DOI: 10.31038/PSYJ.2021312

Introduction

Today’s world is awash in technology. The opportunities for making money through technology emerge when the benefits of the technology can be communicated either to users or to investors or both. In the world of startup accelerators, hungry investors, and the public which has learned to accept the blistering pace of innovation as part of today. It has become increasingly important to communicate one’s invention in a way which convinces the listener and excites the potential investor. It should not be surprising then that there are classes on creating the so-called ‘pitch deck,’ the document designed to excite investors. There is a lot less interest in finding out just what information might excite the prospective purchasers, simply because, in the scheme of things, starting up and raising money are more important than initial and then repeat sales. It is no wonder that many companies know what to say to investors about the financial aspects of the product but do not know what to say about the product itself. That is, these start-ups but also many later stage companies are at a loss to describe the mind of their investors and/or the mind of their customers. This means that, when challenged, virtually all startups struggle, having overlooked the relatively minor effort to really understand in a scientific, discipline manner, people’s reactions to what they are offering. With that in mind, the opportunity arose to use Mind Genomics for a new product idea, patented in the U.S. (Patent No. US 10,340,546 B1) on Jul 2, 2019 (Figure 1). Simply put, the device is a biocompatible, self-recharging micro battery, as small as a grain of rice, that can be safely and effectively implanted within the human body to power at least one implanted medical device, using the patient’s own body fluids. This micro-battery can communicate with the physician, giving advance warning of an imminent heart attack or the presence of a serious communicable disease. The idea is new, revolutionary, and is in the pre-seed phase. The question is what kinds of messages about this product will excite people. To reiterate, messages mean the messages about the product, in terms of what it does, and what that means to the world of health. The Mind Genomics process described here required about four hours of investment, from start to end, a minor monetary investment. The paper describes the process, shows what was learned, and underscores the opportunity to use Mind Genomics and allied sciences of the mind to increase the likely success of an early-stage venture. Similar approaches have been over the last decade to create products, to help horticulture, etc. [1,2].

Figure 1: The patent for the biofuel cell.

Mind Genomics as an Emerging Science

Mind Genomics traces its history to the marriage of several disciplines, beginning with experimental psychology (especially psychophysics), merging with mathematical psychology (conjoint measurement), statistics, and applied consumer research [2]. The objective was and remains to understand the way people make decisions about topics, these topics being of the everyday [3-5]. Mind Genomics focuses on the specifics of a topic, a focus which forces the researcher to deal with the granular aspects, rather than with the grand vision. In other words, the researcher studies the actual product or service itself and, specifically, the minutiae that might otherwise be overlooked in the grand sweep of the ‘elevator pitch.’ Indeed, the granularity of the information provided by Mind Genomics in effect generates a ‘wiki of the topic,’ or an ‘MRI of the mind’ with respect to the product. The importance of such granularity of knowledge is overlooked again and again in the heat of excitement, the haste to get investors, and the omnipresent motivator in business, FUD (fear, uncertainty, doubt). The best way to understand the application of Mind Genomics to startups and pitches, the warp and woof of the venture world, may be through a worked case example, where there is little knowledge at the start of the effort. The body of the paper shows how to understand ‘what works,’ at the level of the structure of the messages (contribution of the individual elements and their combinations), and structure of the mind (influence of who the person is, how the person thinks). The result of the effort is a tool, the PVI (personal viewpoint identifier), which provides the inventor(s) or group raising capital the necessary knowledge of what specific aspects of the invention most promise market success.

As the reader follows the steps, it should be kept in mind that the process takes far longer to describe than it does to execute.  Furthermore, the speed of the processes increases with practice and as the person becomes increasingly facile with the method.

Step 1: Choose the Topic. In this Case the Topic is the Patent, and Specifically What it Does

Step 2: Ask Four Questions Which Tell a Story

It is as this point that many people become ‘stuck, for the simple reason that people are not accustomed to think in a structured and creative manner, viz., to think analytically while at time having fun doing so.

Although one’s obsessive nature may demand that there be five, six, or even more questions, a key aspect of Mind Genomics is the requirement to keep the questions to a number that can be managed easily. In the early days, around year 2000, it was possible to do long interviews by web, with interviews lasting 15 minutes. The longer time required evaluating 48 to 60 vignettes is no longer feasible unless the respondents are well recompensed. Furthermore, as a practical issue, the longer studies with more questions and more answers (viz., 6 questions x 6 answers/question, or 36 elements) somehow ‘never seem to get done.’ They implode because everyone feels that she or he must make a ‘contribution’ to feel part of the process. Too many cooks do really spoil this broth. It is better in terms of process to run three smaller studies in one day, each small, but iterative, rather than that one comprehensive study which never seems to reach the execution phase because of yet another revision and the need for the different parties to agree. The smaller number of elements in the 4×4 design,16 elements, removes much time wasting, back and forth discussion, no matter how deeply people feel that they must discuss and ‘get it perfect’ before the effort, which itself will be done much more quickly, about 1-2 hours.

Table 1 shows the four questions and the 16 elements. These elements may or may not be the correct. One need not know. The underlying Mind Genomics process is quick, powerful, inexpensive. There is no need to be right. One needs to do the study. The data will quickly reveal which types of elements perform well. Subsequent iterations, when they occur, are simply built on the winning elements from the iteration before, the losers discarded, and new elements tried.

Table 1: The four question and the four answers (elements) to each question.

Step 3: Combine the Elements into Vignettes, According to an Underlying Experimental Design

The experimental design specifies exactly 24 combinations, some having two elements, some having three elements, and the remaining having four elements. A vignette can have at most one element or answer from a question, but in many vignettes an answer from one of the four questions is missing. The strategy for this ‘incompleteness’ is that the structure of the combinations, the 24 vignettes, is such that all 16 elements are statistically independent of each other. That means that the absolute contribution of each of the 16 elements can be computed from the regression, making the approach of Mind Genomics exceptionally powerful in the nature of the information that it delivers. Finally, each respondent evaluates a totally unique set of the vignettes, created by permuting or shuffling the elements [6]. The benefit of that permutation approach is that a single study can cover a lot of the different vignettes. The pattern emerges much like the pattern of an MRI in medicine, emerging after combining the different snapshots of the mind, each snapshot from one of the experimental designs. The happy outcome is that the Mind Genomics process needs no basic understanding of the topic. The ease of setting up the Mind Genomics study (minutes), the cost (low), the speed (hours from start to end) make it possible to iterate several times to understand the topic, not by being right at the start, but by iterating to a solution almost painlessly.

Step 4: Select a Rating Question Pertaining to the Topic

Traditional approaches have asked simple, unidimensional questions, either using a rating sale (viz., how interested are you, 1=not interested … 5=interested; how much would you pay? 1=nothing …5= 10$). New methods include selecting an answer from a group of possible answers (viz., 1=sad, 2=happy, 3 = irritated, 4=curious, 5=excited), and so forth.

For this study we explored a two-dimensional answer, dealing with believing the information and buying the product, respectively. The reason was the relevance of these two dimensions both to understand the response to the product/service as marketers and the information to the investors, demonstrating knowledge of the product, and its economic potential.

Here is a new product to help you. It is a small fuel cell for your body!! Read each combination and rate on this scale:

1 = No way

2 = Believe: NO; Buy: NO.

3 = Believe: NO, Buy: YES

4 = Believe: YES; Buy: NO

5 = Believe: YES; Buy: YES

Step 5: Create a Short Classification Questionnaire to Further Understand the Respondent

These small-scale studies are meant to provide quick, deep answers to what interests the respondent, doing so in a study of three minutes or faster. The classification questions, answered at the start of the study, require only information about the respondent (gender, age), as well as an optional third question, with up to four answers. In our Biofuel study, we chose to ask the respondent about her or his medical situation (not relevant, think about situation occasionally, think about situation frequently, or currently monitoring a condition, respectively.)

Step 6: Launch the Product among a Group of Respondents Who are Members of a Large Panel (>10 million) for a 3-minute Experiment in the Form of an Interview

It is a false economy to use one’s own respondents as a panel unless they constitute individual who are otherwise difficult. Both in terms of time and ultimately in terms of money, it is far more practical to use external panels, provided by companies which charge a reasonable, relatively low fee on a per-respondent basis. The panel respondents in this study were provided by Luc.id, Inc., in the United States. The respondents provided can have any desired geographical, age, and other qualifications. The requirement was to work with respondents 50 years or older. These would be the individuals likely to need the product. The self-reported health concern (question 3) would be able to provide the response by individuals who say that they are actively monitoring conditions. Once the researcher has thought about the problem the mechanics involved in setting up, launching, and receiving data are virtually automatic, programmed, simple, fast, and after one or two experiences error-minimizing. It is not the doing, but the thinking which is difficult. Structured thinking of this type, no matter how seeming obvious it turns out to be, must be a conscious, formal part of the development of a communication program about WHY the new idea, and the economic benefits, here specifically, WHO wants the product. The Mind Genomics process forces the respondent to think deeply about the problem, and think quickly, both being important. There is no excuse in Mind Genomics for delay since the process is simple, templated, fast (hours), cost-effective, all leading to iterations. One need not know anything at the start of the iterations, but by three, four, or five iterations, one will have assembled the powerful insights and precise messages.

