DOI: 10.31038/MGSPE.2021112

Abstract

Respondents each evaluated 24 unique combinations of messages (vignettes) pertaining to the choice of a banquet hall, rating each vignette on a Likert scale for likelihood to choose the banquet hall for their next event, and then selecting one of five emotions to show their feeling about the banquet hall as described. The approach, Mind Genomics, revealed the part-worth contribution of each of 36 messages to both choice (question #1) and to feeling/emotion (question #2). Respondents fell into one of three different mind-sets, based upon their patterns of choice, MS1 – Practical aspects of hiring a banquet hall; MS2 – Focus on cost; MS3 – Focus on experience. Membership in the mind-sets did not correlate with any aspects of the respondent as determined by a self-profiling classification. The paper presents a PVI, personal viewpoint identifier, six questions on a 2-point scale, developed by the mind-set segmentation, which assign a new person to one of the three mind-sets.

Introduction

Anthropology and Sociology

The study of everyday events by psychologists, sociologists, anthropologists and even commercially- oriented market researchers, creates a foundation of knowledge about a society. Researchers observe from the outside, often talk to those in the society, and then describe what they see, often in very readable papers and books, occasionally in charts and statistics. Those who observe people in their own environment, talk to them, record these interactions, and perhaps even substantiate some of the information with charts and tables, provide a very readable account of the people as those people go about their daily lives.

The focus of this paper in on a very simple aspect of daily life, specifically hiring a banquet hall. Rather than presenting this topic from the point of view of statistics or economics such as how many halls are rented, whom, for what reasons, and so forth, we approach it from the perspective of the inner, personal aspect, the thinking which may go on in the mind of the prospect who is about to hire the hall.

Banquets and banquet halls present a plethora of interesting aspects to study. Banquets are ceremonial meals. A number of papers present to us the nature of the banquet in ancient times, as well as the role of the banquet and banquet hall today. This paper is not a review of the extensive literature about banquets and banquet halls other than to recognize the rich academic history that the public feasting has enjoyed, whether the academic work focuses on what happened across history [1,2], on the position of banqueting and banquet halls in society [3-5], and even on the nature of banqueting in different cultures [6-8].

There is a practical aspect as well, today’s economy. High-priced banquet activities at wedding receptions have significantly contributed to the growth in the overall profits of the food and beverage (F&B) departments of hotels [9]. Restaurants are, by their very frequency, more important economically than are banquets. Yet, banquets are critical for the hospitality world [10-14]. Almost 70% of the food and beverage revenue of hotels in the U.S. is generated by banquets. Fifty percent of these profits come from weddings in the United States.

When we turn to the topic of ‘banquet’ in the academic literature, searching through Google Scholar we have many more hits, about 51,000, but many of these are relevant to issues of service quality and customer satisfaction on the one hand, and historical aspects of banqueting on the other. Indeed, banqueting as a topic in and of itself is of less interest than banqueting as a social institution, and how it may fit into the complex of daily institutions in one’s culture, whether that culture be current around the world, or part of the daily events of historical times, such as banqueting during the days of the Roman empire, and how the behavior changed with history. There is very little about the topic of ‘banquet halls,’ however.

Despite the paucity of information about banquet halls (not banquets themselves!) in the academic literature, such as the importance of the different reasons for choosing a banquet hall, there are some relevant papers, such as Ling Guan’s master’s thesis in Iowa State University [15] on ‘Push and pull factors in determining the consumer’s motivations for choose wedding banquet venues: A case study in Chongging, China. Guan introduces the notion of two types of factors for the choice of venue; push and pull. four push factors (“seeking relaxation and knowledge”, “fulfilling prestige”, “escaping from daily routine”, and “social networking”) derived from the extracted 10 push items and six pull factors (“budget”, “atmosphere”, “facilities”, “wedding services”, “transportation”, and “service and quality”). Peng & Wang [16] also talk about the wedding banquet as a set of decision with different strategies for decision making.

The alternative to a rigorous academic study of the topic of banquet halls is a study of the commercial banquet hall, how the hall is used, how the hall is chosen, the position of the banquet hall in everyday life. One need only look at the local newspapers of a community, at the websites of business in that community, or just drive around the main streets to get a sense of the nature of the banquet facilities that are offered, how these are presented to the prospective customers. In the same spirit, one can request a market research study about attitudes and practices with regard to banquet halls, or even commission a focus group of banquet hall customers or prospects to discuss their experiences with, and their attitudes towards banquet halls. No doubt such commercially relevant studies have been done, not so much for publication as for business use, to know, for example, how to present one’s banquet hall to customers to increase business.

The Internet is an increasingly popular source of knowledge about daily people, especially because the content of the Internet matches the interests of those who search for information. Popular topics like health have hundreds of millions of sites. When we look for the phrase “banquet halls’ in Google, we end up with approximately 46,000 hits as of July 30, 2020. People also search for the practical aspects of banquet halls such as ‘what are the different types of banquets?, ‘In a banquet hall a good business, what is a banquet hall?’, How do banquet halls charge?’, What happens at a banquet?’, ‘What is a banquet menu?’, and so forth. These are the relevant questions. The websites which address these questions are usually commercial sites, providing an answer to the question, and then offering a particular banquet-related service to the person who is doing the searching.

A good idea of what can be found for in a search for a banquet is the following:

Banquet halls are among the most popular types of venues for events, particularly wedding. The popularity stems from the benefits these halls bring such as stress-free planning for families and couples. These banquet facilities are typically all-inclusive which means a lot of the little details are covered such as catering, seating, decorating and so much more.

https://www.uniquevenues.com/banquet-halls

Applying Mind Genomics to study what appeals to a person asked to choose a banquet facility

Banquets provide a topic which can contribute to the understanding of choice for celebratory events, events that are affordable and which move beyond the everyday dining behavior that one does at a restaurant. Studying the choice of options for a banquet brings into play the recognition of clearly different factors influencing that choice. The different aspects range across topics such as the nature of the information to which people pay attention when choosing a banquet, the influence of the person who is presenting the information, the income of the person giving the banquet, the choice of banquet amenities, and so forth. The reality is that the academic literature on both hospitality and choice, respectively, recognizes the importance of such information, but there is little to be found which treats the options of choice in a banquet hall in a rigorous fashion.

As the 21^st century progresses, and with the plethora of available methods to measure and to analyze, what is missing is a deep psychological understanding of the everyday. We can study the economics and sociology of banqueting, either by reading papers in the scientific literature, by immersing oneself in the veritable flood of advertisements in the popular press and Internet, or from going through the verbatims of in-depth interviews with consumers or the mountains of survey results run by market researchers. Yet, unless there is a sharp focus, one will obtain a great deal of ‘data’ but not understand the ‘mind’ of the customer of a banquet hall. The topic is simply too limited for today’s scientist.

What is needed is a fast, inexpensive, knowledge-creating system to understand any topic where judgment is key, where the data are scarce and where the experience is widespread. Mind Genomics is that key. Mind Genomics was borne out of the desire to understand the mind of the person as that person lives life, doing so in the spirit of Weber to study the ‘whole’ but also in the spirit of experimenting science. Mind Genomics is a branch of experimental psychology, one informed by consumer research in the world of the applied, informed by the statistics of experimental design, and inspired by the images of the MRI (magnetic resonance imagery) in medicine which takes pictures of tissues from different angles, and inspired by genomics which focuses what drives individuation [17-19].

Mind Genomics begins with notion that people may not know what they are thinking but will know It when they see it. The strategy of a Mind Genomics experiment is to combine elements, messages, descriptions of various aspects of the topic (banquet halls), create small vignettes whose composition is known, present the combinations to respondents, obtain a response to the combination, and then estimate the degree to which each element in the combination drives the response. The experiments are quick, affordable, informative, and archival. Most of all, the experiments are targeted to answer question in a direct fashion.

We show the use of Mind Genomics to study banquet facilities, beginning with the topic (Banquet facilities), moving through the selection of questions and answers, and then on to the actual study, the data and analysis, the discovery of mind-sets even within this limited world, and finally the creation of a tool to discover these mind-sets in the population. The demonstration study, which can be developed and run in less than a day, provides the potential to generate a ‘Wiki of the Mind’ for topics relevant to society, in different cultures, different situations.

Explicating the process through the study of choosing a banquet hall

Step 1 – Choose the topic, questions, and answers

The topic is banquet halls. In this particular version of Mind Genomics, we deal with the 6×6 design, comprising six questions, and six answers to each question. Table 1 shows the six different questions, and for each question six different answers. The questions and answers are left to the researcher(s). The questions and answers emerge after about 20-30 minutes of brainstorming among a group of four researchers who have never experienced Mind Genomics research before. Mind Genomics has been developed to promote rapid iterations, so one need not spend a great deal of time thinking about the questions and answers, and perhaps even overthink.

