Where is TikTok Scrolling Us To? - The Cognitive Reshaping of Taiwan's Democratic Politics through Active TikTok Use

Eric Hsu, Researcher, Doublethink Lab

Background

As a rapidly rising short-video social media platform in recent years, TikTok has profoundly impacted the global information ecosystem due to its vast user base and highly engaging content delivery mechanisms. For a long time, many democratic nations have expressed significant concerns regarding the platform’s cybersecurity risks, algorithmic control capabilities, and potential cooperation with authoritarian regimes such as the People’s Republic of China (PRC) and Russia in cognitive warfare.

The short-video information format promoted by TikTok has also significantly altered the information consumption habits of Taiwanese users. Its fragmented and fast-paced dissemination model has amplified information manipulation techniques, including conspiracy theory narratives, further influencing users’ cognitive biases through narrative construction. This creates a more receptive environment for certain ideological perspectives aligned with united front strategies to subtly enter public discourse, helping advance those agendas.

Doublethink Lab has previously released the 2025 Taiwan TikTok User Study: Nationwide Online Surveys — Research Data, with detailed research methods, survey implementation details, and raw data available in the data announcement report.

This comprehensive analysis report focuses on the national online survey data. It uses statistical methods such as cross-analysis, correlation analysis, latent class analysis, and counterfactual inference. The goal is to systematically explore the relationships and potential impacts of active TikTok usage on media consumption habits, views on public and social issues, perceptions of cross-strait and international politics, and other demographic and psychological variables. It aims to progressively address the core question of this project:

What effects and shifts does active TikTok use actually cause?

Key findings

• How often people use TikTok relates in complex ways to their mental health and political views. The clearest pattern is that heavier TikTok users report higher signs of internet addiction.
• When it comes to views on Taiwan’s domestic issues (like concerns about judicial injustice or lack of free speech) and unification with the PRC, agreement with these ideas rises as TikTok use increases, but then drops among the heaviest users. This creates a clear “inverted U-shaped” pattern, showing TikTok’s influence grows up to a point before declining.
• Attitudes such as “PRC is more livable,” “economic interests outweigh national security,” and “cross-strait unification is inevitable” show no significant differences between low-to-extremely high TikTok users but are notably more agreed upon compared to non-users. This indicates a significant binary divide on these issues between TikTok users and non-users.
• TPP supporters exhibit stronger negative sentiments toward the ruling party than KMT supporters do, especially among TikTok users.
• DPP supporters largely believe the KMT benefits from PRC united front tactics, while KMT supporters are split between thinking it benefits either party. TPP supporters overwhelmingly believe the DPP benefits, possibly linked to the “Green-Communist” narrative.
• Most KMT supporters agree Taiwan faces problems like judicial unfairness and lack of free speech, but active TikTok users among them show a significantly higher proportion of “strongly agree” responses compared to inactive users, indicating usage frequency influences the intensity of views.
• Concerns over Taiwan’s judicial system are rising among DPP supporters, possibly due to recent political and social controversies such as perceived interference in the judicial system or judicial outcomes in cases involving child pornography and child abuse, warranting further investigation.
• While most reject statements such as “PRC is more suitable to live” and “economic interests outweigh national security,” TPP supporters are more likely to strongly agree.
• Skepticism about U.S. support for Taiwan has grown mildly among DPP supporters since 2024, with KMT and TPP supporters also showing high skepticism.
• TPP supporters’ dissatisfaction with Taiwan’s democracy and the judicial system, and agreement with statements skeptical of the U.S. now approach or surpass KMT supporters’, indicating possible shifts in political attitudes.
• Most supporters reject sacrificing democracy for cross-strait peace; DPP opposes this most strongly, followed by KMT and TPP.
• Most DPP supporters expect life to change if unification occurs, but two-thirds of active TikTok users in the KMT and TPP camps believe daily life would remain unchanged. Among TPP supporters, the proportion who “strongly agree” is significantly higher than that of KMT supporters, suggesting that some TPP supporters hold a stronger belief that life would remain unchanged after unification.
• Willingness to Fight for Taiwan if PRC invades: After accounting for political affiliation, willingness to fight ranks as DPP, KMT, and TPP supporters. Regardless of TikTok activity, 80–85% of DPP supporters are willing, mostly “strongly willing.” KMT supporters show a more moderate stance, while TPP supporters are slightly less willing, with 30–40% significantly more likely to be “strongly unwilling.”
• Active TikTok users show stronger cross-issue correlations — linking dissatisfaction with domestic issues, more PRC-friendly views on cross-strait relations, skepticism toward the U.S., openness to PRC narratives, more acceptance of ideas like unification inevitability and sacrificing democracy for peace. This suggests TikTok use may shape a distinct ideological and cognitive framework.
• Using Latent Class Analysis, LCA, respondents were categorized into four political attitude groups based on system recognition, perceptions of PRC, views on unification, and value orientations: Status Quo Moderates (36%), Populist Skeptics (18%), Democratic Nativists (24%), and Pro-PRC Stabilizers (22%). These proportions reflect the sample’s distribution and do not represent Taiwan’s overall societal structure.
• Age and Gender Differences: Pro-PRC Stabilizers have the lowest average age (37.3 years), challenging stereotypes of older pro-PRC leanings. Democratic Nativists have the highest female proportion (56.9%), while Populist Skeptics have the lowest (34.7%).
• Party Identification and Attitudes: 64% of Status Quo Moderates lack party preference, indicating political alienation. Pro-PRC Stabilizers and Populist Skeptics show the strongest aversion to the DPP, echoing the notion that “Taiwan’s largest party is the anti-DPP party.”
• Status Quo Moderates have the weakest democratic beliefs, with nearly 30% believing dictatorship may be better in certain contexts. Democratic Nativists have the strongest democratic convictions, with over 80% committed to democratic systems.
• Pro-PRC Stabilizers and Populist Skeptics report higher levels of loneliness and internet addiction, indicating relatively negative psychological states. Life satisfaction and happiness, however, differ very little between groups.
• After controlling for demographic variables, psychological states, and attitudes toward political and public issues, active TikTok use still independently increases the likelihood of aligning with the Pro-PRC Stabilizers (by 2.53%) and Populist Skeptics, while reducing alignment with Democratic Nativists (by 2.59%) and Status Quo Moderates. This demonstrates that active TikTok use directly influences political attitudes, undermining confidence in Taiwan’s system and democracy.

