Insurance by BogdanovaThe Encyclopedia of Insurance and Risk Science

RESEARCH DOSSIER

Latent Consideration Sets in Insurance Choices

Abstract. Insurance choice is ordinarily observed at the end of a much longer decision process: an individual selects one insurer, policy, deductible, liability limit, or health plan, while the set of alternatives seriously evaluated remains hidden. This monograph examines latent consideration sets as a fundamental source of observational ambiguity in insurance demand. It reconstructs the development of consideration-set theory, integrates stochastic attention and search with contingent-claims valuation, and compares evidence from automobile, property, employer health, and Medicare prescription-drug insurance. The analysis distinguishes availability, awareness, eligibility, active evaluation, perceived affordability, and the final conditional choice. It establishes why preference estimates, willingness to pay, observed inertia, price elasticity, and consumer-welfare counterfactuals can be misspecified when researchers impose full consideration on restricted decisions. A tractable two-stage probability representation is derived as an analytical exposition of existing econometric approaches, followed by its identifying assumptions, sensitivity tests, and limits. The monograph critically examines the 2026 evidence on liability limits and deductibles, distinguishing authors' structural estimates from directly observed data. It concludes that credible insurance-choice analysis requires joint attention to the feasible contract menu, the formation of the considered subset, choices conditional on that subset, and the institutional consequences of making an option salient. These are analytical conclusions from the literature, not the announcement of a new theory or a newly estimated empirical result.

Keywords: Insurance choice; latent consideration sets; limited attention; stochastic consideration; insurance demand; deductible selection; liability limits; search; default effects; Medicare Part D; identification; structural estimation; risk preferences; consumer welfare; choice architecture.

Introduction

An insurance purchase records a choice, but a choice is the product of a process whose earlier stages often leave no administrative trace. A selected $500 deductible does not reveal whether the policyholder compared all available deductibles, saw only the default and one alternative, asked an intermediary for one recommendation, or ruled out a larger deductible before calculating any expected payment. A household that retains a health plan may value its provider network, find switching burdensome, fail to notice an alternative, or be automatically re-enrolled. The same terminal selection is compatible with radically different causal accounts. These distinctions affect research on demand, pricing, disclosure, distribution, market power, policy design, and welfare.

This dossier treats the consideration set as the subset of potentially feasible alternatives that the decision-maker actually brings into meaningful comparison at the point of choosing. The set may be stochastic, heterogeneous, context-dependent, and unobserved. It can contain only the incumbent or default; it may be enlarged by a broker, comparison website, enrollment notice, or price shock. Forming that set is analytically distinct from evaluating the consequences of each considered insurance contract. The distinction is established in the economics and marketing literature, including Hauser and Wernerfelt's evaluation-cost analysis (1990), Manzini and Mariotti's stochastic choice model (2014), and Barseghyan, Molinari, and Thirkettle's insurance-choice identification study (2021).

The scientific problem is an inverse problem. The investigator knows the observed selection and some contract and household characteristics, then seeks to infer risk preferences or demand. The result depends on what the investigator assumes was available and considered. An inference of strong risk aversion from an expensive low deductible may instead reflect unequal attention across deductibles; an inference of brand loyalty from renewal may instead reflect unsearched alternatives. Both preference and attention may operate simultaneously. The goal is therefore to represent the ambiguity explicitly, identify the conditions under which it can be reduced, and state what remains uncertain after estimation.

The analysis uses published economic theory, refereed empirical studies, experimental findings, and a clearly identified 2026 working-paper research account. It has international conceptual scope, while every quantitative finding is attributed to its original jurisdiction, product, and study population. No individual transaction dataset was obtained or reestimated for this dossier. Equations and examples below are analytical reconstructions and transparent illustrations; they do not constitute a novel estimator, a new psychological doctrine, or additional empirical observations.

Scope, terminology, and related research

Three distinctions establish the object. The offered menu is the set of options a distributor, platform, employer, or regulated market presents. The feasible menu is the subset a given household can legally and practically purchase, accounting for eligibility, underwriting, affordability, and contractual conditions. The considered set is the subset of feasible contracts that enters deliberation as a live candidate. A product may be advertised but ineligible, eligible but unknown, known but ignored, or inspected and rejected after meaningful evaluation. This dossier reserves “consideration” for the latter decision-relevant stage; awareness and gaze are possible indicators rather than interchangeable measurements of it.

A latent consideration set is unobserved in the estimation data. The word does not imply that researchers have demonstrated a particular psychological state. It describes a missing intermediate variable in the generation of choices. Limited consideration denotes a considered set smaller than the relevant feasible menu. Stochastic consideration permits that subset to change across otherwise similar decisions. Conditional choice describes selection among considered contracts given beliefs, preferences, and information. The concept may be implemented through multiple existing models, whose different restrictions must be evaluated against the data.

The principal Science association is Behavioral Insurance and Risk Perception. The same research object contributes to Insurance Statistics and Predictive Modeling, Insurance Economics, Automobile Insurance, and Insurance Consumer Protection. General search costs, choice defaults, probability weighting, and plan complexity retain distinct research objects elsewhere in the encyclopedia; they enter here only insofar as they cause or help identify consideration-set formation.

Available versus Considered Alternatives

From evoked sets to econometric choice models

The lineage of consideration sets begins with the observation that a consumer may know numerous alternatives but compare only a few. Hauser and Wernerfelt (1990) formalize the tradeoff between the expected advantage of expanding the set and the cost of evaluating another brand. Their original evaluation-cost framework separates evaluating an option, including it in a considered set, and finally consuming it. Applied to insurance, acquiring a quote, understanding exclusions, and selecting an indemnity contract are likewise distinguishable activities. Additional available policies may improve a household's opportunity set without meaningfully improving its decision if their assessment costs prevent entry into consideration.

