Anthropomorphism of the Object (ANTH)

Abstract

The Anthropomorphism of the Object (ANTH) scale is a concise, psychometrically validated seven-item instrument designed to assess the degree to which individuals perceive non-human entities—such as conversational artificial intelligence (AI) agents, chatbots, service robots, digital avatars, and virtual assistants—as possessing human-like attributes, mental capacities, and autonomous agency. Originally operationalized by Cammy Crolic, Felipe Thomaz, Rhonda Hadi, and Andrew T. Stephen (2022) in their foundational study on customer-chatbot interactions in the Journal of Marketing, the instrument addresses a critical measurement gap in human-computer interaction (HCI), consumer psychology, and social robotics. The scale conceptualizes anthropomorphism as a continuous, multidimensional perceptual phenomenon that converges into a robust, parsimonious, unidimensional measurement model. The seven items measure core facets of humanization: perceived vitality or aliveness, human-like qualitative properties, personhood, human experiential feel, personality, independent mentality, and intentional action. Responses are captured using a standard seven-point Likert scale ranging from 1 (“Strongly disagree”) to 7 (“Strongly agree”). Across multiple empirical investigations, the scale has exhibited exceptional psychometric properties, consistently demonstrating high internal consistency (Cronbach’s α typically exceeding .92; McDonald’s ω > .92), rigorous factor structure stability via exploratory and confirmatory factor analyses, and robust convergent, discriminant, and criterion-related predictive validity. This paper provides an exhaustive review of the scale’s theoretical foundations, structural and psychometric properties, validation evidence, scoring mechanics, and research applications.

Keywords

Anthropomorphism, Conversational Agents, Mind Perception, Artificial Intelligence, Human-Robot Interaction, CASA Paradigm, Chatbots, Social Cognition, Consumer Psychology, Psychometrics, Intentionality, Agency

Authors

The scale was developed and validated by a team of leading scholars in marketing, consumer behavior, and digital technology:

• Cammy Crolic — Associate Professor of Marketing, Saïd Business School, University of Oxford. Her research focuses on consumer sensory perception, human-technology interaction, and digital marketing.
• Felipe Thomaz — Associate Professor of Marketing, Saïd Business School, University of Oxford. His work concentrates on digital marketing ecosystems, social media dynamics, and consumer network analysis.
• Rhonda Hadi — Associate Professor of Marketing, Saïd Business School, University of Oxford. Her research investigates the psychological consequences of sensory and technological stimuli, focusing specifically on consumer-technology relationships, embodied interfaces, and humanization.
• Andrew T. Stephen — L’Oréal Professor of Marketing and Vice Dean for Research, Saïd Business School, University of Oxford. His scholarship addresses digital transformation, marketing technology (MarTech), artificial intelligence in customer service, and social commerce.

Purpose

The primary purpose of the Anthropomorphism of the Object (ANTH) scale is to provide a standardized, psychometrically sound, and methodologically efficient instrument for evaluating the extent to which human users project human traits onto technological artifacts and service agents. In modern service environments, organizational touchpoints have shifted from human-to-human interactions to interactions mediated by algorithmic systems, chatbots, digital humans, and autonomous service robots. Consequently, understanding how consumers mentally represent these non-human interactants has become an indispensable objective for researchers and practitioners.

Historically, researchers relied on fragmented, ad-hoc, or overly complex inventories to measure anthropomorphism. Some traditional scales, such as the Godspeed series (Bartneck et al., 2009), emphasize semantic differential items evaluating physical appearance, movement, or basic animacy, which can prove cumbersome or conceptually misaligned when assessing purely textual or screen-based interfaces like conversational agents. Other instruments conflate the fundamental perception of humanness with downstream evaluative judgments such as perceived warmth, competence, or social presence. The ANTH scale was explicitly engineered to overcome these limitations by establishing a unified, seven-item metric that taps directly into the psychological core of anthropomorphism: the simultaneous attribution of life-like qualities, subjective humanness, mental agency, and goal-directed intentionality.

