DOI: 10.31038/PSYJ.2021351

Abstract

The paper presents two studies dealing with attitude towards closing economic gaps, as defined by the poet Percy Bysshe Shelley’s aphorism ‘The rich get richer, and the poor get poorer.’ Both studies worked with sets of 16 different messages, elements that were combined into small vignettes comprising 2-4 elements, the combinations dictated by an underlying experimental design (Mind Genomics). In Study #1 the elements were actual solutions respondents rating the feasibility of the combination of solutions The results from 51 respondents suggest three different mind-sets about what will close the economic gaps ways of evaluating the elements, so-called mind-sets (MS- A1 Business takes lead to create solutions, MS-A2 Can’t think of solutions, MS-A3 Big picture activists). In Study #2 the elements were either specific people, or roles that people fill. The results from101 respondents suggest that there are only two mind-sets about who can close the economic gaps (MS-B1 those who work through power, orders and hierarchy, MS-B2 those who work by convincing others.) The two studies present a complementary pair of approaches to understand the mind of the citizen from the ‘inside out’ when the topic is a societally relevant problem.

Introduction

One need only read the news to get a sense that the economic situation of the middle and the lower classes is becoming increasing dire. Over the past decades, the disparity in income or really in purchasing capabilities have widened, until there is almost a sense of a shrinking middle class, and an increasing group of people who are living from check to check, simply because of the high prices. The awareness of the disparity is decades old [1-3]. The answer is the economy, of course, just like it was in 1992, when William Clinton was elected. The problems of today, 2021, are more severe, however, and the issues far deeper. Economic issues, especially the massive disparity between the rich/ultra-rich and everyone else is codified in the phrase ‘the 1%.’ Furthermore, at the time of this writing, inflation is rearing its ugly head, goods are becoming in short supply because of the ‘supply chain,’ lawless is breaking out across the United States, the country is emerging slowly from the ravages of COVID-19 pandemic, and the nation is divided into the red states and the blue states, the so-called Republican (party) States, and the so-called Democratic (party) states. In other words, the Fraying of America, a term coined by Arthur Kover in work begun a decade ago with Howard Moskowitz, awaiting publication [4].

The traditional answers to the general issue of economic disparity range from laissez-faire (as it is being one today, November 2021, by President Biden, in the United States), to more activist efforts such as government actions [5]. Beyond government action are community/social activities [6], education [7]. All the methods being tried are being stress4r when they move from the almost-hobby nature, serious national application [8].

In the beginning of 2021, Arthur Kover suggested that Mind Genomics be applied to the issue of America’s problems, first to see whether one could create a series of ‘solutions’ and see how they worked with 26 different societal problems, and second to look at the same set of 26 societal problems, but this time look at people (specific individuals or generic titles) to see how they might be perceived as able to solve the problems. This second approach was novel; to identify different individuals, really ‘icons’, combine these icons into small groups, and ask whether the small group would be able to cooperate and arrive at a solution [9]. The ideas for both experiments came in part from conversations about systems thinking and systematic approaches to problems [10].

Mind Genomics – What It Is, Where It Comes From, and How It Works?

The typical approach to social research comprises either observation or studies of large-scale systems, inspired by sociology, or in-depth observation of a small ‘world’ inspired by anthropology. These approaches tend to be observational, looking from the outside in. The observational approaches are complemented by research using surveys, where respondents are instructed to answer many questions about a topic, the questions then tabulated to give a profile of the topic. The observational approaches are also complemented by qualitative research, discussions with the respondent, whether alone (in-depth interview), in pairs (dyads) to allow for interactions, or focus groups with three or more respondents.

The traditional methods are valuable sources of data, but they are not experiments. They are data gathering methods of what exists. They do not show causation, although sometimes causation can be hinted at through so-called causal modeling, an advanced form of statistical regression analysis [11].

Rather than working from the ‘outside-in’ Mind Genomics focuses on the pattern of responses of people to test stimuli, these test stimuli approach for the topic. The researcher in Mind Genomics identifies the topics, identifies relevant ideas in the form of ‘messages’, combines these messages into small, easy to read ‘vignettes’, presents the vignettes to the respondent, obtains the rating of the vignette, and then deconstructs the rating into the contribution of the different messages.

The Mind Genomics approach relies on experiment, on observing the pattern of responses of people to messages dealing with everyday life. The respondent, in turn, is a simple responder, a subject present with this material. The research does not focus on what the respondent says she or he ‘feels’ or ‘thinks’, but simply how the respondent behaves when confronted with the test material.

The foregoing may seem overly subtle and controlled, because it seems so natural to ask questions and to get honest answers. The reality is quite different, however. Most people come with many biases, some to give the ‘right answer’, some to please the interviewer, some to avoid conflict, and so forth. Just as important is the reality that the topics spread across many dimensions, e.g., social, economic, personal, and so forth. The criteria differ from dimension to dimension, but the respondent may not even be aware of these differences.

Mind Genomics was designed to deal with the decision processes of everyday, taking into account the fact that the situations of every day are multi-faceted. Although one might think that a person could adjust the criterion of judgment to be appropriate to the topic, a questionnaire which intersperses different topics becomes hard to deal with, as the criteria demand vary from question to question. A simpler way might be to present the respondent with different stories, doing so rapidly, and request a rating of each story (or combination). One could then attempt to deconstruct the response to the combination, to the vignettes, and estimate the contribution of each component in the vignette, viz., each message or idea. The respondent would not be able to be politically correct. A rapid evaluation of different vignettes would lead to the respondent simply guessing, rather than trying to be correct. Guessing, not trying to give the perfect answer is more typical of everyday behavior.

Its original format, Mind Genomics was set up to look at what drives ‘YES’ for various offers of features, both in products and in services [12,13]. The effort was modeled after the pioneering effort by Wharton professors Paul Green and Yoram Wind [14]. The Mind Genomics process comprised a simple set of features, combined by an experimental design, which prescribed the precise combinations of the features. Each respondent evaluated a unique set of combinations each set a permuted variation of the basic design [15]. It was easy to run these experiments the experiments could be done on a wide variety of topics, and the output was easy to understand, inexpensive to run fast allowing for iteration, and databasing [16,17].

Mind Genomics evolved, from large studies to small, study, easy to set up, and to execute. The focus of the studies evolved from products to social issues. Mind Genomics provided a way to get into the mind of a person, not by the usual observation or questionnaire, but by a simple, hard-to-‘game’ experiment. The respondent would evaluate a set of vignettes (here 24), comprising prescribed combinations of elements, or statements about the topic. The respondent was instructed to read the entire vignette, and the rate the combination on an anchored scale. . Although it sounds difficult to do, and although the respondents attempt to ‘do it right’ and give the ‘correct answer,’ the reality is that only a perfect with perfect memory could even suspect that there was an experimental design controlling the combinations. To most people, the combinations were described as ‘random’, and responded to as such. Most exit interviews revealed that the respondents felt that they just ‘guessed’.

Complementing the elements and the experimental design, was the rating scale. At first the rating sale was a simple 9-point sale, with the assumption that 9 points would allow for more discrimination than a shorter scale of fewer points. Events soon made it clear that the users of the results had no idea what a 6 meant on a 9-point scale. As tractable and sensitive to fine differences the 9-point scale seemed to be, it was hard to understand. Managers would often ask questions which ended up being ‘what does the data mean – please explain). It was to this end that the scale was shorted to five points, and often labelled, usually at both end anchors, ]but now often labelled at each of the five points.

The Worldview of Mind and How It Drives the Design of the Two Experiments

As noted above, traditional research about problems works with the description of a problem, followed either by a discussion about the problems and solutions (qualitative research) or a set of questions dealing with aspects of the topic (survey). The survey questions may be open ended, following the approach of qualitative research, or the questions can be answer on rating scales. The analysis would then present a summary of the discussion or open-ended answers for qualitative research, or a tabulation of answers for the survey.

Mind Genomics follows a different path, combining aspects from three different disciplines, whose aspects it amalgamated into a nascent science with the aim of understanding the mind of the ‘everyday experience,’ and databasing that information.

