Explaining Inconsistent Privacy Effects: How Cognitive–Affective Inconsistency and Ambivalence Shape Online Information Disclosure

Abstract

This study examines why privacy concerns do not consistently deter online information disclosure by focusing on internal evaluative dynamics underlying privacy decisions. Drawing on theories of attitudinal ambivalence and cognitive–affective inconsistency, it investigates how internal tensions shape the translation of privacy concerns into disclosure behavior. Using two-phase data comprising a survey, the research distinguishes between threat-based and coping-based evaluative conflicts by operationalizing ambivalence and cognitive–affective inconsistency across privacy risks, perceived benefits, self-efficacy, and response efficacy. Results from Phase 1, based on 540 Amazon Mechanical Turk participants, indicate that while privacy concerns generally reduce disclosure intentions, this effect is significantly weakened when individuals experience higher levels of cognitive–affective inconsistency and ambivalence. Although ambivalence significantly reduces the magnitude of inconsistency, it has a limited influence on the moderating role of inconsistency. Phase 2 findings further show that under conditions of high ambivalence, cognitive–affective inconsistency related to self-efficacy exerts a significant effect in situation-specific disclosure contexts. By elucidating the dynamic interplay of the internal tensions, this study clarifies when and why privacy concerns fail to predict disclosure behavior and highlights the importance of incorporating internal evaluative dynamics into models of digital privacy decision-making.

1. Introduction

E-commerce has fundamentally reshaped business practices by creating new channels for consumer engagement, a digital transformation that is expected to drive global retail sales beyond $8 trillion by 2027 [1]. This expansion relies heavily on the collection of personal data, ranging from demographic information to behavioral patterns, which allows firms to personalize services and enhance performance [2]. However, such practices also heighten privacy concerns, particularly as advances in artificial intelligence (AI) facilitate large-scale data collection and analysis with minimal transparency [3]. In particular, privacy threats within e-commerce, characterized by the misuse of personal data and opportunistic vendor behavior, further exacerbate public anxiety regarding the loss of control over private information [4,5,6].

In light of these challenges, privacy concerns have emerged as a central theme in understanding how individuals navigate privacy-related decisions and behaviors [7,8]. Commonly defined as apprehension about losing control over personal information, privacy concern has long been viewed as a major deterrent to information disclosure [9,10]. However, empirical findings remain inconsistent: while privacy concerns significantly inhibit disclosure in certain contexts, their influence appears marginal or even negligible in others [11,12,13]. Consequently, the relationship between privacy concerns and actual behavior is widely characterized as weak, unstable, and context-dependent.

Prior studies have examined demographic, contextual, and social moderators (e.g., [14,15]); however, these external factors alone are limited to fully explain why individuals with equivalent privacy concerns exhibit divergent behaviors. Such mixed results suggest that privacy concerns may operate through more intricate psychological mechanisms than previously assumed. A promising explanation lies in internal tensions that destabilize attitudes and weaken their translation into behavior. Misaligned or conflicting evaluations reduce attitude coherence and hinder accessibility during decision making, resulting in greater behavioral variability [16].

This study focuses on two key forms of internal tension: cognitive–affective inconsistency and attitudinal ambivalence. Attitudes comprise both cognitive and affective components, with behavior often guided by the dominant one [17]. Cognitive–affective inconsistency reflects conflict between these components, occurring when beliefs (e.g., “sharing data is risky”) diverge from emotions (e.g., “using apps feels enjoyable”). Such misalignment weakens attitude strength and confidence, increasing behavioral instability [18]. Although previous studies have examined cognition and affect in contexts such as content sharing, security compliance, and social media use [19,20,21], they typically treat cognition as a precursor to affect, overlooking their potential inconsistency and its behavioral implications.

In contrast, attitudinal ambivalence refers to the simultaneous coexistence of positive and negative evaluations toward a single behavior, representing an internal conflict within the individual’s cognitive domain [22,23]. For example, consumers may appreciate the convenience of saving credit card details online while simultaneously fearing data breaches. High ambivalence reduces attitude clarity and accessibility, thereby weakening the predictive power of attitudes and producing unstable behavioral outcomes [24,25,26]. Despite their relevance for understanding the attitude–behavior link, cognitive–affective inconsistency and ambivalence remain underexplored in privacy-related contexts.

Prior research has noted that cognitive–affective inconsistency and attitudinal ambivalence are interrelated forms of internal conflict [16,27]. However, the mechanisms governing their dynamic interplay remain underexplored, particularly in privacy decision-making contexts. This study focuses on ambivalence’s buffering role in the presence of cognitive–affective inconsistency. Higher ambivalence generates psychological discomfort that motivates individuals to engage in deliberative processing, reappraisal, and integrative efforts aimed at reconciling conflicting cognitive and affective evaluations. Through these processes, ambivalence buffers the experienced intensity of cognitive–affective inconsistency by partially integrating competing elements, thereby preventing inconsistency from independently dominating judgment formation [28,29].

However, this buffering process is inherently incomplete. Compared with low ambivalence, which allows individuals to rely on dominant evaluations with minimal deliberation, high ambivalence sustains deliberative processing that heightens attentional focus on discrepancies between cognitive and affective responses. As a result, even after integrative efforts reduce psychological discomfort, residual cognitive–affective inconsistency remains salient. In such contexts, inconsistency does not operate as a direct determinant of information disclosure but instead exerts a stronger moderating influence by conditioning how privacy concerns are translated into disclosure behavior [30].

To address these gaps, this study (a) identifies different types of ambivalences and cognitive–affective inconsistencies, (b) examines their effects on the relationship between privacy concerns and information disclosure, and (c) scrutinizes the effect of the inconsistencies under different ambivalence conditions.

Drawing on Protection Motivation Theory (PMT) [31] as an organizing framework, this study identifies four key cognitive evaluations relevant to privacy decision-making: perceived benefits, privacy risk, privacy self-efficacy, and response efficacy [32]. PMT distinguishes between threat appraisal and coping appraisal. Threat appraisal involves assessing the severity and likelihood of a threat and the perceived rewards of engaging in risky behavior, whereas coping appraisal reflects perceived ability to manage the threat and the expected effectiveness of protective actions. In this context, perceived benefits serve as compensatory positive evaluations that counterbalance perceived risk, while self-efficacy and response efficacy act as protective positive evaluations that mitigate the risk. Privacy risk, in contrast, represents a negative evaluation of potential harm from disclosure. Based on these dimensions, three domain-specific forms of ambivalence and corresponding cognitive–affective inconsistency are derived, and their moderating effects on the relationship between privacy concerns and information disclosure are examined.

The analysis proceeds in two phases. In the first phase, using the full sample, the independent effects of the three types of ambivalence on their respective cognitive–affective inconsistencies, as well as the moderating roles of the inconsistencies in the privacy concerns–disclosure relationship, are tested within parallel threat appraisal and coping appraisal models. In the second phase, participants are classified into high- and low-ambivalence groups via median splits on each standardized ambivalence score. Within each group, the moderating effects of the corresponding cognitive–affective inconsistencies on the privacy concerns–disclosure relationship are examined and compared across conditions.

This study advances understanding of privacy decision-making by examining how different forms of cognitive–affective inconsistency and attitudinal ambivalence shape the relationship between privacy concerns and information disclosure. By explicating the mediating role of ambivalence in the effects of cognitive–affective inconsistency, the study offers a novel psychological explanation for the mixed effects of privacy concerns on disclosure behavior. Although frameworks such as the privacy calculus [33], APCO (antecedents–privacy concerns–outcomes) [34], and dual-process models [35] have enriched research on privacy behavior, they largely emphasize rational evaluation, risk–benefit trade-offs, or heuristic versus systematic processing, implicitly assuming coherent attitudes and stable cognitive guidance of behavior. Persistent empirical inconsistencies in the effects of privacy concerns suggest limits to these assumptions. In contrast, attitudinal ambivalence and cognitive–affective inconsistency capture internal evaluative tensions that operate beneath rational calculation, explaining why individuals with similar levels of privacy concern may behave differently. By accounting for instability in the attitude–behavior link and misalignment between cognitive and affective evaluations, these constructs extend existing models and offer a more psychologically grounded account of privacy decision-making.