Table 1: The raw material for the Mind Genomics study of the banquet hall; six questions and the six answers to each question

The objective of Mind Genomics, the creation of the aforementioned ‘wiki of the Mind’ requires that the test stimuli be relevant for everyday life, and not simplistic statements of the type that one would not typically encounter. That is, the stimuli in Mind Genomics are of the type that would be appropriate for the quotidian, commercial and social aspects, rather than artificially created stimuli, manipulated so that the responses to these artificial stimuli would be able to support a hypothesis about behavior, or disprove it. Thus, one can look at the material in Table 1 in the light of an ethnographic report of the types of messages one might encounter in every-day life.

A key aspect of Mind Genomics is the reality that one need not be ‘right’ at the start of the Mind Genomics experiments. Whereas in most research the effort requires a great deal of planning, the selection of the ‘correct’ test stimuli, the appropriate scales, and the appropriate respondents, the science of Mind Genomics was created to more realistically simulate the exploration of the everyday, where one does not weigh the alternatives in a carefully considered manner, in the fashion called System 2 by Nobel Laureate Daniel Kahneman [20]. Rather, the typical approach is the more automatic approach, called System 1, or in the thought of Harvard psycholinguist George Miller, TOTE, Test, Operate, Test, Exit [21]. One need not be exact to study inexact behavior. One need only be systematic, consistent, affordable, fast, and scalable to create archivable, solid knowledge.

Step 2 – Combine the elements into short, easy to read vignettes, or test concepts, each vignette comprising a maximum of four elements, and a minimum of two elements

Figures 1 and 2 show the same vignette, comprising four elements. The elements are placed one atop the other, left justified, set up to be easy to read. No effort is expended to tie the elements together.

Figure 1: Example of a 4-element vignette, showing the first rating scale (likely to choose)

Figure 2: Example of the same 4-element vignette, showing the second rating scale (select the emotion)

The vignettes are created by an underlying statistical plan called an experiment design [22]. The design for the 6×6 (six questions, each with six answers), comprises 48 vignettes, each element appearing five times, and absent 43 times. A vignette can have at most ONE element or answer from a question. This particular property of incompleteness is necessary to ensure that the 36 elements are statistically independent of each other.

Each respondent evaluates a unique set of 48 vignettes, different from the set evaluated by the other respondents. This strategy is known as a permuted design [23], ensuring that the combinations different across respondents. The strategy is to uncover the underlying pattern in the data (viz., how the elements drive the responses) by testing many combinations only once and putting together the pattern from these many combinations. The strategy is similar to the manner of the MRI (magnetic resonance imagery), which takes many pictures of an object from different angles and combines these by computer to create a 3-dimensional image. The Mind Genomics strategy differs dramatically from the conventional approach of selecting a limited set of vignettes (e.g., 48), to represent all combinations, and then testing the same 48 vignettes with many respondents to average out random error.

Step 3 – Invite the respondents to participate, and collect the ratings

Each respondent is invited through an on-line panel company, specializing in these types of studies. It is important to work with a panel provider because otherwise the response rate is low, and the study may take a week to complete, rather than the more normal hour or two. Panel companies have lists of respondents who have agreed to participate. Most panel companies can tailor the list of respondents by a variety of criteria ranging from simple geo-demographics to self-stated attitudes and behaviors. For this study, the criteria were general, since the focus was on ideas appealing to the general public. Figure 3 shows the introduction to the vignettes, including the scales on which the vignettes will be rated.

Figure 3: Orientation screen, showing instructions to the respondents, the two scales, and the expected time that will be needed to complete the Mind Genomics experiment. The experiment is positioned as a ‘survey’.

Step 4 – Create the database, to prepare it for regression analysis, and for cluster analysis, respectively

The database for the Mind Genomics study comprises one row for each vignette, for each respondent, respectively. Each of the 57 respondents evaluated 48 unique vignettes, generating a set of 2,736 rows of data. The first 36 columns of data corresponded to the 36 elements, with the database showing the number ‘1’ for those elements appearing in the particular vignette, and the number ‘0’ for those elements absent from the particular vignette. Since a vignette could comprise 2-4 elements, only 2-4 cells contained the value ‘1’, and the remainder contained the value ‘0.’ The 37^th column contained the rating 1-9, and the 38^th column contained the choice of the emotion.

The rating scale, values shown in the 37^th column, was transformed to a binary scale, TOP3, following the practice in consumer research and public opinion studies. It is not easy to explain what is meant by any rating scale, unless each point is labelled. Researchers argue over the labelling of each point. An easier way is to bifurcate the scale into a binary scale, with the lower part of the scale (1-6) transformed to 0, and the upper part of the scale (7-9) transformed to 100. The ‘cut-point’ (6,7) is defined arbitrarily. The effect creates a ‘No/Yes’ variable, one easier explain. In order to ensure that it would be possible to run an OLS (ordinary least-squares) regression analysis on the 48 vignettes for each respondent, the Mind Genomics program added a very small random number (<10^-5), to ensure some variation in the dependent variable. Otherwise, the transformed ratings from a respondent might all end up 0 or 100, were the respondent to have confined the rating to the low end of the scale, 1-6, or to the high end of the scale.

The ratings in the 38^th column, the selection of emotions, were treated in a different fashion. There were five emotions. The rating scale is known as a ‘nominal scale,’ with the values 1-5 standing for different emotions. To prepare the data for analysis by OLS regression requires the expansion of the one scale into five new scales, one scale for each emotion, respectively. The expansion added five new columns of data. The respondent had to select one emotion for each vignette. The newly created variable corresponding to that emotion was coded as ‘100’ plus a very small random number. The remaining newly created variables corresponding to the four emotions not chosen were coded as 0 plus a very small random number. Thus, the five emotions were coded to immediately reveal which feeling/emotion was selected.

At this point the data matrix is ready for analysis by OLS (ordinary least-squares) regression, a well-accepted statistical technique, also colloquially known as curve-fitting. The OLS regression is executed in two steps, first to create 57 individual level models in preparation for clustering to reveal mind-sets, and then run with groups of respondents, defined by who they are (e.g., Total Panel, Gender, Age), and by how they think about the topic of banquets, here specifically three mind-sets. This paper presents only the data from the Total Panel, and from three mind-sets, respectively.

Step 5 – Create an individual-level model for each of the 57 respondents, using OLS regression, and then cluster the 57 respondents into two, and then three groups (mind-sets)

One of the continuing themes of Mind Genomics is that people differ from each other in the way they think about a topic. These different ways are called mind-set. Step 5 discovers these mind-sets.

The experimental design ensures that the 36 elements arrayed in the 48 vignettes are set up to allow for OLS regression, relating the presence/absence of the 36 elements to the binary response (TOP3), which was either 0 (rating 1-6) or 100 (rating 7-9). The OLS model was estimated without an additive constant, called forcing the regression through the origin. Estimating the coefficients without an additive constant will make it easier to cluster the respondents into either two mind-sets or into three mind-sets, respectively. The equation for the individual respondent is written as: Top3 = k₁(A1) + k₂(A2) … k₃₆(F6).

The OLS regression is run 57 times, one for each respondent. The preparation stage using OLS regression generates the necessary data matrix comprising 57 rows, one per respondent, and 36 columns, one per element. The matrix is then subject to cluster analysis [24]. The cluster analysis puts the respondents into two groups and then into three groups. Respondents in a cluster show similar patterns of coefficients. Cluster analysis simply provides the solutions but does not decide the number of clusters (mind-sets). It is the researcher who selects the number of mind-sets based upon two criteria. Criterion #1 is interpretability. The strongest performing elements must tell a coherent story. Criterion #2 is parsimony. It is better to have fewer clusters than more clusters, given equal interpretability.

Step 6 – Create the model for the Total Panel and for each of three cluster which emerge

The clustering in Step 5 suggested three groups, also known as mind-sets. Two groups or mind-sets were hard to interpret. The final analysis to understand what drives selection of the banquet hall comes from putting all of the RELEVANT data together into one file and running one grand OLS regression on the data. This time the regression equation has an additive constant whose purpose will be explained below.

Right after the OLS regression on the Total Panel come three separate OLS regressions, first on the data from all respondents assigned by cluster program to Cluster 1 (Mind=Set1), second on the data from all respondents assigned to Cluster 2 (Mind-Set 2), and third and finally, on the data from all respondents repsondents assigned to Cluster 3 (Mind-Set 3). The one equation for each OLS regression is expressed as: Top3 = k₀ + k₁(A1) + k₂(A2) … k₃₆(F6). The analysis below will reveal just how radically different are the mind-sets of these groups. The rationale for the additive constant is that in the absence of specific information about the banquet hall there is still a proclivity to choose the banquet hall. The additive constant measures that proclivity.

Step 7 – Create a model for the Total Panel showing the linkage of each element to every one of the five emotions (scale 2)

Recall that Step 4 above, preparing the data for OLS regression, mapped the second rating scale, selection of emotions, to five new binary scales, one per emotion. The data are ready for OLS regression. This time the dependent variables are the five newly created variables, one per emotion, with the values 0 or 100, depending upon the selection of the emotion. The independent variables are the 36 elements. The model is run without the additive constant, this time because in the absence of elements there is no proclivity to choose an emotion, and therefore the additive constant is by definition 0. The reality is that whether we include an additive constant or not, the coefficients of the elements will show the truly strong performing elements.