Psychophysics

The study states the relation between physical stimuli and perceptions. The notions of psychophysics is that one can ‘measure’ private sensory experience The typical psychophysical study has systematically varied stimuli from a simple physical continuum (e.g.., sound pressure levels of noise, even statements of different amounts of money, or statements about different crimes), and instruct the respondents to assign numbers to represent some perceived aspect such as loudness of the noise, perceived ‘happiness’ or utility corresponding to the different amounts of money, or the seriousness of the crimes. In other words, psychophysics focuses on relating the physical level of the stimulus (e.g., stated amount) to a felt intensity of a response (e.g., degree of happiness, degree of the value of money, ability to buy things, etc.) There is inherent magnitude in both the independent variable, and in the response rating itself.

Experimental Design (Statistics)

Create test stimuli in such a way as to allow the research to gain information about the stimuli by comparing ratings to each other, and by creating a mathematical equation. Mind Genomics works on the response to defined mixtures of stimuli, as we will see below. The experimental design prescribes the specific experimental designs needed for Mind Genomics to create equations at the level of the individual respondent.

Consumer Research

Use consumer research to run surveys (actually experiments which look like surveys) with the results already in the form of a scalable, cross-referenceable database, the foundation of a new science, the mind of the everyday.

Two Studies -What Drives Three Strong Responses – Absolutely Yes, Absolutely No, Don’t Know?

Just to reiterate, our focus now is on the emerging issue of inequality, as summarized by ‘the rich get richer, the poor …’ the topics are HOW can that issue of economic inequality be solved, and WHO can solve it. We will look at the data from the point of what respondent feel will work, won’t work, and can’t even approach to be appropriate in the situation

Study 1: How Solutions Drive Perceived Feasibility

Our first study concerns a series of solutions of different types, taken in part from the summarizations of Baumann & Majeed (2020). Table 1 shows the different solutions, as well as the question ‘driving’ the solution. The important thing to keep in mind is that the solutions are generic. The solutions can work with anything.

Table 1: The four types of solutions, and the four specifics in each type of solution.

We begin with the self-profiling question, and the rating question and answers. The rating question introduces the problem. It is short, to the point. The objective is to have the 16 specifics provide the information that will be rated.

a. A set of self-profiling questions, including age, gender, and the third question below

What is the most effective approach to solve the problem of Economic gap – Rich people get richer, everyone else falls behind.

1=Education Changes 2=Social Movements 3=Business Strategies 4=Government Rules

b. Orientation to the topic and the 5-point anchored rating scale

What is the most effective approach to solve the problem of Economic gap – Rich people get richer, everyone else falls behind.

RATE1=Will encounter resistance … and… Probably won’t work

RATE2=Will not encounter resistance… but … Probably won’t work

RATE3=Can’t honestly decide

RATE4=Will encounter resistance… but … Probably will work

RATE5=Will not encounter resistance … and… Probably will work

The set-up for these Mind Genomics studies is templated, enabling the researcher to follow a simple series of steps to provide the necessary information. Figure 1 shows the set-up template. Figure 2 shows two screens in the set-up template, screens that show the self-profiling classification, and an example of a vignette.

Figure 1: The set-up template for the first Mind Genomics study on the solutions to problems.

Figure 2: Example of the set-up screen for the third self-classification (left) and an example of the set-up page showing a test vignette (right).

This first study was run with 50 respondents. Each respondent rated the set of 24, unique vignettes created by mixing the 16 elements into combinations comprising 2-4 elements. Each question contributed at most one element to a vignette, but for four vignettes contributed no elements to the vignette. Every element appeared five times in 24 vignettes and was absent 19 times. The experimental design was set up to allow for an individual-level regression relating the presence/absence of the 16 elements to the responses. For this project, the preliminary analysis created four dependent variables:

1. RATE1=Will encounter resistance … and… Probably won’t work. When the rating was ‘1’ on the 5-point scale RATE1 took on the value 100. When the rating was not ‘1’ on the 5-point scale, RATE1 took on the value 0. RATE1 corresponds to a belief that the solution will not help solve economic inequity, the problem posed in the introduction.
2. RATE5=Will not encounter resistance … and… Probably will work. When the rating was 5 on the 5-point scale RATE5 took on the value 100. When the rating was not 5, RATE5 took on the value 0. RATE5 corresponds to the belief that the solution will help solve the problem of economic inequality.
3. RATE3=Can’t honestly decide. When the rating was 3 on the 5-point scale RATE3 took on the value 100. When the rating was not 3 on the 5-point scale, RATE3 took on the value 0.
4. RT – The measured response time from the time the vignette was presented to the time the rating was assigned

To ensure that there would be at least minimal variation in the dependent variable, viz., the newly created binary scales (RATE1, RATE3, RATE5), a vanishingly small random number (<10^-5) was added to each newly created binary variable for every case. The added variability does not affect the regression but ensures that there is the requisite variability so that the regression does not crash.

The regression model was run without an additive constant, to allow direct comparisons of the coefficients across groups. The regression equation, estimated using OLS (ordinary least-squares) methods, is expressed as: Dependent Variable=k₁(A1) + k₂(A2) … k1₆(D4)

The self-profiling classification allows us to assign each respondent to gender, to age group, and to the way that problems of this type might be solved. The definition of the subgroups generates 10 different groups. We show only those elements with coefficient of 11 or higher, coefficients that would be clearly significant. The elements and the strong performing coefficients appear in Table 2. The elements are sorted by the sum of the strong performing coefficients. Thus, the strongest performing element in this reduced set of elements is D2 (Provide government funding). The weakest, but still strong performing elements are C4, C1, and A2, all with one strong group, and coefficients of 11.

Table 2: Strong performing elements by element and key self-defined subgroup for RATE5 vs the 16 elements. Only coefficients of 11 or higher are shown.

It is important to note that there is no clear pattern, either by element or by self-classification. Furthermore, half the elements simply fail to drive a perceived ability to drive a strong solution (viz., RATE5). We might have more elements appearing if we create the model based on a combination of RATE4 and RATE5, both saying that the solution will probably be successful, but RATE4 saying it will encounter resistance, and RATE5 saying it will not encounter resistance.

The importance of this first result is that there are no simple solutions. Either the solutions are weak, or the groups are so variable in what the people of the group believe to work that the power of the idea of the solution is attenuated.

An ongoing theme of Mind Genomics is that there exists in everyday experience a different group of ideas which constitutes ‘mind-sets.’ A mind-set comprises a set of ideas which ‘travel together’ and which can be interpreted. That is, the mind-set makes intuitive sense, and tells a meaningful story.

The mind-set emerges from the pattern of responses to the different elements. Once we see which elements emerge together as strong, we may find that the pattern almost ‘jumps out at us.’ When we work with a set of elements for a specific topic, usually about 2-3 mind-sets emerge. There could be more, but the ideal is to work with mind-sets that are interpretable (tell a story), and which are relatively few in number for the topic. Fewer mind-sets are better than many, even though as we extract more and more mind-sets from the same data the story gets clearer, because we focus on narrower and narrower ranges of ideas.

The mind-sets emerge from a simple mathematical analysis, and not from preconceived notions of the researcher. The mind-sets emerge sing the mathematical methods called clustering which puts into separate groups the various objects (viz. respondents) based upon some quantitative criterion. For example, one may put together individuals who show very similar patterns of coefficients. The similarity in the pattern of coefficients from one person to another suggests that these people think in similar fashion.

Our data provides the ideal set up for k-means clustering [18]. Each respondent evaluated 24 vignettes arranged according to an experimental design. We can create an individual level equation for each respondent. The equation will be written as it was before: Dependent Variable=k₁(A1) + k₂(A2) … k₁₆(D4)

The clustering program works with the 51 sets of 16 coefficients, one set for each of the 51 respondents, one coefficient for each of the 16 elements. The clustering program first computes the ‘distance’ between each pair of respondents, defined as (1-Pearson R). The Pearson R is a measure of the strength of a linear relation. If two respondents show a perfect correlated set of 16 coefficients, the correlation is +1 their distance is 0 . The distance is 1-1=0.