Step 7: Prepare the Data for Analysis

Each respondent generates 24 rows of data, one row for each of the 24 vignettes.

For our evaluation of the selling messages for the new biofuel cell, we create three key dependent variables:

Believe/Buy (Rate 5). This variable will be 100 when the rating for a vignette is 5, and 0 otherwise.

Neither Believe nor Buy (Rate 1,2). This variable will be 100 when the rating for a vignette is 1 or 2, and 0 otherwise.

Response time. This is the response time in seconds, to the nearest 10^th of a second, for each vignette.

Table 2 shows the data from the study, data ready for the statistical analysis below. The table shows three rows of data from each of two respondents. The table begins with the row number, the panelist number, and the structure of vignette, viz., which questions does the vignette comprise. The middle of Table 2 comprises 16 columns to code the elements, with the value 1 corresponding to the element present in the vignette, and the value 0 corresponding to the element absent from the vignette. After the 16 columns come two columns, Rating from the 5-point scale, and the measured Response Time, respectively. Beyond 16 columns to code the input variables and the two original responses (Rating, Response Time), we find two new data columns. The first new data column corresponds to the most positive response, rating 5 (Do believe, Will buy). When the respondent selected the rating ‘5’ on the scale, this cell for the vignette in column Rate 5 is given the value 100, but when the respondent selected rating 1,2,3, or 4, this cell is given the value 0. The second new data column corresponds to the two most negative responses, rating 1 or 2. When the rating is 1 or 2, the cell for the vignette in this column is give the value 100. When the rating is 3,4, or 5, the cell is given the rating 0. A small random number is added to the values 0 or 100, simply to introduce some small but necessary variability in the binary ratings, in preparation for the analysis.

Table 2: Example of a data matrix showing three rows of data from each of two respondents.

Step 8: External Analysis – Do the Groups of Respondents or Vignettes differ from Each Other?

By looking at the structure of the vignette we can learn about how the respondent makes decisions. Recall that each respondent evaluated a unique set of 24 vignettes, and that across the set of 121 respondents and 2904 combinations there are relatively few duplicate combinations. The strategy of covering a wide number of combinations (covering the ‘design space’) avoids repeat combinations in favor of more the combinations to be testing. Thus, it is, as yet, difficult to compare two or more groups on the score of the same test stimuli simply because there are few test stimuli. The objective in the design was to create as many unique vignettes as possible. It is possible and instructive to compare the averages of three key dependent variables across all key groups, if only to get a sense of the average values of the key dependent variables. Table 3 shows the averages for the three variables across all the relevant respondents in the group. Table 3 show two additional pairs. The first is the averages from vignettes 1-12 vs the averages from vignettes 13-24. This information tells us whether there is a change in the criteria as the experiment proceeds, with the respondent evaluating 24 vignettes. The second pair of data comes from the responses to vignettes read quickly (non-engaging, response time operationally defined as less than 2.25 seconds) vs read slow (engaging, response time operationally defined as more than 2.25 seconds).

Table 3: Average of the three dependent variables for all vignettes appropriate for the subgroup.

The Total panel generates these three averages across all 2904 vignettes.

Rate5 = 22, viz., 22% of the vignettes are assigned a rating of 5, so that 78% of the vignettes are assigned ratings of 1, 2, 3 or 4, respectively.

Rate12 = 21%, viz. 21% of the vignettes are assigned a rating of 1 or 2, so that 79% of the vignettes are assigned ratings of 3, 4 or 5.

RT = 4.5 meaning that on average the respondent rated a vignette 4.5 seconds after seeing the vignette. This suggests that it took the respondent about 108 seconds, or nearly two minutes to evaluate all 24 vignettes. The rest of the time was occupied with self-profiling classification, and so-forth, making the total time of about three minutes quite reasonable.

Table 3 shows the average responses of the three dependent variables, across Total Panel, Gender, Age, Health Concern (question #3 in the classification), and finally three mind-sets which emerged by clustering together respondents showing similar patterns of strong positive responses towards the elements (mind-sets from clustering). The averages are remarkably similar, except for the responses obtained for the first 12 vignettes vs the second 12 vignettes, and the pattern of rejection (Rate12), for those vignettes read quickly (26% of the vignettes rejected) vs those vignettes read slowly (19% rejected).

Step 9: External Analysis – Does the Structure of the Vignette “Drive the Rating”?

Our first external suggested minor differences across groups, except for the order of the vignette in the set of 24, or the rate of reading the vignette. It helps to understand how the structure of the vignette, the types of elements combined, drives the dependent variables, and whether there are any group differences. Knowing the impact of the interaction gives the researcher a sense of which types of combinations will, in general, drive positive or negative responses from the respondent (or customer or investor). In turn, Knowing the nature of the interaction between structure and subgroup with respect to response time tells us the groups who will be paying close attention to the messaging. It is important to reiterate that, at least yet, we have not seen detailed information about the elements. That information will come later. Our goal here is to better understand the types of messages as they are perceived by the respondents and as they engage the respondent in terms of paying attention. It is also important to note that the respondent does not think in terms of the structure of the vignette, since on average the respondent pays about 4.5 seconds attention to each vignette, sufficient time to read and assign an intuitive, ‘gut reaction.’ Longer response times than 4.5 seconds suggest that the vignette structure presents information which arrests the respondents speed through the interview and may represent a structure of information which strongly engages the respondent. Table 4 presents the averages of the three dependent variables (columns) by the 11 different structures used by the 4×4 Mind Genomics design. Table 4 is divided into three sections, Table 4A for the Rate5, Table 4B for the Rate12, and Table 4C for Response Time. The rows in each section of Table 4 are sorted by the Total Panel, allowing us to get a sense of what interests the respondents, what turns them off, and what engages them. In turn, the subgroups give us a sense of any key differences. To make the inspection of Table 4 easier we have darkened the key cells, respectively cells with averages of 30 or higher for the two binary transformed variables (Rate5, Rate12), and cells with response times of 5.0 seconds or longer. The data reveal patterns very quickly, patterns relevant to investors and marketers alike. The most striking is the importance of the combination of What it is and How it will be paid (AD). The least important is How it works and How will it be paid (BD). The ‘magic’ comes from the combinations of the elements, and that certain combinations are simply strong. When it comes to an outright rejection (Rating12, Not Believe, Not Pay), most of the cells are low. There are a few exceptions, especially for those who are monitoring a condition. When the vignette contains an element of ‘What it is’ and ‘Why should I use it,’ those monitoring a condition find this offensive. In fact, they find both ‘What Why’ and ‘How Why’ to be turnoffs, something that should be remembered when talking to a prospective buyer, but also key information to present to the funding group. Finally, respondents take different amounts of time to process the information. The most engaging vignettes are those with two elements, What How, Why Pay, and How Pay, respectively. The longer vignettes, the one with four elements, but a few with three elements, tend to be glossed over, or at least are less engaging. These results suggest that a great deal of information about the proper presentation can be gleaned simply by understanding the pattern of responses for the three key dependent measures, Rate5, Rate12, and Response Time. Despite the fact that we do not yet know the specific messages to put into the vignette, we should have a sense that the optimal vignette will incorporate ‘What it is’, and ‘How it’s paid for.’

Table 4: How the structure of the vignette ‘drives’ the ratings. The table shows the average values for three key dependent variables, by structure of the vignette (row) and by key group (column).

