YouTube Recommendation Algorithm Perception, AI Regulation Support, and Digital Competence: A Three-Wave Panel Study of Bidirectional Effects and Moderation Effects

Bidirectional Causality between Algorithm Perception and Regulation Support

The emergence of regulations is based on positive or negative perceptions of the regulation subject. Regulations can be enacted to prevent adverse effects or malfunctions, such as those in the media, even without user demand or social discourse. Typically, however, regulations undergo a process in which regulatory discourse, such as public opinion, is formed based on social demand arising from negative user experiences with the regulated subject. Only then are they regulated (Lunt & Livingstone, 2012). For example, if users directly experience issues with YouTube’s recommendation algorithm while using the platform and negative perceptions gradually accumulate, demands for the algorithm’s regulation could rise. Thus, regulation can be enacted through a “bottom-up process” that begins with user awareness of the problems, progress through social discourse, and reaches a point where is regulation in necessary.

According to prior studies, users who experienced the opacity of algorithms provided by platforms perceived the algorithm negatively (Eslami et al., 2019; Gagrčin et al., 2024). Users’ trust in recommendation systems varies significantly depending on their perception of algorithms on social media and other platforms (Wu et al., 2024). Negative perceptions of algorithm manipulation have also triggered resistance behaviors toward algorithms (Hu et al., 2024; Lin, 2025). This discussion further suggests that a forward effect operates between users’ media perceptions and regulations. A forward effect is a general, logical causal relationship in which attitudes form in the direction of one’s experience. For example, people who experience traffic accidents tend to support safety regulations. It should be noted that the forward effect prediction rests on an assumption that warrants explicit acknowledgement. That is because users’ algorithm perceptions do not systematically improve between waves. This assumption is plausible given that algorithmic harm concerns dominated policy and public discourse throughout this period (Cheong, 2024; Shim & Park, 2024), but it represents a boundary condition in the predicted direction.

The forward effect explains the relationship between YouTube’s recommendation algorithm and related regulations. As described, the forward effect clearly illustrates how negative experiences with the algorithm lead to negative attitudes, and regulation support, and result in calls for regulation. However, there are still theoretical gaps within the forward effect. The forward effect cannot rule out the possibility that regulatory discourse, such as policy debates, might reshape users’ perceptions of the recommendation algorithm instead. In other words, the forward effect fails to account for a reverse effect, whereby users’ expectations regarding new institutional introductions could lead them to perceive the platform positively.

The theoretical gap is becoming more problematic in light of the recent policy trends. From 2022 to 2025, the regulation of algorithms became a central focus of major policy discussions worldwide (OECD, 2024). If policymakers translate public concerns into policy immediately without considering how regulation support toward recommendation algorithms could alter users’ perceptions of these platforms, they are essentially relying on a one-way feedback loop, known as the forward effect. Conversely, users’ changing attitudes toward algorithm regulation can influence their perception of algorithms. This could create an alternative feedback route: a top-down process where regulation expectations lead to better recognition of recommendation algorithms through a reverse effect. For example, GAPI, a global AI cooperation organization, has elevated algorithm regulation from a technical issue to a matter of global consensus by emphasizing risk-based regulation, as seen in the EU AI Act. It’s an example of the reverse effect. The reverse effect prediction assumes that regulatory expectations are at least partially met. Where expectations go unmet, null or negative perceptual updating remains plausible. This condition is more likely to be satisfied for transparency regulation support, whose effects are user-facing, than for accountability regulation support, whose institutional mechanisms remain invisible to the users.

How Algorithm Perception Drives Regulation Support

We have seen that users may have a negative experience with media, and that this experience can lead to demands for related regulations. So, when users perceived YouTube’s recommendation algorithm negatively, what regulation responses to they demand or support? Regulations concerning AI systems, such as recommendation algorithms, primarily address algorithmic transparency and accountability principles (Mensah, 2023). In this study, we use the term regulation support to refer to users’ normative beliefs about whether platform providers should implement transparency or accountability actions, with transparency regulation support and accountability regulation support as the two sub-constructs examined.

