DOI: 10.31038/ASMHS.2019315

Abstract

We introduce a system to rapidly explore a topic, focusing both on the direct conscious judgment of information (cognition), and on the time it takes the mind to process the same information (neuroprocessing.) The system begins with the experimental design of easily constructed mixtures of messages. With human respondents, the system measures the cognitive response to these mixtures (ratings), and at the same time, the processing rate of these same mixtures (response-time to assign a rating.) The system is affordable and scalable, working with as few as 10 respondents to as many as several thousand. The outcome data reveal what messages are important, and the response-time to process these same messages. The analysis is virtually automatic, providing a simple, readily used new tool to study decision making. All the tools are standard, easily used by professionals and novices alike, with the results immediately presented in the format of data tables and a PowerPoint® report ready for distribution.

Introduction – the conflict between ‘objective’ and ‘subjective’ in experimental psychology

During the past seventy years, since the auspicious days of the 1950’s shortly after World War II, the field of experimental psychology has been deeply involved in the measurement of subjective experience. During the previous generations it was thought that people could not be accurate instruments to assess the magnitude of external stimuli, although they could react in ways which had desired effects on their life and on their environment. Many professionals believed that people could not act as valid measuring instruments, despite the fact that people could engineer their environment to exacting tolerances. Rather than focusing on the cognitive reactions to stimuli, many experimental psychologists felt that the more appropriate measures were non-cognitive, but rather autonomic nervous system reactions. These were assumed to be more ‘truthful.’ At the very simplest level were measures such as GSR (galvanic skin response), pupil dilation, and heart rate. The feeling was that these measures were more ‘objective indicators’ of one’s reactions to external stimuli, perhaps even better than attitudinal measures. Over time, however, researchers began to recognize that they needed people to respond to the world, using scales, in order to measure the private subjective experience that could otherwise not be measured. During the period, beginning in the 1920’s but accelerating dramatically after World War II, researchers created many different standardized scales in order to measure innate feelings and proclivities. These scales range from political conservatism to fear of new foods, just to give a sense of the range.

The nature of the ‘test stimulus’ – cognitively poor vs cognitively rich

One of the ongoing issues of these experiments is the artificial nature of the stimulus, and the limits of what can be learned. In most studied focusing on what can be learned by ‘objective measures.’ The respondents are presented with test stimuli, either of a meaningless nature in terms of cognition (e.g., lights), or of a modestly meaningful nature in terms of cognitions (e.g., pictures without a context.) It is vital to do so because the typical approach of the scientific method in virtually all fields requires that the researcher isolate the variable to as pure as possible and compare the response of the organism when the variable is present versus when the variable is absent. In this manner, the difference is ascribed to the variable being studied. In such manner one begins to understand the dynamics of the so-called objective measure.

In sum, then, the reactions of the subjects in task involving those cognitively poor stimuli are analyzed to uncover patterns, which help understand how people process information. It must be emphasized here that the knowledge gleaned is from the patterns, the regularities in the response, and not from the response to the individual test stimuli, which, in the real-world, are without any real meaning. It is the opinion of the authors that a new science of the Mind is needed, one which combines the rigor of scientific interventions with test stimuli having meaning. As we will see in the study reported here, quite a bit can be learned about the way people process meaningful information, using direct judgments to understand the process of conscious judging, and using measures of response-time to understand some of the underlying neurophysiological processes.

Mind Genomics – Learning from the reactions to cognitively rich test stimuli

Author HRM was educated as a sensory psychophysicist in the middle 1960’s, with experiments involving the sense of taste. The test stimuli were aqueous mixtures of water with a taste stimulus (e.g., sugar solutions of different concentration), or aqueous mixtures of water with two taste stimuli (e.g. sugar and salt, both dissolved in the same solution.) Sensory psychophysics showed the scientific community that one could learn a great about subjective sensory perceptions. In some extensions of the sensory work, Eugene Galanter pioneered the work in scaling the utility of money [1], and Stevens himself, father of modern psychophysics inspired the use of psychophysical scaling to measure the seriousness of crimes [2].

More relevant insights into the way we think emerged when the researchers began studying responses to combinations of ideas. The combinations of ideas, i.e., mixtures of message, constitute ideal stimuli, simple and inexpensive to create and to test with people. The mixture, in the words of psychologist William James, present a ‘blooming, buzzing confusion.’ The respondent must extract the relevant information quickly from the mixture, and assign a rating to that mixture. The underlying experimental design allows the researcher to estimate the contribution of each element in the mixture. These early studies suggested responses to combinations of messages, created by experimental design, could teach us a great deal about decision-making [3].

Once researchers recognized that they could learn about the respondent’s mind from deconstructing responses to compound mixtures, it was almost a natural step to create a science of decision-making. The science of Mind Genomics was born. Mind Genomics uses the analysis of responses to mixtures of ideas in order to understand the mind of consumers to all sorts of ideas, ranging from the law to religion, to products, and so forth [4, 5] When we combine cognitive measures such as conscious judgments about these mixtures of ideas with measures that reflect neurophysiological processing of information, an easy one being response-time (RT), we may well be able to glean new insights about the way we think. When the stimuli are cognitively rich, e.g., dealing with a meaningful and possibly interesting topic, and when the measures are both conscious ratings and so-called objective physical measures, there is the greater opportunity for patterns to emerge, patterns which would never appear when the stimuli are simplistic, boring, and relatively meaningless. The world of neurophysiological studies for consumer research is beginning to grow dramatically. This paper is part of that trend [6, 7].

