DOI: 10.31038/MGSPE.2022221

Abstract

Mind Genomics was used to assess the response of ordinary people to different prospective strategies involved with the nuclear deal with Iran, in 2016. Each respondent read a unique set of 25 small vignettes comprising systematically varied messages about the nuclear deal, rating each on likelihood for an agreement, and expected emotional response from Iran. From the set of 20 elements only seven elements performed strongly, but not among the total panel, only among emergent mind-sets. These were MS1 (Focus on military aspects, specifically prevention, n=29 respondents), MS2 (Focus on economic development, n=45), and MS 3 (Focus on effective negotiations and diplomacy, n=11). Most of the emotional reactions were negative. The paper suggests that Mind Genomics be used as an iterative, low cost, rapid fashion, to identify strong negotiating points, base upon the mind of the average citizen. The iterations each lasting 3-4 hours, with several iterations possible in a day at low cost, and with deep learning may radically change the process of negotiation. Mind Genomics identifies what specific messages ‘work’. The process can evolve to a joint effort by both parties to the disagreement, and by so doing craft an agreement attractive to both sides, an agreement emerging from the positive responses of the citizens of both sides.

Introduction

The world of US policy the domain of the three branches of the government, and in practice the domain of a host of consultants and others helping to formulate the policy. Often the policy seems well thought out, other times the policy seems to be either poorly thought out, or more of concern, the influence of various parties which dictate aspects of policy for their own interest.

The topic of this paper is the introduction of a tool, Mind Genomics, to help formulate policy by understanding the ‘mind’ of the average citizen, in a way that could tap into the ‘wisdom of the crowd’, and become an iterative, affordable, rapid tool to help policy formulation.

We illustrate the approach by a study run five years ago on responses to policy about Iran. The objective of the study was to demonstrate the potential of what one could learn in a matter of two days, a time that would be shortened to period of 2-4 hours as of this writing (Fall, 2021). The topic of what to do with the fractious government of Iran continues to rear its head. At the original time of the experiment, the last months of the Obama administration, the issue was raised as to what could be done to deal effectively with Iran. Donald Trump was in the midst of pre-election efforts. The research was done to identify key issues and what people wanted as support for the Republican party.

Formulation of Public Policy with the Aid of Polls

Public policy is often announced by a spokesperson for the committee putting forward that policy. It is obvious from the reports both before, and during the birth of the policy, that the policy was ‘crafted’ by a group, and that often the group is bipartisan. There is the phrase ‘horse-trading’ to discuss the back-and-forth negotiations.

At the same time, in the world of politics, whether for candidates or for political issues there are two worlds intertwined. One world is the world of experts, such as individuals from so-called think tanks, who come up with the recommendations. In the United States these individuals are disparaging called ‘Beltway Bandits’, because are housed near Washington. The experts are highly paid to work with the lawmakers and policy makers, to give advice. Occasionally, scientists enter the process as well because the issue is technical [1,2].

At the same time there are the pollsters, who measure public opinion, attitude. The emphasis here is on accurate measurement. Occasionally these pollsters might be asked to consult on policy, but their expertise is accurate measurement. The measurement may occur with well conducted local and national polls, focus groups, individual depth interviews, perhaps coupled with their own observations of what is happening at the time they are doing the research [3].

There are two languages in policy, the language of artisanship in the creation, and the language of statistics and measurement in people’s response to the creating. The language of policy creation is the language of the artisan shop, where the policy is ‘crafted,’ ‘hammered out’, etc., through the interactions and efforts of the individuals involved. The policy is ‘created’ by those tasked with the job. We can contrast this policy of ‘artisanship’ with the language used in measuring responses to the policy, the language of statistics, polls, degree of confidence, measurement of trends, and assignment of reasons for specific patterns of people’s response to the policy. Furthermore, he two languages do not overlaps. There is not much published in terms of scientifically guided iterations in the development of the artisan-crafted policy. The two worlds are different, creation and measurement.

In contrast to the above is the world of product design, especially the world of software design, but engineering in general. The product may be created by an artisan, but that product is special, one time. The true effort is to create products which work, products that have been created by iterations, with the creation coming first, then the testing, then the revision, and the testing again [4]. The key word is ‘testing as part of the iteration,’ something that is not heard publicly in the world of government policy.

A search through the literature reveals a moderate number of papers on policy, but almost none on measurement during the course of policy development in the way that one might iterate in the creation of software. We might we be in different worlds. Policy again and again seems to be crafted as a one-tine reaction, rather than being quickly evolved from iterations and testing propositions in the policy. It is that opportunity, creation and optimization through iteration, which constitutes the contribution of this paper.

Beyond Polls to Experimentation

The notion of experimentation in political science seems at first strange, simply because one thinks of the political order as an emergent, resulting from the confluences of forces and the ‘Zeitgeist,’ the spirit of the times. Philosophers have debated the nature of the political orders, the classes of political orders, and of course both the assumed ‘original political order of man’ (if there ever was one), and the most appropriate political order for a society. The important thing to note is that political order is so critical that it begs for study, whether for itself or knowledge of which allows one to achieve one’s goals.

At the same time, during the past decades there has emerge a notion of experimentation, and the idea of an experimental political science, perhaps of the same type as experimental psychology. The difference is where the material is published, and the nature of the published material. Experimental psychology began to emerge in Germany almost two centuries with the publication of Ebbinghaus’ book ‘On Memory’. The book was filled with the results of experiments, with data that could be studied, reanalyzed, challenged, and ultimately replicated or not.

We can contrast the early beginnings of experimental psychology with the beginnings of experimental political science, whose material appears in book after book, as points of view, substantiated with one or two experiments, or better rethinking of data [5,6]. There are no standard experiments in political science, experiments which constitute the basis of the science. Rather, there is talk, philosophical point of view, the need and from time-to-time re-presenting data, cast in this new light of experimentation. In other words, experimental political science is very much alive, but as hope for the future, not as a daily, simple, scalable system for producing data and knowledge. We are just not ready although the interest is certainly real, as shown by the intellectual vibrancy of the topic, a ‘must’ for breaking through into new territory [7-9].

The Mind Genomics Approach

Mind Genomics is an emerging science with roots in experimental psychology, marketing research and public polling. The fundamental nature of Mind Genomics is of a science of experimentation which discovers the mind of people with respect to a specific micro-topic. The key word is micro-topic, a focus on easy-to-understand ideas. The objective is to quantify decision making from the bottom up, and identify coherent groups, ‘mind-genomes’, based upon different, recurring patterns describing how individuals make judgments about the world of the everyday [10,11].

The part of Mind Genomics emerging from experimental psychology is the focus on the measurement of ideas, the inner psychophysics as it was called by modern day psychophysics pioneer, S.S. Stevens of Harvard University. Psychophysics itself is the search for lawful relations between physical stimuli and subjective responses, so-called outer psychophysics. It is the aspects of psychophysics to which most scientists familiar with psychology and referring to when they refer to psychophysics. Inner psychophysics, Stevens’ dream, was to apply metrics to ideas, to measure ideas.

The part of Mind Genomics emerging from consumer is the use of mixtures of test stimuli which simulate real world stimuli have cognitive meaning. One of the tools of consumer research, coincidentally developed by experimental psychologists Luce and Tukey is ‘conjoint measurement,’ the evaluation of mixtures of stimuli, and the estimation of the contribution of each element in the mixture to the overall response. In the world of commerce, mixtures are importance. They are the substance of which products and services are composed. We don’t buy single ideas, but rather combinations of features and benefits embedded in a product or a service.

The Seven Steps

Mind Genomics follows a templated process comprising seven steps. The steps begin with the creation of raw material, and finish with the identification of strong performing elements, among defined groups of respondents, including new-to-the-world groups of respondents who can be shown to think alike on this topic. The output of the Mind Genomics study may find use in driving a better program of communication of one’s product, or part of an academic effort to create the ‘wiki of the mud for a set of related issues’

Step 1: Define the Problem, Create the ‘Raw Material’, Defined as a Set of ‘Questions’, and a Specified Number of Answers to Each Question

The Mind Genomics effort is an experiment, rather than a questionnaire, although Mind Genomics has often been defined in public terms as a survey’.

The essence of Mind Genomics is to measure responses to defined stimuli, viz., combinations of messages, and these combinations called vignettes. The vignettes are combinations of statements about the topic, in our case policy towards Iran. As a consequence, the Mind Genomics process prescribes the raw material, namely the topic (Iran), a set of ‘questions’ or ‘categories’ which in sequence describe or tell a story, and for each question or category, an equal number of ‘answers.’

