Perceiving Different Types of Bad People:  

How Moral Person Prototypes Influence Moral Impressions  

by  

Sarah Dimakis 

  

A dissertation accepted and approved in partial fulfillment of the   

requirements for the degree of  

Doctor of Philosophy   

in Psychology 

  

Dissertation Committee:  

Robert Mauro, Chair  

Sara Hodges, Core Member  

Gerard Saucier, Core Member  

Erik Girvan, Institutional Representative  

 
 

University of Oregon  
 

Summer 2024 
 
 
 
 
 
 
 



 
 

 
 

 

2 

 
 
 
 
 
 
 

 
 
 
 

© 2024 Sarah Dimakis   
This work is openly licensed via CC BY-NC-ND 4.0. 

 
 
 
  

https://creativecommons.org/licenses/by-nc-nd/4.0/


 
 

 
 

 

3 

DISSERTATION ABSTRACT 
 
Sarah Dimakis 
 
Doctor of Philosophy in Psychology 
 
Title: Perceiving Different Types of Bad People: How Moral Person Prototypes Influence Moral 

Impressions 
 

In response to a complex and information dense world, we organize related information 

into categories (e.g., furniture, animals) to make it easier to apply existing knowledge to newly 

encountered objects and situations (Rosch, 1978). Categories are represented in the mind as 

cognitive prototypes: examples that possess common features of category members (Posner & 

Keele, 1968; Reed, 1972; Rosch & Mervis, 1975). A key feature of prototype-based categories is 

that unobserved characteristics are inferred to category members once the prototype is activated 

(Cantor & Mischel, 1977; Osherson et al., 1990). I propose in this dissertation that we use moral 

person categories to make inferences about the unobserved or unobservable immoral 

characteristics of others, which influences our decisions about with whom to interact and how to 

interact with them. In a series of studies, I demonstrated that people perceive multiple bad person 

categories, examined their properties, and tested a prototype model of moral character evaluation 

against dimensional models. This dissertation is the first rigorous investigation of the structure 

and properties of perceived bad person categories.  

In Study 1, fifty prototypical characteristics of a bad person (e.g., lacking empathy, 

selfishness, racism) were gleaned from previous literature and augmented with additional 

characteristics generated by participants in a free-listing task. In Study 2, participant sorting data 

of characteristics from Study 1 revealed that respondents distinguished between multiple types of 

bad people (e.g., psychopath, abuser, narcissist, bigot). Additionally, categorical and dimensional 

models of bad person characteristics were generated from the responses in Study 2 and tested 

with new data in Study 3. Consistent with prototype-based categories, unobserved immoral 

characteristics that were prototypical of a category were more likely to be attributed to exemplars   

of that category compared to exemplars of competing bad person categories. Further, prototype 

models outperformed dimensional or multidimensional models, but prototype-unidimensional 

dual models performed the best. Thus, people infer prototype-consistent immoral characteristics 

to exemplars of a category, and additionally infer prototype-inconsistent immoral characteristics 



 
 

 
 

 

4 

that are similar in morality to the exemplar (but not similar on other social dimensions such as 

intelligence or sociability). Additionally, further providing support for multiple bad person 

prototype-based categories, large differences were observed in the perception of four bad person 

prototypes regarding their morality, competence, sociability, gender, race, age, physical 

appearance, the innateness and permanence of their immoral characteristics, and the ways 

respondents prefer to interact with them.  

Overall, this dissertation provides strong evidence against the unidimensional model of 

morality commonly used in the moral character evaluation literature and in support of a model of 

moral character evaluation that includes cognitive prototypes. This research may have significant 

implications for why we make errors that others are bad based on harmless or morally irrelevant 

characteristics (e.g., physical appearance, harmless behaviors), including in criminal justice 

contexts. In the future, the methods presented in this paper should be replicated with diverse 

samples to assess commonalities and differences in the cognitive prototypes of bad people across 

relevant subgroups (e.g., liberals versus conservatives, experts versus novices). 



 
 

 
 

 

5 

ACKNOWLEDGMENTS 
 
To start, I wish to express gratitude towards the person who has shaped this project and 

my academic journey the most, my faculty advisor Dr. Robert Mauro. I am profoundly grateful 

for your patience, kindness, encouragement, and wisdom over the past six years. I know I am a 

better researcher and a better person because I got to work with you. I want to thank Annika 

Zwolinski, Allis McAllister, Ciara Wyatt, and Eva Stojanovic, my extremely dedicated 

undergraduate research assistants who helped me generate items and code the responses in this 

paper. I also want to express my sincere gratitude for the members of my committee who have 

provided very insightful feedback on my research: Dr. Sara Hodges, Dr. Gerard Saucier, and Dr. 

Erik Girvan.  

I want to give a special thanks to Dr. Azriel Grysman, whose research lab I was 

extremely fortunate to have worked in as an undergraduate that served as a catalyst for my 

interest in psychology research. I also want to take this opportunity to thank the excellent 

educators I’ve been so fortunate to learn from in my life whose courses taught me the academic 

and life skills I needed to complete this degree, especially Dr. Daniel Anderson, Dr. Courtney 

Gibbons, Dr. Chinthaka Kuruwita, Dr. Tara McKee, Dr. Cameron Brick, Laurel Webster, Patti 

Crouch-Cook, Leon Phillips, Ryan Johnson, and Karen Quinn. 

Finally, I would like to thank my friends and family for your emotional support 

throughout this journey. To my friends and cohort mates Liz, Kim, Kyra, Kellyn, Ruby, Moriah, 

Meg, Andrea, Angela, Cam, and Andrew for being giant sources of happiness and strength. To 

my parents who sacrificed so much to give me the best education they could and who taught me 

to value learning over everything else. To my big sister Maria who taught me long division and 

how to realize you’re being conned. And to my partner Raleigh, our cat Amelia, and our dog 

Zucchini for bringing me so much joy every day. Thank you all.  



 
 

 
 

 

6 

TABLE OF CONTENTS 
Chapter               Page 

 

LIST OF FIGURES ........................................................................................................................ 8 

LIST OF TABLES .......................................................................................................................... 9 
I. INTRODUCTION ..................................................................................................................... 11 

DERIVING A TAXONOMY OF BAD PERSON CONCEPTS ............................................................... 18 
Present Study ........................................................................................................................ 23 

TESTING A PROTOTYPE MODEL AGAINST DIMENSIONAL MODELS OF IMMORAL INFERENCES ... 23 
A Prototype Model ................................................................................................................ 25 
A Multidimensional Model ................................................................................................... 26 
A Prototype-Dimensional Dual Model ................................................................................. 27 
Present Study ........................................................................................................................ 28 

EXAMINING PROPERTIES OF MORAL PERSON PROTOTYPES ....................................................... 28 
Three Dimensions of Person Perception ............................................................................... 28 
Essentialist Attitudes ............................................................................................................. 30 
Social distance ...................................................................................................................... 32 
Present Study ........................................................................................................................ 33 

II. STUDY 1: GENERATING CHARACTERISTICS OF A BAD PERSON ............................ 34 
METHOD .................................................................................................................................... 34 

Participants ............................................................................................................................ 34 
Procedure .............................................................................................................................. 35 
Materials ............................................................................................................................... 36 

RESULTS .................................................................................................................................... 37 
Generating a List of the Prototypical Characteristics of a Bad Person ................................. 38 
Explaining Variance in Conceptions of a Bad Person .......................................................... 39 

DISCUSSION ............................................................................................................................... 40 
III. STUDY 2: EMPRICALLY DRIVEN CATEGORY AND DIMENSIONAL MODELS OF 
BAD PERSON CONCEPTS ........................................................................................................ 49 

METHOD .................................................................................................................................... 49 
Participants ............................................................................................................................ 49 
Procedure and Materials ....................................................................................................... 50 

RESULTS .................................................................................................................................... 53 
Number of Groups ................................................................................................................ 53 
Group Names ........................................................................................................................ 53 
Similarity Matrix ................................................................................................................... 54 
Deriving a Category Model of Bad Person Concepts ........................................................... 56 
Deriving a Dimensional Model of Bad Person Concepts ..................................................... 67 

DISCUSSION ............................................................................................................................... 69 

IV. STUDY 3: TESTING MODELS OF IMMORAL INFERENCES ........................................ 73 
METHOD .................................................................................................................................... 75 



 
 

 
 

 

7 

Participants ............................................................................................................................ 75 
Research Design .................................................................................................................... 76 
Materials ............................................................................................................................... 77 
Procedure .............................................................................................................................. 82 

RESULTS .................................................................................................................................... 83 
Evaluating the Fit of the Empirically Derived Categorical Structure ................................... 89 
Testing Dimensional Models ................................................................................................ 91 
Evaluating the Fit of the Empirically Derived Dimensional Structure ................................. 93 
Comparing Prototype, Dimensional, and Dual Models of Immoral Inferences ................... 96 
Examining Attributes of Bad Person Prototypes .................................................................. 97 

DISCUSSION ............................................................................................................................. 109 

V. GENERAL DISCUSSION .................................................................................................... 122 
IMPLICATIONS .......................................................................................................................... 122 
LIMITATIONS AND FUTURE DIRECTIONS .................................................................................. 127 

Study Good People ............................................................................................................. 127 
Study Other Populations ..................................................................................................... 129 
Design an Intervention for Moral Perception Errors .......................................................... 132 

CONCLUSION ............................................................................................................................ 132 

APPENDIX A: INITIAL 70 BAD PERSON CHARACTERISTICS ........................................ 134 
APPENDIX B: FINAL 50 BAD PERSON CHARACTERISTICS ........................................... 136 

APPENDIX C: ESSENTALIST BELIEFS SCALE .................................................................. 138 
APPENDIX D: SOCIAL DISTANCE SCALE .......................................................................... 139 
APPENDIX E: MEAN AND STANDARD DEVIATION TABLES FOR LIKELIHOOD 
RATINGS ................................................................................................................................... 140 

APPENDIX F:  ERRORS FROM PROCRUSTES ANALYSES .............................................. 145 
REFERENCES CITED ............................................................................................................... 148 
 
 
 



 
 

 
 

 