Relationship between TikTok usage frequency and different variables

Specifically, we assigned the following values for weekly usage frequency: “never” as 0 times/week, “rarely” as 1 time/week, “several times a week” as 3 times/week, “once a day” as 7 times/week, and “several times a day” as 21 times/week. Simultaneously, we defined session durations as follows: 10 minutes or less as 10 minutes, 10–30 minutes as 20 minutes, 30 minutes to 1 hour as 45 minutes, 1–3 hours as 120 minutes, and over 3 hours as 180 minutes. Researchers calculated each respondent’s total weekly TikTok usage in minutes over the past year and categorized users into different levels: 60 minutes or less as “non-users,” 61–180 minutes as “low users,” 181–360 minutes as “moderate users,” 361–840 minutes as “high users,” and over 841 minutes as “extremely high users.” The effective sample sizes for each usage level are provided below.

Researchers initially conducted correlation analyses between TikTok usage frequency and questionnaire items related to internet addiction, various issue perspectives, and well-being measures. Consistent with the 2024 presidential and legislative election analysis, moderate to high correlations were observed between satisfaction with Taiwan’s democracy, certain domestic livelihood narratives with traces of information manipulation, and U.S. skepticism. However, TikTok usage frequency showed only a weak correlation with these variables, indicating a notable discrepancy. Does this suggest that TikTok usage frequency is largely unrelated to users’ agreement with these perspectives and narratives?

Researchers firstly hypothesized that one possible reason is that the relationships between variables may not be linear but rather exhibit non-linear, segmented, or ceiling effects (e.g., correlations stabilize or decline from moderate to high usage). This pattern could manifest as an inverted U-shape (rising then falling) or a plateau effect (no further increase beyond a certain level), which aligns with real-world social media usage behaviors and is entirely plausible in this TikTok usage frequency analysis. Such non-linearity could affect the Pearson correlation coefficients, which assume linear relationships.

To test this, researchers examined the statement “Q16: Taiwanese society has a very serious issue of judicial injustice” across different levels of TikTok users, analyzing differences in means and medians. They found an increasing trend from “almost never use” to “high users,” followed by a decline from “high users” to “extremely high users,” supporting the hypothesis of a non-linear relationship between TikTok usage frequency and various variables. Next, we will further validate this by examining differences in means and medians across different levels of TikTok users for other questions or question groups.

This study analyzes TikTok usage frequency in relation to psychological well-being and attitudes toward political and public issues, identifying four patterns. Significant differences are denoted using the Wilcoxon rank sum test with Bonferroni correction.

◎ Linear Relationship: Higher TikTok usage frequency is associated with higher scores for internet addiction risk (positive correlation).

In the internet addiction question group, Pearson correlation analysis shows moderate to high significant positive correlations between Q5 and Q8. TikTok usage frequency also exhibits mild to moderate positive correlations with each question in the internet addiction group. This indicates that higher TikTok usage frequency is associated with a greater risk of internet addiction.

◎ Inverted U-Shaped Relationship: The agreement scores for the following narratives rise with increasing TikTok usage frequency up to a peak and then decline (inverted U), with high users (6–14 hours per week) as the turning point.

• Agreement with “Taiwanese society has a very serious issue of judicial injustice”
• Agreement with “the Constitutional Court’s interpretations must align with the expectations of the broader Taiwanese public,”
• Agreement with “The current ruling Democratic Progressive Party (DPP) is no different from the Chinese Communist Party (CCP), and Taiwan lacks freedom of speech”
• Agreement with “The current government is the primary producer of misinformation/disinformation in Taiwan”
• Agreement with “even if cross-strait unification occurs, it will not significantly change ordinary people’s lives”

As shown in the figure above, for categories exhibiting an inverted U-shaped relationship, the average agreement scores for these narratives among respondents initially rise linearly with increasing TikTok usage frequency, reaching a peak among high users, and then significantly decline among extremely high users.

◎ Binary Differences in Usage vs. Non-Usage: The differences are primarily between users and non-users, with no significant differences among low to extremely high users.

• Sense of loneliness
• Agreement with “for ordinary people, PRC is more suitable to live than Taiwan”
• Agreement with “in cross-strait communication, Taiwan’s economic interests are more important than national security”
• Impression of the PRC government
• Agreement with “both sides of the strait are one family, and regardless of the form, unification is inevitable in the foreseeable future”

As shown in the figure above, for categories exhibiting binary differences in usage vs. non-usage, respondents in the “almost never use” group have the lowest average agreement scores, significantly lower than those of other usage levels, indicating the least agreement. However, the average scores among low, moderate, high, and extremely high users show no significant differences, suggesting that the variation lies primarily between using TikTok and not using it, rather than the frequency of use.

◎ Specific Group Prominence: Only a specific usage group (high users) shows significant differences, with no notable differences among other groups.

• Satisfaction with Taiwan’s Democracy: High users (6–14 hours per week) tend to be dissatisfied.
• U.S Skepticism: High users have the highest average total score across three related questions.

The research findings reveal that TikTok usage frequency exhibits complex and non-uniform association patterns with various psychological well-being and political and public opinion variables. A few variables, such as internet addiction, show a significant linear relationship with usage frequency, indicating that higher frequency is associated with greater addiction risk.