Step 10: Internal Analysis, to Understand How Elements Drive the Dependent Variables

The original rationale for Mind Genomics was founded on the premise that people could not tell the interviewer what guided their decisions but would likely try to please the interviewer by confabulating one. Such efforts would be especially obvious when the respondent would be asked about the criteria used to guide decision in the routine behaviors, those labelled ‘System 1’ by Nobel Laureate Daniel Kahneman [7]. According to Kahneman, but clearly observed every day, we make thousands of decisions, perhaps many more, simply during our daily lives, doing so virtually automatically. To create a science of the everyday requires an approach beyond observation (too limited, too expensive), and beyond questionnaires and surveys (subject to judgment biases, memory biases, etc.). First a short recapitulation is in order, in order to lay out the rationale for these next steps in the analysis. Mind Genomics works by presenting the respondent with the different vignettes, created by experimental design, doing so in a rapid pace. We saw that the average time for evaluation was approximately 4.5 seconds, from the time that the vignette appeared, and the judgment was assigned. During these 4-5 seconds, on average, the respondent read, thought (almost automatically), and rated the vignette. When these vignettes are presented rapidly, and when the vignettes are created by experimental design, it becomes difficult to think; one simply responds at an intuitive level. The happy result is judgment untainted by most of the cognitive biases which pervade the everyday research. One cannot change the judgment criterion to accord with the specific nature of an element (viz., price versus feature vs benefit, etc.) In contrast, the conventional, one-at-a-time approach the respondent can switch criteria rapidly, depending upon the nature of the element so as to give the ‘right answer.’ Not so with Mind Genomics, which combines these elements into wholes, gestalts, vignettes, each judged, de facto, by the same criterion. The respondent ends up neither able to nor even wants to be ‘correct’ or ‘consistent.’ The respondent simply wants to finish the task, typically doing so in a state of relative indifference, and thus answering honestly, or at least answering in an intuitive way. The benefit is that attempts to ‘game the system’, to ‘please the interviewer,’ to ‘get it right,’ are simply not possible. Armed with the foregoing, we now look at the deconstruction of the vignettes into the part-worth contributions of the elements, doing so by key groups. Each group was self-defined, except by the mid-sets. The mind-sets were discovered by doing the modeling at the individual respondent level, creating 121 models, and then clustering together individual respondents showing the same pattern of coefficients for their 16 elements as those elements drove ‘Rate5,’ viz Believe/Buy. The reader is referred to the in-depth treatments of the clustering method (k-means) in a variety of published papers [8-10].

The actual deconstruction of the vignettes is done by the statistical method of OLS, ordinary least-squares regression. The regression attempts to relate the presence/absence of the 16 elements to the dependent variable. The equations are written as follows:

Rate5 (Believe/Buy) = k₀ + k₁₍A1) + k₂(A2) … k₁₆(D4)

Rate12 (Do Not Believe/Will Not Buy = k₀ +k₁(A1) + k₂(A2)… k₁₆(D4)

Response Time = k₁₍A1) + k₂₍A2) … k₁₆(D4)

The OLS regression uses the entire data set from the relevant respondents to estimate the additive constant (k₀) and the 16 element-linked coefficients (k₁ … k₁₆).

It is important to keep in mind that there is an additive constant for the regression model for Rate5 (propensity to believe/buy in the absence of elements), and for Rate12 (propensity to not believe / not buy in the absence of elements). For response there is no additive constant because there is no propensity to respond to a test vignette unless there are elements in the vignette.

Finally, one could create equations without additive constants for both Rate5 and Rate12. The conclusions would be the same, but the data would be someone less easy to understand.

Armed with the foregoing approach, we use the standard method of OLS regression to create models for nine defined subgroups. These are the ones that are most meaningful:

1. The Total Panel… all respondents who participated.
2. The three mind-sets MS1, MS2, MS3, which emerged from clustering together respondents with similar patterns of coefficients, based upon the models for Rate5 (Believe/Buy). Extensive studies using Mind Genomics suggest that the groupings emerging from this type of clustering generate clear, consistent, and interpretable patterns, pointing to radically different ways of thinking about a topic. These are the so-called mental primaries for a topic, albeit primaries emerging from patterns of response to a granular topic, rather than grand patterns across many topics. Mind Genomics works at the granular level, where the relevant stimuli are likely to be clear, and the relevant behaviors likely to emerge.
3. Gender, age, health groups, those who have no concerns (answer 1 to question #3 in the classification), and those who are monitoring a condition (answer 4 to question #3 in the classification).

Step 10: Define the Meaning of the Coefficients, and Present Data in an Interpretable Format

The analysis in Mind Genomics generates a great deal of data because each element generates nine coefficients, one coefficient for each subgroup. Thus, the analyses involve three different groupings; three dependent variable (Rate5, Rate12, Response Time), nine key subgroups, and 16 elements. This total 3x9x16 or 432 cells of data to inspect to uncover strong performing elements and discern patterns.

To make the analyses easier we do the following:

1. Present the subgroups in a new order, with Total and mind-sets first, because it will be with mind-sets that the major group-differences emerge.
2. Blank out any coefficient which is 0 or lower, because the element with that coefficient is not a driver of Believe/Buy or Not Believe/Not Buy, respectively. A negative coefficient for Rate5 (Believe/Buy) means that the element may either be irrelevant (originally rated 3 or 4) or actively push away (originally rated 1 or 2). In turn, a negative coefficient for Rate12 (Do Not Believe / Would Not Buy) means that the element may be irrelevant (originally rated 3 or 4) or actively push away (originally rated 5)
3. Shade all cells with coefficients of 6 or higher for Rate5 or Rate 12. Shad all cells with response times for the element of 1.5 seconds or longer.

Armed with this information, we can now look at the strong messages for the Biofuel invention. Table 5 is sorted in descending order for the positive elements of the three mind-sets. Occasionally an element appears twice in a table, scoring strongly in two of the three mind-sets (viz., coefficient of +6 or higher).

The rationale for presenting the data in descending order by mind-set is that only through mind-set do we see sufficient strong performing elements which, in turn, seem to cohere together to tell a meaningful ‘story.’ Keep in mind that the mind-sets are created through purely statistical methods, without any connection to what the mind-sets or clusters really mean. The researcher’s task is to select the minimum number of clusters which make sense and uncover the latent pattern. For our data, the assignment of the respondent mind-sets showed the clearest pattern when the clustering was done using the coefficients for Rate5 (believe/buy), and when three mind-sets were extracted. The three clusters thus become three new, non-overlapping groups. The separate data was used to create models for Rate5 (the original basis of the clustering), as well as models for Rate12, and models for Response Time. The compositions of the three mind-sets are fixed at after the clustering analysis. The compositions of the other groups are fixed at the time of classification, viz., the before the actual experiment. Once we know the respondents in each group, it is straightforward to create a summary model or equation for each group for each of the three dependent variables. The final act is to create the summary tables, doing so based on the three mind-sets, which carry most of the interpretable patterns. The other subgroups are presented as backup data. The actual interpretation of the data is not relevant for this research exercise, but is extremely relevant for the inventor, marketer, and the investor. Through understanding what specific aspects do very well (viz., high coefficients for a mind-set) it becomes straight to identify a lot more of the potential of the invention. One knows what to say, how to say it, and now to whom. The researcher may stop at one iteration or move quickly (or slowly) to the next iteration, simply deleting poorly performing elements and them. Over time guesswork turns into solid knowledge.

Step 11: Finding these Mind-sets in the Population

As Table 5 shows, especially Table 5A, it is in the mind-sets that one discovers the important messages. It is clear in this study as in most Mind Genomics studies. that the mind-sets are far more important than anything else about the respondent . Knowing the mind-set enables one to present the necessary information to engage that mind-set, to convince that mind-set, and to avoid saying the wrong thing, something that might immediately turn a prospect into a rejector. The traditional thinking of market researchers, political pollsters and the like is that ‘birds of a feather’ think the same way. That is, lacking a deeper understanding of how people think about the world of the everyday, those looking to understand why people differ from each other in known ways generally look at WHO the person is, what the person may THINK in general terms about a topic (e.g., attitudes towards health, or how a person BEHAVES (viz., what does a person search for on the Internet when exploring a topic). The effort towards classifying a group of people by WHO, by THINK and by BEHAVE is significant, usually reserved for large-scale problems. Mind-Genomics, dealing as it does with the granular aspects of the world, and often with the very ordinary (not in the case of the Biofuel Cell!) suffers from the paradox of being able to discover important mind-sets in the population in the short space of a few hours, but then grapples with the problem of generalizing this discovery so that it moves beyond simply a scientific fact, to be applied, whether for knowledge building, or action generating or hopefully both. The mind-sets are reasonably clear from Table 5 and could become the basis of three distinct sets of communications, whether to a customer or to an investor. ‘What to say to convince’ becomes a matter of research, not a matter of a dearly held opinion, possibly irrelevant or even worse, possibly counter-productive. Just consult Table 5B to see what ‘doesn’t work.’ On the other hand, what is the marketer to do when the distribution of the three mind-sets in the population is similar for each key subgroup, whether gender, age, or even monitoring a condition (Question #3, answer 4)? Table 6 suggests that it will be almost impossible to find the way to assign a new person to the proper mind-set. That impossibility discourages the wider use of a rapid, inexpensive, iterative, knowledge-developing system.