The approach for finding the raw material may range from sheer expertise and ‘off the cuff’ to serious research into what is in published. With the growing interest in Mind Genomics as a fast, iterative process, the movement is towards simple, superficial ideas, some based upon what has been seen or read in public sources, the others based upon one’s own ideas, or the ideas of a creative group, thinking about the topic.

Table 1 shows the list of elements. The structure of the table, four questions, five answers per question, is based on the one of the designs of the Mind Genomics system. The elements were created by author Bitran based upon his on strategic analysis work with his program, Enterprizer(r). It is important to keep in mind that Mind Genomics is a tool which puts the elements to a hard test, as we will see below. The iterative nature of Mind Genomics will allow strong elements to emerge. At the same time, however, the Mind Genomics system is not ‘creative’. And so, a good knowledge of the topic is helpful but not a requirement.

Table 1: The five questions (categories) and four answers (elements) for each question

Step 2: Create Short Vignettes Using Experimental Design

The world of science works by identifying a phenomenon of interest, defining aspects of the phenomenon to be studies, and when possible, isolating those aspects of interest, and measuring them. The aim is to determine the nature of the variable of interest. Doing so means reducing the haze around the variable, the random variation which hides that nature of the variable. The variability itself is unwanted and eliminated through research. The two strategies are to isolate the variable, eliminating extraneous forces which lead to variation, or measure the variable many times, under different situations, and average out the unwanted variation.

When we deal with issues of foreign policy and break out the issues into elements such as those shown in Table 1, the typical research strategy would be to polish each element so that each element is as clear as possible, and as simple to understand as possible. That corresponds to the first effort, measuring the variable which has been made as simple as possible, so other factors do not affect the results. The second is to test that single idea with hundreds of people, one idea at a time with each of the hundreds of people. Averaging the results from the large group should provide a stable measure of the response to the variable.

The one-at-a-time method dictates that the researcher presents the respondent with each of the elements, one element at a time as the phrase says. The respondent is instructed to maintain the same criterion, and with that one criterion rate the element. It does matter whether the element is positive, negative, deals with peace, deals with conflict; the respondent is to use the same rating scale all the time.

An ongoing problem in the on-at-time research is the unnaturalness of single elements. There is no context. The rating is easier when all of the test stimuli, the elements, are of the same type, such as military alliances, or economic alliances, educational strategies, and so forth. The respondent reads the elements, all of the same time, and has no problem evacuating the elements themselves. They are commensurate with each other. The problem arises when the elements are different. The differences may be vast, such as economic policy versus military policy. Although the researcher can instruct the respondent to use the same criterion, it is not clear that the respondent can actually do so.

A better approach, one which removes some of the artificiality of the one-at-a-time method, works by creating combinations of ideas. This is the approach used by Mind Genomics. Rather than forcing the respondent to maintain the same criterion with palpably different types of statements, Mind Genomics puts together the ideas or statements into small easy to read combinations, such as that shown in Figure 1. There is no effort to polish the combination, or to create connectives so that the combination is even more natural looking, appearing like the paragraphs that the respondent is comfortable evaluating. Although the critic might aver that the combination is not polished, that there are no connectives, that some of the laws of grammar are violated, the reality is that the combination forces the respondent to adopt one criterion and keep it because it is impossible in a Mind Genomics experiment to continue to shift judgment criteria to match what ends up seeming to be an ever-changing set of random combinations of ideas. The easiest way is to maintain one’s judgment criteria in the face of ever-changing combinations.

Figure 1: Example of a four-element vignette. Each respondent evaluates 25 unique vignettes. The vignettes for each respondent differ from each other

The combinations themselves may appear to the respondent to be utterly random. Nothing can be further from the truth. The combinations are created according to an experimental design [12]. The experimental design comprises specific combinations, allowing the variables to interact, but making sure that the 20 elements in this particular case are presented so that they are statistically independent of each other. That statistical independence is accomplished by the specific combinations. The design comprises 25 combinations or vignettes. Each vignette has a specified number of elements, at most one element or answer from any question. The vignette structure is:

Two elements in the vignette – 2 of 25 vignettes

Three elements in the vignette – 4 of 25 vignettes

Four elements in the vignette – 11 of 25 vignettes

Five elements in the vignette – 8 of 25 vignettes.

Although some critics might aver that the vignette has to be complete, with one element from each of the five categories, the reality is that respondent has no problem dealing with the sparser vignettes. The problem is the attitude of the researcher who wants completeness.

The basic design of 20 element embedded in 25 vignettes is a very efficient design. The breakthrough is design came around 1998, when the notion emerged of a permutable design. That is, one could create the basic mathematical structure of the design, specifying the combinations, and so forth. Once this was done, I was simple and straightforward to create a basic design, and then permute it, changing the elements, but maintaining the design structure. That meant renumbering the elements but keeping the elements in the same category. Thus, A1 would become A2, A5 would become A4, and so forth. The renumber would be done for all elements. This strategy, described in detail by Gofman & Moskowitz (2010), maintained the structural integrity of the experimental design, but recrafted the design slightly to cover many more of the possible combinations.

Figure 1 shows an example of a four-element vignette. The physical layout is simple, one phrase atop the other. There is no indication of categories or questions, simply a combination of the elements. No effort is made to connect the combinations.

Step 3: Execute the Study (viz., Experiment) Online

The actual study was executed through an on-line panel provider, specializing in recruiting respondents and providing them for these studies. The company, Luc.id Inc., in Louisiana, USA, is an aggregator of respondents from various panels. Working with a panel provider such as Lucid. makes the process easy. Over the past two decades it has become increasingly difficult to recruit one’s own panelists, especially for interview or experiments lasting 10+ minutes. The refusal rate has skyrocketed. As a consequence, the panel providers can deliver a group of respondents, generally filling easy specifications, for a reasonable price.

The respondents were invited to participate. The respondents were shown the following orientation. Note that a link was given for further reading about the JCPOA

By way of background Wikipedia as this this writing (Fall 2020) presents a background to the JCPOA, the Joint Comprehensive Plane of Action, which was signed in 2015.

Under the JCPOA, Iran agreed to eliminate its stockpile of medium-enriched uranium, cut its stockpile of low-enriched uranium by 98%, and reduce by about two-thirds the number of its gas centrifuges for 13 years. For the next 15 years, Iran will only enrich uranium up to 3.67%. Iran also agreed not to build any new heavy-water facilities for the same period of time. Uranium-enrichment activities will be limited to a single facility using first-generation centrifuges for 10 years. Other facilities will be converted to avoid proliferation risks. To monitor and verify Iran’s compliance with the agreement, the International Atomic Energy Agency (IAEA) will have regular access to all Iranian nuclear facilities. The agreement provides that in return for verifiably abiding by its commitments, Iran will receive relief from the U.S., European Union, and United Nations Security Council nuclear-related sanctions. https://en.wikipedia.org/wiki/Joint_Comprehensive_Plan_of_Action

The study was complete by 85 respondents, recruited by Luc.id. The base size of 85 suffices for a simple but often deep focus on the topic. The reason for the need for fewer than the hundreds of respondents in conventional survey work is that the research is searching for patterns, not for a precise measure of one point.

Step 4: Prepare the Data for Analysis by Creating New Binary Variables

The Mind Genomics exercise produces a great deal of data, since each of 85 respondents evaluated 25 different vignettes on two types of attributes, a degree of belief in the meaningful agreement (1=Definitely No … 9=Definitely yes) and a selection of the emotion that would be felt by Iran, if Iran were a person.

Our goal is to link the 20 elements to the ratings and the emotions. We do that in the next section. in this first section we transform the 9-point rating to a no/yes scale. Managers find it easier to work with binary scale, rather than to talk in percentages. Following the convention of previous efforts with Mind Genomics and the 9-point scale, we recode ratings of 1-6 to 0 (low probability), and ratings of 7-9 to 100 (high probability). The recoding could be made more stringent or less stringent. There is no ‘right answer,’ just appeal to previous processes. We do the same type of recoding for the emotions. We recode emotions as positive) negative). (Positive: Happy, Relieved, Victorious; Negative: Defeated, Fearful).

Thus, each vignette ends up with three numbers. One for the binary recode for probability of meaningful agreement, one for a positive emotion, and the complement for a negative emotion. The numbers are either 0 or 100. When it comes to the positive versus negative emotion, one of the two variables will take on the value 100, and the other by definition will take on the value 0.