However, most variables demonstrate non-linear relationships, with the inverted U-shaped pattern being the most common. Specifically, high users (6–14 hours per week) display the strongest critical and agreeing attitudes toward Taiwan’s domestic political issues (e.g., judicial injustice, constitutional interpretations should reflect public expectations, lack of freedom of speech in Taiwan, the ruling government is the main source of disinformation) and unification narratives (e.g., cross-strait unification would not significantly change ordinary people’s lives). In contrast, extremely high users (over 14 hours per week) show a decline in attitude intensity. This may reflect differing motivations for use (e.g., work-related needs versus entertainment-driven use) among extremely high users or suggest an upper threshold for the impact of TikTok usage frequency on these attitudes.

Additionally, certain political views, such as attitudes toward the PRC, cross-strait exchanges, and unification, primarily differ between “users” and “non-users,” with usage frequency having a relatively limited impact on these variables. A few variables, such as satisfaction with Taiwan’s democracy and U.S. skepticism, show significant differences only within specific user groups, reflecting the unique cognitive or attitudinal characteristics of certain TikTok user segments.

Overall, TikTok usage frequency significantly influences subjective well-being and political attitudes, but the effect patterns are complex and varied, encompassing linear, non-linear, and group-specific associations. These findings highlight the multifaceted interactions between social media usage behavior and psychological and political dimensions, warranting further investigation.

Differences in distribution across questions between active and inactive TikTok users among different political orientations

Across the board, regardless of TikTok activity (This study operationally defines “TikTok active users” as individuals who, in the past year, have used TikTok with a frequency of “several times a week or more, with an average session duration of 30 minutes or more” or “several times a day, with an average session duration of 10 to 30 minutes.” All other respondents are classified as inactive TikTok users.), approximately 80–90% of DPP supporters express positive feelings toward the ruling party, with most saying they “slightly like” or “somewhat like” the party. Similar to the polarization trends observed in the 2024 presidential and legislative election cross-analysis, KMT supporters display an opposite pattern, with 82–86% holding negative views of the ruling party, with nearly half (about 46%) saying they “strongly dislike” the party, consistent across both active and inactive users.

Notably, TPP supporters predominantly express negative sentiment toward the ruling party, but with greater intensity. Between 86% and 97% of TPP supporters express dislike, with the strongest “strongly dislike” option reaching nearly 60% among inactive users and approaching 75% among active TikTok users. While the overall distribution trends are similar, the slight differences between active and inactive TPP supporters warrant further investigation.

When asked, “If PRC intensifies its infiltration and united front efforts toward Taiwan, which political party in Taiwan do you think would benefit the most?” most DPP supporters believe the KMT would benefit, though there is a clear distinction between active and inactive TikTok users, suggesting potential differences in how these groups interpret the issue. Among KMT supporters, 30–35% believe the KMT would benefit, roughly equal to the proportion who believe the DPP would benefit.

In contrast, TPP supporters show an opposite pattern to DPP supporters, with 55–60% believing the DPP would benefit and only 20–25% pointing to the KMT. In political science, “intimidation” and “infiltration and inducement” are distinct strategic concepts with different implications. Had the survey framed the question around “intimidation,” results might have aligned more closely with a partisan confrontation pattern — DPP supporters likely viewing the relatively pro-PRC KMT as the beneficiary, while KMT and pan-Blue supporters might argue it strengthens the DPP’s “national peril” narrative, reflecting typical Blue-Green divides.

However, framing the question around “infiltration and inducement” yields different results: DPP supporters still primarily see the KMT as benefiting, KMT supporters are split nearly evenly, and TPP supporters overwhelmingly believe the DPP benefits. This trend may be influenced by narratives like “Green-Communist-Red-Communist alliance,” and the corresponding perspectives and narratives. These perspectives and the narratives behind them warrant deeper examination.

As observed in the project’s data announcement report, democracy remains widely regarded as the best political system across party affiliations. Around 80% of DPP supporters, regardless of TikTok activity, hold this view. KMT supporters show a lower but still significant approval rate of about 55%. TPP supporters exhibit a notable gap: 66.4% of active TikTok users versus 51% of inactive users consider democracy the best system.

Additionally, KMT and TPP supporters are more likely than DPP supporters to believe that the political system doesn’t make a difference. Among respondents with no specific party preference, over 50% still view democracy as the best system, but 25% of active TikTok users and 16% of inactive users believe the political system doesn’t make a difference. While democracy retains broad trust in Taiwan, compared to the 2022 and 2024 election surveys, there appears to be a slight increase in the relatively apathetic attitude of “for me, any political system makes no difference.”

The relationship between party affiliation and satisfaction with Taiwan’s democracy aligns closely with 2024 survey results. DPP supporters show 80–90% satisfaction, while KMT supporters express 70–75% dissatisfaction.

Notably, compared to 2024, TPP supporters’ dissatisfaction has surpassed that of KMT supporters. Both active and inactive TikTok users within the TPP supporter group show higher proportions in the stronger dissatisfaction categories (“strongly dissatisfied” and “somewhat dissatisfied”) than KMT supporters.

Regarding the statement “Taiwanese society has a very serious issue of judicial injustice,” 69.8% of active TikTok users who support the KMT and 75.3% of inactive TikTok users who support the KMT agree, with active users concentrating in the strongest “strongly agree” category (32.4%) and inactive users in the milder “somewhat agree” (35.3%), indicating intensity differences despite broad agreement.

TPP supporters show strong consensus, with over 87% agreeing regardless of TikTok activity, predominantly in the “strongly agree” category (60% for active users, 43.8% for inactive).

Noteworthy, 69.3% of active TikTok users who support the DPP and 75.3% of inactive users who support the DPP also agree, mostly in the milder “somewhat agree” category. This represents a significant increase compared to the 49.68% of DPP supporters who expressed agreement in the 2024 survey. This shift may be related from recent events, including dissatisfaction with the rejection of constitutional justice appointments, concerns about perceived external interference in corruption cases, or frustration with judicial outcomes in cases involving child pornagraphy, child abuse, and child exploitation. These factors deserve further investigation.