During the past four years, since 2016, authors Gere and Moskowitz have introduced and applied a new approach to assign a new individual to one of the mind-sets. The approach is known as the PVI, the Personal Viewpoint Identifier. The PVI uses the table of coefficients (Table 5A), summarizing the coefficients for the mind-sets. The underlying thinking is that the small study presented here provides insight into the basic mind-sets of a topic, viz., combinations of ideas which naturally go together, as can be seen with the small population. These can be likened to ‘mental primaries’, albeit primaries for a limited, quite granular topic, empirically uncovered from experiments. The issue is now to discover the distribution of these primaries across the world, and the lability of these primaries as a function perhaps of experience, of life-situation, etc. A secondary set of goals, not discussed here, is to relate these ‘mental primaries’ to relevant behaviors exhibited by people, viz., the expression of these primaries in everyday life. A third set of goals, also not discussed here, is to these ‘mental primaries’ primaries to genes, to uncover links between genetics and the mind, for defined topics where there is a suspicion that one or another gene might be involved in certain behaviors. Our focus here will simply be the presentation of the PVI, as a ready-to-use tool, one based upon the actual elements used to define the mind-sets. The PVI does not need any theoretical bridge between the mind-sets and the PVI composition. The components of the PVI are the same elements used in the study, with perhaps a slight editing to generalize them where needed. The mathematics underlying the PVI first creates noisy data by adding random variability to the original means, doing so many times, according to a Monte Carlo system. The analysis uses a decision tree to determine which elements and what weighting factors best assign the average individual in each mind-set to the correct mind-set. The scheme which works best across the thousands of perturbations is the scheme used for the PVI. The output emerges in the form of a link to the specific PVI designed for the study. That is the granular data are used as inputs to the assignment program. The PVI emerges with six questions taken from the elements of the study, and put into the form of a statement, with one of two answers. The pattern of the six answers assigns the new respondent to one of the three mind-sets. In studies comprising two mind-sets rather than three, the pattern of the six answers assigns the respondent to one of the two mind-sets. Figure 1 shows the set-up page, configured to be used with the Mind Genomics output. The set-up uses the data from the Mind Genomics study but requires the researcher to assign names to the mind-sets, to create a Yes/No rating question, and an introduction to the task given to the respondent. The actual PVI set-up is presented as an Excel® worksheet, in color, to make the process easy to do, fast, and subject to fewer errors. As part of the set-up, the researcher can specify a video and/or a landing page to which the respondent is immediately directed after the mind-set has been assigned by the PVI program for the specific individual (Figures 2-5).

Table 5: Coefficients relating the presence/absence of the 16 elements to the three dependent variables (5A for Rate5, 5B for Rate 12, 5C for Response Time). The table shows the total panel and key subgroups. Only positive coefficients are shown. Strong performing elements are shown by highlighted cells (coefficient +6 or higher for Rate5 and Rate12; response time of 1.5 seconds or longer).

Table 6: Distribution of the total panel and three mind-sets across the different self-defined subgroups in the population.

Figure 2: The researcher set-up for the PVI, using the output from the Mind Genomics study.

The orientation page for the respondent, as well as background information. The PVI can be configured to send the data to a database, as well as to the respondent, and to a staff person. The orientation page takes approximately 30 seconds to complete. Individual fields of data, e.g., gender, age, telephone, etc. can be suppressed to ensure privacy. Figure 3 shows the actual PVI, with four background or attitude questions, and the six questions emerging from the Monte Carlo algorithm. The PVI questionnaire also takes about 30 seconds to complete.

Figure 3: The orientation page for the PVI (personal viewpoint identifier) The link to the PVI is: https://www.pvi360.com/TypingToolPage.aspx?projectid=1267&userid=2018.

Figure 4: The actual PVI questionnaire, beginning with three questions about one’s attitudes toward health and finishing with six questions.

Figure 5: The feedback from the PVI. The data are stored in a database, along with the information from Figures 2 and 3. The respondent’s mind-set determined by the PVI is shaded (MS1).

The feedback for one respondent. This respondent was assigned to Mind-Set 1 based on the pattern of the responses. Figure 4 shows spaces for both a landing page and a link to a video stored in YouTube. Thus, at the time of deploying the PVI, the researcher may show the respondent a video and drive the respondent to a landing page. In the world of social issues, the PVI becomes a game, wherein the respondent finds out about himself or herself and is exposed to messages through video or landing pages.

Discussion and Conclusion

The ingoing rationale for the paper was the observation that in both the private sector with start-ups and in the public sector with major issues, there seem to be few ways to obtain affordable, solid, actionable data in the realistic framework of need for speed and clarity. The ‘soft’ data from people, used to back up major investments and scientific breakthroughs, seem again and again to be remarkably weak. ‘Subjective, soft data’ are perceived to be a necessary nuisance, either impossible to obtain because the data would take years to obtain or because the data simply is not valued. Often the data presented is qualitative, coming from a limited number of respondents or participants in a set of focus groups or depth interviews. Those data are important to set the stage, but they do not give the inventor, the business owners, the investors, or the government a sense of the ‘there there,’ in the immortal quip of Gertrude Stein. There is no need to discuss the specific data from the study. The data are simply of the type that the process delivers, with the nature of the data similar in general form from study to study, but sufficient in depth for any study to provide the necessary guidance. The study is the first of its kind, from the group associated with the inventor, author Samuel Messinger. Rather than polishing the results, it seemed most appropriate to take the results and explicate them, step by step, so that the paper becomes a guide to interpreting the results, a vade mecum. One outcome is that the reader gets a sense of how to do the study and what will emerge from the study step by step. The other outcome is the ease with which the reader can look at the data tables, to identify what to say, what not to say, whether it matters to whom, and ‘who are the relevant whom’.

Note

The Mind Genomics program (BimiLeap) is available at www.BimiLeap.com.

The Personal Viewpoint Identifier (PVI) is available at www.PVI360.com.

Acknowledgment

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

References

1. Levin LA, Langer KM, Clark DG, Colquhoun TA, Callaway JL, et al. (2012) Using mind genomics® to identify essential elements of a flower product. Horticulture Science 47: 1658-1665.
2. Gofman A (2012) Putting RDE on the R&D Map: A Survey of approaches to consumer-driven new product development. In: A. Gofman & H. R. Moskowitz (Eds.). Rule Developing Experimentation: A Systematic Approach to Understand & Engineer the Consumer Mind, 72-89. Bentham Books. https://doi.org/10.2174/97816080528441120101
3. 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. [crossref]
4. Moskowitz HR, Gofman A, Beckley J, and Ashman H, (2006) Founding a new science: Mind Genomics. Journal of Sensory Studies 21: 266-307.
5. Moskowitz HR, Porretta S, Silcher M (2008) Concept Research in Food Product Design and Development. John Wiley & Sons,
6. Gofman A, Moskowitz H (2010) Isomorphic permuted experimental designs and their application in conjoint analysis. Journal of Sensory Studies 25: 127-145. https://doi.org/10.1111/j.1745-459X.2009.00258.x
7. Kahneman D (2011) Thinking, Fast and Slow. Macmillan.
8. Gere A, Zemel R, Papajorgji P, Moskowitz H, (2019) “Candy Is dandy”: The mind of sexuality as suggested by a Mind Genomics experiment. In Sex, Smoke, and Spirits: The Role of Chemistry pg: 17-31, American Chemical Society.
9. Jain AK, Dubes RC (1988) Algorithms for Clustering Data. Prentice-Hall, Inc.
10. Porretta S, Gere A, Radványi D, Moskowitz H (2019) Mind Genomics (Conjoint Analysis): The new concept research in the analysis of consumer behaviour and choice. Trends in Food Science & Technology 84: 29-33.

DOI: 10.31038/PSYJ.2021311

Abstract

Respondents in Hong Kong evaluated sets of 24 unique vignettes about the Hong Kong Goldfish Market, rating each vignette on ME (agree) or NOT ME (disagree). The vignettes were created by systematically combining elements (messages) from four different categories, into small combinations comprising 2-4 elements each. The experimental design prevented the respondent from ‘gaming’, viz., giving the appropriate answers because the vignette comprises different messages. The deconstruction of the responses by regression showed the contribution of every element to three dependent variables; agree, disagree, and response time (engagement) revealed different types of decision rules for agree versus disagree. Two mind-sets emerged, based on the pattern of responses to the 16 elements; those interested in low price, and the afficionados. The study showed that the engagement times, viz., response times to individual elements were far longer for Mind-Set 2 (afficionados). The study shows the power of simple experiments to create a database of people’s perceptions of the every-day experiences, a ‘Wiki of Daily Life’.

Introduction

An exploration of the world of experimental psychology, social psychology, sociology, and anthropology reveal an ongoing interest in the world of the everyday specially by researchers in sociology and anthropology, as well as social psychology. There is some experimentation by social psychologists and experimental psychologists, generally of a limited nature, often to understand a general principle, and usually with situations that are unusual. The crossing of experimentation with the world of the ordinary is not particularly common, because experimentation looks for general principles of behavior.

During the past two decades, the world of experimentation in psychology has come to embrace new areas, not perhaps in a formal matter, but at least accepting research which shows the everyday. The research may not appear in highly rated journals because of the quotidian nature of the topic, but nonetheless the research plays an important role in our understand of the lives of contemporary people. Some of the work may be found in the topics covered by behavioral economics, other part of the effort may appear in topics covered by consumer research.

