................................................................... .314-2 Willingness to try, Group 2 ............................................................................................... .324-3 Difference after recommendation, Group 1 .......................................................................354-4 Difference after recommendation, Group 2 .......................................................................365-1 Social network visualization ..............................................................................................435-2 Core membership ........................................................................................................... ....435-3 Visual of betweenness........................................................................................................445-4 Network visual of variables ...............................................................................................446-1 Group 1 Network ...............................................................................................................556-2 Group 2 Network ...............................................................................................................556-3 Group 1 Betweenness ........................................................................................................566-4 Group 2 Betweenness ........................................................................................................566-5 Group 1 Closeness .............................................................................................................576-6 Group 2 Closeness .............................................................................................................576-6 Group 1 Degree ..................................................................................................................586-7 Group 2 Degree ..................................................................................................................59 8

PAGE 9

9 Abstract of Thesis Presen ted to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Masters of Science INFLUENCE OF SOCIAL NETWORKS A ND TRUST ON FOOD PRODUCT ADOPTION By Joy Elizabeth Mullady December 2008 Chair: Lisa House Major: Food and Resource Economics This research provides information about so cial networks and wo rd-of-mouth referrals within a confined group. The research examines what influences recommendations from other network members has on others within the networ k to try a new product. The hypothesis of this research is that social network variables will be tter explain the change in a subjects willingness to try a new product. The data was collected wi th a web-based survey on two social groups of college-aged students. The survey is comprised of questions on a persons willingness to try a new product without previous mark eting and then inquires about their relationship with other members in the same social group. Finally, the surv ey asks if a subject w ould be more willing to try a product based on a recommendation by someone in their social dynamic. Prior research on this topic lacks findings that include both anthropologic and ec onomic results in a concurrent study; the combination of the two fields provides deeper insight into consumer behavior. The research at hand is unique in the fact that it is a new combination of two much researched fields. The research is trying to test whether there is a relationship between trust and a subjects willingness to try a new product based on word-of-mouth referral.

PAGE 10

CHAPTER 1 DIFFUSION OF INNOVATION Introduction Out of all the new food products brought to th e retail shelf every year very few succeed. The failure rate of new food products is 70 to 80%. A new product intro duction cost for grocery stores averages approximately $270 each. Each y ear retail grocery stores spend an average of $956,800 to introduce new items that will li kely fail (Frozen Food Digest, 1997). There are many things that influenc e a consumers willingness to adopt1 a new food product, including recommendations made by friends and family; demographic characteristics; and a persons own curiosity and innovativeness. A persons innovativeness is something developed over the life span and there are differe nt levels to ones a doptive behavior of new products. The different levels of innovativeness are broken into fi ve categories: innovators; early adopters; early majority; late majority; and lagg ards. Not only can levels of innovativeness vary among individuals, but for each individual, the level of innovativeness may vary depending on the type of product. Relative advantage is th e degree to which an innovation is perceived as being better than the idea it supersedes (Rogers, 2003). That is the idea of economic profitability, consumers may be more conscious of the decision to adopt a new product or technology when a significant cost is involve d; on the other hand when the product is inexpensive they may be willing to try the product early on. When the price of a new product decreases so dramatically during its diffusi on process, a rapid rate of adoption is encouraged(Rogers, 2003). Recommendations from family and friends also have been shown to influence a consumers adoptiveness. These recommendati ons from people within a consumers social 1 Adopt is defined as regular use of food product. 10

PAGE 11

network give the consumer guidance and reassuran ce that someone they trust believes they will enjoy the product. Family and friends play a la rge role in decisions wi th which everyone is faced; therefore it is no surprise that they tr ust and value the opinions of these members. Consumers receive influences from many differen t directions; some examples would be family, groups, church, and peers. The vast number of gr oups that consumers belong to also plays a role in their willingness to adopt new ideas. A group can easily influence members to try and explore new products. Within these groups trust between members plays a role in whether or not a consumer would be willing to adopt the new product another peer would recommend. Trust is seen in social situations throughout a consumer s network of people; consumers must value and trust the opinions of peers othe rwise why would they accept recommendations from said peers. A consumers social network, tr ust, and willingness to try2 new things all play a role in their adoptive behavior of new products. There are also five defined classes of so cial behavior for when a person adopts new products such as innovators, early adopters, earl y majority, late majority and laggards. The classes of social behavior are based on a bell shaped curve wide ly publicized by Everett Rogers. The curve is broken into five diffe rent categories, the first of which is innovators. Innovators can be described as people on the cutting edge of technology (Orr, 2003). The implementation and confirmation of the new product by innovators plays a large role in the early adopters decision to adopt. Early adopters use the information gath ered by their predecessors; if they observe that the information has been effective they are encour aged to adopt (Orr, 2003). Within the category of early adopters are found the opinion leaders. Opinion leaders are perceived by others to make informed decisions. The early majority follo ws suit with the opinion leaders of the early 2 Try is defined as willingness to consume new or different products. 11

PAGE 12

adopters group. The early majority is consider ed conformity-loving and will follow suit based on the previous trends of the early adopters (Orr, 2003). Therefore if the opinion leaders among the early adopters adopt a product a nd recommend it, the early majority is likely to follow suit. The late majority and the laggards are considered eith er very traditional or isolated in their social system. If they are traditional they a skeptical of how this ne w product will affect old traditions and if they are isolates their l ack of social awareness decreases the awareness of social benefits (Orr, 2003). Innovation is assured through a soci al process, marked by an opinion leaders adoption. The well informed leaders are able to communicate their approval or disapproval of the product and their opinion is conveyed to the masses. This s hows that opinion leaders are an important subset of the social system and reco mmendations are taken into account by consumers. Related to this concept is the concept of soci al networks. Social networks are referred to as groups of individuals that share a common space and interact on a regular basis with each other. Within a network of people trust plays a role with members attitudes towards people. Social networks and their members have been s hown to influence members willingness to try new things. Trust likely plays a role in memb ers willingness to accept recommendations from others. If a member has a low level of trust the member may be less likely to adopt the new product; or if the trust is high the member may be willing to try it immediately. Trust has multiple definitions, such as: assured reliance on the character, ability, strength, or truth of someone or something, in which confidence is placed, dependence on something future or contingent, a charge or duty imposed in faith or confidence or as a condition of some relationship (Merriam-Webster, 2008). The definition that will be relevant to the resear ch at hand someone in which confidence is placed; theref ore when the word trust is used this is meaning the paper is referring to. 12

PAGE 13

Word-of-mouth communication is one of the most widely accepted notions in consumer behavior and plays an important role in shaping consumers attitudes (Reingen & Brown, 1987). Word of mouth is the interac tion between people on an interpersonal level. The relationship between social structure and influence needs to be examined further. Reingen and Brown also suggest that network analysis has been deve loped into a powerful method for investigating linkages between micro and macro level processes. Our goal was to combine methods commonly used in both anthropology and economics to establish the variables th at affect consumers willingness to try new products based on word-ofmouth referrals. By defining the influencing factors of a social ne twork, this research attempts to isolate the variables that will affect consumers willingness to try new products based on word of mouth referrals. Researchable Problem The goal of this thesis is to answer the question, what influences consumers willingness to try new food products, including the impact of word-of-mouth referrals? This question was addressed empirically by estab lishing factors affecting consum ers product adoption, including social network variables. Attrib utes of different groups throughout the University of Florida will be studied in a social network framework to de termine what influences a subjects likelihood to respond to word-of-mouth referrals in regard to trying new food products. Variables that will be used to determine the impact of the referral will include trust, a subject s own adoptiveness, and the strength associated between subjects within the social network. The main hypothesis that will be tested is that social network variables will increase the ability to correctly predict a subjects willingness to listen to recommendations from fellow subj ects. The general focus of the equations developed for this research is as follows: 13

PAGE 14

14 H1: Difference= (knowledge, age, gender, primary shopper, variables describing the recommenders place within a network; variables de scribing the referrers place in the network) H2: Difference= (knowledge, age, gender, primary shopper) Where difference is equal to the difference in willingness to try a new food product before and after a recommendation.

PAGE 15

CHAPTER 2 LITERATURE REVIEW Word-of-mouth communication is one of the most widely accepted notions in consumer behavior and plays an important role in shaping consumers attitudes (Reingen, & Brown, 1987). The purpose of Reingen and Browns study was to examine from an interpersonal network perspective, the roles that tiestrength and homophily may play in macro and micro word of mouth processes. In their study, they tested nine different hy potheses to study the strength of weak and strong ties. The study was conducted on a who told whom network of information traced out for three piano teachers. None of the teachers used formal marketing techniques for their services; instead they relied on positive word of mouth to generate market transactions. Reingen and Brown found that without formal ma rketing techniques in place the piano teachers were receiving enough word of mo uth advertising or praise that a formal technique was not necessary. To conclude Reinge n and Brown suggest that network analysis has been developed into a powerful method for investigating linkage s between micro and macro level processes. Going further they indicate the relationship between social structure and word of mouth influence needs to be more closely examined by including more complex measurement of social structure (Reingen, & Brown, 1987). Reingen and Kernan (1986) studied a service marketer: a piano tuner who did not engage in any formal marketing communication. Ther e were two phases to the study. The first identified the referral network, and the second produced data on the social network. The hypothesis of this study was confirme d that the actors potential sources of referral, the stronger the tie, the more likely it is to be activated for the flow of referral (Reingen & Kernan, 1986). The stepwise logistic regression analysis revealed that for both data sets the greater the importance attached to a social relation the more likely it was to be activated for a referral within 15