Finally, vanishingly small random number is added to every newly created binary scale. this random number does not affect the results but does avoids a problematic statistical issue emerging from OLS (ordinary least0squares) regression occurring when the respondent selects all ratings for question 1 (meaningful agreement) either 1-6 or 7-9 (all 0’s or all 100’s across the 25 vignettes) or select all emotions as positive or all emotions as negative.

Step 5: Compute Means for to Better Understand the Patterns of Responses

By Step 5 we have already put the data into a form that makes it easy to compare average ratings (the focus of this step), and to link the elements to response (focus of Step 6).

We can explore the quality of the data by computing averages, considering both the number of elements in a vignette as a stratifying variable, and considering the order of testing as a stratifying variable. Even without knowing which elements are present in a vignette, we can ask whether there are any differences in the average ratings assigned to vignettes of 2,3,4 or 5 elements respectively, in terms of the binary transform of likelihood of agreement (TOP 3), and for the average Positive and average Negative emotions.

To answer the foregoing questions means simply to identify each vignette in two ways, first by the number of elements (2, 3, 4 or 5, respectively), and second by the position in the respondent’s sequence (first third, second third, final third).

Table 2 that there is no difference by position in terms of how it covaries with either likelihood to reach agreement (Q1) of emotion selected.

Table 2 also shows no effect of number of elements in terms of affecting the likelihood to reach agreement. There is, however, a quite strong and inverse covariation between the number of elements in the vignette and the selection of a positive emotion. Shorter vignettes are perceived as more likely to generate a positive emotional response by Iran, perhaps because the terms are defined, and the permission is direct. That is, shorter vignettes may leave less ‘wiggle room’, ‘and less ‘fine print’ in the agreement.

Table 2: Average values for TOP3 (likelihood of an agreement), and emotions selected (positive, negative) versus order of testing, and the number of elements in a vignette

The final topic of our surface is analysis is to get a sense of how the respondents feel about what they are reading. Question 1 allows us a sense of whether respondents feel optimistic about the process, viz., that it will happen, or feel pessimistic. Question 2 gives us a sense of their emotions. Let us average the ratings from their reactions to their own 25 vignettes, independent of what is in the vignettes. (Although, we know that each element appears equally often in the 25 vignettes; it’s just the combinations which vary).

Figure 2 shows a scatterplot of the average score for ‘reach agreement’ (% rating 7-9) vs. the average percent of selections of a positive emotion. Figure 2 shows a concentration of respondents on the left, with low average value of TOP3. We conclude from this that the individual respondents, on average, feel that the agreement will not be reached. There is no sense, however of a preponderance of emotions. Respondents simply do not seem to be able to figure out what the feelings of the Iranians will be a finding which should not surprise. Response can feel strongly about the outcome but not feel strongly about the expected feelings emerging from that outcome.

Figure 2: Scatterplot showing the average ratings for reach agreement (abscissa, TOP3) versus the percent of times that a positive emotion will be experienced by the Iranians

Step 6: Relate the Elements to the Ratings

As of today’s state-of-the-art, the pinnacle of the analysis is the ability to relate the presence / absence of the 20 elements to the response, whether the response be the TOP3 (strong likelihood of that there will be an agreement), or the selection of a positive emotion, and finally the selection of a negative emotion. Mathematically, the selection of positive versus the selection of negative emotions is complements of each other. We will be dealing with both, because in our presentation of data will look only at strong performing elements driving positive emotions, and strong performing emotions driving negative emotions, and in turn NOT presenting data from elements which do not strongly engage of or the other.

The experimental design allows us to create both group models and individual-level models relating the presence/absence of the 20 elements to the response. The original design was set up to allow a simple regression equation to describe the data: Response=k₀ + k₁(A1) + k₂(A2) … k2₀(E4). Recall that each respondent evaluated a unique set of 25 vignettes, comprising a permuted variation of the original design, a variation known to ‘work’, viz., to mathematically identically to the original design.

The first analysis created models relating the presence/absence of the elements to the actual rating of Question 1 on the 9-point scale. Although we will be looking at a transformed variable (TOP3 instead of the 9-point rating), it is instructive to see the degree to which our 85 respondents generate data which is consistent. We measure consistency by estimating the equation, and computing the goodness of fit, the multiple R, the multiple correlation. The multiple R goes from 0.00 (no fit of the variables to the ratings; totally inconsistent results) to +1.00 (perfect fit of the variables to the ratings, totally consistent results which trace the ratings precisely to the presence/absence of the elements).

Figure 3 shows the distribution of the 85 ratings. We can feel confident about the data. Even though most respondents feel that they are ‘guessing,’, that they cannot figure out the ‘correct answer,’ our estimation of consistency suggests that the results are reasonably consistent.

Figure 3: Consistency of the results for the 85 respondents, shown by the Multiple R statistic estimated from the individual-level multiple linear regressions

Step 6: Divide the Respondents by the Pattern of the Coefficients to Create Mind-sets

Our last analysis divides the respondents by the pattern of their coefficients. For each respondent we create a model or equation whose dependent variable is TOP3, previously defined as taking on one of two values. The values depend upon the original rating of Q1, the probability of reaching an agreement. Recall that ratings of Q1 1-6 were coded 0, ratings of 7-9 were coded 100.

The database generated from the individual-level regressions comprises 85 rows, one row corresponding to each respondent. Each row comprises 21 columns, one column for the additive constant, and 20 columns for the 20 coefficients. The objective of clustering is to divide this group of 85 ‘objects,’ viz respondents into a limited number of non-overlapping groups, the clusters or mind-sets, based upon a mathematical criterion. The criterion does not require the researcher to know the ‘meaning’ of the measures, viz., in this case the coefficients, but simply to have each object quantified on each measure. Thus, we have 85 objects (people) on 20 measures (coefficients). We do not consider the additive constant in the process.

The clustering program is a heuristic. There are many different clustering programs. The program used here is k-means [13], with the objective of putting the 85 people into either two groups (analytic pass 1) or three groups (analytic pass 2). The criteria are that the profiles of the 20 averages (one per coefficient A1-E4) should be ‘far away from each other’, and the distance between the objects or people in a cluster should be as small as possible. The criterion for distance is (1-Pearson Correlation Coefficient, R). The Pearson R shows the strength of a linear relation between two variables, taking on the value +1 (viz., Distance=0) when they are perfectly linearly related, and taking on the value -1 (viz., distance=2) when they are perfectly inversely related.

Our criteria for choosing the ‘best’ number of clusters combines a desire for parsimony (fewer clusters are better than more clusters), and interpretability (the clusters must tell a coherent story, and the stories of the clusters must differ from one another).

The two-cluster solution, although parsimonious, seemed too jumbled. There was no clear story. The three-cluster solution seemed a bit better. A four-cluster solution was virtually no different in types of groups than the three-cluster solutions. That is, two of the clusters in the four-cluster solution seemed quite similar. The decision was to work with a three-cluster solution.

In the language of Mind Genomics, the cluster becomes a mind-set, a way of responding to a limited set of related stimuli. The min-sets are constructed from the patterns of the coefficients form the 85 respondents who participated in this study. Over the years, the mind-sets which emerge from these focused, quite small studies, continue to repeat. The repetition comes about because when we abstract the type of individual based upon the pattern of responses, we end up with just a few really quite different groups. The psychologists called the ‘archetypes’, but the archetypes emerging from Mind Genomics are based on small, single-focus studies. Yet, again and again, these mind-sets continue to appear in many different ways. The great anthropologist, Joseph Campbell [14], would call this the ‘hero with a thousand faces.’

Step 7: The Total Panel and the Mind-sets

The Mind Genomics effort naturally brings with it many numbers for the study 21 numbers for each group, or 84 numbers for the combination of total panel and the three mind-sets. The objective of these studies is to find patterns, and not to overwhelm ourselves with numbers which may end up disguising the patterns in the dense undergrowth of numbers. To counteract the death by wall of numbers were show only positive coefficients of 8or higher. These strong performer in a Mind Genomics study. We may be losing some information by this stringent cutoff, but a coefficient of +8 or higher is strongly significant from the regression modeling, with a t statistic approaching 2.0.

Table 3 shows the total panel and the three mind-sets, created for the results from Question 1, on the likelihood of an agreement. The cluster uses the coefficient emerging when TOP3 is the dependent variable. The table shows base size first, then the additive constant, and then the strong performing elements for each mind-set.

Table 3: Performance of the strong performing elements for total panel and three emergent mind-sets. Only the seven elements with coefficients of +8 or higher are shown

The additive constants are 32-38 meaning that without additional information, but just knowing that there are negotiations, about one in three responses to the vignettes are 7-9. We know this because the additive constant tells us the likelihood of a rating of 7-9 in the absence of elements, and is a purely theoretical, computed value. Nonetheless, the additive constant gives us a good sense of basic response. It is remarkable that all three mind-sets agree so well. This is unusual. The agreement means we are dealing with specifics.