Recently, author Moskowitz has suggested that a new discipline of experimental psychology be developed, one called Mind Genomics [1] Moskowitz et. al., 2007). The effort is to understand the drivers of decision about the quotidian topics of everyday life, not so much to develop grand schemes of behavior, but rather simply to catalog the myriad different aspects of daily behavior. The effort looks at the world of the granular, from the bottom up, from the specific, limited topics which make up the warp and woof of our daily behavior, topics which would not ordinarily be thought to be the appropriate topics of psychology. The topics might be those of interest to an anthropologist who describes these topics in a discussion of behavior, but issue to explore the topic in depth, to understand the facets of a topic, and the different ways to look at the topic.

Early efforts to study daily behavior focused on things, primarily things one might eat or drink. The objective was to understand the different patterns of preferences regarding food. The underlying reason for choosing food was that studies of preferences for food and studies of eating in restaurants and dining halls were well known, well accepted by the world of scientists [2]. The effort soon expanded to other situations, usually products (e.g., financial industry; [3], but eventually moved into experience itself. As far back as 1957, for example, sociologist William Foote Whyte was recording the sociology of the everyday, producing such classics as the Organization Man [4]. The effort continues, with such evaluation of landmark groups, such as the Baby Boomers [5].

Science need not be focused on the traditional topic, such as foods, topics with a long intellectual and academic history. Anyone with a sensitivity to social issues is typically interested both in general patterns and in specific stories. The general patterns are usually reported in good scientific fashion, with appropriate statistics, general conclusions, and foundational knowledge. The stories are often more interesting, sketchy but real world, and used to make the general science more interesting. The science of Mind Genomics, used here to evaluate a common pet market, the Hong Kong (HK) Goldfish Market, provides a way to introduce rigor into the study of what otherwise might simply an interesting vignette, introduced as part of a deeper presentation of a culture, a society, or a hobby. The HK Goldfish Market has been the topic of studies, primarily of a sociological/business nature [6,7] There does not seem to be a deep understanding of the response of individuals to the both the market itself, and to their spending patterns, perhaps because of the localized nature of the topic, and the fact that the ‘mind’ of the individual regarding the HK Goldfish Market is simply not sufficiently important in the world of science.

Despite its minor position as a topic of study, the HK Goldfish Market is an important market. According to the Wikipedia article ‘History of Goldfish Market, Hong Kong, 2012’:

Hong Kong is one of the leading exporters of Goldfish and other tropical fish for aquarists and fish keepers around the world. In Tung Choi Street Goldish shops have congregated for many years…Originally the interest in fish, particularly goldfish, in Hong Kong was related to the needs of Fung Shui, the ancient Chinese system to bring harmony to a house. Under this system Goldfish are particularly important so therefore there has always been a demand for goldfish in Hong Kong more than typical in other countries… During the 1970s and 1980s the keeping of goldfish alongside other types of tropical fish such as butterfly fish became a very suitable hobby for the majority of Hong Kong’s population who lived in flats in high rise apartments. Without the space to keep pet cats or dogs the keeping of tropical fish, both freshwater and less commonly sea fish, became very popular. (Source: Wikipedia, 2020)

The Mind Genomics Approach to Understanding the HK Goldfish Market

Mind Genomics is an emerging science, philosophically descended from experimental psychology, with an admixture of sociology, anthropology, consumer research, and statistics. The notion is that one should be able to create a ‘wiki of the mind’, a searchable database of how people think and make decisions in their daily lives. These decisions are made with respect to the ordinary events of lives, the daily flotsam and jetsam which constitutes the everyday, the banal activities. As noted above, seminal works emerge from the observation of everyday behaviors, such as eating, using financial services, going to work, and so forth. Those are the inspirations. What Mind Genomics provides is a rapid, focused, experimentation-inspired approach to fill in the gaps, to under the features driving a person’s decision. One might think of Mind Genomics as the experimental science of the everyday, or the mechanism by which to create a ‘Wikipedia of Everyday Life.’

The topic is a cartography or limited exploration of a topic, here the perception of the Hong Kong Goldfish Market as responded by people in Hong Kong. The issue to understand their behavior towards the Goldfish market in terms of spending, and what they would like to see in the market. The ingoing vision was to treat the topic as a combination of anthropology (individual behavior), sociology (dealing with a well known establishment), psychology (how does the respondent think about the topics, what topics or messages engage), and economics (what does the person spend, and how does that interact with features of the market.)

It is important to note that Mind Genomics can be scaled to cover many different aspects of society, from the combined points of view of anthropology, sociology, psychology, and economics. The studies are small (base sizes of 20-30 respondents suffice for a basic understanding), are quick to set up (30 minutes), quick to execute in the field with a panel provider (approximately 60 minutes), and with data that are clear and easy to understand, returned to the researcher in both a PowerPoint® report ready to share, as well as a database in Excel® read for further analysis. As such, the approach of Mind Genomics presented here for the Hong Kong Goldfish Market is a template for many such studies, creating in its wake that ‘wiki of the mind’, or more correctly a ‘wiki of the mind and society’ so relevant to record and understand daily life in an era.

Mind Genomics follows a series of simple steps, with the steps ‘templated’ on the computer interface (www.bimileap.com). Mind Genomics forces the researcher to think of a topic in terms of totality, then break the topic down to four questions which tell a story, and then provide four answers to each question. The rationale is that the sequence forces the researcher to think about the problem in an analytical fashion, rather than in a holistic fashion. The result of the Mind Genomic will be the importance of each of the answers to driving a response. Rather than generating a single answer, Mind Genomics will generate the ‘underlying plans’ of the topic, allowing the person to understand the topic in depth, and to reconstruct the topic in new ways. The Mind Genomics interface has been simplified to follow an easy-to- use template, with the suggest type of answer shown as a suggestion, easily overwritten by the researcher who ‘gets the general idea of what is needed at that step’ from the suggestion. By the end of 2-3 ‘tries,’ viz., set-ups of different topics, the research is proficient, the researcher’s mind forever ‘rewired’ to think in a structure yet creative manner. The explanation for the 2-3 tries is that in every skill there is a learning period. It is difficult to create a computer program forcing a person to exercise ‘creative and critical thinking’, and have a person perform excellently the first time the person uses the program.

Step 1 – Define the Topic

Step 1 is easy, because it requires the researcher to pick a topic of interest, without doing any ‘disciplined thinking.’ The topic here is the above-mentioned HK Goldfish Market, a topic that can be addressed by young researchers and older researchers alike. The fact that the topic is easy approachable, and NOT fear-inducing, allows students to realize that science and research can be fun, and be ‘theirs.’ Students can work with Mind Genomics to explore virtually any topic of interest to them, make discoveries, and ‘own’ knowledge and creative thought, rather than simply hearing about the joy and knowing, thinking, and creating.

Step 2 – Create Four Questions which Tell a Story

It is at this juncture that the researcher is challenged to think in both a critical way and in a creative way. The questions must pertain to the topic (Hong Kong Goldfish Market) but must tell a story which is connected. Experience with this stage suggests that it is at Step 2 that most beginners become frustrated because they never have been required to think in this deconstructive, analytic fashion, breaking down a topic into components. They may have been exposed to topics with components but have never had to exert themselves to define a topic in terms of components. It is at Step 2 when people feel challenged, overwhelmed, and want to drop the topic because they are out of their ‘comfort zone.’ Those who continue, those who do two or three set ups of different studies, report that they ‘overcome’ this block, and feel that the demands of Step 2 force them to learn how to think in a different way, a more structured way, a way that makes them feel proud.

Step 3 – For Each Question, Instruct the Researcher to Provide Four Answers, Preferably Phrases

As Table 1 shows, the phrases for beginners tend to be short, and not descriptive. With practice, however, the researcher feels liberated, and grows more creative. Figure 1 shows the questions and the answers for one question. Table 1 shows the four questions and the four answers. It is important to emphasize that Mind Genomics studies are easy and quick to set-up, inexpensive to run. These simple studies, really scientific experiments, lend themselves to iterations. The life lesson is that nothing is permanent, and that experience can be shown to build ultimate success. In terms of Mind Genomics, one can repeat the study several times, several iterations, at each iteration keeping elements or ideas which just showed themselves to ‘work’, viz., perform well or in interesting ways, and in turn discarding and replacing elements which perform poorly, or which do not teach anything. The elements, phrases in Table 1, represent a first effort to explore the HK Goldfish Market using Mind Genomics. It is important to note that no Mind Genomics experiment is ever ‘too early’ or ‘too late’ in the process of developing an understanding of a topic. The iterative nature of Mind Genomics encourages exploration to the depth of understanding one wishes to achieve.