DPP supporters generally disagree with the statement “The current ruling Democratic Progressive Party (DPP) is no different from the Chinese Communist Party (CCP), and Taiwan lacks freedom of speech,” with a 14.6% gap between active (76.1%) and inactive (90.7%) TikTok users in disagreement. KMT supporters show an opposite pattern, with 72.1% of active and 54.5% of inactive users disagreeing, active users showing higher agreement. TPP supporters’ agreement (74.6% active, 75.5% inactive) surpasses KMT supporters, a shift from 2024 when their agreement was lower, indicating a significant change deserving further exploration.

Over 70% of respondents, regardless of TikTok activity, disagree with “ For ordinary people, PRC is more suitable to live than Taiwan.” DPP supporters show over 85% disagreement, with over 55% in “strongly disagree.” KMT supporters have 60–75% disagreement, with milder intensity. TPP supporters show 53–60% disagreement, with higher proportions in “somewhat agree” and “strongly agree” compared to DPP and KMT.

DPP and KMT supporters predominantly disagree with the statement: “In cross-strait communication, Taiwan’s economic interests are more important than national security.” 80.3% of inactive TikTok users who support the DPP disagree with the statement, with 46.5% choosing “strongly disagree.” Active users who support the DPP show slightly lower disagreement rates at 71%.

KMT supporters show broad disagreement (65–70%), mostly selecting “somewhat disagree” (around 40%). TPP supporters are evenly split, with higher “strongly agree” proportions than DPP and KMT.

DPP supporters overwhelmingly disagree (over 80%) with “The current government is the primary producer of misinformation/disinformation in Taiwan,” most selecting “strongly disagree.” KMT active (75.3%) and inactive (66.9%) users agree, with active users concentrating in “strongly agree” (34.2%) and inactive in “somewhat agree” (28.8%). TPP supporters show higher agreement than KMT (87.3% active, 76.6% inactive), with over 60% of active users and 40% of inactive users selecting “strongly agree”, a sharp contrast to DPP supporters.

In the U.S. Skepticism question group, consisting of three statements, cross-analysis by political party affiliation shows a significantly different distribution pattern compared to previous surveys. As noted in the project’s data announcement report, the overall distribution has shifted toward greater agreement, indicating a clear increase in Taiwan’s societal acceptance of narratives that express skepticism toward the United States.

Specifically, DPP supporters, who have historically been less likely to agree with such narratives, have shown a substantial rise in agreement, from about 10–20% in the previous year to over 40%, and in some cases nearly 60%, for the statement “The United States is not trustworthy and only wants to exploit and drain Taiwan.” This applies to both active and inactive TikTok users, though most responses are concentrated in the milder “somewhat agree” category. However, for the statements “The United States uses Taiwan to provoke PRC and attempts to push Taiwan into war” and “If PRC were to use force to attack Taiwan, the United States would definitely not send assistance to Taiwan,” over half of DPP supporters still disagree, suggesting that their fundamental trust in the U.S. role has not been fully undermined.

In contrast, KMT supporters show a consistent agreement rate of 65–80% across the three statements, indicating relatively stable support for narratives that question U.S. intentions or reliability. Notably, active TikTok users among KMT supporters have the highest proportion of “strongly agree” responses, with over 30% selecting this option across all three statements, demonstrating both widespread agreement and stronger conviction compared to their inactive counterparts.

Additionally, it is noteworthy that supporters of the TPP also show levels of agreement with statements skeptical toward U.S.’s intentions or reliability that are closer to, or slightly higher than, those of KMT supporters. Among TPP supporters, active TikTok users exhibit even higher agreement, with over 80% and up to nearly 90% endorsing these statements, surpassing the agreement levels of active TikTok users among KMT supporters. However, their intensity is slightly lower than that of KMT supporters, with responses more concentrated in the “slightly agree” category, followed by “strongly agree.”

The survey was conducted from March 10 to March 20, 2025. Since the January 13, 2024 election, the international landscape has undergone several significant and far-reaching changes, which likely influenced shifts in Taiwan’s public attitudes toward these narratives. These factors warrant further in-depth analysis and tracking.

Regarding the statement “for the sake of cross-strait peace, it’s okay to give up the democratic system,” supporters of all three major political parties generally disagree, with DPP supporters showing the highest level of disagreement. Over 60% of KMT and TPP supporters also lean toward disagreement.

Further analysis reveals that among DPP supporters, 73.1% of active TikTok users disagree, notably lower than the 83.4% of inactive users, with the difference primarily in the strongest “strongly disagree” category (37.9% for active vs. 46.2% for inactive). KMT supporters are more evenly distributed across “strongly disagree,” “slightly disagree,” and “slightly agree” options. TPP supporters, however, show a significantly higher proportion of “strongly agree” responses compared to DPP and KMT supporters, indicating that some TPP supporters have a relatively positive attitude toward prioritizing peace over maintaining the current democratic system.

The attitudes of DPP supporters contrast sharply with those of KMT and TPP supporters regarding the statement “even if cross-strait unification occurs, it will not significantly change ordinary people’s lives” Regardless of TikTok activity, 65–75% of DPP supporters disagree, with approximately 35% in both active and inactive groups concentrating in the strongest “strongly disagree” category, indicating a highly negative stance toward this statement.

In contrast, KMT and TPP supporters show relatively similar distribution trends. About 65% of their active TikTok users tend to agree with the statement, primarily in the milder “slightly agree” category, while only about 48% of inactive users agree, revealing a clear gap. This suggests that active TikTok use may be associated with higher levels of agreement.

Furthermore, TPP supporters have a significantly higher proportion of “strongly agree” responses compared to KMT supporters, indicating that some TPP supporters hold a stronger belief in the notion that life would remain largely unchanged after unification.