Step 8 – Create a PVI (personal viewpoint identifier) to assign new people to one of the three mind-sets

Study after study in the world of Mind Genomics reveals that the respondents in the same mind-set fall into different geo-demographic, behavioral, and attitudinal groups. The Mind Genomics study for banquet halls is such a narrow topic that we would not expect the mind-sets to distribute by gender, age, and so forth. Thus, there needs to be a way to assign a new person to one of the mind-sets which emerge, if only to improve the nature of the interaction between sales/manager of the banquet hall and customer hiring the services of the banquet hall for a particular event. The topic of banquet halls is very narrow, and it is quite unlikely that the scientific literature can provide much guidance about how to identify the mind-sets, if even the scientific (or business) literature recognizes the existence. The PVI will be used to create a tool comprising six question which will assign a new person to one of the three mind-sets.

Results

Total panel versus emergent mind-sets

Respondents cannot easily tell the researcher what is important versus what is not important. Yet, the coefficients from the model reveal immediately how strongly each element drives the TOP3, the rating ‘I would choose’.

We begin with the additive constant. The additive constant tells us the conditional probability of a person responding 7-9 on the 9-point scale, viz., I would likely choose this banquet hall from my next event. Keep in mind that most respondents are not in the immediate situation of choosing a banquet hall. We would expect, therefore, that the respondents have not thought about the banquet hall. What is remarkable, but not surprising, is the very low additive constant, 6, one of the lowest for total panel in the many Mind Genomics studies conducted by author HRM. At a basic level the respondents are simply not interested in the banquet hall. The respondents are being honest. They are reading the vignettes and, not being interested in a banquet hall, they respond that they are not interested.

Beyond the additive constant is the contributory power of the different elements. The typical standard errors are around 3-4 for the coefficients of the elements. Table 2 shows the strong performing coefficients for the 36 elements by Total Panel, and by the three mind-sets which emerge, and discussed below. The elements are shown in descending order based upon the three mind-sets. To allow patterns to emerge, we will remove any coefficients lower than +8. We lose some of the fine-grained information, but the patterns more clearly emerge.

Table 2: Performance of the elements by Total Panel and by the three mind-sets. Only strong performing elements are shown, with coefficient of 8 or higher.

We can sort the elements based upon the Total Panel, and pull out the very strong performing elements, those elements with coefficients of +10 or higher. The coefficient tells us the increased (or decreased) percent of responses of magnitude 7-9 beyond the percent shown by the additive constant when the element is inserted into the vignette. For example, the additive constant is 6 for the Total Panel, meaning that in the absence of elements we expect to see 6% of the ratings be 7-9. Incorporate the element D4, Open Bar included at no extra cost, and we can expect an additional 16% of the responses (viz., 22%) to be 7-9. Choose D3 instead, Only $150 per person for closed event, and we can expect to lose 9% of the respondents, viz., end up with a sum of -3, viz., virtually none of the rating being 7-9. Put both D4 and D4 in together, and we can expect the percent of ratings 7-9 to be 6% (additive constant) + 16% (Open Bar …) – 9% ($150 per person) or 13%.

There are seven very strong elements. They make sense. These are the elements with coefficients of +10 or higher. There is no single ‘theme’ appropriate to these seven elements. They range from open bar to photographer, to flexibility, and so forth. It should be kept in mind that without the Mind Genomics experiment it would be highly unlikely for anyone to be able to predict that each of the seven elements would be perform well. It takes an experiment to reveal the winners.

Open Bar included at no extra cost

Professional photographers upon request

Want to have a fun night out with your friends… Host a theme party at our fascinating hall

Want an unforgettable prom night… Host it here and it will be a night you will never forget

Need a last-minute party? We are fast, fun and friendly

All the information is on our website…only a click away

All rooms are customizable to fit your party size

There are elements with coefficients of 0 or lower. These are not necessarily ‘bad’ but may be either ‘bad’ (ratings of 1-3) or ‘irrelevant’ (ratings of 4-6). We would have to analyze the data from the opposite side of the scale (1-3 transformed to 100; 4-9 transformed to 0). That analysis in the reverse direction would tell us whether the negatives are truly ‘bad’ (1-3) or merely irrelevant (4-6.)

The more revealing results emerge when we consider the nature of the mind-sets, what attracts these mind-sets, and the existence of an underlying theme for each mind-set. Respondents may or may not know about mind-sets. Indeed, unless the topic of banquet halls comes up as a focus of one’s conversation and plans, it is unlikely that one would even pay attention to what’s important about a banquet hall unless challenged to answer. The Mind Genomics experiment will, however, reveal these mind-sets in about 5-10 minutes.

The clustering program (Step 6 above) suggested three groups of respondents, based upon the pattern of coefficients. The three right-most columns of data show the strong performing elements for each of the mind-sets. An element performing well in two mind-sets appears, twice, once for each mid-set in which is performs well. The name to be given to the mind-set comes from the nature of the elements which perform best.

Mind-Set 1, with 33 respondents, has an additive constant of 13. The respondents are not particularly interested in the banquet hall but respond positively to elements of a very general nature. One gets a sense that Mind-Set 1 comprises individual who respond to general, practical information, but are not thinking of the specifics of an event.

All the information is on our website…only a click away

Want an unforgettable prom night… Host it here and it will be a night you will never forget

Want to have a fun night out with your friends… Host a theme party at our fascinating hall

Need a last-minute party? We are fast, fun and friendly

Easy to find parking near the hall

Mind-Set 2, with 10 respondents, has an additive constant of -48, but shows 29 of the 36 elements scoring very well. Mind-Set 2 is definitely interested in the particulars of the banquet. The very low additive constant is deceptive and is a statistical aberration to correct for the exceptionally high-scoring elements appropriate for a person who actively contemplates hiring a banquet hall. We can get a sense of their seriousness by looking at the elements which generate the exceptionally high coefficients of 30 or above. These are the respondents who are in the target audience.

Open Bar included at no extra cost

Most affordable rates per hour …4 hours per night

All rooms are customizable to fit your party size

Our professional event planner takes care of interior decorations

Keep More in your wallet with our flexible offers

All fees and taxes included in final price

Only $120 every additional hour

Take a step outside onto our one of a kind designed balcony

Professional photographers upon request

Mind-Set 3 with 14 respondents shows the highest additive constant, 27. They are modestly interested in the notion of a banquet hall, but the highest scoring elements give a sense of a casual shopper looking at the offer of a banquet hall. The only element which strongly interests them are ‘All rooms are customizable to fit your party size.’ One gets a sense that this is information which is ‘nice to know’ but the respondents are not excited by the idea of a banquet hall in the same way that respondents in Mind-

Set 2 are.

All rooms are customizable to fit your party size

Lost your valuables? We keep them safe in our safe

Open Bar included at no extra cost

Have a nice view of the great scenery

Linking elements to feelings/emotions

In the world of consumer research, the recognition of emotion as a key driver of purchase has a long history. Indeed, a great deal of the consumer research is on the emotions evoked by products, services, and advertisements. When emotions are invoked the typical approach instructs the respondents to select the one or two emotions which ‘fit’ the vignette. The standard analysis measures the percent of respondents assigning one or another emotion to the test stimulus, either for the entire test stimulus, or during the period that the test stimulus is being evaluated, e.g., for a video.

The Mind Genomics approach differs. The goal is to link each individual element to every one of the emotions. It is impossible to do that in a simple way because there are several different emotions, and the respondent’s job is extremely difficult to divide up the emotional feeling into the different parts. One might have the respondents select the most dominant emotion experience, the approach used here, or instruct the respondent to check ‘all which apply,’ not used here, but also feasible.

The mapping of the one rating (select the single emotion which best describes the feeling) into the five alternative emotions, allows the researcher to link the 36 elements to emotions. As noted in Step 7above, the analysis uses OLS regression, without incorporating the additive constant. The coefficients in the OLS equation can be considered to be the linkage between the element and the emotion. The linkages for the total panel appear in Table 3. Again we focus only on the strong linkages, this time defined as +8 or higher, based upon previous experiences with linkages in Mind Genomics.

Table 3: Linkages between the elements and the five feelings/emotions. Only strong linkages are show (>= 6)

Most of the elements link strongly to feeling/emotion, as shown by the shaded cells. A few elements link strongly to two emotions. For example, element F3, Have the wedding of her dreams, a phrase not particularly interesting to anyone not contemplating marriage, links both to disinterested and to interested, presumably because of the very specific nature of the phrase, viz., a wedding.

Assigning a new person to a mind-set at any time in the knowledge development process

Knowing the nature of the mind-sets for topics of the everyday builds up the ‘Wikipedia of the Mind’. Beyond knowledge, however, is the opportunity to use the information for further research, or for applications such as sales. Rapidly and affordably uncovering the nature of the underlying mind-sets of people on any topic that can be dimensionalized is one distinct contribution of Mind Genomics. The second contribution is the ability to assign a new person to one of the mind-sets.