In applied and empirical contexts, the ANTH scale enables researchers to assess:

• Customer Service Dynamics: How anthropomorphic design cues (e.g., human names, conversational fillers, avatars, empathetic linguistic framing) alter consumer expectations, tolerance for service failure, and emotional reactivity. Notably, Crolic et al. (2022) revealed an anthropomorphism “double-edged sword,” whereby humanizing a chatbot amplifies consumer anger and blame attribution when service delivery fails.
• Human-Computer Interaction (HCI) and Social Robotics: The extent to which embodied or disembodied artificial agents elicit social responses from users, validating the degree of psychological engagement fostered by specific conversational UI/UX designs.
• Ethical and Regulatory Dimensions of AI: Monitoring how deceptive or hyper-realistic anthropomorphism affects consumer vulnerability, over-trust, cognitive manipulation, and privacy compliance in digital marketplaces.
• Experimental Manipulation Checks: Serving as an efficient, sensitive manipulation check in laboratory and field experiments examining human-AI interaction.

Psychological Construct

The psychological construct assessed by the ANTH scale is perceived anthropomorphism, defined conceptually as the cognitive tendency to imbue the real or imagined behaviors of non-human entities with human-like characteristics, motivations, intentions, emotions, or psychological states (Epley, Waytz, & Cacioppo, 2007). Rather than viewing anthropomorphism as an objective property residing within the machine, the construct is operationalized as an inductive perceptual and psychological inference made by the human observer.

The construct encompasses several interrelated psychological sub-dimensions that collectively constitute the perception of an entity as human-like:

1. Perceived Animacy and Vitality

At its foundational level, anthropomorphism involves categorizing an entity as animate rather than inanimate. The scale taps this layer through items evaluating whether the object “seemed to be alive.” This dimension measures the primitive cognitive heuristic wherein an interactive object displaying reactive, real-time responsive behavior triggers biological animacy-detection systems in human social cognition.

2. Qualitative Humanness and Personhood

Beyond rudimentary animacy, human beings possess distinct qualitative markers—phenomenological “felt” humanness, person-like qualities, and overall personhood. The ANTH scale addresses this phenomenological threshold through items measuring whether the agent “had humanlike qualities,” “seemed person-like,” and “felt human.” These items capture experiential judgments regarding whether interacting with the system simulates the holistic perceptual experience of engaging with a fellow human member of our species.

3. Individuation and Personality

A core milestone of anthropomorphic inference is the attribution of stable, distinctive behavioral, emotional, and tonal patterns, captured by the item measuring whether the agent “had a personality.” When users detect idiosyncratic linguistic styles, consistent emotional valences, or recognizable personal temperaments in an algorithm, they transition from viewing the system as a generic, deterministic computational tool to perceiving it as an individuated social agent.

4. Mental Capacity and Mind Perception

A central tenet of advanced social cognition is the theory of mind—the capacity to attribute mental states to others. The ANTH scale assesses this via the item capturing whether the agent “possessed its own mind.” Drawing directly from cognitive psychology and mind perception literature (Gray, Gray, & Wegner, 2007), this facet determines whether the interactant is viewed as possessing an internal cognitive architecture capable of autonomous deliberation, knowledge storage, and consciousness, as opposed to operating merely as a passive input-output script.

5. Agency and Intentionality

The pinnacle of anthropomorphic attribution is intentionality—the belief that an entity formulates desires, plans, goals, and acts deliberately to attain them. Measured by the item assessing whether the object “acted with its own intentions,” this dimension directly predicts moral and legal accountability. When users perceive that an AI acts with intrinsic intentionality, they naturally apply social norms, hold the agent responsible for its errors, and experience acute frustration or resentment when its performance falls short of conversational expectations.

Theoretical Framework

The theoretical architecture underpinning the ANTH scale draws upon three foundational pillars of modern cognitive psychology, consumer behavior, and communication theory: the Three-Factor Theory of Anthropomorphism, Mind Perception Theory, and the Computers Are Social Actors (CASA) Paradigm.