These two regulation principles are functionally distinct. Transparency regulation support refers to obligations on platform providers to disclose algorithmic decision-making processes, data usage practices, and recommendation criteria (Larsson & Heintz, 2020). By directly addressing the information asymmetry that underlies negative algorithm perception, transparency enables users to understand why specific content is recommended and how their personal data informs those decisions (Zarouali et al., 2021). Accountability regulation support, by contrast, entails governance principles that require platforms to be answerable to external bodies through audits, complaint systems, and periodic reporting (Reynolds & Hallinan, 2024). Unlike transparency, which enables users to direct observation through visible changes, accountability operates through indirect systems through institutional intermediaries whose remain invisible to users at the user interface level (Ferrari et al., 2025; Goodman & Trehu, 2022). In this sense, accountability regulation support’s effect is real but experienced indirectly, if at all, by ordinary users. More importantly, this functional asymmetry would directly predict earlier reverse effects for transparency regulation support, but moder delayed effects for accountability regulation support.

This distinction suggests that negative perceptions of YouTube’s algorithm may prompt demands for transparency and accountability, which operate different principles. Thus, even identical complaints may lead to different policy demands. Although user demands for regulation may differ depending on the type of regulation, these demands are generally triggered by users’ awareness of the harm caused by algorithms. It is true that these demands are emerging. Algorithm harms—including filter bubbles, misinformation, and privacy violations—create conditions where negative perceptions translate into regulation support. This is the point where users’ negative perceptions of YouTube can translate into support for regulation. This would create a forward effect in the context of YouTube’s recommendation algorithm. Thus, this study poses the following research question and proposes research hypotheses.

• RQ1. Does negative YouTube algorithm perception at Time t predict increased support for (a) transparency regulation and (b) accountability regulation support at Time t+1?
• H1a. Negative YouTube algorithm perception at Time t will positively predict transparency regulation support at Time t+1.
• H1b. Negative YouTube algorithm perception at Time t will positively predict accountability regulation support at Time t+1.

How Regulation Support Shapes Algorithm Perception

This study focuses on the reverse effect that follows the forward effects. This reverse effect is based on top-down process. This effect forms the field of cognitive psychology. Gregroy (1974) explains, human cognition is not merely a passive process of receiving sensory input but an active process of construction. Individuals with strong expectations about a system perceive that these expectations significantly influence their evaluation of the system (Clark, 2013; Raftopoulos, 2001). Similarly, regulation support such as transparency or accountability can make a platform’s recommendation algorithm appear more positive when expectations are met. This occurs through the users’ active interpretive process rather than mere sensory reception.

Do all types of regulation reconstruct perceptions of recommendation algorithms in the same way, then? More specifically, do user expectations regarding transparency and accountability regulations lead to the same evaluations of algorithm perception? As discussed above, Ferrari et al. (2025) highlighted a potential asymmetry between the effects of transparency and accountability, emphasizing the inherent observability of these two-regulation types. When users support transparency regulations and increase their level of scrutiny of platforms, they can perceive the tangible effects of these regulations through changes to the interface, enhanced privacy protections and public statements. This tangible experience enables positive perceptions of the algorithm to be updated based on the evidence users observe and their expectations of transparency.

In comparison, accountability regulations make it difficult for users to experience the substance of the regulations first-hand. Those supporting accountability regulations cannot observe whether complaint handling systems function properly, whether algorithmic audits are initiated according to standards or whether platforms face penalties such as fines for violating user protection regulations. Consequently, unlike with transparency, users who support accountability may fail to update their positive perception of the algorithm if there is no observable evidence. This is because their expectations of accountability and the available evidence do not align, even if substantial regulation progress has been made. The following research questions and hypotheses were established by the study to empirically demonstrate the reverse effect, in line with the discussion.

• RQ2. Does support for (a) transparency regulation and (b) accountability regulation at Time t predict improved algorithm perceptions at Time t+1? Do effects differ between transparency and accountability? Do effects differ between transparency and accountability regulation support?
• H2a. Transparency regulation support at Time t will positively predict YouTube algorithm perception at Time t+1.
• H2b. Accountability regulation support at Time t will show weaker effects on YouTube algorithm perception at Time t+1 compared to transparency regulation support.