When we look at the column for total panel, we find NO strong performing elements that disappointing finding does not mean that we failed in this attempt, although it might mean failure. Our success in the study comes after we deconstruct the total panel into the three groups, based upon patterns of coefficients, not upon magnitude of coefficients. That is, our three mind-sets would have emerged if all of the coefficients were equally reduced by 20 points. In such a case three mind-sets would emerge from the patterns, but NO elements would emerge as being strong.

Before we go into the three mind-sets, which is now quite simple, its worth remarking that we began with 20 elements, the best guesses from people involved. Yet, only seven of the 20 elements emerged as strong, no elements emerged as strong for total, and surprisingly, each strong performing element appeared strong only in one of the three mind-sets.

The min-sets are easy to describe. One simply looks at the strongest element.

Mind-Set 1=Focus on military aspects (prevention) – 29 of the 85 respondents

Mind-Set 2 Focus on economic development – 45 of the 85 respondents

Mind-Set 3 – Focus on effective negotiations and diplomacy – 11 of the 85 respondents.

We move now to the elements which drive strong positive and strong negative responses. The coefficients in Table 4 emerge from six regressions. The six regression comprised three regressions for the selection of a net positive emotion, and three regressions for the selection of a net negative, in both cases two regressions for each mind-set, respectively. The regression model was run without the additive constant, because of the previously conventions in Mind Genomics practice, that emotions and other selections emerging from the nominal scales are estimated without coefficients.

This time we look only the elements which drive a percent selection of 16% or more, for either a positive or a negative emotion. Table 4 shows us that only one mind-set, MS1 (focus on military aspects, prevention) feel that there will a strong positive response. All three mind-sets feel that there will be a strong negative emotion from Iran.

Table 4: Strong positive and negative emotions selected by the respondents from the three mind-sets as they think about the feeling emerging from Iran, as driven by the element. Only coefficients of +16 or higher are shown

Discussion and Implications

When this study was executed in 2016, Mind Genomics was just beginnings its broader application to international relations, having begun in 2012 with studies of the Israeli Palestinian conflict. The realization at that time, confirmed by many subsequent studies in a variety of areas, is the relative paucity of solid information about the mind of the citizen in the world of social issues, the mind of the customer in the world of commerce, the mind of the patient in medicine, the mind of the client in legal and business issues, and so forth. There were dozens of polls, dozens of learned volumes on key issues, the ongoing broadcasting, and increasing ‘natter’ of the media with ‘talking head’ proclaiming the same new, spun one or another way.

A cursory content analysis of the literature, of the media, and so forth brings out facts, histories, opinions, and the voice of the citizen. The voice of the citizen, however, appears to be limited to simple factoids, statements, voting on issues. Furthermore, there seemed to be a desire to compare changes, and by that comparison to get a sense of where things were going. In other words, the focus was on the macro, with little content, and the depth was assumed to emerge by observing the path of the macro trends over time, perhaps with an effort to see how the trend covaries with exogenous factors, like world order world economics, and so forth. And perhaps even the world’s ‘Zeitgeist’ although Zeitgeist might be more the bias of the analyst than the reality of the items. There are examples of iterated efforts, such as China’s policy [15], but these iterations are large-scale, in the manner of iterating products, rather than ideas.

Enter Mind Genomics, here presented as the first experiment on international relations, at a time when Mind Genomics was conceived of as a one-off process, requiring a lot of thinking, a great deal of expertise for choosing the ‘right material’, and the careful efforts which accompany a scientific project. There were 85 respondents, rather than the customary hundreds of respondents, but that is not a problem. the problem here is the fact that the Mind Genomics study at that time was considered as a final effort, a one time ‘deep dive’ into the mind of the citizen. And the results are what they were, pointing to different mind-sets, but with remarkably few elements performing strongly, either in terms of driving agreement or driving emotions.

The methods of Mind Genomics have been proven again and again, in the legal, medical and commercial realms [16-18]. In those realms, the efforts of Mind Genomics have evolved from one-off, large-scale studies with 36 elements down to the current size of 16 elements (four questions and four answers to each question). The notion of the ‘final experiment’ has given way to Mind Genomics as a fast, iterative, learning=based process. Within that world-view, this study would be updated by a series of short studies, each requiring about 60 minutes to set up on publicly available program (www.BimiLeap.com), and then executed with 50-100 respondents automatically with 60-90 minutes, and the entire data set totally analyzed 10 minutes, and returned to the researcher. One might imagine the use of the iteration as a way both to arrive at good ideas, acceptable to both sides, as well as a consensus-building method, wherein both sides cooperate, and thus build good will.

In the evolution of political science, and the evolution of knowledge of people, these early studies by Mind Genomics of political issues show the potential of a systematic exploration of a topic. When that exploration becomes inexpensive, quick, easy to execute on the internet, and most importantly, ITERATIVE, we have the potential a new political science, one based upon data, extending across many countries, many people, over time, and many topics [19-21]. What was one study in 2016 could well generate a wiki of the mind for the topic of dealing with Iran, that ‘wiki’ filled with data, topic-related, and searchable for specific results and for general patterns.

Acknowledgments

The author would like to acknowledge the help of four associates who helped to design the study.

Joseph Bitran of Enterprizer®

Alice Duggan

Tawfik Hamid

Richard Sciacca

References

1. Garrison JA (2003. Foreign policy analysis in 20/20: a symposium. International Studies Review. 5: 155-202.
2. Rametsteiner E, Pülzl H, Alkan-Olsson J, Frederiksen P (2011). Sustainability indicator development—Science or political negotiation? Ecological Indicators. 11: 61-70.
3. Gadarian SK (2010) The politics of threat: How terrorism news shapes foreign policy attitudes. The Journal of Politics. 72: 469-483.
4. Wynn DC and Eckert CM (2017) Perspectives on iteration in design and development. Research in Engineering Design 28: 153-184
5. Druckman JN, Greene DP, Kuklinski JH, Lupia A (2011). Cambridge Handbook of Experimental Political Science. Cambridge University Press.
6. Morton RB, Williams KC (2010) Experimental political science and the study of causality: From nature to the lab. Cambridge University Press.
7. Kinder DR, Palfrey TR (1993) On behalf of an experimental political science. In: Experimental Foundations of Political Science 1-39.
8. Kittel B, Luhan W, Morton R (2012). Experimental political science: Principles and practices. Springer.
9. McDermott R (2002) Experimental methods in political science. Annual Review of Political Science. 5: 31-61.
10. Gere A, Radvanyi D, Moskowitz H (2017). The Mind Genomics Metaphor-From Measuring the Every-Day to Sequencing the Mind. International Journal of Genomics Data Minin. IJGD
11. Radványi D, Gere A, Moskowitz HR (2020) The Mind of Sustainability: A Mind Genomics Cartography. International Journal of R&D Innovation Strategy (IJRDIS). 2: 22-43.
12. Moskowitz HR, Gofman A, Beckley J, Ashman H (2006) Founding a new science: Mind Genomics. Journal of Sensory Studies, 21: 266-307.
13. Likas A, Vlassis N, Verbeek JJ (2003) The global k-means clustering algorithm. Pattern Recognition 36: 451-461.
14. Campbell J (1949). The Hero with a Thousand Faces, New York, Pantheon
15. Leutert W (2021) Innovation through iteration: Policy feedback loops in China’s economic reform. World Development. 138: 105-173.
16. Wren JE and Williams TC (2009) Selling blue elephants to the jury: Potential application of rule developing experimentation in litigation. Baylor Law. 61: 1
17. Gabay G, Moskowitz HR (2019) “Are we there yet?” Mind-Genomics and data-driven personalized health plans. In: The Cross-Disciplinary Perspectives of Management: Challenges and Opportunities. Emerald Publishing Limited.
18. Milutinovic V, Salom J (2016) Mind Genomics: A Guide to Data-Driven Marketing Strategy. Springer.
19. Kaarbo J (2003) Foreign policy analysis in the twenty-first century: back to comparison, forward to identity and ideas. International Studies Review. 5: 156-202.
20. Horiuchi Y, Smith DM, Yamamoto T (2018) Measuring voters’ multidimensional policy preferences with conjoint analysis: Application to Japan’s 2014 election. Political Analysis. 26: 190-209.
21. Rapport A (2017) Cognitive Approaches to Foreign Policy Analysis. In Oxford Research Encyclopedia of Politics.

