Information Load and Decision Efficacy

• Information Load and Decision Efficacy is defined by measuring cognitive burden through self-reported scales and quantitative feature counts to assess decision performance.
• Excess information can both clarify and overwhelm, with studies illustrating non-linear impacts on accuracy, response time, and overall mental efficiency.
• Adaptive interfaces using dynamic sub-selection and transparency policies improve decision outcomes by minimizing overload and aligning information presentation with user capabilities.

Information load is defined as the quantitative and qualitative burden imposed on the cognitive resources (attention, working memory, executive functions) of a decision-maker by the volume, complexity, and format of presented data, cues, or explanations. Decision efficacy captures the outcome quality—accuracy, efficiency, reliability—of human decisions made under varying information loads. Across domains from clinical diagnostics to financial trading and collective survey design, the functional relationship between these constructs is highly non-linear: excess information may both clarify and overwhelm, shaping user autonomy, engagement, and outcome optimality.

1. Formal Constructs and Measurement Paradigms

Research operationalizes information load through direct, self-reported mental effort scales (e.g., 7-point Likert for cognitive load (Herm, 2023), NASA TLX subscales for mental demand and stress (Rezaeian et al., 28 Jan 2025)), or indirectly via the count and complexity of features, options, or explanation elements presented to a user (Huang et al., 2024, Cheng et al., 6 Mar 2025). Quantitative frameworks model load as cardinal measures—for a subset selection mask $b \in \{0,1\}^d$, $I(b) = ||b||_1$—or as a complexity scalar $\kappa$ scaling intrinsic information difficulty (Du et al., 18 Jun 2025).

Decision efficacy is multifaceted: commonly quantified as percentage accuracy (task performance), task time (seconds to decision), and composite metrics such as mental efficiency,

$$ME = \frac{Z(\mathrm{Perf}) - Z(\mathrm{Effort}) - Z(\mathrm{Time})}{\sqrt{3}},$$

with each component $Z(\cdot)$ standardized within participant (Herm, 2023). In clinical and collective-choice domains, additional metrics include reliance rates, edit distances (for navigation cost in surveys), and utility-derived outcome measures (Langtry et al., 2024, Cheng et al., 6 Mar 2025).

2. Cognitive Load Theory and Mechanisms of Impairment

Information load interacts with human cognitive architecture through both intrinsic (task and stimulus complexity) and extraneous (interface design, presentation format) channels. Sweller’s Cognitive Load Theory emphasizes the partition of load into intrinsic, extraneous, and germane components (Cheng et al., 6 Mar 2025). High information load can exceed available cognitive capacity, inducing selective attention failures

$$\Pr\{\text{Process }I_j\} = \frac{\exp(-\gamma L(I_j; \kappa))}{\sum_{k=1}^M \exp(-\gamma L(I_k; \kappa))},$$

processing errors

$$\text{Error}_{ij} = \epsilon_{ij} \times g(L(I_j; \kappa), C_i),$$

and strategic overload avoidance, such as satisficing (Du et al., 18 Jun 2025, Cheng et al., 6 Mar 2025).

The "Transparency Paradox" formalizes autonomy depletion via stochastic geometric Brownian motion:

$dA_t = \mu(I_t) A_t dt + \sigma A_t dW_t,$

where information-dependent drift $\mu(I_t) = \mu_0 - \beta I_t - \gamma I_t^2$ captures the diminishing returns and eventual negative impact of excessive information on user engagement and perceived control (Margondai et al., 20 Jan 2026).

3. Decision-Theoretic and Value-of-Information Formulations

Decision theory frames the optimal use of information through the constructs of Expected Value of Revealed Information (EVRI) and Expected Value of Displayed Information (EVDI) (Horvitz et al., 2013):

$$EVRI(e; E, E) = \sum_j p(H_j\,|\,E \cup e)\,u(A^*(E \cup e), H_j, t(E \cup e)) - \sum_j p(H_j\,|\,E)\,u(A^*(E), H_j, t(E)).$$

$$EVDI(e;E,E) = \sum_i p(A_i \mid E \cup e) \sum_j p(H_j \mid E) u(A_i, H_j, t(E \cup e)) - \sum_i p(A_i \mid E) \sum_j p(H_j \mid E) u(A_i, H_j, t(E)).$$

These quantify the marginal gains in expected utility from incrementally displaying, highlighting, or suppressing evidence, while factoring cognitive review time and user expertise.