In the world of consumer research, sociology, and the like, there is a pervasive belief that ‘birds of a feather flock together,’ or more specifically, people who ‘look alike think alike.’ This notion underlies some of the strategies of companies such as Facebook, which observe a specific behavior of different people, and find commonalities among these people in terms of “who they are.” One can market to this newly created group by marketing to people which look like them, on a variety of dimensions, so-called look-alikes.

It is attractive to search for assignment rules which put new people into newly developed mind-sets. Those who work with so-called ‘Big Data’ believe that it is possible to do so, if only the algorithms and the computing power is sufficient, only if the data to be processed is sufficient, and only there are enough cases so that one is not working with random noise. The attraction of such ‘hope’ is enormous, despite the fact that the use of Big Data is to assign individuals to specific, granular segments, of a type that may not even be suspected.

Such marketing strategies make sense when one looks at behavior from the ‘outside-in’, searching for common properties of a group of people which are of interest. Mind Genomics works from the opposite direction, knowing that there are certain ways of ‘thinking’ about a topic, and then determining how a new person thinks about the topic. The approach is not to find lookalikes, but rather to create a simple test which assigns a person to a mind-set. The test, called a PVI, personal viewpoint identifier, is applied to the data of the original group of respondents which defined the mind-set, and then applied to new individuals.

Table 4 shows the distribution of the respondents by total panel and by the three mind-sets. The respondents completed a short classification questionnaire at the start of the study. The questionnaire provided the respondent’s gender, age, and the reason(s) that the respondent might be interested in hiring a banquet hall. The respondents could select any number of the different reasons for hiring a banquet hall. It is clear from Table 4 that the mind-sets are found in all of the groups into which the respondent can define herself or himself. It is not easy, if even possible, to develop a simple assignment rule to mind-set based simply upon the pattern of who the respondent says she or he IS, or what she or he would hire a banquet hall.

Table 4: Distribution of the respondents into groups (gender, age, reason for hiring a banquet hall.)

Mind Genomics provides an entirely different approach to assigning people to groups, one working from the bottom up, from the simple granularity of an experiment that can be conducted in an hour or two, or several iterative experiments that can be conducted in a day or so. The approach used by Mind Genomics is called the PVI, the personal viewpoint identifier.

The PVI uses the basic data from the mind-sets (see Table 2), perturbing the data with random numbers added to the coefficients. Through a Monte-Carlo simulation the PVI identifies six elements which best differentiate among the mind-sets, based on a two-point scale. The PVI then generates 64 patterns. Each pattern maps to one of the two or three mind-sets.

Figure 4 shows the introductory page to the PVI, which obtains relevant information about the respondent. Each question can be suppressed. Some of the information is appropriate to marketing, other information, e.g., about time, relevant to scientific studies about decision making and diurnal rhythms and so forth. Figure 5 shows the actual PVI itself, comprising background questions about attitudes and usage, as well as the six question, which appear in randomized order, different for each respondent. The six questions appear after the background questions. Finally, Figure 6 shows the feedback to the respondent regarding the mind-set to which the respondent is assigned. The information from Figures 4-6 are maintained in a user-accessible database. Thus, the PVI provides both ‘gamification’ to interest the respondent, and archival information for subsequent work. The respondent can also be immediately led to a video or to a ‘landing page’ on a website, depending upon the mind-set to which the respondent has been just assigned.

Figure 4: Introductory page to the PVI

Figure 5: The background questions and the six PVI questions

Figure 6: Results from the PVI, sent to the respondent. The mind-set to which the respondent is assigned is shown by the shaded cells.

Discussion and conclusion

The Mind Genomics experiment detailed here represents a divergent way of thinking in social science. As noted in the introduction, researchers in the social science usually forego the experimental method in favor of observation when the topic is everyday behavior. The psychologists and economists might study everyday behavior, but with the exception of researchers such as Robert Shiller [25], few economists and psychologists study the simple, every-day experience, which constitutes the warp and woof of social life.

The story changes a little when we move beyond anthropology, sociology and classical economics to some aspects of experimental social psychology, behavioral economics, and occasionally consumer research, respectively. The experimentation focuses on what might be considered unusual perturbations of the everyday, such as varying prices or prices expressed in different ways. The objective is, through the experiment, to express some generality of behavior, such as the fact that people like prices specified by text rather than by numbers When it comes to consumer research the effort is no so much on experimentation at all as on the role of the test stimulus (the nature of the banquet hall, etc.) in driving choice.

Mind Genomics provides the researcher with a new set of tools, systematic experimentation that can be done using simple to vary stimuli (viz., combinations of messages, or combinations of visuals [26], and a simple respondent task (e.g., read/look, and then rate the combination on the basis of criteria laid out in a rating scale provide to the respondent.) In doing so, Mind Genomics may open up our understanding of ‘typical’ human behavior of the everyday, a rich source to understand the ‘mind of man.’

Statistical appendix – establishing consistency of response

The impetus for this statistical appendix comes from the continuing question by those introduced to the method, name ‘can people really do this task?’ We are accustomed to many people in the applied profession of consumer research who believe that ordinary people simply cannot be consistent when evaluating mixtures of messages. The ingoing assumption is that unless the messages are simple, the respondent becomes totally confused. This assumption is reinforced by exit interviews in which respondents ‘complain’ that they did not do the study correctly because they did not ‘know’ what the researcher wanted to hear, and thus they feel that their answers were random.

In order to address these ongoing issues, one can point to the data and show that the pattern of responses is meaningful, at least at an intuitive level. An example of this is element D4, ‘Open Bar included at no extra cost’ with a coefficient of +16 for the Total Panel. Such suggestions of valid results appear in study after study, but are not considered sufficiently rigorous. Some critics want statistical evidence of reliability, such as split half reliability.

One way to show consistency of responses, which may be a form of validity, is to demonstrate that the ratings accurately ‘track’ the stimuli, at the level of the individual respondent. That is, within the confines of the Mind Genomics experiment, one can create a model for each respondent relating the numerical ratings on the 9-point scale to the presence/absence of the 36 elements. The degree to which the ratings ‘track’ the elements is a measure of the consistency of the ratings, and in some respects to the validity of the respondent’s ratings within the confines of the Mind Genomics experiment.

The approach to establish consistency is to compute the Pearson multiple correlation for each respondent, showing the strength of a linear relation between the elements and the ratings. The Pearson multiple linear correlation goes from a perfect +1 (perfect linear relation), down to 0 (no apparent relation).

Figure 7 shows the distribution of multiple linear correlations for the respondents, based on 48 observations, and 36 predictors. The R values were computed by relating the presence/absence of the 36 elements to the 9-point rating, using OLS (ordinary least-squares) regression .More than half of the multiple linear correlations are 0.6 or higher, suggesting that despite the apparent difficulty of choosing the relevant information from the vignette, respondents act in a consistent manner, and in a seeming interpretable, rational manner as well.

Figure 7: Distribution of the multiple Pearson R values across the 57 respondents.

Acknowledgement

Attila Gere wishes to acknowledge and thank the Premium Postdoctoral Research Program of The Hungarian Academy of Sciences

References

1. Dunbabin KM (2003) The Roman Banquet: Images of Conviviality. Cambridge University Press.
2. Graham JW (1967) A banquet hall at Mycenaean Pylos. American Journal of Archaeology 71: 353-360.
3. Anglo S (2017) Overcrowding at court: A Renaissance problem and its solution: temporary theatres and banquet halls. In Architectures of Festival in Early Modern Europe 201-212.
4. Chau AY (2008) The sensorial production of the social. Ethnos 73: 485-504.
5. Meads C (2001) Banquets Set Forth: Banqueting in English Renaissance Drama. Manchester University Press.
6. Edwards W (1987) The commercialized wedding as ritual: A window on social values. Journal of Japanese Studies 13: 51-78.
7. Lau CK, Hui SH (2010) Selection attributes of wedding banquet venues: An exploratory study of Hong Kong prospective wedding couples. International Journal of Hospitality Management 29: 268-276.
8. Schoenfeld S (1987) Folk Judaism, elite Judaism and the role of bar mitzvah in the development of the synagogue and Jewish school in America. Contemporary Jewry 9: 67.
9. Adler H, Chienm TC (2005) The wedding business: a method to boost food and beverage revenues in hotels. Journal of Foodservice Business Research 7: 117-125.
10. Choi JH, Choi YJ (2017) The effect of service scape of banquet hall at special class hotels on customer satisfaction and positive word-of-mouth intention. Korea Tourism Research Association 32: 125-145.
11. Kim YK (2016) A study on the relationship among service quality and customer satisfaction of wedding hall restaurants, and recommendation intensity-focus on the moderating effect of wedding hall and hotel image. Culinary Science and Hospitality Research22: 252-266.
12. Kim GC, Lee YJ (2009) A study on the importance and performance of the display of dining space for Hotel banquets. Culinary science and hospitality research 15: 173-187.
13. LaFleur T, Hyten C (1995) Improving the quality of hotel banquet staff performance. Journal of Organizational Behavior Management 15: 69-93.
14. Noone BM, Kimes SE, Renaghan LM (2003) Integrating customer relationship management and revenue management: A hotel perspective. Journal of Revenue and Pricing Management 2: 7-21.
15. Guan L (2014) “Push and Pull Factors in Determining the Consumer’s Motivations for Choosing Wedding Banquet Venues: A Case Study in Chongqing, China”. Graduate Theses & Dissertations. 13851.
16. Peng MC, Wang RY (2017) Evaluating wedding banquet halls using a novel multiple-criteria decision-making model. Advances in Management and Applied Economics 7: 13-28.
17. Moskowitz HR (2012) ‘Mind genomics’: The experimental, inductive science of the ordinary, and its application to aspects of food and feeding. Physiology & behavior 107: 606-613.
18. Moskowitz HR, Gofman A, Beckley J, Ashman H (2006) Founding a new science: Mind genomics. Journal of sensory studies 21: 266-307.
19. Moskowitz HR, Gofman A (2007) Selling Blue Elephants: How to Make Great Products that People Want before they even Know they Want Them Pearson Education.
20. Kahneman D (2011) Thinking, Fast and Slow. Macmillan.
21. Miller GA, Galanter E, Pribram KH (1960) Plans and the Structure of Behavior. Henry Holt and Co.
22. Box GE, Hunter WH, Hunter S (1978)Statistics for Experimenters (Vol. 664). New York: John Wiley & sons.
23. Gofman A, Moskowitz H (2010) Isomorphic permuted experimental designs and their application in conjoint analysis. Journal of Sensory Studies 25: 127-145.
24. Jain AK, Dubes RC (1988) Algorithms for Clustering Data. Prentice-Hall, Inc.
25. Shiller RJ (2017) Narrative economics. American Economic Review 107: 967-1004.
26. Gofman A, Moskowitz HR, Fyrbjork J, Moskowitz D,  Mets T (2009) Extending rule developing experimentation to perception of food packages with eye tracking. The Open Food Science Journal 3: 66-78.
27. Moskowitz MR, Ashman H, Minkus-McKenna D, Rabino S, Beckley JH (2006) Databasing the shopper’s mind: Approaches to a ‘mind genomics’. Journal of Database Marketing & Customer Strategy Management 13: 144-155.