8 

LIST OF FIGURES 
 
Figure Page 
 
1. Computerized Similarity-Sorting Task .................................................................. 52 

2. Distribution of Groups Created in Initial Sort ....................................................... 54 

3. Elbow Plot of Within-Cluster Variance for Initial Sort ......................................... 57 

4. Plot of Average Silhouette Width for Initial Sort .................................................. 57 

5. Dendrogram of Hierarchical Cluster Analysis Ward’s Method (k = 3)................. 59 

6. Dendrogram of Hierarchical Cluster Analysis Ward’s Method (k = 7)................. 60 

7. Dendrogram of Hierarchical Cluster Analysis Ward’s Method (k = 10)............... 61 

8. Sorting Agreement for Seven-cluster Solution ...................................................... 65 

9. Three-Dimension Solution from MDS (Morality, Competence Dimensions) ....... 70 

10. Three-Dimension Solution from MDS (Morality, Sociability Dimensions) ......... 71 

11. Example Instructions for Introducing an Exemplar ............................................... 76 

12. Likelihood Exemplars Have Unobserved Characteristics ..................................... 85 

13. Recreating of Category Structure Derived in Study 2 ........................................... 89 

14. Morality and Competence Ratings for Exemplars and Characteristics ................. 94 

15. Morality and Sociability Ratings for Exemplars and Characteristics .................... 95 

16. Imagined Gender for Each Prototype .................................................................... 99 

17. Imagined Race for Each Prototype ........................................................................ 100 

18. Imagined Age for Each Prototype .......................................................................... 101 

19. Essentialism Ratings by Prototype ......................................................................... 106 

20. Desired Social Distance by Prototype .................................................................... 107 

 



 
 

 
 

 

9 

LIST OF TABLES 
 
Table Page 
 
1. Bad Person Characteristics Unconstrained Listing Ranked by Frequency ............ 42 

2. Bad Person Characteristics Unconstrained Listing Ranked by Mean Badness ..... 44 

3. Descriptives and Zero-Order Correlations of Bad Person Characteristics ............ 47 

4. Most Frequent Labels Given to Describe Initial Groupings .................................. 55 

5. Membership Values from C-Means Analysis ........................................................ 66 

6. Research Design Demonstrating How Exemplars Are Defined ............................ 77 

7. Pool of Characteristics Describing Exemplars ...................................................... 80 

8. Factor Loadings of Judgments of Exemplars and Characteristics ......................... 82 

9. Model Predicting Likelihood of Psychopath from Category Membership ............ 86 

10. Model Predicting Likelihood of Abuser from Category Membership ................... 87 

11. Model Predicting Likelihood of Bigot from Category Membership ..................... 88 

12. Model Predicting Likelihood of Narcissist from Category Membership .............. 88 

13. Model Fit Metrics for Time of Responding to Inferences ..................................... 91 

14. Model Fit Metrics for Models Likelihood of Inference ......................................... 98 

15. Means (SDs) for Differences Between Cognitive Prototypes ............................... 103 

16. Participant Generated Unobserved Characteristics of Bad Person Exemplar ....... 111 

17. Participant Generated Unobserved Characteristics of Good Person Exemplar ..... 112 

18. Participant Generated Unobserved Characteristics of Psychopath Exemplar ....... 113 

19. Participant Generated Unobserved Characteristics of Abuser Exemplar .............. 114 

20. Participant Generated Unobserved Characteristics of Bigot Exemplar ................. 115 

21. Participant Generated Unobserved Characteristics of Narcissist Exemplar .......... 116 



 
 

 
 

 

10 

22. AI Images from Appearance Descriptions of Basic Prototypes ............................ 117 

23. AI Images from Appearance Descriptions of Superordinate Prototypes ............... 118 



 
 

 
 

 

11 

I. INTRODUCTION 
 

The label bad person is used to describe a wide range of individuals who violate moral 

norms. This includes those in positions of power who organize and benefit from suffering on a 

massive scale, such as genocidal dictators, terrorist organization leaders, cult figures, sex 

traffickers, and drug lords. It includes common criminals directly responsible for inflicting harm 

onto others, such as serial killers, mass shooters, sex offenders, batterers, stalkers, scammers, and 

burglars. Even those who have not broken any laws are sometimes referred to as bad people, 

including political adversaries, members of marginalized communities, billionaires, atheists, 

‘climate criminals,’ schoolyard bullies, and self-absorbed popular ‘mean girls.’ 

The people listed above differ in a lot of important ways. Some are violent, while others 

are too focused on themselves. Some have high intelligence that makes it easy to deceive others, 

while some individuals’ immorality stems from their perceived uninformed or unintelligent 

beliefs. Some get close to their targets through their warmth and charm, while others are 

reclusive and disagreeable. Some are seen as capable of reform, while others are perceived to 

have immoral characteristics that are hardwired in their brain or DNA. Given the wealth of 

important information to attend to, how do we know what to expect and how to act when we 

encounter people like these? 

In response to an information dense world, we organize related information into 

categories (e.g., furniture, animals) to help us know how to interact with and what to expect from 

new stimuli in our environment (Rosch, 1978). Categories are represented in the mind as 



 
 

 
 

 

12 

cognitive prototypes: average or typical examples with features abstracted from experiences1 

with category members (Posner & Keele, 1968; Reed, 1972; Rosch & Mervis, 1975) and as 

exemplars: specific category members that a person has interacted with or learned about (Medin 

& Schaffer, 1978; Medin & Smith, 1981). A new stimulus is granted category membership 

according to the degree of similarity between the new stimulus and the mental representation of 

the category (Dopkins & Gleason, 1997; Rosch & Mervis, 1975). For example, when we 

encounter a chair we have never seen before, we recognize it as a chair because it is similar to 

our mental representation of a chair. Once we have categorized it as a chair, we immediately 

know how to interact with it. A chair can be sat on. 

In the social domain, we categorize people into groups defined by shared attributes, such 

as women, Canadians, extroverts, runners, and accountants. These social categories and their 

mental representations (e.g., prototypes, stereotypes2) enable us to make quick inferences and 

 
1 Experiences with category members does not need to be direct. For categories where it is rare 

to encounter a category member, the cognitive prototype may be entirely constructed from 

indirect experiences. For example, we are likely to receive more information about serial killers 

from the media (e.g., news, true crime documentaries) than from direct experiences because of 

how rare it is to encounter a category member.  

2 Both stereotypes and prototypes shape how we perceive and interact with strangers. While 

stereotypes are (usually negative) beliefs and attitudes about members of social categories (e.g., 

women, Asians), prototypes refer to the mental representations of concepts within and outside 

the social domain (Hilton & Hipple, 1996; Rosch, 1976). In this research, I use the term 



 
 

 
 

 

13 

interaction decisions about people we have limited knowledge of (Bodenhausen et al., 2012; 

Cantor & Mischel, 1977; Macrae & Bodenhausen, 2000). For example, upon learning that your 

neighbor is an accountant, you might infer that he has the stereotypical characteristics of 

accountant -- for instance, that he is boring (Tilburg et al., 2022) -- and interact with him 

accordingly. You might assume that he is trustworthy enough to approach at the block party, 

although a boring conversationalist.3 

An important social interaction decision is to decide whether to approach or to avoid 

someone. Moral psychologists have theorized that people make this distinction when we divide 

people into good people and bad people, where good people are typically those we want to form 

relationships with and bad people are typically those we want to avoid (Hartman et al., 2022; 

Heiphetz, 2020; Helzer & Critcher, 2018; McHugh et al., 2022). Just like we categorize a new 

chair based on its resemblance to the chair prototype, I propose in this dissertation that we 

evaluate moral character based on how closely a target resembles our mental representation of a 

good person or bad person. Once a categorization is made, I hypothesize that we are more 

inclined to approach, praise, and form relationships with those we perceive as good and to avoid 

and punish those we perceive to be bad. 

In addition, I propose that there are more than two categories in the moral person domain. 

In general, categories representing a concept are hierarchically structured, with a basic level 

 
prototypes instead of stereotypes as I am referring to the mental representations of social 

categories rather than beliefs or attitudes about social categories. 

 
3 I mean no offense to accountants. Your job is very valuable, and I’m sure you are really fun to 

be around. 



 
 

 
 

 

14 

where categories are highly informative and distinct from each other, a superordinate level 

where categories are more abstract, more different from each other, and have more members, and 

a subordinate level where categories are more concrete, more similar to each other, and have 

fewer members (Rosch, 1978). For instance, if furniture is at the superordinate level, then chair 

would be a category at the basic level, and desk chair would be a category at the subordinate 

level.  

We should expect a similar type of hierarchy for bad people. Given the diversity of 

members in the bad person category, the mental representation of a bad person is not sufficiently 

detailed to make nuanced inferences and interaction decisions. For example, just as describing a 

chair as furniture would not provide specific advice on how to interact with the object, 

categorizing someone as a bad person would lead to uncertainty surrounding how to interact 

with and what to expect from them. For instance, while some bad people pose an immediate 

physical threat to our safety, others are merely untrustworthy or unreliable friends or business 

partners.  

Therefore, I propose that we also perceive moral person basic categories that are less 

inclusive than the superordinate bad person category, such as psychopath, bigot, pedophile, 

dictator, terrorist, and narcissist. Lower-level moral person categories have richer mental 

representations compared to superordinate categories because they contain additional, category 

specific information. And since basic categories optimize cue validity, or the ability for a given 

piece of information to predict membership into a category (Rosch et al., 1976), I propose that 

the formation and maintenance of basic categories plays a crucial role in moral character 

evaluation. 



 
 

 
 

 

15 

Psychologists researching moral character evaluation define moral character as an 

individual’s perceived disposition towards morally relevant thoughts, feelings, choices, traits, 

and behaviors (Cohen & Morse, 2014; Goodwin et al., 2014; Hartman et al., 2022). Nearly all 

research on moral character evaluation has neglected to study moral person categories or has 

concentrated on the superordinate good person and bad person categories (e.g., Hartman et al., 

2022; McHugh et al., 2022). Most researchers use a simple, unidimensional model of moral 

character where targets can be placed and ranked against each other on a moral dimension from 

extremely immoral to extremely moral (e.g., Brandt & Reyna, 2011). 

Primarily, researchers have examined whether a single piece of information can influence 

moral impressions by manipulating one characteristic in a scenario and keeping all other 

information constant. For example, by having participants indicate on a unidimensional scale of 

morality where they would place targets who make different decisions in moral dilemmas 

(Critcher et al., 2020; Uhlmann et al., 2013), who have different faces (Rosa et al., 2021; 

Todorov et al., 2015), or express different emotions in a moral situation (Anderson et al., 2021; 

Stearns & Parrott, 2012). When they observe a significant difference between groups, they 

record the variable as a factor that influences moral impressions. 

However, a unidimensional model of moral character evaluation is insufficient because it 

assumes that members from different moral person categories who score the same on the 

unidimensional scale are more similar than they are perceived to be. For example, two people 

perceived to have the same moral character on a unidimensional scale may be expected to violate 

different moral norms, be assessed to have different capabilities to carry out their intentions, or 

to be avoided to different extents. Thus, I propose that a more complex model (e.g., 

multidimensional, multicategory) of moral character evaluation could more powerfully explain 



 
 

 
 

 

16 

how people infer unobserved immoral characteristics in others and make decisions about with 

whom to interact and how to interact with them.   