2. Literature Review

2.1. The Effect of Privacy Concerns and Explanations for Inconsistencies

A substantial body of research has explored the consequences of privacy concerns across diverse digital contexts, including e-commerce [36], personalization services [37], privacy protection behaviors [38], and information disclosure [39]. However, empirical findings remain inconsistent. Some studies report that privacy concerns significantly reduce individuals’ willingness to disclose personal information [33,40], whereas others find little or no effect [41,42]. Similar inconsistencies are evident in online purchasing and social networking [43]. For example, privacy concerns negatively predicted information sharing on Facebook in some studies [44,45] but showed marginal or nonsignificant effects in others [46,47,48], particularly over time [49]. Findings also vary across domains such as mobile app adoption [50] and travel-related information sharing [51], as well as across cultural contexts. Chen [52] observed significant predictive effects of privacy concerns in Hong Kong but not in the United States, and Krasnova et al. [53] found similar cross-national differences between German and U.S. respondents. Even within the same platform, privacy concerns appear to shape some behaviors, such as restricting profile visibility, but not others, such as actual disclosure [46]. Collectively, these findings suggest that the inconsistent impact of privacy concerns cannot be fully explained by contextual or demographic variations alone, highlighting the need to identify deeper psychological mechanisms that govern privacy-related decisions.

To account for these discrepancies, prior research offers several perspectives. The behavioral valuation perspective posits that privacy exerts limited influence because individuals place relatively low value on it, willingly exchanging personal information for immediate benefits such as convenience, goods, or personalization [54,55]. By contrast, the behavioral distortion perspective emphasizes cognitive biases and heuristics (e.g., bounded rationality, illusory control, or situational cues) that distort decision-making and attenuate the role of privacy concerns [14,56,57,58]. Other scholars distinguish between opinion-oriented and behavior-oriented explanations [59]. Opinion-oriented accounts highlight limited digital literacy and misconceptions about data practices, which reduce the salience of privacy risks [60,61,62]. Behavior-oriented accounts instead argue that privacy attitudes are inherently poor predictors of disclosure, independent of knowledge [63]. Finally, other researchers point to the influence of technological design, contending that social media platforms are deliberately engineered to prompt disclosure and thereby diminish the actual effect of users’ privacy concerns on their behavior [13,15].

2.2. Cognitive–Affective Inconsistency

Cognitive–affective inconsistency arises when cognitive evaluations conflict with affective reactions toward the same object [17]. For instance, individuals may recognize the health benefits of diet food yet dislike its taste. Such inconsistency weakens the coherence of attitudes and undermines their predictive power for behavior. Millar and Tesser [64] demonstrated that inconsistent cognitive and affective components caused divergent behaviors based on the emphasis, amplifying the role of the pre-evaluation focus; however, consistent components yielded similar behavioral outcomes. Similarly, Schleicher et al. [65] found that affective–cognitive consistency strengthened the job satisfaction–performance relationship, a finding echoed by Visser and Coetzee [66]. More recently, Conner et al. [16] examined both inconsistency and ambivalence across three large-scale prospective studies and found that inconsistency had stronger and more stable effects on behavior. A meta-analysis confirmed that when both were included in models, inconsistency remained the sole significant predictor.

These findings identify cognitive–affective inconsistency as a key determinant of when attitudes reliably predict behavior, offering a promising framework for understanding the unstable relationship between privacy concerns and information disclosure.

2.3. Attitudinal Ambivalence

Classical attitude theories, including the functional theory of attitudes, assume attitudes as stable evaluative representations that guide behavior [67,68,69]. However, empirical evidence demonstrates considerable variability in the strength of this link. For instance, Kim and Hunter [70] reported attitude–behavior correlations ranging from 0.02 to 0.85, indicating that attitudes are not always reliable predictors.

One major source of this inconsistency is attitudinal ambivalence, the simultaneous presence of positive and negative evaluations toward the same object [71,72]. Ambivalence reduces clarity, accessibility, and strength of attitudes, thereby diminishing behavioral predictability [24,25,73,74]. Evidence across domains supports the disruptive role of ambivalence: ambivalence reduces attitude stability in contexts ranging from ethnic and gender attitudes [75,76] to health behavior and consumer choice [77,78]. In a digital context, ambivalence was found to influence online transactions [79] and protection behavior [80]. Further research shows that ambivalence moderated the link between attitude and organizational behaviors, including fair trade consumption [81], counterproductive behaviors toward the organization itself [82], and entrepreneurial intentions [83]. In particular, longitudinal evidence suggests that ambivalence erodes attitude stability over time [84,85] and influences responsiveness to social and normative cues [86].

Collectively, these findings underscore that ambivalence reduces the strength and consistency of the attitude–behavior relationship, providing another potential explanation for the inconsistent behavioral effects of privacy concerns.

2.4. Conceptual Distinctions Between Ambivalence and Cognitive–Affective Inconsistency

Attitudinal ambivalence and cognitive–affective inconsistency both reflect internal evaluative tension that can disrupt the attitude–behavior link, yet they are conceptually and structurally distinct [16]. Attitudinal ambivalence refers to conflict within a single attitudinal component, most commonly cognition, arising from the simultaneous endorsement of strong positive and negative evaluations of the same object or behavior [73,74]. For instance, an individual may strongly perceive both the benefits and the privacy risks of disclosing personal information. Because these opposing evaluations coexist within the cognitive component, ambivalence constitutes intra-component conflict, reducing evaluative clarity and motivating deliberative, resolution-oriented processing [87].

Cognitive–affective inconsistency, in contrast, reflects conflict between different attitudinal components, capturing misalignment between cognitive evaluations and affective responses toward the object or behavior [64]. This form of tension does not require opposing cognitions [88]. For example, an individual may hold uniformly negative beliefs about disclosure risks (low ambivalence) while simultaneously experiencing positive affect toward using a digital service. In this case, evaluative conflict arises from divergence between cognition and affect, producing inter-component conflict that undermines attitudinal coherence and weakens the reliability with which attitudes guide behavior [89,90].

Because these tensions arise at different structural levels, ambivalence and cognitive–affective inconsistency are related but nonredundant. Either may occur in the absence of the other. An individual may strongly endorse both the benefits and risks of disclosure (high ambivalence) while experiencing neutral or well-integrated affect, resulting in ambivalence with relatively low inconsistency. When mixed cognitive evaluations coexist with affective responses that diverge from their overall balance, both ambivalence and cognitive–affective inconsistency are present, yet they remain analytically distinct.

These structural differences correspond to distinct psychological roles. Ambivalence primarily disrupts evaluative clarity and motivates deliberation aimed at resolving internal conflict, whereas cognitive–affective inconsistency undermines attitudinal coherence and conditions whether attitudes are effectively activated to guide behavior at the point of decision [16]. Together, these distinctions clarify why ambivalence and cognitive–affective inconsistency represent independent mechanisms through which internal conflict shapes privacy-related decision making. To aid clarity,

Table 1. Key Distinctions between Ambivalence and Cognitive–Affective Inconsistency.