Comparing judgments with the time needed to make those judgments

This paper compares the content of judgments with the time needed to make the judgments. The approach uses Mind Genomics, measuring both the response to the test combinations (vignettes), and the time needed to assign the response. The experiment goes deeper, in two ways. First, the analysis separately deconstructs the response (rating), and then the response-time, into the part-worth contribution of the elements, to determine how each element or messages ‘drives’ the responses. Second, the analysis creates two Mind-Sets for the respondents based on how the elements drive the ratings (clustering on cognitive judgments), and then a separate set of two Mind-Sets for the same respondents, this time based on how the elements drive the response-time (clustering on neurophysiological data.)

We present our approach, using a small, web-based experiment with 25 respondents, set up, executed, automatically analyzed, and automatically reported in a matter of 45 minutes. We deliberately keep the study small to see how much information and insight can be extracted from a simple, cost-effective effort. Our long-term objective is to lay the foundation to easy-to-do studies, combining cognitively meaningful stimuli, judged with relevant scales by ordinary people, with the co-variate of response-time measured at the same time. We attempt to demonstrate that Mind Genomics can make researchers out of almost anyone (scalability of use), can do so inexpensively, and can investigate almost any topic where the ‘mind is king.’

The steps for the process appear in Table 1, along with the rationale for each step

Table 1. The research process combining Mind Genomics and measures of response-time.

The study

The Russian- Ukrainian conflict of 2018, began some years back [8, 9]. The topic, a geo-political conflict, is meaningful in terms of everyday life, but not widely understood. Even when the respondents are not familiar with the topic, the test stimulus (vignette) presents sufficient information for the respondent to make judgments based upon what is presented, and based upon their own understanding of current events, whether deep or only superficial. Thus, the messages can talk about peace and war, as realistic, but rather ‘remote’ topics. Our comparison of cognitive measures (ratings) and neurophysiological measures, will make sense in terms of dealing with ‘real world’ issues.

Table 2 shows the set of four questions, and the four answers to each question. (Table 2) also shows the number of ‘key words’ (information) in each element.

Table 2. The four questions and the four answers to each question. The topic is the Russian Ukrainian conflict in 2018.

One of the key features of Mind Genomics is that it enables the researcher to use relatively few respondents for exploratory studies, such as the one reported here, or many respondents to define a topic area with a large, representative sample of respondents. The power of Mind Genomics, and its ability to work with few respondents, comes from the use of ‘permutable’ experimental designs, with each respondent presented with a full experimental design, different in combinations from the experimental design of the same material presented to another respondent [10] Thus, even with as few as 25 respondents, one can cover a wide space of 600 alternative combinations of messages, far more than most conjoint studies ever attempt to explore [11].

The respondent’s experience

The respondent reads each of the 24 vignettes, rating each vignette as a totality. The Mind Genomics APP (BimiLeap) records the rating and the response-time. (Figure 1) shows an example of the layout of the study on smartphone. The respondent can also participate with a personal computer or a tablet. There are biases in surveys. One of these biases is the desire of the respondent to please the researcher or the interviewer, by giving politically appropriate, non-confrontational answers to questions. This tendency to please the interviewer is promoted both by a personal interview, and by having questions on the interview which allow a person to slant her or his answers in the appropriate way. In contrast to the foregoing, Mind Genomics experiments are virtually impossible to ‘game.’ Mind Genomics experiments are done in the privacy of one’s home, on a computer, away from other people, so there is no interviewer bias. More importantly, however, Mind Genomics studies are impervious to the desire to be ‘politically correct.’ Test stimuli continually change, with ever-changing combinations appearing one after another.

Figure 1. Example of the respondent experience. The figure shows the presentation of a vignette on a respondent’s smartphone. The Mind Genomics study can be done using any device which can show websites, such as smartphones, personal computers, and tablets, respectively.

Results

A very simple first analysis computes the average rating, and the average response-time, respectively, for each of the 24 positions. Every respondent evaluated 24 unique vignettes. It is not the vignette itself which interests us, but rather whether there is a position effect. Figures 2A-2C show that there is no clear position effect for the average 9-point rating by position (Figure 2, left panel), nor for the average binary-transformed value by position (Figure 2, middle panel.) There is a clear position effect for the response-time, RT. The first position shows a higher average response-time, perhaps because the respondent is discovering what to do (Figure 2, right panel.) The strong position effect means that it will be more judicious to consider, where appropriate, the data without taking into account the first test vignette, i.e., the vignette in position #1.

Figure 2. The covariation with response order of ratings of the 9-point scale (left panel), the binary transformed scale (middle panel), and the response-time (right panel.)

The deconstruction of the responses is done by OLS, ordinary least-squares. The independent variables are the presence/absence of the 16 elements or answers to the four questions. They take on the value ‘1’ when present in a vignette, or ‘0’ when absent. The dependent variables are either the binary values (0/100) after transformation of the original 9-point ratings, or the response-time in seconds, from the time the vignette appeared to the time that the respondent assigned a rating.

Clustering respondents on the basis of ratings of the elements

Clustering is a well-accepted method in statistics to divide objects by their patterns. The software is readily available [12]. The r clustering method is a matter of choice. The clustering used here computes a measure of ‘distance’ between each pair of respondents based upon the Pearson correlation between their corresponding 16 coefficients, one per element. The distance is expressed as (1-Pearson R.) When two respondents show a perfect linear relation, the Pearson R is +1 and the distance is 0. When two respondents show a perfect inverse relation, the Pearson R is -1 and the distance is 2.

Table 3 shows the results for the deconstruction of the binary values, for the total panel and for three mind-set segments emerging from clustering the respondents based on the set of 16 coefficients generated from the individual models. The dependent variable was always the respondent’s rating, either 0/100.

Table 3. Parameters of models relating the presence/absence of the 16 elements in vignettes to both the binary-transformed ratings (also called Top3), and to the response-time (RT). The table shows the results from the total panel and from two Mind-Sets (segments) emerging from clustering. The clustering was done based upon the coefficients for the ratings (binary transformed, Top3) of the 16 elements, from the 25 respondents.