Table 1: The four questions and four answers to each question (elements).

Figure 1: Distribution of responses (left panel) and response times (right panel).

Step 4 – Combine the Answers into Vignettes according to an Experimental Design

Mind Genomics differs from the typical way one would study the HK Goldfish Market. The conventional practice is that the researcher would identify aspects about a topic, create questions pertaining to each aspect, and instruct the respondent to answer the battery of questions, one question at a time. The pattern of answers gives a sense of how the respondent feels about the topic. This approach, known as ‘isolate and study’ may work for most topics, but when it comes to study aspects of daily life it is impossible to prevent a respondent from changing the criteria of judgment. An example comes from two of the questions in Table 1 (Question A, Question B) deal with price. Two of the Questions in Table 2 (A,B) deal with spending, and two deal with features (C,D). It is hard to use the same criterion to judge these two types of questions.

Table 2: examples of three vignettes created by experimental design, and the binary coding of the vignette and the data to prepare for statistical analysis.

The experimental design mixes the different answers into vignettes, combinations, comprising both statements about spending and pricing (Questions 1,2), and statements about features (Questions 3,4). Table 2 shows an example of the vignettes and the underlying experimental design. A respondent shown this combination maintains a single focus, a single criterion, when judging the entire vignette. The analysis turns out to be much simpler, much more direct, as we see below.

The texts on experimental designs provide different recommended designs. The specific design used for Mind Genomics is a so-called main-effects design, permuted into 500 different designs having the same structure, but featuring different combinations of the 16 elements. The benefit of the permuted design is that is covers a great deal of the design space, the possible combinations, a strategy to discover underlying patterns [8].

The actual experimental design comprises four independent variables (the four questions), and four ‘options’ or ‘levels’ of each independent variable (viz., the four answers or elements). There are 16 elements in total. Each respondent evaluates a unique set of 24 vignette created according to a main-effects design, in which the 16 elements are each presented five times, in 24 vignettes, and absent 19 times from the 24 vignettes. The design ensures that a vignette comprises 2-4 elements, at most one element from each question. The design further ensures that the 16 answers or elements will be presented in a way that makes them statistically independent of each other. The property of statistical independence is important when one wants to deduce the contribution of each of the 16 elements to the overall rating, using the method of OLS (ordinary least-squares) regression. Ensuring that the 16 elements are independent of each other at the start makes the analysis quite straightforward, virtually automatic, with the results ‘figuratively’ jumping out at the researcher.

In many scientific studies the objective is to obtain data which has as little extraneous variation as possible, so-called error variability. To the degree that the researcher can reduce the error variability, the patterns underneath will emerge more clearly. The standard way to do this error reduction is to either suppress the noise by careful testing (impossible to do with people), or to average out the variability by testing the same set of vignettes with hundreds of people, so that the random variation cancels out. The Mind Genomics worldview goes contrary to the traditional approaches. Mind Genomics is metaphorically an ‘MRI of the mind.’ The patterns of responses generated from the different respondents (here 30 different responses) give information from different perspectives to the same topic. The information can be combined by computer to generate a much more robust, comprehensive, multi-aspect view of the problem. The patterns emerging from the Mind Genomics effort literally ‘jump out’ as we will see below.

Step 5 – Create an Orientation Paragraph

The paragraph introduces the topic and provides the respondent with the scale. The best practices for orientation paragraphs depend upon the specific use. For most situations, the less one says the better. The rationale for ‘saying little’ is that the key information should come from the elements. The paragraph below presents the orientation:

Everyone these days is talking about the HK Goldfish Market. We would like you to have fun with us. We are doing a study on what people REALLY think about the HK Goldfish Market. Please read the whole screen below, and rate the combination… There is no right or wrong. JUST YOUR OPINION, no one else. Don’t think too long…just look, read, rate!

How do YOU feel about this set of statements TOGETHER>?

1=1 = Doesn’t agree at all with MY opinion of the HK Goldfish Market …

5=5 = Perfectly agrees with MY opinion of the HK Goldfish Market

Step 6 – Run the Study

The study can be run among friends, or through a panel service. This study here was run with a panel service, with respondents from Hong Kong, familiar with the HK Goldfish Market. The respondents who agree to participate open the link, read the introduction, complete a short introductory survey (classification) about age gender, and frequency of visiting the HK Goldfish Market. The respondent then rates the 24 different vignettes created for the respondent, doing so in about three to four minutes. The computer records the rating and measures the time between the appearance of the vignette on the computer screen and the respondent’s rating. The actual interview, the experiment, required about three minutes of the respondent’s time.

Results

Mind Genomics data provide a rich bed of test stimuli. Each respondent evaluated 24 different vignettes. The 30 respondents generated 720 different combinations of elements, answers to the question, with, each combination designed to be different from all the others. An initial analysis revealed that three respondents assigned virtually the same rating to all 24 vignettes that were presented to them. The data from these three respondents were eliminated, leaving 27 respondents, sufficient for a quite rich analysis, as will be see below.

Step 7: Transform the Rating Data to Two Binary Scales, the Agreement Scale, and the Disagreement Scale, Respectively

Over the past decades, researchers have come to rely on two types of scales. The first is a graded scale, called a Likert scale, or category scale. The notion is that the scale comprises a set of discrete points. The respondent is required to rate the test stimulus assigning a rating point. The presumption is that the scale points are equally spaced in terms of psychological distances, and thus averaging and statistics are acceptable. The 5-point scale is an example. Consumer researchers recognize that these scale points are neither equally spaced, nor in fact can be readily interpreted by anyone. Users of the scale always ask, for example, “what does Rating X (e.g., 5) mean on the 5-point scale?” In the absence of extensive studies of the scale, it is easier to divide the category or Likert scale into ranges calling one part 0 and the other part 100. The division may not be truly equal, but managers understand the notion of two scale points. For this study, the analysis divided the scale in two ways, both generating a two-point scale easy to understand:

Describes Me scale. Ratings 1-3 transformed to 0 (does not describe me), ratings 4-5 transformed to 100 (does describe me). A small random number is added to every transformed scale value in order to ensure that the data will never generate a situation where all of the transformed ratings are either 0 or 100. There must be some variation in the scale data for the subsequent regression analyses to work.

Does NOT Describe Me Scale. Ratings 1-2 transformed to 100 (do not agree; does not describe me), ratings 3—5 transformed to 100 (agree; does NOT not describe me)

Step 8: External Analyses – Looking at the Patterns of the Data, but Not at individual Elements

As a note about the practice of science, it is always a good practice to plot one’s data, whether the study comprises a few hundred data points, such as this study on the HK Goldfish Market, or the study comprises hundreds of thousands, or even millions of data points. The first step should be to familiarize oneself with the data, to explore the data, to become familiar with it, to find general patterns. Only then, after familiarization, does it make good research sense to start testing for differences, to make substantive conclusions about patterns and so forth.

The first external analysis plots out the distribution of responses, as shown in Figure 1. The left panel shows the distribution of ratings on the 5-point Likert scale. The right panel shows the distribution of responses times. Figure 1 suggests that there are more agreements (describes me, ratings 4-5) and far fewer disagreements (does not describe me, ratings 1-2). Figure 1 further suggests that the responses evaluate the vignettes quite quickly, most taking about 2 seconds or less to rate a vignette. The large number of ratings at 7 seconds correspond to those vignettes which took longer than 7 seconds. The assumption was that these vignettes represent situations during which the respondent was not paying attention to the task.

The summary data shown in Figure 1 tells us just a little about the different aspects of the HK Goldfish Market. We understand that the phrases generate agreement, and that the information is easy and quickly processed. As of yet, we do not know the ‘internal’ aspects of the data, specifically the ‘mind’ of the respondent who is assigning the rating. We can see from ‘outside’, but we do not necessary get a sense of what is going on ‘inside.’ To get a sense of what is going on in the respondent’s mind requires us to understand how the specific elements in the vignette ‘drive’ the ratings. The learning emerging from linking elements to responses which give us a sense of how the respondent is thinking (rating), and what is engaging the respondent’s attention (response time). The former, ratings, is under the control of the respondent’s conscious mind. The latter, response time, is not under control of the respondent’s conscious mind, but rather an uncontrolled behavior reflecting attention to, and engagement with, the task.

Continuing our ‘external analysis’, we can learn more from the data, specifically the average responses. We create four new averages, one for each respondent, based upon the 24 vignettes rated by the respondent. Although each respondent evaluated different 24 unique vignettes, we can get a sense of the general response to the topic. The four new averages are, respectively, the ratings (1-5), the response times (after truncation to move all responses times to a maximum of 7 seconds), average Top2 (Describes ME, viz., agree), and average Bot2 (Does not describe ME, viz., disagree). We will find deeper insights when we plot these averages by respondent.