1. Epley, Waytz, and Cacioppo’s Three-Factor Theory (2007)

The overarching conceptual framework for the scale is rooted in the cognitive-motivational model developed by Nicholas Epley, Adam Waytz, and John T. Cacioppo. This model posits that anthropomorphic attribution is governed by three primary psychological drivers:

• Elicited Agent Knowledge: Humans possess vast cognitive schemas about human functioning, which serve as the primary baseline and default anchor when interpreting ambiguous external stimuli. When a chatbot presents human cues (e.g., first-person pronouns, dynamic responses), the human schema is activated, and individuals naturally map human traits onto the object.
• Effectance Motivation: The fundamental human desire to comprehend, anticipate, predict, and control one’s environment. By projecting human-like intentions, personalities, and minds onto an algorithm, users reduce cognitive uncertainty, making the machine’s responses intelligible within familiar social interaction frameworks.
• Sociality Motivation: The basic evolutionary need for social affiliation and connection. In periods of isolation, stress, or interpersonal deprivation, individuals become more prone to perceiving social presence, empathy, and life within non-human artifacts.

2. Gray, Gray, and Wegner’s Theory of Mind Perception (2007)

The ANTH scale integrates the structural insights of mind perception, which demonstrates that people evaluate minds along two distinct, orthogonal dimensions: Agency (the capacity for self-control, planning, thought, intentionality, and moral responsibility) and Experience (the capacity to experience sensations, emotions, warmth, hunger, pain, and joy). While conventional technological artifacts are typically viewed as lacking both dimensions, conversational and intelligent agents occupy an ambiguous boundary zone. By explicitly capturing both the subjective feeling of humanness (experience-adjacent items) and intentional, autonomous deliberation (agency items: “possessed its own mind,” “acted with its own intentions”), the ANTH scale provides comprehensive coverage of this dual-component cognitive framework.

3. The CASA Paradigm (Reeves & Nass, 1996; Nass & Moon, 2000)

The Computers Are Social Actors framework establishes that individuals automatically apply social rules, behavioral heuristics, and interpersonal expectations to interactive computing technologies, even when consciously aware that the computer lacks genuine sentience. Crolic et al. (2022) extended the CASA paradigm by integrating Expectancy Violations Theory and attribution theory: when technology is endowed with human traits, consumer expectations regarding the agent’s comprehension and problem-solving abilities escalate dramatically. When the agent fails to resolve a customer problem, this heightened attribution of intentionality and mind causes customers to perceive the failure as willful neglect, escalating negative emotional outcomes such as anger and dissatisfaction.

Validity

The psychometric validity of the ANTH scale has been rigorously evaluated across laboratory experiments, online panels (e.g., Prolific, Amazon Mechanical Turk), and real-world field settings, demonstrating high standards of construct, convergent, discriminant, and criterion-related validity.

Construct and Structural Validity

Structural validation was established by Crolic et al. (2022) across multiple independent studies involving varying chatbot designs, customer service scenarios, and experimental conditions. Confirmatory factor analysis (CFA) consistently supports a robust unidimensional model, indicating that the seven items load heavily on a single latent factor representing the perceived anthropomorphism of the target object. The average variance extracted (AVE) reliably surpasses the conservative .50 threshold (often exceeding .70), signifying that the scale captures substantially more construct variance than measurement error variance.

Convergent Validity

Convergent validity has been established through strong, statistically significant correlations with alternative, established measures of humanization and social presence. Specifically, the ANTH scale correlates positively and strongly with:

• The anthropomorphism subscale of the Godspeed inventory (Bartneck et al., 2009; $r \approx .75 – .84$).
• The Mind Attribution Scale (Waytz, Gray, Epley, & Wegner, 2010; $r \approx .68 – .79$).
• Measures of perceived social presence (e.g., Gefen & Straub, 1997; $r \approx .65 – .76$).
• Explicit single-item measures of perceived human-likeness ($r > .80$).

Discriminant Validity

Discriminant validity was verified using both the Fornell-Larcker criterion and the Heterotrait-Monotrait (HTMT) ratio of correlations. The ANTH scale reliably discriminates from conceptually adjacent but distinct constructs:

• Perceived Competence: CFA models demonstrate that anthropomorphism remains distinct from perceived machine competence (HTMT < .65), showing that an agent can be seen as highly anthropomorphic yet incompetent, or highly competent yet purely mechanical.
• Perceived Warmth: While anthropomorphic agents frequently project warmth, the two constructs are empirically separable (HTMT < .70), demonstrating that warmth is an evaluative social trait, whereas anthropomorphism is a broader ontological attribution.
• System Usability and Perceived Usefulness: Correlations with constructs from the Technology Acceptance Model (TAM) remain moderate ($r < .45$), confirming that ANTH does not merely mirror general system satisfaction or functional utility.