After clustering to reveal the two pairs of Mind-Sets, each respondent was assigned to the appropriate Mind-Set for the binary rating, and the appropriate Mind-Set for Response Time. For all models, the data from the first vignette (Response Order 1) was discarded. Then, the first vignette tested by each respondent was eliminated from the data set, and OLS regression was run on all the data from all the respondents in the particular Mind-Set. This is the so-called Grand Model. The analysis thus generated four Grand Models.

Table 3 shows the parameters of the Grand Models created from the group data (Total Panel, Respondents in MS1, and Respondents in MS2). The first three columns of data show the coefficients from the binary models (called Top3). The second three columns of data show the coefficients from the response-time (RT) models for the same respondents, segmented using their ratings of the vignettes.

It is clear that there are two Mind-Sets, based on clustering respondents according to their ratings of perceived likelihood of war. Mind-Set 1 feels that war will break out if there is direct military action. Mind-Set 2 feels that war will break out if there is an arms build-up.

Associated with each of these elements is also a response-time measure. Those response times of two seconds or longer are shown in bold and shaded. These are elements which ‘stop’ the respondent, engaging the respondent. We do not know whether the respondent could verbalize that these particular elements are engaging, but the regression analysis deconstructs the response time into the contribution of these elements (Table 3).

The response-time data become more interesting when the response-time coefficient is plotted against the binary-transformed or Top3 coefficient, either for the total panel, or for the Mind-Sets. (Figure 3) shows a clear pattern for total panel, as well as for the two Mind-Sets. As the perception of ‘likelihood of war’ increases (abscissa) the response-time of the element diminishes. For this cognitively relevant task, evaluation of the likelihood of war, we see a definite pattern relating a neurophysiological-based measure, response-time, to a judgment criterion, likelihood of war. The more likely the sense of ‘war’ breaking out, the faster the response time, when the plot is at the level of the 16 individual elements. In other experiments by author HRM, dealing not with critical events but with ordinary products, like yogurt, this straightforward pattern does not emerge (see appendix to this paper).

Figure 3. The relation between the coefficient for response-time (RT) for the element (ordinate) and the coefficient of the same element from the model for ‘likelihood of war’ (abscissa.) The Mind-Set segments, MS1 and MS2 were obtained by segmenting the 25 respondents based upon the coefficient for Top3, the binary-transformed response of the rating scale.

Does segmentation on the basis of response-time produce meaningful patterns?

Just as one may cluster the respondents based on their judgments of what they perceived to drive the likelihood of war, so one may cluster the same respondents on the pattern of what drives response-times. The mechanics of clustering remain the same. The only differences are the nature of the models, and the interpretation of the meaning of the segmentation.

The clustering process begins by building a model for each respondent, using all 24 vignettes, despite the bias encountered with the first vignette. There is no other option. Each respondent generates a pattern of 16 coefficients, which can be divided into two (or more) clusters. Table 4 shows the parameters of the models for the 16 elements, for models using as the dependent measure response-time (first three data columns), and the binary transformed rating (Top 3, second three data columns.)

Table 4. Parameters of models relating the presence/absence of the 16 elements in vignettes to both the binary-transformed ratings (also called Top3), and to the response-time (RT). The table shows the results from the total panel and from two Mind-Sets (segments) emerging from clustering. The clustering was done based upon the coefficients for response-time of the 16 elements, from the 25 respondents.

In order to assess the ‘meaningfulness’ of the segmentation based on response-time, it is necessary to look at the nature of the clusters in terms of what is responded to most rapidly. It appears that the words ‘United States’ drive the fastest response for Mind-Set1, and the words’ United Nations’ and ‘NATO;’ drive the fastest response for Mind-Set2.

It might well be that the segmentation and clustering on the basis of cognitive responses identify group differences due to ‘ideas’, whereas segmentation and clustering on the basis of response-time identify group differences due to specific ‘words.’ Finally, Figure 4 shows the relation between the coefficient for response-time for the element (ordinate) and the coefficient from the model for ‘likelihood of war.’ This time the Mind-Set segments MS1 and MS2 come from the segmentation by response-time. Mind-Set 1 in (Figure 4), focusing on the search for words, shows a clear relation between response-time and belief that war will break out. Mind-Set 2 in Figure 2 shows no such relation (Table 4).

Figure 4. The relation between the coefficient for response-time (RT) for the element (ordinate) and the coefficient of the same element from the model for ‘likelihood of war’ (ordinate.) The Mind-Set segments, MS1 and MS2 were obtained by segmenting the 25 respondents based upon the coefficients for Response-Time.

Applying the approach – assigning a new person to a mind-set

Our small study here identified a potential pair of mind-sets in the population, those who believe that the path to war occurs by direct action (Mind-Set 1) versus occurs by military build-up (Mind-Set 2.) We used only 25 respondents, but we were able to uncover two mind-sets when we clustered on the basis the coefficients derived from the ratings. The analogy here is the discovery of basic colors, the red, yellow and blue, with a small set of test stimuli. Mind Genomics allows us to identify these basic mind-sets even with a small group of respondents.

The next level of effort is to use this discovery of two mind-sets to understand the world. Examples of such understanding and fundamental problems to be addressed in light of our small discovery are:

How do these two mind-sets distribute around the world, by age, by gender, by government, by personal history?

Over time, does a person remain in the same mind-set? Are these two mind-sets fixed, or can a person first be a member of one mind-set, but through life experience change into the other mind-set?

Is there a relation between membership in a mind-set and education?

If one can do many of these studies on the political world, then can one extract other mind-sets for other topics, such as negotiation, and study the membership pattern of a single individual across many mind-sets?

