Analytical Capacity: making sense of complexity.

A regional commercial team opens its monthly review with an apparently straightforward problem: margin has fallen sharply in two markets. Finance points to discounting, sales blames a change in product mix, operations highlights freight costs, and the customer data shows an unusual rise in returns. Every explanation is plausible and the underlying data is genuine, yet the team still has to determine which information deserves attention, whether the figures are comparable, what question should be investigated first and how much confidence the available evidence can support.

The quality of the eventual conclusion depends on work that starts before anyone begins building an argument. Some evidence has to be sought and some ignored, fragmented information needs a usable structure, and the problem itself may need to be reformulated once the first assumptions stop fitting the evidence. Patterns have to be interpreted without converting coincidence into causation, and competing explanations need to survive a reasoning process strong enough to expose their weaknesses. Even after that, the final conclusion can still fail if its confidence or scope exceeds what the evidence justifies.

We define Analytical Capacity as the capacity to critically process information and draw logical conclusions. Within the Unified Skills Map, we organise this capacity through six interrelated dimensions: Selecting information, Organising information, Framing the analysis, Interpreting information, Structuring reasoning, and Reaching sound conclusions. Contemporary research provides substantial support for the functions represented by all six. The academic literature has not established them as six independent psychological faculties, and evidence from reasoning, judgment, metacognition, expertise, information science and problem representation points towards a process that is recursive and strongly dependent on knowledge and context (Andreucci-Annunziata et al., 2023; Ananth, Baer and Deichmann, 2025).

That distinction matters because analytical capability is often compressed into a single familiar label such as critical thinking, problem solving or intelligence. Those constructs overlap parts of analytical work while leaving other parts outside their scope. Critical-thinking research itself contains substantial variation in definitions and measurement, and cognitive reflection shares meaningful variance with intelligence and numeracy without being reducible to either (Otero, Salgado and Moscoso, 2022; Sobkow, Olszewska and Sirota, 2023). Analysis therefore deserves to be examined through the work it actually requires, with these neighbouring constructs used where they illuminate a specific part of that work.

Analysis starts with the problem you think you have

Imagine that the commercial team immediately accepts the question, “Why did discounting reduce margin?” The wording already contains a causal assumption. Evidence about product mix, returns or freight may now receive less attention because the investigation has been framed around an explanation that has not yet been established. A technically sophisticated analysis can proceed from that point and still become increasingly precise about the wrong problem.

Framing the analysis concerns the construction of the problem that analytical effort will address. In well-structured tasks, the question, relevant variables and acceptable evidence may already be specified. Real professional problems frequently arrive in a less cooperative form. The analyst has to decide what the question is, which boundaries matter, whose definition of the problem is being inherited, what time horizon is relevant, which assumptions are provisional and what would count as an adequate explanation.

Recent organisational research follows problem representations as they develop and change over time. Ananth, Baer and Deichmann (2025) synthesise research across several traditions and show how problem finding, framing and formulation shape the development of complex problem representations over time. Chen, Sharma and Muñoz (2023), tracing the evolution of research problems, likewise show that formulation changes through continuing judgments about scope, significance, specificity and centrality. These studies are not tests of a generic “framing ability,” yet they provide strong evidence for the function we are trying to capture: the representation of the problem governs what subsequent analysis can see.

Framing can fail very early when an inherited description reflects organisational politics or concentrates attention on a visible symptom while the process generating it remains outside the frame. Anchoring can keep attention fixed on the first plausible explanation, and a narrow scope can remove the variable that eventually explains the result. Once those choices have been made, later evidence is filtered through them.

A defensible frame remains revisable as the investigation develops. Pham, Magistretti and Dell'Era (2023), studying ill-defined problem framing, show that analogy, association and abductive operations can help people construct alternative representations. Some of those processes overlap with Creative Capacity because opening a different frame may require generating a possibility that was absent from the original problem definition. Their analytical value depends on whether the revised frame gives the evidence a better structure and improves the quality of the investigation. The boundary between analytical and creative work is functional here, because the same cognitive operation can serve different purposes inside a larger process.

Once a workable problem has been constructed, Selecting information determines which evidence enters that process and what further evidence needs to be found. Selection includes active search, source evaluation, relevance judgment, recognition of missing information and the decision to stop searching. A narrow definition that treats selection as choosing from a fixed pile of documents misses much of the analytical work involved.