In this dissertation project, I tested and compared the unidimensional model to two 

additional models of moral character evaluation: a prototype model and a multidimensional 

model. In a unidimensional model, a target is inferred to have unobserved characteristics that 

align with their perceived morality. For example, a moderately bad person would be assumed to 

have moderately bad characteristics, and less so extremely bad or slightly bad characteristics. In 

contrast, in a prototype model, a target who is similar to a cognitive prototype is inferred to 

have unobserved immoral characteristics that are prototype consistent. For example, a target who 

has characteristics similar to the psychopath prototype would be inferred to have unobserved 

characteristics consistent with the psychopath prototype. In a multidimensional model, 

additional dimensions of person perception (e.g., competence, sociability) can explain why two 

individuals with the same morality are inferred different immoral attributes. For instance, an 

immoral, competent, and sociable person would be more likely to be inferred competent and 

sociable characteristics compared to an immoral, incompetent, and unsociable person.   

Fully understanding the process of moral character evaluation is vital because person 

perception errors in the moral domain can have serious consequences. Failing to recognize 

someone’s capacity for harm can lead one to be the victim of deception, exploitation, or 

violence. Conversely, falsely accusing someone of poor character can result in a missed 

opportunity to connect with and benefit from a partnership, or the injustice of wrongful 

accusations or punishment. Thus, developing a deeper understanding of the mechanisms that 

underlie moral person judgments can inform interventions aimed at preventing harmful errors 



 
 

 
 

 

17 

from occurring. In summary, my dissertation had three goals that will deepen our understanding 

of those mechanisms: 

1. Derive a Taxonomy of Bad Person Concepts. Expand our understanding of how 

people store immoral person characteristics by deriving basic categories of the bad 

person concept. Deriving a taxonomy of bad person concepts from a relatively 

homogeneous population would establish that individuals within a subculture share 

mental representations of more than one type of bad person. 

2. Test a Prototype Model Against Dimensional Models of Immoral Inferences. 

Explain how unobserved immoral characteristics are inferred about a target from a set 

of observed immoral characteristics. Testing whether immoral characteristics are 

inferred to a target because the initial and inferred characteristics are perceived to be 

prototypical of the same type of person (i.e., prototype model), because of their 

proximity to the target on a dimension or dimensions (i.e., dimensional models), or 

because of prototypes and proximity (i.e., a dual model) would improve our 

understanding of moral character evaluation. 

3. Examine Properties of Moral Person Prototypes. Evaluate whether we perceive 

and interact with various types of bad people differently. Testing for differences in 

perceived morality, competence, sociability, gender, age, race, physical appearance, 

beliefs regarding the immutability of their immoral characteristics, and desire for 

social distance of characterizations of exemplars would deepen our understanding of 

why people make basic category distinctions in the moral domain.   

  



 
 

 
 

 

18 

Deriving A Taxonomy of Bad Person Concepts 
 

Knowing what are the basic categories in a domain is useful for understanding how 

people make everyday distinctions between different members of a superordinate category 

(Bodenhausen et al., 2012; Rosch et al., 1976). For social categories, identifying the basic level 

of a taxonomy can inform our understanding of how people naturally describe and understand 

others. People prefer to describe others at the level of abstraction that is inclusive of the greatest 

number of people but still predictive of behavior, as opposed to being purely evaluative (John et 

al., 1991). For example, in the personality hierarchy stingy, selfish, unkind, and bad, people 

prefer to describe others as unkind, which is the most abstract word in the hierarchy that still 

informs what kind of behavior the person is likely to engage in. Similarly, people may prefer to 

refer to those who violate moral norms with terms that are more predictive of behavior than bad 

person, such as psychopath, narcissist, or sex offender. However, while prior theory has alluded 

to “an infinite number of categories” in the moral domain beyond the superordinate categories 

good and bad (p. 133; McHugh et al., 2022), there have been no sufficiently rigorous empirical 

investigations to derive a moral person taxonomy.   

The very first draft of a moral person taxonomy may go as far back as the 4th century 

BCE. Scholars have reconstructed writings from one of Aristotle’s students, Theophrastus, who 

penned highly entertaining character sketches of 30 different types of men (Anderson, 1970). 

Theophrastus depicted each man with a disagreeable characteristic and described how each type 

of man could be identified across different situations. Many of his characters would be 

considered a bad person by today’s standards. For example, the Man without Moral Feeling does 

not feel guilt or remorse when he violates moral norms, the Insincere Man compliments others to 

their face and then speaks ill of them behind their back, the Arrogant Man believes he is superior 



 
 

 
 

 

19 

to others, and the Shamelessly Greedy Man hoards more resources than he needs. Other 

characters possess negative but not necessarily immoral traits, such as the unhygienic Repulsive 

Man who goes to bed without showering and the Talkative Man who constantly expresses what 

he is thinking, even in socially inappropriate situations.  

Although the exact purpose (e.g., entertainment, theatre, personal vendetta) of 

Theophrastus’s character sketches is unknown, at some point The Characters was used for moral 

instruction. The preface of most modern translations depicts Theophrastus informing his 

interlocutor Polycles that the character sketches will “help [our children] to choose the best 

people to know and be with, and they will have something to aim at” (Anderson, 1970, p. 3). 

Thus, the highly detailed examples of morally flawed individuals that Theophrastus sketched out 

many centuries ago may have been used or intended to be used to instruct children on who to 

avoid. 

Additionally, throughout history Christian theologists have sought to compile a list of 

sins or vices that lead to immoral action to use in their moral instruction. Although the specific 

vices that are considered sinful have evolved over time, the most influential and well-known set 

emerged during the Middle Ages: The Seven Deadly Sins – pride, envy, anger, avarice, sloth, 

gluttony, and lechery (Bloomfield, 1952). Although Christian theology specifies that these are 

sins that everyone must resist for a virtuous life, it may be that people perceive seven different 

sinners: a person who believes they are superior to others (pride), a person who is excessively 

envious of others’ success and possessions (envy), a person who cannot control their anger, 

hatred, and desire for vengeance (anger), a person who hoards wealth and possessions (avarice), 

a person who lacks motivation and ambition (sloth), a person who overindulges on food and 



 
 

 
 

 

20 

drink (gluttony), and a person who violates moral norms such as marital fidelity in the pursuit of 

sexual pleasure (lechery). 

In modern social psychology research, Moral Foundations theorists have proposed that 

there are at least six categories of moral violations, similar to the seven sins from Christian 

theology: harm, inequality, disproportionality, disloyalty, subversion, and impurity (Atari et al., 

2022; Graham et al., 2013). If categories of moral violations map directly onto norm violators, 

then Moral Foundations Theory would expect six types of immoral agents: a type of person who 

physically harms others (harm), a type of person who treats some people worse than others 

(inequality), a type of person who takes more than they deserve (disproportionality), a type of 

person who is disloyal to their groups and allies (disloyalty), a type of person who disobeys 

authority and disrespects tradition (subversion), and a type of person who defiles objects, bodies, 

or institutions that are considered pure (impurity). 

The validity of this six-category model of moral violations was tested by asking 

participants to rate the likelihood that an agent would violate a moral norm, given that they 

violated a different moral norm (Landy & Bartels, 2018). A cluster analysis revealed that people 

perceive three types of bad people: a person who violates moral and social norms (e.g., cheating, 

hurting others, being impolite), a person who does disgusting behaviors, usually of a sexual 

nature (e.g., watching deviant pornography), and a person who uses power to restrict or deny the 

freedoms of others (e.g., forcing their daughter to take premed classes). This study gives initial 

support for the claim that people distinguish between different types of bad people. However, the 

authors’ main objective was to evaluate a taxonomy of moral behaviors, not to derive basic 

categories of bad people. As a result, they limited their characteristics to behaviors and did not 

provide any cognitions (e.g., beliefs, intentions) or affect (e.g., guilt, empathy), which are 



 
 

 
 

 

21 

especially powerful signals of moral character (Critcher et al., 2020; Helzer & Critcher, 2018). 

Thus, Landy and Bartels (2018) may be underestimating the actual number of perceived basic 

categories, as they did not test whether people distinguish more finely between categories based 

on perceived cognitions and affect. 

Additionally, personality research may offer insight into a moral person taxonomy insofar 

that perceived categories in part reflect underlying realities concerning different antagonistic 

personality traits individuals possess. For example, the Dark Triad proposes three antisocial or 

harmful personality traits that individuals possess to differing extents: psychopathy, 

Machiavellianism, and grandiose narcissism (Paulhus & Williams, 2002). Thus, people may 

perceive three types of bad people: psychopaths, who have emotional deficiencies and 

inclinations toward impulsive, thrill-seeking behavior, which can lead them to harm others 

without guilt or remorse (Dolan & Fullam, 2006; Hare et al., 1990; Holmqvist, 2008), 

Machiavellians, who manipulate and exploit others through careful planning (Jones & Paulhus, 

2014; Monaghan et al., 2016; Rauthmann & Kolar, 2012), and grandiose narcissists, whose 

entitlement and grandiose sense of selves may lead them to become aggressive if their positive 

self-concepts are threatened (Bushman et al., 2003; Gentile et al., 2013). 

Further, others have proposed additional “dark traits” such as sadism, spitefulness, greed, 

perfectionism, and dependency that could be perceived moral categories (Marcus & Zeigler‐Hill, 

2015; Mededović & Petrović, 2015). For example, people may perceive the moral category 

sadists who derive pleasure from causing others psychological or physical pain (Baumeister & 

Campbell, 1999; Buckels et al., 2013) and spiteful people who are willing to take some cost to 

themselves to harm others (Marcus et al., 2014). It is reasonable to expect that any number of 

these categories may emerge as perceived moral person categories.  



 
 

 
 

 

22 

Additionally, prior work on personality structure derived from type-nouns may inform 

the expected structure of perceived categories. For instance, US students indicated the extent to 

which a large set of type-nouns (i.e., nouns that define a class of people with shared attributes, 

such as lawbreakers or psychopaths) described themselves and others (Saucier, 2003). A factor 

analysis revealed that an eight-category solution fit the data reasonability well, revealing a 

personality structure that included more than one morally relevant factor: an egocentric factor 

(e.g., snob, gossip, eavesdropper) and a delinquency factor (e.g., lawbreaker, pothead, drunk). 

Insofar as moral perception in part reflects reality, this work may indicate that US students 

perceive a type of person that is negative and critical, and another type of person who breaks 

laws. However, given that participants were young adults who were instructed to describe people 

they know, there may be additional factors that were not captured by the methodology that are 

important in criminal contexts (e.g., serial killer, psychopath) or may emerge in other samples. 

For example, a German study of type-nouns found a factor for men related to cowardice (Henss, 

1998). 

Recently, Hartman et al. (2022) theorized a taxonomy of moral person types with three 

superordinate categories villain, hero, and victim, and basic categories saint, striver/sinner, 

demon, opportunist, virtuous victim, and culpable victim. In this taxonomy, basic categories 

represent different levels of prototypically to the superordinate category. For instance, a demon is 

evil in all situations, while an opportunist is inclined towards evil but whether the inclination is 

present or not will depend on the situation. This model suggests that the only important 

differences that people observe between basic moral person categories are valence (i.e., whether 

a target is perceived to be good or bad) and strength (i.e., whether a target is perceived to be 



 
 

 
 

 

23 

capable of acting on good or bad intentions). However, no empirical test of this category 

structure has been conducted. 