The clustering was done using RATE5 as the dependent variable. The first step in the clustering was to run the 50 regression models, each without the additive constant, as noted above. The second step was to apply the k-means clustering, and extract three mind-sets. Two mind-sets produced a more parsimonious set, but the stories were not clear, viz., interpretability was not sufficient.

Finally, the k-means clustering program assigned each of the 51 respondents to one of the three clusters or mind-sets, based upon a measure of cohesiveness of the cluster. After each respondent was assigned to one of the three non-overlapping clusters, it was a simple matter to run four equations for each cluster, using only those respondents assigned to the cluster. The four equations were RATE1, RATE5 (Table 3), and RATE3 and Response time (Table 4).

Table 3 presents the results for Total panel and for the three mind-sets. Based upon the strong performing elements, we can call the mind-sets as following:

Table 3: Strong performing elements for total and for each mind-set, based upon the model for RATE1 (encounter, resistance and won’t work), and based upon the model for RATE5 (encounter no resistance, will work).

Mind-Set A1=Based on Rate 5: Business Takes the Lead

The business has to be open to new ideas, receptive to solving the problem as part of the business flow and be open to innovation. Avoid activism. The only solution which is problematic is listening to the voice of young people. There are those in Mind-Set3 who think it will work, and those who think it won’t work, based upon the strong performance of element A2 (Promote the voice of young students) for both RATE1 and RATE5.

Mind-Set A2– Can’t Think of Anything

Mind-Set 4 is interesting simply because nothing seems to have a chance of working. On the other hand, when it comes to this mind-set thinking about what absolutely won’t work, viz., how they perform on RATE1 (resistance/won’t work) they ae negative to the ideas which seen perfectly reasonable to others.

Mind-Set A3 – Big Picture Activists

They want major change, which can be through business practice, major philanthropic donations from business, or even through riots. They don’t believe in slow activist movements.

Table 4 presents the strong performing elements for RATE3 (cannot decide), and for response time (RT). The models were once again the standard linear models, without an additive constant. The dependent variable for RATE3 was the binary transformed value for ratings that were either 3 (transformed to 100), or not 3 (transformed to 0). The dependent variable for response time, the number of seconds did not need any added very small random number because there was clear variation among the different response times.

Table 4: Strong performing elements for total and the three mind-set segments, for RATE3 (can’t decide) and RT (response time).

In contrast to the interpretations for RATE1 (NO) or RATE5 (YES), the elements driving RATE3 do not tell a coherent story. There are three strong performing elements for Total Panel, and four strong performing elements for each mind-set. In no mind-set do we see a story.

The elements driving long response times are not related to the mind-set itself, but tend to of two types, either starting a riot or protest, or create a self-help movement Both of these seem emotionally evocative, suggesting that the response time measure is not a measurement of good/bad, but rather of the startle-value of the idea, coupled with the ability of the idea to paint a suggestive word picture.

Table 5, showing the distribution of respondents in the three emergent mind-sets reveals no simple pattern. It often comes as a surprise that when we penetrate a topic, people faced with the same topic find radically different points of view when they evaluate specifics. These different points of view emerging from a ‘micro-topic’ often fail to emerge when the topics so large as to avoid specifics. Thus the 17 people who say that problems can be solved by business strategies do not fall into Mind-Set A1 (Business Takes the Lead). Only 5 of 17 respondents are assigned to the correct mind-set. Similarly, of the nine respondents who way that the problem can be answer by social movements, only two are assigned to Mind-Set A3 (Big Picture Activities).

Table 5: Distribution of the respondents across the three mind-sets for study 1 (Solutions).

Study 2 – How People as Icons or Emblems Drive Perceived Feasibility of Solutions

The second study moved from actual solutions, albeit general ones, to individuals who represent prospective problem solvers. The underlying thinking was that although people may not ‘know’ what solution to a problem ‘feels right’, they may have a feeling of WHO can solve their problem. Some of the thinking behind Study 2 comes from the notion that there might be ‘archetypes’ which emerge, based upon those who are perceived to be able to solve the problem [19,20].

Following the same Mind Genomics approach of a topic, four questions, and four answers to the questions, we did the same type of study. We begin with the self-profiling classification, the introduction to the topic, and the five-point anchored rating scale:

a. A set of self-profiling questions, including age, gender, and the third question below

Which political description fits YOU best?

 1=Old time Republican 2=Trump Republican 3=Democrat 4=None

b. The topic but the rating scale and the answers changed to fit the issue of solution providers, rather than solutions themselves:

What will happen when these people work together to solve this problem: Economic Gap: Rich people get richer, everyone else falls behind

RATE1=Cannot cooperate … and … No real solution will emerge

RATE2=Cannot cooperate … but … Real solution will emerge

RATE3=Honestly cannot tell

RATE4=Can cooperate … but … No real solution will emerge

RATE5=Can cooperate … and … Real solution will emerge

This time, however, we replace the questions and answers with those in Table 6.

The analysis for Study 2 on People as icons or emblems was done in precisely the same fashion as was done with Study 1 on problem solutions. Thus, the two studies can be compared, at least in their general morphologies, regarding the number and magnitude of coefficients emerging as strong drivers, the nature of the mind-sets.

Table 6: The four types of emblematic problem solvers, and four specific people or groups for each type.

In contrast to the relatively sparse number of very strong performing elements for actual, albeit general solutions (Table 2), putting people in as problem solvers, and building models for RATE5 versus elements (no additive constant) shows many more strong elements (Table 7) The stronger performers are the ‘usual suspects. What is remarkable is that at the time of this study, when President Biden was doing reasonably well at the polls, and there were no looming disasters, President Biden was seen as a problem solver only by those who called themselves Democrats. Surprisingly, so did former President Trump, and only among Democrats. He scored poorly everywhere else.

Table 7: Strong performing elements by element and key self-defined subgroup. Only coefficients of 11 or higher are shown.

The clustering of respondents on the basis of the pattern of coefficients for RATE5 (RATE5=Can cooperate … and … Real solution will emerge) produced some strong surprises. First, no elements scored strongly on RATE1 (Cannot cooperate … and … No real solution will emerge) nor on RATE3 (honestly cannot tell). The failure to score strongly on these two response points suggests that people ‘know’ who they believe and trust, but their critical thinking may stop there. The data suggest an asymmetry in thinking between positives (people who are respected and probably liked), and negatives (people who are disrespected and probably disliked). Furthermore, only two clusters or mind-sets were needed. A three-cluster solution revealed two quite similar mind-sets, differing only in one of two elements.

Table 8 shows the strong performing elements for RATE5, and for response time, by total panel, and by the two mind-sets emerging from study 2. The important thing to notice is the set of high coefficients for RATE5 meaning that the respondents feel strongly about their answers, AND the short response times. There is very little ‘shock value’ of people, except Mother Theresa, who would not be typically thought of as a problem solver.

Table 8: The strong performing elements for RATE5 and for Response Time (RT) for study 2, with the elements being people and the rating scale being ability to cooperate and solve the problem. RATE1 and RATE3 generated virtually no strong performing elements.

The group membership is more interesting for this second experiment (Table 9). The self-proclaimed Democrats appear equally in the two mind-sets, Mind Set B1 (working through orders) and Mind-Set B2 (working by convincing.) The self-proclaimed Republicans (both regular and Trump Republicans) appear far more frequently in Mind-Set B1 (working through orders).

Table 9: Distribution of the respondents across the three mind-sets for study 1 (Icons, Emblems).

Discussion and Conclusion

The original motivation for these studies was an interest how we think about solving social problems. The approaches to problem solving generally talk about strategies, about success stories. The strategies and success stories are so individuated that they either lack flavor entirely because they are generic (viz., strategies, such as points about solving issues), or they are so specific as to leave one wondering what to do. Furthermore, a glance at the literature about problem solving for social solutions did not bring up the role of the individual thinker, but rather the role of the situation, and the role of the expert.

The objective here was to approach the topi of problem solving of social issues from two angles, first specifics and then individuals. The specifics make sense; they are types of actions that can be taken to solve a problem. Which ones would work in the case of certain social issues, of which the economic inequality described here is one of them?