DOI: 10.31038/MGSPE.2021111

Abstract

Respondents evaluated systematically varied combinations of messages (vignettes) about healthful pasta. Each respondent evaluated a unique set of 48 vignettes constructed from 36 messages about different aspects of health carbs, rating it both on purchase intent and selecting the price willing to pay. Each respondents rated each vignette on both purchase intent and price would pay. Two sets of three mind-sets emerge, one based on purchase intent, one based on price would pay. Three patterns of mind-sets emerge; focus on cognitive (Brain) performance, focus on a healthier/more enjoyable life (Life), and focus on taste and sensory pleasure (Comfort Food). These mind-sets exhibit different patterns of what is important to them when making a judgment.  The paper shows the ability of the emerging science of Mind Genomics to probe deeply into what seems at first to be a simple topic, healthful pasta, and the ability to reveal profound differences in the way people think about this supposedly simple, limited topic.

Introduction

During the past sixty-plus years, the notion that people ‘differ’ from each other in predictable ways has gained increasing popularity in the world of consumer research. The ever-present variation among people, observed by authors, philosophers, social scientists, biologists, not to mention governments and their politicians, no longer represents a source of irritating variability in the world of ‘nomos’, of laws which apply the same to all people. Rather, the person-to-person variations, the world of ‘idio’, is falling increasing under the scrutinizing lens of the researcher, who is searching for rules explicating that person-to-person variability.

More than sixty years ago, famed consumer researcher William D. Wells suggested that it might be possible to divide people not so much by who they ARE, but rather by how they THINK. This was the auspicious beginnings of the field of Psychographics [1], which eventuated into classifications such as PRIZM by Claritas [2], with its 68 different personas. The number of personas might seem unusually high in light of the normal desire of people to simplify the world, but the objective of PRIZM and such types of classifications is to deal with the remarkable diversity of the minds of people, doing so in a way, which makes the diversity tractable.

The move in business and research is find many mind-sets for a topic, and in so doing account for the substantial variability across people. Mind-sets, the search for ‘basic groups’, moves the reality of variability from merely an irritating fact-of-life to a front-and-center position as a signal that there are fundamental groups in the population to be discovered.  The only issue is the desire to account for the variability by find ever more ‘basic factors’ in an attempt to explain as much of the variation as possible. The aforementioned system, PRIZM, is just such an example.  The general focus, then, is to take a subject with a large number of aspects, like the way one thinks about life, or health, and break it down to simplistic groups. The key words are wide range, many groups, account for as much variability as possible

This paper moves in the opposite direction, looking at a topic that might be considered ‘narrow’ at the very outset, healthful pasta, and trying to divide this narrow topic even further. The objective, therefore, is to work with a narrow world that people might already consider to be ‘hard to segment because of its specificity’, and uncover even deeper-lying mind-sets.

Mind Genomics and the study of psychological granularity

We approach the topic of micro-segmentation looking at one product, healthful pasta. Pasta itself is a very large category around the world, historically important during the past centuries [3][4], and a source of continuing innovation. We are accustomed to pasta as being the carrier for a flavor-imparting and mouthfeel-imparting product, most generally that product being a sauce. Pasta can also be a carrier for vegetables, for meats, and even eaten without anything, except perhaps with a bit of melted butter to add flavor.

Our focus, micro-segmentation, means that we want to look at the different ways that a person can perceive pasta.  One might ask the respondent to talk about pasta, and the different aspects of pasta, but it’s not clear that this in-depth interview or focus group interview will be able to uncover the micro-segments or realize their existence and nature when the research happens upon them.  Conventional consumer research, whether interviews, surveys, and so forth, simply are instruments which, in the end, are just too dull to work.  The issue is just how many ways can one talk about pasta, and what kind of questions can one ask, and then just recognize when something really new emerges.

Mind Genomics works in a variety of fashions radically different from the approaches used by consumer research. The approaches of Mind Genomics have been explicated in a number of earlier publications [5][6][7]

Experiment, not survey

Mind Genomics comprises experiments, not surveys, not discussions. Mind Genomics mixes together statements about the topic, pasta, these statements (or elements) having been combined by an underlying menu (experimental design) which dictates the combinations. For our case history with helpful pasta, we have to limit ourselves to the specific number required by the underlying experimental design. The Mind Genomics design requires six questions and six answers to each question (aka elements, messages) Table 1 shows the six questions, and the six answers to each question, or a total of 36 elements. These elements will be the raw material of the study.

One of the important aspects of Table 1 is the depth of information, the granularity of the statements. The granularity matches the type of information presented to people in every-day life, the type of information which makes the stimulus more real. People do not live in a world of abstractions, a world where the detail is sacrificed to a general phrase which is so general as to not to relate very much to the actual experience. For example, in Table 1 we might use the phrase ‘good carbohydrates’ to give the reader a sense of the stress on good for the mind, good for the body, good for everyday life. Yet, the term ‘good carbohydrates,’ pales in in comparison to the phrases in Table 1, which add color, texture, and a sense of reality.

Table 1

If there is any single aspect of Mind Genomics which can be said to be of major import to the world of knowledge development, that aspect might just be the study using granular messages, rather than the study using generalities, hollowed-out messages, presenting a general idea, but one without a sense of experience, richness and evocative meaning. In other words, the building blocks of Mind Genomics, are ‘cognitively rich.’

Vignettes, combinations of elements as the test stimulus

The respondent reacts to the combinations. The stimulus is presented in simple format, with no attempt to create a coherent picture, and a pleasant reading experience (See Figures 1,2). The Mind Genomics experiment comprises the presentation of disparate pieces of information, pieces which may be joined in the mind of the respondent, pieces which may be concordant or discordant. All the respondent has to do is read and rate the combination, on either one rating scale (Figure 1; Purchase Intent) or on two rating scales (Figure 1 for Purchase Intent; Figure 2 for Price.  The typical commissioning professional of the research , the market researcher or the marketer, or product developer, often expresses a desire for fully formed, polished concepts, and just one or two of them, one to be selected as the ‘better’ and presumably (but not directly stated), the ‘best’ of that could be. The Mind Genomics approach flies in the face of that conventional system, presenting combinations of factoids. It is the task of the respondent to roam through the information and assign the rating.

Figure 1: Example of a test combination of elements (vignette) rated on purchase interest

Figure 2: Example of the same test combination of elements, rated on price would pay

Experimental Design of test stimuli

The vignettes are constructed according to an experimental design [8]. The experimental design specifies the combinations to be created for each individual, so-called vignettes. For this study of 6 questions, 6 answers per question, comes to a set of 60 vignettes. The vignettes comprise as few as two elements, and as many as four elements, designed in such a way that each element appears equally often, that the 36 elements are statistically independent of each other, and the number of 2-element, 3-element and 4-element vignettes are always the same across respondents. A permutation scheme [9] changes the specific combinations, on a respondent-by-respondent basis. The combinations for each respondent are different from each other, allowing the Mind-Genomics system to test many of the possible combinations at least once, sometimes twice. This approach differs at its core from the conventional approach in research which selects a limited number of combinations, testing that limited set of combinations many times in to reduce the variability, viz., by averaging.