DOI: 10.31038/MGSPE.2022214

Abstract

We present an approach to database the mind of the citizen on topics, using a set of interlocked studies created through Mind Genomics, in which the elements stay the same, but the topic changes. The database allows the creation of models, equations relating the rating of systematically varied vignette to the presence/absence of 16 elements (statements), as well as estimating the effects of features describing the respondent (gender, age, belief in what method best solves social problems). The data uses experimental design to the create the test stimuli (vignettes), dummy variable regression analysis to show the contributions of the elements and the features describing the respondent, as well as clustering to create new to the world mind-sets, different ways to look at a topic. The paper closes with the suggestion of how to create these databases in either an ad hoc fashion, or preferably in a systematized way, year over year.

Introduction

The academic dealing with social issues, especially problems dealing with the plethora of economics-based problems, is simply enormous, and cannot be straightforwardly summarized. What is missing, however, appears to be an integrated approach to studying these social issues from the mind of the typical citizen. There are various public polls of consumer sentiment), and one-off polls, really studies, about current issues, usually sponsored by an organization involved in public affairs and conducted by a market research company using strict rules of consumer research (e.g., Axios polls conducted by IPSOS, a marketing research conglomerate well known to fits work in the area).

The information about public issues, e.g., studies about what bothers people, appears in documents, summarized, and simplified for public consumption by the media. The rest of the information may go into the innumerable topic-related books published by commercial publishers, or go into reports circulated to politicians and other public servants. Studies such as the Quinnipiac polls [1] are executed year after year, and the database compiled, both for those interested in current problems as well as those interested in the study of changing social scene over time. Some of the thinking can be traced to the discipline known as SSM, soft systems methodology [2].

For studies done an ad hoc basis, there is no reason to create this integrated database of the mind. Such as concept might be interesting, but it does not fit the view of those who want to focus on the moment, and report what is happening in the ‘her and now.’ Those who database their information would be more likely to appreciate an integrated database, contents able to be cross-referenced. Classic books on the mind of the citizen in society might have benefited from the availability of such a database, although that statement is more of a conjecture than a point of fact. Yet, we might consider how earlier efforts might have been enhanced by this type of data, such as the pioneering book by [3]. The current precis of their book, available in 2022, describes the research effort for which the database might be invaluable:

Presents a review and analysis of theoretical and empirical issues in the mechanisms and functions of interpersonal behaviors and their development in social encounters. The relationship of social cognitive structures in the individual to societal structures, developmental, emotional, and economic aspects of interpersonal relations… [4]

The vision of Mind Genomics to provide a database of the citizen’s mind began in the early 2000’s. At that time there was a growing interest in the mind of the citizen about social issues. Twenty years, ago, however, the focus was simply on understanding social issues from the inside of a person’s mind. The senior author participated in studies of response to the voting platform of candidates e.g., the voting platform of Kerry [5]. Inspiration for the work came from the newly emerging interest in computers for data acquisition and the use of experimental designs to create combinations of ideas that the respondent would then evaluate [6]. The effort continues today, suggesting that there is an underlying current of acceptance of conjoint measurement to understand citizen minds [7], along with the recognition that understanding mind-sets can help impact education for students, and create a better world [8].

At the same time, there was an obvious lack of integrate databases about the mind of the citizen about ordinary problems of daily life. The media as well as the journals were and continue to be populated with either continuing stories in the case of media, or well executed but one-off studies by academics using the most powerful social science tools. The senior author executed one large scale study on different situations causing anxiety, using the Mind Genomics tool described below, finding the approach to generate a reasonable integrated database. That database revealed far more than would have been revealed by 15 disconnected studies on the same topics. The success of integrated the 15 parallel studies into a single database called ‘Deal With It’ (for colloquiality) generated the vision that one could use the disciplined approach by the emerging science of Mind Genomics to create a database of the citizen mind, and perhaps make a contribution to the emerging discipline of citizen science [9-11].

The Mind Genomics Approach and Its Use in a Societal Issues Database

Key issues facing the citizen are often approached by researchers using qualitative (depth) interviews, either with single individuals or groups, usually to get a sense of ‘what’s happening in the mind of the citizen.’ Beyond that there may be polls or surveys about the topic. Beyond that is the sociological approach of looking at people in groups, as well as studies of the way a society works. There are no databases to speak of which go into the mind of the citizen, at least no systematized databases updated on a yearly basis, across aspects of the citizen’s life.

Traditional research answers the various questions in an adequate way, but often the data is in a somewhat disorganized format because there is the need to tell a coherent story after digesting and integrating the various sources and types of information. The astute, insightful investigator can pick up the thread of the story, and, with the right data, weave the story together so it morphs into a compelling narrative. When the topic is of sufficient importance, other efforts may be initiated to fill the gaps, and round out the topic.

What is missing from the foregoing is a systematic way to explore the world of the citizen from the inside of the citizen’s mind, doing so with groups of related topics, doing so with people around the world, and on a systematic basis. The data produced by a systematic approach can become invaluable, supplying insights, revealing patterns, increasing our factual knowledge, and promoting the discovery of patterns. If, perchance, the approach is also affordable, then society has the capability to profile itself, worldwide, over time, creasing a database that might well reveal short term and long term patterns.

Mind Genomics as an Affordable, Efficient, Scalable System

The entire Mind Genomics process is templated, from start to finish, including the analysis. Through the templating, the technology forces the researcher to learn a new way of disciplined thinking, a way which ends up being an algorithm for solving a problem, or even for innovation. We begin with the three steps, shown in Figure 1.

Figure 1: The first three templated steps for set up, choose the topic (left panel), select four questions (middle panel), and generate four answers to each question (question 2 on the right panel)

Step 1: Choose the Topic

Figure 1 (left panel) shows that the study topic is ‘Loss of Hope.’ The database includes the results from these five economics-oriented studies, chosen from the full set of 26:

1. College Expense – Education for people in College is too expensive.
2. Economic Gap – Rich people get richer, everyone else falls behind.
3. Loss of hope – People who have no hope that anything they do will help their lives.
4. Poverty – Poverty so that some people don’t have enough to eat.
5. Social Security – People not sure that Social Security will last.

Step 2: Select Four Questions or Dimensions Which ‘Tell a Story’

Our ‘story’ is not a story but rather four sources of solutions (education, social, business, and governments.

Step 3: Create the Element, Four Specifics from Each Type of Solution, or 16 Elements

The set of 26 studies dealt with the solutions of social problems. The solutions were to be appropriate to ‘solving’ the fundamental or underlying issues which led to the problems, not to the actual specific solutions, which would be topic-specific, and would defeat the purpose of an integrated database incorporating many problems. Table 1 shows the four different solutions (education change; social movements; business strategies and government involvement), posted as a question, and for each solution, four specifics.

Table 1: The four types of solutions, and the four specific solutions for each type

Step 4 – Create the Self-profiling Classification Questions to Learn the Respondent’s Gender, Age, and Optional Behavior Provided by the Third Question

The third question in the self-profiling question. The actual topic of the study was given (Loss of hope), and then the four alternatives. The same format applied to all studies. Only the topic of the actual study change.

Preliminary Question: What is the most effective approach to solve the problem of Loss of hope – People who have no hope that anything they do will help their lives

1=Education Changes

2=Social Movement

3=Business Strategies

4=Government Rules

Step 5 – Create the Test Combinations Using Experimental Design

Conventional research works with single ideas (idea screening or promise testing), or with completed ‘concept’ or even advertisements. Typically, one has no way of knowing what ideas will win, or how a concept will score. The astute researcher limits risk by narrowing down the effort, generating a good sense of what answer will be obtained, and choreographing the research to accept or reject the ingoing hypothesis. Thus, in the end, most research is not so much to ‘discover’ and to confirm, presumably because most researcher is subtly based upon a ‘pass/fail’ system.

Mind Genomics is different. Mind Genomics screens ideas, combinations of the elements in Table 1, almost metaphorically in the way an MRI takes pictures of the tissue from different angles, and then combines these pictures at the end to produce an in-depth visual representation. No one picture is correct. Rather, it is the many different combinations which are processed to generate a pattern. The MRI does not ‘test,’ but rather recreates from different angles. Mind Genomics uses the same approach, albeit metaphorically, by testing different combinations of the elements, getting reactions, and from the patterns of the reactions, showing elements which drive solutions to the problems, and elements which did not.