Information-credibility research makes the difficulty measurable. Ou and Ho's (2024) meta-analysis of 85 empirical studies found that perceived credibility is strongly related to message quality and source credibility, while message fluency also affects how credible information feels. Fluency is dangerous because an easy-to-process explanation can acquire persuasive force without gaining evidential strength. A confident executive summary, polished consultant slide or fluent AI-generated paragraph can therefore influence selection before the analyst has examined provenance or underlying evidence.

Search eventually has to stop, and Ilani, Nowkarizi and Arastoopoor's (2024) systematic review identifies perceived information sufficiency, task characteristics, time, search environment and individual factors among the conditions that influence that decision. Premature stopping can freeze an analysis around the first coherent account, while uncontrolled search can consume time without materially changing the judgment. Good selection requires a defensible relationship between the analytical question, the value of additional evidence and the cost of continuing the search.

The misinformation literature provides unusually observable tests of some of these processes. Steinfeld's (2023) eye-tracking research connects digital literacy and visual attention to source metadata with successful misinformation identification. Experimental work also shows that source-credibility cues and accuracy prompts can improve discernment under some conditions (Pennycook and Rand, 2022; Prike, Butler and Ecker, 2024). These settings offer evidence about source evaluation without defining the whole dimension. In a workplace investigation, the same analytical function may concern which customer interviews are representative, whether a dashboard has changed its denominator, whether a supplier's figure is independently verified or which missing document would materially change the conclusion.

Organising information addresses a related problem because disconnected fragments are difficult to interpret, yet every structure imposed on evidence already makes analytical choices. A category can place superficially similar cases together while concealing differences that later prove important, and aggregation can erase variation that explains the result. Visual representations create the same risk when they foreground one relationship, suppress another or draw an assumed causal connection as if it had already been established.

Research on external representation and expertise shows why organisation matters. Srivastava, Srivastava and Chandrasekharan (2021) found that the strategies students used to construct concept maps were associated with the quality of the resulting representations. Classic expert-novice research also shows that expertise changes how problems are categorised, with experts more likely to organise information around structurally meaningful principles available through their domain knowledge (Chi, Feltovich and Glaser, 1981). Contemporary work on analogical reasoning similarly emphasises relational structure and shows why surface similarity can mislead mapping between problems (Holyoak, Ichien and Lu, 2022).

Organising information is therefore the construction and revision of a representation that makes relevant relationships, contrasts, dependencies and gaps easier to inspect. The strongest representation preserves uncertainty that still matters instead of removing it for visual neatness. A commercial team that collapses all markets into one average may make the data easier to read while destroying the pattern that explains the margin decline. Analytical organisation succeeds when reduced complexity reveals consequential structure without manufacturing a cleaner reality than the evidence supports.

Evidence has to become meaning

A well-framed problem and an organised evidence base still leave the analyst with a difficult task: deciding what the information means. Interpreting information concerns the movement from observed evidence towards warranted understanding. The processes involved change with the kind of problem being analysed. A financial analyst may be comparing trends and ratios, a clinician may be integrating diagnostic cues, and an operations team may be deciding whether a delay pattern reflects capacity constraints, supplier instability or a measurement change.

The literature around interpretation is correspondingly broad, with statistical reasoning, causal inference, diagnostic reasoning, scientific reasoning and judgment under uncertainty examining different kinds of evidence-to-meaning relationships. Their diversity is useful because it prevents interpretation from becoming a vague talent for “spotting patterns.” Human beings are highly capable of perceiving relationships that are irrelevant, coincidental or produced by the way information has been represented.

Representation can change inferential performance even when the underlying evidence remains equivalent. Kunzelmann et al. (2022), studying Bayesian diagnostic reasoning in medical students, found that frequency-based visualisations improved performance and efficiency relative to less supportive representations. The result connects Organising information directly with Interpreting information: a better representation can make the probabilistic structure of evidence easier to use, while a poor representation can create difficulty that looks like a reasoning deficit.