Both approaches, the Mind Genomics evaluation of many combinations and the conventional many-replicate approaches focus on the same objective – to identify how each element drives the response, doing so by reducing noise. The conventional approach averages out the noise but limits the number of vignettes to what turns out to be very few. The conventional approach assumes that the combinations selected truly provide a ‘good sample’ of the full set of data. I contrast, the Mind Genomics approach allows for a noisy measurement of each point, because each point has only one measurement. However, across many respondents the Mind Genomics study evaluates many of the possible combinations, allowing the pattern to emerge. A good metaphor for the Mind Genomics approach is the ‘MRI of the mind.’

The respondent experience

The respondent is oriented in the evaluation through a simple description, provided more as a formality than as a deep introduction to the topic (Figure 3). The respondent reads the vignette and rates the vignette on two scales, purchase intent and price would pay, defined as shares of stock (Table 2). The respondent rates the vignette on the first, and then the second scale appears. The respondent rates the vignette on the second scale and then next vignette appears. Respondents have no problem sifting through any size vignette, reading what is presented, and making their judgment. Indeed, to the respondent, there is no sense of complete versus incomplete. The vignette is simply a collection of elements to be read as a unity and then rated.

Figure 3: The orientation page, instructing the respondent what to do.

Table 2: The rating scales

Prepare data for analysis

Each respondent evaluates 48 vignettes, created from the 36 elements. All elements appear an equal number of times across the48 vignettes. The respondent’s rating is assigned according to a Likert Scale (Rating Scale 1) or in terms of different dollar values (Rating Scale 2).  For each respondent and each vignette, the rating scales are transformed prior to analysis by OLS (ordinary least-squares) regression.

a. For rating scale #1 (purchase), the first transformation (TOP3) shows the likelihood of the response ‘I’ll buy this product as described by the vignette.’ The 9-point Likert Scale is transformed to a binary scale, with ratings of 1-6 transformed to 0 to denote either ‘not buy’ or ‘may buy / may not buy.’  Ratings of 7-9 are transform to 100 to denote ‘will probably or definitely buy.’ The transform from a category or Likert scale to a binary scale follows the approach of consumer researchers and public opinion pollsters who find that it is easier for their audiences to understand no/yes, rather than the meaning of say a 6.3 on a 9-point scale.

b. For rating scale #1 (purchase), the second transformation (BOT3) shows the likelihood of the response ‘I will not buy this product as described by the vignette.’ The 9-point Likert Scale is transformed to a binary scale, but ratings of 1-3 are transformed to 100 denoting ‘not buy’ and ratings of 4-9 are transformed to 100, denoting ‘may/may not buy or probably/definitely buy. We will be interested in the elements which drive a respondent away from buying, towards actively rejecting the product. The best way to discover the ‘drivers’ of rejection is to look at the part of the underlying rating scale dealing with active rejection.

c. For rating scale #2 (price), we transform the rating value to the dollars selected. This gives us a sense of how much people are willing to pay.

When we look at the actual data from our 151 respondents, each of whom evaluated the 48 vignettes, we see a distribution in each of these dependent variables.  We see that the respondents distribute their ratings on the 9-point scale, and that there quite a number of vignettes which score well, assigned a rating of 7-9 (see Figure 4, left panel). We also see that despite the high ratings of purchase intent, the respondents do not feel that the shares of stock in the company making the product are worth very much (Figure 4, middle panel). Finally, when we plot price of the share on the ordinate versus purchase intent on the abscissa, two measures of acceptance, on involving behavior, the other invoking economics, we see the expected relation between purchase intent (‘I like it more’) and price willing to pay (Figure 4, right panel).

Figure 4 gives a sense of the general type of information provided by data in which the test stimuli have little or no cognitive richness, but are rather test stimuli, the responses to which are measured and summarized.  There is little to be gained from an in-depth analysis of the data at this point because the data has little cognitive richness. We can say that the patterns appear to follow one or another structure, but we cannot actually feel that we are entering into the ‘mind’ of the respondent. The researcher could develop a picture of some aspect of the mind of the respondent by looking at the patterns of purchase intent vs price for different groups, such as males versus females, and so forth. The researcher would then learn that for a specific group (to be named after analysis), the respondents in that group are likely to say that they would pay a fair bit MORE for the product as the purchase rating goes from level A to level B, whereas a complementary group would not pay a fair bit more for the same change in purchase rating, from Level A to Level B. As long as one can measure purchase intent and price on many stimuli one can create these graphs for the total panel, for any subgroup, and in turn show differences in pattern, and then hypothesize about what might be responsible for those group-to-group differences in the patterns of the data.

Figure 4: Distribution of ratings for purchase, of price willing to pay, and the ‘smoothed and summarized’ relation between price willing to pay and purchase intent. The data come from the full group of 151 respondents, each of whom evaluated 48 unique vignettes.

The Mind Genomics ‘project’ moves in a deeper direction, putting numbers on the individual elements which constituted the building blocks of the vignettes. The deep goal of Mind Genomics is to put numbers onto these elements, numbers which are meaningful to the respondent and the researcher alike, numbers which tell a story, and shed light on the decision-making process. The cognitively rich array of elements in Table 1 provides the matrix of messages. The nature of the respondent’s mind can be understood a bit more deeply when these different elements have numbers attached to them. When these numbers attached to the elements emerge from behavior rather than from direct evaluation of the elements in a survey, the insight into the mind is even deeper. When the elements compete with each other, the resulting numbers show the ‘drawing power’ of each element.

Transform the data to prepare for regression modeling

The experimental design, creating as it does 48 vignettes for each person, allows for a statistical analysis which relates the presence/absence of the elements to the dependent variable. The data matrix is set up as a set of rows, specifically 48 rows for each respondent, each row corresponding to one of the 48 vignettes. The elements are the columns, 36 columns altogether, one column for each element. For each vignette, a column can either show the value ‘0’ when the element is absent from the vignette, or a ‘1’ when the element is present in the vignette.  The next two columns correspond to the actual rating assigned by the respondent. The final three columns correspond to transformed data. The first of the final three columns is labelled TOP3, taking on the value 0 when the rating was 1-6 (viz., not buy or may/may not buy), and taking on the value 100 when the rating was 7-9 (probably/definitely buy). The second of the final three columns is labelled BOT3, taking on the value 100 when the rating was 1-3 (definitely Not buy/probably Not buy), and taking on the value 0 when the rating was 4-9 (might/might not buy, probably/definitely buy). The third and last of the final three columns is labelled PRICE corresponding to the price defined by the number shares x dollars/share.  Table 3 shows a portion of the data table, rotated for the sake of space, with the data in the table ready for analysis by OLS (ordinary least-squares.)

The data are now ready for analysis by OLS (ordinary least-squares) regression. The objective is to create a mathematical equation of the form below, the equation showing how each of the 36 elements drives the response. The 36 elements will be the independent variables, the three newly created variables (TOP3, BOT3, Dollar Price) will be the dependent variables. The matrix show in Table 3 is ready for analysis, both at the level of each of the 151 respondents, and at the level of all of the respondents, or some defined subset of the responses.

Table 3: Example of data from the study, along with the transformation, and ready for OLS (ordinary least-squares) analysis

Create individual level models for TOP3, BOT3, and Price, respectively, generating three sets of 151 models or equations, each set comprising 36 coefficients

Each model is an equation of the form: Transformed Rating = k₁(A1) + k₂(A2) … k₃₆(F36)

The equation is absent the additive constant, viz., goes through the origin. This form of the equation makes it easier to compare coefficients for TOP3 and BOT3.  The equation shows us the number of transformed rating points attributed to each element, when that element is included in the vignette. Thus, when the coefficient is + 11, we interpret this to mean that 11 transformed rating points are added to the rating. For the case of TOP3, a +11 means that when the element is incorporated into the vignette, an additional 11% of the respondents will assign the vignette the rating of 7-9. For the case of BOT3, a+11 means that when the element is incorporated into the vignette, an additional 11% of the respondents will assign the vignette the rating of 1-3.  Finally, for PRICE, when the coefficient is +11, the incorporation of the element into the vignette will increase the number of dollars by 11.

Create the three models (equations) for the Total Panel

Table 4 shows the coefficients for the 36 elements, sorted by the coefficients for TOP3, interest in purchasing the pasta product. The important thing to observe about Table 4 is the sense of ‘knowing the mind of the respondent,’ simply by knowing the text of the elements. The modeling provides the numbers. It is the numbers which allow us to sort the data and to get a sense of which elements most likely will drive purchase, which elements will prevent purchase, as well as which elements are most valuable versus least valuable.  Note that the Mind Genomics output presents what could be an overwhelming volume of numbers. In order to let patterns emerge, Table 4 presents only coefficients of 7 or higher for TOP3 and for BOT3.