Mind Genomics works by experimental design, systematic combinations of answers to problem. The problem is presented, and then the Mind Genomics program presents different combinations of these solutions. The respondent simply rates the combinations on a scale. The respondent ends up doing the rating by intuition, rather than trying to guess what the right answer is

Step 6 – Create an Orientation Paragraph, Introducing the Respondent to the Topic

For most research it is not necessary to create a long-set up. A short paragraph, even a single sentence will do the job. For this study we put together a more general paragraph, which could work with the different problems. The orientation ended with the specific problem, here shown in italics, but in normal font in the actual study. Figure 2, left panel, show how the orientation is typed into the BimiLeap template. The actual text follow, with the topic of the study in bold.: America is full of unsolved issues. You will see a list of possible actions to solve a problem: Loss of hope – People who have no hope that anything they do will help their lives. Please use the scale below to tell us what will happen when the solutions are applied to deal with this problem: Loss of hope – People who have no hope that anything they do will help their lives

Figure 2: The template acquisition form for the orientation (left panel), the rating scale (middle panel), and an example of one of the vignettes (three elements; right panel)

Step 7 – Create the Rating Scale

The rating scale makes up five labelled points. The enables the researcher to deal with two dimensions, resistance (no/yes), and work (no/yes). Figure 2 (middle panel) shows the templated screen to type in the rating scale.

What is the most effective approach to solve the problem of Loss of hope – People who have no hope that anything they do will help their lives.

1=Will encounter resistance … and… Probably won’t work

2=Will not encounter resistance… but … Probably won’t work

3=Can’t honestly decide

4=Will encounter resistance… but … Probably will work

5=Will not encounter resistance … and… Probably will work

Step 8: Present Each Respondent with 24 Vignettes

Figure 2 (right panel) shows an example of a vignette. Each respondent begin with the self-profiling classification, then read the orientation page, and then rated 24 vignettes on the 5-point scale. The program presented the combination, acquired the rating (5-point scale), and the number of seconds, to the nearest tenth of second, between the time the vignette was presented and the time that the response was made.

Step 9 – Prepare the Data for Statistical Modeling

A key benefit of Mind Genomics is ‘design thinking.’ Rather than getting data and testing hypotheses, Mind Genomics is set up to create a database. The data itself forms rows of data. Each respondent generates 24 rows of results, with the following columns.

a. Columns 1-3: The columns record the topic, the respondent identification code, the age, gender, and answer to the third classification questionnaire. These are the same for the 24 rows of data for that respondent.

b. Columns 4-19: There are 16 elements that could be incorporated into a vignette. Each of the next 16 columns corresponds to an element, with the value ‘1’ inserted when the element appeared in that vignette, else the value ‘0’ inserted when the element was absent from that vignette. The experimental design prescribed which set of 2-4 elements would appear. Thus, any row would show two, three, or four ‘1’s,’ and the rest 0’s.

c. Columns 20-22: The respondent rated the vignette on the 5-point scale. The next three columns show order of test (1-24) of the vignette, the rating assigned by the respondent (1-5), and the number of seconds to the nearest 10^th of second between the appearance of the vignette and the respondent’s rating (0-8 seconds; all times > 8 seconds were truncated to 8).This is called the RT, the response time.

d. Columns 23-27: The datafile was manually reshaped by augmenting it with five new variables (R1-R5), showing which rating was assigned. For example, when the respondent to rate the vignette ‘5’, R5 took on the value 100 (with a vanishingly small random number added), whereas R4, R3 R2 and R1 each took on the value 0 (also with a vanishingly small random number added). The random number is a prophylactic measure for the downstream regression models, yet to come.

e. Columns 28-31: Four new variables were created, allowing the database to feature a single variable emerging from both instances of an answer. For example, the phrase ‘Probably will work’ appears in R4 and R5. Thus, R45 takes on the value ‘100’ (plus the vanishingly small random number) when the rating was either 4 or 5, respectively. R45 takes on the value 0 (Plus the vanishingly small random number) when the rating was 1,2 or 3, respectively. The four newly created variables of this type are:

R45 Probably will work

R25 will not encounter resistance

R12 Probably won’t work

R14 will encounter resistance

Step 10 – Use Clustering to Create New to the World Mind-sets, Individuals Who View the World the Same within this Specific Framework of Problems and Solutions

At the start of the on-line experiment, the respondent completed a small, three-question self-profiling questionnaire, to record gender, age, and select of which approach would be the best way to solve the problem. A hallmark of the Mind Genomics approach is to let pattern of responses to the granular issue generate possibly new-to-the-world groupings of respondents, not based on who they are, but based on how they respond to the granular issues (here solutions to problems). These groups are mind-sets. The respondent may or may not even be aware of belonging to a mind-set, but the response pattern to the 24 vignettes will reveal that membership, after the responses to the vignettes are deconstructed into the part-worth contributions of each of the 16 elements.

Clustering is a well-accepted group of statistical methods which divide objects into non-overlapping groups based upon patterns of features shared by the objects. In our case the pattern of features will be the degree to which each of the 16 elements drives the response. The elements will be coded as 0’s and 1’s, in the database, and the criterion variable will be R45, the rating of ‘probably will work’. The analysis, purely mathematical, will create a profile of 16 numbers (coefficients) for each respondent, each coefficient attached to one of the16 elements. The clustering program [12] will put respondents into two groups, and then three groups, based strictly on mathematical criteria, not judgment. It will be the job of the researcher to select which set of groupings makes sense (two groups vs. three groups). The criteria will be parsimony (the fewer the number of groups or clusters, the better), and interpretability (the groups must make sense).

The novel approach here is that the clustering will be done on the coefficients of all 257 respondents. Thus, the clustering will look at the way the respondents feel a problem can be solved, with the problem varying by experiment, and clearly stated at the start of the experiment. Psychologists call this process priming [13].

The method for creating clusters follows the rules of statistics. The total data-set includes five studies, slightly more than 50 respondents per study. Recall that the 24 vignettes for each respondent were laid out by an experimental design. Even though the combinations were different for each respondent, the mathematical structure was the same. This is called a permuted design [14]. The benefit of the individual level experimental design it that it allow the researcher to use OLS (ordinary least-squares) to relate the presence/absence of the 16 elements either to the rating, to the binary transformed rating, or to response time

When OLS regression is applied to the data, one respondent at a time, using the option of ‘no additive constant,’ the individual level regression appears as:

Binary Response (R45)=k₁(A1)+k₂(A2)+… k₁₆(D4)

The foregoing equation expressed how the 16 different answers shown in Table 1 can be combined to estimate the rating of R45, the newly created variable ‘probably will work.’ Thus, the regression analysis extracts order from the data, allowing patterns to appear. High coefficients suggest that when the element in inserted into the vignette, the rating is likely to be either 4 or 5, both corresponding to probably will work. Low coefficients suggest that when the element is inserted into the vignette, the rating is likely not to be 4 or 5.

The individual level, and for that matter the group models, do not have an additive constant. This revised form allows the direct comparison of the 16 elements. It is vital to be able to compare the elements, side by side, across groups. The additive constant is more correct statistically, but makes the comparisons difficult. Thus, for this r set of analyses we choose not to use the additive constant, even though the model will not fit the data as well.

The five studies were treated identically, considered simply as part of a one big study. To the regression analysis the structure of the inputs and output was identical. At the end of the regression analysis the result was a data matrix forming 16 columns, one for each element, and 257 rows, one for each respondent. The numbers in the data matrix were the coefficients.

A k-means clustering program divided the 257 respondents into two groups and three groups based upon the distances between the respondents. Respondents separated by large distance, to be defined now, were put into different clusters, or mind-sets The ‘distance’ between people was operationally defined as (1-Pearson Correlation), computed on the 16 coefficients of pairs of respondents, not matter whether they were i the same study or different studies. the structure of their data allowed that.

The Pearson correlation shows the strength of a linear relation between two objects (e.g., respondents). The value of R varies from+1 through 0 to -1. R takes on the highest value, 1, when the two objects are perfectly linearly related. R take on the lowest value, -1, when two objects are perfectly inversely related. The distance between two people goes from a low of 0 when the 16 pairs of coefficients generate a Pearson R of+1 (D=0), to a high of 2 when the 16 pairs of coefficients generate a Pearson R of -1 (D=2).

Step 11 – Create Group Models, Incorporating All the Data from a Group

The final modeling consist of creating a general model which uses all 16 elements as predictors, as well as study, gender, age group, belief in what is the best way to solve the problem, and finally mind-set. The modeling thus puts all the variables on the same footing, allowing the researcher to instantly understand the contribution or driving power of each element or respondent feature to three selected dependent variables. These three variables are R45 (probably will work), R3 (can’t make a decision), and RT (response time in seconds).