Manski's (1977) random utility foundations provide a conceptual basis for separating probabilistic observations from deterministic preference orderings. Subsequent limited-attention work makes the unobserved choice subset a distinct explanatory component. Masatlioglu, Nakajima, and Ozbay (2012) show through revealed-attention analysis why inferences from menu choice must incorporate what the agent attended to; Manzini and Mariotti (2014) introduce independent stochastic consideration and establish identification results under that model's restrictions. These are substantive theoretical contributions by their authors. This dossier compares and applies them without renaming their properties as new principles.

A useful general lesson follows from this history. Full-menu utility maximization is a special description of a decision, with a maintained assumption about information and opportunity. When the selected contract is interpreted as the best among every possible contract, the analyst silently makes consideration a certainty. The consequence is structural: observed absence of demand for an alternative becomes evidence of weak preference for it, even when its probability of reaching comparison is low.

The empirical-search literature, synthesized by Honka, Hortaçsu, and Wildenbeest (2019) in Empirical Search and Consideration Sets, places consideration-set formation within sequential or simultaneous information acquisition. That connection matters to insurance because many prices are personalized and cannot be inferred from advertisements alone. A household may need to provide a driver history, estimated mileage, property characteristics, or medical information before receiving a usable quote. The relevant menu may therefore materialize through search rather than exist as a fully known table of alternatives.

The insurance-specific choice environment

Insurance contracts differ along numerous dimensions: premium, deductible, limit, coinsurance, exclusions, waiting periods, claim service, insurer solvency, provider network, pharmacy formulary, renewal terms, and coverage triggers. Their quality is also state-contingent. The lowest premium is not necessarily the cheapest contract over the loss distribution; the contract with the largest nominal limit may be poor value after exclusions; a health plan's provider-network consequences may dwarf a simple premium comparison. These features increase the informational requirements of meaningful comparison.

The feasible set must be measured at the decision-maker level. In automobile insurance, state-specific legal minimums and insurer underwriting may constrain offered liability limits. In an employer health plan, employment eligibility and the employer's menu delimit real options. In Medicare Part D, residence, enrollment rules, plan year, subsidy status, and formularies affect feasible choices. A model that labels unavailable plans “ignored” inflates inferred inattention. Conversely, a model that treats only displayed options as feasible may miss eligible plans hidden by a comparison interface.

A contract-menu researcher should record at least five successive states: technically possible, legally and commercially eligible, disclosed or encountered, considered, and selected. These states are neither necessarily nested across time nor perfectly measured; for one decision episode their substantive ordering is a disciplined way to characterize the evidence. An excluded insurer cannot simultaneously count as a feasible rejected choice. An insurer noticed incidentally need not count as a contract seriously compared.

A latent-set explanation and its rivals

Imagine three hypothetical auto policies with annual premiums of $800, $900, and $1,100, respectively, and identical contract quality apart from a deductible schedule. Observing purchase of the $1,100 policy can be rationalized by strong preference for its deductible, a nonprice brand attribute, uncertainty about loss probability, a default, switching costs, or exclusion of the cheaper alternatives from the considered set. The observation alone does not rank these explanations. Distinguishing them requires variation in exposure to alternatives, valid measures of attributes, independent measures of beliefs, or credible experimental interventions.

This is important because the considered set is itself an outcome. Search costs, preferences, prior experiences, loss histories, age, literacy, interface design, channel, price, and social recommendations can affect which contracts are considered. Preferences can even determine attention: a highly risk-averse consumer might seek only low-deductible plans; another consumer might attend only to the lowest visible premiums. Treating consideration as random noise statistically independent of preferences can therefore produce its own specification error.

Search and brand consideration

Honka's (2014) auto-insurance search-cost study estimates a consumer's decision about how many insurers to search and whether to switch. In the U.S. data studied, the paper reports estimated search costs of approximately $35–$170 and an average switching cost of about $40. These are model-based estimates for the study environment, not universal prices of attention, and they pertain principally to insurer search rather than the selection of a deductible within an already chosen policy.

Honka and Chintagunta (2017) use consideration-set and price information in the U.S. automobile insurance industry to compare simultaneous and sequential search. Their empirical evidence favors simultaneous search under their framework, and the paper shows that choosing the wrong search process distorts simulated consideration shares and purchase shares. The implication is that an estimated attention parameter can inherit misspecification from the assumed search technology. A consumer who collects several quotes together and a consumer who stops searching after a good first quote may display similar final choice yet different counterfactual responses to new advertising or price dispersion.

The distinction between insurer-level search and within-insurer contract comparison deserves special attention. Shopping among carriers and selecting among liability limits are nested or interacting decisions, not necessarily one simple list. The number of insurers quoted and the number of deductible alternatives compared may be measured separately. A single observed policy reflects choices about product category, carrier, coverage type, limit, deductible, and riders. Aggregating these into one undifferentiated menu can hide where attention becomes restricted.

Attention and Consideration Probabilities

Two-stage probability representation

Let an individual be indexed by i, the set of feasible insurance contracts by F_i, and the unobserved nonempty consideration set by C_i, with C_i a subset of F_i. Let x_i denote observed consumer and market attributes. A general decomposition of the probability of selecting contract j is

Pi(j∣Fi,xi)=∑C⊆Fi: j∈CPr⁡(Ci=C∣Fi,xi) Pi(j∣C,xi).P_i(j\mid F_i,x_i)=\sum_{C\subseteq F_i:\,j\in C}\Pr(C_i=C\mid F_i,x_i)\,P_i(j\mid C,x_i).