Regarding the question, “If PRC invades Taiwan militarily, would you be willing to fight to defend the country?” the survey results show that, regardless of whether they are active TikTok users, over 55% of respondents express willingness to fight.

When political affiliation is factored in, DPP supporters exhibit the highest willingness, with approximately 80–85% indicating readiness to fight. Among them, 37.6% of inactive TikTok users and 49.3% of active users fall into the “very willing” category, demonstrating a firm commitment to defending Taiwan against a military attack.

KMT supporters display a more moderate stance, with about 45–50% expressing willingness, regardless of whether they are active TikTok users or not. Their responses are primarily split between “somewhat unwilling” and “somewhat willing.” Notably, the proportion of inactive users selecting “very unwilling” (24.4%) is significantly higher than that of active users (11.5%).

TPP supporters show slightly lower willingness than KMT supporters, with only 32.5% of active TikTok users and 42.8% of inactive users indicating readiness to fight. Additionally, their “very unwilling” proportion is relatively high, exceeding 30% and approaching 40%, reflecting a stronger reluctance among TPP supporters.

Overall, compared to the 2024 election survey, this analysis finds that TPP supporters now surpass KMT supporters in expressing agreement with critical views on domestic issues and narratives skeptical of the U.S., a shift from the previous year when KMT supporters were the primary high-agreement group for these narratives. While DPP supporters show a slight shift toward greater agreement on some issues compared to 2024, neutral voters (used as a reference group) align more closely with DPP positions on most issues. Whether this reflects a significant ideological shift among TPP supporters within a year or a notable restructuring of their supporter base requires further clarification.

Due to the study design, the sample ratio of active to inactive TikTok users is approximately 1:1.5, potentially over-representing active users and affecting overall distribution trends for certain variables. To verify this, researchers conducted cross-analyses within the active (1,085 respondents) and inactive (1,527 respondents) groups, finding that distribution trends for party affiliation and issue perspectives are largely consistent between the two groups.

Nevertheless, Active TikTok users who support TPP show higher agreement on certain issues compared to inactive users who support TPP. This may suggest that active TikTok use amplifies cognitive effects for specific political groups on certain topics, warranting further in-depth exploration.

Differences in patterns between active and inactive TikTok users

Although the data announcement report has already detailed the differences in response distributions between active and inactive TikTok users across various questions, confirming significant differences in distribution patterns for most variables except for “satisfaction with Taiwan’s current democracy,” “perception of Chinese citizens,” and “relative deprivation,” we aim at this stage to use correlation analysis to compare whether the relationships among variables within the active and inactive user groups exhibit notably distinct patterns.

• Relationship between income, subjective economic satisfaction, and relative deprivation

The analysis results show that within both active and inactive TikTok user groups, subjective economic satisfaction (Q41), perceived fairness between effort and reward (Q42), and relative deprivation (Q43) exhibit moderate to high positive correlations (correlation coefficients ranging from 0.48 to 0.714). However, when examining the relationship between real wage income and these three indicators, only a mild positive correlation is found between “real wage income” and “relative deprivation” among active TikTok users, with other relationships showing no significant correlation.

This suggests that wage income, regardless of whether individuals are active TikTok users, does not strongly influence their satisfaction with economic conditions, belief in the rewards of effort, or sense of relative deprivation. This indicates that wage income has limited impact on these subjective evaluations, and active TikTok use does not significantly alter the correlation patterns among these variables.

• Relationship between preference for the ruling party and satisfaction with Taiwan’s democracy

Based on previous election survey projects, the two most explanatory dimensions — “preference for the ruling party” and “satisfaction with Taiwan’s democracy” — exhibit similar correlations across different groups. Among active TikTok users, the correlation coefficient between these two variables is 0.618, indicating a significant and strong positive correlation. Similarly, among inactive TikTok users, the correlation coefficient is 0.642, also reflecting a significant and strong positive correlation. This suggests that, regardless of TikTok activity, a higher preference for the ruling party (DPP) is associated with greater satisfaction with Taiwan’s current democratic system, demonstrating a highly consistent relationship pattern across both groups. Furthermore, the close correlation coefficients between the two groups imply that active TikTok use does not significantly alter the strength or direction of the positive relationship between these variables, indicating limited influence of TikTok usage on this association.

• Cross-Issue correlation analysis

Overall, compared to the previous analysis of the relationship between preference for the ruling party and satisfaction with Taiwan’s democracy within active and inactive TikTok user groups, this study similarly finds no significant changes in the direction of correlations between issues, narratives, and variables across the two groups. However, notable differences in the strength of correlations are observed.

For instance, compared to inactive users, active TikTok users exhibit stronger correlations in domestic issues (e.g., Taiwan’s judicial injustice, lack of freedom of speech in Taiwan, the ruling government as a source of disinformation) and cross-strait narratives (e.g., PRC is more suitable to live, economic interests in cross-strait exchanges outweigh national security). Additionally, active users show stronger correlations in their agreement with U.S Skepticism (e.g., the U.S. intends to exploit Taiwan) and PRC propaganda (e.g., Taiwanese are part of the Chinese nation). In other words, active users who agree with one statement are more likely to agree with others to a greater extent. Similar patterns are observed between narratives skeptical of U.S. and PRC propaganda, as well as issues related to Taiwan’s resistance to PRC (e.g., peace justifies abandoning democracy, cross-strait unification is inevitable).

These findings suggest that active TikTok use may shape a distinct cross-issue ideological perspective and cognitive framework within the user group.

Latent class analysis on the political attitude of Taiwanese people

Through the analyses in the preceding chapters, which examined political party affiliations, TikTok usage frequency, and correlation analyses within active and inactive TikTok user groups, we have observed differences in issue perspectives between active and inactive TikTok users across various party preferences. However, moving beyond the traditional framework of partisan biases, we aim to further explore the latent group patterns among Taiwan’s general population.