Does membership in the mind-set co-vary with any exogenous, measured behavior, such as political activism?

And perhaps, most controversial, is there a relation between the genetics of an individual (e.g., revealed by chromosomal mapping) and membership in a mind-set?

One approach to predicting mind-set membership looks at the pattern of coefficients for the mind-sets (Table 3), and selects elements showing the greatest differentiating power, i.e., the biggest difference for the average panelist. Each selected element is then edited to become a question, to be answered NO or YES, or some other appropriate pair of responses for the same type of binary decision. The questions are incorporated into a short questionnaire (Figure 5, left panel.) The pattern of responses shows which mind-set is the likely mind-set of the respondent (Figure 5, right panel.) The approach is simple, quick, and works on summary data. The important thing to keep in mind is that the objective is to have the respondent rate single elements that are most discriminating between two mind-sets or among three mind-sets. It will be the pattern of ratings which will end up being most appropriate for a person in a specific mind-set. The algorithm will then assign the new person to the mind-set segment most likely to generate the pattern just obtained from the new respondent, the person waiting to be assigned.

Figure 5. The PVI, the personal viewpoint identifier, showing the questions and simple answers, used to assign a new person to one of the two mind-sets. The platform independent, online-based personal viewpoint identifier of the study is currently available directly through the following link: http://162.243.165.37:3838/TT03/.

Discussion and conclusions

During the past decades scientific inquiries have grown more expensive, longer, often harder to implement, and with results limited to a specific topic, almost ‘filling a hole in the literature.’ Mind Genomics, as we have presented it here, is evolving in an independent direction. Mind Genomics takes a ‘snapshot of reality’ in terms of the reactions of people to cognitively meaningful messages or ideas about a single topic of experience, relevant to the person’s life. The ideas in this paper are issues which are best put into the world of ‘current events’ but the ideas can range from moral issues to economic issues, education, and so forth.

The positive news from the study is that is appears quite possible to use small, inexpensive, easy-to-run studies to quantify how people respond to the world around them, and at the same time prevent the system from being ‘gamed.’ The further positive news is that, even with the low base size, it is often quite easy and affordable to uncover emergent mind-sets, putting the potential of discovery in the hands of experimenters without the concomitant cost. Thus, Mind Genomics in the current format, the BimiLeap APP, democratizes research, putting research and discovery into the hands of everyone. The negative news is that Mind Genomics cannot uncover a general clear relation between response-time, a physiological measure, and the response to elements, a cognitive measure. The two mind-sets created according to response-times emerge, as they must from clustering, but they make only modest intuitive sense. Although we can easily see differences in response pattern to elements after segmenting the pattern of coefficients for ratings, we see no correspondingly clear differences between two mind-sets emerging from the pattern of coefficients for response-times. Our first effort, using the physiological measure of response-time to understand mental processing, must be considered only modestly successful. We emphasize here that the only difference in the two clustering efforts, the first based on coefficients for ratings, the second based on coefficients for response-time, is the nature of the measure, cognitive versus so-called neuro or physiological. It may be that response-time in this form has to be further analyzed, incorporating other variables besides the element itself. The predictor variables might be the element and some morphological features of the elements as well. That effort is left to future research [13–15].

References

1. Galanter E, Pliner, P (1974) Cross-modality matching of money against other continua. In Sensation and measurement, Reidel Pg No: 65–76.
2. Stevens SS (1975) Psychophysics: Introduction to its perceptual, neural and social prospects. New York, John Wiley.
3. Box GEP, Hunter WP, Hunter JS (1978) Statistics for experimenters, New York, John Wiley.
4. 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.
5. Moskowitz HR, Gofman A, Beckley J, Ashman H (2006) Founding a new science: Mind genomics. Journal of sensory studies 21: 266–307.
6. Fugate DL (2007) Neuromarketing: a layman’s look at neuroscience and its potential application to marketing practice. Journal of Consumer Marketing 24: 385–394.
7. Genco SJ, Pohlmann AP, Steidl P (2013) Neuromarketing for dummies. John Wiley & Sons.
8. Charap S, Colton TJ (2018) Everyone loses: The Ukraine crisis and the ruinous contest for post-Soviet Eurasia. Routledge.
9. Russell W (2018) Russian Relations with the “Near Abroad”. In Russian Foreign Policy Since 1990 (pp. 53–70). Routledge.
10. Gofman A, Moskowitz H (2010) Isomorphic permuted experimental designs and their application in conjoint analysis. Journal of Sensory Studies 25: 127–145.
11. Moskowitz HR, Silcher M (2006) The applications of conjoint analysis and their possible uses in Sensometrics. Food quality and preference 17: 45–165.
12. De Hoon MJ, Imoto S, Nolan J, Miyano S (2004) Open source clustering software. Bioinformatics 20: 1453–1454. [crossref]
13. Bercík, Jakub, Elena Horská, Wang WY, Ying-Chun Chen (2015) How can food retailing benefit from neuromarketing research: a case of various parameters of store illumination and consumer response. In 143rd Joint EAAE/AAEA Seminar, March 25–27, 2015, Naples, Italy, no. 202714. European Association of Agricultural Economists.
14. Lee N, Broderick AJ, Chamberlain L (2007) What is ‘neuromarketing’? A discussion and agenda for future research. International journal of psychophysiology 63: 199–204.
15. Stipp H (2015) The Evolution of Neuromarketing Research: From Novelty to Mainstream: How Neuro Research Tools Improve Our Knowledge about Advertising. Journal of Advertising Research 55: 120–122.

Appendix

Nine recent studies with base sizes of 25-50 respondents, conducted in the same way as the current study. The graphs show the relation for the total panel, between the coefficient of response-time (ordinate) and the coefficient for interest (binary transformed, abscissa). The models were created from the ‘total panel data’ after the data in the first position was eliminated from the data set, leaving only 23 vignettes evaluated by each respondent.