Criterion-Related and Predictive Validity

The scale exhibits robust predictive validity, effectively predicting key psychological, behavioral, and emotional outcomes. In Crolic et al. (2022), ANTH scores successfully moderated the relationship between service failure and customer anger. In failed service encounters, higher ANTH scores predicted greater blame attribution to the chatbot, which in turn significantly mediated heightened consumer anger, reduced repurchase intentions, and lower overall customer satisfaction. When interactions were successful, higher ANTH scores predicted enhanced rapport, perceived empathy, and customer delight.

Reliability

The internal consistency and measurement reliability of the ANTH scale have been verified across numerous empirical studies. Across diverse technological contexts—ranging from text-based chatbots and voice-based virtual assistants (e.g., Apple Siri, Amazon Alexa) to embodied humanoid service robots (e.g., SoftBank’s Pepper)—the instrument has demonstrated exceptional reliability coefficients:

• Cronbach’s Alpha (α): Across the empirical studies reported by Crolic et al. (2022), the Cronbach’s alpha coefficients for the seven-item index ranged between .92 and .96, comfortably exceeding the standard psychometric threshold of .70 for research and .90 for applied assessments.
• McDonald’s Omega (ω): Because Cronbach’s alpha assumes tau-equivalence, researchers have computed McDonald’s composite omega (ω), which regularly demonstrates values between .93 and .96, confirming high composite reliability under congeneric assumptions.
• Item-Total Correlations: Corrected item-to-total correlations for all seven items consistently exceed .70, with most falling within the .75 to .88 range, indicating that each item makes a substantial contribution to the latent construct without evidence of redundancy or collinear artifact.
• Test-Retest Stability: In repeated-measure experimental paradigms without intervening manipulations, the scale demonstrates robust test-retest reliability ($r_{tt} > .85$), reflecting stable perceptual judgments of a given interactive system.

Factor Analysis

The psychometric dimensionality of the ANTH scale was established using rigorous exploratory factor analysis (EFA) and confirmed through structural equation modeling (SEM) and confirmatory factor analysis (CFA).

Exploratory Factor Analysis (EFA)

Initial EFA using principal axis factoring and maximum likelihood estimation with both orthogonal (Varimax) and oblique (Promax) rotations unambiguously yielded a single-factor solution:

• Kaiser-Meyer-Olkin (KMO) Measure: Values typically exceed .91, indicating superb sampling adequacy.
• Bartlett’s Test of Sphericity: Highly significant across all validation samples ($\chi^2, p < .001$), confirming substantial shared variance among items.
• Eigenvalues and Variance Explained: A single primary factor routinely accounts for 68% to 78% of the total variance, with the initial eigenvalue typically exceeding 4.80, and the second eigenvalue falling well below the 1.0 Kaiser criterion cut-off (typically < 0.65). The scree plot displays a classic “elbow” after the first factor.

Confirmatory Factor Analysis (CFA)

Confirmatory factor analyses across diverse validation datasets confirm that the unidimensional model exhibits excellent fit to empirical data, meeting or exceeding modern cutoff standards proposed by Hu and Bentler (1999):

• Comparative Fit Index (CFI): Ranges between .975 and .994.
• Tucker-Lewis Index (TLI): Ranges between .962 and .990.
• Root Mean Square Error of Approximation (RMSEA): Values consistently fall between .038 and .058 (with 90% confidence intervals staying below .08).
• Standardized Root Mean Square Residual (SRMR): Values consistently range between .018 and .032.
• Factor Loadings: Standardized factor loadings (λ) across all seven items are uniformly high, ranging from .76 to .93 ($p < .001$). Typical standardized loadings are summarized below:

Measurement invariance testing across participant demographics (e.g., age cohorts, gender) and agent modalities (text-only conversational agents vs. embodied voice/avatar systems) has demonstrated full metric and scalar invariance, confirming that differences in ANTH scores reflect genuine differences in anthropomorphic perception rather than measurement artifacts.