This is an application of the law of total probability. It does not by itself identify the distribution of considered sets or the conditional preference model. Many joint distributions of latent consideration and conditional choice can generate the same observed purchase shares. An empirical model acquires identifying content only through restrictions, measurements, or interventions.

In a common stochastic-attention specification, each available contract k enters consideration with probability q_ik, potentially conditional on characteristics, and the decision-maker selects the highest-valued considered option. If independent attention is assumed, the probability of a specific nonempty subset C is proportional to the product of attention probabilities for included and excluded contracts:

Pr⁡(Ci=C∣Ci≠∅,Fi,xi)=∏k∈Cqik∏k∈Fi∖C(1−qik)1−∏k∈Fi(1−qik),C≠∅.\Pr(C_i=C\mid C_i\neq\varnothing,F_i,x_i)= \frac{\prod_{k\in C}q_{ik}\prod_{k\in F_i\setminus C}(1-q_{ik})} {1-\prod_{k\in F_i}(1-q_{ik})},\qquad C\neq\varnothing.

The denominator normalizes away an empty set. Other models instead give the incumbent a mandatory place or specify a fallback to nonpurchase; those produce different observational implications. Independence is an assumption of a particular model, not a universal property of attention. Household-level latent types, bundle presentation, advertising, or broker scripts can make attention events correlated. Manzini and Mariotti's formal stochastic model and Abaluck and Adams-Prassl's empirical identification analysis illustrate different structured approaches.

Conditional on a set, a standard logit representation might take the illustrative form

Pi(j∣C,xi)=exp⁡[Vij(xi)/σ]∑k∈Cexp⁡[Vik(xi)/σ],j∈C.P_i(j\mid C,x_i)=\frac{\exp[V_{ij}(x_i)/\sigma]}{\sum_{k\in C}\exp[V_{ik}(x_i)/\sigma]},\qquad j\in C.

The systematic payoff V_ij must incorporate relevant expected premiums, risk exposure, nonprice features, and possibly utility over loss states; σ is a scale parameter. Logit is one estimation choice, not an inference that actual insurers or households possess Gumbel-distributed preferences. Under this two-stage model, price affects final purchase both through conditional valuation and, if q_ij depends on price or price-related signals, through attention.

A useful decomposition follows from differentiating the purchase probability with respect to a product attribute. In schematic form,

∂Pi(j)∂x=∑C[∂Pr⁡(Ci=C)∂xPi(j∣C)+Pr⁡(Ci=C)∂Pi(j∣C)∂x].\frac{\partial P_i(j)}{\partial x}= \sum_C\left[ \frac{\partial\Pr(C_i=C)}{\partial x}P_i(j\mid C) +\Pr(C_i=C)\frac{\partial P_i(j\mid C)}{\partial x} \right].

The first term captures movement of alternatives into or out of comparison; the second captures choice substitution within considered sets. This is standard probability calculus applied to limited-consideration choice. It exposes why reduced-form demand elasticity alone does not tell researchers which mechanism caused the response.

Default-specific attention

A default-specific structure separates periods of passive retention from active reconsideration. Let d_i be the default and a_i the probability of an active search episode. Then the purchase probability admits the simple conceptual mixture

Pi(j)=(1−ai)1{j=di}+aiPi(j∣active search).P_i(j)=(1-a_i)\mathbf 1\{j=d_i\}+a_i P_i(j\mid\text{active search}).

The active-search term may itself integrate over smaller considered sets. The default can affect the probability of waking up, the considered alternatives, or preferences after active evaluation. A default's impact therefore need not be a single switching-cost coefficient. This distinction is central to Abaluck and Adams-Prassl's (2021) Medicare Part D analysis, which models both passive default retention and limited consideration during active searching.

A persistent choice can arise from the convenience of retaining an incumbent without any additional information processing. It can also arise because the consumer compared alternatives and found the incumbent superior. Administrative data indicating only that the same policy renewed cannot distinguish these circumstances. Studies estimating one generic inertia coefficient should label it a composite unless additional observables separate inattention, genuine switching burdens, and valuations of service continuity.

Awareness, gaze, quotation, and meaningful comparison

Measures of attention require careful interpretation. A quote request is a stronger signal of evaluation than an impression of an advertisement, but a requested quote may have been obtained only to satisfy an intermediary's requirement. A click on a plan card establishes some exposure; it may reflect curiosity, mistake, or a genuine candidate. A postpurchase question asking which plans were considered can be contaminated by rationalization and imperfect recall. Edenbrandt and coauthors (2022), studying awareness and consideration with eye tracking, report substantial differences between visual and stated consideration. Their experiment does not concern insurance, but it directly cautions against equating observed visual attention with selection-relevant consideration.

For insurance research, the most defensible measurement design triangulates traces: displayed menu, scroll and click logs, quotes requested, details expanded, documents opened, stated shortlist, and actual purchase. The resulting measures are indicators with different error structures. The investigator should report which stage a measurement actually records, its timestamp relative to the decision, the possibility of automated default selection, and how missing browsing activity was treated. A zero click in an agent-mediated sale means something different from zero click in an entirely digital sale.