DOI: 10.31038/ASMHS.2019314

Abstract

We present a new approach, Mind Genomics, to understanding the needs of prospective users with respect to a teaching APP designed to promote improved memorization of important texts. Using small-scale experiments, using the systematically varied messages in the form of stories or vignettes, Mind Genomics uncovers the customer-requirements of the APP. These vignettes are combinations of ideas about the product, its use, and the benefits to be obtained. The pattern of reactions to these vignettes reveals which specific features and benefits ‘drive interest.’ Mind Genomics does not require the respondent to intellectualize the need, an intellectualization which introduces response biases, and perhaps demand an answer that the respondent may not know. Rather, the deconstruction of the pattern of the immediate responses assigned almost automatically and without deep thinking, clearly reveals the underlying needs. The results from this small-scale study suggest three radically different mind-set segments. Mind Genomics, finds application where the respondent’s job is to make decisions, and where one would like to reduce the biases due to what the respondent expects the appropriate answer to be. We show how Mind Genomics can become an early-stage, rapid, affordable, and scalable system for deep understanding of human judgments.

Introduction

The Psychology of Memory and its Position in the 21^st Century Information World

Studies of memory lie at the historical foundation of experimental psychology. Among the earliest reported publications is Hermann Ebbinghaus’ book on Uber die Gedachtnisse, Memory, reporting in detail his extensive work with memory drums and rote learning [1]. Memory and its association with learning has not lost its allure for researchers, and has become increasingly important once again, almost two centuries after Ebbinghaus excited a waiting world with his experiments. This new excitement does not deal with the academic studies of memory and the myriad effects of the stimulus and the person as influences. Rather, this new excitement about memory revolves around the realization that in this new world of instant information, critical thinking, not rote memory, is important

As technology continues to improve, educators focus increasingly on technological aids to education, called in some circles ‘Ed-Tech.’ Computers promise to accelerate the development of thinking. For some areas such as information retrieval, computers have now, at least in the minds of some people, supplanted human memory as a key for one’s learning. As if to say: One need not ‘remember’ anything. Google®, Google Scholar® and other technologies can store and recall more in a moment than a person could remember in a lifetime. Indeed, as this 21^st century progresses, we see education in a maelstrom, as the new technologies conflict with old ways of learning.

One of the victims of this accelerated change in the way education is practiced comes from the loss of memorized information which comprised a person’s basic storehouse of knowledge. We no longer read very much, and our attention span is coming under suspicion as weakening. We are not disciplined in what we read, what we learn, since it is clear that computer-savvy young person, even as young as 8–10 years old, can extract enough information from web-sources that she or he can write a paper based on that “research”. Of course, their thinking won’t be as good as someone who has processed the information by thinking about it, but nonetheless the information will be there. Despite the plethora of information easily available, there is still the need for knowledge, memorized, processed, and incorporated into one’s mind, readily available for use in coping with the everyday [2, 3].

The foregoing is the negative part of today’s evolution, the loss of one’s store-house of information. Joshua Foer, who was the 2006 Memory Champion of the USA [4], co-founded a site, “Sefaria”, a storehouse of Jewish classic texts, searchable and clearly presented to any learner. When asked: “Why does anyone need to memorize nowadays, we have Google and Sefaria!”, Foer is said to have replied “Our memory is not like a passive bank account, in which the more you put into your account, the more you have to withdraw. Rather, our memory is like a lens, through which we see the world. What we remember actively guides our thinking, deepening our understanding. The more we remember – the better we think.” [5]

There is a positive part to the computer revolution as well. With the aid of machines, we can learn faster. Machines which provide feedback can become coaches, indeed tireless ones. A properly programmed machine can become a valuable ‘coach, when it can take the stimulus input, present it, acquires feedback on the subject’s reactions, and continue to do so, tireless, efficiently, hour after hour after hour.’

Psychology of Performance versus Psychology of Communication

Experimental psychologists are accustomed to studying processes, such as how we learn, the variables which drive the rate of learning and forgetting, and so forth. The focus of experimental psychology is on the person as a ‘machine, ’ with the goal to understand how this machine operates. The scientific literature of experimental psychology thus deals with well-contrived experiments, constructed to isolate, understand, and quantify aspects of behavior, such as learning and memory.

Less attention is given to what people ‘want’ in their lives. When we talk about a learning aid, we talk about what it does. The design of the machine, the so-called ‘customer requirements’ are left either to studies of human factors or studies of marketing, whether basic or applied. The discipline of Human factors studies the changes in behavior at the nexus of man-and-machine. Marketing studies what people want, with the goal of applying that knowledge to solve a specific, practical problem.

The study presented here incorporates aspects of experimental psychology, human factors, and marketing. The study here is an experiment, to explore how statements about features of a machine drive consumer’s responses. The experiment here was done in the spirit of human factors, to understand the aspects of the man-machine interaction. Additionally, the experiment was done in the spirit of marketing, to understand the types of mind-sets which may want different things, and the nature of the communications appealing to each mind-set.

Solving the Problem Using Experimental Design of Ideas (Mind Genomics)

Traditional methods to understand consumer requirements use a variety of different methods, ranging from an observation of what is being currently to (field observation), to focus groups which discuss needs, to questionnaires which require the respondent to identify what is important from a list of alternatives, and down to so-called A/B tests where the respondents experience alternative instantiations of a product, and the researcher observes which instantiation performs better, makes the changes, and commissions another A/B test.