In systems design under uncertainty, on-policy Value of Information (VoI) analysis similarly quantifies the economic and performance impact of measurement-induced information increments (Langtry et al., 2024):

$$EVII_P(e) = E_{z}[ E_{\theta|z}[ u(P_u(\pi(\theta|z)), \theta) ] ] - E_{\theta}[ u(P_u(\pi(\theta)), \theta) ].$$

4. Empirical Evidence: Explanation Granularity and Interface Design

Herm et al. (Herm, 2023) empirically demonstrate that low-information, local explanations ("Why?", "Why-Not?") reduce mental effort, maximize accuracy ($\sim$87%), and minimize time-cost, yielding superior mental efficiency over global, high-information types. Table below summarizes their findings:

Explanation	Median Effort	Accuracy (%)	Mean Time (s)	Mental Efficiency
Baseline	6.0	48.6	72.6	-0.34
How	5.0	55.0	51.7	-0.15
How-To	5.0	65.0	49.8	-0.11
What-Else	4.0	68.0	60.1	-0.08
Why	2.0	87.0	34.5	+0.34
Why-Not	3.0	81.0	38.9	+0.23

Excessive explanation detail, particularly in novices, degrades mental efficiency. In clinical contexts, the introduction of complex visual and probabilistic cues can amplify perceived stress without commensurate accuracy improvement (Rezaeian et al., 28 Jan 2025). High AI confidence scores drive overreliance and lower cognitive load, but risk decreased diagnostic performance through automation bias (Rezaeian et al., 28 Jan 2025, Cau et al., 2 May 2025). Hybrid and personalized interfaces—combining feature-based transparency with counterfactual contrast and confidence calibration—optimize efficacy under variable user profiles and dynamic uncertainty (Cau et al., 2 May 2025, Huang et al., 2024).

In survey-based collective decisions, two-phase “organize-then-vote” interfaces reduce navigational and operational load, shifting cognitive effort toward strategic preference construction and away from mechanical satisficing (Cheng et al., 6 Mar 2025). Edit-distance metrics and NASA-TLX scores confirm that scaffolding and chunking of information presentation improve engagement with complex decision tasks.

5. Dynamic Sub-Selection and Adaptive Transparency Policies

The Dynamic Information Sub-Selection (DISS) framework (Huang et al., 2024) formalizes instance-level adaptation of information presented to black-box decision-makers, optimizing the trade-off between information load and performance reward:

$$\pi^* = \arg\max_{\pi \in \Pi} \mathbb{E}_{(x, y) \sim \mathcal{D}} \left[ r(y, M(x \odot \pi_b(x), \pi_o(x))) \right],$$

with regularization explicitly penalizing information overload. Empirical results consistently show that policies minimizing information load (by feature selection, option masking, or explanation pruning) often recover near-oracle decision efficacy—especially under simulated overload, risk aversion, or simplicity bias.

The transparency paradox and adaptive transparency principles (Margondai et al., 20 Jan 2026) extend this to time-evolving autonomy budgets, with optimal information presentation determined by instantaneous thresholds:

$$u^*(A, I, t) = \begin{cases} u_{\max}, & V_I(A, I, t) > \tfrac{c}{\alpha} \ 0, & V_I(A, I, t) < \tfrac{c}{\alpha} \end{cases}.$$

Dynamic, personalized adjustment and real-time cognitive state monitoring yield superior user engagement and cumulative decision-quality than static, maximal transparency.

6. Implications for Interface Design, Policy, and Future Research

Design principles systematically derived from these studies include:

1. Limit displayed “chunks” to cognitive bandwidth: 5–9 items for time-critical decisions (Horvitz et al., 2013), one option per step in collective surveys (Cheng et al., 6 Mar 2025).
2. Prioritize local, contrastive explanations; suppress global views for novices or high-stakes contexts (Herm, 2023, Cau et al., 2 May 2025).
3. Calibrate confidence signals; shield users from automation bias by exposing low-confidence cues and demanding deliberate user response (Rezaeian et al., 28 Jan 2025, Cau et al., 2 May 2025).
4. Scaffold organization (categorization) before consolidation (commitment) to reduce search cost and prevent early disengagement (Cheng et al., 6 Mar 2025).
5. Model and respect user’s working memory and autonomy limits; employ threshold-based dynamic policies to prevent cognitive overload and depletion (Margondai et al., 20 Jan 2026).
6. Use decision-theoretic evaluation (EVRI, EVDI) to select and highlight only high-utility evidence (Horvitz et al., 2013).

Empirical validation, including laboratory studies and natural experiments (e.g., financial disclosure reform (Du et al., 18 Jun 2025)), confirms that information complexity and overload systematically impair outcome speed and accuracy, disproportionately affecting less sophisticated users and generating large-scale welfare losses. Policies that structure, prune, or adapt displayed information consistently restore or amplify decision efficacy.

7. Controversies and Theoretical Limits

Recent meta-analyses (Hullman et al., 2024) caution that claims about bias or decision loss under variable information load depend critically on experimenter specification of decision problems—state space, signal structure, priors, and utility functions. Without full normative characterization and user access to sufficient information, observed “inefficiencies” may reflect experimental artifact rather than cognitive failure.

In energy systems optimization, the marginal value of information from monitoring is found to be minimal (EVSI $<$2% total cost) compared to standard population-profile design—even when operating under high load uncertainty (Langtry et al., 2024). This suggests practical upper bounds to information acquisition in many policy domains.

---

The corpus establishes that information load exerts a controllable, quantifiable influence on decision efficacy. The relationship is mediated by cognitive architecture, user expertise, interface design, and adaptive policy. Optimal decision support requires joint calibration of information volume, granularity, and timing—empirically justified across domains, and grounded by rigorous decision-theoretic and cognitive models.