Bias research provides many examples of interpretive failure, although the strongest modern reading is more contextual than the familiar catalogue of universal human irrationalities. Base rates may be neglected in some laboratory tasks, yet Link and Raab (2022) found expert beach-volleyball decision-makers using base-rate information in real sequential decisions. Ng, Lee and Lovibond (2024) show that measurement format itself can exaggerate apparent illusory-causation effects. The tendency to misinterpret evidence is real, but its expression depends on task design, representation, expertise and the ecology in which judgment occurs.

Expertise complicates the picture because relevant knowledge often improves interpretation by giving the analyst schemas, causal models and disciplinary standards that a novice does not possess. It helps distinguish an ordinary fluctuation from a meaningful anomaly and makes some implausible explanations easier to reject. The same accumulated knowledge can also create expectations that narrow attention or encourage premature closure. Expertise changes the interpretive landscape; it does not guarantee that every interpretation will be correct.

A disciplined interpretation therefore preserves competition between plausible meanings long enough for evidence to discriminate among them. Correlation can suggest a relationship without settling causality. An anomaly may represent an error, a rare case or an early sign of a new pattern. A trend can change because the underlying phenomenon changed, because the population changed or because the measurement process changed. Interpretation gains quality through the analyst's ability to keep these possibilities connected to the evidence and to the limitations of what that evidence can establish.

Reasoning has to carry the weight of the claim

Once evidence has been interpreted, the analyst has to connect observations and propositions into an argument strong enough to support the claim being made. Structuring reasoning has the clearest relationship with mature research traditions in deduction, induction, abduction, causal reasoning, analogy, argumentation and hypothesis comparison. These traditions differ in formal structure, but they share a concern with whether conclusions are warranted by the premises and evidence from which they are derived.

Plausibility, validity and evidential support answer different analytical questions. An explanation can fit the observed facts while competing explanations fit them equally well; deductive validity preserves the relationship between premises and conclusion even when a premise is inaccurate. Empirical work adds another layer because evidence can favour one account without eliminating uncertainty, leaving analytical quality dependent on the type and strength of inference appropriate to the question.

Coherence can become deceptive at this point when a manager builds a compelling story in which a competitor's price cut caused a sales decline, customer sentiment confirms the story and subsequent discounting appears to explain the recovery. The narrative hangs together, but its coherence says little about whether other explanations were tested. A seasonal shift, channel mix, distribution disruption or simultaneous marketing campaign may fit the observations equally well. Reasoning becomes stronger when the structure exposes which evidence actually discriminates among those accounts.

Research on confirmation processes has examined this vulnerability for decades. Nickerson's (1998) integrative review shows how selective search and interpretation can protect a focal hypothesis, while Wason's (1960) classic experiment demonstrated how people can test hypotheses by seeking confirming evidence while neglecting observations capable of challenging them. Contemporary intervention work shows that some of these errors can be reduced locally. Rodríguez-Ferreiro, Vadillo and Barberia (2023), for example, found that a targeted intervention reduced illusory causal judgments even when participants were not controlling the sampling of evidence, suggesting that training altered how causal evidence was interpreted and integrated.

Externalising reasoning can also help because hidden inferential jumps become inspectable. Nesbit and Liu's (2025) systematic review identified 124 higher-education studies of argument mapping, including 102 empirical studies. The literature is heterogeneous, but a substantial portion examines whether mapping premise-conclusion structures can improve critical-thinking performance. Argument maps are useful here as one example of a wider principle: reasoning can be developed and assessed more effectively when the relationship between evidence, assumptions, alternatives and conclusions is made visible.

The capacity we call Structuring reasoning therefore concerns the construction of inferential relationships whose strength matches the claims they are supporting. It includes comparing alternatives, testing assumptions and incorporating counterevidence where the problem requires it. Logical coherence contributes to that work without providing a sufficient standard on its own.

A sound conclusion may still contain uncertainty

Analysis eventually has to produce an answer that another person can use. Reaching sound conclusions concerns the relationship between that final claim and everything that supports it. The conclusion has to answer the analytical question, remain proportionate to the evidence and reasoning, and carry an appropriate level of confidence.

Confidence introduces a separate problem because subjective certainty can drift away from accuracy. Drummond Otten and Fischhoff (2023), across three studies of scientific reasoning, found average confidence exceeding accuracy by roughly 22 to 25 percentage points, although higher-performing participants were generally better calibrated. Better reasoning can therefore improve calibration without eliminating overconfidence.

