A collective textbook-guide by Colaboratoria
cognitive distortions and systemic errors in science
Information literacy in the study of digital international relations:
- Yury KolotaevAuthor
The emergence of digital methods of analysis is accompanied by a large number of costs and distortions that can undermine effective scientific work. Identifying such problems is closely tied to the concepts of digital and information literacy, which highlight recurring distortions in research practice: systemic errors in science, errors in applying digital methods and interpreting digital data, and cognitive limitations.
/01
Information and Digital Literacy
The prospects and opportunities that come with the development of digital methods of analysis open broad horizons for researchers in international relations. They make it possible to identify previously hidden communicative patterns in the study of global processes while relying on clear empirical data.
At the same time, the emergence of new methodology is accompanied by a large number of costs and distortions that can harm effective scientific work. Attempts to shed light on these problems come down to the concepts of digital and information literacy, which highlight recurring errors in scientific work—arising both from emerging analytical techniques and from enduring historical challenges inherent in the scientific discipline.
Information and digital literacy constitute a comprehensive set of competencies essential for effectively utilizing data, sources, and digital technologies throughout research activities.
This set includes an understanding of the capabilities and limitations of current methods of scientific analysis, as well as the ability to critically assess the results of their application.
Information and digital literacy comprise several key components; a thorough understanding of these elements is essential for minimizing common errors in research.
Systemic errors in science
Errors in applying digital methods
Errors in interpreting digital data
Cognitive limitations
/02
Systemic Scientific Errors
The problem of scientific distortions, whether broadly or specifically within (digital) international relations, includes several key aspects such as spurious correlations, scale errors, bias, and non-representativeness. These distortions can significantly affect research quality and the conclusions drawn from it.
Key scientific biases
Research on digital international relations based on big data and social media analysis can lead to spurious correlations stemming from the nature of statistical analysis itself. For example, a researcher studying digital diplomacy observes growing support for a state’s positions on social media and interprets this as the success of its foreign policy.
However, such support may be caused by temporary trends or data manipulation rather than real changes in public opinion. The use of hashtags to promote certain ideas often creates the illusion of broad support, even when it is merely the result of coordinated activity by a small number of users.
Hence the general rule: correlation does not automatically imply causation. Verification and checking are always required!
However, such support may be caused by temporary trends or data manipulation rather than real changes in public opinion. The use of hashtags to promote certain ideas often creates the illusion of broad support, even when it is merely the result of coordinated activity by a small number of users.
Hence the general rule: correlation does not automatically imply causation. Verification and checking are always required!
Errors in choosing scales (nominal, ordinal, interval) are also a problem in digital international relations. Building any metrics within a study is tied to reference points that appear most useful.
However, such reference points do not always reflect an objective picture. This also concerns the use of existing indices or ratings that fail to capture all necessary aspects and sometimes present qualitative parameters as quantitative ones.
For example, a "soft power" index, if inconsistently designed, may include subjective indicators such as "government effectiveness," which in itself depends on other indices and (in the absence of explanatory tables or clarifications) is not directly quantifiable or may be distorted. This leads to unreliable conclusions about a country’s real influence in international affairs.
However, such reference points do not always reflect an objective picture. This also concerns the use of existing indices or ratings that fail to capture all necessary aspects and sometimes present qualitative parameters as quantitative ones.
For example, a "soft power" index, if inconsistently designed, may include subjective indicators such as "government effectiveness," which in itself depends on other indices and (in the absence of explanatory tables or clarifications) is not directly quantifiable or may be distorted. This leads to unreliable conclusions about a country’s real influence in international affairs.
Bias in digital diplomacy can manifest through selective data sampling for analysis. In such cases, the researcher chooses only data that confirm their political stance while ignoring contradictory facts (this phenomenon also directly relates to the cognitive biases discussed below).
For instance, when a researcher examines solely successful instances of international cooperation while disregarding failures, the assessment of cooperative potential becomes unduly optimistic.
For instance, when a researcher examines solely successful instances of international cooperation while disregarding failures, the assessment of cooperative potential becomes unduly optimistic.
Non-representative sampling is a significant problem for all sciences, including the social sciences and humanities. In the digital space, many studies are based on data from a limited number of platforms or user groups, which may not reflect the real picture of public opinion.
For example, activity on a particular social network likely differs from the views of people who do not use that platform. As a result, there is a risk that conclusions will be based on incomplete information.
For example, activity on a particular social network likely differs from the views of people who do not use that platform. As a result, there is a risk that conclusions will be based on incomplete information.
Thus, scientific errors become more complex as digital tools for analyzing international relations develop. These challenges require more careful approaches to research methodology and data analysis to ensure the reliability of conclusions and recommendations.
/03
Errors in Applying Digital Methods and Interpreting Digital Data
Levels of information analysis
To analyze digital processes, it is necessary to consider three key levels of information processing and representation in the digital environment, each of which imposes its own limitations.
Three key levels:
This level involves analyzing digital platforms and technologies essential for research. Limitations include technical parameters (e.g., algorithms, privacy settings) and the availability of digital resources across countries. At the media level, it is necessary to select platforms and technologies that ensure data reliability and representativeness. For example, social media data should be analyzed with regard to algorithmic filtering and access restrictions.