Digital Competence as Multidimensional Construct and the Role of Moderation

As discussed earlier, research on digital competence requires breaking down users’ technical abilities into specific dimensions rather than consolidating them into a single, unified concept. The European Parliament defines digital competence as ‘the ability to use information society technologies confidently and critically for work, leisure and communication’ (European Parliament & Council of the European Union, 2006, p. 16). Given the diverse application areas of digital competence, the development of digital competence tailored to these various domains is also required. Hwang et al. (2022) refined digital competence within the Korean context, delineating it into seven dimensions: basic technical skills, everyday use, critical understanding, sharing and production, social participation, rights protection, and security competence.

This study positions digital competence as a moderator rather than an antecedent of algorithm perception, for two reasons. First, digital competence is a relatively stable individual characteristic that exists prior to and shapes how users interpret algorithmic experience. It is not caused by those experiences. Second, the main question of this study is not whether digital competence affects regulation support, but for whom and under what conditions the relationship between algorithm perception and regulation support operates. This is a boundary condition question for which moderation is the theoretically appropriate specification (Valkenburg et al., 2016). Each dimension of digital competence is linked to a specific moderating principles—coping resource availability (rights protection), sensitivity to algorithmic harm cues (general competence), and capacity to interpret platform-level evidence (critical understanding). That predicts not whether but how strongly and in which direction the perception-attitude relationship manifests across competence levels.

Moderating Effect of General Digital Competence

General digital competence encompasses the everyday use and basic understanding of technologies such as smartphones and Internet, rather than specialized expertise in a particular field (Hwang et al., 2022). This general digital competence can help people understand and respond to issues with recommendation algorithms and engage with regulatory discourse. For forward effects, high digital competence enables users to recognize subtle algorithmic issues that low digital competence users might miss (Sundar & Marathe, 2010). Technical experience can also trigger negative machine heuristics when algorithm malfunction (Molina & Sundar, 2024), reinforcing pathway from negative perceptions to regulation demands.

Correspondingly, the relationship between regulation and perception restructuring may manifest divergent patterns depending on the digital competence gap, not only in the forward direction but also in the reverse direction. Users with high general digital competence can alter their perception of regulation, either positively or negatively, based on their ability to monitor recommendation algorithms. Prior studies ( Jung & Park, 2024) have shown that digital competence is associated with positive attitudes towards new technologies. This suggests that competent users can engage with regulation discussions using an interpretative framework that considers how regulation could serve to improve recommendation algorithms. When users with high digital competence back transparency or accountability regulations, they might link these with better recommendation algorithms, in turn, positively rethink recommendation algorithms based on these expectations. The following research questions and hypotheses were established from this discussion.

• RQ3a. Does general digital competence moderate (a) forward effects (algorithm perception → regulation support) and (b) reverse effects (regulation support → algorithm perception)?
• H3a-1. General digital competence will moderate the effects, so negative perceptions of the algorithm will more strongly predict regulation support (transparency and accountability) among high-competence users than low-competence users.
• H3a-2. General digital competence will moderate reverse effects, so regulation support more strongly predict improved algorithm perceptions among high-competence users than low-competence users.

Moderating Effect of Right Protection Competence

The concept of right protection competence is similar to the idea of external political efficiency, which is defined as the belief that the government will respond to citizens’ political demands (Gil de Zúñiga et al., 2017). An individual’s experience of expressing political demands through institutional channels like elections or petitions, and seeing outcomes reflected in reality, plays a key role in shaping external political efficacy.

In this study, the ability to protect rights is tied to users’ trust in recommendation algorithm regulation and their belief that their demands can prompt changes to theses algorithms, much like external political efficacy. Capacity for rights protection encompasses knowledge of regulation compliant systems (e.g., media review committee), the ability to report phishing, fraud and copyright infringement, an understanding of rights protection processes (Hwang et al., 2022; Shim & Park, 2024). Users with higher levels of this competence can hold platforms to account for algorithmic issues and their consequences more effectively than users with lower competence through regulatory discussion.