Dimension	Attitudinal Ambivalence	Cognitive–Affective Inconsistency
Scope/Structure	Intra-component (within cognition: positive and negative evaluations coexist) [71]	Inter-component (misalignment between cognitive evaluations and affective responses) [17]
Psychological Origin	Conflict from competing cognitive elements (e.g., benefits vs. risks) [71,72]	Divergence from emotional residues or contextual influences [64]
Measurement Approach	Capturing the intensity and equivalence of positive and negative evaluation: Griffin formula [72]	Operationalized as the absolute difference between cognitive evaluation and feelings about the attitude object [16].
Behavioral Implications	Reduces attitude clarity and stability; motivates deliberation and resolution-seeking [73,74,91]	Undermines attitude coherence; conditions whether attitudes guide behavior at decision point [16,64]

summarizes the key conceptual distinctions between the constructs.

2.5. The Relationship Between Cognitive–Affective Inconsistency and Ambivalence

A handful of empirical studies have directly examined the relationship between cognitive–affective inconsistency and attitudinal ambivalence. Conner et al. [16] found that both constructs significantly moderated attitude–behavior links across behavioral domains such as eating, smoking, and physical activity when tested separately; however, when modeled together, only cognitive–affective inconsistency retained unique explanatory power. More reliable influence of cognitive–affective inconsistency may lie in the fact that the inconsistency reflects structural misalignment within the attitude itself, whereas ambivalence captures the conflicts of cognitive evaluations alone. Nohlen et al. [27] demonstrated that ambivalence alone does not necessarily evoke internal conflict; rather, conflict becomes salient when cognitive–affective inconsistency is high, and situational cues highlight evaluative discrepancies.

These findings indicate that ambivalence and cognitive–affective inconsistency are interrelated and context-dependent rather than entirely distinct constructs. Both play critical roles in determining the stability and predictive strength of attitudes, but the mechanisms through which they jointly shape privacy-related behavior remain insufficiently understood, highlighting the need for further empirical examination.

In sum, despite extensive research, the effect of privacy concerns on online disclosure remains inconsistent across different contexts. Current explanations primarily focus on external factors, largely overlooking internal psychological mechanisms like cognitive–affective inconsistency and attitudinal ambivalence. These constructs capture the conflict and uncertainty individuals face when deciding to share personal information. Examining their interrelationship clarifies why privacy concerns often fail to predict behavior and advances theoretical understanding.

3. Theoretical Foundation

Traditional attitude theories, such as the functional theory of attitudes, conceptualize attitudes as stable evaluative representations stored in memory that guide judgment and behavior [67]. Within this view, attitudes are often assumed to be unidimensional and consistent predictors of behavior [68,69,92,93]. However, alternative perspectives, including the dual-component model of attitude [64,94], challenge this assumption by emphasizing that attitudes comprise two interrelated yet distinct components: a cognitive component, reflecting beliefs and judgments about an object, and an affective component, representing emotional reactions toward it. When aligned, these form a stable attitude that reliably predicts behavior. Misalignment, however, causes cognitive–affective inconsistency, weakening attitude strength and confidence, thus increasing behavioral variability [18]. In privacy decisions, where rational recognition of online service benefits may conflict with emotional fears of data misuse, this framework explains why similar privacy concerns lead to different behaviors. Alignment enhances the predictability of privacy concerns, while conflict diminishes their effect.

In contrast, attitudinal ambivalence reflects within-component conflict, defined as the coexistence of strong positive and negative cognitive evaluations toward the same object or behavior [71,72]. Ambivalence triggers psychological discomfort by activating clashing evaluations, reducing evaluative clarity and creating discomfort [24,25,74]. When individuals are confronted with the necessity of action, such as deciding whether to disclose personal information to access a digital service, this conflict becomes especially consequential. According to the Action-Based Model of Dissonance, the coexistence of incompatible evaluative tendencies impedes action readiness, generating a state of action indecision that is experienced as aversive arousal and psychological tension [95,96]. This discomfort signals a threat to effective functioning and motivates individuals to resolve the evaluative conflict to enable decisive action.

Building on this logic, the Model of Ambivalence-Induced Discomfort (MAID) [29] proposes that ambivalence activates an internal alarm system that disrupts heuristic or automatic processing and prompts more deliberative, systematic evaluation. During this resolution-oriented processing, individuals actively engage in effortful mental work, such as re-evaluating the relative importance of perceived benefits and risks or reassessing their affective reactions toward disclosure [16,29]. Through these processes, cognition and affect are often pulled toward greater alignment, thereby reducing the absolute level of cognitive–affective inconsistency. In this sense, ambivalence can mitigate the magnitude of inconsistency by motivating individuals to harmonize their internal evaluations.

Crucially, however, a reduction in the level of inconsistency does not imply a reduction in its behavioral relevance. On the contrary, ambivalence fundamentally alters the role that cognitive–affective inconsistency plays in decision making, shifting it from a direct destabilizing force to a conditional mechanism that shapes how attitudes are translated into behavior. Because ambivalence keeps both cognitive evaluations and affective reactions simultaneously salient and accessible, individuals become more sensitive to any remaining discrepancy between what they think and how they feel. Under deliberative processing, even relatively small mismatches are no longer ignored as noise; instead, they are treated as diagnostically meaningful signals of unresolved conflict that require further consideration [29,97].

In this study, the Model of Ambivalence-Induced Discomfort (MAID) provides the integrative theoretical framework for explaining how different forms of evaluative conflict shape privacy-related behavior. Specifically, MAID elucidates the psychological process through which attitudinal ambivalence activates discomfort, disrupts automatic attitude–behavior translation, and induces more deliberative processing, thereby conditioning the influence of cognitive–affective inconsistency on disclosure decisions.

4. Hypotheses and Research Model

4.1. Research Model

To operationalize the cognitive content within which ambivalence and cognitive–affective inconsistency arise, this study draws on PMT [31] as an analytical structuring framework to distinguish between two broad categories of cognitive evaluations that are particularly salient in privacy-related decisions. According to the theory, individuals evaluate both the potential threats they face and their capacity to cope with these threats when deciding whether to engage in protective behaviors [32,98]. Threat appraisal involves assessing the perceived severity and vulnerability of a risk, as well as the potential rewards of engaging in risky behavior, whereas coping appraisal pertains to evaluating one’s ability and efficacy to manage or mitigate the threat. Within this framework, perceived benefits, privacy self-efficacy, and response efficacy represent favorable cognitive evaluations, as they either compensate for or reduce the perceived harms of disclosure, whereas privacy risk constitutes an unfavorable evaluation reflecting potential negative consequences.

Building on this distinction, the study develops two separate models to reflect these appraisals within the privacy context. In the threat appraisal model, perceived benefits and privacy risks are combined to form a single cognitive evaluation for capturing cognitive–affective inconsistency. It represents the trade-off between rewards and threats associated with information disclosure. The equivalency and saliency of perceived benefits and privacy risks are used for portraying ambivalence. While the inconsistency captures the extent to which rational assessments of gains and losses are misaligned with emotional reactions toward information disclosure, attitudinal ambivalence is conceptualized as the coexistence of equivalent benefits and privacy risk evaluations.

In contrast, the coping appraisal model centers on individuals’ perceived capacity to manage privacy threats. Two cognitive evaluations are developed by pairing privacy risk with privacy self-efficacy and response efficacy, respectively. These combinations reflect internal (privacy self-efficacy) and external (response efficacy) coping resources relative to perceived threats. The inconsistencies between these evaluations and affect capture how misalignment between cognitive assessments of one’s coping ability and emotional responses influence privacy decision-making. Correspondingly, two forms of attitudinal ambivalence are conceptualized: one between privacy self-efficacy and privacy risk, and another between response efficacy and privacy risk.

Across both models, MAID provides the underlying psychological mechanism explaining how ambivalence alters processing depth and conditions the behavioral role of cognitive–affective inconsistency. PMT is used solely to differentiate the cognitive domains within which these internal tensions arise, allowing for a systematic examination of how ambivalence and inconsistency operate across distinct evaluative contexts. The conceptual framework is illustrated in Figure 1.