When dealing with a seemingly diverse group of respondents who may share certain commonalities in values or ideology tendencies, Latent Class Analysis (LCA) serves as an effective exploratory statistical method to identify these hidden patterns. LCA estimates potential latent classes based on respondents’ response patterns across a set of observed variables (e.g., questionnaire items). These classes can be understood as implicit subgroups reflecting respondents’ psychological structures or political attitudes. Compared to traditional clustering techniques (e.g., K-means clustering), LCA is built on a probabilistic model, making it particularly suitable for handling categorical data (e.g., Likert scales) and capable of estimating the probability of each respondent belonging to a specific class, thus providing more flexible and inferentially robust classification results.

In this study, we selected seven questions related to political attitudes to construct a four-class latent class model, aiming to depict the potential differentiation in respondents’ political stances and cognitive orientations.

The seven questions included in this study cover trust in Taiwan’s current system, attitudes toward PRC society as a whole, perceptions of cross-strait unification scenarios, and stances on the value trade-off between freedom and national security. These constitute an ideal set of indicators for observing the contours of political attitudes in contemporary Taiwanese society.

Based on the results of the Latent Class Analysis (LCA), the four-class model achieves the best balance between statistical fit and substantive interpretability. The response distributions across the seven questions for each class exhibit clear and significant differences, indicating that this classification effectively distinguishes four groups with distinct political attitude tendencies. Please note that the percentages of the four political attitude groups reflect the internal distribution within the survey data and represent trends within the dataset, not the precise structure of Taiwanese society as a whole.

Based on the response patterns across the selected questions, we have named and described the four latent classes as follows:

• Status Quo Moderates (approximately 36%): This group generally disagrees or holds neutral stances on questions such as whether the government is the main source of disinformation, judicial injustice and whether TikTok should be banned, indicating a basic level of trust in the democratic system. They also firmly oppose Chinese unification and governance models, demonstrating a relatively stable sense of local identity and support for democracy.
• Populist Skeptics (approximately 18%): This group strongly believes the government is a source of disinformation and cyber armies, perceives judicial injustice, and supports the idea that Constitutional Court’s interpretations must align with the expectations of the broader Taiwanese public, reflecting pronounced skepticism toward the system and populist tendencies. On PRC-related issues, opinions are divided, with some showing pro-PRC leanings and others opposing, indicating internal inconsistency but an overall high distrust in Taiwan’s current system.
• Democratic Nativists (approximately 24%): This group exhibits highly consistent attitudes across the seven questions, strongly trusting the democratic system, rejecting claims that disinformation mainly from the government, judicial injustice, and PRC superiority narratives, and firmly opposing unification. They also support banning TikTok, reflecting strong support for Taiwan’s current democratic system and local values, making them the most distinct democratic nativist group in this study.
• Pro-PRC Stabilizers (approximately 22%): This group holds attitudes similar to Populist Skeptics on domestic issues, showing low trust in the government and judicial system, but with a more moderate tone. They have a relatively positive view of the PRC, believing that cross-strait unification would not significantly affect their lives, and do not support banning TikTok. Overall, this group exhibits a higher tolerance for the PRC governance.

Through the results of Latent Class Analysis (LCA), we were able to delineate four distinct political attitude types from a seemingly divergent pool of respondents. Detailed differences across variables for these four attitude types can be found in the linked tables.

To further understand the differences in sociodemographic and psychological characteristics among these four political attitude types, this study compiled descriptive statistics for variables such as gender, age, education level, party affiliation, TikTok usage frequency, and psychological indicators. The analysis reveals clear differences in group structure, political attitudes, and psychological states.

In terms of TikTok usage, Pro-PRC Stabilizers and Populist Skeptics have higher proportions of active TikTok users (52.7% and 51.9%, respectively), compared to lower proportions among Status Quo Moderates and Democratic Nativists (37.1% and 32.1%, respectively). This suggests that TikTok, as a platform with Chinese affiliations, is more favored by the former two groups, while Democratic Nativists exhibit a more reserved attitude toward its use. Whether TikTok usage strengthens pro-PRC, populist, or anti-establishment tendencies will be further explored in the next section using counterfactual analysis.

In terms of gender distribution, Democratic Nativists have the highest proportion of females (56.9%), while Populist Skeptics have the lowest (34.7%). Although the average age difference across the four groups is minimal, Pro-PRC Stabilizers have the lowest average age (37.3 years), which contradicts the common perception that pro-PRC individuals are predominantly older, suggesting that traditional views linking pro-PRC attitudes to specific party affiliations may overlook heterogeneity and potential blind spots.

Among Status Quo Moderates, 64% have no specific party preference, indicating a notable sense of alienation from the party system. Regarding “least liked party,” Populist Skeptics and Pro-PRC Stabilizers show the strongest aversion to the Democratic Progressive Party (DPP), perhaps lending credence to the recent saying that “Taiwan’s largest party is the anti-DPP party.” Status Quo Moderates have a more balanced dislike for both the DPP and Kuomintang (KMT), with nearly half expressing no particular aversion, further highlighting their detachment from traditional Blue-Green divides. In contrast, Democratic Nativists primarily dislike the KMT, followed by the Taiwan People’s Party (TPP), indicating not only rejection of the KMT’s authoritarian history and pro-PRC leanings but also reservations toward the TPP, which is relatively closer to the KMT.

In terms of psychological indicators, the four groups show minimal differences in subjective well-being indicators such as daily happiness and life satisfaction. However, Populist Skeptics and Pro-PRC Stabilizers report higher levels of loneliness and internet addiction. Whether these psychological tendencies are causally linked to their political attitudes requires further clarification.