Psychological and institutional determinants

Consideration may depend on working-memory demands, numeracy, perceived loss severity, probability beliefs, financial salience, familiarity with brands, trust, advice, prior claims, and social context. It also depends on interface and market organization: ordering, displayed premiums, number of alternatives, comparison defaults, recommendation labels, and intermediary incentives. Some attributes influence both utility and attention. A low premium can make a plan more desirable after evaluation and more likely to appear at the top of a price-sorted screen. An exclusion restriction claiming that price affects only one stage is therefore an empirical hypothesis requiring justification.

Theoretical work on revealed attention (Masatlioglu, Nakajima, and Ozbay 2012) and menu expansion under limited consideration (Lleras et al. 2017) establishes that adding options need not improve actual selection when it changes attention allocation. The welfare meaning of an added insurance plan thus depends on what other plans it displaces from the shortlist and whether it is comprehensible and relevant to the policyholder.

Risk Preferences and Choice Menus

State-contingent valuation of an insurance contract

Insurance differs from routine retail choice because the object of purchase transfers losses across uncertain future states. A stylized contract specifies a premium p_j and an indemnity schedule I_j(L) for uncertain loss L. Under expected utility, its valuation can be expressed as

Vij=Ei[ui(Wi−pj−L+Ij(L))]+sij,V_{ij}=\mathbb E_i\left[u_i\bigl(W_i-p_j-L+I_j(L)\bigr)\right]+s_{ij},

where W_i is available wealth and s_ij represents appropriately interpreted nonfinancial contract attributes. More elaborate models allow subjective loss probabilities, uncertain service quality, reference dependence, and distorted decision weights. The valuation of a deductible or liability limit is jointly shaped by the premium difference, the probability and severity of covered loss, and the marginal value of wealth in relevant states.

The essential identification problem is that the researcher usually sees a selected indemnity schedule, not the full contingent comparison. Under an assumed complete menu, selection of low deductibles can be interpreted as a willingness to pay for lower residual risk. Under latent consideration, the same data may imply that high-deductible alternatives rarely entered the subset, particularly where plans are displayed in bundles or incumbents are sticky. Consequently, estimates of the coefficient on risk exposure can absorb the effect of omitted consideration.

Barseghyan, Molinari, O'Donoghue, and Teitelbaum (2013) estimate a risk-preference model using deductible choices and find an important role for probability distortions in the U.S. insurance environment studied. Their analysis establishes that conventional curvature alone need not account for observed deductible decisions. It does not establish that all unusual choices are caused by attention; nor does a subsequent limited-consideration estimate invalidate the underlying evidence on probability weighting. Multiple mechanisms can jointly determine behavior, and their empirical separation depends on identifying restrictions.

Preferences and consideration may be correlated

A common modeling convenience assumes that the process generating the shortlist is independent of the household's latent tastes, conditional on observed covariates. Insurance provides reasons to question that convenience. A household with strong concern about severe liability losses may deliberately ask for higher limits; someone attracted by a familiar insurer may rarely solicit rival quotes. Consumers who understand financial exposure may both choose high deductibles and search more broadly. In these cases, attention heterogeneity and risk-preference heterogeneity are jointly generated.

Barseghyan, Molinari, and Thirkettle (2021) explicitly study a risk-choice model with unobserved consideration, including circumstances in which the consideration process depends on preferences and observed conditions. They establish sufficient conditions for semi-nonparametric point identification and illustrate the estimator using property-insurance purchases. The force of their result lies in identifying a richer model from structured variation, not in suggesting that any observed insurance dataset automatically identifies every preference and attention parameter.

A 2026 manuscript by Barseghyan, Cohen, and Feng studies insurance risks large and small using auto liability limits and collision and comprehensive deductibles. Its account describes a latent two-type link between preference and consideration heterogeneity across coverage contexts. The paper reports 322,448 household–policy observations from one U.S. auto insurer over nearly two decades; the numbers characterize its observed sample. The implied types, attention probabilities, and utility rankings are model estimates conditioned on functional form and identification choices. This manuscript is dated March 23, 2026 and is treated here as a working-paper study, not as an independently replicated or known peer-reviewed result.

The interpretive danger of preference stability tests

Suppose a household selects a small deductible for collision but a large deductible for comprehensive coverage. Preference-only analysis may treat this cross-coverage divergence as a rejection of stable risk preferences. Yet the household may face different menus, communicate with different employees, receive different defaults, or inspect a small number of different options. Barseghyan, Prince, and Teitelbaum (2011) documented cross-context deductible-choice differences using an expected-utility benchmark. Their findings are evidence about the fit of that benchmark to the studied data. Later models that allow hidden consideration illuminate an additional explanation for such divergences, while leaving open genuine differences in risk perceptions and liquidity needs.

At a methodological level, a stability test of preferences should condition on or bound relevant variation in opportunities and attention. Otherwise it jointly tests preference stability, consideration stability, belief stability, menu measurement, and the imposed functional form. Rejecting the composite null does not identify which component failed. A careful paper reports that composite structure explicitly rather than attributing the entire rejection to psychology.

Deductible versus limit decisions

Deductible decisions usually trade relatively frequent smaller out-of-pocket amounts against premium savings. Liability-limit decisions typically address rarer but potentially much larger indemnity differences, potentially influenced by assets exposed to judgments, legal minimums, and expectations about loss severity. The opportunity and attention processes also differ: the default deductible may be preselected; the liability-limit menu may contain standardized tiers or require explicit broker discussion. A universal single coefficient for attention across both decisions would require evidence rather than convenience.