This level concerns the content and narratives created and disseminated through digital technologies. Limitations here relate to cultural differences, language barriers, and information interpretation. At the discourse level, it is important to consider how narratives and terms are understood in different cultural contexts. This helps avoid linguistic and substantive distortions.
This level analyzes how researchers perceive, process, and interpret information. Limitations include biases arising during data analysis and the impact of digital technologies on cognitive abilities—for example, gullibility or excessive dependence on AI-based technologies. For this reason, at the level of cognitive mechanisms it is necessary to implement methods of data verification to minimize bias and ensure reproducibility of results.
A comprehensive understanding of algorithms and technologies allows researchers to effectively assess their suitability at each stage of analysis:
-
At the media level: assessing how social media algorithms determine reach and data accessibility
- At the discourse level: examining how algorithms can amplify certain narratives, creating informational distortions
- At the level of individual cognitive mechanisms: considering how personal experience may lead to overestimating or underestimating observed phenomena
Digital literacy: international context
- Digital tools and platforms may operate differently across jurisdictions.
For example, European data protection standards (GDPR) impose strict restrictions on the use of personal data, affecting tool and methodology choices as well as sampling.
- International studies require special attention to transparency and reproducibility.
Algorithms developed in one country may be unsuitable for data from another due to "cultural," "structural" (i.e., compatibility and accessibility-related), or linguistic differences. Recognizing and reflecting these limitations in the research text helps ensure the reliability of results.
- Orientation toward international data governance standards.
International data governance standards such as the FAIR principles (Findable, Accessible, Interoperable, Reusable) are especially relevant in joint projects involving researchers from different countries. It is also important to ensure transparency when working with cross-border data, avoid unauthorized use of information, and follow ethical standards—specifically, ensure user consent and anonymization (see Chapter 17).
Framing and priming effects
When working with digital platforms, framing and priming effects play a major role. They are directly connected to the formation of opinions and perceptions of international reality, especially in digital international relations. These concepts reflect how information is presented in the media space and what reaction it can trigger. Simultaneously, framing and priming are influenced by the perspective and structure of the information intermediary.
Framing — a way of structuring information through the choice of language, context, metrics, or visualization.
Framing seeks to highlight certain aspects of an event or problem, both within a single text and across an entire platform. For example, using terms with different emotional connotations ("challenge" instead of "crisis," "optimization" instead of "reduction") can radically change perceptions of a problem even if quantitative data remain unchanged.
Priming — a way of presenting information whereby prior context or stimuli influence the perception and interpretation of subsequent information.
For example, if a platform is pre-configured to search for or prioritize "crisis" data, it begins to assign greater importance to negative aspects while ignoring positive ones. Similarly, an information search or data structuring algorithm "tuned" to a particular type of retrieval may avoid information that falls outside the preset context.
On the one hand, this simplifies the search for relevant information, but on the other hand it creates a risk of reducing data representativeness. In data analysis, priming typically occurs by establishing context, such as through the formulation of initial hypotheses or the selection of training data for models.
Framing and priming on social media involve how information is presented, such as through headlines, hashtags, engagement metrics, or keyword choices. These elements determine how users interpret what is happening.
Example:
In research on public opinion about climate change, using the term "global warming" instead of "climate crisis" may influence emotional response and perceived urgency differently.
In research on public opinion about climate change, using the term "global warming" instead of "climate crisis" may influence emotional response and perceived urgency differently.
Metrics such as mention counts or engagement levels can also create a false sense of importance for a given topic if the overall context and audience size are not considered. In other words, framing effects are intensified when researchers analyze social media data without taking platform features into account.
Social media algorithms can amplify priming and framing if they are trained on data that already contain biases. If a machine learning system analyzes news that predominantly uses sensational or negatively framed headlines, it may misinterpret the overall tone of discussion, ignoring more neutral aspects.
Methods for detecting and minimizing framing and priming effects:
- using multi-channel data to validate findings;
- researchers' awareness of how preliminary hypotheses and queries affect results;
- reviewing language and examining terminology to detect emotional undertones and adjust towards more neutral expressions;
- testing different scenarios and using multiple alternative interpretations to reduce framing influence;
- integrating data from different social networks to identify general trends and exclude the "algorithmic bias" of a single platform;
- comparing social media data with surveys, news, or statistical reports to verify conclusions.
Critical thinking
When analysing social media data, it is essential to consider various sources of potential error or bias that may affect sampling. These include both intentional and unintentional misinformation, fraudulent content, and other forms of informational noise. Unethical handling of data may lead to distorted statistics or manipulated graphs.
To address these problems, critical thinking serves as the main tool and lies at the core of information literacy, representing an essential competency for handling data amid widespread fake news and digital disinformation. Critical thinking involves activities such as verifying the origins of information—like evaluating the author’s credibility, examining referenced data, and considering how detailed the content is—as well as analysing possible underlying motives or interests that authors or sources may have.
A key practice for critical assessment is comparing information across different sources to detect discrepancies. Recognizing manipulative techniques and falsified data is a more advanced skill that also relies on awareness of one’s own cognitive biases that can be used for manipulative purposes.