Beyond the seven main questions used to construct the latent classes, clear clustering patterns are also observed in other questions related to democratic systems and cross-strait relations. Notably, Status Quo Moderates have the weakest belief in democracy, with nearly 30% believing that authoritarian regimes may be better in certain contexts. Conversely, over 80% of Democratic Nativists consider democracy the ideal system under all circumstances. Populist Skeptics have the lowest satisfaction with Taiwan’s current democratic system and the highest levels of U.S. skepticism, while Pro-PRC Stabilizers most strongly agree that “unification would not bring significant changes” and are the least willing to fight for Taiwan.

In summary, this chapter uses descriptive statistics to outline the demographic and psychological characteristics of the four latent political attitude types. The results show that Pro-PRC Stabilizers and Populist Skeptics share some similarities in TikTok usage frequency, aversion to the DPP, and psychological states, particularly in higher levels of loneliness and internet addiction tendencies. Status Quo Moderates exhibit low party identification and uncertainty in democratic beliefs, while Democratic Nativists demonstrate the most stable support for democracy and relatively positive psychological states. Notably, the younger age profile of Pro-PRC Stabilizers challenges the traditional association of pro-PRC attitudes with older generations. The next chapter will use counterfactual analysis to further examine whether active TikTok use significantly influences individuals’ political attitude classifications.

Counterfactual analysis: assessing the impact of active TikTok use on latent political attitude classes

When exploring whether social media use influences political attitudes, relying solely on statistical models to observe their associations is often insufficient. This is because both social media usage behavior and political attitudes may be simultaneously influenced by other underlying variables, known as confounders. Without properly addressing these confounders, it is challenging to determine whether observed differences are due to social media use itself or stem from pre-existing background differences among users.

To more accurately assess the impact of active TikTok use on political attitudes, this study employs the Inverse Probability of Treatment Weighting (IPTW) method within a counterfactual analysis framework. IPTW simulates a scenario where “everyone has an equal chance of being an active TikTok user,” allowing us to estimate the potential causal effect of TikTok usage on the latent political attitude classes.

IPTW is a propensity score-based weighting method designed to achieve balance in covariates between the “treatment” (active TikTok users) and “control” (inactive TikTok users) groups, mimicking the conditions of an experimental design. Specifically, we first use a logistic regression model to predict the probability of each respondent being an active TikTok user (i.e., the propensity score). Then, we apply weights based on these scores to make the distributions of active and inactive groups more comparable in terms of background characteristics.

Through this weighting process, we can approximate the logic of counterfactual inference: when two groups are matched on background factors such as gender, age, education, and party preference, and the only difference is whether they actively use TikTok, what differences in political attitudes emerge?

• Propensity Score Model: Predicting the Probability of Active TikTok Use

This study employs three sets of variables to construct a model predicting the probability of each respondent being an active TikTok user:

1. Demographic variables: Age, education level, gender, region of residence, and income.
2. Psychological states: Internet addiction score and loneliness.
3. Political and public issue attitudes: Most supported political party, most disliked political party, favorability toward the ruling party, confidence in democracy, satisfaction with Taiwan’s current democracy, agreement with economic interests outweigh national security, U.S. skepticism score, agreement with sacrificing democracy, belief that cross-strait unification is inevitable, and trusted sources of information.

These background variables may simultaneously influence both whether respondents actively use TikTok and their latent political attitude class, necessitating statistical balancing. The results of the propensity score model indicate that active TikTok use is highly correlated with these variables, confirming significant differences between active and inactive user groups across multiple variables in the absence of treatment.

Subsequently, we applied IPTW to weight the sample, simulating a scenario where the two groups are comparable in background characteristics. Post-weighting, we assessed the balance of all covariates, finding that the Standardized Mean Difference (SMD) for all variables was less than 0.1, meeting common balance standards. This indicates that the weighting effectively reduced systematic differences between the two groups, approximating the ideal conditions of a randomized experiment.

• Impact of Active TikTok Use on Latent Political Attitude Classes

After confirming that the weighted sample approximates experimental design conditions, we constructed a multinomial logistic regression model with the latent political attitude classes as the dependent variable to analyze the impact of TikTok usage behavior on class membership. The results show that, even after controlling for potential confounders and achieving balanced sample distributions, active TikTok use still significantly influences respondents’ political attitude types, supporting a preliminary inference of a potential causal relationship between active TikTok use and political attitudes. The estimated predicted probabilities based on the model are as follows.

After balancing the influence of confounding variables (the three groups of demographic variables, psychological states, and political and public issue attitudes mentioned above), the distribution of political attitudes among active TikTok users compared to inactive users shows significant differences. Active users have a 2.53% higher probability of belonging to the Pro-PRC Stabilizers class, marking the largest increase, and a slightly elevated probability of being in the Populist Skeptics class. Conversely, active users are less likely to belong to the Democratic Nativists class, with a decrease of 2.59%, the largest reduction, and also show a declining trend in the Status Quo Moderates class.

These results indicate that, even after controlling for other background conditions, active TikTok use itself exerts an influence on political attitude tendencies, particularly in weakening trust and satisfaction with Taiwan’s current system and democracy. This aligns with concerns regarding the political content and algorithmic mechanisms prevalent on TikTok, providing important empirical evidence for understanding how digital platforms shape and reshape citizens’ political identities and attitudes.

For detailed results on model coefficients and odds ratio interpretations, please refer to the linked appendices and data.

Conclusion

TikTok’s influence on political attitudes shows non-linear, phased fluctuations and polarizing trends, indicating that its immersive usage and narrative-driven content reshape users’ perceptions of institutions, national identity, and future imaginaries. Active users, especially on cross-strait and domestic issues, are significantly more likely than inactive users to agree with statements such as “the PRC is more livable” and “economic interests take precedence over national security.” This influence transcends party lines, notably affecting TPP supporters and even some DPP supporters. The significant impact of TikTok on Taiwanese people’s political attitudes, democratic beliefs, and psychological states underscores that it functions as more than an entertainment platform. It is a critical arena for reshaping political cognition and institutional trust.