Present Study 
 

My dissertation project aims to contribute to the literature by first establishing that people 

distinguish between different types of bad people based on more complete information about 

their cognitions, affect, behavior, and personal characteristics4. This study is necessary, as 

existing taxonomies of moral character types have been theoretical, derived to understand 

personality structure rather than person perception, or empirically driven from behaviors only.  

Testing a Prototype Model Against Dimensional Models of Immoral Inferences 
 

The unidimensional model of moral character evaluation that is widely used in the moral 

psychology literature often predicts that two targets with identical moral character scores will be 

perceived as more similar to each other than they are actually perceived to be. For instance, two 

 
4 It's essential to clarify that I do not aim to derive a cross-culturally valid taxonomy of 

moral person types. Just as moral norms differ significantly across cultures (Graham et al., 2009; 

Haidt et al., 1993; Janoff-Bulman & Carnes, 2013; Rai & Fiske, 2011; Schein & Gray, 2015), I 

anticipate substantial cross-cultural disagreement regarding which types of bad people are 

perceived and how they are cognitively represented. To properly test models of moral inferences, 

it is necessary to understand the taxonomic structure of the participants within the culture being 

sampled. Therefore, this project derives basic categories from the same pool of participants that 

were then used to test models of moral judgment, but the taxonomic structure derived should not 

be interpreted as one that is universally shared. I expand on this point in the general discussion. 

 



 
 

 
 

 

24 

individuals who are both considered to be a moderately bad person in a unidimensional model 

may differ in additional ways that are important to accurately predict their behavior, infer 

unobserved or unobservable characteristics, recommend punishment, and make informed 

decisions about how and when to interact with them. By measuring and conceptualizing moral 

character on a unidimensional scale, researchers cannot fully explain how observers make 

nuanced moral judgments and decisions that are important in real-world contexts. 

 For example, researchers using a unidimensional measurement of morality have found 

that we sometimes make moral character judgments that do not reflect the morality of the 

observed behavior, such as characters in vignettes being rated as more immoral than their 

behavior warrants (Tannenbaum et al., 2011; Uhlmann & Zhu, 2014) or more moral than their 

behavior merits (Schaumberg & Mullen, 2017). Of interest here, many studies have found that 

behaviors or characteristics that are considered relatively harmless can be perceived as strong 

evidence of a target’s poor character. For instance, animal-abusers were rated as having poorer 

character than wife-abusers, despite wife-abuse being rated as more immoral than animal-abuse 

(Uhlmann & Zhu, 2014). In another study, job candidates who asked for a $40,000 marble table 

with their face engraved on it were rated as having poorer integrity than candidates who asked 

for a $40,000 cash bonus (Tannenbaum et al., 2011).  

 A unidimensional model of moral character cannot explain these findings. A 

unidimensional model would predict that a target’s morality should be perceived as similar to the 

morality of the observed characteristics, and targets should be most likely to be inferred 

characteristics that are similar in morality to the target. However, other models may be better 

able to explain moral character evaluation: a prototype model, a multidimensional model, and a 

dual model.   



 
 

 
 

 

25 

A Prototype Model  
 

According to prototype theory, categories are represented in the mind as a summary of 

attributes and their relations, abstracted from direct and indirect experiences with members from 

the category (Posner & Keele, 1968; Reed, 1972; Rosch & Mervis, 1975). Further, membership 

is granted based on the degree of similarity between the new stimulus and the cognitive 

representation of the category (Rosch & Mervis, 1975). Thus, if a target’s observed 

characteristics are similar to a perceiver’s prototype of a bad person category, then the target 

would be classified into the category and transferred unobserved or unobservable characteristics 

that are consistent with the perceiver’s mental representation of the category (Cantor & Mischel, 

1977; Osherson et al., 1990; Smith & Zárate, 1992). Therefore, a prototype model of moral 

character evaluation would posit that similarity to a prototype would predict the likelihood that a 

target is inferred unobserved immoral characteristics.  

Reflecting on surprising findings in the moral character evaluation literature, Schein and 

Gray (2018) noted, “[c]onsider a man who masturbates with a dead chicken. Although this act is 

‘harmless,’ parents likely would not want him babysitting their daughter for fear of the harm he 

might do to her.” Thus, relatively harmless characteristics could lead to severe moral 

commendation of a target’s character when the target is assumed to have immoral characteristics 

that were not originally observed (i.e., future harmful behavior). For instance, a prototype model 

would predict that someone who masturbates with a chicken would be categorized as a sex 

offender or psycho and inferred other characteristics of their category, and thus receive more 

condemnation than the behavior warrants. 

  Additionally, consider the finding that an animal abuser was perceived as having worse 

moral character than a wife abuser, even though wife abuse was rated as more morally wrong 



 
 

 
 

 

26 

than animal abuse (Uhlmann & Zhu, 2014). Animal abuse is widely known to be part of the 

MacDonald Triad, factors that when observed together are theorized to (although not empirically 

supported) predict a serial killer (Parfitt & Alleyne, 2020). While animal abuse may not be rated 

as more immoral than spousal abuse, the perceiver may be inferring unobserved characteristics 

based on perceived similarity to a prototype (e.g., a psychopathic serial killer), and therefore rate 

the target as more immoral than the action itself would warrant. In another example, when the 

interviewee requested a marble table with their face on it, it may have activated the perceiver’s 

narcissist prototype, whereas simply asking for more money did not (Tannenbaum et al., 2011). 

Thus, a prototype model, which posits that targets are inferred unobserved characteristics when 

they are prototype consistent, may better explain inferences compared to a unidimensional 

model. 

A Multidimensional Model  
 

Another model that may offer additional explanation above a unidimensional model is a 

multidimensional model. For instance, Hartman et al. (2022) proposed strength as a second 

dimension of moral character, which concerns differences in the ability to carry out moral or 

immoral intentions and seemingly captures the competence dimension of person perception (e.g.,  

Brambilla et al., 2021; Goodwin, 2015). Given that moral person perception is expected to utilize 

the same mental processes involved in general person perception, we should also expect that 

people are attuned to the third dimension of person perception, sociability, which reflects the 

target’s outward extroversion and charm (Brambilla et al., 2021; Goodwin, 2015). Judgments 

about sociability are relevant to encode because they can predict the target’s capacity to integrate 

into society, persuade others of their moral beliefs, and deceive others into perceiving them as 

virtuous. Thus, a multidimensional model, which posits that targets would be inferred 



 
 

 
 

 

27 

unobserved characteristics that are similar to the target on multiple dimensions (e.g., morality, 

competence, and sociability), may better explain inferences compared to a unidimensional 

model. 

A Prototype-Dimensional Dual Model 
 
 A prototype model of moral character evaluation expects that category membership will 

predict the likelihood that a target will be inferred a characteristic, such that targets are more 

likely to be inferred characteristics that are consistent with their category and less likely to be 

inferred characteristics that are inconsistent with their category. Further, if the categorical 

structure has a strict hierarchy, all category-inconsistent information is equally likely to be 

inferred. A dimensional model of moral character evaluation offers that the distance from the 

observed and inferred characteristics on one or more dimensions (e.g., morality, competence, and 

sociability) predicts the likelihood that a target will be attributed an unobserved characteristic. 

According to a dimensional model, category membership does not offer any additional 

explanation. 

 However, these models may be too simple, as it is possible that both category and 

distance provide explanation for why characteristics are inferred. For instance, it is possible that 

certain categories are perceived to be closer in the dimensional space to some categories 

compared to others. For example, people may perceive a bigot and a narcissist to be closer in 

morality, competence, and sociability than they are to a psychopath, and therefore infer that a 

bigot has narcissistic traits to a higher degree than psychopathic traits. Thus, a dual model that 

includes both dimensional and prototype components may best explain moral character 

evaluation.  

  



 
 

 
 

 

28 

Present Study 
 
  My dissertation project aimed to contribute to the literature by explaining why, given a 

set of initial immoral characteristics about a target, additional immoral characteristics are 

inferred. I hypothesized that the prototype, multidimensional, and dual models would provide a 

better explanation for immoral inferences than the unidimensional model that is widely used in 

the literature. 

Examining Properties of Moral Person Prototypes  
 

The final goal of my dissertation project was to evaluate whether there are meaningful 

and systematic differences between how we perceive and interact with different types of bad 

people. Cataloguing these differences provides valuable insight into why we employ basic 

categories to make more accurate judgments and decisions. Additionally, by examining the 

properties of bad person prototypes, I have illuminated harmless characteristics (e.g., physical 

appearance, social identities) that can lead to a false positive categorization that the target intends 

or desires harm. If I would have failed to find differences between exemplars of various bad 

person categories, then a parsimonious conclusion would be that there is one category: bad 

person. 

Three Dimensions of Person Perception 
 

Impressions of a target can be explained almost entirely by the perceiver’s independent 

assessments of the target’s warmth and competence (Abele et al., 2021; Abele & Wojciszke, 

2014; Cuddy et al., 2008; Koch et al., 2016). Warmth refers to how the target interacts with 

others, including their friendliness, sociability, honesty, and trustworthiness, and competence 

refers to the target’s ability to complete their goals, including their confidence, assertiveness, 

intelligence, and skill. Recently, researchers have argued for the separation of the warmth 



 
 

 
 

 

29 

dimension into discrete evaluations of morality and sociability (Brambilla et al., 2021; Brambilla 

& Leach, 2014; Goodwin, 2015; Leach, 2007). A three-dimensional model better explains the 

primacy of morality in social judgment and accounts for the reality that morality and sociability 

are often independent, such as a gregarious thief or a principled introvert. Since morality, 

competence, and sociability are recognized as the primary dimensions of general person 

perception, they are expected to be important in our understanding of the distinctions people 

make about different types of bad people.  

Generally, people with positive characteristics are inferred to have unrelated positive 

characteristics and people with negative characteristics are inferred to have unrelated negative 

characteristics, i.e., the “halo effect” (Thorndike, 1920). For example, people who violated moral 

norms like cheating or shoplifting were inferred to have low ability, an effect that was driven by 

perceptions of low intelligence (Stellar & Willer, 2018). Thus, I hypothesized that people who 

are described to have prototypical bad person characteristics (e.g., lacks empathy) will also be 

inferred to have lower competence and sociability than people who are described to have 

prototypical good person characteristics. However, I hypothesized that not all basic categories 

are perceived the same, such that prototypical members of basic bad person categories can 

diverge significantly on morality, competence, and sociability from each other, depending on the 

characteristics of the prototype.  

Morality. The moral dimension is composed of lower-level traits like unfair, dishonest, 

unprincipled, untrustworthy, and disloyal (Brambilla et al., 2021; Goodwin, 2015). Perceiving 

the morality of others is important because it provides perceivers with information about whether 

the target intends to be harmful or helpful to the perceiver and others (Fiske et al., 2007). Some 

bad person categories (e.g., a psychopath) are expected to be rated as more immoral than other 



 
 

 
 

 

30 

bad person categories (e.g., a narcissist), based on how dangerous and harmful they are 

perceived to be. 