PAGE 16

groups (Reingen & Kernan, 1986). Reingen and Kern an (1986) also found that consumers with multiple potential sources for referral were more likely to be activated if the referral came from a source with a significant tie. Shaw (1965) also suggested that product adoption can, in many cases, be fueled by communication between tastemakers/ innovators and a lower status group. Shaw points out the fact that the flow of comm unication is a two way street; thus a new product or service might be adopted initially by tastemakers/ innovators. Therefore the presence of innovators in the small informal groups is an important factor fo r early product development. This is supported by studies from other anthropologists and sociologists who show the rate of adoption by upper class respondents was more conservative or slower than the rate by lower class respondents in adopting a television (Shaw, 1965). The crucial question Shaw tried to answer was: who are these influential individuals and how does marketi ng research locate them? Shaw also suggested that initial adoption of a product could start at the top, middl e or lower rungs of society; he measured class mobility as how many friends the individual had in different walks of life (1965). Udry and Conley (2004) examined social netw orks among farmers within four villages in developing Ghana. Within each village 60 marri ed couples were chosen, and interviewed 15 times during the course of two years. The st udy explores determinates of the particular economic networks. Udry and Conley use a met hod of analysis for sample size that is fixed across sample villages. That is, the probability of observing any link in village V is (S/N)2. The sampling is based on villages and captures only links between villages. Development was shaped by networks of information being passe d from one household to the next and some information would make it to the marketplace wher e it was spread through other villages. Udry and Conley discuss that their findings exemplify how economic development is being shaped by 16

PAGE 17

networks of information, capital and influence in these communities. These authors also discuss that the data of social netw orks is egocentric; the links between individuals is very informational among the connections through networks. This demonstrates that all social networks are composed of subgroups; each indivi dual within the network has clusters throughout their own social network. Coleman et al. (1957) provided findings of co mmunication within a so cial network when they take a group of physicians and the social st ructure through which th ey are linked. This study is one of the few that attempts to exam ine the relationship betw een networks and their members. In their study, Coleman et al. de signed a method of survey research involving structured interviews with a sample of physicians However, they encountered a problem in that the survey needed to treat individuals as indepe ndent units of observation. This was corrected for by asking each doctor three sociometric questions that were used to establish links within the medical community. Coleman et al. did this to assure that each docto r named by other doctors was included within the sample. The results were that even stronge r relations were found between doctors when they turned to social influences. This showed that the highly integrated doctors seemed to have learned from one anothe r while the less integrated learned from outside sources such as trade publications and medical jo urnals. Coleman et al.s findings support the hypothesis that social network variab les play a role in an actors willingness to listen to product recommendations. Reingen et al. (1984) suggeste d that students of consumer behavior focused on social influence to determine consumer brand choices. Reingen et al. (1984) used an example of a sorority to demonstrate that brand congruence is influenced by relationships within social networks. Reingens study focused on 49 members of a sorority at a Southwestern university in 17

PAGE 18

which 75.5 % of the subjects lived in the sorority house. Subjects filled out a self-administered questionnaire at one of the monthly meetings (a controlled environment because nonmembers are not allowed at these meetings). Subjects were as ked to list the name of the brand they use for hygiene products and the br and they prefer. Rei ngens study showed us the influence consumers have on others within a group when exhibiting bra nd consumption for products that were likely visible in at least one social relationship. There were significan t findings for private products such as shampoo and toothpaste (Reingen et al ., 1984), however, there were no statistically significant findings for public products (such as cl othing and watches). This is the opposite of what previous literature suggests. Overall the results of this study strongly suggest that brand congruence depends heavily on a complex mix of t ype of product, type of social relation, and type of social structural unit (Reingen et al., 1984). Doriean (2003) discussed how the core simulation model implements two social processes: the group process determines clusters of actors that minimize total group imbalances, and the simulated actors use social know ledge about cluster membership to make social choices that minimize their individual social imbalances. Ac tors are independent beings within a group, and therefore make their own decision based on the ties they have with other actors. The simulation completed by Doriean suggests th at though initial contentiousness is relevant, the modes of communication are important fo r balance theoretic dynamics. Fisher and Price (1992) used a different model in their resear ch to explain that social factors are likely to mediate the effects of adve rtising and other marketing strategy variables on consumers adoption decisions. The personal component of the m odel comprises an individuals belief that engaging in a behavior leads to salie nt personal outcomes and an evaluation of these outcomes (Fisher & Price, 1992). The normative component of the model comprises beliefs 18

PAGE 19

19 about the likelihood and importance of the social consequences of undertaking a behavior. They conclude that the study provide s evidence that perceptions of consumption visibility and expectations of social approval can vary wide ly for a single product (Fisher & Price, 1992). Busken (1998) discussed that the level of a buye rs trust is higher if he talks to buyers who talk to many other buyers, because the message will spread at a much higher rate in the network assuming that people keep on spreading the messa ge. Busken (1998) also discusses findings from previous studies such as Colemans di ffusion of innovation and Granovetters (1983) established result that the level of trust that can be placed is higher if the network has a higher density. There are many articles discussing word-ofmouth referrals and many articles discussing social networks, although there ar e no studies or articles publishe d with the extensive insight of the two combined. The question that has not yet been answered is one which helps to define which variables within social networks influe nce a consumers willingness to try a new product based upon the referral of an act or within a social network.

PAGE 20

CHAPTER 3 THEORY Product Adoption Product adoption occurs in stages, with c onsumers first going through a decision making process. One of the beginning steps in pr oduct adoption is produc t knowledge. Product knowledge refers to the amount of information a c onsumer has stored in his or her memory about particular product classes, product forms, brands models and ways to purchase them (Peter & Donnelly, 2003). The initial level of product know ledge may influence how much information is attained when deciding to make a purchase (P eter & Donnelly, 2003). The second stage, the decision stage occurs when a person decides to adopt the new technology. The implementation stage occurs when a consumer puts the innovation to use (Rogers, 2003). Until this point, the innovation decision was simply a me ntal exercise. The confirma tion or post-purchase evaluation stage follows the implementation stage. The innova tion decision process theo ry which states that innovation diffusion is a proce ss that happens over time through five different stages: knowledge, persuasion, decision, implementation and confirmation. The knowledge stage is where a person gathers the information needed about a new technology to help the decision process. Persuasion is where the pe rson is influenced by early adopters. The diffusion of innovation is a marketing con cept that was developed by Everett Rogers. It is described as a process by which an innovati on is communicated through channels over time. The diffusion of innovation theorized that innovati ons would spread through society in a bell shaped curve (Rogers, 1962). The adoption of an innovation usually follows a normal bell-shaped curve ( Figure 3-1 ) when plotted over time on a frequency basis (Rogers, 2003). If the cumulative number of adopters is plotted, the result is an S shaped curve ( Figure 3-2 .) (Rogers, 2003). 20

PAGE 21

According to the theory of innovation, this S-curve repres enting product adoption can be divided into five different sec tions. These five sections are: innovators, early adopters, early majority, late majority, and laggards. This is created by the fact that people adopt the innovation at different rates. The rate at which people adopt is represen ted by the bell-shaped curve see Figure 3-1 ., which can also be broken into five cate gories: innovators; early adopters; early majority; late majority; and laggards. Different classes of consumers tend to gather information from earlier classes that adopt a new technology. The information gathered aids in the decision making process. In general consumers recognize a need for a product, search for information about alternatives to meet the need, evaluate the information, ma ke purchases, and evaluate the decision after the purchase(Peter & Donnelly, 2003). Putting these curves together, on e can see that the innovators and the early adopters select the innovation first, followed by the majority and finally the laggards, until a technology is common. A key element in the innovation of a pr oduct is consumers attitude toward the new product. Consumer Decision Behavior Consumer behavior studies consumers purchasin g behavior. It examines when, what, why and how consumers make their purchasing decisions Consumer behavior is revealed in many different disciplines of soci ety. Marketing, psychology, anthr opology, and economics all explore this topic. Understanding the theo ry behind consumer behavior opens doors to infinite potentials within many different markets. The marketing pot ential of understanding consumer behavior is one of great magnitude; this would allow industry executives to target their customers with full understanding of when or why th ey will purchase a new product. Psychology and anthropology explore this topic to discover an explanation of why consumers behave the way they do. Economics utilizes the skill provided with marketi ng and wants to be able to explain the demand 21

PAGE 22

curve, a major feature of economics. Consumers behavior affects many different categories in various disciplines and being ab le to explain this theory by combining methods from these disciplines may provide more in-depth understand ing. Explaining consumer behavior is one of the goals of this research as it is trying to delve into reasons behind why a consumer would consume or try a new product based on word of mouth referrals from fellow consumers. Word Of Mouth Referral Word of mouth refers to pa ssing information between consumers via talking. Word of mouth communication consists of three essen tial parts: that it contains interpersonal communication: that the content from a marketing perspective is commercial: and even though the content is commercial the communicat ors are not motivated commercially (Kirby & Marsden, 2006). Word of mouth referrals are most commonly used to describe positive referral rather than negative advertisi ng. Word of mouth marketing is on a more personal basis than more formal ways of advertising, such as te levision ads, newspaper or magazine ads, and billboards. The communication between individua ls is thought to have an added layer of credibility, a trust variable in most cases. Res earch has shown that indi viduals are more willing to accept referrals from another individual who is known to them than from a stranger. Word of mouth referrals can spread rapidly throughout communities and are great sources of free marketing to companies. People listen to word of mouth because it is part of their normal information search about brands, when ente ring the market place (Kirby & Marsden, 2006). The theory behind word of mouth referrals is that the news of a product will spread among consumers of similar groups. This is slightly like the domino affect with product referral. Consumers have also been more willing to try new products based on another consumers recommendation of the product if the recommender is close to the said consumer. Consumers are most influenced by family members and frie nds, these are consider ed strong ties; where 22

PAGE 23

weaker ties tend to help the spread of a message through society (Kirby & Marsden, 2006). Consumers also tend to seek out word of mouth referrals more often when the purchase decision is risky (Kirby & Marsden, 2006). The more unders tanding there is of word of mouth referrals the greater marketing advantage. Being able to establish the innovators or trend setters within groups would help with social marketing and theori es such as word of mo uth referrals. Also with the increasing use of the Wo rld Wide Web and social networki ng sites, it is inherent that there are further studies into the as pect of social marketing channels. Social Networks Udry and Conley (2004) state that there is a st rong spatial element to social networks and it is useful to visualize their shap e in real space. Th e location of each subject is determined by the average position of the plot he/she cultivates a nd is indicated by a dot on a graph. Information networks appear to be characterized by pres ence of focal individuals, whom are linked to multiple ties in the sample. Brown and Reingen (1987) suggest that geographic theory is highly useful when showing relationships between indivi duals. The graphs Reingen uses are simple to explain since a point represents an individual an d each line represents a pairs relationship (1987). Wellman (1983) breaks down social network analysis into ma ny different parts and five main principles: 1. Ties are often asymmetrically reciprocal, differi ng in intensity and content. This indicates that although person A indicates they know pe rson B at a certain level, person B may not identify the same level (if any) of knowle dge with person A. Although ties are rarely symmetric they are usually reciprocated in a generalized way, and ar e a stable part of a social system. 2. Ties link network members directly as well as indirectly, causing us to look at larger network structures. 23