4.2. Hypotheses

4.2.1. Effect of Privacy Concerns on Information Disclosure

Privacy concerns have long been recognized as a key deterrent to personal information disclosure in digital environments. Individuals who perceive higher risks of data misuse such as unauthorized access tend to demonstrate a reduced willingness to share personal information [99]. Empirical studies consistently support this negative association across diverse contexts [45,58,100,101]. Collectively, these findings indicate that elevated privacy concerns significantly inhibit individuals’ intentions and behaviors related to self-disclosure.

H1: 

Privacy concerns are negatively associated with information disclosures.

4.2.2. Effect of Cognitive–Affective Inconsistency

The inconsistency between cognitive evaluations and affective responses that jointly constitute an attitude reflects a misalignment between rational beliefs and emotional reactions [16]. This study focuses on the cognitive evaluations underlying threat and coping appraisals, identifying three distinct forms of cognitive–affective inconsistency. Within the threat appraisal model, inconsistency is conceptualized as the misalignment between the integrated benefit–privacy risk evaluation and affect toward information disclosure. This form of inconsistency captures the extent to which rational assessments of potential rewards and risks diverge from emotional reactions to disclosure. In the coping appraisal model, two additional inconsistencies are examined: one between privacy self-efficacy and privacy risk, and another between response efficacy and privacy risk. These inconsistencies reflect the degree to which cognitive evaluations of internal and external coping capacities are misaligned with affective responses.

When cognitive evaluations within these appraisals conflict with affective reactions, the integrative foundation of privacy attitudes becomes unstable, thereby weakening the motivational force of privacy concerns in guiding disclosure behavior. Individuals experiencing such inconsistency may struggle to align their cognitive beliefs with their emotional orientations, making it more difficult to translate privacy concerns into coherent behavioral intentions. Consequently, their disclosure decisions become more variable and context dependent, reflecting a weakened attitude–behavior relationship under conditions of cognitive–affective misalignment [18].

H2: 

Cognitive–affective inconsistency of benefits-risk evaluation and affect weakens the effect of privacy concerns on information disclosure.

H3: 

Cognitive–affective inconsistency weakens the effect of privacy concerns on information disclosure.

H3(a): 

Cognitive–affective inconsistency between privacy self-efficacy–privacy risk and affect weakens the effect of privacy concerns on information disclosure.

H3(b): 

Cognitive–affective inconsistency between response efficacy–privacy risk and affect weakens the effect of privacy concerns on information disclosure.

4.2.3. Effect of Attitudinal Ambivalence

Attitudinal ambivalence refers to the simultaneous presence of both positive and negative evaluations toward an object or behavior [72]. Unlike cognitive–affective inconsistency, which arises from misalignment between cognition and emotion, ambivalence resides within the cognitive system itself, occurring when conflicting evaluations coexist at the same level of processing.

From the perspective of attitude strength, the stability of attitudes over time is a critical determinant of their ability to predict behavior [71]. However, attitudinal ambivalence undermines this stability by accessibility, coherence, and certainty of evaluative judgments [102]. When individuals simultaneously recognize both favorable and unfavorable attributes, their overall attitude becomes less certain and more context-dependent. This internal conflict lowers attitude accessibility, meaning that attitudes become less readily retrievable from memory during behavioral decision-making [97,103]. Empirical evidence consistently shows that high levels of ambivalence weaken the strength, stability, and predictive power of attitudes [29,102].

In the context of privacy and information disclosure, individuals experiencing strong ambivalence are likely to display inconsistent or unstable disclosure behavior. Their privacy attitudes lack internal coherence, making them less reliable predictors of whether they will disclose personal information in a given situation. Similar to cognitive–affective inconsistencies, this study examines three distinct forms of attitudinal ambivalence: (1) ambivalence between perceived benefits and privacy risk, (2) ambivalence between privacy self-efficacy and privacy risk, and (3) ambivalence between response efficacy and privacy risk. These conflicting evaluations are expected to weaken the motivational linkage between privacy concerns and disclosure behavior. Consequently, it is hypothesized that:

H4: 

Attitudinal ambivalence between benefits and privacy risk weakens the effect of privacy concerns on information disclosure.

H5: 

Attitudinal ambivalence weakens the effect of privacy concerns on information disclosure.

H5(a): 

Attitudinal ambivalence between privacy self-efficacy and privacy risk weakens the effect of privacy concerns on information disclosure.

H5(b): 

Attitudinal ambivalence between response efficacy and privacy risk weakens the effect of privacy concerns on information disclosure.

4.2.4. Effect of Attitudinal Ambivalence on the Level of Inconsistency

Attitudinal ambivalence generates psychological discomfort and threatens action readiness [24,74]. When individuals face a consequential decision, such as whether to disclose personal information, this discomfort motivates efforts to restore evaluative coherence in order to enable decisive action [28,95]. According to the MAID, ambivalence activates systematic, resolution-oriented processing aimed at reducing internal conflict, rather than maintaining or amplifying it [95]. During this process, individuals reweigh cognitive evaluations of benefits and risks and reassess their affective reactions, which facilitates greater alignment between cognition and affect.

As a result, higher ambivalence is expected to reduce the discrepancy between cognitive evaluations and affective responses over time, as individuals actively work to reconcile these components. Importantly, this reduction reflects an attempt to resolve discomfort rather than the absence of conflict at the outset. Thus, although ambivalence initially signals evaluative conflict, it ultimately motivates integrative processing that attenuates the level of cognitive–affective inconsistency.

H6: 

Ambivalence between benefits and privacy risk decreases the cognitive–affective inconsistency of benefits- risk evaluation and affect.

H7: 

Ambivalences of coping appraisal decreases cognitive–affective inconsistencies.

H7(a): 

Ambivalence between privacy self-efficacy and privacy risk decreases cognitive–affective inconsistency between privacy self-efficacy–privacy risk and affect.

H7(b): 

Ambivalence between response efficacy and privacy risk decreases cognitive–affective inconsistency between response efficacy–privacy risk and affect.

4.2.5. The Moderating Effect of Inconsistency Under Different Ambivalences

While ambivalence motivates individuals to reduce internal conflict and may therefore attenuate the level of cognitive–affective inconsistency, it simultaneously alters how remaining discrepancies between cognition and affect influence behavior. When ambivalence is high, individuals lack a dominant evaluative orientation that can automatically guide action, because positive and negative cognitive evaluations are both salient and competing [29,74]. According to the MAID [29], the absence of a dominant attitude weakens automatic attitude activation and increases reliance on controlled, deliberative processing to form behavioral intentions. In this deliberative state, individuals become more attentive to internal coherence and evaluative alignment because unresolved conflict threatens decisiveness and effective action.

Under such conditions, even relatively small discrepancies between what individuals cognitively believe (e.g., privacy concerns) and how they affectively feel about disclosure become psychologically salient. Cognitive–affective inconsistency serves as a diagnostic cue indicating unresolved internal conflict, which disrupts the translation of privacy concerns into disclosure behavior. In contrast, when ambivalence is low and either benefits or risks dominate, behavior is guided primarily by the dominant evaluation through automatic processes, leaving little opportunity or motivation for internal inconsistencies to exert influence [97]. Thus, ambivalence amplifies the moderating role of cognitive–affective inconsistency, even as it may reduce its absolute magnitude.

H8: 

The effect of cognitive–affective inconsistency is different by the level of ambivalence.

H8(a): 

The effect of cognitive–affective inconsistency is greater under high ambivalence between benefits and privacy risk.

H8(b): 

The effect of cognitive–affective inconsistency is greater under high ambivalence between privacy self-efficacy and privacy risk.

H8(c): 

The effect of cognitive–affective inconsistency is greater under high ambivalence between response efficacy and privacy risk.