The general model is expressed as:

Dependent Variable=k₁(A1)+k₂(A2)+… k₁₆(D4)+k17(College Expenses)+k18(Economic Gap)+k19(Loss of Hope)+k20(Poverty)+k21(Social Security)+k22(Female)+k23(Male)+k24(Age 17-29)+k25(Age 30-49)+k26(Age 50-64)+k27(Age 65-80)+k26(Business)+k27(Education)+k28(Government)+k29(Social Movement)+k30(Mind-Set 1)+k31(Mind-Set 2)+k32(Mind-Set 3)

The foregoing equation is easy to estimate, even for large data sets. It is important to keep in mind that the 16 elements (A1-D4) were designed to be statistically independent and thus always appear in the equation. Not so, however, with the other variables. In every regression model, exactly one of the classifications from each group will be missing, and given a value 0 by the regression. That is because the coefficients for the classification features are relative, not absolute. Thus, when looking at males versus females, there is a variable called male, and another variable called female. One of them will have a coefficient showing its relative contribution (viz. female). The other will be set to 0 (viz. male)

How do the Respondents Distribute Across the Different Classification Criteria?

Table 2 shows the distribution of the respondents across the five studies, each column corresponding to a study. The rows correspond to the distinct groups into which a respondent could be put, either from the up-front self-profiling classification, or from the clustering into three mind-sets.

Table 2: Distribution of respondents across the five studies

What is the Pattern of Ratings Assigned by the Respondents in the Separate Groups?

Our first analysis focuses on the pattern of ratings, something that would be the natural first step of any research. We have five ratings (R5, R4, R3, R2, R1, the simple five point scale), as well as four combining scales: Probably will work (R45), Won’t encounter resistance (R25), Probably won’t work (R12), and Will encounter resistance (R14), respectively.

Faced with the data, and absent consideration of the underlying experimental design, the standard analytics would begin by compiling a list of frequencies of ratings by key subgroups (Table 3). After doing that, the typical analysis might look for departures, such as groups in the studies seeming to depart from the general pattern.

Table 3: Percent of responses for each group assigned to original ratings (adds to 100), and then both positives (Probably work, Encounter no resistance), and negative (Probably not work, Encounter resistance)

This surface analysis looks at the pattern for the Total Panel versus the pattern for a specific group, such as the study topic ‘Loss of Hope’, which seems aberrantly positive. The surface analysis provides observations, but little in the way of deep insight.

How the Ratings Change with Repeated Evaluations

Conventional research often asks a limited number of questions perhaps in a randomized order to forestall order bias. The data from these five studies across 50+respondents and 24 vignettes per respondent allow the researcher to get a sense repeating the same task 24 times. The skeptic would say that it is impossible, and that no one can be consistent across 24 vignettes. That skepticism brings up the question of just what happens when the respondent continues to focus on the same topic for 24 vignettes; the vignettes are all different from each other, so we cannot look at the ratings for the same vignette over time. But we can look at the average ratings of vignettes in the same position of time, to see whether we can find a pattern of average vs. time, recognizing of course that the no two vignettes are alike. The issue is whether there is a noticeable position effect.

To understand the issue of stability with repeated exposure to the same problem we looked at the average rating by position. Rather than looking at 24 positions, we reduced the 24 positions to six by creating six sets of positions (e.g., 1-4, 5-8, et.) and then averaging the four ratings for each respondent to generate six new ‘ratings’.

The foregoing analysis allows us to create averages of ratings for each of the six orders, doing so for all respondents in a study, and by each of the five studies. Figure 3 shows the scatterplot of average rating of the 5-point scale versus the new set of six positions. There is a clear order effect, stronger for some (e.g., College Expenses and Economic Gap, less clear for others such as Loss of Hope, Poverty and Social Security). The reason for the differences of average ratings by order of testing is not clear because the five studies were done in the same way.

The change in average rating is important to deal with. It is not usually addressed in conventional research, where the topic is only broached once, and rated. There is nothing to discuss in Figure 3, because we only have a surface measure. However, we deal with Figure 3 as part of a later analysis.

Figure 3: Change in the average rating over the 24 vignettes by each of the five studies

Creating Enhanced Models for the Study Using OLS Regression

In Step 11 above we presented the expression for the enhanced regression model, considering both the elements, as well as the study, gender, age, selected belief about the best solution, and mind-set. The 16 elements are presented as 0’s and 1’s, the remaining factors (study through mind-set) as category variables which can be deconstructed into separate variables.

As noted above, the equation is:

Dependent Variable=k₁(A1)+k₂(A2)+… k₁₆(D4)+k₁₇(College Expenses)+k₁₈(Economic Gap)+k₁₉(Loss of Hope)+k₂₀(Poverty)+k₂₁(Social Security)+k₂₂(Female)+k23(Male)+k24(Age 17-29)+k25(Age 30-49)+k26(Age 50-64)+k27(Age 65-80)+k26(Business)+k27(Education)+k28(Government)+k29(Social Movement)+k30(Mind-Set 1)+k31(Mind-Set 2)+k32(Mind-Set 3)

We run the regression equation by total in the next analysis. In the appendix, we present the parameters of the model by study, by gender, by age, by belief in the best solution, and by mind-sets, as well as by the first and last set of vignettes (to deal with the issue of just what changes as the person evaluates the vignettes)

Table 4 shows the coefficients of models for R45 (Probably work), R3 (Cannot answer), and RT (Response time), respectively. Our first set of analyses focuses only on the coefficients of the 16 elements.

Table 4: Models for total panel relating the presence/absence of the 16 elements, the different self-profiling classifications, and the topic study to R45 (probably solve), R3(cannot decide) and RT (response time in seconds)

The first data column, labelled RT45 corresponds to the coefficients for the rating of ‘probably can solve,’ viz., R4 and R5 combined. Surprisingly, eight of the 16 elements generate coefficients of 12 or higher. There are surprises, such as B2 (create a riot to overthrow the government.) This element might not have appeared had the respondents simply rated what ideas would lead to a possible solution, presumably because of an ‘internal editor’ which tries to be politically correct, and automatically attach a negative response to the element. It is only because the element is embedded in mixture of other elements that the respondent becomes far less capable to be politically correct, simply because it is impossible to be so when confront with what sees a ‘blooming buzzing confusion.’ The analogy here might be the emergence of negative qualities when a respondent interprets a Rorschach blot. Negative ideas are not easily suppressed in the narrative.

C3 Embedding issue within business operations

B1 Create self-help movements

C4 Big spending philanthropic initiatives by businesses

D1 Create laws and legislation to prevent the issue

C2 Rely on business innovation to provide the solution

B4 Promote social media activism

B3 Create a riot to overthrow the government

D4 Incentivize behaviors…tax breaks

The second column, labelled R3, shows the strong elements driving ‘I cannot decide’.. There are no strong performers, viz elements which generate coefficients of+12 or higher. There are two elements which come close. However these are elements which confuse respondents. We would not have really known that, except for the power of this emergent dataset that we are creating.

B1 Create self-help movements

D3 Public outreach through mailers and mass messaging

The third column, labelled RT, shows the reaction time ascribable to each element. The Mind Genomics algorithm measured the time from the appearance of the element to the rating, and used that as a dependent measure. Again, these are estimated times needed to read the element and contribute to the decision. The response time can be s a measure of engagement, of reading the information and thinking about it. the response time itself is neither good nor bad, but simply a measure of behavior. The elements which require time to process are those dealing with actions that the person takes:

B1 Create self-help movements

B3 Create a riot to overthrow the government

B2 Start a protest and improve conditions within the government

Our next analysis looks at the contribution of ‘the group’ of respondents. Following the set of 16 elements (sorted in order) we see four groups. These are the four ways that we divided the respondents, ahead of the study itself. These are age, gender, preferred method of solving problems (all three from the self-profiling classification), and then topic of the study.

One option of each set is always assigned the value 0 because these alternatives in each set are not statistically independent of each other. The respondent must belong to one of the four ages, be one of the two genders, select one of the four preferred methods for solving problems, and take part in one of the five studies. Consequently, incorporating these variables into the regression program meant leaving one of the options out for each group. That option is not estimated in the larger equation, but instead is left out, and in the reporting is automatically assigned the value 0 for its coefficient. It makes no difference which four options are selected. All the coefficients for the options are estimated with respect to the optionally deliberately omitted from the estimation, and automatically assigned the value 0. The four options are Age 65+; Male; Social Movement; Social Security.