Although a great deal of consumer research assumes that people ‘know’ what they want, the reality is that they do not. Kahneman & Egan [6] suggest that we operate with at least two systems of decision-making, the ‘Fast’ and the ‘Slow’, respectively, called ‘System 1’ and ‘System 2.’ In our regular lives we are presented with compound situations containing many different cues, situations to which we must respond quickly. We have no time to weigh alternatives in a considered fashion. The rate at which these compound situations come at us can be numbing when we stop to count them. Consider, for example, driving quickly, and the many decisions that must be made, especially when maneuvering in traffic.

The complexity of decision making, the involvement of Systems 1 and 2, respectively, in this emotionally tinged topic of learning to remember makes it imperative that we move away from simplistic methods of ‘asking people what they want, ’ and, instead, do an experiment in which what people want emerges from the pattern of responses, without any intellectualization on the part of the individual.

Mind Genomics eliminates the problems encountered with many of the approaches which require the respondent to intellectualize what may be impossible to intellectualize, much less to communicate. The objective of Mind Genomics is to identify the importance of alternative features of an offering by presenting many descriptions of the offering, instructing respondents to consider each description as a possible product, and then to rate the description. The respondent is not instructed to reveal the reasons for accepting or reject each specific alternative, but rather, almost in a non-analytical way, rapidly evaluate the offering quickly, almost automatically, as one does with small purchases. The pattern of reactions to the different offerings reveals what features of the offering are important, and what features are irrelevant, or even off-putting.

The Contribution of Experimental Design

The experiments in Mind Genomics are patterned after the way nature presents its complexity to us, but in a more structured format. Mind Genomics studies combine individual pieces of information, ‘messages’ or ‘ideas, ’ doing so by experimental design [7]. The combinations, vignettes, are presented to the respondent who is encouraged to make a decision, doing so rapidly, e.g., rate the vignette on an attribute. The experiment comprises the presentation of a set of these vignettes, here 24 in total, to each respondent, who reacts to the vignette, rates, and moves automatically to the next vignette, repeating the process. The experimental design, in turn, enables the researcher to deconstruct the rating into the contribution of the individual elements, the messages. To the respondent, the array of alternative vignettes evaluated in the space of five minutes or so might seem to be a numbing set of randomly combined ideas, but nothing can be further from the truth.

As will emerge from the analysis of responses to a description of a new APP, Shanen-Li, designed to help memory, consumer demands emerge quite clearly from the descriptions. Consumers are asked simply to be participants to evaluate ideas. They are not asked to be experts, nor even to proffer their opinion, but simply to give their immediate, so-called ‘gut’ reaction to each vignette or test combination.

The Mind Genomics Process – Setup

The first step in Mind Genomics asks four questions, and for each question, requires four simple answers, or a total of 16 questions. The questions are never presented to the respondent. Only the answers are presented in combinations, as we will see below. The questions provide a structure to generate the answers. It is the answers which provide the necessary information about the Shanen-Li APP.

A parenthetical note is appropriate here. When one begins the process of creating a Mind Genomics experiment, the notion of question and answer is easy to comprehend. The questions which, in sequence, tell a story, are themselves difficult to create, at least for the first two or three studies. The answers themselves are easy to create once the questions are formulated. Over time, and with repeated experience, the novice begins to think in this more orderly fashion of telling stories through questions and providing the substance of those stories through the answers. In a sense, the Mind Genomics process may somehow ‘train’ the user to think in a new, structured way, one which forces a discipline where there may not previously have been discipline.

One of the key features of Mind Genomics is that one need not know the ‘right answers’ at the start of the process, a requirement which is often the case for more conventional studies. Rather, Mind Genomics system is designed to be iterative, inexpensive, and rapid. That is, one can do a study in a matter of a few hours, identify the important messages or elements, discard the rest, and, in turn, incorporate new elements to the next iteration. Within the space of a day it is possible to do 3–4 iterations, and by the end of the four iterations one should have come upon the strongest messages. In this paper we present the first iteration in order to demonstrate the nature of the process and the type of learning which emerges.

Assembling the Raw Materials to Tell ‘Stories’

The first step in a Mind Genomics study consists of asking a set of questions which ‘tell a story.’ As noted above, this first step may seem easy, but it is not as simple as one might think. The objective is to summarize the nature of the stimulus through questions. Table 1 shows the four questions for the first study. These questions give a sense of a story. The rationale for these questions beyond ‘telling the story’ is to evoke answers, or elements, the messages containing the actual information which will appear in the test vignettes. The questions never appear in the study. They are only an aid to structure the vignette, and to stimulate the researcher to provide the meaningful elements which convey information, in this case information about the APP.

Table 1. The four questions and the four answers to each question.

Each question, in turn, requires four answers to the question. As Table 1 shows, the questions are simple, and the answers are equally simple. Every effort is made to avoid conditional statements, and statements which require a great deal of thinking. Furthermore, the answers are phrased in every-day language, in words that a person might use to describe the APP or the experience of memorization.

When doing these types of studies, one often feels ‘lost’ at the start of the process. Our educational system is not set up to promote critical thinking of a Socratic nature, the type of thinking required by Mind Genomics. The notion of telling a story through questions is strange, as if the notion of providing alternative answers which may be ‘what is, ’ and ‘what could be.’ Nonetheless, with practice the exercise soon becomes easier, although it is not quite clear at this writing (2019) whether this Socratic approach can replace the traditional thinking, or whether the approach can be practiced more easily with repeated efforts.

Creating Stories by Experimental Design (Systematic Combinations)

People often respond based upon what they think the right answer either ‘IS’, or what they believe the appropriate answer to be. Questionnaire-based research is especially prone to mental editing, response biases, based on belief, or based on the covert, often un-sensed desire to please the interviewer. Computer-administered questionnaires may compensate for the latter, because there is no interviewer, but rather a machine. It may be difficult for the respondent even to respond to machines when the topic is emotionally tinged.