5. Study Design and Data Collection

5.1. Measurement Items

The constructs in this study were measured using items adapted from established scales, with minor wording modifications to suit the study context and participants. Privacy concerns were assessed using items from Malhotra et al. [39]. Affect toward information was measured using the items from Chaiken & Baldwin [104]. Privacy risk items were drawn from Xu et al. [105], while perceived benefits were measured using items from Xu et al. [106]. Privacy self-efficacy and response efficacy were measured with the items of Youn [32] and Johnston and Warkentin [107], respectively. All items used a seven-point Likert scale ranging from 1 (“strongly disagree”) to 7 (“strongly agree”). Gender and age were included as control variables.

To capture the misalignment between cognitive and affective components of attitudes, this study operationalizes cognitive–affective inconsistency as the discrepancy between cognitive evaluations and affective responses toward information disclosure [104]. Affect was measured using five items on a seven-point Likert scale which indicates feeling state from negative affect to positive affect. The mean of these items was used as the affect score. Cognitive evaluations were assessed through perceived benefits, privacy risk, privacy self-efficacy, and response efficacy, each averaged across its respective items. To ensure comparability with the affective scale, benefit-related evaluations were rescaled to create a continuum from 1 to 7, where higher values represent more positive cognitive appraisals relative to perceived risk. Benefits and privacy risk were combined using the formula: Benefits/Risk = (Benefits − Privacy Risk)/2 + 4.

Cognitive–affective inconsistency was then calculated as the absolute difference between the rescaled cognitive evaluation and the corresponding affect score. Along with the threat and coping appraisal model, three types of inconsistency were computed based on the combinations of benefits and privacy risk with affect, privacy self-efficacy and privacy risk with affect, and response efficacy and privacy risk with affect. The same rescaling procedure was applied to the self-efficacy/risk and response efficacy/risk combinations before calculating their respective inconsistencies.

In line with Thompson et al. [72], attitudinal ambivalence was measured by concentrating on the similarity and intensity of positive and negative cognitive evaluations, two opposite evaluations of information disclosures [67]. Ambivalence was calculated using the following formula [72,108]: Ambivalence = (positive + negative)/2 − |positive − negative|.

Attitudinal ambivalence decreases when the similarity of positive and negative beliefs increases. Because each belief was measured with multiple items, the average of the responses of items measuring the belief was calculated, and then attitudinal ambivalence was calculated. The list of all the measurement items is presented in Appendix A.

5.2. Data Collection

To test the proposed model and hypotheses, data were collected using Amazon Mechanical Turk (MTurk). MTurk is an online crowdsourcing platform that enables researchers to recruit participants for surveys and experiments. It is widely used in information systems and behavioral research because it provides timely access to diverse adult internet users familiar with digital services, online transactions, and information disclosure. MTurk was chosen for this study due to its relevance to privacy-related decision-making and its features for screening participants, attention checks, and response-quality controls, which enhance data reliability and internal validity. These procedures align with prior research and help ensure robust empirical analyses.

Participants were required to have experience disclosing personal information for purchasing products on e-commerce or social networking platforms. To minimize potential bias from inattentive or unfaithful responses, MTurk workers were first screened based on their reputations. Participants were provided with brief information about the study, including its goals, social impact, and potential risks, and were asked to provide informed consent before participating. A cross-sectional survey was then administered, in which respondents reported their privacy concerns, affect toward information disclosure, perceived benefits, privacy risks, privacy self-efficacy, and response efficacy, along with their willingness to disclose personal information.

6. Analyses and Results

6.1. Demographic Statistics

A total of 582 MTurk workers residing in the U.S. participated in the survey. Among them, 42 were excluded due to incorrect MTurk IDs or failure to provide the correct security code, essential for tracking participants in this study, leaving 540 valid responses. As shown in

Table 2. Descriptive statistics.

Total Observation	Items		Frequency/ Average	Percent/Standard Deviation
540	Gender	Female	205	38.0%
		Male	335	62.0%
	Age		36.7	10.2

, approximately 38.0% identified as female, with a mean age of 36.7 years.

The measurements were assessed in terms of construct reliability, and convergent and discriminant validity. To establish indicator reliability, loading values of constructs’ items were first checked to see if there was an item whose value is equal to or lower than 7 [109]. Then, construct reliability was examined using Cronbach’s alpha and composite reliability, with the common threshold of 7 [110]. Convergent validity was assessed using the average variance extracted (AVE), with the threshold of 0.50 [109]. As shown in

Table 3. Analysis of Construct Reliability.

Constructs	Cronbach Alpha	Composite Reliability	AVE
Perceived Benefits	0.949	0.963	0.867
Privacy Risk	0.918	0.942	0.802
Privacy Self-efficacy	0.926	0.946	0.815
Response Efficacy	0.906	0.933	0.778
Privacy Concerns	0.909	0.936	0.785

Notes: AVE = Average Variance Extracted.

, each construct demonstrated good reliability and context validity.

Next, the discriminant validity of the reflective constructs was assessed using the heterotrait-monotrait (HTMT) ratio proposed by Henseler et al. [111]. According to this method, HTMT values should ideally be below 0.85 for optimal discriminant validity, or below 0.90 to be considered acceptable. Upon calculating the HTMT values (see

Table 4. Discriminant validity.

	BEN	PRC	RES	RISK	SEL
BEN
PRC	0.277
RES	0.374	0.222
RISK	0.385	0.822	0.267
SEL	0.086	0.057	0.531	0.091

Note: BEN = perceived benefits; PRC = privacy concerns; RES = response efficacy; RISK = privacy risk; SEL = privacy self-efficacy.

), all values were found to meet the criteria.

Variance inflation factors (VIFs) were computed for assessing multicollinearity among predictor constructs for the threat and coping model, respectively. For the threat model, VIF values were below the threshold of 3.3 [112]. However, for the coping model, VIFs of ambivalences and cognitive–affective inconsistencies were above the threshold but below 5.0, suggesting moderate but acceptable multicollinearity [113].

Common method bias (CMB) was assessed using multiple diagnostic procedures. First, Harman’s single-factor test was conducted, revealing that the single factor explained 43.3% of the variance in the threat appraisal model and 36.9% in the coping appraisal model. Both values fall below the commonly accepted 50% threshold, suggesting that CMB is unlikely to be a serious concern. Second, a marker variable approach was applied following Lindell and Whitney [114]. Cultural belief in individualism was selected as the marker variable, as it is theoretically unrelated or only weakly related to the focal constructs. Correlations between the marker variable and the substantive constructs did not exceed 0.30 [115]. Moreover, inclusion of the marker variable resulted in only marginal changes in explanatory power, increasing R² by 0.1% in the threat model and 0.2% in the coping model. Together with the full collinearity VIF results, these findings indicate that CMB is unlikely to substantially affect the study’s conclusions.

6.2. Phase 1: The Effect of Inconsistency and Ambivalence

In phase 1, the moderating effects of different forms of cognitive–affective inconsistencies and ambivalences were examined, defined by the threat and coping appraisal model, using partial least squares–structural equation modeling (PLS-SEM). PLS-SEM was employed for data analysis using SmartPLS 4. This approach was chosen because the study emphasizes the examination of moderating effects involving ambivalence and cognitive–affective inconsistency, rather than the assessment of global model fit. In addition, the latent variables exhibited significant deviations from normality, as indicated by the Jarque–Bera (JB) test [116]. Given its robustness to non-normal data and its suitability for estimating interaction effects in complex models, PLS-SEM was deemed appropriate for the present analysis. As shown in

Table 5. Jarque–Bera (JB) Normality Test.