The capacity for right protection is expected to have different moderating effects on forward and reverse effects. For forward effects, when users experience negative algorithms, their perceived vulnerability to recommendation algorithms may motivate regulatory demands. This may be independent of institutional knowledge. Protective Motivation Theory suggests that coping appraisal (response efficacy and self-efficacy) predicts protective behavior more strongly than threat appraisal (Floyd et al., 2000; Tsai et al., 2016). Users with limited rights protection competence lack alternative coping strategies beyond systemic regulatory demands. When coping resources are scarce, protection motivation intensifies (Rippetoe & Rogers, 1987). In contrast, users with high rights protection competence can access alternative coping strategies, which could reduce their need for systemic regulation change. Consequently, the psychological pathway from negative perceptions of recommendation algorithms to support for regulation may be weakened. Importantly, this coping-based mechanism operates regardless of regulation type. Unlike critical understanding competence which differentially amplifies forward effects depending on the observability of transparency versus accountability violations, rights protection competence shapes the intensity of regulatory demand through coping resource availability rather than through the interpretive capacity required to diagnose specific regulatory failures. Users with low rights protection competence would lack individualized coping strategies across both regulatory fields, making systemic regulation support their primary available response to perceived algorithmic harm in both transparency and accountability contexts. Accordingly, the amplifying effect of low rights protection competence on forward effects is expected to be consistent across regulation types.

In the reverse effect, rights protection competence plays a key moderating role in regulating problem recognition and regulation support. Studies on external political efficacy suggest that individuals’ political participation and evaluations of institutions improve as their trust in institutions and belief in their responsiveness increases (Balch, 1974; Pollock, 1983). In a similar vein, users with high rights protection competence expect their regulatory demands to compel platform changes. This expectation could generate an expectancy effect, improving perceptions of current algorithms even before actual institutional improvements occur. Based on the above discussion, this study established following research questions and hypotheses.

• RQ3b. Does rights protection competence moderate (a) forward effects and (b) reverse effects differently?
• H3b-1. Forward effects may be stronger among low rights protection competence users who lack alternative coping processes beyond demanding systemic regulation change.
• H3b-2. Rights protection competence will positively moderate reverse effects, so regulation support (transparency and accountability) predicts improved algorithm perceptions among high rights protection users, but not among low rights protection users.

Moderating Effect of Critical Understanding Competence

Critical understanding competence is the ability to evaluate information and identify reliable sources, as well as spot deliberate manipulation (Hwang et al., 2022). It involves assessing information for reliability, discerning misinformation from truthful content (Koh, 2024). Recent algorithm literacy studies suggest that critical understanding enables users to mitigate risks and maximize opportunities when using the media. This involves cognitive, emotional, and behavioral skills (Gagrčin et al., 2024).

Critical understanding competence can moderate forward and reverse effects, like rights protection competence. But the moderating effect patterns for these two effects may differ due to differences in observability. Regarding the forward effect, the moderating effect of critical understanding competence will differ across transparency and accountability regulation. The asymmetric moderating pattern predicted in H3c-1 follows from differences in the observability of transparency versus accountability violations. Transparency violations, such as privacy issues manifest concrete, interface-level anomalies that users can encounter and recognize without specialized interpretative capacity (Eslami et al., 2019; Lin, 2025). Because even users with relatively low critical understanding competence can identify obvious transparency failures, competence offers limited additional leverage on the relationship between negative algorithm perception and transparency regulation support. The forward effect for transparency is therefore expected to operate uniformly across competence levels.