The same calibration problem appears among professionals, as Kelly and Mandel (2024) showed in a study of 70 intelligence analysts before and after calibration training. Training improved calibration for interval estimates in which participants had initially been overconfident, while performance changed differently on binary-choice judgments and improvement did not transfer consistently across the two formats. Calibration is partly task-dependent, which makes a generic “be more confident” or “be less confident” intervention analytically crude.

A sound conclusion can therefore be qualified, conditional or explicitly incomplete. The available evidence may support a probability range, leave two explanations unresolved or justify further investigation before a stronger claim is made. False precision creates the appearance of analytical completion by hiding uncertainty that still belongs in the answer. Underconfidence can be damaging as well when strong evidence is presented so tentatively that decision-makers cannot distinguish it from speculation.

Reaching a sound conclusion also clarifies the boundary between analysis and decision making because an analysis may establish that a proposed investment carries a materially higher probability of delay than the alternative. The decision still depends on expected value, strategic priority, risk tolerance, resources and consequences outside the truth of the analytical claim. Analytical Capacity should provide a defensible account of what the evidence supports. Decision making then uses that account alongside goals and constraints.

The six dimensions keep reopening one another

The six dimensions are useful because they expose different kinds of analytical work. Their relationship becomes misleading when they are read as six stages completed once in a fixed order.

The commercial margin example becomes more realistic once the dimensions start reopening one another. The team initially frames the issue around discounting, selects transaction data and organises performance by market. The pattern looks convincing until interpretation reveals that returns were booked differently after a system change. That discovery changes which information is relevant and forces the team to reorganise the data. A second analysis weakens the discounting explanation, so competing hypotheses are generated and the original frame is widened to include product mix and logistics. Reasoning then reveals that two markets still behave differently enough to resist one explanation, and the provisional conclusion is narrowed accordingly.

This kind of movement is consistent with research on problem formulation, information search and metacognitive regulation. Later analytical activity changes earlier activity because each new representation alters what the analyst knows about the problem. Framing guides evidence selection, but new evidence can make the frame untenable. Organisation supports interpretation, while interpretation can reveal that the organisation has hidden a relevant distinction. A tentative conclusion can expose an evidential gap and reopen the search.

Metacognition operates across these movements by monitoring understanding, strategy and confidence. Stebner et al. (2022) show that metacognitive strategies can transfer under designed conditions while spontaneous transfer remains limited, and Guo's (2022) meta-analysis finds that metacognitive prompts can improve self-regulated learning and learning outcomes. Monitoring helps an analyst recognise that a route is failing or that confidence has moved ahead of evidence. It cannot supply the missing domain knowledge or make an invalid inference valid by itself.

The resulting architecture is recursive: the six dimensions remain analytically distinguishable because different failures can occur in each, yet their practical interdependence allows a weakness discovered in one dimension to change the work required elsewhere.

Analysis works through knowledge

Analytical Capacity becomes empty if it is imagined as a content-free ability that can produce expert judgment anywhere. A skilled analyst entering an unfamiliar medical, legal, engineering or financial domain still lacks the concepts, causal models, conventions and standards of evidence that make much of the information intelligible.

Domain knowledge enters every dimension by helping the analyst recognise what evidence is relevant and which source should be treated cautiously. It provides categories and schemas for organising information, while making some representations available that a novice would not construct. Interpretation depends on knowing which relationships are meaningful in the domain and which apparent patterns are ordinary. Reasoning draws on domain-specific assumptions, and conclusion quality depends partly on understanding how much the available evidence can legitimately support within that field.

The relationship between expertise and problem representation has been visible since foundational expert-novice studies. Chi, Feltovich and Glaser (1981) found that physics experts and novices categorised problems differently, with experts organising more readily around underlying principles while novices relied more on surface features. Contemporary disciplinary research continues to show how domain-specific representations shape reasoning, including work on scientific and chemical problem solving (Talanquer, 2022; Karch and Sevian, 2022).

Accumulated knowledge still leaves meaningful differences in how people search for evidence, test alternatives, monitor confidence and revise a frame. Otero, Salgado and Moscoso's (2022) meta-analysis shows substantial overlap between cognitive reflection and cognitive ability, while Sobkow, Olszewska and Sirota (2023) find distinguishable latent factors among reflection, numeracy and fluid intelligence. The evidence supports an interaction among cognitive resources, analytical strategies and domain knowledge, with no single master variable explaining analytical performance.