	BEN	PRC	RES	RISK	SEL
JB statistics	51.46	37.61	22.60	3.21	14.27
p-value	0.000	0.000	0.000	0.201	0.001

, all variables except privacy risk deviated from a normal distribution. Because traditional maximum likelihood estimation in covariance-based SEM (CB-SEM) relies on normality, violation of this assumption can compromise the reliability of parameter estimates [117]. Furthermore, the potential differences in parameter estimates between CB-SEM and PLS-SEM were considered [118]. Reinartz et al. [119] (p. 338) note that “parameter estimates do not differ significantly from their theoretical values for either ML-based CB-SEM (p-values between 0.3963 and 0.5621) or PLS (p-values between 0.1906 and 0.3449).” PLS-SEM is less sensitive to non-normality, and bootstrapping with 5000 subsamples using the bias-corrected and accelerated method further enhances the reliability and generalizability of the results.

The results, summarized in

Table 6. Effects of Inconsistencies and Ambivalences.

Exogenous	Threat Model	Coping Model	Hypothesis Test Result
PRC → IDB	−0.335 *** (0.053)	−0.336 *** (0.049)	H1: Supported
INC-BA × PRC → IDB	−0.128 ** (0.037)		H2: Supported
INC-SA × PRC → IDB		−0.161 (0.113)	H3(a): Not supported
INC-RA × PRC → IDB		−0.244 * (0.122)	H3(b): Supported
AMV-BR × PRC → IDB	−0.095 * (0.039)		H4: Supported
AMV-SR × PRC → IDB		−0.163 * (0.076)	H5(a): Supported
AMV-RR × PRC → IDB		−0.077 (0.080)	H5(b): Not supported
AMV-BR → INC-BR	−0.422 *** (0.041)		H6: Supported
AMV-SR → INC-SR		−0.285 *** (0.047)	H7(a): Supported
AMV-RR → INC-RR		−0.433 *** (0.045)	H7(b): Supported
GEN	−0.013 (0.076)	0.046 (0.037)	Control Variables
AGE	0.041 (0.035)	−0.007 (0.077)
Adjust R2 (IDB)	0.270	0.263

Notes: PRC = privacy concerns; INC-BA = inconsistency between benefits/risk and affect; INC-SA = inconsistency privacy self-efficacy/risk and affect; INC-RA = inconsistency between response efficacy/risk and affect; AMV-BR = attitudinal ambivalence between benefits and privacy risk; AMV-SR = attitudinal ambivalence between privacy self-efficacy and privacy risk; AMV-RR = attitudinal ambivalence between response efficacy and privacy risk; IDB = information disclosure behavior; AGE = age; GEN = gender; The signs of the interaction term coefficients were to be negative for interpretation; The number in parentheses is standard deviation; * p < 0.05, ** p < 0.01, *** p < 0.001.

, indicate that privacy concerns were negatively associated with information disclosure across both the threat and coping appraisal models, supporting H1. This relationship was further weakened by the inconsistency between benefits-risk evaluation and affect, providing evidence for H2. In the coping appraisal model, the inconsistency between privacy self-efficacy-privacy risk and affect had a negligible effect, whereas the impact of privacy concerns diminished as the inconsistency between response efficacy–privacy risk and affect increased, supporting H3(b) but not H3(a).

Attitudinal ambivalence also moderated the effect of privacy concerns. Ambivalence between benefits and privacy risk significantly reduced the influence of privacy concerns, supporting H4, and a similar pattern was observed for ambivalence between privacy self-efficacy and privacy risk, supporting H5(a). In contrast, the interaction between response efficacy and privacy risk was negligible, failing to support H5(b). Across both appraisal models, all three forms of attitudinal ambivalence were negatively associated with cognitive–affective inconsistencies, consistent with H6, H7(a), and H7(b). Overall, these findings highlight that both cognitive–affective inconsistency and attitudinal ambivalence play critical roles in weakening the motivational impact of privacy concerns on information disclosure.

6.3. Phase 2: The Effect of Inconsistency Under Different Levels of Ambivalence

Following established ambivalence research [72], participants’ perceived benefits and privacy risks were first standardized to place both evaluations on a common metric and to reduce potential scale-usage bias. Standardization ensures that ambivalence reflects relative evaluative strength within the sample rather than absolute response tendencies.

Using the standardized benefit and risk scores, an ambivalence index was computed to represent the degree to which strong positive and negative evaluations coexist. This approach reflects two core properties of ambivalence emphasized in prior research: evaluative intensity and balance [16]. Higher values on this index indicate greater ambivalence, characterized by concurrently strong and competing benefit and risk perceptions, whereas lower values indicate clearer evaluative dominance or weak, non-diagnostic evaluations.

Participants were then classified into high- and low-ambivalence groups using a median split on the ambivalence index. Individuals above the median were categorized as high in ambivalence, reflecting substantial internal conflict between perceived benefits and risks. Individuals at or below the median were categorized as low in ambivalence, encompassing cases in which either one evaluation clearly dominated or both evaluations were weak. This grouping strategy follows prior research that treats ambivalence as a structurally distinct psychological state rather than a purely continuous moderator [66,104].

Participants were also categorized into high- and low-ambivalence groups for privacy self-efficacy and response efficacy, respectively, following the same standardization and ambivalence classification procedure applied to benefit and risk evaluations. Then, the effect of cognitive–affective inconsistency was examined across the groups.

As shown in

Table 7. The Effect of Inconsistency under Different Levels of Ambivalence.

Exogenous		High Ambivalence	Low Ambivalence	Hypothesis Test Result
Benefits Model	PRC → IDB	−0.214 *** (0.055)	−0.538 *** (0.052)
	INC-BA × PRC → IDB	−0.137 * (0.054)	−0.157 ** (0.048)	H8(a): Not supported
Privacy Self-efficacy Model	PRC → IDB	−0.286 *** (0.053)	−0.495 *** (0.055)
	INC-SA × PRC → IDB	−0.143 ** (0.054)	−0.070 (0.050)	H8(b): Supported
Response Efficacy Model	PRC → IDB	−0.274 *** (0.053)	−0.486 *** (0.052)
	INC-RA × PRC → IDB	−0.120 * (0.052)	−0.123 * (0.048)	H8(c): Not supported

Notes: PRC = privacy concerns; INC-BA = inconsistency between benefits/risk and affect; INC-SA = inconsistency between privacy self-efficacy/risk and affect; INC-RA = inconsistency between response efficacy/risk and affect; * p < 0.05, ** p < 0.01, *** p < 0.001.

, Tthe results indicated that the effect of privacy concerns on information disclosure was greater in low ambivalence in all the models, confirming the effect of ambivalence. The moderating effect of cognitive–affective inconsistency under high ambivalence was significantly more pronounced within the privacy self-efficacy model alone. In contrast, the effects within the benefits and response efficacy models were only marginally higher or statistically indistinguishable. Consequently, the data provided empirical support for H8(b) only.

6.4. Additional Analysis

6.4.1. Multiple Models

To ensure the robustness of the model, several alternative specifications were developed and estimated, as summarized in

Table 8. Analysis Results of Multiple Models.