Ability.  The competence dimension consists of lower-level traits like unintelligent, 

illogical, unorganized, unathletic, uncreative, and unclever (Brambilla et al., 2021; Goodwin, 

2015). Perceiving the competence of others is an important moral concern because it provides 

information about whether the target can fulfill their immoral or moral desires and intentions 

(Fiske et al., 2007). Further, Hartman et al. (2022) argued that immoral agents who are perceived 

to have low ability may be perceived more akin to victims than villains, as they are seen 

primarily as a product of their unfortunate circumstances. Some moral categories (e.g., a con 

artist) may be perceived to be highly intelligent and clever, while other categories may be 

perceived as irrational and ignorant (e.g., a bigot).  

Sociability. The sociability dimension includes lower-level traits like unsociable, 

unhappy, disagreeable, unfunny, and unenthusiastic (Brambilla et al., 2021; Goodwin, 2015). 

Judgments of sociability are considered to provide a similar function to judgments of 

competence because they provide information about the target’s ability to attain friends, allies, or 

supporters that could help them enact their moral or immoral desires or intentions (Landy et al., 

2016). Some members of basic categories (e.g., a psychopath) may be perceived as solitary 

individuals on the fringes of society, while others (e.g., a narcissist, a manipulator) may come 

across as very charming and sociable, which allows them to get close to their targets.  

Essentialist Attitudes 
 

People, to differing extents, view others as having innate, immutable essences that cannot 

be observed directly but can explain their behavior and make them different from other types of 

people (Gelman, 2004; Rhodes & Moty, 2020). In particular, past research has found that people 



 
 

 
 

 

31 

view others as having “moral essences” that are perceived to be biologically driven, stable, 

essential to who a person is, differentiate bad people from good people, and can explain why 

people violate or adhere to moral norms (Heiphetz, 2020; Lynch et al., 2019; Strohminger & 

Nichols, 2014). 

Essentialist attitudes are important to study in the moral domain because they predict 

endorsements of how severely offenders should be punished (Martin et al., 2022; Martin & 

Heiphetz, 2021; Vel-Palumbo et al., 2019). For instance, in cases where the crime is relatively 

minor (e.g., theft, property destruction), individuals described as being biologically predisposed 

to criminal behavior were recommended harsher punishments compared to individuals who 

committed the same crime but were not described as biologically predisposed (Martin & 

Heiphetz, 2021). In another study, people who more strongly endorsed essentialist attitudes 

about criminal behavior were more likely to endorse the death penalty (Martin et al., 2022).  

Although previous researchers have not directly examined whether people hold varying 

essentialist views about different types of bad people, their research has provided indications that 

we should expect to find them. Mainly, some immoral characteristics seem to be more 

essentialized than other immoral characteristics. For instance, Martin and Heiphetz (2021)’s 

manipulation of essentialists attitudes had no effect for major crimes like “murder,” which could 

be explained by an understanding that people already possess highly essentialized attitudes 

regarding murderers. In another study, respondents indicated that it would be a more dramatic 

departure from their friend’s ‘true self’ if they were to suddenly become racist or cruel, 

compared to if they started stealing or cheating on a spouse (Strohminger & Nichols, 2014).  

The discovery that some immoral characteristics are perceived to be more biologically 

driven and stable than others may reflect underlying perceived differences in the types of bad 



 
 

 
 

 

32 

people believed to possess those characteristics. For example, we might expect some types of 

bad people (e.g., a psychopath, a pedophile) to be perceived as having fixed immoral 

characteristics that are a function of biology, while other types of bad people (e.g., the narcissist) 

may be perceived to have traits that are nurtured and mutable.  

Learning which categories are highly regarded as essential will further our understanding 

of how moral judgments of people change or fail to change over time. While members of less 

essentialized categories could be perceived to change over time and potentially even lose 

membership to their category, members of highly essentialized categories may be perceived to 

have a stable categorization. For instance, if the pedophile’s sexual attraction towards children is 

perceived to be biologically determined and fixed, then someone who molests a child may 

always be perceived as a pedophile, even as decades pass without re-offending. Additionally, 

members of highly essentialized categories may be more susceptible to harsher punishments, 

including endorsement for the death penalty, even when the punishment may not be proportional 

to the offense committed. In contrast, people may perceive members of less essentialized 

categories as probable or even likely to mature out of their category, and therefore will interpret 

instances of good behavior as genuine evidence that they no longer belong in the category.   

Social distance 
 

Another important distinction between different types of bad people may be how 

comfortable we are interacting with them. While we may seek to exclude sex offenders from 

living in our neighborhoods and advocate lifelong imprisonment for psychopaths, we may feel 

more comfortable coexisting with other types of bad people. People who display strong moral 

principles are often disparaged by observers who feel morally inferior in comparison (e.g., 

vegans; Minson & Monin, 2012). In contrast, individuals who exhibit less stringent moral 



 
 

 
 

 

33 

standards do not threaten the perceiver’s positive moral self-concept, and thus the observer may 

be more willing to form a relationship with such individuals, as long as they do not pose a 

genuine threat to their physical or mental health. In some cases, people may even benefit from 

relationships with wrongdoers, as they offer a way to reap the benefits of moral transgressions 

without direct involvement in the wrongdoing (Melnikoff & Bailey, 2018). Additionally, we 

often tolerate or in some instances desire narcissistic (Brunell et al., 2008; Grijalva et al., 2015) 

and psychopathic (Uhlmann et al., 2013) traits in leaders. 

Present Study  
 

My dissertation project tested whether there are perceived differences in morality, ability, 

sociability, essentialist beliefs, desired social distance, gender, race, age, and physical 

appearance between exemplars of various bad person categories. By cataloguing the attributes of 

four different prototypes, I established that we have cognitive prototypes in the moral person 

domain and provided preliminary insights that can inspire and inform future research questions 

in the field.  

  



 
 

 
 

 

34 

II. STUDY 1: GENERATING CHARACTERISTICS OF A BAD PERSON 
 

The purpose of Study 1 was to generate a list of prototypical characteristics (e.g., 

behaviors, emotions, beliefs, desires, motives, physical appearance) of a bad person in the same 

relatively homogenous population that will be used to derive (Study 2) and test (Study 3) 

categorical and dimensional models of moral character evaluation. Using a relatively 

homogeneous population was necessary because people from different cultures or subcultures 

are expected to have different conceptions of a bad person that would weaken the tests of moral 

character evaluation in Study 3. For example, if someone does not think that racism or 

homophobia makes someone a bad person, then they may be less likely to infer to that person 

unrelated immoral characteristics than someone who does think racism or homophobia makes 

someone a bad person. A secondary goal of Study 1 was to conduct an exploratory analysis to 

test which individual difference variables (e.g., moral values, conservatism, belief in pure evil) 

can explain differences in conceptions of bad people. 

Method 
 
Participants  
 

University of Oregon students were recruited online from the Psychology/Linguistics 

Human Subjects Pool to take a 30-minute survey and were compensated with course credit. Of 

the 183 participants aged 18 to 70 (M = 20.14, SD = 4.71), 69.3% identified as a woman, 26.8% 

identified as a man, and 3.8% identified as non-binary or “other.” Additionally, 66.1% identified 

as White, 12.6% identified as Asian, 11.5% identified as Hispanic, 3.8% identified as Black, and 

6.1 % specified an ethnicity other than those listed. Finally, the sample skewed liberal, with 

68.3% identifying as extremely or moderately liberal, 25.1% identifying as moderate, and 6.6% 

identifying as extremely or moderately conservative. According to recent demographics provided 



 
 

 
 

 

35 

by the university registrar, the sample generally represented the racial breakdown of the typical 

University of Oregon student but oversampled female students.  

Because I did not conduct an inferential test for this study, I did not run a power analysis 

to determine sample size. I aimed to collect between 100 to 200 participants, which is adequate 

to survey a broad range of students and is standard across studies that use this methodology to 

derive prototypical traits (e.g., Tilburg et al., 2022; Walker & Pitts, 1998). Thirty-five 

participants were removed from the exploratory analysis explaining variance in the conceptions 

of bad people for failing at least one attention check. 

Procedure 
 
In an online study, participants provided descriptions of a bad person with the following 

instructions: 

Instructions: Picture in your mind a "bad person." Think about a real person (not a 

fictional character), ranging from a person you would consider to be somewhat bad to 

one you would consider to be extraordinarily bad. Now think of other people, each of 

whom you would consider to be a "bad person." In the blanks below, describe the 

characteristics of these bad people, such as their behaviors, emotions, beliefs, desires, 

motives, intentions, traits, relationships, abilities, and physical appearance.  

The participants were provided 20 blank spaces to write in descriptors. They were told to 

list as many descriptors as come to mind (one per blank) and that they could click the “add 

characteristic” button to add more. If they did not add at least 10 characteristics, an error 

message appeared that said that they needed to have at least 10 characteristics to continue. On 

the next page, the participants were given a rating matrix table of each characteristic they had 

written and were told to “rate the extent to which you would consider someone a bad person if 



 
 

 
 

 

36 

they expressed the characteristic” on a scale of 1 (definitely not a bad person) to 9 (definitely a 

bad person).  On the following page, they rated their free listed characteristics again but with 

good person replaced with bad person, i.e., on a scale of 1 (definitely not a good person) to 9 

(definitely a good person). 

After the unconstrained listing and rating task, participants were asked to rate 70 

additional characteristics on the two 9-point scales described above. The characteristics were 

drawn from theory and previous research on which characteristics cue negative moral character 

judgments (see Appendix A). I included the 70 additional characteristics to ensure that attributes 

that are perceived to be prototypical of bad people but are not easily accessible (e.g., due to low 

prevalence in the general population) are considered in the creation of the final list of 

prototypical characteristics.   

Finally, in order to conduct an exploratory analysis to examine how conceptions of what 

makes a person a bad person differ across individual difference variables, participants filled out 

a short survey to collect their moral values, conservatism, belief in pure evil, gender, age, 

ethnicity, political orientation, and political affiliation. 

Materials 
 

Moral values were measured with the 36-Item MFQ-2 (Moral Foundations 

Questionnaire-2; Atari et al., 2022). The MFQ-2 measures the extent to which an individual 

endorses moral principles related to harm (e.g., “caring for people who have suffered is an 

important virtue”), equality (e.g., “the world would be a better place if everyone made the same 

amount of money”), proportionality (e.g., “I think people who are more hard-working should end 

up with more money”), loyalty (e.g., “I think children should be taught to be loyal to their 

country”), authority (e.g., “I think obedience to parents is an important virtue”), and purity (e.g., 



 
 

 
 

 

37 

“I think the human body should be treated like a temple, housing something sacred within”) on a 

scale of  1 (does not describe me at all) to 5 (describes me extremely well). People who score 

high in the proportionality, loyalty, authority, and purity domains diverge from the typical liberal 

undergraduate student and may therefore be expected to have different conceptualizations of a 

bad person (Graham et al., 2009). 