That interaction also sets a hard limit on transfer. A person may learn a broadly reusable process such as checking alternative explanations or expressing confidence probabilistically, yet competent use in a new domain still requires knowledge of what counts as plausible, relevant and diagnostic there. Gobet and Sala (2023) review the broader cognitive-training literature and conclude that far-transfer claims have often exceeded the quality of the evidence. Stebner et al. (2022) similarly find that metacognitive transfer benefits from designed support because spontaneous application is unreliable.

Analytical processes can therefore travel farther than expert conclusions. Someone may carry habits of source checking, assumption testing and calibrated confidence from one context to another. The quality of the resulting analysis still depends on whether those processes can operate on sufficiently accurate knowledge of the new domain.

Human analysis is bounded, contextual and emotionally exposed

The standard for Analytical Capacity cannot be an imaginary analyst with unlimited attention, perfect memory and enough time to evaluate every possible explanation. Simon's (1955) account of bounded rationality remains relevant because real analysis always takes place under constraints on information, computation and time. Because attention and time are limited, analysts have to select evidence, simplify representations and eventually stop searching. The analytical question is whether those reductions are appropriate to the stakes and structure of the problem.

Heuristics belong inside this picture as tools whose usefulness varies with context. A shortcut that performs well in a stable, feedback-rich environment can become dangerous when the environment changes or when the cue being used loses validity. Link and Raab's (2022) expert base-rate study shows why laboratory demonstrations of bias have to be interpreted within the task ecology in which the judgment occurs. Strong analysis requires sensitivity to when a familiar shortcut fits and when the structure of the problem demands more deliberate work.

Cognitive control supports that flexibility without defining analytical quality itself. Egner's (2023) review distinguishes processes involved in maintaining task focus from those involved in switching when conditions change. An analyst needs enough stability to resist irrelevant distraction and enough flexibility to abandon a failing representation. Switching constantly would be as unhelpful as persevering indefinitely; the value lies in matching control to the analytical situation.

Creative Capacity becomes especially relevant when the existing explanation space is too narrow. Alternative hypotheses, counterfactuals and reframes have to be generated before they can be evaluated. A purely evaluative system can compare only the options already present. Analytical Capacity contributes the evidential discipline that tests those possibilities, while creative processes can expand the set from which the analysis works. Pham, Magistretti and Dell'Era's (2023) research on problem framing shows how closely these functions can interact in ill-defined problems.

Emotion influences analysis through attention, risk perception, motivation, confidence and persistence. Research does not support one universal direction in which emotional involvement simply degrades judgment. Bartholomeyczik, Gusenbauer and Treffers' (2022) systematic review and meta-analysis finds differentiated effects of incidental emotions on decisions under risk and uncertainty, and Duque, Cano-López and Puig-Pérez (2022) report similarly heterogeneous effects of stress and cortisol across decision tasks. Emotional salience can draw attention to a genuine risk or narrow attention too aggressively; anxiety can encourage further checking or push the analyst towards premature certainty. Emotional Capacity matters because regulation can help keep these influences from taking control of the evidential standards being applied.

Motivated reasoning adds another source of exposure because people care about some conclusions independently of their truth. Prior beliefs, identity, incentives and organisational pressure can affect which evidence is sought or believed. Maguire et al. (2022), in preregistered work on politically salient COVID-19 judgments, found strong effects of prior beliefs without supporting a simple claim that greater cognitive sophistication necessarily amplifies politically motivated reasoning. Analytical vulnerability emerges from the interaction among motivation, knowledge, context and reasoning, with no single mechanism explaining every failure.

Analytical Capacity can be developed, within limits of transfer

A useful developmental framework needs stronger evidence than the observation that people become better with experience. The research base on critical-thinking instruction, problem-based learning, metacognitive prompting, argument mapping, calibration and debiasing provides evidence that several component processes can improve through deliberate intervention.

Liu and Pásztor's (2022) meta-analysis finds a positive overall effect of problem-based learning on critical-thinking outcomes in higher education. Guo (2022) reports positive effects of metacognitive prompting in computer-based learning, with intervention design and feedback affecting results. More recent synthesis around argument mapping identifies a substantial experimental literature in which inferential structures are made explicit and learners practise constructing or evaluating them (Nesbit and Liu, 2025). Targeted interventions can also improve local source evaluation and reduce particular causal illusions (Lu et al., 2023; Rodríguez-Ferreiro, Vadillo and Barberia, 2023).