Exogenous	Model 1	Model 2	Model 3	Model 4	Model 5
PRC → IDB	−0.337 ***	−0.355 ***	−0.334 ***	−0.291 ***	−0.297 ***
INC-BA × PRC → IDB	−0.223 ***		−0.172 ***	−0.168 ***	−0.158 ***
INC-SA × PRC → IDB	−0.121 **		−0.040		−0.114 *
INC-RA × PRC → IDB	−0.150 ***			−0.132 **	−0.173 ***
AMV-BR × PRC → IDB		−0.204 ***	−0.114 *	−0.164 **	−0.146 **
AMV-SR × PRC → IDB		−0.094 *	−0.028		−0.021
AMV-RR × PRC → IDB		−0.103 *		−0.018	−0.021
AMV-BR → INC-BR			−0.422 ***	−0.422 ***	−0.422 ***
AMV-SR → INC-SR			−0.504 ***		−0.504 ***
AMV-RR → INC-RR				−0.541 ***	−0.541 ***
GEN	−0.002	0.000	−0.000	−0.016	−0.008
AGE	0.043	0.036	0.031	0.052	0.041

Notes: (a) PRC = privacy concerns; INC-BA = inconsistency between benefits/risk and affect; INC-SA = inconsistency privacy self-efficacy/risk and affect; INC-RA = inconsistency between response efficacy/risk and affect; AMV-BR = attitudinal ambivalence between benefits and privacy risk; AMV-SR = attitudinal ambivalence between privacy self-efficacy and privacy risk; AMV-RR = attitudinal ambivalence between response efficacy and privacy risk; IDB = information disclosure behavior; AGE = age; GEN = gender. (b) The signs of the interaction term coefficients were to be negative for interpretation. * p < 0.05, ** p < 0.01, *** p < 0.001.

. To remove possible collinearity among the constructs, an orthogonalization procedure was employed. All variables were mean-centered before creating interaction terms, and residualized predictors for privacy self-efficacy, response efficacy, and their respective interactions were obtained through regression. When multiple inconsistencies were included simultaneously (Model 1), they collectively and significantly weakened the effect of privacy concerns on information disclosure. Likewise, in the model including multiple ambivalences (Model 2), ambivalences significantly reduced the influence of privacy concerns. When both inconsistencies and ambivalences were incorporated together (Model 5), nearly all inconsistencies and most ambivalences, except the ambivalences involving privacy risk with privacy self-efficacy and response efficacy, significantly attenuated the effect of privacy concerns. The negligible effects of these two ambivalences may reflect minor information loss resulting from the orthogonalization procedure.

6.4.2. Group Comparisons for Examining Ambivalences and Inconsistencies

To further assess the robustness of the moderating effects of ambivalence and cognitive–affective inconsistency, participants were categorized into high and low groups for each construct, and the effects of privacy concerns were compared using independent-sample t-tests. Standardized scores were first computed for each form of ambivalence and inconsistency, after which participants were divided into high and low groups. Positive standardized values were classified as high ambivalence or low inconsistency, whereas negative values were classified as low ambivalence or high inconsistency. To ensure clearer group distinctions, cases with standardized values between −0.1 and 0.1 were excluded from the analysis. As shown in

Table 9. The Effects of Privacy Concerns across Different Inconsistencies and Ambivalences.

	Coefficient		Standard Deviation		Difference
	High	Low	High	Low
AMV-BR	−0.251	−0.627	0.054	0.04	0.376 ***
AMV-SR	−0.338	−0.559	0.058	0.049	0.221 ***
AMV-RR	−0.310	−0.579	0.051	0.048	0.269 ***
INC-BA	−0.265	−0.591	0.046	0.048	0.326 ***
INC-SA	−0.341	−0.564	0.052	0.050	0.223 ***
INC-RA	−0.306	−0.573	0.050	0.050	0.267 ***

Notes: AMV-BR = attitudinal ambivalence between benefits and privacy risk; AMV-SR = attitudinal ambivalence between privacy self-efficacy and privacy risk; AMV-RR = attitudinal ambivalence between response efficacy and privacy risk; INC-BA = inconsistency between benefits/risk and affect; INC-SA = inconsistency between privacy self-efficacy/risk and affect; INC-RA = inconsistency between response efficacy/risk and affect; *** p < 0.001.

, the results revealed that the effect of privacy concerns on information disclosure was significantly weaker among participants exhibiting high ambivalence or high inconsistency compared to those with lower levels of these constructs.

6.4.3. Group Comparisons of the Relationship Between Ambivalences and Inconsistencies

Ambivalence scores were standardized and ranked to classify participants into high and low ambivalence groups. To ensure a clear distinction between groups, cases with standardized values between −0.1 and 0.1 were excluded from the analysis. Based on this classification, participants with higher benefit–risk ambivalence were compared to those with lower ambivalence in terms of their inconsistency of benefit–privacy risk and affect using independent-sample t-tests. The same procedure was applied to the other forms of inconsistency. As shown in

Table 10. The Effect of Ambivalences on Cognitive–Affective Inconsistencies.

	AMV-BR		AMV-SR		AMV-RR
	High	Low	High	Low	High	Low
Mean of Inconsistency	1.162	2.092	1.182	1.888	1.171	2.088
t-value	−8.593 ***		−6.684 ***		−7.908 ***

Notes: AMV-BR = attitudinal ambivalence between benefits and privacy risk; AMV-SR = attitudinal ambivalence between privacy self-efficacy and privacy risk; AMV-RR = attitudinal ambivalence between response efficacy and privacy risk; *** p < 0.001.

, participants exhibiting higher ambivalence consistently demonstrated lower cognitive–affective inconsistency across all three conditions. These findings may support that elevated ambivalence destabilizes cognitive evaluations by undermining their confidence and coherence, thereby reducing their accessibility and weakening their capacity to guide behavior.

7. Discussion

This study investigated how cognitive–affective inconsistency and attitudinal ambivalence jointly shape the relationship between privacy concerns and information disclosure within the dual framework of threat and coping appraisals. The results offer several important insights into the distinct psychological mechanisms through which these constructs operate.

First, the findings reveal that cognitive–affective inconsistency exerted a stronger moderating effect on the relationship between privacy concerns and information disclosure than did attitudinal ambivalence, across both the threat and coping appraisal models. This suggests that discrepancies between cognitive evaluations and affective responses exert a more decisive influence in attenuating the behavioral impact of privacy concerns. While attitudinal ambivalence captures cognitive conflict, cognitive–affective inconsistency reflects a deeper misalignment across mental systems: between what individuals believe and what they feel. Such cross-system discordance disrupts attitude stability, weakens attitude–behavior consistency, and impairs the integration of cognitive assessments into decision-making processes [25]. The stronger moderating role of cognitive–affective inconsistency therefore implies that affective dissonance can overshadow cognitive conflict, rendering privacy attitudes less coherent, less retrievable, and ultimately less predictive of disclosure behavior. In other words, when individuals “know” privacy is risky but “feel” unbothered, the motivational power of privacy concerns diminishes.

Second, the relatively weak moderating effects of inconsistency between self-efficacy or risk and affect may indicate that these constructs are more cognitively stable and less susceptible to emotional misalignment. Self-efficacy, grounded in self-reflection and accumulated experience, tends to be internally coherent and resilient to affective fluctuation. Individuals typically possess direct and confident access to their efficacy beliefs, allowing these cognitions to dominate even in the presence of conflicting emotions. Likewise, risk perceptions, often formed through deliberate evaluation and reinforced by knowledge or past exposure, maintain cognitive accessibility and are less easily disrupted by transient affective states. The marginal moderating effects of these inconsistencies suggest that when individuals hold well-defined efficacy or risk beliefs, the impact of affective discordance on privacy-related decision-making is limited, highlighting a potential avenue for future research.

Third, in the coping appraisal model, ambivalence between response efficacy and privacy risk exhibited only a negligible moderating effect. A plausible interpretation is that, while response efficacy informs beliefs about the general effectiveness of protective measures, it is less directly tied to an individual’s personal sense of agency in mitigating privacy threats. Within digital environments, people may rely more heavily on self-efficacy than on response efficacy, which concerns the perceived adequacy of external systems or tools [120]. This plausible explanation could be examined in future research.