In a previous paper author Kover and Moskowitz introduced the idea of Projective Iconics, doing so within the realm of Mind Genomics [9]. The idea was to move beyond the rational to the emotion in the assessments of problems and solutions. The traditional methods for dealing with problems appeared to be all rational, left brain oriented with the utility of solving the problem (soft benefits), or harder, more economically measurable benefits. The test stimuli were always problems, the solutions were generally tangible, except for some feelings, and the evaluation was rational.

A different way had to be developed, one which would encompass something deeper than rational solutions, the act. We were taken with the adage than investors often say that they bet on the jockey, not on the horse. That is, it is the person leading the solution might be just as important as the solution itself. In that way, was born the version of Mind Genomics used here, labelled Projective Iconics. Rather than having solutions, we have combinations of problem solvers would that approach work?

Study 1 using the standard solutions suggests that people can evaluate good versus poor solutions. That is, across the set of respondents there are s number of solutions which clearly are not perceived to work, viz., RATE1, and another set of solutions which may or may not work, but people cannot decide. And, of course, quite a number of solutions which ae believed to work, especially when the total panel is broken out in subgroups. There are also a great number of solutions which are deemed not to work.

Study 2 upends the pattern, by suggesting that when we move from concrete solutions to icons on whom people can project their feelings, we are able to identify people or groups who can solve the problem but find it hard to assign people to groups who cannot solve the problem.

The conclusion here is that there is a profound difference in the way we think about the solution to problems, with far more concreteness when we talk about the actual solution, and far emotion when we talk about the problem solvers themselves.

References

1. Durham Y, Hirshleifer J, Smith VL (1998) Do the rich get richer and the poor poorer? Experimental tests of a model of power. The American Economic Review 88: 970-983.
2. Kotler PT, Lee NR (2009) Up and Out of Poverty: The Social Marketing Solution. Pearson Prentice Hall.
3. Baumann S, Majeed H (2020) Framing economic inequality in the news in Canada and the United States. Palgrave Communications 6: 1-11.
4. Kover A, Bejarano LER, Moskowitz H (2021) Social and business problems through the lens of projective iconics: Introducing a new systematics to understand and quantify perceptions of social issues. Psychology Journal Research Open 3: 1-9.
5. Wright G (2018) The political implications of American concerns about economic inequality. Political Behavior 40: 321-343.
6. Babu S, Pinstrup-Andersen P (2007) Social innovation and entrepreneurship: Developing capacity to reduce poverty and hunger. Twenty twenty (2020) focus brief on the world’s poor and hungry people/International Food Policy Research Institute (IFPRI).
7. Knapp MS, Turnbull BJ, Shields PM (1990) New directions for educating the children of poverty. Educational Leadership 48: 4-8.
8. Thorbecke E, Charumilind C (2002) Economic inequality and its socioeconomic impact. World development 30: 1477-1495.
9. Kover A, Moskowitz HR, Pagajorgji P (2021) The Fraying of America, in Review.
10. Stroh DP (2015) Systems thinking for social change: A Practical Guide to Solving Complex Problems, Avoiding Unintended Consequences, and Achieving Lasting Results. Chelsea Green Publishing.
11. Cliff N (1983) Some cautions concerning the application of causal modeling methods. Multivariate Behavioral Research 18: 115-126.
12. 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.
13. 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.
14. Green PE, Krieger AM, Wind Y (2001) Thirty years of conjoint analysis: Reflections and prospects. Interfaces 31: S56-S73.
15. Gofman A, Moskowitz H (2010) Isomorphic permuted experimental designs and their application in conjoint analysis. Journal of Sensory Studies 25: 127-145.
16. Milutinovic V, Salom J (2016) Mind Genomics: A Guide to Data-Driven Marketing Strategy. Springer.
17. Salom J (2021) Mind Genomics with big data for digital marketing on the internet. Handbook of Research on Methodologies and Applications of Supercomputing, 282-289. IGI Global.
18. Likas A, Vlassis N, Verbeek JJ (2003) The global k-means clustering algorithm. Pattern recognition 36: 451-461.
19. Harootunian JA, Quinn RJ (2008) Identifying and describing tutor archetypes: The pragmatist, the architect, and the surveyor. The Clearing House: A Journal of Educational Strategies 82: 15-20.
20. Bradshaw TK (2007) Theories of poverty and anti-poverty programs in community development. Community Development 38: 7-25.

DOI: 10.31038/PSYJ.2021344

Abstract

460 New York City based respondents participated in a Mind Genomics study to identify the messages which promote registering to vote. Each respondent evaluated 48 different vignettes, combinations of messages, created from a base of 36 messages. The vignettes for each respondent were unique, prescribed by an underlying permuted experimental design. The Mind Genomics design enables discoveries of mind-sets in the population (segmentation), and synergies among pairs of elements (scenario analysis). Data from the total panel revealed no strong performing elements driving intent to register to vote. Data emerging from three mind-sets revealed strong-performing elements for each mind-set. Scenario analysis, an analytic strategy which reveals synergies between elements. revealed the existence of far stronger messaging which could emerge by combining specific pairs of elements. The data and straightforward analytic process suggest that systematic exploration of issues in public policy can quickly create a repository of archival knowledge for the science of policy, as well as direct recommendations of actions to be taken. The speed of the approach furthermore allows the method to be even more powerful, as the iterations retain the strong performing elements, eliminate the weak performing elements, and replenish with new, hitherto untested messages.

Introduction

The case history we present grew out of a student competition to create more effective messaging regarding voting, specifically getting people to say that they intend to register to vote. Pollsters and other political professionals often have a sense of what is important to the voter, in terms of substantive topics, such as the economy, the looming issues with health care, and so forth. There is a plethora of possible messages from which to choose, with the problem being which specific topical message for which candidate. However, the important question on the table is, in the first place, how to get people to register to vote. For the more diffuse issue of ‘voting itself’, like the issue of ‘health maintenance itself,’ we deal with a more difficult problem. There is no pressing need, no issue to solve, no ‘pain points’ to address. Indeed, it is the exact opposite. There is an indifference to the democratic process, one that need not be explained nor studied, and whose origins are not relevant unless those origins can be marshalled to help identify an actionable solution. In other words, the general issue of ‘registering to vote’ is more difficult to understand [1]. There is no pressing fear on the part of the population. Rather, there is a creeping indifference, something which alarms a few people, but is irrelevant to many others until the consequences of such indifference destabilize the country or state or city, and the citizen’s pain begins [2]. The year-on-year decline in those who do not vote has been noted by a variety of sources [3,4]. The issues holding people back range from economics [5] to social alienation (Engler & Weisstanner, 2021), to inconvenience and forgetfulness in the wake of other commitments [6], all occurring in the advanced economies where there is freedom. The situation in the United States is interesting because at the same time that voting is deemed to an important civic duty, registering for voting entails passively registering to serve on a jury, an opportunity to do one’s duty, but not a popular one [7]. In other countries the change in voting over years emerges as a mixed set of patterns. There are a variety of countries where the voting is declining, and others where the voting is increasing. And then there are the dictatorship, where it is mandatory to vote, and of course to agree with the slate offered by the party. The increasing apathy of voters over the years has not gone unnoticed. In 2016, coauthor Markovitz, teaching a marketing class, used Mind Genomics to identify the messages that one could use, and the venues for those messages, both with the objective to increase voting. The idea way to find the different media used for each respondent, identify the strongest messages for the respondent (or group of respondents, called mind-sets), and then recommend the messages for each group, and the place to pick the messages. This dual strategy, optimize the message, and identify the right media, was done by the marketing class, and the results recommended [8].

The reanalysis presented looks more deeply at the nature of respondents, and the possible existence of synergies between elements.