Table 4: Coefficients for the three models (TOP3, BOT3, PRICE) for the total panel

Highest TOP3: Look good & feel great at any age with a whole-grain fortified pasta diet… delivers plenty of energy & lifelong weight control

Highest BOT3: Whole grain fortified pasta is not always the most palatable taste & texture… a bit grainy

Highest PRICE: Look good & feel great at any age with a whole-grain fortified pasta diet… delivers plenty of energy & lifelong weight control

As we read through the data in Table 4 it is no necessary to accept or reject hypotheses. Mind Genomics presents us with a list of the different elements, and their scores. There is no necessity to begin with any hypothesis that must be falsified. There may be absolutely no prior knowledge at all about healthful pastas, in which case these would be the results from the pioneering efforts. The real thinking can now begin, to look at the winner versus the losers, and create grounded hypotheses, results from simple experiments. Furthermore, the experiment will provide a great deal of additional knowledge and insight, as we soon will see in the subsequent sections.

Divide the 151 respondents into three complementary mind-sets, based upon the pattern of coefficients for a dependent variable

The division into three mind-sets was done in order to compare the nature of mind-sets for TOP3, BOT3, and PRICE, respectively. The three sets of coefficients give us a sense of how the 36 elements drive positive interest in purchase (TOP3), drive negative interest in purchase (BOT3), and drive price that would be paid (PRICE).

Quite often there is an assumption that people differ in what they like, and so the respondents are divided by convenient geo-demographic data such as gender, age, where the respondent lives, and so forth. Occasionally the respondents are divided by what they say they feel to be important (e.g., taste versus price versus convenience versus health), such information obtained by an additional questionnaire administered at the time of the evaluation. A third and equally common way to divide the respondents is by what they say they have done, either in consumption or in purchase.  All three ways of dividing people end up showing differences in the pattern of responses to the elements, but the patterns are quite noisy, and the underlying ‘story’ is hard to discern.  It is not clear whether the variation is noise, or a weakly attenuated signal. What is clear, however, is that WHO A PERSON IS DOES NOT PREDICT HOW A PERSON RESPONDS TO SPECIFIC MESSAGES.

Once the coefficients are created for a variable, e.g., TOP3, we use clustering to divide the 151 respondents into exactly three groups. The choice of three groups is done by fiat, to divide the respondents into ‘fine grained groups,’ not too many and not too few.  The clustering is done by minimizing the ‘distance’ between pairs of respondents within a cluster, based upon a measure of distance (D = (1-Pearson Correlation)), with the distance based upon the coefficients.  Two respondents whose 36 coefficients show the same exact pattern of responses to the messages, generate a Pearson Correlation of +1, and a distance between then of 0 (D = 1-1 = 0). In contrast, two respondents whose 36 coefficients show precisely opposite patterns of responses generate a Pearson Correlation of -1, and a distance between them of 2 (D = 1 – – 1 = 2).

We run the separate k-means clustering program separately for each of the three sets of coefficients [10]. These three separate analyses each generates its own group of three mind-set. The three mind-sets for each of the three dependent measures (TOP3, BOT3, PRICE) need no necessary relation to each other. For example, two respondents falling into the same mind-set for one dependent variable (e.g. TOP3) need not fall into the same mind-set for the other dependent variables (e.g. BOT3 and PRICE, respectively).

Once we define the mind-sets for each dependent variable, we then run three regressions, again without the additive constant, one regression equation for all the data from Mind-Set1, a second regression for all the data from Mind-Set2, and finally a third regression for all the data from Mind-Set3.

Themes of the mind-sets.  Our first analysis of the mind-sets considers the themes. There are really three themes:  Brain function, Life performance, and Pasta as food, respectively. Each general theme is positive for some mind-sets and negative for others. The elements were all written as positive descriptors, so the negatives come from people’s dislike of the content of the message, not from the structure of the message.

Brain function

TOP3 Mind Set 1 – Good brain function

PRICE Mind-Set 2 – Values pasta for better brain function

Life performance

TOP3 Mind Set 2 – Good product for energy, socializing, good thinking, better overall life.

PRICE Mind-Set 3 -Better looking, better life, better performance

PRICE Mind-Set 1 – Values strength, health, positive outlook, no weight gain

BOT3 Mind-Set 1 – Turned off by pasta seen as a functional fuel for behavior

BOT3 Mind-Set 3 – Turned off by pasta for mood

Comfort food

TOP3 Mind-Set 3 – Pasta as comfort food

BOT3 Mind Set 2 – Turned off by novel taste, wheaty

Strong performing elements in the three mind-sets

Mind Genomics studies generate a great deal of data, results which are interesting in and of themselves because the test stimuli are meaningful. In the interests of clarity and space, the mind-set data will be reduced to show only the strong performing elements for the specific mind-set, and the performance of those strong performing elements for the Total Panel as well. In the interests of simplicity, we now present only those elements with coefficients of +10 or higher.

The data appear in Table 5 (TOP3), Table 6 (BOT3), and Table 7 (PRICE), respectively. The strong performing elements are shown in shaded cells.  It is not necessary to go through each table, but rather simply look at the name assigned to the mind-set to get a sense of the commonality. The names themselves, however, are not what is important. Rather, it is the membership of the individuals in the mind-set and the nature of the commonality in the mind-set, which commonality may or may not be simple to discover.  The fact that the study focused on what might be considered a ’micro-topic,’ healthful pasta may be the cause both of richness of information about the micro-topic, but also harder-to-name subsets of this micro-topic, which is quite unified to begin with.

Table 5: Coefficients for TOP3 by mind-set and total panel. Only the strong performing elements for the mind-sets are shown

Table 6: Coefficients for BOT3 by mind-set and total panel. Only the strong performing elements for the mind-sets are shown.

Table 7: Coefficients for PRICE by mind-set and total panel. Only the strong performing elements for the mind-sets are shown.

The surprising resilience of cognitive economics patterns – price vs purchase (TOP3)

Figure 4 above suggests a monotonic increasing function of PRICE versus rated purchase intent on the 9-point scale. The underlying data is a scattergram from all of the respondents. The curve shows the smoothed relation, estimated by the smoothing function of the Systat statistical analysis program [11].  The pattern makes intuitive sense; for healthful pasta people say that they would pay more for a product that they like.

The same analysis can be done for the data from the three groups of mind-sets, derived in term from TOP3, from BOT3, and from PRICE, respectively. Figure 5 shows these nine smoothed curves. Again, Purchase Intent refers to the 9-point rating scale, Price refers to the dollars, and the data contain all the vignettes for the relevant mind-set. What is remarkable about Figure 5 is the dramatic similarity of the patterns, no matter how the respondents are divided. There may be slight variation, but the patterns are almost identical.

Figure 5: Relation between Price willing to pay (ordinate) and 9-point rating of purchase (abscissa), for three dependent variables (TOP3, BOT3, PRICE), each generating three mind-sets. The curves emerge from smoothing the raw data to show the underlying pattern.

Finding these mind-sets in the population

One of the hallmarks of Mind Genomics is that the mind-sets distribute in apparently random ways through the population, a fact which disturbs the traditional researcher searching for a co-variation between HOW THE PERSON THINKS ABOUT A TOPIC (the mind-set) and WHO THE PERSON IS, OR WHAT THE PERSON DOES, OR EVEN THE PERSON’S GENERAL ATTITUDES (GENERAL PSYCHOGRAPHICS.) In only very rare cases do we find strong co-variation between the mind-sets and other factors about a person. We should not be surprised at this lack of co-variation. The old adage ‘birds of a feather flock together’ does not seem to hold true when we focus on the deep variation between people on a topic, even people living in the same household. People may share some general attitudes, but it is rare, if ever, to find two people who agree completely on the granular aspects of any topic.

Table 8 shows the distribution of the three mind-sets for each dependent variable in terms of total, gender, age, and where the person lives.  Table 9 shows the cross-tabulation of the mind-sets. Any value greater than 40% of the total panel is darkened. For example, there are 68 males. We would expect an equal 1/3 distribution, of 23,23,23 for the three mind-sets. We have chosen the 40% cut-off as worthy of note. Thus, for 68 respondents, 40% means more than 27.2 respondents. We round to the lower whole number (27). It Is clear from both tables that knowing a person’s membership in either a geo-demographic subgroup (Table 8) or in a mind-set (Table 9) does not allow us to easily predict the person’s membership in any other type of mind-set.

Table 8: Cross-tabulation of membership in gender, age, and residence by respondents in the total panel, and in the three groups of three mind-sets each. There is no clear pattern.

Table 9: Cross-tabulation of membership in the three groups of three mind-sets each. There is no clear organizing pattern allowing prediction of mind-set membership from knowledge of other mind-set membership.

Predicting the profile of mind-set memberships using the PVI (personal viewpoint identifier)

During the past four years it has become increasingly obvious to authors Moskowitz and Gere that the practical applications of Mind Genomics would increase in number and scope when one could move beyond the limited number of respondents and apply the mind-set ‘clustering’ or ‘segmentation’ to the world at large.

Initial observations of how researchers were using segmentation revealed that the segments emerging from typical studies were very large, very general, and lacking the granularity. The typical segmentation appeared to emerge from the top down, so that one could divide people into general personas. The division would be made on the basis of questionnaires about general topics, leading to a limited number of personas, general groups of people [2][12][13]

The segmentation made interesting reading, the personas were described in detail, but there was no clear way to link these personas to the specific topic, especially when the topic is so granular as healthful pasta, and more so when the topic does not yet even exist.