The same household can show domain-specific behavior through four channels: objectively different stakes; subjectively different loss distributions; different true contract values; and different shortlists. The 2026 working paper's cross-context analysis is significant because it attempts to estimate these channels jointly. Its findings should not be converted into unconditional claims that households universally prefer maximum liability coverage or that consideration explains every unusual deductible pattern.

The role of dominance

An alternative is economically dominated when another feasible alternative yields weakly greater relevant outcomes in every pertinent state and strictly greater outcomes in at least one state, under the appropriate definition of preferences, costs, and contract attributes. Dominance offers a strong diagnostic when contracts genuinely share all material nonfinancial features. Bhargava, Loewenstein, and Sydnor (2017), in Choose to Lose, examine 23,894 employee choices in a U.S. employer health-plan menu and report widespread selection of financially dominated options under their menu comparison. The authors associate the behavior substantially with misunderstandings of insurance. Their sample size is taken from the published journal version; an earlier circulated draft reported a slightly different count.

Even dominated choices do not uniquely establish a latent consideration-set mechanism. Misunderstanding a plan while considering it, attending to a subset without understanding it, belief errors about medical spending, and unexpected restrictions on plan equivalence are conceptually different explanations. The strongest inference from verified dominance is that the complete-information benchmark fails under the study's maintained equivalence assumptions. Additional measures of awareness, comprehension, and evaluation are necessary to allocate that failure to particular mechanisms.

Observed Deductibles and Liability Limits

Evidence matrix and what each study observes

The empirical studies can be grouped by what their investigators directly see. Insurer-shopping studies sometimes record which carriers received quote requests; deductible studies commonly record the selected policy and menu but have weaker observation of consideration; health-plan administrative studies observe a menu, enrollment, claims, and sometimes a survey; laboratory and platform studies can observe the ranking and exposure of alternatives. These data-generating differences matter as much as sample size for claims about latent attention.

Research setting Observed choices and ancillary data Main inference and boundary
U.S. auto-insurer shopping (Honka 2014) Search and purchase information used in a structural search/switching model Search costs are estimated for insurer selection; they do not directly measure within-policy deductible attention.
U.S. auto-insurer search (Honka and Chintagunta 2017) Considered carriers, quoted prices, marketwide price distribution, and selected carrier Comparison of simultaneous and sequential search within a specified search framework.
U.S. property insurance (Barseghyan, Molinari, and Thirkettle 2021) Risky contract choices with contract covariates Point identification under specified variation and structural restrictions; considered sets remain latent.
U.S. automobile coverage (Barseghyan, Cohen, and Feng 2026) 322,448 household–policy observations, liability limits and physical-damage deductibles Joint preference-attention estimates reported in a working paper; no independent replication in this dossier.
U.S. employer health plans (Bhargava, Loewenstein, and Sydnor 2017) 23,894 employee decisions from a menu containing dominated contracts Robust evidence of choice problems under maintained comparability; cannot by itself assign every error to inattention.
U.S. Medicare Part D (Abaluck and Adams-Prassl 2021) Prescription-drug-plan choices and attributes over time An identified structured limited-consideration model; numerical results depend on its population and counterfactual assumptions.
Health-plan information and hassle costs (Handel and Kolstad 2015) Administrative choices/claims linked to consumer information surveys Direct information measures improve separation of preferences from informational and hassle frictions.

The table is a source-critical comparison rather than a statistical meta-analysis. The outcomes differ in products, country-specific institutions, information, and estimation methods. Their coefficients and welfare quantities cannot legitimately be pooled into a universal numerical effect of consideration sets. The studies are especially valuable when their differences are used to determine which conclusions generalize and which mechanisms demand independent measurement.

Automobile insurer shopping

The auto-insurance market illustrates both the usefulness and the ambiguity of search records. A shopper can gather multiple quotes and then select the cheapest acceptable contract; another may recognize only one carrier and continue it. Honka's (2014) structural estimates of consumer search and switching quantify the contribution of search costs to retention in her sample. That estimate relies on a model for how search costs determine the number of searched carriers. Without an observed or identified search process, price differences and brand choice alone can also be explained by unmeasured brand utilities or by heterogeneous valuation of insurer service.

Honka and Chintagunta (2017) exploit distributions of observed prices within the consideration set to distinguish search strategies in auto insurance. Simultaneous sampling determines the set before seeing all prices; sequential sampling can stop as favorable quotes arrive. If a researcher imposes sequential search on data generated by simultaneous comparison, estimated selection into sets and inferred carrier market shares may be biased. Their counterfactual exercise explicitly documents this danger for the largest insurers in their study. The result is about the identification and interpretation of search strategy in a specific U.S. market, not proof that all automobile insurance consumers search simultaneously.

Property-insurance selection

Barseghyan, Molinari, and Thirkettle's (2021) discrete choice under risk is pivotal because it links rigorous identification analysis with a field application involving insurance. By allowing unobserved heterogeneity in consideration and risk aversion, and certain dependence on observed and unobserved characteristics, the study directly addresses why traditional risk-preference identification can fail in insurance purchase data. Its formal results have scope conditions: support of risk and price variation, specification of the menu, sufficient distributional restrictions, and the maintained choice process. An estimate from a dataset with few premium variations or only a single incumbent option cannot inherit identification simply because the econometric equation resembles theirs.

For property insurance, relevant confounders include whether all deductible levels were truly offered to a property, insurer-specific underwriting restrictions, financing or mortgage requirements, and bundling with other lines. Separating these from attention is essential because a supposedly “ignored” higher deductible might have been unavailable or disallowed. For retrospective research, insurer rate manuals and individual quote logs may be as important as the final contract file.