These literatures support targeted developmental gains within a highly variable intervention landscape. Constructs, duration, tasks and outcome measures differ substantially across studies. Improvement on a test that closely resembles the training can reflect near transfer, better strategy use, greater familiarity or additional domain knowledge. Far transfer requires evidence from meaningfully different tasks and contexts.

Kohmer et al. (2025), examining generic and domain-specific critical online reasoning, directly engages this problem by testing training and transfer across contexts. Stebner et al. (2022) likewise show that metacognitive transfer can be supported while remaining unreliable when learners are expected to recognise its relevance spontaneously. Gobet and Sala's (2023) broader review provides a strong warning against interpreting improvements on trained cognitive tasks as proof of generalized capacity gains.

Development is therefore strongest when practice targets the processes we want to improve and gives learners a reason to use them again in varied settings. Explicit reasoning criteria can make quality inspectable, while worked examples can reveal how an expert frames and structures evidence. Comparing strong and weak analyses can expose subtle differences in evidential support. Feedback can address both the answer and the route used to reach it. Repeated confidence judgments can make calibration visible, while carefully varied tasks create opportunities to recognise structural similarities across contexts.

Professional development can use the same principles without turning analysis into classroom exercises. A procurement team can compare two supplier investigations that reached different conclusions from the same evidence. A marketing team can revisit a failed campaign diagnosis after new data is introduced. A manager can be required to document which evidence would change a recommendation before presenting it. These designs create practice in the analytical processes themselves while keeping the work anchored in relevant Knowledge Domains.

Assessment has to reveal the analytical route

Recording whether the final answer is correct captures too little of what Analytical Capacity is meant to represent. A correct conclusion can come from sound analysis, prior familiarity, guessing or an invalid path that happens to terminate at the right answer. A wrong conclusion can emerge from a defensible process working with incomplete or misleading evidence. Outcome quality remains essential, but it cannot diagnose the whole capacity.

Performance-based assessment offers a stronger route because the task can expose the analytical process as it unfolds. Van Damme et al. (2023), using large international performance-based critical-thinking data, demonstrate that complex thinking can be assessed across institutional and national settings while also showing variability associated with educational and background factors. The broader measurement literature still warns against assuming that one score provides a pure readout of a domain-general capacity.

A serious Analytical Capacity assessment would therefore use several authentic or scenario-based problems and collect evidence from intermediate work. The assessor could observe which sources the person chooses, whether missing evidence is noticed, how information is represented, what assumptions are made explicit, whether alternative explanations are considered and how confidence changes as evidence develops. Constructed responses are especially useful because they reveal the relationship between claims and evidence instead of asking the participant only to recognise a correct option.

The six dimensions leave different traces in a performance task. Search choices, source rejection, gap recognition and stopping rationale provide evidence for Selecting information. Organising information becomes visible in the representation itself, including whether tables, maps or categories expose relevant relationships without hiding contradictory evidence. The quality and scope of the analytical question, along with any revision after new evidence appears, provide evidence about Framing the analysis. Patterns, causal claims, base rates and treatment of uncertainty then reveal aspects of Interpreting information. For Structuring reasoning, the assessor can examine premise-evidence relationships, alternative explanations and the handling of counterevidence, while Reaching sound conclusions becomes visible in evidence-conclusion fit, calibration, stated limitations and willingness to revise.

Assessment can become more diagnostic when the task is perturbed after a provisional conclusion, for example by introducing one credible piece of counterevidence. The assessment can then observe whether the person updates appropriately, identifies which part of the reasoning has changed and adjusts confidence in proportion to the new evidence. This creates a direct test of analytical revision that a static multiple-choice instrument cannot provide.

Self-report can supplement this evidence by describing habits or dispositions, but it should not be mistaken for demonstrated analytical performance. A person who reports that they “always consider alternative views” has supplied information about self-perception. Demonstrating the capacity requires a task in which a plausible alternative is available and the person's reasoning reveals whether it was genuinely examined.