Accountability violations present a basically different interpretive challenge. Accountability principles operate through institutional channels that are often invisible to ordinary users, making failure difficult to detect (Cheong, 2024; Goodman & Trehu, 2022). Recognizing such failures requires metacognitive awareness, which is the ability to perceive not only the occurrence of harm but also the breakdown of institutional oversight (Dogruel et al., 2022). Consistent with Molina and Sundar (2024) finding that sophisticated users are more likely to attribute algorithmic harm to platform oversight failure rather than mere technical error, we argue that critical understanding competence meaningfully amplifies the forward effect specifically in the accountability regulation support. Users with higher critical understanding competence are better equipped to make consequently more likely to translate negative algorithm perception into accountability regulation support.

For the reverse effect, users must detect changes on the platform to form attitudes toward transparency regulations (Reynolds & Hallinan, 2024). This includes explanations for recommendations, clear privacy disclosures and content labelling. Users with high critical understanding competence can monitor these changes and verify improvements. Based on previous studies, user’s high algorithm literacy discovers differences or the absence of content in their news feeds through interaction with others (Rader & Gray, 2015; Velkova & Kaun, 2021), deepening their critical reflection through participation in communities of practice (Cotter, 2024; Morris, 2023).

In the context of accountability, the reverse effect may also lead to greater disparity based on a lack of understanding due to constraints on observability. Users have limited means of directly checking whether complaint handling systems are functioning or audits are being conducted. However, those with a strong critical understanding can gather clues about accountability improvements through sources of information such as corporate blogs and traditional media reports (Bishop, 2020).

On the other hand, users with low critical understanding competence may miss these indirect signals and lack the cognitive pathways to connect possible improvements in accountability with their attitudes towards recommendation algorithms. With this discussion, this study established following research questions and hypotheses.

• RQ3c. Do critical understanding competence moderate (a) forward effects and (b) reverse effects differently, and do these patterns vary between regulation support?
• H3c-1. Critical understanding competence will not significantly moderate transparency-related forward effects but will positively moderate accountability-related forward effects.
• H3c-2. Critical understanding competence will positively moderate reverse effects (both transparency and accountability), so regulation support predicts improved algorithm perceptions among high critical understanding users but not among low critical understanding users.

Participants and Procedure

This study used three-wave national panel data from the Intelligent Information Society User Panel Survey (Korea Communications Commission (KCC) and Korea Information Society Development Institute (KISDI), 2022 to 2024).¹ The survey employed stratified multi-stage sampling of Korean internet users and smartphone users aged over 20. The baseline survey was in August 2022 (Wave 1, N = 5,378), with follow-up surveys in September 2023 (Wave 2, N = 4,581), and August 2024 (Wave 3, N = 4,437). Analyses retained participants completing all waves with complete responses (N = 2,269, retention rate = 42.2%).

The sample included 44.5% male, 55.5% female, median age of 40-49 years. Education varied: high school or below (37.4%), college degree (61.7%), graduate degree (.9%). Monthly income was also varied: none (2.0%) to 9 million KRW or more (15.6%), median 3-4 million KRW. Sample characteristics mirror national statistics. The broad representativeness of the panel is supported by the near-universal smartphone penetration among Korean adults. According to a nationlally representative survey (Korea Gallup, 2022), smartphone usage reached 97% overall (male: 99%, female: 95%), and 100% among those aged 18–59, suggesting that access to internet-based surveys is not systematically stratified by gender or age in this population.

Attrition Analysis

The analytic sample (N = 2,269; retention rate = 42.2%) closely mirrors the original Wave 1 (N = 5,378). Gender distributions were similar (analytic: 44.5% male; original: 44.1%; X² (1) = .18, p = .670) and education distributions were comparable. Two small but significant differences emerged. The analytic sample slightly underrepresents the youngest and oldest age groups (X² (5) = 17.98, p = .003) and reports marginally lower income (analytic: M = 5.50; original: M = 5.64; t = -2.06, p = .039). Given the small magnitude of these differences and the inclusion of demographic controls in all analyses, attrition is unlikely to substantially bias the findings.

Measures

All measures used 5-point Likert scales (1 = strongly disagree to 5 = strongly agree for attitude items; 1 = very poor to 5 = very good for competence items) unless otherwise noted.