Additionally, by distinguishing between active and inactive TikTok user groups and conducting internal cross-issue correlation analyses within each, we found that while the direction of correlations across variables remains consistent between the two groups, there are clear differences in the strength of these correlations. For example, among active TikTok users, agreements with narratives skeptical of U.S. intentions and reliability shows a strong positive correlation with agreement with Chinese propaganda, whereas this correlation is only weak among inactive users. A similar pattern is observed between U.S. skepticism and cross-strait-related discourses (e.g., “sacrificing democracy for peace” and “cross-strait unification is inevitable”). Active TikTok use indeed appears to construct a distinct cross-issue ideological perspective and cognitive framework.

The study also finds that active TikTok use is positively correlated with internet addiction and leads to significant shifts in attitudes toward issues such as judicial injustice, unification imaginaries, and skepticism toward the U.S. This is particularly pronounced among the Pro-PRC Stabilizers and Populist Skeptics groups, who exhibit higher agreement and more extreme expressions. Through Latent Class Analysis (LCA), the study identified four political attitude groups, and subsequent counterfactual analysis revealed that active TikTok use itself increases the likelihood of users aligning with the Pro-PRC Stabilizers and Populist Skeptics classes while significantly reducing the probability of belonging to the Democratic Nativists and Status Quo Moderates classes. This confirms that active TikTok use inherently carries the risk of reshaping political attitudes and eroding democratic beliefs.

To address TikTok’s subtle influence on democratic beliefs, cognitive biases, and political socialization, the government and civil society must adopt a society-wide strategic communication approach, collaborating to strengthen democratic defense mechanisms. Otherwise, digital platforms may unwittingly reshape public cognition and the political spectrum, posing an irreversible challenge to Taiwan’s democratic resilience.

Limitations

Sampling limitations

This survey was conducted using an online questionnaire. To ensure data quality and validity, in addition to the preliminary error-checking mechanisms built into the EZChoice system, the research team designed a logic verification table (covering logical consistency and relational checks) based on the survey content and performed multiple logical validations on the data. Furthermore, the survey data were weighted and adjusted based on the latest demographic statistics from the Ministry of the Interior, using variables such as respondents’ gender, age group, education level, and geographic region to enhance the sample’s representativeness of the population. However, as this survey was conducted online, certain sampling limitations may still exist. For instance, respondents’ internet usage habits may influence their willingness to participate, or certain demographic groups may be underrepresented in the online sample.

Sensitivity of public politics, cross-Strait Issues, and international relations perspectives

Some questions in this questionnaire address public politics, cross-strait issues, and international relations, topics that are highly polarized in Taiwanese society, where differing political ideologies may lead respondents to hold sharply divergent views. In this context, some respondents may feel distrust toward the questionnaire content or the survey itself. Although this online survey was commissioned to a politically neutral polling company to mitigate this effect, such sensitivities may still pose potential challenges to the survey results.

Prejudices and social desirability bias among TikTok/Douyin Users

At the outset of the questionnaire, respondents were screened based on their weekly usage frequency and single-session duration of TikTok/Douyin to distinguish between active and inactive users. This design may make respondents aware of the survey’s targeted focus, potentially prompting them to associate it with negative societal perceptions of TikTok. In this scenario, some respondents may adjust their responses to avoid being labeled as “pro-PRC” or “irrational,” leading to social desirability bias. Such bias may affect the authenticity of the data, particularly for questions involving political attitudes or values.

Quantitative analysis

Although this study employs diverse quantitative methods such as cross-tabulation, correlation analysis, latent class analysis (LCA), and counterfactual analysis to comprehensively depict the potential relationships between TikTok usage behavior and related variables, quantitative analysis itself has certain limitations. Even though the sample size is sufficient and weighted for demographic structure, finer-grained subgroup analyses (e.g., LCA) may still be affected by insufficient sample sizes, potentially impacting the stability and explanatory power of the estimates.

Furthermore, while counterfactual analysis can simulate potential outcomes under different scenarios, its results heavily depend on model assumptions and the selection of variables, and unobserved confounding factors may still be present during the inference process. Therefore, the statistical inferences and observed associations presented in this study should be regarded as preliminary explorations of trends and potential mechanisms. Future research will require additional experimental validation and longitudinal tracking to enhance the robustness of the conclusions.

Acknowledge

This project expresses special gratitude to Dr. Zonghong Lin, a full-time research fellow at the Institute of Sociology, Academia Sinica, for providing numerous suggestions and insights during the research planning stage regarding questionnaire design, research logic, and subsequent analysis. We also thank Yun-Ru Chen for assisting with the complex and in-depth statistical analysis of this project, and, together with Jun-Ying Li, research assistant at the Institute of Sociology at Academia Sinica, for providing crucial support in questionnaire design and data statistical analysis for the online survey.

Additionally, we extend our gratitude to the Public Opinion Program team at the Academia Sinica Research Center for Humanities and Social Sciences and the PRC Impact Survey team for providing valuable data from the Well-Being Survey and telephone surveys, respectively, which served as references for this project’s questionnaire. The research design of this project also partly drew upon the Taiwan Communication Survey Database and the Taiwan Social Change Survey. We express our sincere respect and gratitude to the many scholars and teams who have contributed to these fields in the past.

Furthermore, we thank DSIGroup Co., Ltd., for conducting the online survey on behalf of Doublethink Lab in accordance with the planned specifications. We also appreciate the friends who provided various forms of assistance and suggestions for this project; without everyone’s support, this survey would not have been possible. All matters arising from this project and its report are the sole responsibility of Doublethink Lab.

Contact us

Doublethink Lab: [email protected]

Social Engagement Researcher- Eric Hsu: [email protected]

Attachments

◎ 2025 TikTok 抖音使用者調查＿全國網路調查 原始數據/Raw data
◎ 2025 TikTok 抖音使用者調查＿全國網路調查 譯碼簿/ Code book