PAGE 24

3. The structuring of social ties creates nonrandom networks hence network clusters, boundaries and cross-linkages arise. This explains that most individuals are not members of only one group therefore there are cross-linkages or connected clusters. 4. Cross linkages connect clusters as well as individuals. Asymmetric ties and complex networks distribute scarce re sources differentially; unequal access to scarce resources may in turn increase the asymme try of ties (Wellman, 1983). 5. Networks structure collaborative and competitive activities to secure scarce resources, in a system with finite resources; social networ ks compete for the access to such resources. Social networks can be defined by different variables that expl ain the ties or links between actors. Many scholarly sources use Lin Freeman s definition of central ity variables (Wellman, 1983). Each measure is defined below: Centrality is the measure of the extent to which a particular node or actor dominates a network (Wellman, 2008). This is calculated by co mparing centrality of the most central point to the centrality of other points. Degree, betweenne ss and closeness are all measures of centrality. Degree is the number of connect ions a node or actor has within a network. As degree increases the number of actors a nod e is tied to relationally increas es. An example would be the more people someone knows within the ne twork the higher their degree would be. Coreness is an example of a degree measure. Coreness is the measure of how centrally located an actor or node is loca ted to a subset of the group. An example would be a clique, whether the actor is a part of the clique or not. Closeness is the sum of the distance to each pe rson within the network. Therefore people with a large closeness sc ore are on the outside of the group, not di rectly tied to others within the group. Betweenness is the measure of the extent to which a node lies on the geod esics between each other pair of nodes (Wellman, 2008). Betw eenness can help explain how centrally located an individual is within a closed network, whether the actor is far away from the node that ties a 24

PAGE 25

group together such as the pres ident of a group. The betweenne ss score increases the farther away this person is from the center of the group. Consider an example of a pyramid where the betweeness score increases the further down the pyramid the subject is, while the subjects betweenness score decreases the closer to the top of the pyramid the subject is. 25

PAGE 26

26 Figure 3-1. Diffusion of Innovation curve Figure 3-2. S-shaped curve re presenting product adoption

PAGE 27

CHAPTER 4 DATA Data for this study were collect ed through an internet survey developed in Survey Crafter Professional (Survey Crafter, Inc., 2008). The survey was completed by two student organizations at the University of Florida. Th ese two groups took the surv ey to fulfill the studys need for small whole networks. Each group had no more than 30 students who interacted with each other on a regular basis. Participants in the survey were asked a bout their willingness to try new food products, whether they were the primary grocery shopper in their household and demographic questions such as age, race, gender, and ethnicity. They were then asked to identify other students from within the group he or she knew, as well as how many times within the last month they saw the other students in a social sett ing. In addition to gathering information on knowledge, students were asked how much they truste d people they saw on at least one social occasion. The last series of questions delved into how likely th ey would be to try three new products based on recommendations from students they knew. The study was completed by two campus organiza tions that were compensated 150 dollars for their participation. The first group consis ted of 23 members with 16 completed surveys, creating a response rate of 70 %. The second group consisted of 21 members, of which 13 completed the survey, for a 62 % response rate. Answers to the demographic questions ( Table 4-1 ) indicate that the two groups were primarily Caucasian. The age ranged from 19 to 22 in group 1, and 18 to 23 in group 2. The majority of respondents were primary shoppers with more than 50% indicating this in each group. Groups 1 and 2 had no one indicate they were of Hispanic descent. 27

PAGE 28

Respondents were asked to identify how likely th ey were, on a 5-point Li kert scale, to try three new products: a candy bar; a sandwich from a fast-food restaurant; a meal at a new sitdown restaurant ( Figures 4-1 4-2 ). Most respondents indicated they would be unlikely to try a new candy bar. Respondents from groups 1 and 2 said there was some chance to a fifty-fifty chance of trying a new sandwich. Groups 1 thought that there was some chance they would try a new relatively expensive sit down restaurant, wh ile those in group 2 t hought it was likely they would try a new sit down restaurant. Tables 4-2 and 4-3 show the survey responses i ndicating how well each respondents knows every other respondent within each group, ba sed on a 1 to 5 scale, with 1 being they do not know the person, and 5 being highly knowledg eable. In group 1 person P had the highest average knowledge of each person in the group. Th e range of scores for group 1 were 1.1 to 4.2. In Group 2 there was an indicati on that people slightly knew each other with the average being between 2.5 and 2.9. On average each student knew the other students 3.23, in group 1. Person K in group 1 had the highest knowledge score of the other members within the organization. There was a range of 1.31 to 4.19 of knowledge be tween members in group 1. Group 2 had an average of 2.71 level of knowle dge between members. Person H and person D had the same knowledge score spread across members in th e organization, 3.46. The range of knowledge within group 2 was 1.92 to 3.46, within group 2 ev eryone had at least some knowledge of all other members. Tables 4-5 and 4-6 show how many times within the last 30 days each respondent has interacted socially with another respondent from the group. This question was asked only if the person had indicated that they knew the subject. Person P in group 1 seemed to interact the most often with other group members. Person P intera cted with person H over 40 times in the last 28

PAGE 29

month. Person M in group 1 interacted with person H 30 times in the la st month. Many of the members in group 1 said they interacted with another member of the group over 20 times in the last month, C, D, E, I, J, M, O, P. Group 1 had a range of 0 interactions to 10 interactions a month on average; the average was 5.65 interactions a month. Group 2 had a range of 2.24 interactions to 4.23 interactions a month. Group 2 did not have as many interactions with other memb ers as group 1; this groups average was 3 interactions in the last 30 days Person E said they interacted on average 7 different times within the past thirty days with other members. Pe rson H was the member that other members most often said they had interaction between group members. Respondents were only asked how well they trus ted people they had at least interacted with one time socially. Therefore, if they did not know the person, the space is left blank because this question was not asked of them. Tables 4-8 and 4-9 show the amount of trust each respondent has in each member in the group. The tr ust variable was chosen on a scale of 1 to 5, 1 being do not trust, and 5 being highly trust. Respondents in group 1 indi cated a high trust level of the members within the group, with a range of trust levels from 3.67 to 4.53. Group 2s scores indicated a trust leve l of an average 3.5. After answering the questions about knowledge and trust, each respondent was reminded of their answers to these questions and asked again to indicate how likely they were to try these three products a candy bar, a new sandwich at a fa st food restaurant, and eat at a new relatively expensive restaurant after a recommendation from people th ey indicated they knew. Figures 4-4 and 4-5 show the difference between their sc ores and the new score based on the recommendations of others in groups 1, and 2 respectively. 29

PAGE 30

For group 1, the most frequent occurrence wa s no change in the willingness to try. However, for all products, there were some cases where willingness to tr y increased, and others where it decreased. This was the same for group 2. 30

PAGE 31

Table 4-1. Demographic profile of groups Race Group 1 Group 2 Caucasian 93.75% 92.31% African American 0.00% 7.69% American Indian 0.00% 0.00% Asian 0.00% 0.00% Other 6.25% 0.00% Gender Male 43.75% 76.92% Female 56.25% 23.08% Age 18 0.00% 15.38% 19 12.50% 23.08% 20 31.25% 23.08% 21 43.75% 15.38% 22 12.50% 7.69% 23 0.00% 15.38% 24 0.00% 0.00% 25 0.00% 0.00% Primary Shopper No 18.75% 30.77% Yes 81.25% 69.23% Hispanic Descent No 100.00% 100.00% Yes 0.00% 0.00% 0246810 Unlikely Some Chance Fifty fifty chance Likely Highly probable Frequency Restaurant Sandwich CandyBarFigure 4-1. Willingness to try, Group 1 31

PAGE 32

02468 Unlikely Some Chance Fifty fifty chance Likely Highly probable Frequency Restaurant Sandwich CandyFigure 4-2. Willingness to try, Group 2 Table 4-2. Knowledge level of Group 1 ID A B C D E F G H I J K L M N O P A 0 1 1 3 3 2 1 3 3 2 3 4 4 3 3 4 B 1 0 1 1 1 1 1 2 1 1 3 2 1 1 1 3 C 1 2 0 5 5 4 2 3 3 3 3 3 3 2 4 4 D 3 1 5 0 5 4 1 4 4 3 3 3 4 3 3 4 E 4 1 5 5 0 4 1 4 3 3 3 3 4 2 2 5 F 3 1 5 5 5 0 1 5 5 5 5 5 5 5 5 5 G 1 1 1 1 1 1 0 2 2 2 2 2 1 2 1 2 H 4 2 3 4 4 3 1 0 4 4 5 4 5 4 4 5 I 4 2 4 5 5 5 1 5 0 4 5 5 5 5 5 5 J 4 1 3 3 4 4 1 5 4 0 4 4 5 4 4 5 K 5 4 4 5 5 4 1 5 5 4 0 5 5 5 5 5 L 5 2 4 4 4 4 1 5 5 5 5 0 5 5 5 5 M 4 1 3 3 3 4 1 5 4 4 4 4 0 4 4 5 N 3 1 2 2 3 3 1 4 5 3 5 4 4 0 5 5 O 4 1 5 4 4 4 1 5 5 5 5 5 5 5 0 5 P 4 2 3 3 4 4 2 5 5 5 5 5 5 4 4 0 Avg 3.1 1.4 3.1 3.3 3.5 3.2 1.1 3.9 3.6 3.3 3.8 3.6 3.8 3.3 3.4 4.2 32