1. Stability of judgment across the array of evaluations: Are there respondents who change their minds during the course of the Mind Genomics evaluation? If so, how much do they change their mind? Are there those who increase their interest in voting with repeat evaluations, and if so, what messages appeal to them? And are there those whose interested decreases with repeat evaluation, and if so what messages appeal to them, but also what messages turn them off.
2. Mind-Sets: Can we discover intrinsically different, structurally meaningful mind-sets of voters in the population of respondent? One of the foundations of Mind Genomics is its approach to uncover new-to-the-world mind-sets, different ways of making decisions about the same facts. Rather than differentiating voters on the basis of WHO they are, we focus on the way they weight information to make their decision, either YES – Register to vote, or NO – Do not register to vote, respectively.
3. Interactions of messages: Can we identify synergisms between elements, so that with deep knowledge we can find those ‘nuggets’ of messages with the ability to break through the indifference?

Mind Genomics as a New Way to Accelerate Impossible-to-Game Measurement

Mind Genomics began in the world of experimental design, with the pioneering work of mathematical psychologists and statisticians R. Duncan Luce and John Tukey [9]. The objective was to create a new form of fundamental measurement. Their treatment is mathematical and filled with axioms. What is important to note is the word ‘conjoint’. The goal was to measure individual quantities by the behavior of mixtures of these quantities. In other words, to create variables, mix them, measure the reaction to the mixture, and then estimate the part-worth contribution of each element. Although conjoint measurement may seem a little too theoretical, the reality is that within a few years, consumer researchers at Wharton and other places (Green, Wind, etc.) would apply a version of Conjoint Measurement to features of services and products [10]. The engine of analysis would move from theoretical issues to practical applications in the world of marketing to focus on services and products. The early versions of conjoint measurement involved difficult-to-execute studies, where the respondent would compare two ‘bundles’ of ideas or offers and select one. The study required that the researcher know what to test ahead of time and know what to combine to get the best results, such knowledge coming from both experience with the topic. As a result, the early conjoint methods were cumbersome, requiring a significant knowledge of the topic with the study providing a little extra information.

There was a clear need to create a knowledge-development system, which could start at ‘ground zero’, with no knowledge, be easy to implement, be robust statistically, and be iterative. Thus was born the Mind Genomics approach, used here [11].

The foundations were simplified:

1. Conceptualize the problem as a mix-and-match, rate, deconstruct, evaluate, discard, replace, move on. The steps were in part modeled after the classic books Plans and Structure of Behavior, by Miller, Galanter, and Pribram [12]. Their abbreviation for the process was TOTE, Test, Operate, Test, Exit
2. The system should work with no starting knowledge and should NOT require much in the way of thinking by the researcher. All of the ‘hard’ work would be done in the template, the hard work being the up-front thinking of some ideas. The rest is mechanical [13,14].
3. The process would become a discovery tool, open to inexpensive, rapid iteration, so that one would build up a great of knowledge at every iteration. The iterations should take no more than a few hours
4. The data to be shown were collected by students, with little experience in the topic of voting or public polling, but who were able to create a powerful knowledge base in the matter of days.

Explicating the Mind Genomics Methods through a Case History

Step 1 – Create the Raw Materials

Mind Genomics works by presenting the respondent with specific combinations of messages, viz., so-called ‘elements.’ Step 1 creates these elements. The process begins by the selection of a topic (convincing people to register to vote). The process then proceeds by creating a set of questions which ‘tell a story’, and in turn a set of ‘answers’ or ‘elements’ for each question. In this study, we used a version of Mind Genomics set up for six questions, each question having six answers. The questions are not really questions, per se, but rather what one might call ‘topic sentences’ in writing and rhetoric. They move the account along. Ideally, they should fall into a logical order. Table 1 presents these six questions, and the six answers for each question.

In the Mind Genomics study, the questions are not shown to the respondents. As a result, the answers or elements must ‘stand on their own.’ During the evaluation, the respondent will find it easy to ‘graze’ through the different answers presented in the test combinations and make a judgment. The structure of the question, its clarity, is far less important than the structure of the answer, the element. Ideally, there should be no subordinate clauses, as few connectives as possible, and very little if-then thinking. In other words, simple declarative statements are best.

Table 1: The raw material for the Mind Genomics study, comprising six questions, and six answers (elements) to each question.

Step 2: Create Vignettes, the Stimuli to be Evaluated

One of the foundations of Mind Genomics is that the respondents should be required to evaluate vignettes, combinations of elements created according to an underlying experiment design [15]. The experimental design is a set of recipes, in this case 48 different recipes or vignettes for each respondent. Of these, 36 vignettes comprise four elements, with no question contributing more than one element. The remaining 12 vignettes comprise three vignettes, again with no question contribute more than one element. One of the differences between Mind Genomics and conventional research is the way that the underlying patterns are uncovered, viz., in terms of dealing with variability or ‘noise.’ The standard scientific approach is to suppress the noise by doing one element at a time so the respondent can focus on the element, or by testing the same vignette with many respondents, so that the variability can be averaged out. In both cases the research must perforce be limited to the 48 vignettes chosen, so it is good research practice to know a lot about the topic, so that the choice of the elements and the creation of the vignettes is ‘close to as good as it can be.’ The strategy seems adequate, unless of course one does not know much about the answer and does not even know where to start. In such a case, there is a reluctance to spend a lot of money on solid research. The Mind Genomics approach is quite different. The ingoing assumption is that the research should cover as wide a space of alternative combinations as possible, rather than be focused on a small, and presumably promising area. This strategy of covering a wide swath of the ‘design space,’ the world of possible combinations, is accomplished by a permutation strategy [15]. The basic mathematical structure of the experimental design is maintained, but the actual combinations differ. The happy consequence is that each respondent evaluates a different portion of the design space. That is, each respondent evaluates all elements, each element five times in different combinations, but it is the combinations which vary. Only at the end, when the ratings are deconstructed into the contribution of the individual elements do we get a consensus value for each element, the coefficient which is the key to the analysis, the ‘secret sauce’ in the parlance of business.

It is worth noting here that the systematic permutation and the potential for iteration means that the researcher really does not have to know, or even ‘guess’ what are the correct elements, and what are the combinations which will be most productive to reveal the answers to the problems. Rather, the underlying computer program for Mind Genomics will create the combinations for a respondent, present these combinations to the respondent, get the ratings, and store the data. The process is fast, the creation of the different sets of combinations is automatic, built into the system, allowing the entire process, from start to finish, from creating the elements to evaluating the analyzed results, to occur in a matter of hours, or a day at most.

Step 3 – Create the Additional Material for the Study

This material included the orientation page, comprising a short introduction to the topic, as well as a 9-point rating scale. As we see below, the orientation creates very little expectation on the part of the respondent about what the correct answer will be. It will be the task of the elements (Table 1), combined into vignettes (Step 2) which will drive the response. The orientation is simply a way to introduce the respondent to the task. The orientation for this study is simply the question ‘How likely are you to register to vote based on the information above?’. The respondent’s task was simple; read the vignette and rate the vignette. There was not deep information about the need for voting, etc. That information would be provided by the elements. The actual ‘look’ of the question appears below. Note that the vignette occupied the top of the screen, and the rating scale occupied a small section of the bottom of the screen:

In addition to the orientation and rating, the respondents were instructed to fill out a short questionnaire on who they were, and gave the researcher the permission to contact them, and to append additional third-party data of a non-confidential source. That additional information augmented the information obtained in the Mind Genomics experiments, allowing the researcher to understand the preferences and way of thinking of individuals based upon WHO they are, and WHAT they do. Such information is the typical type of information served up in studies. By itself the information informs but does not guide directly. Coupled with understand the important elements to drive a person to say she or he will register to vote, the information becomes far more valuable. One can then prescribe, rather than just describe.

Step 4 – Execute the Experiment

The respondents were from New York City participants who were members of a nation-wide panel company, Luc.id. Since around 2010 it has become increasingly obvious that it is virtually impossible to do online research, even with short interviews of more than 30 seconds without compensating the respondent. The days of massive responses to studies are finished, simply because people are both starved for time, and inundated with on-line surveys for every ‘trackable behavior’ of economic relevant. The refusal rate for interviews is skyrocketing. Thus, the use of online panel providers has dramatically increased, removing the onerous tasking of finding respondents for these short studies.