Mind Genomics moves in a different direction, using experiments to understand the mind of the respondent. In a Mind Genomics experiment, the respondent is presented with combinations of messages, one message atop the other, such as that shown in Figure 1. The respondent’s job is simply to rate the combination on a scale, without having to explain WHY the rating was assigned. It is hard at first for a respondent to evaluate this type of mix of messages because people have been taught to deconstruct compound stimuli, and then to evaluate each part of the compound stimulus. The notion of rating an artificially combined set of messages moving in different directions is at first strange, but then becomes very easy by the time the respondent rates the second or third vignette.

Figure 1. Example of a vignette and a rating scale for the APP.

Although the vignette in Figure 1 appears to have been designed by randomly throwing together different combinations, the truth is the opposite. The 24 vignettes for a respondent are carefully crafted so that the 16 elements, the independent variables, are statistical independent of each other, and that each element appears an equal number of times.

Table 2 shows schematics for the first eight vignettes for respondent #1. The vignettes are first presented in the original design format (top section), and then shown in a binary expansion (middle section). The regression program cannot work with the original design, expressed in terms of the questions and answers in each vignette. It is necessary to recode the design so that there are 16 independent variables, which, for any vignette, take on the value 0 when absent from the vignette, and take on the value 1 when present in the vignette.

Table 2. Part of the actual experimental design..  The table shows the first eight vignettes of 24 for the first respondent.

The bottom of Table 2 shows the two ratings and the response time. The first rating is the number on the anchored 9-point scale. The second number is the transformed rating. The transformation is done so that ratings of 1–6 are transformed to 0 and ratings of 7–9 are transformed to 100. Afterwards, a very small random number (<10^–5) is added to the binary transformed ratings to ensure that the OLS (ordinary least-squares) regression will work, no matter whether the respondent uses the entire 9-point scale, or limits the ratings to the low part of the range (1–6), or limits the ratings to the high part of the range (7–9), respectively The transformation makes it easy for researchers and managers to understand the meaning of the numbers. Researchers and managers want to learn whether a specific variable, in our case a message, drives the answer ‘no’ (not interested) or yes (interested).

Each of the respondents is assigned a different experimental design, created by permuting the elements [8] the same mathematical structure and robustness of design is maintained, but the specific combinations change. This strategy differs from the typical research approach which ‘replicates’ the same test stimuli across many respondents in order to obtain a ‘tighter’ estimate of the central tendency. With more respondents, the standard error drops, and the researcher can be more certain of the repeatability of the result. This statistical strength is achieved by repeating the experiment with a limited number of test stimuli, chosen to represent the wide range of alternative combinations. Mind Genomics works in a totally different way, covering a lot more of the space, albeit with fewer estimates of any single combination of elements, i.e., fewer replicates of the same vignette. Often there is only one estimate of the vignette. The rationale is that it is better to cover a wide range of alternative stimuli with ‘error’ than a narrow and perhaps unrepresentative range of stimulus with ‘precision.’

Individual Differences: Average Liking of the Vignette versus Average Response

Do individuals who like the ideas about the Shanen-Li APP respond any faster (or slower) than individuals who don’t like the ideas? In other words, is there a discernible pattern at the level of the individual respondent, so that those who like an idea (on average) respond faster or slower than those who don’t like an idea?

Figure 2 shows a plot of the average rating of liking (average binary response) versus the average number of seconds (average response time). Each filled circle corresponds to one of the 50 respondents. It is clear from Figure 2 that, at the level of the individual respondent, there is no clear relation between how much a person ‘likes’ an idea presented by the vignette and how rapidly the person responds to the vignette. Those who, on average, don’t like the idea of the APP respond quickly or respond slowly as those who, on average like the idea.

Figure 2. Relation between response time (ordinate) and liking of the idea (abscissa). Each filled circle corresponds to one respondent, whose ratings and response times, respectively, were averaged across the 24 vignettes.

Modeling

A deeper understanding of the dynamics of decision making emerges when we deconstruct the ratings (here the binary transformation) into the contribution of the individual elements. The experimental design ensures that the 16 elements are statistically independent of each other at the level of the individual respondent. Combining the data from the 50 experimental designs into one grand data set comprising 1200 observations, 24 for each of 50 respondents, allows us to run one grand analysis using OLS (ordinary least-squares) regression. OLS will deconstruct the data into the part-worth contribution of each of the 16 elements,

Table 3 shows the results of the first analysis, wherein the dependent variable is the binary transformed data (ratings of 1–6=0; ratings of 7–9=0), and wherein the independent variables are the 16 elements. The elements take on the value 1 when present in a vignette, and the value 0 when absent from a vignette.

Table 3. Coefficients of the model relating the presence/absence of the 16 elements to the binary transformed model for ‘like using this APP.’

The equation estimated by OLS regression is expressed as: Binary Rating = k₀ + k₁(A1) + k₂(A2) … k₁₆(D4)

The additive constant is the expected percent of times that the binary value will be 100, in the absence of elements. All vignettes comprised at least two and at most four elements, so the additive constant is a purely estimated parameter. Nonetheless, the additive constant can be thought of as a ‘baseline’ value, namely the likelihood of a positive response towards the APP in general.

The additive constant is 44.41, meaning that in the absence of specific information; we are likely to see almost half the responses being strongly positive. The value 44.41 is a bit shy of 50. The T-statistics tells us the ratio of the additive constant to the standard error of the additive constant. The T-statistic can be thought of as a measure of signal to noise, of the value of the additive constant to the variability of the additive constant. The ratio is 5.71, quite high, with the probability of seeing such a high ratio being virtually 0 if the ‘true’ additive constant were really 0.

When we look at the individual elements for the total panel, we find that the coefficients are quite low, with the highest coefficient being 1.66. The coefficient tells us the expect increase or decrease in the percent of respondents who say that they would be interested in the APP if the element were to be included in the vignette. We begin with the additive constant (44.41) and add the individual coefficients of the elements.