PAGE 33

Table 4-3. Knowledge level of Group 2 ID A B C D E F G H I J K L M A 0 2 3 3 2 3 3 4 2 4 4 3 3 B 2 0 4 3 3 3 2 3 3 3 4 3 3 C 3 4 0 3 3 3 3 3 3 3 4 3 4 D 3 4 3 0 3 5 3 4 4 4 5 3 4 E 3 4 4 3 0 3 3 3 4 4 3 3 3 F 2 3 2 4 2 0 3 4 3 3 3 3 2 G 4 3 3 3 2 3 0 3 3 3 2 3 3 H 3 3 4 4 3 5 3 0 4 4 4 3 5 I 2 3 3 3 3 3 2 3 0 3 3 2 3 J 2 2 2 2 2 2 2 2 2 0 2 3 2 K 2 4 3 4 2 3 2 2 3 2 0 2 2 L 3 2 2 3 2 2 3 3 3 3 3 0 3 M 2 2 3 2 2 2 3 3 2 2 2 2 0 Avg. 2.38 2.77 2.77 2.85 2.23 2.85 2.46 2.85 2.77 2.92 3.00 2.54 2.85 Table 4-4. Socially interacted within last 30 days, Group 1 Name A B C D E F G H I J K L M N O P A 3 2 0 1 1 0 1 0 3 0 0 1 B 1 1 0 1 C 0 40 20 10 0 1 1 1 1 1 1 0 1 1 D 6 30 30 10 10 10 0 10 5 15 0 0 10 E 10 12 25 8 15 10 8 10 8 10 5 4 15 F 1 6 6 4 3 1 2 1 0 1 1 1 3 G 2 2 1 2 3 1 3 H 2 0 0 2 2 0 3 3 4 4 10 2 1 10 I 8 0 4 8 10 4 12 8 25 10 10 8 8 11 J 5 0 1 5 0 10 5 5 10 10 0 2 50 K 4 0 1 4 4 0 6 20 2 2 6 2 2 6 L 5 0 1 3 4 1 5 5 2 3 5 3 10 10 M 6 0 1 1 0 30 3 3 5 5 1 1 20 N 3 0 1 2 0 3 4 2 6 4 3 25 5 O 6 1 3 3 1 6 8 4 6 25 3 25 8 P 15 0 3 7 10 1 0 40 20 20 20 15 30 15 15 Avg. 5.9 0 5 8 7 3 0 10 7 4 7 6 8 5 5 10 33

PAGE 34

Table 4-5. Socially interacted within last 30 days, Group 2 ID A B C D E F G H I J K L M A 1 1 1 1 1 3 15 1 1 1 1 1 B 1 3 1 1 1 1 1 2 1 2 1 1 C 1 3 1 1 1 3 1 2 1 2 4 D 2 2 2 2 1 2 2 3 2 5 2 2 E 1 9 1 9 3 9 9 9 12 5 5 12 F 2 3 3 6 2 2 5 3 2 2 2 2 G 5 4 4 4 4 4 4 5 4 4 4 5 H 2 2 2 2 1 3 1 2 2 1 1 2 I 4 4 4 5 4 5 4 5 5 5 4 5 J 3 3 3 2 3 1 2 3 1 1 3 3 K 2 5 4 7 1 1 3 1 1 L 1 1 1 2 1 2 1 2 2 2 1 1 M 5 5 6 4 4 5 5 5 5 5 5 4 Avg. 2.2 3.2 3 3 2 2 3 4 3 3 3 2 3 Table 4-6. Trust between group members, Group 1 ID A B C D E F G H I J K L M N O P A 4 4 4 4 5 5 5 B 2 3 3 C 5 5 5 4 4 4 4 5 4 5 5 D 3 5 4 4 4 4 4 3 4 5 E 3 5 5 4 3 3 4 4 3 4 2 2 5 F 3 5 5 5 3 4 4 4 5 4 4 4 G 2 2 2 2 2 2 2 H 4 4 4 5 5 5 4 4 4 5 5 I 4 4 4 4 3 4 3 5 4 4 4 4 4 J 4 2 3 4 4 4 4 4 4 5 K 3 3 3 3 4 5 3 4 4 4 5 5 L 5 4 4 4 3 3 4 3 3 4 5 5 5 M 4 3 3 4 3 4 3 4 3 4 5 N 4 3 4 4 5 4 5 4 4 5 5 O 5 5 3 3 4 5 5 5 5 5 5 5 5 P 5 4 4 4 3 5 5 5 5 5 5 4 5 Avg. 3.9 4.4 3.8 3.9 3.7 3.7 4.1 3.8 4.1 3.9 4.3 3.7 4.4 4.5 34

PAGE 35

Table 4-7. Trust between group members, Group 2 ID A B C D E F G H I J K L M A 3 3 3 3 3 4 4 3 4 4 4 4 B 3 4 3 3 3 3 3 4 3 4 3 3 C 4 5 4 4 4 5 4 4 4 4 5 D 3 4 3 3 5 4 4 5 4 5 3 4 E 4 4 5 4 4 4 4 5 5 4 4 5 F 3 4 4 5 3 3 4 4 3 3 3 3 G 3 3 3 3 3 4 3 4 3 3 3 3 H 4 4 4 4 4 4 4 4 4 4 4 4 I 3 4 3 4 3 4 3 4 4 4 3 4 J 4 4 4 4 4 4 4 4 4 4 4 4 K 1 3 3 4 1 1 3 1 1 L 3 2 2 2 2 2 2 3 3 3 2 2 M 4 4 4 4 4 4 5 5 4 4 5 4 Avg. 3.25 3.67 3.50 3.67 3.20 3.50 3.64 3.67 3.92 3.50 3.83 3.55 3.50 0 20 40 60 80 100 120 Decreased >1 Decreased by 1 No changeIncreased by 1 unit Increased > 1 Candy Sandwich RestaurantFigure 4-3. Difference after recommendation, Group 1 35

PAGE 36

36 0 10 20 30 40 50 60 70 80 90 100 Decreased >1 Decreased by 1 No changeIncreased by 1 unit Increased > 1 Candy Sandwich RestaurantFigure 4-4. Difference after recommendation, Group 2

PAGE 37

CHAPTER 5 METHODS Social Network Networks are sets of connected units such as people, organizations, or states. Each network is made of relational ties. Any form of interaction between two people can be considered a relational tie; a tie indicates that a person has some knowledge of the other person. A relational tie can be consider ed talking, colleges, friendship, partnership, marriage, adultery etc. Each person can belong to many different networks and there is no limitation to the number of groups anyone can belong to. In a network diagram a node is an individual actor within a network; an example of this are the red circles in the network diagram shown in Figure 5-1 The relational ties between actors or nodes are represented by the arrows from one circle to another circle. Each of these lines shows a relationship between two actors, and can be on e-directional or two. For example, person 22 indicates they know person 13, so there is an arrow from 22 to 13, but 13 didnt indicate they knew 22, so there is no arrow pointing towards 13. When looking at actors 17 and 36, there are arrows at both ends, indicating they both knew each other. Networks are cohesive, densely knit, and tightly bound. The networks that will be st udied in this research are considered whole networks as opposed to personal networks. Within each network there are different variab les that measure the strength and centrality of networks. These variables are represented by different measures su ch as core, between, degree, and closeness. Core is defined as being part of the cent ral group or not. An example of core would be a high school clique A high school class, you could identify students in a clique, or not in a clique. Those in the clique would be in the core, the others would be out of the core. There can be more than one clique with in a group, and they can even overlap. Figure 5-2 is an 37

PAGE 38

example of core membership w ithin the group; the circles are inside the core group while the triangles are not inside the core. The triangl es know each other and only one knows someone in the core group, represented by a tie. Betweenness is another way to measure the centrality of a group or network. Betweenness is defined as the measure of the extent to which a node lies on the geodesics between each other pair of nodes (Wellman, 2008). Betweenness ca n help explain how centrally located an individual is within a closed network, whether the actor is far away fr om the node that ties a group together such as the pres ident of a group. The betweenne ss score increases the farther away this person is from the center of the gr oup. A visual example of this is shown in Figure 53. The dot in the middle is an actor with a sma ll betweenness score, because in this case the person is the center of the group. Node 8 in Figure 5-4 has a high betweenness score because there are more nodes (or people) be tween this individual and everyone else in the network. This is because the actor is farther away from the cente r of the group but still connected to the center group. Degree is defined as the number of connections a node or actor has within a network. As degree increases the number of actors a node is tied to relationally increase s. An example would be the more people someone knows within the network the higher their degree would be. Figure 5-4 shows that node 7 has the highes t degree because it has the most connections to actors within the group. Closeness is the sum of the distance to each pe rson within the network. Therefore people with a large closeness sc ore are on the outside of the group, not di rectly tied to others within the group. Nodes 1 and 2 are an example of this variable in Figure 5-4 with the highest closeness score of the group because they are on the outside with a small amount of connections. Node 4 38

PAGE 39

and 5, 1 and 2, and 3 and 6 are all structurally equivalent, meaning they have the same amount of ties to other actors. Node 10 is the most peripheral with the leas t connection of all. Empirical Models The study uses a discrete choice variable over an ordinary least square model because the discrete choice variable is broke n into three categories and will provide more insight into the explanation of the re spondents choice. Following the basic framework in the literature3, the probit model starts with the specification of a latent form for a binary c hoice model that involves two (mutually exclusive and collectively exhaustive) alternatives. In th is case, I am interested in the consumers willingness to try new products. The latent form for individual consumer i is: (1) ),0(~ '* N XYi iii where Yi is the unobserved latent variable for individual i, Xi is Kx1 vector of explanatory variables, is the corresponding parameter vector, and i is a stochastic error normally distributed with mean zero and a variance-c ovariance structure that is uncorrelated and homoskedastic: (2) nI E2'where is a NxN matrix of variances and covariances, 2 is the variance of the error, and In is an identity matrix. Then, the indicator for the latent variable is (3) otherwise y Xifyi i i i 0 1 3 See Greene (2008); Maddala (1983); Wooldridge (2006). 39

PAGE 40

where, for the case of willi ngness to try new products, ( Yi ) will be equal to one if the consumer tried a new product and zero otherwis e. The probability that the late nt variable is greater than zero is expressed as: '1 1 0* i i i i i i iX X XP XPYP (4) where is the cumulative density function (cdf ) of the standard normal distribution. To estimate this model, the maximum likelihood function is derive d using the underlying assumption that each observation is drawn inde pendently from a Bernoulli distribution with some success probability. Th e log-likelihood function is (5) '1log1 logi i i iX yXy LogL This function is maximized using the ma ximum likelihood estimator (MLE) to obtain estimates of the parameters of interest (). Note that the assumpti on of independent draws is consistent with the specification assumed for the variance covariance matrix of the error term in (5). The goodness of fit measure examines how well the regression line fits a set of data. There are positive and negati ve estimates of i, these residuals are best estimates when they are close to the regression line. Ordered Probit The study uses an ordered probit over a probi t model because with the ordered probit the results show more explanation of a respondents choice. A probit woul d require collapsing the three categories into two, for a binary response taking away insight into respondents choices. 40