The study encompassed 460 respondents, with an interview lasting about 8-10 minutes. The compensated panelists generally do not ‘drop out’ of the study mid-way, as is the case for unpaid volunteers, where it is difficult to get panelists, and difficult to retain panelists to finish the task.

Table 2 gives a sense of the depth of information obtain about each respondent. Some of the questions were asked of the respondent at the time of the interview. Other questions were answered by third-party data purchased for the project.

Table 2: Example of some direct self-profiling classification questions and additional third-party data available and matched to the respondents by matching email addresses. A total of 145 additional data points were ‘matched’ to the study data of each of the 460 respondents.

Step 5: Create Models Which Relate the Presence/Absence of the Elements to the Rating

The respondent rated the vignettes on a 9-point scale. One might ordinarily wish to relate the presence/absence of the elements to the 9-point rating. The issue there is that we do not know, intuitively, what a 7 means, or what a 2 means, etc. We do know that the higher numbers mean that the respondent is more likely to register to vote, and that the lower numbers mean that the respondent is less likely to register to vote. That information is directional, but not sufficient.

In consumer and social research circles, there has been a movement to re-code scales such as the 1-9 or similar scales, to make the interpretation easier. We created six new binary scales, as follows:

The statistical analysis OLS (ordinary least squares) regression needs some minimum amount of variation in the dependent variable. Across the entire set of 460 respondents, it is very likely that the respondents will not generate the same rating (e.g., TOP3, all respondents rating the 48 respondents 7-9). If the respondents were to somehow do so, the statistical analysis would crash. On the other hand, for individual respondents, it is likely that a respondent might confine all ratings to 1-3, making BOT3 always 100. IN that case, the OLS regression would crash when creating a model or equation for that one respondent, bringing the entire processing to a halt.

To forestall the problem of a ‘crash; we add a vanishingly small number to each of the binary transformed variables that we just create ensuring that the actual transformed ratings vary a very little but do vary around the levels of 0 and 100, respectively. There is no meaningful effect on the regression coefficients emerging after performing this small prophylactic adjustment, but we prevent crashes. Indeed, without this adjustment, about 5% of the respondent models ‘crash’ because the respondent’s transformed numbers either all map to 100 or all map to 0.

The experimental design at the level of both the individual and at the level of the group allows us to create equations relating the presence absence of the elements to the transformed, binary ratings. We create six equations, each expressed as:

Binary Rating = k₀ + k₁(A1) + k₂₍A2) … k₃₅(F5) + k₃₆(F6)

Each equation is characterized by its own additive constant, and its own array of 36 coefficients. We interpret the additive constant as the expected percent of the respondents who will register to vote or not register to vote (according to the variable definition), albeit in the absence of elements. The additive constant is a purely estimated parameter but can be used as an index for predilection to register to vote. We expect increasing magnitudes of the additive constant as we go from TOP1 (Definitely intend to register to vote) to TOP3 (Intend to register to vote), and we expect a decreasing magnitude of the additive constant as we go from BOT3 (intend not to register to vote) to BOT1 (definitely not intend to register to vote). For the first analysis, we create six equations or models, based on the data from the total panel, and using each of the newly created binary variables as a dependent variable. With 36 elements, and six dependent variables, the OLS regression generates a massive amount of data (six additive constants, 216 coefficients, viz., 36 coefficients for each of the binary variables). That amount of information overwhelms the researcher, disguising patterns where they exist. To uncover the pattern, we blanked out all coefficients of 3 or lower, only to end with no strong performing elements. For the Total Panel only, we looked at elements with coefficients of +2 or higher. For all other analyses of coefficients, we look at elements with coefficients of +3 or higher. We begin first with the additive constant, the estimated propensity to register to vote, in the absence of elements. As we expected looking for individuals who feel strongly about voting generates a low additive constant of 10 (viz., for TOP1). We are likely to find only about 10% of the responses to be a ‘9’, in the absence of elements. When we make the criterion easier, accepting a 7, 8, or 9, (viz., TOP3) the additive constant jumps to 27.

Table 3 shows us that despite our efforts to find motivating elements, only six elements passed the relatively easy screen, viz., a coefficient of +2. The coefficient of +2 is very low in the world of Mind Genomics. Table 3 shows that the effort to find drivers of voting produced only three elements which show any promise, using the data from the total panel, and the promise they show is less than enthusiastic.

A4 You’ll be done registering in 5 minutes or less.

B2 You can get immediate answers for voter registration questions by calling, 1-800-FOR-VOTE.

D6 Voting allows you to be an advocate for your family and for your community: control your leadership and control your life.

When we move to the response ‘Not Register to Vote’ we see a similar pattern. The additive constants are similar in magnitude and go in the right direction. The strong statement about not voting, a rating of 1, captured by the variable BOT1, suggest that 10%, saying they would not definitely register to vote when the criteria are made less strict.

These are the three elements, presumed at the start of the experiment to drive positive voting, but instead drive the opposite, not registering to vote:

D3 Be an example for those who look up to you.

E5 You may have your things unpacked, but you haven’t moved in until you’ve registered.

F2 Government “of the people and by the people” requires your participation. The solution to your problems starts by registering.

Table 3: “Strong’ performing elements from the total panel, defined operationally as a coefficient of at least +2. Only those coefficients are shown. Missing elements failed to generate any coefficients of 2.0 or higher for any of the binary dependent variables.

Do People Change Their Stated Likelihood of Voting During the Interview?

Having now looked at the data from the total panel, and finding very little, we must pursue the reason why we fail to discover strong elements from the total panel. If we did not have the underlying structure, we would not know how weak the data are from the total panel. We would simply choose the strongest performing vignette and work with that vignette. Such an approach characterizes the research where the stimuli are put together, without structure. If we have one or two or even three or four elements varying, we might make a good guess, but we could not be sure. Mind Genomics take us in a different direction, to uncover the performance of the elements. It is those elements which constitute the building blocks of revised potentially better performing elements. Our first analysis looks at the (possible) change in the rating assigned by the respondents as the interview or experiment progresses. Recall that each respondent evaluated 48 unique vignettes, each vignette comprising 36 combinations of four elements (one from each of four questions), and 12 combinations or vignettes of three elements (one from each of three questions). By design, and by the systematic permutation of the vignettes, respondents saw different vignettes. We cannot measure change in the response to a specified vignette which most likely appeared just a few times, but we can measure the relation (if any) between the average rating assigned by the respondent and the order in the study (rating 1-9 for each vignette, order 1-48).

Our analysis uses OLS regression, done at the level of the individual respondent. For each respondent we know what was assigned to each vignette rated by the respondent, as well as the order of testing. We express the relation as: Rating (9-point scale) = k₀ + k₁(Order of Testing). The slope, k₁, tells us the effect of repeating the interview. We are interested in the sign of the slope, k₁, and then the magnitude of the slope. When k₁ is positive we conclude that the respondent becomes more interested in registering to vote as the interview or experiment goes on. It may be linked to the respondent being more sensitive to messaging. When k₁ is negative we conclude that the respondent becomes less interested in registering to vote as the interview or the experiment on. The respondent may be turned off. In turn, the magnitude of the slope, viz. the numerical value of k₁, tells us how many rating points on a 9-point scale will be added to the rating or subtracted from the rating for each additional vignette evaluated. Figure 1 shows the estimated magnitude of change in the rating assigned by a respondent across the 48 vignettes. Most of the respondents show a small change in the rating from vignette #1 to vignette #48. Most the respondents are within +/- two points on the 9-point rating scale. Keep in mind that the regression analysis generating the data was did not look at the actual range, but simply the pattern of changes manifesting itself at the individual respondent level.

Figure 1: The distribution of expected ranges to be expected as the respondent proceeds to evaluate 48 vignettes. Most of the range lies between an increase of 2 points to a decrease of 2 points from first vignette to last vignette.