Finally, the moderating effect of cognitive–affective inconsistency was stronger in the high ambivalence condition only for the privacy self-efficacy model, whereas its effect was weaker or statistically indistinguishable across ambivalence levels in the benefit and response efficacy model. This asymmetry may reflect differences in the perceived reliability and confidence of the underlying evaluative information. Privacy self-efficacy reflects internally generated assessments of personal capability, which tend to be relatively concrete, self-referential, and stable. Under high ambivalence, discrepancies between cognitive evaluations and affective reactions may become particularly salient, amplifying the moderating role of cognitive–affective inconsistency. In contrast, evaluations of benefits and response efficacy rely more on external information, which may be perceived as less transparent, less verifiable, or contextually constrained. Consequently, cognitive–affective inconsistency may appear to add limited incremental explanatory power in these domains. Future research could test this potential explanation for the observed pattern.

8. Theoretical and Practical Implications

8.1. Theoretical Implications

This study makes several important theoretical contributions to privacy research, attitude theory, and information systems scholarship. First, this study advances privacy research by demonstrating that the behavioral relevance of privacy concerns is jointly shaped by attitudinal ambivalences and cognitive–affective inconsistencies, which operate through distinct psychological mechanisms. Ambivalences weaken the privacy concern–disclosure relationship by reducing evaluative clarity and confidence, whereas cognitive–affective inconsistencies undermine this relationship by disrupting coherence between cognitive assessments and affective responses. By modeling both moderators simultaneously, the study shows that unstable privacy behavior cannot be fully explained by within-cognition conflict alone and that cross-system misalignment constitutes an additional and theoretically distinct source of attitude weakness.

Second, this study advances attitude–behavior theory by examining different forms of cognitive–affective inconsistency and attitudinal ambivalence. Whereas prior research has largely focused on overall ambivalence or generalized inconsistency, the findings suggest that different forms of internal tension have independent and context-dependent effects on behavior. Specifically, these forms of evaluative conflict each reduce the motivational impact of privacy concerns, even when overall attitude strength is high. The results underscore the importance of incorporating multiple, contextually relevant forms of internal tension into attitude–behavior models, highlighting that attitude strength alone does not fully account for behavioral outcomes. By examining different forms of internal tensions, this study extends multi-component attitude frameworks and encourages theorists to consider how the type and context of evaluative tension shape attitude-driven behavior.

Finally, the study contributes to attitude theory by highlighting the interplay between ambivalence and cognitive–affective inconsistency in shaping attitude–behavior links. The results show that higher ambivalence is associated with lower levels of inconsistency, suggesting that these constructs are related but not redundant. Importantly, ambivalence’s influence on the moderating effect of inconsistency is limited, indicating that inconsistencies can still weaken the attitude–behavior relationship even when ambivalence is present. This finding refines multi-component attitude models by demonstrating that distinct forms of evaluative tension may coexist and operate in partially independent ways, emphasizing the need to account for both ambivalence and inconsistency when explaining variability in attitude-driven behavior.

8.2. Practical Implications

This study offers several practical implications for enterprises, particularly digital platforms, e-commerce firms, and service providers that rely on user data and voluntary information disclosure. First, the findings suggest that enterprises should move beyond strategies that merely reduce perceived privacy risks or increase perceived benefits and instead focus on achieving coherence between users’ cognitive evaluations and affective responses. When cognitive assessments and emotional reactions point in different directions, privacy concerns lose their motivational force. Interface designs that emphasize emotional reassurance without reinforcing cognitive understanding may therefore backfire by increasing cognitive–affective inconsistency. Enterprises should pursue convergent design, in which informational content (e.g., privacy explanations and risk disclosures) and affective cues (e.g., visual tone, language warmth, and feedback timing) consistently reinforce the same evaluative direction.

Second, the results indicate that managing ambivalence alone is insufficient to ensure consistent disclosure behavior. Although reducing ambivalences can lower internal conflict, the findings show that cognitive–affective inconsistency continues to weaken disclosure decisions even at low levels of ambivalence. This cautions enterprises against relying solely on trade-off simplification strategies, such as emphasizing benefits over risks. For example, dashboards that transparently summarize both benefits and risks may reduce ambivalence by explicitly delineating the consequences of disclosure and contextualizing their relevance, thereby enabling users to cognitively integrate competing considerations. However, when such information is delivered in an overly reassuring or playful manner, it may inadvertently increase cognitive–affective inconsistency by dampening appropriate concern and misaligning emotional responses with cognitive evaluations.

Third, the domain-specific findings offer targeted design guidance. The stronger conditional effect of inconsistency under high ambivalence in the privacy self-efficacy model suggests that enterprises should pay particular attention to users’ perceived control and competence. Interfaces that enhance self-efficacy, such as granular privacy controls, clear feedback on user actions, and customizable data-sharing options, should also ensure that these features evoke affective confidence rather than confusion or anxiety. Misalignment between perceived control and emotional comfort can undermine the intended effects of empowerment-oriented privacy tools.

Finally, the findings caution against overreliance on affective nudges alone. While affect-based design elements (e.g., friendly language, trust badges, or reassuring visuals) may increase short-term disclosure, they risk eroding long-term trust if they systematically create misalignment with users’ cognitive appraisals. Sustainable privacy management requires affect–cognition coherence, not affective suppression of concern.

9. Conclusions

This study examined how cognitive–affective inconsistency and attitudinal ambivalence jointly moderate the relationship between privacy concerns and information disclosure, drawing on Protection Motivation Theory (PMT). By distinguishing between threat and coping appraisals, the analysis revealed that inconsistencies between cognitive evaluations and affective responses significantly weaken the predictive power of privacy concerns on disclosure behavior. Notably, the moderating effect of cognitive–affective inconsistency was stronger than that of attitudinal ambivalence, underscoring the dominant role of emotional misalignment in privacy decision-making. In contrast, the negligible effect of ambivalence between response efficacy and privacy risk suggests that individuals rely more on self-efficacy—an internally accessible belief—than on external efficacy perceptions when evaluating privacy-related coping capacity. Furthermore, the finding that ambivalence reduces cognitive–affective consistency highlights how internal cognitive tension diminishes the stability and accessibility of evaluative beliefs.

Theoretically, this study contributes to privacy research by integrating emotional and cognitive dimensions within the PMT framework, demonstrating that internal misalignments can erode the strength of privacy attitudes. It also bridges theories of attitude structure and decision-making, offering an insightful account of how ambivalence and inconsistency interact to shape behavioral outcomes. Practically, the findings emphasize the importance of reducing emotional–cognitive misalignment in privacy communication and interventions, such as through affectively aligned risk messaging and efficacy-enhancing education strategies.

Although this study advances understanding of internal tensions in privacy decision-making, several limitations warrant consideration. First, the cross-sectional design precludes causal inference. Although the theoretical model posits that cognitive–affective inconsistency and ambivalence attenuate the influence of privacy concerns, longitudinal or experimental designs are needed to establish temporal precedence and dynamic interplay among these constructs. Second, cognitive–affective inconsistency was measured using an absolute difference score, a common approach in inconsistency research [30], but this method may reduce reliability, obscure the direction of inconsistency, and fail to distinguish qualitatively different psychological states. Future research could adopt alternative operationalizations, such as latent interaction or residual-based approaches. Third, reliance on self-reported measures may introduce social desirability or recall bias. Participants might underreport disclosure behavior or overstate privacy concerns due to normative pressures. Future research could incorporate behavioral trace data (e.g., actual disclosure logs) or implicit measures (e.g., response latency, physiological arousal) to enhance validity. Fourth, the sample was drawn from U.S.-based MTurk workers, limiting generalizability. Cultural variations in privacy valuation, trust in institutions, and emotional expressiveness e.g., [52] may moderate the observed effects. Cross-cultural replications, particularly in collectivist or high-uncertainty-avoidance societies, are essential. Finally, the study focused on e-commerce and social platform contexts. Privacy dynamics may differ in high-stakes domains such as healthcare, finance, or government services, where perceived risk and coping resources vary significantly. Extending the model to these contexts would strengthen its external validity.