PAGE 41

An ordered probit gives more information for the three categories to explain a respondents change in willingness to try each of the three new products. The ordered probit model is used when y is an ordered response, then the values we assign to each value of y are no longer arbitrary (Wooldridge, 2002) The rating scale conveys useful information and even though we can not say that the difference between four and two is twice as important as the difference between one and zero. The ordered probit model for y can be derived from a latent variable model (Wooldridge, 2002). For the standard normal assumption for e, we derive the conditional distribution of y given we simply compute each response probability: x; (5-1) P( y=0 x) = P( y* x) = P (x +e 1 x) = (1 x ) The ordered probit model does not contain an intercept in the formulation of the model. The parameters and can be estimated by maximum likelihood. With the ordered probit models we are interested in the response probabilities P(y = j x) P(y = j x) is unambiguously determined by the sign of k does not always determine the di rection of the effect for the intermediate outcomes, 1,2,.,J-1 (5-2) (Wooldrige, 2002). When trying to estimate the partial effects of the xj is due to the strong assumptions that y* given x satisfies the classical linear model assumptions without these assumptions, the ordered probit estimator of would be inconsistent (Wooldridge, 2002). Selectivity Bias In this study the groups of st udents were not randomly sele cted, and they only slightly compare to the demographic breakdown of the Un iversity of Florida po pulation. The three groups in this study when combined are approximate ly equal percentages of the male and female student population at the University of Florid a. The race demographi cs are, however, not representative of the student popul ation at the University of Flor ida. Therefore, due to these 41

PAGE 42

results the conclusions made by th e study cannot be translated to or drawn across to the entire student body at the University of Florida. Limitations There are limitations to the re search on closed social groups. One of the main limitations to this research is that it cannot be trans posed or compared to the entire population of undergraduate students across the Un ited States. Another limitation to this research is that the data is composed of specific subgroups of th e undergraduate population at the University of Florida and therefore the results can only be generalized to the underg raduate population at the University of Florida. Because the study is not a representative sample of the undergraduate population the data results are lim ited to the study participants. 42

PAGE 43

Figure 5-1. Social network visualization Figure 5-2. Core membership 43

PAGE 44

44 Figure 5-3. Visual of betweenness Figure 5-4. Network vi sual of variables

PAGE 45

CHAPTER 6 EMPIRICAL RESULTS Social Network Model The first stage of analysis consisted of anal yzing the data with Uc inet (Borgatti, 2002). After collecting the data, the first step was to clean and transpose the data into matrices which Ucinet could understand. Within each group 4 diffe rent variables were cal culated to explain the relationships between group members. These were Freeman degree, Freeman Betweenness, Closeness and Core/Periphery. The network diagram for groups 1 and 2 are shown in figures 6-1 and 6-2 In these figures, each actor is represented by a node. Each line drawn out from a node represents a tie to another actor within the group, a nd the arrow at the end of the line denotes the direction of the relationship. Group 1 consists of 16 members and group 2 has 13 members. Figures 6-4 and 6-5 show the betweeness score of each actor in the groups. The betweenness score is defined as to help explain how centrally located an individual is within a closed network, whether the actor is far away from the node that ties a group together. The President of a group would be expected to have a high betweenness score as many people would indicate a knowledge of this person, compared to a newcomer who might know only one or two people, therefore having a lower betweenness score. As the be tweenness scores decrease, the actor is further to the out side of the group. In Figure 6-4 the size of the square is representative of the size of the betweenness variable. For group 1, betweenness scores ranged from 0 (3 people with relatively few ties) to 4.303 (the largest square in the diagram). Within group 2, many of the ac tors received the same betweenness score as they were nearly all equally connected to each other. The range of scores for betweenness for this group was only 0.00 to 0.339. 45

PAGE 46

Closeness is the sum of the distance to each pe rson within the network. Therefore people with a small closeness score are on the outside of the group, not direct ly tied to othe rs within the group. Figures 6-7 and 6-8 show the closeness scores for each group. The range of closeness scores for group 1 is 6.250 (two actors with few ties to the group) to 100.0 (3 actors, including the one with the highest betweenness score). For group 2, closeness scores ranged between 85.74 to 100.00. Again, as group 2 indicated many ties to each other, many members had the highest closeness score of 100.0 (10 of 13 actors). Degree is represented in figures 6-10 and 6-11 Degree is defined as the number of connections a node or actor has within a network. As degree increases the number of actors a node is tied to relationally increases. Degree shows that as relationships within the group increase the score of each actor wi th more relationships increase. Again, for group 1, we see the center node with the highest degree for group 1 (255.0), but degree defines the difference between the two outliers, where one has a degree of 3.0 (only has a few connections) compared to the other with 14.0 (low compared to most, but with more connections than the person in the top left of the diagram. The range for degree in group 2 is smaller, from 40.00 to 145.00. The final variable calculated is core membership. Of the 16 members of group 1, seven are members of the core and nine are not. In gr oup 2, 6 of the 13 member s, are in the core. Econometric Results This study was conducted to determine if there are social network variables that can help explain a persons willingness to tr y a new food product. In order to determine this, one model was estimated, using network variables and demo graphic variables. The model was estimated three times, one for each of the th ree products being tested (candy ba r, sandwich, and restaurant). These three equations were repeat ed for each of the two groups studied, resulting in 6 models being estimated. The ordered probit model will be estimated using two different samples for 46

PAGE 47

three different products (candy, sandwich and rest aurant) resulting in 6 estimates or models. In all 6 models, the dependent variable was the di fference between the willingness to try before and after the recommendation. The model was: Diffx = f (knowledge, trust, gender, primary shoppe r, own closeness, own betweenness, own core, own degree, closeness, betweenness, core, degree) Where Diffx is the difference in willingness to try a new candy bar, sandwich or restaurant after a recommendation. Gender is defined as the relationship between recommender and respondent. Trust is defined as the trust leve l the respondent has give n the recommender. Knowledge is defined as the know ledge level the respondent has given to the recommender. Primary shopper is a dummy variable representing if the respondent is the primary shopper of the household or is not. Own closeness, own be tweenness, own core, and own degree are the network variables representing the respondents position in the network. Closeness, betweenness, core, and degree are the networ k variables representing the recommenders position in the network. These models were estimated using Limdep 6 statistical software (G reene, 1998). Results are presented in Tables 6-3 through 6-9 For the candy bar, the actual model pred icted the impact of the recommendation on willingness to try better than the nave prediction in both group 1 and 2. For group 1, the model predicted the dependent variab le correctly 78.84 % of the tim e (compared to 44.97 % nave prediction). For group 2, the model was correct 80.77 % of the time (compared to 55.13 % nave prediction). For the sandwich, the model predicted the impact of the recommendation on willingness to try better than the nav e prediction. For group 1, the actual model wa s correct 80.95 % of the 47

PAGE 48

time (compared to 50.79 % nave prediction). For group 2, the model was correct 78.85 % of the time (compared to 51.92 % nave prediction). For the restaurant, the model predicted the imp act better than the na ve prediction for both groups. Within group 1, the model predicted 67.2 % correct, while the nave prediction was only 39.68 % correct. In group 2, the model pr edicted 63.46 % correct versus 38.46 % nave prediction. In both groups and all products the model pr edicted better than the nave prediction. As with the candy and the sa ndwich models, model one pred icted the impact of the recommendation on willingness to try the restaura nt better than both model 2 and the nave prediction. In group 1, the model with the network variables was co rrect 60 % of the time (compared to 40 % nave prediction),while the m odel without network variables was only correct 58 % of the time. For group 2, the model with network variables was co rrect 61 % of the time (compared to 38 % nave prediction), but the mode l without network variab les was only correct 46 % of the time. There were three categories to choose from within each model, category 0 indicated a decrease in willingness to try. Category 1 i ndicated no change in willingness to try the new product. Category 2 indicated an increase in willingness to try the new product. Within the candy bar equation, there is a pred icted probability of 0.0049 that category 0 is chosen within group 1. There is a predicted probability of 0.7054 that category 1 is chosen within group 1. Thirdly within Group 1 there is a predicted pr obability of 0.2946 that category 2 is chosen. The predicted probabilities for Group 2 are 0.0120, 0.3208, and 0.6672 for the three categories respectively. 48

PAGE 49

Within the sandwich equation the predicte d probability was 0.2628, 0.5402, and 0.1970 that each of the three categor ies was chosen respectively with in Group 1. Group 2 predicted probability was as follows 0.0910, 0.8356, and 0.0734 for each category respectively. The restaurant equation had predicted pr obability of 0.1470, 0.0964, and 0.7566 within Group 1 for each category respectively. Group 2 had predicted probabilities of 0.2186, 0.3282, and 0.4532 for each category respectively. Candy Bar Equation Results Results for the equations included for the di fference in willingness to try the candy bar after recommendation are presented in tables 6-3 and 6-4 for groups 1 and 2 respectively. The dependent variable included three categories of responses to the recommendation to try a new candy bar. In the first categor y (Y=0), respondents decreased their willingness to try the candy new bar. In the second category (Y=1), re spondents willingness to try the new candy bar remained the same and in the third category (Y =2), respondents willingness to try the new candy bar increased after the recommendation. The four non-network variables included in each equation were gender, primary shopper, trust, and knowledge. Gender was not found to be statistically significant in either of the two groups. Respondents trust in th e recommender was statistically si gnificant and positive in both of the two groups. Respondents knowledge of recommender was found to be statistically significant for Group 1. Respondents trust in the recommender was st atistically significant for all groups, except for Group 1 at y=0. If a respondent indicated th ey trusted the recommender, the respondent was 39% less likely to have no change in willingness to try a new candy bar and 41% more likely to increase their willingness to try a new candy bar based on th e recommendation. Respondents were 2% less likely to decrease their willingness to try and 21% less likely have no change in 49