Thus far we know the behavior of the respondent and can differentiate those respondents who are likely to increase versus decrease their ratings. We do not know anything about their criterion for making their judgments. We could ask the respondents to tell us their criteria, but it’s unlikely that they could tell us. The interview is so short, the vignettes judged so quickly, and the attention to the topic only modest while the interview is going on. Despite what might be wished for by novice researchers, most experienced researchers in these types of studies KNOW that their respondents are barely interested in the topic and are answering automatically to stimuli which much seem to them like a ‘blooming, buzzing confusion’. Those are the words of Harvard psychologist William James, when describing how a baby must perceive the world. Fortunately, Step 2 above tells us that despite the response of the respondent (or professional) asked to describe the test stimuli, there is a strongly laid structure underlying each respondent’s set of 48 vignettes. The structure prescribes exactly which elements belong in each vignette, doing so down to the level of a single respondent. We divide the respondents into three groups, defined qualitatively as those with positive range (one point or greater increase in the rating from vignette #1 to vignette #48), those with a flat range (between -1 and +1 point across 48 vignettes), and those with negative range (one point or greater decrease in the rating across 48 vignettes). The first become more interested in registering to vote, the second don’t really change their rating, and the get turned off.

Table 4 shows the strong performing elements for each group. Again, we select only those elements which have a breakthrough coefficient, now defined as +4, but which could easily be changed. The objective is to reduce the ‘wall of numbers’ to a limited set with the patterns coming through.

We see the following patterns emerging:

1. There are breakthrough elements for Groups 1 (positive range) and Group 3 (negative range), but no strong elements for Group 2 (flat range)
2. Despite the differences between the groups, and the differences in the patterns of the additive constants, there is no clear ‘story’ about what is driving Group1 (positive range) vs. Group 3 (negative range).

Table 4: Strong performing elements for three groups created on the basis of the range of the 9-point rating to be observed as the respondents proceeds to rate vignette 1 to vignette 48.

We conclude that if there is a story, it is deeper than the observed patterns of responses. Looking at large morphological differences in the patterns of responses gives us a lot of data, a lot of comparisons, but sadly no insight.

Uncovering Underlying Mind-sets based on the Pattern of Coefficients

One of the hallmark features of Mind Genomics is its focus on the decision-making of the everyday, and the recognition that the variability often observed in the data may be result in part from the combination of underlying groups with different criteria. A good metaphor is white light without color. One who looks at white light would say that it is colorless, but the structure of white is that emerges from three primary colors, red, blue, and yellow, respectively. Continuing the metaphor, what if the lack of strong, interpretable patterns in the data come not so much from lack of patterns, nor from intractable variability, but rather from the class of different mind-sets, having different criteria. The failure to uncover strong patterns may be the result of mutual cancellation. Mind Genomics researchers have worked out simple ways to identify these mutually exclusive primary groups, without the benefit of ‘theory’ about how the topic actually works, but simply on the basis of ‘hands-off’, clustering. Recall that each respondent evaluated a unique set of 48 vignettes, embodying the 36 elements in different combinations, with the data from each respondent constituting a complete experimental design. That is, each respondent both evaluated different combinations, but the mathematics of each set of combinations allows us to create a model for that individual [15].

To create these primary groups, or ‘mind-sets’, we followed these steps, adapting the Mind Genomics process, but incorporating two dependent variables simultaneously, register to vote (TOP2), and not register to vote (BOT2).

1. Create the mind-sets on the basis both of drivers of registering to vote, and drivers of NOT registering to vote. That is, we were interested in moving beyond one direction (drivers of registering to vote)
2. For each of the 460 respondents, create a model for TOP2 relating the presence/absence of the 36 elements to the TOP2 value. Create another model for BOT2. We thus have 460 pairs of coefficients, each pair comprising 36 coefficients.
3. When estimating the model for each respondent, do not use the additive constant. The rationale is that we will be combining the two sets of 36 coefficients to create a set of 72 coefficients for each respondent. All the information must be available solely in the coefficients. The technical appendix shows that estimating the coefficients without an additive constant produces the same pattern of coefficients as estimating the coefficients with an additive constant. The only difference is the magnitude of the coefficient. Figure 2 in the Technical Appendix shows the high co-variation between the two sets of coefficients, estimated for the same data, one without and one with the additive constant, respectively.
4. Create the 460 rows of data, comprising 36 coefficients for TOP2, and 36 coefficients for BOT2. Each respondent now has 72 coefficients.
5. Use principal components factor analysis to reduce the size of the matrix, by extracting all factors with eigenvalues of 1 or higher. This produced 19 factors.
6. Rotate the factors by a simplifying method, Quartimax, to produce a set of 19 new factors, rather than 72. Each respondent becomes a set of 19 numbers, the factor scores in the structure, rathe than a set of 72 numbers. We van be sure that the 19 factors are independent of each other.
7. Extract two and three clusters, or mind-sets, based on strictly numerical criteria [16]. The cluster method is the k means clustering, with the measure of distance between any two people defined by (1-Pearson Correlation between the two people on the 19 factors). In practical terms, any clustering method will do the job, since the clustering is simply a heuristic to divide the 460 respondents into similar-behaving groups
8. The principal component factor analysis allows us to create models for two segments (mind-sets) corresponding to the two-cluster solution, and three segments (mind-sets) corresponding to the three-cluster solution. The three-cluster solution was clearer. One could extract ore clusters, or mind-sets, but we opted for parsimony.

Create the six equations, with additive constants, for each of the three mind-sets, using the respondents allocated to the mind-sets. Eliminate all elements which fail to exhibit a coefficient of +4 in any model. This step winnows out most of the elements. The elements which remain show strong performance, and also suggest an interpretation, something not seen the previous data because variables did not have ‘cognitive richness’.

Table 5 suggests that there are three subtly different groups

MS 1 – A sense of voting is easy, fun, like sports. Don’t want to be reminded of the ‘seriousness’ of voting

MS 2 – Make it easy, make it simple, learn. They are ready. Nothing really turns them off.

MS 3 – Hates lines, make it easy. That’s all. Avoid talking about social responsibility. It’s a turnoff

The additive constants suggest that Mind-Set 1 (voting as fun) is most likely to register, without messages

Mind-Set 2 is likely to register. Nothing really turns them off.

Mind-Set 3 can be swayed by the right or wrong messages

The three mind-sets differ both in the pattern of likelihood to register and in the topics which turn them off, if there are any. Only Mind-Set 3 really responds in a way that suggest they are turned off.

Table 5: Strong performing elements for three emergent mind-sets (coefficient > = 4).

Figure 2: Scatterplot based on the data from the total panel, showing the strong co-variation of the 36 coefficients when estimated with an equation with an additive constant, vs. absent an additive constant.

Synergisms in Messages – Increasing the Likelihood of Mind Set 1 to Say They Will Register to Vote

As noted above, most conjoint measure studies with experimental design focus on a limited set of combinations, with the respondent testing all or only some of the combinations. None of the methods use permuted designs. It is the permuted design which allows the research to explore a great deal of the design space. One of the unexpected benefits is the ability to identify synergism and suppressions between pairs of elements. It to the study of interactions, and the search for synergism that we now turn.