The extent to which an individual believes that some people are evil was measured with 

the 22-Item BPE or Belief in Pure Evil Scale (Webster & Saucier, 2013). Participants rated items 

such as “evil people have an evil essence, like a stain on their souls, which is almost impossible 

to get rid of” on a scale of 1 (Disagree Very Strongly) to 7 (Agree Very Strongly). People who 

score low on this scale may be less likely to condemn others for characteristics that have not 

caused harm (e.g., having a belief but not acting on it). 

Conservatism was measured with the 12-Item Social and Economic Conservatism Scale 

(SECS; Everett, 2013). The SECS measures the extent to which an individual feels positively or 

negatively about different social (e.g., religion) and economic (e.g., business) issues on a 100-

point scale, where higher scores are considered more conservative. People who score high on this 

scale are expected to have diverging views from the typical liberal undergraduate student on 

which behaviors, cognitions, and affect make an individual a bad person (e.g., Graham et al., 

2009).  

Results 
 

The primary goal of this data analysis was to create a non-redundant list of the most 

prototypical characteristics of a bad person from the unconstrained listing and rating task and the 

70-item rating task. A secondary objective was to conduct exploratory analyses to test whether 



 
 

 
 

 

38 

people have different conceptions of a bad person based on their moral values, political 

orientation, and beliefs regarding the existence of pure evil.    

Generating a List of the Prototypical Characteristics of a Bad Person 
 

On average, participants freely listed 15.56 (SD = 4.74) descriptors, for a total of 1211 

unique descriptors. A team of four researchers (three undergraduate students and the principal 

investigator) reduced the list based on the following procedural rules for simplifying prototypical 

characteristics adapted from Walker and Pitts (1998). Disagreements over which characteristics 

should be considered synonymous were resolved through lengthy discussion over a period of 

several weeks. 

1. Drop modifiers (e.g., completely lacks empathy turns into lacks empathy). 

2. Separate compound phrases (e.g., lacks empathy and is manipulative turns into two 

separate characteristics, lacks empathy and manipulative).  

3. Collapse words and phrases that have synonymous meanings in common use into one 

(e.g., lacks empathy and lacks sympathy became lacks empathy). 

4. Drop idiosyncratic descriptors that cannot be combined with another category (e.g., 

Arkansas native), where fewer than 5 participants used the descriptor. 

After removing idiosyncratic descriptors, 103 distinct descriptors remained. To determine 

the final 50 characteristics, I compiled three lists: the top 50 freely recalled descriptors rated by 

frequency (Table 1), the top 50 freely recalled descriptors rated by mean badness rating, and the 

top 50 closed list descriptors by mean badness rating (Table 2). To match the closed list 

descriptors to the freely recalled descriptors, three coders independently matched the closed list 

items with the free listed items and disagreements were resolved through discussion.  



 
 

 
 

 

39 

Forty-four characteristics were on at least two lists and were therefore prioritized to be on 

the final list. Five of the forty-four items were removed for being too multifaceted or too general 

(e.g., narcissistic, bigoted, toxic, cruel, and harms others). Next, based on high rankings across 

multiple lists, abuse was split into domestic abuse and child abuse, sadism was split into animal 

sadism, sexual sadism, and general sadism, and child predator was split into child molestation 

and sexual attraction towards children. The final seven items were added because they were rated 

extremely high on a single list (i.e., necrophilia, cannibalism, scamming) or rated close to 50 on 

more than one list (e.g., exploitative, jealousy, substance abuse, psychotic). None of the final 

characteristics were rated as strongly indicative of both a good person and a bad person.  

To write items for the final list, I used pre-written items from the current study when 

available; otherwise, new items were created based on discussions with the research team about 

the meaning of each freely recalled characteristic. A few items were slightly reworded after 

discussion with the research team raised concerns that they could be misinterpreted. The final list 

of items can be found in Appendix B.  

Explaining Variance in Conceptions of a Bad Person 
 
 In an exploratory analysis, I tested whether an individual’s moral values, conservatism, 

and belief in pure evil are related to their conception of a bad person (Table 3). Overall, 

conservatives were less likely to endorse that having racist, transphobic, sexist, or xenophobic 

beliefs makes someone a bad person, but were more likely to endorse that being an atheist, 

unpatriotic, being addicted to substances, trading sex for money, and committing petty crimes 

makes someone a bad person. These findings suggest that conservatives and liberals may have 

different moral person category structures. For instance, a bigot may be more likely to emerge in 



 
 

 
 

 

40 

a liberal sample, whereas a welfare queen may be more likely to emerge in a conservative 

sample.   

Belief in Pure Evil (BPE) positively predicted nearly all items where the person fantasized about 

but did not act on their immoral desires (e.g., rape, cannibalism, necrophilia, pedophilia, and 

infidelity), suggesting an individual difference in the perception that a person’s badness can be 

dictated by their desires alone, even in the absence of evidence that the person has or will act on 

those desires. Further, social conservatism, r(146) = .30, p <.001 and economic conservativism, 

r(146) =.19, p =.023, were both significantly related to BPE, which suggests that conservatives 

may be more inclined to judge a person’s moral character based on their desires alone. 

Discussion 
 
 In Study 1, participants listed and rated characteristics that are prototypical of a bad 

person.  The freely listed descriptors were polished by dropping modifiers, separating compound 

phrases, collapsing synonyms, dropping idiosyncratic responses (n < 5), and resolving 

discrepancies between coders. Then, the freely listed descriptors were compared to the closed-list 

ratings to generate a final list of 50 prototypical characteristics of a bad person. These 50 

characteristics were used in the next study to develop a taxonomy of the bad person concept.  

Exploratory analyses revealed that conservatives and liberals differed in their judgments 

of what constitutes a bad person, particularly for judgments concerning racism, substance abuse, 

and patriotism. Additionally, differences in moral values and the belief that some possess an evil 

essence also influenced which characteristics were considered prototypical of a bad person. 

These findings support the idea that there will be differences between population subgroups in 

their conceptions of bad person prototypes, especially across the political spectrum. Thus, as 



 
 

 
 

 

41 

suspected, investigations of moral person prototypes will have to be made within relatively 

homogeneous subgroups.  

  



 
 

 
 

 

42 

Table 1 

Bad Person Characteristics from Unconstrained Listing Ranked by Frequency 
 
  Frequency Badness Ratings 
Rank Characteristic N Prop M SD 
1 Rude 130 71.0% 6.83 1.25 
2 Lying 117 63.9% 6.68 1.36 
3 Selfishness 96 52.5% 6.63 1.29 
4 Mean 91 49.7% 6.81 1.43 
5 Manipulative 82 44.8% 7.59 1.05 
6 Narcissistic 65 35.5% 7.09 1.50 
7 Easy to anger 64 35.0% 5.86 1.40 
8 Egotistical 63 34.4% 6.38 1.33 
9 Physically violent 50 27.3% 7.92 1.07 
10 Abusive 47 25.7% 8.72 0.62 
11 Racist 44 24.0% 8.55 1.30 
12 Self-centered 42 23.0% 6.71 1.11 
13 Aggressive 42 23.0% 7.40 1.17 
14 Lacks empathy 40 21.9% 7.15 1.29 
15 Judgmental 38 20.8% 6.50 1.29 
16 Harmful 36 19.7% 7.86 1.44 
17 Sexist 36 19.7% 8.42 1.40 
18 Greedy 35 19.1% 6.94 1.33 
19 Hateful 35 19.1% 7.57 1.29 
20 Controlling 34 18.6% 6.85 1.33 
21 Careless 33 18.0% 6.15 1.54 
22 Cruel 33 18.0% 8.00 1.60 
23 Bullying 32 17.5% 7.47 1.19 
24 Unemotional 29 15.8% 6.62 1.97 
25 Ignorant 29 15.8% 5.69 1.20 
26 Negative 28 15.3% 5.96 1.73 
27 Inauthentic 28 15.3% 5.96 1.40 
28 Jealous 28 15.3% 6.50 1.43 
29 Distant 26 14.2% 6.23 2.20 
30 Close-minded 26 14.2% 6.58 1.60 
31 Infidelity 25 13.7% 7.36 1.85 
32 Spitefulness 24 13.1% 6.71 1.27 
33 Homophobic 24 13.1% 8.17 1.79 
34 Stealing 21 11.5% 7.52 1.60 
35 Bigoted 21 11.5% 8.14 1.06 
36 Freeloader 21 11.5% 7.90 1.14 
37 Untrustworthy 20 10.9% 6.10 1.21 
38 Entitled 20 10.9% 6.55 1.05 
39 Does not care about others 19 10.4% 6.74 1.66 



 
 

 
 

 

43 

  Frequency Badness Ratings 
Rank Characteristic N Prop M SD 
40 Disloyal 19 10.4% 6.83 1.25 
41 Toxic 18 9.8% 7.33 1.57 
42 Psychotic 18 9.8% 6.22 1.48 
43 Power hungry 18 9.8% 7.22 1.52 
44 Two-faced 17 9.3% 7.88 0.99 
45 Few relationships 17 9.3% 5.24 1.68 
46 Annoying 17 9.3% 5.88 1.54 
47 Deceitful 16 8.7% 7.13 1.36 
48 Unforgiving 16 8.7% 6.44 1.09 
49 Murder 16 8.7% 8.75 0.58 
50 Loud 16 8.7% 4.31 2.02 

 
 
  



 
 

 
 

 

44 

Table 2 
 
Bad Person Characteristics from Unconstrained Listing Ranked by Mean Badness 
 
Rank Characteristic Badness ratings Frequency 
  M SD N Prop 
1 Child predator  9.00 0.00 7 3.8% 
2 Transphobic 8.77 0.60 13 7.1% 
3 Murder 8.75 0.58 16 8.7% 
4 Rape 8.73 0.59 15 8.2% 
5 Abusive 8.72 0.62 47 25.7% 
6 Sadistic 8.69 0.63 13 7.1% 
7 Racist 8.55 1.30 44 24.0% 
8 Sexist 8.42 1.40 36 19.7% 
9 Bad intentions 8.40 0.55 5 2.7% 
10 Homophobic 8.17 1.79 24 13.1% 
11 Bigot 8.14 1.06 21 11.5% 
12 Cruel 8.00 1.60 33 18.0% 
13 Physically violent 7.92 1.07 50 27.3% 
14 Freeloader 7.90 1.14 21 11.5% 
15 Two-faced 7.88 0.99 17 9.3% 
16 Harmful 7.86 1.44 36 19.7% 
17 Opposing political beliefs 7.85 0.99 13 7.1% 
18 Lawbreaker 7.83 1.75 12 6.6% 
19 Superiority complex 7.80 1.40 10 5.5% 
20 Emotional abuse  7.80 1.30 5 2.7% 
21 No guilt or remorse 7.80 1.10 5 2.7% 
22 Xenophobic 7.67 3.27 6 3.3% 
23 Gaslighting 7.63 1.19 8 4.4% 
24 Manipulation 7.59 1.05 82 44.8% 
25 Hateful 7.57 1.29 35 19.1% 
26 Stealing 7.52 1.60 21 11.5% 
27 Gossiping 7.50 1.08 10 5.5% 
28 Dominance 7.50 1.38 6 3.3% 
29 Bullying 7.47 1.19 32 17.5% 
30 Aggressive 7.40 1.17 42 23.0% 
31 Religious intolerance  7.40 3.58 5 2.7% 
32 Possessive 7.40 0.89 5 2.7% 
33 Cheats 7.36 1.85 25 13.7% 
34 Toxic 7.33 1.57 18 9.8% 
35 Condescending 7.29 1.44 14 7.7% 
36 Humiliates others 7.25 1.16 8 4.4% 
37 No accountability 7.25 1.22 12 6.6% 
38 Power hungry 7.22 1.52 18 9.8% 
39 Lacks empathy 7.15 1.29 40 21.9% 