PAGE 50

their willingness to try a new candy bar base d on a recommendation within Group 2. Group 2 respondents were 24% more likely to increase their willingness to try a new candy bar based on the recommendation. Respondents knowledge of the recommender was al so statistically significant for group 1, respondents likelihood to maintain or increas e their willingness to try the candy bar. Respondents were 16% more likely to have no change and 17% less lik ely to increase their willingness to try the candy bar if they knew the recommender. Two groups of network variables were included as independent variables in the models. One represented the respondents network position; the second represented the recommenders position. In both groups, the re spondents own degree was statistically significant, while betweenness was statistically si gnificant in Group 1 and closen ess and core membership were statistically significant in Group 2. Though th e respondents own degr ee was statistically significant in both groups, the marginal effects were 1% or less for all categories in Group 1. Each unit increase in the degree score increased the likelihood the respon dent would not change their willingness to try the candy bar by 2% in Group 2 and decreased their likelihood they would increase their willingness to try the ca ndy bar by 2%. For each unit increase in the respondents betweenness score in Group 1, they were 19% more likely to have no response to the recommendation and 20% less li kely to increase th eir likelihood to try the new candy bar. For each unit increase in the respondents own cl oseness score in Group 2, they were 1% less likely to have no response to the recommendati on and 1% more likely to increase their willingness to try the candy bar. Respondents own betweenness proved statistically significant within Group 2. In Group 2, if the respondent was in the core, their likelihood to decrease their willingness to try decreased by 4% and their likelihood to have no response to the 50

PAGE 51

recommendation decreased by 37%. Members of th e core were 42% more likely to increase their willingness to try the new candy bar. Th e second set of network variables described the recommenders place in the network. Closeness, betweenness, degree, and core were not statistically significant for either group. Sandwich Equation Results Results for the model with network variables included for the explana tion of difference in willingness to try based on a recommendation of a new sandwich are presented in tables 6-5 and 6-6 The dependent variable incl uded three categories of responses to the recommendation to try a new sandwich at a fast food restaurant. In th e first category (Y=0), re spondents decreased their willingness to try the new sandwich. In the second category (Y=1), respondents willingness to try the new sandwich remained the same. In th e third category (Y=2), respondents willingness to try the new sandwich increased. The results for the sandwich model were di fferent than the candy model; however, the results for the non-network variab les were largely similar. Ge nder was again not statistically significant in either group. Trus t in the recommender was statisti cally significant in both groups, as with the candy bar model, however knowledge of the recommender was only statistically significant for the second group in the sandwich models. Primary shopper was again significant for the second group only. In Group 1, if the res pondent indicated they trusted the recommender, they were 17% less likely to decrease their wil lingness to try the sandwic h and 20% more likely to increase their willingness to try. In Group 2, if the respondent indicated they trusted the recommender, they were 3% less likely to de crease their willingness to try the sandwich and 19% more likely to increase their willingness to try. The knowledge of the recommender was statistically significant within Group 2, with respondents 3% less likely to decrease their willingness to try a new sandwich from a fast food restaurant 19% more likely to increase their 51

PAGE 52

willingness to try a new sandwich. If the res pondent was a primary shopper in Group 2, they were 19% less likely to decrease their willingne ss 27% more likely to increase their willingness to try the sandwich. For the network variables representing the re spondent, the results were different for the sandwich ( Tables 6-5 and 6-6 ) compared to the candy bar. Own-closeness and core membership were still statistically signif icant for Group 2, however, othe r network variables were not statistically significant for Group 2. The only network variable that was statistically significant for Group 1 was core membership. For each one unit increase in the respondent s own closeness score within Group 2, respondents were 1% less likely to have no change in their willingness to try a new sandwich and 1% more likely to increase their willingness to try a new sandwich. If a respondent was a member of the core, they were 11% and 4% less likely to decrease their willingness to try the sandwich in groups 1 and 2, respec tively. Members of the core were 12% and 25% more likely to increase their willingness to try (Groups 1 a nd 2 respectively). As with the candy model, the second set of network variables, describing the referrers positi on in the network, were not statistically significant. Difference in Restaurant The results for the model with network vari ables included to expl ain the difference in willingness to try a restaurant based on a recommendation are presented in tables 6-7 and 6-8 The dependent variable included three categories of responses to the recommendation to try a new relatively expensive sit down restaurant. In the first category (Y=0 ), respondents decreased their willingness to try the new restaurant In the second cate gory (Y=1), respondents willingness to try the new restaurant remained the same. In the third category (Y=2), respondents willingness to try the new restaurant increased. 52

PAGE 53

As with the sandwich and candy bar models, gender was not si gnificant in either group. Respondents trust in recommender was statistically significant for Group 1 only. As trust increased, the respondent was 24% less likely to d ecrease their willingness to try a new relatively expensive sit down restaurant, and 24% more lik ely to increase their willingness to try. Knowledge of the recommender was statistically significant within Group 2 only, similar to the candy bar model. Respondents were 18% less lik ely to decrease their willingness to try the restaurant in Group 2 as knowledge of the reco mmender increased by one unit. In Group 2, respondents were 8% more likely to not respond to the recommendation and 10% more likely to increase their willingness to tr y the restaurant. If the respond ent was the primary shopper, they were 13% and 42% less likely to decrease their w illingness to try a new relatively expensive sit down restaurant in groups 1 and 2, respectively. Members of Group 1 were 10% more likely to increase their willingness to try a new relatively expensive sit down restaurant. Respondents from Group 2 were 25% more likely to not res pond to the recommendation and 17% more likely to increase their willingness to try a new relatively expensive sit down restaurant. Results for the network variables were again di fferent for the restaurant model. Closeness was statistically significant for both groups, while betweenness and core membership were statistically significant for Group 2, but not for Group 1. Degree was not statistically significant for either model. Respondents within Group 1 were 4% more likely to decrease and 4% more likely to increase their willingness to try a new relatively expensive sit down restaurant with each unit increase in their closeness score. The respondents own closeness within Group 2 was statistically significant as well, though the imp act of these effects was less than 1% for all categories. For Group 2, as a respondents be tweenness score increased, they were 177% less likely to try a new relatively expensive sit down restaurant when there is a unit decrease in 53

PAGE 54

betweenness. Respondents are 75% more likely to not respond and 103% more likely to increase their willingness to try a new relatively expensiv e sit down restaurant w ith an increase in the betweenness score. Members of the core in Group 2 were 46% less likely to decrease their willingness to try the restaurant and 13% more likely to have left th eir willingness to try unchanged. Members of the core were 33% more likely to increase their willingness to try a new restaurant. Like the candy bar and sandwich models, the third set of variables described the location of the recommender within the social network. Cl oseness, betweenness, degree, and core were not statistically significan t in Groups 1 and 2. Finally, a Wald test was run to determine if the addition of the network variables was helpful in determining the possible outcomes. Th e Wald test is a way of testing the significance of particular explanatory vari ables in a sta tistical model (Kyngas, Rissanen, 2001). If a particular group of explanatory va riables test significant, the conclusion is that parameters associated with these variables ar e not zero, so that variables should be included in the model. If the Wald test group of variables is not signifi cant then these explanat ory variables should be omitted from the model (Kyngas, Rissanen, 2001). The Wald test was run on the network variables within each of the six models. Table 6-10 is a representation of the outcomes. For Group 1 the variables were signif icant within the equation difference in candy bar. Within Group 2, these variables were significant in each of the three equations at a 99% level. 54

PAGE 55

Figure 6-1. Group 1 Network Figure 6-2. Group 2 Network 55

PAGE 56

Figure 6-3. Group 1 Betweenness Figure 6-4. Group 2 Betweenness 56

PAGE 57

Figure 6-5. Group 1 Closeness Figure 6-6. Group 2 Closeness 57

PAGE 58

Figure 6-6. Group 1 Degree 58

PAGE 59

Figure 6-7. Group 2 Degree Table 6-1. Model prediction nave prediction1 model prediction1 nave prediction2 model prediction2 Group 1 diff_candy 41.27% 65.61% 41.27% 65.61% diff_sand 50.79% 70.37% 50.79% 68.25% diff_rest 39.68% 59.79% 39.68% 57.67% Group 2 diff_candy 55.13% 78.85% 55.13% 57.05% diff_sand 51.92% 74.36% 51.92% 59.62% diff_rest 38.46% 60.90% 38.46% 46.15% 59

PAGE 60

Table 6-2. Difference candy regression results, Group 1 Coefficent Standard Error P Gender1 -0.12 0.21 0.57 Refer Trust 1.25 0.15 0.00 Refer knowledge -0.52 0.17 0.00 Primary shopper 0.12 0.27 0.66 Own closeness 0.07 0.08 0.42 Own core -0.01 0.29 0.98 Own betweeness -0.59 0.20 0.00 Own degree 0.01 0.01 0.00 Closeness -0.01 0.10 0.95 Betweenness -0.10 0.19 0.60 Core -0.16 0.31 0.61 Degree 0.01 0.01 0.10 Marginal Effects Y=00 Y=01 Y=02 Gender1 0.00 0.04 -0.04 Refer Trust -0.02 -0.39* 0.42* Refer knowledge 0.01 0.16* -0.17* Primary shopper -0.00 -0.03 -0.04 Own closeness -0.00 -0.02 0.02 Own core -0.00 0.00 -0.00 Own betweeness 0.01 0.19** -0.19* Own degree -0.00 -0.00** 0.01* Closeness 0.00 0.00 -0.00 Betweenness 0.00 0.03 -0.03 Core 0.00 0.04 -0.05 Degree -0.00 -0.00 0.00 Number of Observations 189.00 Log likelihood function -94.89 60