Instrument / Measurement Tool

• Test Type: Self-report perceptual rating scale; post-interaction psychometric questionnaire.
• Target Population: Consumers, end-users, study participants, or system operators interacting with non-human agents (e.g., chatbots, virtual assistants, robots, algorithms, brands, digital avatars).
• Administration Format: Digital / online questionnaire (Qualtrics, Decipher, REDCap), paper-and-pencil, or embedded post-interaction survey widget.
• Completion Time: Approximately 1 to 2 minutes.
• Item Count: 7 items.
• Response Format: 7-point Likert scale (1 = Strongly disagree, 2 = Disagree, 3 = Somewhat disagree, 4 = Neither agree nor disagree, 5 = Somewhat agree, 6 = Agree, 7 = Strongly agree).
• Scoring Rules:
  • All 7 items are positively keyed (no reverse-scored items).
  • An overall index of perceived anthropomorphism is computed by calculating the arithmetic mean of all seven items:
    
    ANTH Index = (Item 1 + Item 2 + Item 3 + Item 4 + Item 5 + Item 6 + Item 7) / 7
  • Composite scores range from 1.00 to 7.00, with higher scores reflecting greater perceived anthropomorphism of the target entity.
  • In structural equation modeling (SEM), items can be treated as congeneric manifest indicators loading onto a single latent construct.
• Adaptation Guidelines: Researchers examining non-chatbot targets may substitute the term “the chatbot” with the appropriate target object noun (e.g., “the robot”, “the virtual assistant”, “the avatar”, “the system”, or “the brand”) without compromising the psychometric integrity of the underlying latent construct.

Permissions & Fee and Test Year

The Anthropomorphism of the Object (ANTH) scale was published in 2022 in the Journal of Marketing (American Marketing Association). As an academic psychometric measure published within peer-reviewed scholarly literature, the scale is freely accessible and available for non-commercial academic research, pedagogical use, and empirical investigations under standard fair use conventions, provided that appropriate bibliographic attribution is accorded to the original authors:

Crolic, C., Thomaz, F., Hadi, R., & Stephen, A. T. (2022). Blame the bot: Anthropomorphism and anger in customer-chatbot interactions. Journal of Marketing, 86(1), 132–148. https://doi.org/10.1177/00222429211034407

Commercial entities or organizations seeking to integrate the scale into proprietary commercial diagnostics, enterprise software platforms, or commercial monitoring suites should consult the copyright policies of the American Marketing Association (AMA) or contact the lead authors for permission.

References

• Bartneck, C., Kulić, D., Croft, E., & Zoghbi, S. (2009). Measurement instruments for the anthropomorphism, animacy, likeability, perceived intelligence, and perceived safety of robots. International Journal of Social Robotics, 1(1), 71–81. https://doi.org/10.1007/s12369-008-0001-3
• Crolic, C., Thomaz, F., Hadi, R., & Stephen, A. T. (2022). Blame the bot: Anthropomorphism and anger in customer-chatbot interactions. Journal of Marketing, 86(1), 132–148. https://doi.org/10.1177/00222429211034407
• Epley, N., Waytz, A., & Cacioppo, J. T. (2007). On seeing human: A three-factor theory of anthropomorphism. Psychological Review, 114(4), 864–886. https://doi.org/10.1037/0033-295X.114.4.864
• Gefen, D., & Straub, D. W. (1997). Gender differences in the perception and use of e-mail: An extension to the technology acceptance model. MIS Quarterly, 21(4), 389–400. https://doi.org/10.2307/249720
• Gray, H. M., Gray, K., & Wegner, D. M. (2007). Dimensions of mind perception. Science, 315(5812), 619. https://doi.org/10.1126/science.1134475
• Hu, L. T., & Bentler, P. M. (1999). Cutoff criteria for fit indexes in covariance structure analysis: Conventional criteria versus new alternatives. Structural Equation Modeling: A Multidisciplinary Journal, 6(1), 1–55. https://doi.org/10.1080/10705511909540118
• Nass, C., & Moon, Y. (2000). Machines and mindlessness: Social responses to computers. Journal of Social Issues, 56(1), 81–103. https://doi.org/10.1111/0022-4537.00153
• Reeves, B., & Nass, C. (1996). The media equation: How people treat computers, television, and new media like real people and places. Cambridge University Press.
• Waytz, A., Gray, K., Epley, N., & Wegner, D. M. (2010). Causes and consequences of mind perception. Trends in Cognitive Sciences, 14(8), 383–388. https://doi.org/10.1016/j.tics.2010.05.006

Items of the Scale