 
 

 
 

 

45 

Rank Characteristic Badness ratings Frequency 
  M SD N Prop 
40 Intrusive 7.14 1.46 7 3.8% 
41 Deceitful 7.13 1.36 16 8.7% 
42 Narcissistic 7.09 1.50 65 35.5% 
43 Troubled Upbringing 7.00 2.00 6 3.3% 
44 Greedy 6.94 1.33 35 19.1% 
45 Disloyal 6.89 1.82 19 10.4% 
46 Controlling 6.85 1.33 34 18.6% 
47 Rude 6.83 1.25 130 71.0% 
48 Mean 6.81 1.43 91 49.7% 
49 Irrational 6.80 0.84 5 2.7% 
50 Does not care about others 6.74 1.66 19 10.4% 

 
 
  



 
 

 
 

 

46 

Table 3 
Descriptives and Zero-Order Correlations Between Individual Variables and Badness Ratings of Bad Person Characteristics  
 

 Badness Moral Foundations Conservatism  

Item M SD Care Equality Prop Loyalty Authority Purity Social Economic BPE 

Child molestation 8.99 .12 .15 .05 .05 .13 .12 .08 -.02 .11 .09 
Murder 8.97 .16 .16* .10 .09 .10 .07 .09 -.15 .01 .05 
Rape 8.89 .39 .25** .29*** .03 -.04 -.05 -.04 -.14 -.22** .06 
Child abuse 8.87 .43 .26** .18* -.04 .05 .05 .03 -.06 -.14 .01 
Partner abuse 8.69 .63 .30*** .21** .04 -.01 .01 .02 -.10 -.15 .11 
Hitman 8.67 .79 .22** .01 .00 .11 .06 .15 -.02 -.05 .13 
Animal sadism 8.56 1.06 .07 .31*** -.13 -.12 -.13 -.12 -.19* -.28*** .01 
Rape fantasies 8.52 1.03 .20* .23** -.01 .08 .05 .07 .09 -.05 .17* 
Cannibalism 8.51 1.05 .01 .00 .10 -.01 .09 .06 .08 .08 .21* 
Sadism 8.47 .88 .31*** .29*** .02 .03 -.04 .00 -.16 -.20* .11 
Homophobic 
behavior 8.47 .84 .25** .27*** -.01 -.11 -.16 -.12 -.30*** -.17* .00 

Racist behavior 8.45 .95 .23** .31*** .00 -.11 -.13 -.15 -.30*** -.17* -.04 
Racist beliefs 8.40 1.04 .15 .23** -.08 -.29*** -.30*** -.30*** -.45*** -.18* .07 
Scammer 8.30 1.02 .14 .07 .09 .08 -.05 .02 -.12 -.09 .14 
Necrophilia 8.24 1.28 .18* .20* .07 .06 -.04 .03 -.07 -.04 .17* 
Sexist behavior 8.18 1.18 .16 .24** -.07 -.20* -.22** -.32*** -.43*** -.09 -.03 
No guilt 8.14 1.13 .25** 0.18* .03 .00 -.06 -.03 -.15 -.16 .26* 
Homophobic 
beliefs 8.11 1.29 .19* .35*** -.08 -.19* -.22** -.34*** -.47*** -.20* .00 

Terrorism 8.11 1.20 .14 .03 .01 .05 .06 -.06 .03 .05 .18 
Transphobic 
beliefs 8.09 1.28 .26** .43*** -.19* -.27*** -.31*** -.34*** -.51*** -.38*** .05 

Pedophilia 8.09 1.36 .14 .11 .11 .06 .05 .06 .08 .01 .29*** 
Sexist beliefs 8.03 1.26 .35*** .37*** -.04 -.08 -.16* -.20* -.41*** -.31*** .03 



 
 

 
 

 

47 

 Badness Moral Foundations Conservatism  

Item M SD Care Equality Prop Loyalty Authority Purity Social Economic BPE 

Exploitative 8.01 1.09 .23** .27*** -.10 .00 -.09 -.09 -.19* -.26** -.02 
Transphobic 
behavior 7.97 1.46 .31*** .42*** -.15 -.26** -.30*** -.33*** -.48*** -.40*** .01 

Stalking 7.95 1.11 .18* .20* -.06 -.10 -.19* -.13 -.25** -.15 .04 
Bullying 7.90 1.00 .27*** .22** .11 .03 -.01 -.04 -.09 -.13 .17* 
Money seeking 7.89 1.41 .18* .30*** -.16* -.12 -.29*** -.15 -.32*** -.23** .01 
Infidelity 7.71 1.24 .20* .16* .10 .13 .07 -.03 -.06 -.09 .01 
Controlling 7.65 1.11 .36*** .11 .03 .12 .09 -.02 -.09 -.14 .07 
Imperialism 7.55 1.46 .37*** .45*** -.12 -.08 -.18* -.09 -.30*** -.30*** -.03 
Xenophobia 7.49 1.45 .29*** .48*** -.11 -.22** -.29*** -.20* -.44*** -.35*** .01 
Entitled 7.49 1.35 .24** .33*** -.16 -.13 -.21** -.21* -.45*** -.22** .05 
No empathy 7.40 1.45 .18* .17* -.21* -.02 -.05 .00 -.24** -.17* .16 
Sexual sadism 7.39 1.66 .09 .10 .02 .08 .08 .02 -.03 .00 .13 
Strategic lying 7.16 1.40 .21* -.06 .06 .18* .13 .10 .04 -.03 .19* 
No accountability 7.14 1.32 .18* .04 -.01 .00 -.06 -.11 -.18* -.13 .05 
Judgmental 7.01 1.25 .25** .13 .00 .06 -.02 -.02 -.13 -.17* .19* 
Property damage 6.98 1.26 .09 -.09 .16 .20* .22** .21* .20* .02 .31*** 
Infidelity 
fantasies 6.97 1.52 .12 -.01 .13 .11 .17* .14 .14 .00 .18* 

Spiteful 6.90 1.29 .23** .09 .13 .21* .17* .21* .11 .02 .12 
Disloyal 6.90 1.32 .19* -.06 .14 .24** .18* .06 .11 .04 .11 
Only cares about 
themselves. 6.85 1.35 .27*** .18* -.05 .09 -.10 -.05 -.20* -.16* .10 

Hypocrisy 6.84 1.32 .14 .06 .04 .17* .02 -.09 -.08 .11 .02 
Harms 
environment 6.81 1.57 .30*** .27** .00 .00 -.13 -.06 -.19* -.22** .10 

Pathological  
lying 6.75 1.49 .04 -.14 .09 .12 .08 .16 .14 .06 .35*** 



 
 

 
 

 

48 

 Badness Moral Foundations Conservatism  

Item M SD Care Equality Prop Loyalty Authority Purity Social Economic BPE 

Arrogance 6.66 1.41 .22** .13 .08 .09 -.04 -.05 -.08 -.12 .10 
Selfish 6.65 1.47 .12 -.05 .06 .18* .18* .17* .09 -.04 .22** 
Capital 
punishment 6.63 1.73 .11 .06 -.10 .06 .00 .05 -.04 -.08 .06 

Lawbreaker 6.62 1.52 .06 -.26** .11 .24** .22** .11 .25** .07 .21* 
Ideological 
violence 6.47 1.63 .17* .09 -.06 .08 .03 .05 -.02 -.10 .18* 

Greedy 6.41 1.46 .18* .13 -.09 -.06 -.09 -.15 -.20* -.12 .06 
Power seeking 6.26 1.61 .24** .17* -.16 -.01 -.04 -.06 -.22** -.15 .11 
No gratitude 6.12 1.45 .15 -.05 .01 .23** .13 .04 -.02 .02 .25** 
Unreliable 6.11 1.34 .08 .09 .07 .08 .13 -.09 -.10 .03 .11 
Pyromania 6.05 1.69 .07 -.15 .08 .23** .21** .27*** .12 .05 .22** 
Gossips 5.95 1.47 .05 -.15 .24** .28*** .10 .17* .21* .09 .23** 
Kleptomania 5.95 1.67 -.04 -.32*** .21* .28*** .32*** .38*** .31*** .19* .30*** 
Impulsive 5.70 1.44 -.01 -.11 .08 .05 .09 .10 -.02 .03 .08 
Conformity 5.55 1.54 .09 -.03 .00 .03 -.04 -.06 -.06 .02 .14 
Reclusive 5.49 1.97 .09 -.06 -.05 .19* .01 -.03 -.03 -.02 .16 
Temper 5.44 1.53 -.03 -.11 -.02 -.04 -.02 .00 -.11 .00 .09 
Bragging 5.43 1.40 .15 -.05 .02 .09 .04 -.03 -.15 -.05 .14 
Nationalism 5.32 1.90 .20* .32*** -.13 -.17* -.33*** -.21* -.38*** -.21* .04 
Attention seeking 5.12 1.44 .14 -.06 .01 .02 -.04 -.12 -.12 .00 .11 
Loses friends 5.11 1.53 .01 -.18* -.12 -.01 .04 -.04 -.03 -.05 .13 
Sexual fetishes 5.07 1.87 -.01 -.06 -.04 .17* .12 .21** .03 .11 .23** 
Prostitution 4.99 2.06 .04 -.27** .22** .33*** .36*** .35*** .30*** .28*** .16 
Substance 
addiction 4.47 1.69 -.13 -.25** .05 .16* .26** .31*** .24** .24** .09 

Unpatriotic 3.56 1.59 -.14 -.39*** .10 .31*** .29*** .26** .33*** .30*** .28*** 
Atheism 2.81 1.75 -.14 -.29*** .04 .26** .23** .27*** .34*** .24** .14 



 
 

 
 

 

49 

III. STUDY 2: EMPRICALLY DRIVEN CATEGORY AND DIMENSIONAL MODELS OF 

BAD PERSON CONCEPTS 

 
To learn whether people distinguish between different types of bad people, Study 2 

utilized a similarity-sorting task to derive a taxonomy of bad person concepts. Previous research 

had empirically derived a moral person taxonomy based on behaviors alone and found three 

categories (Landy & Bartels, 2018). The present study expanded on this approach by including 

(alongside behaviors) perceived beliefs, desires, and emotions of the targets, based on the 

characteristics provided in Study 1. To identify candidates for the basic categories of bad people 

in the given population and determine which traits were perceived to be most central to each 

category, I conducted multiple types of clustering analysis. To reduce and interpret the 

dimensions people use to distinguish immoral person characteristics from each other, I 

conducted a dimensional reduction analysis. The clustering and dimensional reduction analyses 

generated categorical and dimensional models of moral characteristics that were then tested in 

the final study. 