PAGE 61

Table 6-3. Difference candy regression results, Group 2 Coefficent Standard Error P Gender1 0.19 0.26 0.45 Refer Trust 0.74 0.18 0.00 Refer knowledge 0.25 0.19 0.19 Primary shopper 0.38 0.30 0.20 Own closeness 0.04 0.01 0.00 Own core 1.32 0.28 0.00 Own betweeness -1.05 1.14 0.35 Own degree -0.06 0.02 0.00 Closeness -0.00 0.01 0.91 Betweenness 1.01 0.98 0.30 Core -0.25 0.26 0.34 Degree -0.00 0.01 0.93 Marginal Effects Y=00 Y=01 Y=02 Gender1 -0.01 -0.06 0.06 Refer Trust -0.02** -0.21* 0.24* Refer knowledge -0.01 -0.07 0.08 Primary shopper -0.01 -0.10* 0.12 Own closeness -0.00** -0.01* 0.01* Own core 0.03 0.30 -0.34 Own betweeness -0.04* -0.37* 0.42* Own degree 0.00*** 0.02* -0.02* Closeness 0.00 0.00 -0.00 Betweenness -0.02 -0.29 0.32 Core 0.00 0.07 -0.08 Degree 0.00 0.00 -0.00 Number of observations 156.00 Log likelihood function -75.37 61

PAGE 62

Table 6-4. Difference sandwich regression results, Group 1 Coefficent Standard Error P Gender1 -0.00 0.20 0.99 Refer Trust 0.96 0.13 0.00 Refer knowledge -0.07 0.15 0.63 Primary shopper -0.19 0.26 0.45 Own closeness -0.07 0.08 0.38 Own core 0.59 0.28 0.03 Own betweeness 0.11 0.19 0.56 Own degree -0.01 0.01 0.22 Closeness 0.04 0.10 0.67 Betweenness -0.07 0.17 0.69 Core 0.23 0.28 0.42 Degree 0.00 0.01 0.83 Marginal Effects Y=00 Y=01 Y=02 Gender1 0.00 0.00 -0.00 Refer Trust -0.17* -0.03 0.20* Refer knowledge 0.01 0.00 -0.01 Primary shopper 0.03 0.01 -0.04 Own closeness -0.01 0.00 -0.01 Own core -0.11** -0.02 0.12** Own betweeness 0.02 -0.00 0.02 Own degree 0.00 0.00 -0.00 Closeness -0.01 -0.00 0.01 Betweenness 0.01 0.00 -0.01 Core -0.04 -0.01 0.05 Degree -0.00 -0.00 0.00 Number of Observations 189.00 Log likelihood function -114.01 62

PAGE 63

Table 6-5. Difference sandwich regression results, Group 2 Coefficent Standard Error P Gender1 0.38 0.27 0.17 Refer Trust 0.98 0.21 0.00 Refer knowledge 0.99 0.24 0.00 Primary shopper 2.00 0.37 0.00 Own closeness 0.06 0.01 0.00 Own betweenness 0.79 1.07 0.46 Own core 1.23 0.30 0.00 Own degree -0.04 0.02 0.02 Closeness 0.00 0.01 0.54 Betweeness -0.62 1.01 0.54 Core -0.21 0.27 0.44 Degree 0.00 0.02 0.77 Marginal Effects Y=00 Y=01 Y=02 Gender1 -0.01 -0.06 0.08 Refer Trust -0.03** -0.16 0.19* Refer knowledge -0.03** -0.16 0.19** Primary shopper -0.19* -0.08 0.27* Own closeness -0.00** -0.01 0.01* Own betweenness -0.02 -0.12 0.15 Own core -0.04** -0.21 0.25* Own degree 0.00 0.01 -0.01 Closeness -0.00 -0.00 0.00 Betweenness 0.02 0.10 -0.12 Core 0.01 0.03 -0.04 Degree -0.00 -0.00 0.00 Number of Observations 156.00 Log likelihood function -70.94 63

PAGE 64

Table 6-6. Difference restaura nt regression results, Group 1 Coefficent Standard Error P Gender1 0.06 0.19 0.73 Refer Trust 0.90 0.13 0.00 Refer knowledge -0.19 0.15 0.20 Primary shopper 0.44 0.25 0.07 Own closeness -0.16 0.09 0.06 Own core 0.02 0.26 0.94 Own betweenness -0.05 0.18 0.76 Own degree 0.00 0.01 0.96 Closeness 0.03 0.10 0.80 Betweenness 0.06 0.17 0.73 Core -0.13 0.27 0.62 Degree 0.01 0.01 0.26 Marginal Effects Y=00 Y=01 Y=02 Gender1 -0.02 0.00 0.02 Refer Trust -0.24* 0.00 0.24* Refer knowledge 0.05 -0.00 -0.05 Primary shopper -0.13*** 0.03 0.10** Own closeness 0.04** -0.00 -0.04** Own core -0.01 0.00 0.01 Own betweenness 0.01 -0.00 -0.01 Own degree -0.00 0.00 0.00 Closeness -0.01 0.00 0.01 Betweeness -0.02 0.00 0.02 Core 0.03 -0.00 -0.04 Degree -0.00 0.00 0.00 Number of Observations 189.00 Log likelihood function -130.09 64

PAGE 65

Table 6-7. Difference restaura nt regression results, Group 2 Coefficent Standard Error P Gender1 -0.16 0.23 0.49 Refer Trust 0.09 0.16 0.56 Refer knowledge 0.58 0.17 0.00 Primary shopper 1.23 0.27 0.00 Own closeness 0.02 0.01 0.00 Own coreness 1.65 0.27 0.00 Own betweeness 5.78 1.00 0.00 Own degree -0.01 0.02 0.67 Closeness 0.01 0.01 0.47 Betweenness 0.38 0.87 0.66 Core -0.27 0.23 0.26 Degree 0.01 0.01 0.56 Marginal Effects Y=00 Y=01 Y=02 Gender1 0.05 -0.02 -0.03 Refer Trust -0.03 0.01 0.02 Refer knowledge -0.18* 0.08** 0.10* Primary shopper -0.42* 0.25* 0.17* Own closeness -0.01* 0.00** 0.00* Own coreness -0.46* 0.13 0.33* Own betweenness -1.77* 0.75** 1.03* Own degree 0.00 -0.00 -0.00 Closeness -0.00 0.00 0.00 Betweenness -0.12 0.05 0.07 Core 0.08 -0.04 -0.05 Degree -0.00 0.00 0.00 Number of observations 156.00 Log likelihood function -102.18 65

PAGE 66

66 Table 6-8. Significant variab les in regression models Variable Group 1 Candy Group 2 Candy Group 1Sandwich Group 2Sandwich Group 1 Restaurant Group 2 Restaurant Gender + Refer Trust + + + + + Refer Knowledge + + + Refer P. Shopper + + + + Own Closeness + + + Own Betweenness + Own Degree + Referrer Close Referrer Between Referrer Core Own Core + + + + Referrer Degree Table 6-9. Wald Test Group 1 Equation Chi-Square P-Value Candy 15.46 0.05 Sandwich 6.77 0.56 Restaurant 10.64 0.22 Group 2 Equation Chi-Square P-Value Candy 56.10495 0.00 Sandwich 50.86239 0.00 Restaurant 71.1305 0.00

PAGE 67

CHAPTER 7 CONCLUSIONS The goal of this thesis was to investigate if the use of social network variables would improve the power of prediction of the respons e to recommendations a bout food products. In each of the six cases (3 food products across 2 groups), the network variables improved model fit. The model with networ k variables included descriptor s about the respondents and recommenders position within the network. Th ese variables were be tweenness, closeness, degree, and core membership. Two variables, the respondents trust in th e recommender, and the re spondents knowledge of the recommender were significant five out of six occasions (the exception was the model representing the restaurant fo r Group 2 for trust and the model representing the sandwich for Group 1). The relationship was positive, indicati ng that as the trust th e respondent had in the recommender increased, the likelihood to increas e their willingness to try the new food product (candy bar, sandwich, and restaurant) based on a recommendation in creased. This is expected as one is likely to place more credibility in info rmation obtained from those people in whom they have trust. Knowledge of the referrer was also significant in predicting a difference in the willingness to try food products, however, the way knowledge influenced response to recommendations varied. For Group 1, the rela tionship was negative, implying as knowledge of an individual increased, their response to the recommendation to try a product decreased. At first this seems to be against intuition, as one mi ght expect people to respond to recommendations from those they know more. However, this corresponds to netw ork literature that i ndicates people may respond more to those that are more different than they (that they may know less) as those people introduce new information to the respondent. However, this variable ha d a positive relationship 67

PAGE 68

in Group 2. Whether or not the respondent was the primary shopper was significant for Group 2 in all cases, but was only significan t in Group 1 in the restaurant model. Primary shoppers were designated if the respondent does fifty % or more of the shopping, and might be expected to play a role in the purchase of new products. In th e four cases where the marginal effects were significant, primary shoppers were more likely to respond to reco mmendations with an increase in willingness to try the products. When examining the network variables repr esenting the respondent s position in the network, betweenness and core membership were significant four out of six times. These variables were signifi cant in Groups 1 and 2. A responde nts own betweenness score had a negative relationship, meaning as the betweenness score gets smaller, the respondent is more likely to respond positively to recommendations a bout the candy bar. This indicates that the people with lower betweenness scores, those who more on the periphery of the groups, responded more positively to recommendations. A respondents own core membership was signifi cant three times in si x equations. In all three cases, core members responded positively to recommendations. As respondents are more on the outside of a group, the betweenness score is lower, the more influenced they are by others recomme ndations to try a new food product. Also, each respondent that falls into the core of a group is more likely to listen to recommendations from other members to try a new food product. Marketer s should target the core members of groups. The core members influence each others deci sions to respond to recommendations and will therefore act as a ripple affect and influence the outside members of the group. The marketing message that should be taken fr om this study is that any pers ons recommendation of a product from within a group can have a ripple effect on th e entire group message, therefore the marketers 68

PAGE 69

should target core group members and watch the ripple effect take place within a group to try a new food product. In summary, the network variables improved the ability of the model to predict a subjects willingness to try a new candy bar, sandwich from a fast food restaurant, or eat at a new relatively expensive sit down restaurant based upon a recommendation compared to a model without network variables. There have been a few lessons learned from this research that would be helpful for future research on this subject. One co mplication was with a third group of data that was attempted. In this group, there were very few respondents, with relatively little knowledge of each other. Due to this, some of the network va riables could not be used. For ex ample, the variable representing core membership would have been singular as it wa s impossible to identify a core in the group. In the future the groups need to be larger to allow for more dive rsity in the core membership. Additionally, as many respondents did not know each other, there were not many observations in the study, and the model could not be estimated (i t lacked degrees of freedom). With network data, each time one individual know s another, a row of data is created. Hence, small groups can create a large dataset as if ther e are ten members, and they each know each other, there would be 90 observations. But if there is a group of ten members, and they each only know one other person, there are only ten observations. The th ird group where data collection was attempted was a class instead of a social organization. This proved to be difficult because there was less incentive for the members to respond, and they were less likely to know each other. Hence, for future studies, one recommendation would be to focus on using social or ganizations instead of classes. 69