Liability limits: rare high-severity states

The 2026 Barseghyan, Cohen, and Feng working paper reports that its auto-insurance sample spans liability limits as well as collision and comprehensive deductibles. The authors estimate a joint preference-and-consideration specification and report that the dispersion of selected liability limits is explained primarily by consideration heterogeneity in their preferred analysis. Their structural estimates imply that about 70% of households prefer more liability coverage and about 30% the opposite under the estimated model. These percentages describe latent preference rankings inferred by the authors, not directly surveyed rankings of all consumers or a random sample of the world's insurance buyers.

The economic interpretation is plausible in context. Liability protection may preserve significant wealth in low-probability, high-severity scenarios; premiums paid for larger limits can be modest relative to potential protection. A wide spread of selected limits may therefore be difficult to reconcile with a preference-only model using observed pricing and wealth data, especially if the model constrains the utility curvature that can produce those patterns. The possibility of heterogeneity in awareness of available tiers offers another source of dispersion.

Several rival explanations must nevertheless be scrutinized. Households may disagree about suit likelihood or exposure beyond primary limits; asset attachment laws and bankruptcy protections may matter; some may misunderstand umbrella coverage or already hold overlapping policies; brokers may emphasize different limits; coverage tiers may be constrained by eligibility; and the insurer may selectively observe households that have already chosen to buy. The authors' structural fit and counterfactuals compare specified alternatives. They are powerful evidence for the relevance of consideration in that sample, but the model cannot logically exclude all mechanisms that were not independently measured.

A further distinction concerns the meaning of “more coverage.” Higher limits generally offer additional indemnity in some loss states, but they normally carry an additional premium. It follows that a preference for higher limits is conditional on the offered premium schedule and beliefs about liability exposure. A statement that additional limit is universally strictly better would be unwarranted without holding premium and all other features constant.

Collision and comprehensive deductibles

Collision and comprehensive insurance involve claim triggers and loss distributions different from third-party liability. Low deductibles transfer more small-to-moderate losses to the insurer, while higher deductibles reduce the premium and increase retained losses. The 2026 paper analyzes choices across these contexts and reports that latent consideration patterns are correlated with preferences. Its cross-coverage analysis can explain some correlations that preference-only analysis would interpret as context-instability. The analysis does not show that every correlation is attentional or that loss beliefs are uniform across coverage types.

This finding should be read alongside Barseghyan et al.'s (2013) deductible-based probability-weighting estimates and Barseghyan, Prince, and Teitelbaum's (2011) cross-line stability test. These studies pursue different inferential targets under different maintained specifications. The later work changes the set of explanatory possibilities, providing a reason to examine the robustness of risk-preference estimates to unobserved consideration rather than replacing the earlier studies' original research questions.

A concrete estimation hazard arises when a common insurer interface preselects one deductible. The household may interpret it as the recommended level, compare only adjacent values, or fail to see a less prominent high-deductible alternative. Price sensitivity estimated on the full nominal menu will then conflate a conditional response to prices with a response to the probability that prices are ever compared. For insurers, the displayed menu may be an endogenous element of contract design; for researchers, its exact historical version is therefore a potentially indispensable source document.

Health insurance and Medicare Part D

Health-plan selection introduces contingent benefits that are difficult to compare using premiums and deductibles alone. Provider networks, plan formularies, cost sharing across services, caps, and expected treatment needs may differ. Abaluck and Gruber (2011), examining Medicare Part D plan choices, document systematic patterns of choice inconsistent with their full-information benchmark, including disproportionate weight on premiums relative to expected out-of-pocket costs. Those observed patterns motivate research on informational frictions; they do not identify latent consideration probabilities on their own.

Handel and Kolstad (2015) improve the measurement problem through linked administrative and survey data, estimating the role of information frictions and perceived hassle costs alongside risk preferences. Their study shows why measuring what consumers know matters when inferring risk aversion from insurance contracts. If the econometrician models all unexplained selection as taste, limited awareness or misunderstanding may be misclassified as a strongly held preference. The interpretation of an insurance choice should therefore use information measurements wherever credible observations are available.

Handel's (2013) analysis of adverse selection and inertia demonstrates that increasing active switching can have market-level consequences under risk-based sorting. In the particular employer setting studied, reducing inertia could worsen adverse selection sufficiently to harm aggregate welfare. This is an important warning against evaluating a choice intervention only by whether the consumers who switch obtain a preferred plan; insurance markets are coupled through premiums and costs. It is not a prediction that reducing inertia always harms insurance buyers.

Abaluck and Adams-Prassl (2021) study Medicare Part D using asymmetric demand responses to identify consideration probabilities under structured models. The paper reports an estimated average of 1.97 considered plans each year in its relevant specification, while only about 15% of individuals actively consider a plan other than their default; conditional on active search, the estimated considered set is substantially larger. Those quantities are different conditional objects and cannot be equated. They arise from a calibrated structural framework for U.S. Medicare Part D, not a census in which every plan inspected by each consumer was observed.

Their “smart default” analysis illustrates the welfare importance of targeting: assigning every consumer to a lower-cost plan provides little average welfare gain under the preferred model, whereas a policy focused on consumers with sufficiently large potential savings yields substantial gains. The precise $300 threshold they evaluate belongs to their Medicare Part D counterfactual and should not be generalized mechanically to auto, property, or private health insurance. The broader analytical lesson is that who receives a nudge, what becomes salient, and the cost of adaptation jointly determine welfare.