Method 

Participants 

  One-hundred and thirty-one undergraduate students were recruited from the University of 

Oregon Psychology/Linguistics Human Subjects pool for a 1-hour in-person study and were 

compensated with course credit. Ten participants were removed from the sample for taking the 

survey in less than 17 minutes, which was determined to be too quick to read every item. The 

sample size was determined a priori based on the recommendation of at least 20 participants per 

expected cluster, though a smaller sample is needed for more separate clusters and for the fuzzy 



 
 

 
 

 

50 

clustering algorithm (Dalmaijer et al., 2022). Past research using this method has used a 

comparable sample size or smaller (Brewer et al., 1981; Pilkonis, 1988; Walker & Pitts, 1998). 

The sample demographics were similar to Study 1. Students ranged from 18 to 28 (M = 

19.00, SD = 1.39) and were majority white, with 60.3% reporting as White, 17.6% reporting as 

Hispanic, 12.4% reporting as Asian, 7.4% reporting as Black, and 2.3% reporting as biracial or 

“other.” Additionally, 65.3% of participants reported that they identified as a woman and 33.1% 

reported that they identified as a man, and 1.6% identified as nonbinary or “other.” The sample 

leaned liberal, with about 61.2% reporting as politically liberal and 5.8% reporting as politically 

conservative. 

Procedure and Materials 
 

Participants completed a computerized similarity-sorting task on large monitors in a 

computer lab5. When the participants arrived at the laboratory in small groups, the experimenter 

demonstrated the sorting task with physical cards and described the task. Participants were told 

that the task measured how many types of people they perceive, and that they would be sorting 

negative characteristics into piles, such that each pile represented a different type of person. The 

experimenter refrained from describing the traits using moral descriptors (e.g., bad, evil, 

 
5 The similarity-sorting task was computerized by using Pick, Group, & Rank questions 

on Qualtrics, with custom javascript to save the participants’ categorization solution to 

embedded survey data and to hide groups until participants clicked on an ‘add a pile’ button so 

that participants could choose how many groups to make. Custom css script enhanced the user 

experience of the task and emulated the experience of the original task with physical cards. 

 



 
 

 
 

 

51 

immoral) as to not bias the participants’ responses. Participants were directed to make as many 

piles as they saw fit, with as many characteristics per pile as they saw fit. 

Participants first completed a practice trial to adjust to the task where they sorted 10 

animal characteristics (e.g., excellent swimmers, four limbs, can fly) into piles “based off whether 

[they] think that the characteristics would normally show up in the same type of animal.” Once 

they sorted all the characteristics, they were given an example of a three-category solution and 

told that they may have had more or fewer categories, depending on how they defined their 

categories. 

On the next page, participants read a short description of the study that repeated the 

researcher’s verbal instructions to motivate them to complete the very cognitively demanding 

task and orient them to the main study. On the following page, the screen displayed a pile of 50 

blue rectangles representing index cards with negative moral person attributes printed on them 

(generated in Study 1), one gray box labeled “Pile 1,” and a button that said, “add a pile” (Figure 

1). The blue cards were “shuffled” and laid on top of each other such that each participant saw 

the cards in random order and could drag them into piles one at a time. The computers were large 

enough that participants could create at least 20 piles comfortably without needing to scroll. If a 

pile had more items than could be viewed, participants were directed to hover over the box and 

scroll to review the items. When all cards were sorted, they were able to click to the next page 

after agreeing that they had reviewed their piles, and they felt confident that each represented a 

different type of person they perceive. 

 

 
  



 
 

 
 

 

52 

Figure 1 
 
Computerized Similarity-Sorting Task 
 

 
 

Next, participants were shown a list of the characteristics they chose for their first pile 

and were asked if they could further divide the pile into two or more piles, such that each pile 

represented a type of person. If they indicated yes, they were displayed a second screen that 

allowed them to do a second sort, such that the characteristics from the first pile were located to 

the left of the screen on blue cards, and participants could create as many new piles as they saw 

fit and put as many cards in each new pile as they saw fit. Then, on a new page, participants were 

displayed a screen with their Pile 1 characteristics printed and the fill-in-the blank question 

“What would you call someone who had all of these characteristics?” On the following page, 

participants were asked to generate new characteristics for their pile. This task was repeated until 

all of their piles (if desired) were further divided into more piles.  



 
 

 
 

 

53 

Then, a page was displayed to participants that printed their initial piles, including the 

contents and the name they gave each pile. They were asked to consider whether any of their 

original piles could be combined to create a bigger pile that represents a type of person. Then, 

they completed a third sorting task with the names of their groups on blue cards, with the ability 

to create as many piles as they saw fit, with as many initial groups per pile as they saw fit. 

Finally, participants rated all 50 characteristics based on “the extent to which you would 

consider someone a bad person if they expressed the characteristic” on a scale that ranged from 1 

(definitely not a bad person) to 9 (definitely a bad person) and answered standard demographics.  

 
Results 
 
Number of Groups 
 
 On average, participants created 6.74 (SD = 2.84) piles in the initial sort, with a minimum 

of 2 piles and a maximum of 18 piles. The distribution was skewed to the right, suggesting that 

the typical person initially distinguished between 5 to 7 types of people from the characteristics 

given (Figure 2). Further, after giving participants the opportunity to further divide their initial 

piles, they created on average of 9.26 (SD = 4.52) groupings, with a minimum of 3 piles and a 

maximum of 23 piles. When given the opportunity to combine their initial piles, they created on 

average 3.55 (SD = 1.42) groupings, from a minimum of 1 pile to a maximum of 8 piles. 

Group Names 
 
 For each initial group, participants were prompted to generate a label for the pile based 

on what they would call someone who had all characteristics listed. The names were cleaned by 

removing any modifiers or articles (e.g., “a very evil psychopath” turned into “psychopath”) and 

correcting spelling errors. If a participant wrote several names (e.g., “a racist, homophobe, and 



 
 

 
 

 

54 

transphobe” or “a narcissistic psychopath”) or gave a paragraph describing the person, the name 

was counted as a unique entry. Overall, the most common category names were “narcissist,” 

“psychopath,” and “abuser,” and each were used by about a third of the participants (Table 4).  

Figure 2 

Distribution of Groups Created in Initial Sort 

 

Similarity Matrix 
 

Next, I created a 50 x 50 triangular similarity matrix where each element represents the 

number of participants who grouped two characteristics in the same pile, such that high numbers 

indicated that two characteristics were frequently in the same pile and 0 indicated that two 

characteristics were never put in the same pile. I also created a total similarity matrix that sums 

the frequencies from the initial and secondary sorts, removing any redundant entries (i.e., 



 
 

 
 

 

55 

participants who chose not to split or sort a pile). Next, I created a distance matrix from the 

similarity matrices using the `dist` function from the `stats` package in R, where distance 

between each item pair was modeled by Euclidean distance. 

I ran the analyses with both the distance matrix from the initial sort and distance matrix 

from the total sort. However, the results were extremely similar, and therefore I am only 

reporting the results using the similarity matrix from the initial sort. 

Table 4 
Most Frequent Labels Given to Describe Initial Groupings 
 

Label Frequency Percent 
Narcissist 41 33.9% 
Psychopath 37 30.6% 
Abuser 36 29.8% 
Addict 20 16.5% 
Conservative 20 16.5% 
Psychotic 20 16.5% 
Selfish 15 12.4% 
Pedophile 14 11.6% 
Manipulator 12 9.9% 
Bigot 9 7.4% 
Cannibal 9 7.4% 
Bully 8 6.6% 
Criminal 8 6.6% 
Greedy 8 6.6% 
Sociopath 8 6.6% 
Dictator 7 5.8% 
Insecure 7 5.8% 
Mentally ill 7 5.8% 
Predator 7 5.8% 
Creep 6 5.0% 
Entitled 6 5.0% 
Self-centered 6 5.0% 
Liar 5 4.1% 
Racist 5 4.1% 
Sadist 5 4.1% 

 
  



 
 

 
 

 

56 

Deriving a Category Model of Bad Person Concepts 
 
 Hierarchical Cluster Analysis. Agglomerative hierarchical cluster analysis clusters 

items into groups by identifying items that are close together based on the imputed distance 

matrix, and then merges the nearest items iteratively until all items are part of the same cluster. 

The analysis results in a hierarchical structure of groups that can be modeled by a dendrogram, 

such that the y-axis represents the Euclidean distance between clusters.  

For the hierarchical cluster analysis, I used the `hclust` function from the `stats` package 

in R and used Ward’s minimum variance method to link clusters. Ward’s method prioritizes 

minimizing the within-cluster variance when merging two clusters together, which results in 

more compact clusters compared to other popular methods like single, complete, or average 

linkage. The `agnes` function from the `cluster` package calculated that the Ward’s method had 

the highest agglomerative coefficient for the data, which indicates that it produced the strongest 

clustering structure (i.e., within-cluster similarity is high and between-cluster similarity is low) 

compared to the single, complete, and average linkage methods.  

 Next, to determine the optimal number of clusters, I used the `fviz_nbclust` function from 

the `factoextra` package in R to model the within-cluster variance (i.e., WSS) and between-

cluster variance (i.e., silhouette width). The within-cluster sum of squares measures how 

compact each cluster is and can be visualized by an elbow plot. The plot demonstrated that there 

are three tightly knit clusters, and that additional clusters do not provide a dramatic reduction in 

variance (Figure 3). The between-cluster variance can be modeled by average silhouette width, 

which is calculated by averaging the silhouette width for each item. Silhouette width values that 

are close to one indicate that the item is well suited for its own cluster and is far from 

neighboring clusters. The plot of average silhouette width by cluster agreed that three was the 



 
 

 
 

 

57 

optimal number of clusters, such that additional clusters decreased the between-cluster variance 

(Figure 4). 

Figure 3 
 
Elbow Plot of Within-Cluster Variance for Initial Sort 

 

Figure 4 

Plot of Average Silhouette Width for Initial Sort 

  



 
 

 
 

 

58 

Even though the three-cluster solution