One of the first questions emerging from the introduction of Top2 and Bot2 is the degree to which these averages parallel the averages that would have been obtained by working with the original 5-point Likert scale. That is, when we average responses for the binary scales, do we see the same pattern as we would see when we average the ratings themselves? Or does the binary transformation lose so much granular information that the transformation creates new problems of ‘meaning’, despite the easier interpretation is easier! Figure 2 shows us the plot from the 27 respondents. Each circle corresponds to one of the 27 respondents. We would make the same decision based upon the patterns of all three plots. The only difference is that the binary transforms of ME (Top2) and Not ME (Bot2) are less clear because we exclude all ratings of ‘3’ from both. Yet can be fair confident that our qualitative conclusions will be the same when we use the 5-Ponit Likert Scale or the Binary Transformed Scale. The binary scale will be easier to interpret, however.

Figure 2: Average ratings for 24 vignettes for each of 27 respondents. Each circle is a respondent. The patterns are similar for Likert Rating Scale vs Binary Scale, noisier for the two binary scales.

The second external analysis searches for relations between response time and either the actual ratings on the original 1-5 scale, or the binary transformed scale. Figure 3 shows a noisy but discernible relation between the average response time from the respondents and the average transformed rating assigned by the respondent. It should be kept in mind that these results come from averaging response times and ratings from a unique set of 24 vignettes for each respondent.

Figure 3: Relation between average binary scale (Top2 – Agree, ME; Bot2 – Disagree, NOT ME) and response time. Each point corresponds to the average of one respondent’s rating of 24 vignettes.

On average, respondents who showed the highest average agreement showed the shortest response times.

On average respondents who showed the highest average disagreement showed the longest response times.

Step 9: Internal Analysis Relating Each Element to the Binary Transformed Rating’s

Had the data been simply combinations of elements without cognitive ‘meaning’ the analysis would have stopped at the external analysis, simply because there is nothing to be learned from the properties of a specific stimulus. The stimulus would just be part of the set of stimuli picked for the analysis to discern a general pattern.

Mind Genomics moves beyond the external, simply because the elements themselves have cognitive richness, meaning in what they communicate, meaning in their sentence structure, meaning in the words, and so forth. The richness need not be explicated at the start of the Mind Genomics study. It suffices only that the elements be chosen for a reason germane to the topic. In this study, there are two questions pertaining to pattern of spending and amount of spending, and two questions about attitude, specifically what one wants in the market, and how one feels shopping in the market. These questions are never asked directly, but rather represent by cognitively rich statements to which the respondent reacts by assigning a rating, doing the assignment rapidly in what Nobel Laureate Daniel Kahneman called System 1 behavior [9].

The experimental design enables the research to create equations relating the presence/absence of the 16 elements to the four dependent variables, whether these be the actual ratings on the 5-point Likert scale, the binary transformed ratings for Agree (Describes ME, Top2), the binary transformed ratings for Disagree (Does NOT Describe Me, Bot2), or response time.

The first analysis using OLS (ordinary least-squares) regression creates equations for each individual, with the binary transformed rating of Agree (Top2) as the dependent variable. The rationale for the individual-level modeling is that the 27 different models will generate the data needed to divide the respondents into two complementary groups, mind-sets, based upon what specific elements they feel describes them.

The general form of the equation is: Top2 = k₀ + k₁(A1) + k₂(A2) + k₃(A3)…k₁₆(D4)

The foregoing equation can be estimated at the level of each respondent (27 different equations), or at the level of groups such as Total Panel (one equation), gender (two equations), age (two equations), and finally mind-sets emerging from clustering the respondents by the pattern of their coefficients (two equations).

The OLS (ordinary least-squares) regression model emerges with the additive constant, and 16 coefficients. The ‘rules’ for interpreting the parameters are as follows when the dependent variable is either the Top2, ( Agree, ME) or the Bot2 (Disagree, NOT ME)

The additive constant is the estimate percent of times that the rating will be ‘describes me’ (viz., 4 and 5), when there are no elements. Clearly the experimental design ensures that each of the 24 vignettes comprises 2-4 elements, so the additive constant is a purely estimated parameter. One can consider the additive to be a baseline likelihood to agree (Top2) or a baseline likelihood to disagree (Bot2) even before information is presented. In some ways the additive constant can be considered an indication of the way people think about a topic, in general, without specifics.

The coefficient shows the additional percent of responses are added to the dependent variable when the element is inserted into the vignette. Thus, a coefficient element of +6 means that an additional 6% of the responses will be added to the response when the element is inserted. A coefficient of -5 means that 5% fewer of the responses will move away from the dependent variable.

The additive constant and the coefficients sum together. Thus, for the situation of the dependent variable Bot2 (Disagree, Not Me), when the additive constant is 37 and the coefficient is -6, the expected percent of responses for Bot2 for that 1-element vignette is 37 – 6 or 31. In this case the element takes away from Bot2. In contrast, when the coefficient is +9, then the expected percent of responses for Bot2 for that 1-element vignette is 37+9 or 46, meaning there will be 46% of the Bot2 responses for that 1-element vignette.

The vignettes comprised 2-4 elements, meaning that one can combine 2-4 elements to create a new vignette, and estimate the likely rating, making sure that the elements come from different questions. The estimated value is simply the arithmetic sum of the additive constant and the elements.

The strategy for presenting the results will be to show only the elements which are positive, viz., greater than 0, for either Top2 or Bot2, for any subgroup. The rationale is that we are interested in learning about the pattern of elements which drive the response. Showing only positive numbers lets the patterns emerge clearly. Highlighting strong performing elements (coefficients of +8 or higher for the binary transformation) further allows the patterns to emerge in greater relief.

A further strategy when presenting the data will be to sort the data in descending order by the two mind-sets, highlighting the strong performing elements by shading the cells in which coefficients are +8 or more. A coefficient of +8 corresponds to an element which is around two standard errors beyond the coefficient of 0, and suggests a strong impact of the element on the binary rating

Table 3 presents the results for the Top2 binary variable, defined as: Agree, or ME. we begin with the additive constant

1. Total Panel – About Half of the responses will agree (additive constant = 46).
2. Males show a strong propensity to agree (additive constant = 70), females show a moderate propensity (additive constant = 43).
3. Younger respondents (17-29) and older respondents (30+) show similar propensities to agree (additive constants of 45 for younger and 42 for older).
4. Respondents who never frequented the HK Goldfish Market show a strong propensity to agree (additive constant = 76), frequent shoppers show a moderate propensity to agree (additive constant = 40).
5. Mind-Set 1 (focus on low price, easy maintenance) show a moderate propensity to agree (additive constant = 40), Mind-Set (True Afficionados) show a strong propensity to agree (additive constant = 63).

In terms of the performance of the elements, many elements are positive, meaning that the respondent feels that they describe the respondent’s feelings. There are, however, a great number of elements with zero or negative coefficients.

1. The most consistently strong element is C4: My wish for the HK Goldfish Market: ready-made aquariums and/or aquarium maintenance.
2. A variety of other elements emerge as strong for the different geo-demographic and behavioral groups, but not consistent pattern that lends itself to easy identification.
3. When we divide the respondents by the pattern of what describes them, creating two mind-sets, we find two clear groups. Mind-Set 1 focuses on easy maintenance, appearing to spend less money or no money. They have a lower additive constant, 40, meaning that they are not likely to agree, to feel that the phrases in the vignette apply to them. In contrast, Mind-Set 2 spends a lot of money, and wants high quality, interesting fish, and equipment. Furthermore, Mind-Set 2 has an additive constant of 63, meaning that they are ready to agree. We present the mind-sets as the final two columns of data, sorting the table by mind-set to reveal the patterns, which emerge clearly after the sorting. There is no such clarity of pattern for any other grouping of the respondents, viz. WHO they are or what they DO.

The approach to creating mind-sets using Mind Genomics has been previously described in previous papers [10]. The ingoing assumption is that respondents with similar patterns of coefficients for the 16 elements on Top2 (agree) belong to the same mind-set. Respondents with dissimilar patterns of coefficients belong in different mind-sets. The data from these studies suggest two clearly different mind-sets. Using mind-sets to organize data is not limited to Mind Genomics but has been shown to be a stimulus to creative thought [11].

4. By focusing only on the positive elements, and highlighting the strong performers, the nature of the mind of the respondent becomes clearly with respect to the HK Goldfish Market, in a way hard to capture by conventional anthropological observation, sociological analysis, or market research. One begins to sense the structure of the people for this granular part of the Hong Kong ‘every day.’

Table 4 present the reverse scale focusing on disagree. The groups are the same, total, gender, age, frequency of visit, and the two emergent mind-sets from clustering the respondents on Top2. Again, only the positive coefficients are shown except for the additive constants.