Field interventions, comprehension, and the presentation of options

Information access can increase effective consideration, yet better comparative information only influences choice when consumers understand and use it. Kling, Mullainathan, Shafir, Vermeulen, and Wrobel (2012) study comparison friction in Medicare drug plans and demonstrate why a usable individualized comparison differs from nominal access to a large listing. Their experimental evidence supports designing assistance around the actual comparison task, rather than assuming that providing a document makes every listed alternative operative in decision-making.

Dellaert, Johnson, Duncan, and Baker (2024) investigate the interaction of ordering and partitioning in online health insurance interfaces. Their choice architecture experiments indicate that dividing a menu can direct attention toward good alternatives when a reliable ranking places them prominently, while harmful ranking can make partitioning counterproductive. The effect is joint: the accuracy of the recommendation model conditions the welfare consequences of the attention filter. An interface that reduces search effort may also conceal superior alternatives if its personalization is faulty or biased.

Quiroga Gutierrez and Boes (2024) provide experimental evidence on information provision in health insurance choices, further illustrating the importance of designing and assessing specific decision support rather than equating information quantity with decision quality. Such experiments have value in isolating the effect of presented information, but external validity to actual insurance renewal, broker advice, and long-term claims consequences requires separate evidence. The investigator should not substitute a hypothetical one-session comprehension measure for the long-term welfare of a real contract without articulating that bridge.

Comparative implications and transportability

The studies above overwhelmingly concern U.S. automobile, employer health, Medicare prescription-drug, and property insurance environments. Their methods can be applied internationally, but estimated set sizes, search costs, welfare effects, and institutional responses depend on national insurance law, public subsidies, distribution systems, contract standardization, consumer protection rules, and digital market infrastructure. A result for Medicare Part D should be labeled specifically as such; public mandatory coverage or a national comparison website in another country may alter both the true menu and the probability of active search.

Meaningful international extension should document the jurisdiction and year of the applicable rules, the actual menu facing each population, the function of brokers and agents, whether default renewal is lawful and practiced, and whether personalized quotes can be generated without substantial costs. Insurance contracts are not merely retail labels. Regulatory or market rules can create real constraints that look like limited attention when omitted from a dataset.

Identification and Structural Estimation

What is and is not identified from purchases alone

Suppose only the final policy j is observed and there is a large feasible menu. The researcher seeks both a vector of taste parameters θ and the probabilities q that alternatives enter consideration. Without additional restrictions, it is possible to adjust preference parameters and consideration probabilities in offsetting ways while maintaining the same purchase shares. The conditional selection equation contains a latent mixture, and mixture components can be observationally equivalent. It follows that a good statistical fit cannot establish a particular psychological mechanism by itself.

For example, a low observed purchase share for a high-deductible policy can result from disutility for residual loss or near-zero consideration probability. Under a simplified case with deterministic conditional preference order, a contract that is never considered has zero purchase probability regardless of its payoff; a preferred contract that is commonly considered may still be selected rarely if an even more preferred one is usually present. Identification therefore requires more than observing the ranking of selected products. Model assumptions about independence, mandatory defaults, exogenous price variation, and stability of attention directly affect conclusions.

Manzini and Mariotti (2014) derive menu-based stochastic identification results for their random-attention representation. Masatlioglu, Nakajima, and Ozbay (2012) study preference inference with revealed inattention. These results should be viewed as formal conditional theorems in their models. Violations of the assumed menu independence or stability can change which restrictions hold. Applied insurance data often contain endogenously selected menus and unobserved eligibility constraints, so testing the relation between theoretical objects and actual quote behavior is central.

Identification through asymmetric demand responses

Abaluck and Adams-Prassl's (2021) identification from asymmetric responses provides a particularly relevant strategy. In the considered classes of discrete-choice models, full consideration implies a symmetry property in cross-demand responses under standard maintained regularity conditions. If a rival's improved characteristics change the probability of its entering consideration, the effect on other alternatives may become asymmetric. The authors derive constructive restrictions that identify attention and preference components from such departures, validate the approach experimentally where considered sets can be observed, and apply it to Medicare Part D.

The empirical opportunity is valuable because data on exact consideration histories are often absent. Nonetheless, every claim of identification must state the functional conditions, menu and price variation, and types of attention process to which the theorem applies. Asymmetry in observed demand can also arise from misspecified nonprice attributes, dynamic selection, correlation in unobserved utilities, price endogeneity, or failures of the structural restrictions. Robustness analysis must address these alternatives rather than presenting asymmetry as a universal fingerprint of psychological inattention.

Identification under risky outcomes

Barseghyan, Molinari, and Thirkettle (2021) derive sufficient identification conditions for insurance choices under risk with heterogeneous consideration and risk aversion. Their contribution directly confronts the problem of simultaneously learning the distribution of risk preferences and the probabilities of comparing contracts. It is analytically distinct from assuming a simple exogenous attention filter. Particularly relevant are the dependence between observed price/contract characteristics and consideration, and whether preferences and consideration are correlated through latent heterogeneity.

An applied researcher should evaluate the actual support of premiums, loss probabilities, and observed alternatives. Some insurers price in regulated tiers, leaving little independent variation in premium differences; some alternatives are available only to high-risk policyholders; underwriting classifications may jointly determine price and feasibility. Valid point identification in one study cannot be transplanted to a second dataset that lacks the necessary variation. The scientifically appropriate outcome may be partial identification or a set of sensitivity bounds.