What is remarkable is the low value of the coefficients for the total panel. The highest performing element is B3, ‘The student experiences a sense of success, that turbocharges motivation.’ The coefficient is only 1.66, i.e., about 2. The T statistic is 0.35, meaning that it’s quite likely that the real coefficient is 0.

There are some elements which, in fact, are negative, pushing respondents away.

Now there is a plan to succeed
Students become masters faster than they can imagine

Looking at Key Subgroups

We now move to an analysis of subgroups, specifically gender, age, and then mind-set segments. The respondent gender and age are obtained directly from the experiment. Respondents are instructed to give their gender (only male versus female), and to select the year of their birth.

For mind-set segments, we use the well-accepted method of cluster analysis [9] to discover complementary groups of respondents which respondent differently and meaningfully to the 16 elements. The experimental design allowed us to create an individual-level model relating the presence/absence of the elements to the binary-transformed ratings. Each respondent generates a unique pattern of 16 coefficients. We combine respondents into complementary groups with the property that the patterns of coefficients in a group (mind-set segment) are similar to each other, and differ from the average patterns for the other groups. The actual segmentation uses a measure of distance between respondents defined as (1-Pearson Correlation). When two patterns perfectly correlate (Pearson Correlation = 1), the distance is 0. When two patterns perfectly inversely correlate (Pearson Correlation = -1), the distance is 2.0.

Table 4 shows the additive constants and the strong performing elements for each defined subgroup. What should become immediately apparent is that:

Table 4. Strong performing elements by subgroups.

1. The additive constant is modest, except for the respondents who are age 26+ (a small group). The respondents accept the ideas of a tutoring APP of this sort, but it will be the elements which must do the hard work.
2. The strong-performing elements really occur among the mind-sets. That is, the opportunities do not lie among the respondents based on gender or age, but based on mind-sets.
3. We do not know the mind-sets ahead of time. The mind-sets must be extracted through analysis of patterns, only after the experiment has been run.
4. The key to success for this product is the array of mind-sets emerging from the segmentation. Even with the mind-sets, only a few elements drive interest, but when they do, they do strongly
5. At the end of the paper, we present an approach to discover these mind-sets in the population.

The Speed of Comprehension and Decision Making

Before the respondent rates a vignette, we assume that the respondent has read and comprehended the material in the vignette. Although the responses occur very rapidly, suggesting very quickly reading and decision making, we can still uncover the relation, if any, between the element and the speed at which that element is comprehended. Figure 3 shows that many of the vignettes are responded to within a second or two.

Figure 3. Distribution of response times for the 1200 vignettes. Response times of faster than 8 seconds were truncated to be 8 seconds, under the assumption that the respondent was otherwise engaged when participating in the experiment, at least when reading the vignette.

The response time it does not tell us much. We do not understand the dynamics of response time, specifically in this experiment, why some vignettes took longer times, some took shorter times. One way to discover the answer deconstructs the vignette into the contribution of the different elements to response time, in the same way that we deconstructed the contributions of the elements to the binary transform. The only difference is that we write the equation without an additive constant. The equation, written below, expresses the ingoing assumption that without elements in a vignette, the response time is essentially 0.

Table 5 presents the same type of table as presented by Table 3, namely a full statistical analysis of the elements, showing their coefficient, the t statistics (a measure of signal to noise), and the p value for the coefficient. The difference between Tables 3 and 5 is that for the binary rating, the model contains an additive constant because the ingoing assumption is that there is a predisposition towards the topic of a memory-training APP, but there is no predisposition in the case of response

Table 5. Coefficients of the model relating the presence/absence of the 16 elements to the binary transformed model for ‘response time.’ All response times of 8 seconds or more for vignette were transformed to 8 seconds.

It is clear from Table 5 that most of the elements are reacted to quickly, as suggested by the coefficient, which is a measure of the number of seconds. The fastest elements are those which paint a word picture of an activity, and which may be visualized. The slowest elements are those which talk about less concrete topics, such as motivation and feelings.

Subgroups – Do They Respond at Different Speeds to the Elements

When looking at the deconstructed response times in Table 4 we see that virtually all response times range between one-half second and one second, respectively. There is no sense of any large differences between elements. A one-half second difference is still quite rapid. The story is quite different, however, when we look at subgroups defined by gender, by age, and then by mind-set segment. Table 6 shows those elements which catch the respondent’s attention, operationally defined as taking more than 1.15 seconds for the element to be ‘processed. Combining a high scoring element for interest with a high scoring element for response times means bringing together an interesting element which maintains the respondent’s attention during the stage of ‘grazing for information.’

Table 6. Elements showing slow response times, suggesting that they ‘catch’ the respondent’s attention.

Discovering the Mindsets in the Population

Mind-sets distribute through the population. The traditional approach to produce development and marketing often believed that ‘you will need or believe based upon WHO YOU ARE.’ The notion of ‘WHO YOU ARE’ may be a result of the person’s socio-economic situation or may be a result of a person’s general psychographic profile [10]. The premise of Mind Genomics is that people fall into different groups, Mind-Sets, not necessarily based on who they are, nor on what general things they believe. Table 7 shows that even for this small base of respondents, the three Mind-Sets distribute in almost similar ways across interests, gender, and age. A different method is needed to identify Mind-Sets emerging from these focused studies. It is unlikely that the Mind-Sets for this new-to-the-world division into Mind-Sets for this APP can be found in the analysis of so-called Big Data. A different method is need, the PVI, Personal Viewpoint Identifier, described below.

Table 7. Distribution of the 50 respondents in the three Mind-Sets, by interest for memorizing, by gender, and by age