Traditional segmentation is interesting, but really inactionable at the granular level simply because the segmentation is created with general propositions in mind, not with respect to a specific product. In a deep philosophical sense, one might say that traditional segmentation is imbued with sociology, organizing the world at large, the ‘nomothetic,’. In contrast, Mind Genomics segmentation is imbued with psychology, organizing the response to a granular topic, focusing on the individual, the ‘idiographic’.

In order to make the Mind Genomics segmentation more usable the analogy used is that Mind Genomics segmentation discovers groups of ‘mental primaries’, and not groups of people. The research would reveal these primaries, combinations of ideas, as shown by the three groups of mind-sets. Those mind-sets would be primaries, like color primaries. One needs a tool, a mental colorimeter, as it were, to assign a new person to one of these three primaries. With three sets of mind-sets, one for each dependent variable, the tool would have to assign a new person to one mind-set for each dependent variable.

The approach to create these assignments uses Monte Carlo simulation of the data, with 20,000 iterations. In the actual study, there were three such PVI’s created, one for each of the three dependent variables. The objective was t create a six-question tool for each dependent variable. The six questions are taken from the 36 elements used to create the mind-set segmentation for the dependent variable.  The six questions are answered on an anchored, two-point scale. With six questions and a two-point scale, the PVI questionnaire generates exactly 64 combinations. Each combination links with one of the relevant three mind-sets.

Figure 6 shows the introduction to the PVI. The respondent simply presses a link, and is taken to the introduction, which requests participation and then background information.  The information will be stored in a database for later use.

Figure 6: Introduction to the PVI, showing the request for permission and for background information. As of this writing (Summer, 2020) the link is:
https://www.pvi360.com/TypingToolPage.aspx?projectid=215&userid=2018

Figure 7 shows the actual PVI, with three sets of six questions. Each respondent will receive the same PVI, but the order of PVI’s will be randomize across respondents, as will the order of questions within each PVI.  With three PVI’s, one per dependent variable, there are six orders of the PVI’).  Second, within each PVI, the order of the questions will be randomized. For each PVI there are 6! Orders of questions, i.e., 6x5x4x3x2x1 or 720 orders

Figure 7: The actual PVI, for the three dependent variables.

Finally, Table 10 shows an example of feedback for a respondent, who completed the three PVI’s. The segment membership and the feedback are shown by the shaded cells. The researcher has the option to provide no feedback at all, to provide mind-set membership, or provide mind-set membership, feedback as well as information about the segments to which the respondent does not belong!

Table 10: Feedback for one respondent based upon pattern of answers to the three PVI’s. The shaded cells show the mind-set to which the respondent belongs, and the feedback for that mind-set.

Discussion and conclusions

When the topic of healthful pasta was first proposed some years ago and the study run (late 2012), the notion of Mind Genomics as a cartography was in its infancy. The research objective at the time was to determine what specific messages pertaining to pasta would prove to be most compelling. The effort at that time, only nine years ago, was to explore messaging, with the objective that here was a tool, Mind Genomics, which could provide a great deal of deal on many alternative messages. Up to then, the conventional wisdom was either to test single messages (so-called promise testing) or test fully formed concepts, polished, dense paragraphs, presenting a few ideas in a well-executed, almost seamless package. The idea was novel — discover through systematic experimental design powerful albeit mind-set specific messages.

When one looks at the richness of the data, the first question which emerges is ‘what do these data tell us about good carbs, or healthy pasta?’ This first question is the scientific aspect. The second question is ‘how do I use these data to help people enjoy a healthier diet’. The third question is ‘how do I used these data for commercial purposes.’  There are other questions of a research nature, such as ‘why do people fall into the mind-sets they do,’ ‘why is the price-purchase intent curve so similar across mind-sets,’ and the ever-recurring question ‘are these mind-sets stable, and does a person person’s mind-set ever change?’

What do these data tell us about good carbs?

This first question is answered by an abundance of data. The elements below are the strong performing elements. Rather than working with one theme at a time, and either saying that theme (e.g., brain power) is important or not important, Mind Genomics works with many themes. Thus, the information which emerges is far richer, a landscape of information rather than a single image, a single idea. The notion of the cartography, the landscape, is dramatically different from the more traditional, focused, hypothetico-deductive system, which considered one hypothesis or theme at a time, and through experiment attempting to falsify it [14] In Mind Genomics, the effort is to explore a broad landscape, not investigate each location off the landscape in a sequence of seemingly disconnected experiments, only later put together by a met-analysis

In terms of specifics, here are promising messages.

Look good & feel great at any age with a whole-grain fortified pasta diet… delivers plenty of energy & lifelong weight control

Put radiance back into your skin with whole-grain fortified pasta, which helps you sleep better and stop night-time problems

Stop feeling tired, eat whole grain fortified pasta for extra stamina …no need to cancel your evening plans anymore!

Vitamins and anti-oxidants in fortified whole-grain pasta improve brain power & thinking… reduce risk of cognitive impairment

Fortified whole-grain pasta contains complex carbohydrates – the “good carbs” which are essential to healthy brain function … boosts your brainpower … keeps you mentally sharp

Double pleasure without the guilt: whole grain lasagna is not only heavenly delicious, but healthy and nutritious too!

After a hearty whole grain fortified lasagna.no need to worry about weight gain

How do I use these data to help people enjoy a more healthful diet?

A growing issue in today’s world is the unhealthiness of the diet, the growing issue of obesity, and the need to create a better way to eat so that obesity and diet-related diseases such as diabetes do not ravage our society.  The discovery of strong performing messages allows those who communicate about healthful living to discover ‘what to say’ and ‘what not to say.’  One might surmise that any nutrition professional would know what to say, and that the information shown. Nutrition professionals know the science behind the food but may not know what messages convince. Indeed, as Table 11 shows, what appeals to the total panel may appeal strongly to only some of the respondents, and not to others.  It is here that the PVI, the personal viewpoint identifier, emerges with the power to assign a new person to a mind-set, and thus know at the start of the relationship with that new person the kinds of messages that will resonate.

Table 11: Strong performing messages may appeal to the total without appealing to all mind-sets or may appeal strongly to only one mind-set.

Are there ‘rules’ about how many mind-sets exist, and does a person’s mind-set change over time?

As Mind Genomics experiences increasing application, with more studies and more situations, questions of the number of mind-sets emerge, as well as the invariance of a mind-set. There is no fixed n umber of mind-sets. Each topic can be investigated in depth, to generate an array of different mind-sets. Unlike basic colors, of which there are only three (red, yellow, blue), mind-sets emerge for virtually any topic where decisions are made on the basis of information.  This study on good pasta shows one can take one topic and ‘drill down’ to create at least three mind-sets. There are smaller topics within good pasta, such as brain function, which themselves can generate mind-sets.   As for the invariance of mind-sets over time for a single person, that question remains a topic for the next generation of investigators.

Acknowledgement

Attila Gere thanks the support of the Premium Postdoctoral Researcher Program of the Hungarian Academy of Sciences

Reference

1. Barry J, Weinstein A (2009) Business psychographics revisited: from segmentation theory to successful marketing practice. Journal of Marketing Management25: 315-340.
2. Brower M (2018) Claritas (formerly Nielsen) Segmentation and Market Solutions. The Charleston Advisor. 18: 15-20.
3. Shelke K (2016) Pasta and Noodles: A Global History. Reaktion Books.
4. Webb D (2019) Pasta’s history and role in healthful diets. Nutrition Today. 54: 213-220
5. Moskowitz HR (2012) ‘Mind genomics’: The experimental, inductive science of the ordinary, and its application to aspects of food and feeding. Physiology & behavior107: 606-613. [crossref]
6. Moskowitz HR, GofmanA Beckley J, Ashman H (2006) Founding a new science: Mind genomics. Journal of sensory studies. 21: 266-307.
7. Moskowitz HR, Gofman A (2007) Selling Blue Elephants: How to Make Great Products that People Want Before They Even Know They Want Them. Pearson Education
8. Box GE, Hunter JS (1961) The 2 ^k—p fractional factorial designs. Technometrics. 3: 311-351
9. Gofman A, Moskowitz H (2010) Isomorphic permuted experimental designs and their application in conjoint analysis. Journal of sensory studies. 25: 127-145.
10. Jain AK, Murty MN, Flynn PJ (1999) Data clustering: a review. ACM computing surveys (CSUR). 31: 264-323.
11. Wilkinson L (2010) Systat. Wiley Interdisciplinary Reviews: Computational Statistics. 2: 256-257.
12. Kahle LR (1986) The nine nations of North America and the value basis of geographic segmentation. Journal of Marketing 50: 37-47.
13. Wells WD (1975) Psychographics: A critical review. Journal of marketing research12: 196-213.
14. Lakatos I (1976) Falsification and the methodology of scientific research programmes. In Can theories be refuted? (pp. 205-259). Springer, Dordrecht.