The coefficients for each of these four groups can only be compared within the group, not to the other groups, and not to the elements. Nonetheless, we still get a sense of the effects. For example, when it comes to the coefficients for R45, Probably Work, respondents end up generating higher rating when the study topic is “Loss of Hope” with a coefficient of+11. This is independent of all other factors, including elements and ways of classifying the respondent. In contrast ‘economic disparity’ is the least likely to be solved, at least from these data, with a coefficient of -5.

Looking at the differences between the coefficients for R45, we can conclude that:

1. Age 17-29 is the most positive (+7 ), whereas age 50-64 is most negative (-8)
2. There are no big differences across the four groups, based on the way they define themselves in terms of what best solves the problem.
3. There is no difference in gender
4. The is a substantial difference in the topic. The coefficient is highest for Loss of Hope (+11) meaning in general people are optimistic that this can be solved. The coefficient is lowest for economic gap disparity (-5) meaning people are least optimistic that this can be solved.

One again it is important to note that this type of information could not be easily obtained from conventional data sources, but becomes a simple byproduct of the data base, trackable over time, and across cultures and events.

We could do the same analysis for R3, the inability to make a judgment. There are no noteworthy group differences in R3, in the way there were for R45.

Finally, the analysis for RT for the age groups suggest that the response for the youngest respondents (age 17-29) is dramatically faster than the response time for the two older groups (age 50-64, age 65-80). The coefficient for age 17-29 is -0.6. The coefficient for age 50-64 is+1.0 On average, the speed is the response of the older respondents is 1.6 seconds longer for each element.

On the Nature of Micro and Macro Differences among the Three Emergent Mind-sets

The standard analysis by Mind Genomics usually reveals dramatically different, clearly explainable differences across the different mind-sets. Table 5 shows the performance of the elements of these three mind-sets, and the labelled assigned to each. This type of information become increasingly important as the researcher tries to uncover macro pattern among people. It is straightforward to uncover macro patterns when one has commensurate data for all the individuals, as one has here, based on the 16 coefficients.

Table 5: Performance of the 16 elements by the three mind-sets

Traditionally, Mind Genomics stopped after showing the underlying mind-sets and their coefficients. Do we learn any more from knowing the average coefficient in a mind-set, not of the coefficient, but of the different groups?. Are the groups similar, or do the groups differ from each other?

Table 6 shows that there remains heterogeneity across similar groups, even within a mind-set. The variation in coefficients has already been reduced by the clustering, which generated three mind-sets. The remaining variation, that due to the age, gender, preferred solution, and topic, is more of a baseline ‘adjustment’ value, like the intercept in an equation. One might say that the variables of age, gender, preferred solution of the problem and study topic, respectively are simply additive correction factors of different magnitudes.

Table 6: The pattern of coefficients for the total panel, and for the three different mind-sets

The Nature of the Differences between the First and the Last Sets of Four Vignettes

Recall that Figure 3 shows the change in the average rating from the start of the evaluation to the end of the evaluation. Each of the filled circles corresponded to the average of R45 for a set of four vignettes (positions 1-4, 5-8, 9-12, 13-16, 17-20, 21-24). Below figure shows clear that there is an effect. Ordinarily the researcher would report this observation, and move on. The modeling approach allows us to create a full model for each of the six sets of four vignettes. We can create the grand model for the first quartet of vignettes (order 1,2,34), for the last quartet (order 21,22,23, 24) and discover the magnitude of the effect by subtracting the coefficients (Difference=Coefficient for Position 21-24 MINUS coefficient for position 1-4).

Table 7 shows the largest differences for the three dependent variables. There is no need to explain the differences. The intent here is simply to show that these deeper questions can be explore though the database in a way that allows the research to uncover patterns, perhaps unexpected ones, and from that effort generate a working hypothesis.

Table 7: “Large”differences between corresponding coefficients for Positions 21-24 MINUS Position 1-4. The table shows only those major differences, for three dependent variables, R45, R3, and RT

Discussion and Conclusions

The goal of this paper is to demonstrate a new way of thinking about social issues, one which moves out of the realm of hypothesis testing, and more into the realm of databasing, with objectives to record the citizen’s mind in a new way, and as a byproduct lead to hypothesis generation. The novelty of the approach is the facile, rapid, affordable, and scalable creation of databases having to do with different topics in the same domain.

The It! studies of two decades ago began this effort, but at that time it the value of having the precise elements across all topics was ignored. The It! studies attempted to customize the elements, but at the same time maintain a logical structure spanning all the studies. The result was that each study had to be analyzed separately. The emergence of similar mind-sets across foods [9] was encouraging, but the further analytic power emerging from directly comparability was missing. It was a matter of hoping that the same mind-sets would appear, rather than creating the conditions to use all the data to create a common set of groups spanning all the experiments.

The next logical step can be the expansion of the database across more people within a country, countries beyond the United States, and the creation of the database year after year, or even in an ad hoc way during period of social change. The simplicity and affordability of the database approach as demonstrated here allows for the expansion of this databasing approach to other verticals. In that spirit, the other verticals will feature other topics, and so the topics will change to fit the vertical.

The long term view of the process maybe something like creating a collection of perhaps eight such databases, each dealing with a ‘vertical,’ viz different facets in of life, each vertical comprising perhaps seven different but precisely parallel studies (topics in the database), each study run with 100 respondents (rather than 50), and study created to be exactly alike and run the same way in 20 countries. This totals 8 (databases/one per vertical) x 7(studies per database) x 100(respondents per study) x 20(countries) or 1,120 studies, each study run with 100 respondents. Verticals could be situations such as conflicts, negotiations, social problems, empowering citizens, enhancing education, and the like. The cost would be minimal (1,120 studies x 400-$600$ per study as of this writing, Winter, 2022, according to www.BimiLeap.com).

The potential to understand society, its problems, its issues, and opportunities to create a better world through knowledge is the key deliverable from these studies. One might end up with keys which allow groups of people to understand each other, information about communications between hostile parties in conflict situations, along with the ability to update the information, focus that information, or expand the scope as the need arises.

Acknowledgement

Author HRM gratefully acknowledge the conversations about these databases with his colleagues, too many to list, and most to the unwavering encouragement of his wife, Arlene Gandler, who inspired the vision of databasing for world issues during the early, foundational years of Mind Genomics.

References

1. Searles K, Ginn MH, Nickens J (2016) For whom the poll airs: Comparing poll results to television poll coverage. Public Opinion Quarterly 80: 943-963.
2. Holwell S (2000) Soft systems methodology: other voices. Systemic Practice and Action Research 13: 773-797.
3. Foa UG, Foa EB (1974) Societal structures of the mind. Charles C Thomas
4. PsycINFO Database Record 2016 APA (Foa and Foa, 1974)
5. Moskowitz HR, Gofman A (2007) Selling Blue Elephants: How to Make Great Products that People Want Before They Even Know They Want Them. Pearson Education, 2007.
6. Hunt JD, Abraham JE, Patterson DM (1995) Computer generated conjoint analysis surveys for investigating citizen preferences. In Proceedings of the 4th International Conference on Computers in Urban Planning and Urban Management, Melbourne, Australia, 13-25.
7. Stadelmann S, Dermont C (2020) Citizens’ opinions about basic income proposals compared–A conjoint analysis of Finland and Switzerland. Journal of Social Policy 49: 383-403.
8. Lilley K, Barker M, Harris N (2015) Exploring the process of global citizen learning and the student mind-set. Journal of Studies in International Education 19: 225-245.
9. Moskowitz HR, Beckley J (2006) Large scale concept response databases for food and drink using conjoint analysis, segmentation, and databasing. The Handbook of Food Science, Technology, and Engineering, 2. ed. Y H Hui chapter 59 Taylor and Francis
10. Foley M, Beckley J, Ashman H, Moskowitz HR (2009) The mind-set of teens towards food communications revealed by conjoint measurement and multi-food databases. Appetite 52: 554-560. [crossref]
11. Moskowitz HR (2004) Evolving conjoint analysis: From rational features/benefits to an off-the-shelf marketing database. Marketing Research and Modeling: Progress and Prospects (215-230). Springer, Boston, MA.
12. Likas A, Vlassis N, Verbeek JJ (2003) The global k-means clustering algorithm. Pattern Recognition 36: 451-461.
13. Molden DC (2014) Understanding priming effects in social psychology: An overview and integration. Social Cognition 32: 243-249.
14. Gofman A, Moskowitz H (2010) Isomorphic permuted experimental designs and their application in conjoint analysis. Journal of Sensory Studies 25: 27-145.