PAGE 70

Though this data was not repres entative of the University of Florida, or any group of students, and therefore can not be extrapolated, we still learned much from the research. This data collection proved very informative on findi ng which network variables help explain a respondents willingness to try a new food product. The data that was collected from social organizations was easier to coll ect and provided more statistica lly significant results of the network variables with more variation. The network data does not need to be collected on both the recommender and the respondent. The variables that we re useful were the respondents network variables, where they were compared to others in the network. Th e variables associated with the recommenders position in the network did not give any useful information in explaining a respondents willingness to try a new food product based on a r ecommendation. These variables may be able to be excluded from further re search when explaining a respo ndents willingness to try a new product based on a recommendation. Though there are things to be learned from this type of research, collecting data for network studies is more difficult than it appears. Originally, the goal had been to recruit more students from classes at the Un iversity of Florida. Data collection began in Fall 2007 and continued to Summer 2008. E ach semester one to two surveys were developed and data was collected. Likely due to the length of the surv ey, many of the attempts to collect data failed when most members of the network did not respond to the survey. Another issue with the data collection method being used was that the classe s were given no incentive to take the survey. Another difficultly with collecting this type of data is the size of the dataset. If there are 25 members in a class, there are 25*24, or 576 possible relationships. This causes two problems. First, each member must answer questions abou t each other. As the total number of members 70

PAGE 71

71 gets larger, the burden on the respondent (and time to complete) grow s exponentially, making it less likely to get completed obs ervations. Additionally, the burd en on the person working with the data grows. Finally, using a web survey prov ided a convenient way to collect data, but also created problems. If there was an issue with th e website, as was the case with certain internet browsers, respondents became frustrated and quit the survey. Second, without personally interviewing, the respondents had the opportunity to not complete the survey, making it difficult to get whole group responses.

PAGE 72

APPENDIX A WEB SURVEY Template for Web-Based Survey for Social Network Survey of Students and Food Choices Participants will link to a page that shows the in formed consent and asks if they are 18 years or older. By selecting continue, they agree to participate and the followi ng questions are asked: 1) Name __________________ 2) Gender: Male (0) Female (1) 3) Age: _____ 4) Are you the primary shopper for groceries in your household? (the primary shopper is the person responsible for at least 50% of food purchased for the hous ehold, alone or sharing the task with another household member.) Yes (1) No (0) 5) Please indicate your race: White (1) Hispanic (3) American Indian (5) Black or African American (2) Asian (4) Other_________ (6) 6) On a scale of 1 to 5, how likel y are you to purchase a new candy bar? Definitely would not buy Probably would not buy Might or might not buy Probably would buy Definitely would buy 1 2 3 4 5 7) On a scale of 1 to 5, how likely are you to purchase a new sandwich at a fast food restaurant? Definitely would not buy Probably would not buy Might or might not buy Probably would buy Definitely would buy 1 2 3 4 5 8) On a scale of 1 to 5, how likely are you to go to a new sit-down restaurant? Definitely would not go Probably would not go Might or might not go Probably would go Definitely would go 1 2 3 4 5 72

PAGE 73

73 Respondents will then be presented with a randomi zed list of student names from their course and asked the following questions: On a scale of 1 to 5; Please answer the following questions: 9) How well do you know this ( insert classmates name here in a randomized order ) on a scale of 1 to 5? 1= Do not know (skip to question 11) 2= know slightly (go to question 10) 3= know (go to question 10) 4= know well (go to question 10) 5= Know very well (go to question 10) 10) How well do you trust (insert classmates name here in a randomized order) ? 1= Do not trust 2= trust slightly 3= trust 4= trust well 5= Trust very well 11) If the following (insert classmates name here in a randomized order) recommended a candy bar, how likely would you be to purchase the candy bar? 1= Definitely would not purchase 2= Probably would not purchase 3= Might or mi ght not purchase 4= Probably would purchase 5= Definitely would purchase 12) If the following (insert classmates name here in a randomized order) recommended a sandwich, how likely would you be to purchase the sandwich? 1= Definitely would not purchase 2= Probably would not purchase 3= Might or mi ght not purchase 4= Probably would purchase 5= Definitely would purchase 13) If the following (insert classmates name here in a randomized order) recommended a restaurant, how likely would you be to go eat at the restaurant? 1= Definitely would not go 2= Probably would not go 3= Might or might not go 4= Probably would go 5= Definitely would go

PAGE 74

LIST OF REFERENCES Aldrich, J. H., & Nelson, F. D. (1984). Linear Probability, Logit, and Probit Models. A Sage University Paper. Borgatti, S. E. (2002). Ucinet 6 for Windows: Software for Soci al Network analysis. Harvard, MA: Analytic Technologies. Burt, R. (1987). Social Contagion and Innovation: Cohesion Vers us Structural Equivalence. The American Journal of Sociology 92 (6), 1287-1335. Buskens, V. (1998). The Soci al Structure of Trust. Social Network 20, 265-289. Coleman, J., Katz, E., & Menze, H. (1957). The Diffusion of an Innovation Among Physicians. S ociometry 20 (4), 253-270. Engel, J. E., Blackwell, R. D., & Kegerreis, R. J. (1969). How Information is Used to Adopt an Innovation. Journal of Advertising Research 9 3-8. Fisher, R., & Price, L. (1992). An Investigation into the Soci al Context of Early Adoption Behavior. The Journal of Consumer Research 19 (3), 477-486. Frozen Food Digest. (1997, July 1). All Business (San Fransisco, Ca) Retrieved August 15, 2008, from http://www.allbusiness.com/ma rketing/market-res earch/631186-1.html Granovetter, M. (1983). The Strength of Weak Ties: A Network Theory Revisted. Sociological Theory 1 201-233. Greene, W. (1998). LIMDEP. Version 6.0 BeUport, NY: Econometric Software, Inc. Gujarati, D. N. (2004). Basic Econometrics (Fourth ed.). New York: Tata McGraw-Hill. House, L. A., & House, M. C. (2007). Do Reco mmendations Matter? Social Networks, Trust and Product Adoption. EAAE Forum on Innovation and System Dynamics in Food Networks. Hummon, N. P., & Doriean, P. (2003). Some Dyna mics of Social Balance Processes: Bringing Heider Back into Balance Theory. Social Networks 25, 17-49. Kirby, J., & Marsden, P. (Eds.). (2006). Connected Marketing. Oxford: Elsevier. Kyngas H., R. M. (2001). Blackwell Science Ltd. Malden, Ma: Blackwell Publishing. Retrieved 10 24, 2008, from Blackwell Publishing: http://www.blackwellpublishing.co m/specialarticles/jcn_10_774.pdf Merriam-Webster. (2008, July). Merriam-Webster Dictionary. Retrieved June 2, 2008, from Merriam-Webster Online Dictionary: http:/ /www.merriam-webster.com/dictionary/Trust 74

PAGE 75

75 Orr, G. (2003, March 18). Stanford University. Retrieved June 2, 2008, from http://www.stanford.edu/class/symb sys205/Diffusion%20of%20Innovations.htm Peter, P. J., Donnelly Jr., J. H. (2003). A Preface to Marketing Management. Boston: McGrawHill Irwin. Reingen, P. H., Foster, B., Brown, J., & Seidma n, S. (1984). Brand Congruence in Interpersonal Relations: A Social Network Analysis. Journal of Consumer Research 11, 771-783. Reingen, P., & Brown, J. (1987). Social Ti es and Word of Mouth Referral Behavior. The Journal of Consumer Research 14 (3), 350-362. Reingen, P., & Kernan, J. (1986). Analysis of Referral Networks in Ma rketing: Methods and Illustration. Journal of Marketing Research 23, 370-378. Richins, M. L. (1983). Negative Word of M outh by Dissatisfied Consumers: A Pilot Study. Journal of Marketing 47, 68-78. Rogers, E. (2003). Diffusion of Innovation. New York: Free Press. Shaw, S. J. (1983). Behavioral Science Offe rs Fresh Insights on New Product Acceptance. Journal of Marketing 47, 68-78. Survey Crafter, Inc. (2008). Survey Crafter 4.0.8. Acton, MA, USA. Udry, C. R., & Conley, T. G. (2004, May). Social Networks in Ghana. Wellman, B. (1983). Network Analysis: Some Basic Principles. Sociological Theory 155-200. Wellman, B. (2008). Networks for Newbies. Sunbelt XXVIII. St. Petersburg. Wooldridge, J. M. (200). Econometric Analysis of Cross section and panel data. The MIT Press.

PAGE 76

BIOGRAPHICAL SKETCH Joy Mullady received her undergraduate degree fr om the University of Florida College of Agricultural and Life Sciences, with a major in food and resource economics and a concentration in international economics in 2006. Joy was born in Mt. Vernon, New York. At the age of two she moved to New Jersey after a transfer from he r fathers job. She moved to Texas at age 4, due to another transfer from her fathers job. By the age of eight my family had moved to Florida, where we currently live. She is one of the middle siblings, in a group of four. Her sister Maurine is the oldest and is a lawyer, who graduated with he r undergraduate degree from the University of Florida. Maurine is followed by her older brother William who is two years older; he currently lives in Washington D.C. and is atte nding the Art Institute of Design. James is the youngest; he is only 18 and is headed off to college this year at SUNY, Purchase College. Joy has currently resided in the stat e of Florida for the past 16 years. Joy was blessed to see many different states and travel around Europe for a summer as well. She was a research assistant intern whose projects included updating seafood statistics, buildi ng surveys, gathering data on Hispanic culture and more. She also interned at the Hillsborough County Extension Office as a Four H leader teaching Money Wise to childr en ages 8-11. Joy has been learning and exemplifying her leadership sk ills since a very young age. 76