Analysis

Cross-lagged panel models examined reciprocal relationship between YouTube algorithm perception and regulation support. Separate models for Period 1 (2022-2023) and Period 2 (2023-2024) estimated: (a) forward effects (perception → attitudes; RQ1), (b) reverse effects (attitudes → perception; RQ2), and (c) autoaggressive paths. Hierarchical regression models included demographic controls (age, gender, education, and income). Continuous variables were standardized (M = 0, SD = 1).

For moderation (RQ3), median- split comparisons divided respondents into high and low groups for general digital competence, rights protection, and critical understanding. Z-tests compared coefficients between groups. Moderator values were time-matched to the predictor. That is, digital competence measured at Wave 1 (T1) was used for Period 1 analyses (T1 → T2), and the digital competence measured at Wave 2 (T2) was used for Period 2 analyses (T2 → T3). This approach ensures temporal ordering consistency and reduces potential confounding from changes in moderator values over time.

Descriptive Statistics and Correlations

Table 1 shows the means, standard deviation, and correlations. YouTube perception increased from 2022 (M = 3.29, SD = .37) to 2023 (M = 3.38, SD = .39), then declined in 2024 (M = 3.32, SD = .33). Transparency and accountability attitudes increased linearly, though growth decelerated (Period 1: +2.5% and +2.2%; Period 2: +0.8% and +1.4%).

Table 1.

Descriptive Statistics and Correlations(N = 2,269)

Within-wave correlations remained weak to moderate (transparency: r = .23-.37, average r = .28; accountability: r = .15-.35, average r = .25), peaking in 2023 then declining by 2024.

Meanwhile, across-wave stability revealed moderate stability. Specifically, YouTube perception (r = .20 in 2022-2023 and r = .33 in 2023-2024), transparency attitudes (r = .19 in 2022-2023 and r = .17 in 2023-2024), and accountability attitudes (r = .13 in 2022-2023 and r = .22 in 2023-2024) showed moderates stability. These patterns indicate meaningful stability within individuals alongside substantial changes over time, which are ideal conditions for cross-lagged panel analysis (Hamaker et al., 2015).

Algorithm Perception Predicting Regulation Support (RQ1)

To test if negative algorithmic experience drives regulatory support, we examined forward effects from YouTube perception to regulation support across two periods.

The transparency pathway (H1a) operated immediately in Period 1. Contrary to H1a, Individuals who viewed YouTube’s algorithm more positively in 2022 expressed stronger support for transparency regulation support the following year (β = .056, SE = .021, p < .001). While the hypothesis predicted that negative perceptions would drive regulation demand, the significant positive coefficient indicates the opposite direction. This effect nearly doubled by Period 2, with perception in 2023 predicting attitude in 2024 at β = .107 (SE = .022, p < .001), a 91% increase from Period 1. Thus, H1a was not supported in the predicted direction.

The accountability pathway (H1b) showed different timing. During Period 1, algorithmic experience had no impact on accountability attitudes (β = .017, SE = .022, n.s.), rejecting the hypothesis for this early phase. The pathway emerged only in Period 2, where more positive perception in 2023 significantly predicted stronger accountability support in 2024 (β = .080, SE = .022, p < .001). The pathway emerged only in Period 2, where more positive perception in 2023 significantly predicted stronger accountability regulation support in 2024 (β = .080, SE = .022, p < .001). This represents a 371% increase from Period 1.

Regulation Support Predicting Algorithm Perception (RQ2)

We next tested whether regulation support shape how people perceive algorithmic behavior, examining reverse effects that would indicate top-down cognitive processing. For transparency (H2a), this pathway was absent during Period 1. Attitudes in 2022 showed no relationship with subsequent perception in 2023 (β = .021, SE = .021, n.s.), However, Period 2 revealed a striking emergence: transparency attitudes in 2023 predicted improved YouTube perception in 2024 (β = .074, SE = .021, p < .001), a 252% increase from Period 1.

The accountability pathway (H2b) traced the opposite trajectory. During Period 1 individuals with stronger accountability attitudes perceived YouTube more favorably the following year (β = .046, SE = .021, p < .05), supporting H2b for this period. By Period 2, this effect vanished (β = .005, SE = .021, n.s.), a 90% decline from Period 1.