Moskowitz and Gofman [11] introduced the notion of ‘scenario analyses for Mind Genomics. The guiding notion is that pairs of elements may synergize with each other, but the synergy could be washed out in a larger design. A better way to find out whether elements synergize is to select one of the questions (e.g., F), and separate all the data in the study into one of seven different strata, specifically all those vignettes where there is no F (by design), all those vignettes where F is held constant at F1, all those vignettes where F is held constant as F2, etc. Our starting data, therefore, is a set of several strata. We will end up running seven equations of the same type, one equation for each stratum. The equation will have only 30 independent variables (A1-E6), because for each stratus there is a single value of F, a single element. The set of elements from F are no longer independent variables. They simply exist in the vignette, or in the case of F0 deliberately left out of the vignettes. We can now select a target population, e.g., Mind Set 1, and run the regression seven times, once for each stratum. We will choose the most stringent dependent variable, TOP1 (definitely register to vote). The independent variables will be A1-E6, 30 out of the 36 variables. The additive constant is still the expected percent of response TOP1 (rating of 9, definitely register to vote), in the absence of the elements. The coefficients are the incremental percent of responses ‘will register to vote’ when the element appears in the vignette. Armed with that information let us now run the reduced model on each of the seven strata. Table 6 shows the coefficients. The columns correspond to the seven different strata. The rows correspond to the elements which show coefficients of at least +10 in one stratum. These are elements which are expected to synergize. To make navigating easier, and to uncover the strong performing combination, we present only those cells with positive coefficients of +4 or higher. The simplest way to discover combinations is to search for the shaded cells with the highest coefficient and add that high coefficient to the additive constant. The result will be the estimated score for that pair of elements as the key message. There are several very strong combination, combinations that we would not have guessed, first in the absence of mind-set segmentation, and second, in the absence of ability to uncover synergistic (or suppressive) combinations. A good example is the synergistic pair (F4, B6), and then ‘finished off’ with element D5. Table 6 suggests that the total score for TOP1 (definitely would register to vote) would be 11 for the additive constant, 15 for the synergistic pair (F4, B6, or 26 points. There is room for one more element, which we are free to choose, as long as the element makes intuitive sense and fits with F4 and B6. One example could be D5:

F4 = Big issues don’t slack during busy times. Neither should you. Go register today.

B6 = Everything you like and everything you don’t like in government came from elected officials. Your vote does matter.

D5 = Regret is the result of knowing you could have done more for your life. Register today, regret nothing tomorrow.

A possibly better strategy emerges when we look at F2 as an introductory phrase. The element itself does not bode well (additive constant of -1), but it synergizes with four of the six elements show in the stub (row) of Table 6.

F2 Government “of the people and by the people” requires your participation. The solution to your problems starts by registering.

B2 You can get immediate answers for voter registration questions by calling, 1-800-FOR-VOTE.

C4 New York City is first in a lot of things but ranks 46^th in voting. Lead the movement. #NYCVotesTheMost

D2The best things in life come free. Registering will satisfy your lifestyle by improving physical and mental wellness.

E4 The best things in life come free. Registering will satisfy your lifestyle by improving physical and mental wellness.

Table 6: Strong pairwise- interactions between elements F1-F6, and the remaining elements. The data come from respondents in Mind-Set 1.

Discussion and Conclusions

The world of public policy requires that the citizens perform their duties. Some of these duties are mandatory, such as military service, education, and obeying the law. Some are rights, not necessarily duties, such as registering to vote. Ask any group of people about how they feel about registering to vote, and you are likely to get a range of answers, from affirmation of patriotism, to indifference, to the absolute dislike of registering to vote because it is at once disinteresting, a duty, and worst of all, it puts one on the list for jury duty, another public service not in great favor. The sentiments around registering to vote are often simply measured as ‘yes/no’, e.g., will you register to vote or not register to vote. In this Mind Genomics study (really experiment), we have gone into the topic as it were a product or service, being offered to the respondent. We have used the language often used to ‘convince,’ only to discover that across the entire panel respondents, there are really no strong messages. If voting were a service or a product, we would ‘go back to the drawing board’ and try again

The speed, simplicity, cost, and templated structure of Mind Genomics, especially with smaller versions of the study presented here, 16 rather than 36 elements, makes it now possible to iterate through, testing different messages of a ‘public service’ nature. Public service messages may be viewed as a necessary evil, to be checked off, even though they contain little of a sales nature, and are primarily exhortations to do one’s duty and to be good citizens. Or, as Mind Genomics suggest, public service messages may provide the necessary matrix of ideas to use as a way to understand and to motivate the citizen. The study run here itself constitutes a larger-than-usual study in terms of the ideas explored. The results suggest a lack of knowledge of ‘what really motivates people,’ or more correctly a lack of understanding of people who are the targets of communication for a topic which is at best unromantic, quotidian, ordinary, and perhaps even potential negative because it could lead to jury duty. How interesting, however, the study becomes when we peel back the layers, understand the minds of people through segmentation, and through understanding of synergies where two messages combine to do far more than one expected. This type of information, collected across different types of studies, in an iterative process, builds a bank of knowledge for messaging about the common weal, the common good. Having a process such as Mind Genomics embedded in our societal life and in our political process offers far greater benefits to society than we can imagine today. Just imagine messaging for the social good, and doing it expeditiously, inexpensively, effectively.

Technical Appendix Relation between Coefficients Estimated with vs. without the Additive Constant

Mind Genomics is founded on the use of experimental design and OLS (ordinary least-squares) regression. Experimental design creates the test stimuli (vignettes) by specifying the specific combinations. OLS regression deconstructs the response to the vignettes, to estimate the part-worth contribution of each element to the respondent. Traditional Mind Genomics has worked with OLS regressions estimated with an additive constant. The constant is a measure of the likelihood of the response in the absence of the stimulus, a purely theoretical parameter. In this study, comprising six questions, each with six answers or elements, the experimental design called for 48 vignettes. Each element appeared 5x in the 48 vignettes and was absent 43 times. We can estimate two equations for the Total Panel, or indeed two equations for any subgroup, both equations using the same data.

Equation 1: Equation with the additive constant

Binary Dependent Variable (e.g., TOP2) = k₀ +k₁(A1) + k₂(A2) … k₃₆(F6)

Equation 2…which looks exactly like equation 1, but has no additive constant

When we estimate the coefficients, and plot one set against the other in a scatterplot, Figure 2 tells us that the patterns are the same, although the coefficients are higher when there is no additive constant. Figure2 shows an almost perfect co-variation of coefficients estimated in the two ways (R=0.94), with different values, however for the same element.

Acknowledgments

The authors wish to acknowledge the contributions of those students at Pace University, New York, who designed, executed, and wrote up the study for presentation to the Office of the Mayor of New York City. The strategy and messaging were used prior to the elections to encourage New Yorkers to register to vote in the upcoming election. The students then presented the study to professional direct marketers, at a direct marketing conference. In their own words ‘this is a great, practical tool which taught us a lot as we used it and got to see what we could accomplish.’

References

1. Harder J, Krosnick JA (2008) Why do people vote? A psychological analysis of the causes of voter turnout. Journal of Social Issues 64: 525-549.
2. Gershtenson J, Plane DL, Scacco, J.M. & Thomas, J., (2013) Registering to vote is easy, right? Active learning and attitudes about voter registration. Journal of Political Science Education 9: 379-402.
3. Jennings W, Wiezien C (2018) Election polling errors across time and space. Nature Human Behaviour 2: 276-283.
4. Rubado ME, Jennings JT (2020) Political consequences of the endangered local watchdog: newspaper decline and mayoral elections in the United States. Urban Affairs Review 56: 1327-1356.
5. Kaufman RR, Haggard S (2019) Democratic decline in the United States: What can we learn from middle-income backsliding? Perspectives on Politics 17: 417-432.
6. Dale A, Strauss A (2009) Don’t forget to vote: Text message reminders as a mobilization tool. American Journal of Political Science 53: 787-804.
7. Knack S (1993) The voter participation effects of selecting jurors from registration lists. The Journal of Law and Economics 36: 99-114.
8. Xu AJ, Wyer Jr RS (2012) The role of bolstering and counterarguing mind-sets in persuasion. Journal of Consumer Research 38: 920-932.
9. Krantz DH (1964) Conjoint measurement: The Luce-Tukey axiomatization and some extensions. Journal of Mathematical Psychology 1: 248-277.
10. Green PE, Krieger AM, Wind Y (2001) Thirty years of conjoint analysis: Reflections and prospects. Interfaces 31: S56-S73.
11. 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.
12. Miller GA, Galanter EH, Pribram KH (2017) (Orig. 1960). Plans and the Structure of Behavior, Routledge.
13. 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.
14. Moskowitz HR, Gofman A, Beckley J, Ashman H (2006) Founding a new science: Mind Genomics. Journal of Sensory Studies 21: 266-307.
15. Gofman A, Moskowitz H (2010) Isomorphic permuted experimental designs and their application in conjoint analysis. Journal of Sensory Studies 25: 127-145.
16. Xu R, Wunsch D (2008) Clustering (Vol. 10). John Wiley & Sons.