In contrast to the clarity of results from Table 3, showing agreement (ME), the pattern of additive constants and coefficients in Table 4 is confusing. The difficult of discovering a clear pattern may emerge because responses focus on what they agree with. What they fail to agree upon may either be irrelevant, or important. In either case, respondent appears to focus on using only one side of the scale. The respondent may not be ‘weighing’ the entire set of elements to come up with a single composite judgment, but rather may simply focus on finding the key element, ignoring everything else. In such a case the pattern would be one-sided, clearer when the respondent focuses primarily on agreement. To test this hypothesis may simply require a parallel study with the respondent instructed to focus either on agreement in one test cell, or disagreement in another test cell.

Table 3: Coefficients for the ‘Top2’ model, relating the presence/absence of the elements to the Top2 value. In the interest of clarity, only the positive coefficients are shown.

Table 4: Coefficients for the ‘Bot2’ model, relating the presence/absence of the elements to the Bot2 value. In the interest of clarity, only the positive coefficients are shown.

The final analysis of groups looks at the response time, defined operationally as the number of seconds between the appearance of the vignette on the screen and the assignment of a rating by the respondent. The modeling is the same as that for Top2 and Bot2, with one exception. The exception is that the additive constant is omitted from the model for the response time vs elements. The rationale is that in the absence of elements there is no response.

Table 5 shows the estimated response times assignable to each element. The first data column, for Total Panel, shows all 16 coefficients. The range of coefficients goes from a low of 0.3 seconds (A3: Spend money at HK Goldfish Market: Spend less than 5 years ago) to a high of 0.9 seconds, such as D2 (The HK Goldfish Market: Big variety of shops), C1 (My wish for the HK Goldfish Market: Larger variety of animals). B3 (Annual Spend at HK Goldfish Market: 2,500 HKD – 5,000 HKD at the shops), and so forth. There is no clear pattern for the Total Panel, other than perhaps that the elements describing the offerings tend to engage the respondent a little long.

Table 5: Coefficients for the ‘Response Time’ model, relating the presence/absence of the elements to the measured response time. With the exception of the Total Panel, onlypositive coefficients are shown, in the interest of clarity and simplicity.

As done for Tables 2 and 3, the rest of Table 4 shows only those elements which are deemed to be ‘engaging,’ viz., show response times of 1.0 seconds or longer. The cut-off of 1.0 seconds is strictly an operational, giving a sense of the types of elements which engage.

1. Males seem to be more engaged by elements dealing with price. Females seem to be more engaged by elements dealing with features.
2. Young respondents are far more likely to be engaged by elements, older respondents are not.
3. Those who say they never frequent the HK Goldfish Market are not engaged by any elements. Those visit frequently are engaged by only one element, D4 (The HK Goldfish Market: Find exclusive/specialty fish)is
4. Mind-Set 1, (focus on low price & easy maintenance), is engaged by two elements, D2 and C2, dealing with variety. Mind-Set 2 (true afficionados) is engaged by both price and features, showing deeper engagement as reflected by response time. The deepest engagement is 1.9 seconds, B1 (Annual Spend at HK Goldfish Market: Less than 1,000 HKD at the shops). This element may surprise and intrigue, because it is so contrary to the behavior and interests of the afficionado.

Generalizing the Results – Finding these Mind-sets in the Population for Science and Business. With a small group of 27 respondents, the distribution of respondents into mind-sets will be error prone. The small base size of respondents finds it best use as a tool to uncover hitherto-unexpected mind-sets. The small number of respondents used for discovery does not suffice to estimate the proportion of these mind-sets across the population, especially in different countries. Thus, with small base sizes, the distribution of mind-sets across relevant subgroups is at best a rough estimate. Table 6 shows this distribution. Mind-Set 1 comprises most of the respondents. Furthermore, that the most outstanding aspect of the distribution is that of the seven respondents in Mind-Set 2 (True Afficionados), six are older respondents, far more than would have been expected.

Table 6: Distribution of respondents from the total panel and two mind-sets across gender, age, and frequency of visiting the HK Goldfish Market.

Given the distribution of mind-sets shown by Table 6, how can the researcher or digital marketer assign a new person to one of the two mind-sets for this granular topic of a pet market? Recently, author Moskowitz in collaboration with Hungarian researcher Attila Gere developed an approached called the PVI, the Personal Viewpoint Identifier. The PVI is based upon the data from Table 4, the coefficients from the transformed data (Top2). The PVI uses simulation and decision trees to create a system which assigns a new person to one of the two (or three) emergent mind-sets from a Mind Genomics study [10-14].

Figure 4 shows the introduction to the PVI. These data can be customized, so that the data are entirely anonymized, or the data can include such information at telephone or email, for follows-ups. Figure 5 shows the set of informational questions about the respondent, and then the six questions comprising the PVI itself. The four first questions, ‘information’, are equivalent to the types of questions researchers ask about attitudes and usage for topics of interest. Figure 6 shows the format of the template used to transfer data from the Mind Genomics study to the PVI. Note that after the data from the respondent are stored in a database, and the respondent is sent an email of results, the respondent may be guided to a video stored in YouTube, or to a landing page. Thus, the PVI serves both as an information-gathering system, and as a tool for e-commerce.

Figure 4: Introduction to the PVI for a ‘pet market’. https://www.pvi360.com/TypingToolPage.aspx?projectid=1269&userid=2

Figure 5: Classification questions and the PVI itself. The first four questions are classification (attitude and usage). The second six questions constitute the PVI.

Figure 6: The Excel® based template, allowing the researcher to select the elements, the classification questions, the binary PVI questions, and the post-PVI experience with a video or landing page. The PVI is computed after the template is completed.

Discussion and Conclusions

For most of the history of psychology, experiments have presented the respondent with artificial situations to uncover rules of behavior. The experiments are crafted from theory, to prove or disprove a hypothesis. The study presented here on the HK Goldfish Market reveals the potential of increasing our understanding of the granular, every-day, unremarkable experience, revealing patterns of decision-making, and emergent understanding at several levels.

Taking its cue from consumer research, anthropology, sociology, as well as statistics, the newly emerging science of Mind Genomics works in a different way, one that might be called a cartographic analysis. The objective is to not to develop general hypotheses about behavior, and either show that they describe the data, or falsify the hypothesis. Rather, Mind Genomics uses the methods of experimental science to understand how people react.

The experiments in Mind Genomics are easy to perform, and the subject matter is boundless. As a consequence one need not create a hypothesis and test that hypothesis by manipulation to prove or disprove the hypothesis, or even conjecture It is adequate to act like an explorer, a cartographer, mapping the land, finding interesting areas, unusual formations, and the ‘stuff’ worth talking about. Mind Genomics as a science should appeal to those who are not interested in the traditional tasks of ‘filling holes in the literature,’ nor responding to calls to answer key issues. Instead, and in the spirit of the early Baconian philosophers of natural science, it is sufficient to map the topic, to study the different aspects, without being forced to justify one’s scientific curiosity by first putting up a hypothesis to be proved or disproved, the hallmark of today’s hypothetico-deductive method (Grimes, 1990).

References

1. 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. [crossref]
2. Warde A (2016) The practice of eating. John Wiley & Sons.
3. Carruthers BG, Kim JC (2011) The sociology of finance. Annual Review of Sociology 37: 239-259.
4. Whyte W (1957) The Organization Man. Garden City. NY Doubleday.
5. Moschis GP, Mathur A (2007) Baby boomers and their parents: Surprising findings about their lifestyles, mindsets, and well-being. Paramount Market Publishing.
6. Han Z, Lai LW, Fan J (2002) The ornamental fish retail market in Hong Kong: its evolution and evaluation. Aquaculture Economics & Management 6: 231-247.
7. Lam KKH, Lai LWC (2002) Goldfish (Chin‐yu or Kin‐yu) culture practice in Hong Kong. Aquaculture Economics & Management 6: 275-293.
8. Gofman A, Moskowitz H (2010) Isomorphic permuted experimental designs and their application in conjoint analysis. Journal of Sensory Studies 25: 127-145.
9. Kahneman D (2011) Thinking, fast and slow. Macmillan.
10. Porretta S, Gere A, Radványi D, Moskowitz H (2019) Mind Genomics (Conjoint Analysis): The new concept research in the analysis of consumer behaviour and choice. Trends in Food Science & Technology 84: 29-33.
11. Boaler J (2015) Mathematical mindsets: Unleashing students’ potential through creative math, inspiring messages and innovative teaching. John Wiley & Sons.
12. Grimes TR (1990) Truth, content, and the hypothetico-deductive method. Philosophy of Science 57: 514-522.
13. Moskowitz HR, Gofman A, Beckley J, Ashman H (2006) Founding a new science: Mind Genomics. Journal of Sensory Studies 21: 266-307.
14. Moskowitz Howard R, Sebastiano Porretta, Matthias Silcher (2008) Concept research in food product design and development. John Wiley & Sons.