/04
Cognitive biases in research
Cognitive biases — systematic deviations in perception, memory, interpretation, and analysis of information that arise due to the limitations of human cognition.
These biases often result from automated thought processes intended to simplify complex information. To some extent, similar patterns can be observed in algorithms/AI, including those related to "hallucination" or data distortion.
In scientific research, cognitive biases can reduce analysis quality. They are especially common when working with large datasets or complex models. Cognitive biases are directly linked to the general scientific biases mentioned above and to critical thinking.
Examples of cognitive biases and their impact on scientific research
A tendency to favour information that confirms existing beliefs or hypotheses while ignoring data that contradict them.
Ignoring overall statistical information (the base rate) in favor of isolated or vivid examples. For instance, when evaluating the effectiveness of an information campaign, a researcher may focus on a few successful cases while ignoring failure statistics or control group data. This leads to an inflated assessment of effectiveness.
A tendency to rely on an initially encountered value (an "anchor") when subsequently evaluating information. For example, when assessing network support for a digital actor, a researcher may accidentally treat initial raw values (reposts and reactions to a single post) as a baseline and ignore refined statistics across the full sample. This is especially common when early results appear plausible or match expectations.
Cognitive biases can also be "transferred" to AI algorithms. This happens due to incorrect model configuration, bias in training data, or the choice of evaluation criteria. For instance, if one viewpoint (ethical or political) predominates in training data, AI, in the absence of corrective mechanisms, will reproduce that bias.
The base rate fallacy can also be reproduced. Forecasting algorithms that ignore base statistics overestimate the importance of rare, salient data. In other words, insufficient consideration of overall event probabilities leads to distorted outputs and forecasts.
Cognitive biases embedded in AI can significantly reduce forecast accuracy, model adequacy, and the quality of conclusions. Distortions in training can make models less robust to real-world data. Therefore, cognitive biases are an important factor affecting research quality both from the researcher’s perspective and from the perspective of using digital technologies.
/05
Practical Approaches to Minimizing Systemic Errors, Biases, and Media Effects
Teaching information and digital literacy
- Developing basic data skills: understanding statistics fundamentals, how algorithms work, and information processing methods.
- Reflecting on critical thinking and learning to recognize bias and misinformation in data.
- Promoting an ethical approach to research by emphasizing data confidentiality, responsible algorithm use, and scientific integrity.
Using "multimodal" approaches
A multimodal approach involves collecting and comparing data from different sources, such as social networks, official reports, and news media. This reduces the impact of algorithmic bias typical of social networks and allows validation of conclusions by comparing them with independent sources.
This helps reduce the risk of errors caused by the limitations of a single source. In particular, cross-validation is required—i.e., checking data whereby findings from one source are confirmed using another (for example, analyzing overlaps in trends identified by different data-processing algorithms).
Checklists for assessing the reliability of data sources
- Who is responsible for creating the content, and what is the reputation of the source?
- May I confirm whether these data are original?
- Accurate classification of collected data: Is the objective to inform, manipulate, or elicit a reaction?
- Are there any observable contradictions or evident biases within the identified patterns and correlations?
- Do the claims and data align appropriately with the broader contextual landscape?
Information literacy has emerged as an essential component for effective scientific research in the context of digital international relations. It contributes to improving research quality, minimizing risks, and strengthening trust in science. Using a multi-level analytical approach, accounting for systemic scientific biases, and maintaining a critical attitude toward both cognitive processes and algorithms helps increase the accuracy of research in the digital age.
Practicum
- Describe the three levels of data processing (media, discourse, individual perception) and the limitations associated with each of them. How can social media algorithms distort research results at each of these levels?
- What is the difference between framing and priming? Provide examples from international relations where these effects can lead to erroneous conclusions.
- How can cognitive biases be transferred to AI algorithms? What risks does this create for research in digital diplomacy?
- How does applying the FAIR principles increase the reliability of research? Suggest actions to assess the credibility of a data source.
- Give an example of data manipulation that can be detected using critical thinking.
- Imagine that a researcher analyzes the effectiveness of a climate-related information campaign using data from only one platform. What systemic and cognitive errors might arise?
- Griftsova, I. N., Sorina, G. V. (2022). Critical thinking against pseudoscience. Chelovek, 33(1), 7.
- Kahneman, D., Sibony, O., Sunstein, C. (2021). Noise: The Flaw in Human Judgment. AST.
- Khodikova, N. A. (2021). Critical thinking and information literacy. Culture and Security, (3), 11−15.
- García-Carmona, A. (2023). Scientific thinking and critical thinking in science education: Two distinct but symbiotically related intellectual processes. Science & Education, 1−19. link.springer.com/article/10.1007/s11191−023−460−5
- Gruwell C., Ewing R. (2022). Critical Thinking in Academic Research — Second Edition. Minnesota State Colleges and Universities. open.umn.edu/opentextbooks/textbooks/1190
- Scientific Research and Communication — Scientific Information Literacy. UCONN. guides.lib.uconn.edu/c.php?g=1 067 492&p=9 147 563