A higher-order Analytical Capacity score would eventually require psychometric evidence showing enough covariance across dimensions and tasks to justify aggregation. Our conceptual architecture gives a strong basis for designing those measurements; empirical validation would still have to establish whether the six dimensions are sufficiently distinct for diagnosis and sufficiently coherent for a reliable overall score.

AI changes where analytical work is concentrated

Generative AI can produce summaries, comparisons, explanations and recommendations at a speed that makes some traditional analytical tasks feel almost costless. The availability of those outputs changes the distribution of work because information can arrive already organised, interpreted and expressed as a conclusion. The human analyst then faces a different problem: deciding whether that apparently finished reasoning deserves to be trusted.

Recent evidence is already showing why the answer depends on how AI is used. Li, Cui and Hagedorn's (2026) systematic review synthesised 67 empirical studies published from 2022 to 2025 on ChatGPT and critical or creative thinking in higher education. More favourable outcomes appeared when use was scaffolded around reflection, justification and inquiry, while less structured use was more often associated with cognitive offloading and superficial engagement. The underlying literature remains dominated by short-term and heterogeneous designs, which makes long-run claims about enhancement or deskilling premature.

Human-AI decision research shows a related problem with reliance. Lu, Wang and Yin (2024), across three preregistered experiments, found that supplying additional opinions could reduce over-reliance while increasing under-reliance. More advice did not produce a monotonic improvement in judgment because users still had to decide which source deserved weight. Dunning, Fischhoff and Davis (2024) likewise examine when people appropriately heed AI recommendations, showing why trust and reliance have to be calibrated against actual system performance.

Explanations create another trap when fluency is confused with verifiability. Fok and Weld (2024) argue from the human-AI evidence that explanations help complementary performance mainly when they reduce the cost of verifying whether the recommendation is correct. A coherent explanation can increase understanding or trust without giving the user the domain evidence needed to evaluate accuracy. Salimzadeh, He and Gadiraju (2024) further show that task uncertainty and complexity can change reliance on AI, which is especially consequential when the user has difficulty determining whether the output is right.

AI changes the evidence environment by making provenance checks for synthetic content part of Selecting information, while automated clustering and summarisation make Organising information easier to outsource and therefore more important to inspect for hidden disagreement or uncertainty. Framing carries a different risk because a poorly posed question can be elaborated fluently without ever becoming the right analytical question.

Interpretation and reasoning then face the problem of generated plausibility: an explanation may sound coherent before its evidence has been verified, so claims need traceable support and alternatives still need to be tested. Reaching sound conclusions places the final burden on calibration, because the analyst's confidence has to remain grounded in evidence even when the system uses highly certain language.

The practical value of Analytical Capacity therefore changes as machines absorb parts of information handling. Verification, framing, source judgment and uncertainty become harder to avoid because the analyst can receive a polished answer before establishing whether the underlying analytical work has actually been done.

Building better analysis

Analytical Capacity describes the work through which information becomes a conclusion that deserves its confidence. That work begins with the construction of the problem and continues through selective evidence gathering, representation, interpretation and inference. A conclusion closes the process only provisionally because credible new evidence can reopen any part of it.

The six dimensions give this work a usable structure without turning analysis into a mechanical sequence. The evidence environment begins with Selecting information, while Organising information and Framing the analysis shape what can be seen and what problem the evidence is being asked to explain. As the investigation develops, Interpreting information turns structure into warranted meaning and Structuring reasoning tests whether the inferential relationships can carry the claims placed on them. Reaching sound conclusions controls the final relationship between claim strength and uncertainty. Any of these dimensions can expose a weakness that sends the analyst back into another part of the process.

Development therefore needs to reach the processes beneath the finished answer. People need opportunities to investigate problems whose frame is not already supplied, work with evidence of uneven quality, make assumptions visible, test competing explanations and revise confidence when the evidence changes. Relevant knowledge remains indispensable because analytical processes need something accurate to operate on, and transfer becomes increasingly difficult as contexts move farther from the conditions in which those processes were learned.

The same discipline becomes more consequential in an information environment capable of generating plausible analysis on demand. A fluent answer can compress the visible effort of analysis while leaving the epistemic work unresolved. As information generation becomes cheaper, analytical quality depends increasingly on determining which question deserves investigation, which evidence deserves weight, what that evidence can support and how strongly the resulting conclusion should be held.