How Digital Competence Shapes These Relationships (RQ3)

General Digital Competence (RQ3a)

We also tested if digital competence moderates both causal directions, entering high versus low competence groups based on median splits.

Period 1 rejected H3a-1. Low-competence users showed stronger forward effects (transparency β = .124, SE = .042, p < .01; accountability β = .139, SE = .043, p < .01) than high-competence users (transparency β = .024, SE = .025, n.s.; accountability β = -.040, SE = .024, n.s. ; z = -2.06, p < .05 and z = -3.60, p <.001). This asymmetry disappeared entirely by Period 2. Both groups now showed significant positive forward effects (high competence: transparency β = .122, SE = .030, p < .001, accountability β = .099, SE = .031, p < .01; low competence: transparency β = .095, SE = .031, p < .01, accountability β = .062, SE = .032, n.s.), with no meaningful difference between them (z = .62 and z = .84, both n.s.).

Table 2.

Cross-Lagged Effects between YouTube Algorithm Perception and AI Regulation Support Across Two Time Periods

H3a-2 was rejected due to non-significant moderation across both periods. For transparency, high-competence users showed β = .067 (SE = .030, p < .05) compared to low-competence users’ β = -.002 (SE = .031, n.s.). The group difference between high-competence and low-competence users was not significant (transparency reverse: z = 1.60, n.s.). For accountability, high-competence users showed β = .072 (SE = .030, p < .05) versus low-competence β = .046 (SE = .032, n.s.), again with non-significant moderation (accountability reverse: z = .61, n.s.).

Period 2 continued this pattern of non-significant moderation, providing only marginal evidence for H3a-2. First, for transparency reverse effects, high-competence users showed β = .094 (SE = .030, p < .01) compared to low-competence users β = .035 (SE = .031, n.s.), with group differences z = 1.38, n.s,. Second, for accountability reverse effects, high competence users showed β = .031 (SE = .030, n.s.) versus low-competence users β = -.033 (SE = .031, n.s.), with z = 1.48, n.s.

Table 3.

Digital Competence Moderation Effects (Period 1: 2022-2023)

Rights Protection Competence (RQ3b)

Testing whether awareness of digital rights operates differently than general digital competence produced distinct temporal patterns.

Period 1 strongly supported H3b-1. Low-competence users showed stronger effects (transparency: β = .133, SE = .032, p < .001; accountability: β = .094, SE = .032, p < .01) while high-competence users showed negative effects (transparency: β = -.018, SE = .028, n.s., accountability: β = -.060, SE = .029, p < .05), creating highly significant group differences (z= -3.56, p < .001 for both). Period 2 showed complete convergence. Both groups showed positive effects (high: transparency β = .120, SE = .029, p < .001, accountability β = .085, SE = .029, p < .01; low: transparency β = .101, SE = .031, p < .01, accountability β = .078, SE = .031, p < .05) with moderation disappearing entirely (z = .43 and z = .16, both n.s.), now supporting the original prediction.

The reverse direction showed temporal evolution. H3b-2 was rejected in period 1 but supported in period 2. Period 1 showed no meaningful moderation. For transparency, high-competence users showed β = .064 (SE = .029, p < .05) compared to low-competence users’ β = .027 (SE = .030, n.s.), but the group difference was not significant (transparency: z = .88, n.s.). For accountability, high-competence users showed β = .052 (SE = .029, n.s.), with no significant moderation (z = -.55, n.s.). By Period 2, however, clear asymmetric patterns emerged. High-competence users demonstrated substantial reverse effects (transparency: β = .137, SE = .030, p < .001; accountability β = .077, SE = .030, p < .05) while low-competence users showed weak or negative effects (transparency: β = .035, SE = .031, n.s., accountability: β = -.045, SE = .031, n.s.), creating significant moderation (z = 2.36, p < .05 and z = 2.85, p < .01).

Table 4.

Digital Competence Moderation Effects (Period 2: 2023-2024)