A practical audit sequence for insurance datasets

A defensible empirical investigation starts with the feasible menu as an observed or reconstructed object. Identify which insurer, deductible, liability limits, and riders each policyholder could actually have purchased at the relevant date. Where exact menus are unavailable, retain documented uncertainty rather than labeling all absent purchases as ignored alternatives. Distinguish a quote absent because of ineligibility from a quote not requested, and mark defaults in the purchase record.

The second task is measurement of information exposure: pricing page position, broker presentation, sequence of quote requests, click logs, comparison tables, mailing records, and app sessions. The analyst should assess selection into these records. Digital traces may miss offline advice, and a broker-managed purchase can record only the broker's activity. Survey measures should identify awareness, active comparison, and stated shortlist separately, preferably before the final selection. The eye-tracking comparison by Edenbrandt et al. (2022) illustrates that different attention measurements are not equivalent.

The third task is model construction with explicit rival specifications: full consideration; default-specific attention; alternative-specific consideration; correlated latent types; and, where feasible, endogenous search. Estimation should compare predictive fit on held-out choices and economically interpretable cross-price responses, while keeping attention and utility mechanisms conceptually distinct. A flexible attention model can absorb arbitrary discrepancies if it has no restrictions, so superior in-sample fit alone is an inadequate argument for mechanism identification.

The fourth task is design-based identification where feasible. Randomized ordering, verified price notices, targeted default changes, or disclosure interventions can shift consideration without intentionally changing the contract's intrinsic coverage. However, even an interface intervention may alter perceived endorsement, trust, and valuation, violating an exclusion restriction that assumes attention is its only channel. The researcher should directly measure comprehension and beliefs and use negative controls or mechanism-specific outcomes when possible.

The fifth task is validation. Compare model-implied considered-set sizes with quote logs or prospectively asked shortlists, while acknowledging measurement error. Evaluate whether estimated risk preferences predict behavior in different coverage contexts. Test whether predicted responses to premium changes materialize in later data. Check whether the attention mechanism generalizes across digital and broker-mediated channels. Document nonresponse, missing not-at-random browsing histories, and the consequences of excluding consumers who ultimately purchase no policy.

The sixth task is normative. Keep the objective used to evaluate choices explicit: expected utility under calibrated beliefs, consumer-experienced utility, financial expenditure, insurer costs, or social surplus with adverse selection. Without this definition, a welfare ranking of two consideration regimes can be driven by an unannounced interpretation of what consumers truly value.

A transparent sensitivity exercise

Consider an illustrative menu of three deductible contracts, A, B, and C. Suppose a researcher observes that A is selected by 60%, B by 30%, and C by 10% of a sample. Under full consideration, these shares could be fitted through a distribution of utility differences. Under limited consideration, they could also arise if many consumers have B or C as a latent favorite but rarely see it. Observed shares alone supply no unique allocation of the 60/30/10 pattern across preferences and attention.

A structured sensitivity analysis can vary the probability that every feasible contract enters consideration, the dependence of attention on a default, and the correlation between attention and risk preferences. One may then report a range of willingness-to-pay or risk-aversion estimates rather than a single apparently precise number. The calculations should be based on a specified sample, offered prices, and validated choice model. The percentages here are hypothetical and carry no empirical estimate; their function is to expose why observed purchase shares are not a direct survey of unconstrained preferences.

Methodological reliability and reproducibility

Research-grade evidence requires a record of the precise source version. A publisher's abstract may confirm a finding's existence and central claim without allowing verification of a robustness table, confidence interval, or sample exclusion. A working-paper date must be retained rather than silently converted to a journal publication date. Replication data, when provided, should be distinguished from verified successful independent replication. A proposed robustness procedure remains a proposed procedure until actually run.

For this monograph, bibliographic identity and the central source claims were checked against publisher records, institutional research descriptions, and accessible article material. The 2026 paper was inspected in its March 23 working-paper version, but its underlying microdata and estimation code were not independently reproduced. No statistical model was fitted, confidence interval recalculated, or new causal estimate generated here. These are substantive scope limits relevant to the strength of the dossier's conclusions.

Welfare and Generalization Limits

The relevant welfare counterfactual

Insurance welfare is sensitive to which counterfactual is being evaluated. Providing a better-priced alternative, increasing its probability of being considered, forcing a switch, helping the household understand its costs, and redesigning the risk pool are distinct interventions. They may alter different objects in the empirical model. A correct welfare calculation therefore states whether it changes the feasible menu, attention probabilities, conditional preferences, subjective beliefs, premiums, insurer costs, or equilibrium sorting.

Let C_i be the consideration set generated under status quo policy z_0 and C_i' the set generated under intervention z_1. A schematic comparison of expected consumer welfare can be expressed as

ΔWconsumer=E[Ui(ji(z1),z1)−Ui(ji(z0),z0)]−Ki(z1,z0),\Delta W^{\mathrm{consumer}}= \mathbb E\bigl[U_i(j_i(z_1),z_1)-U_i(j_i(z_0),z_0)\bigr]-K_i(z_1,z_0),

where K_i represents relevant adaptation, information, or switching costs and utilities are measured under an explicitly stated welfare standard. This is an accounting representation of a counterfactual, not a novel social-welfare formula. For market-level analysis, the calculation may also require changes in insurer expected costs, premiums, risk composition, subsidies, and externalities. Merely expanding considered sets does not establish the sign of the full change.

Preference welfare and decision-process welfare

If preferences conditional on full information are treated as normative, a person who selected a dominated plan because it was the only one examined may gain by receiving a better alternative. If the person values stability, continuity, privacy, or the avoidance of burdensome choice, an imposed switch may reduce utility even when expected expenditure declines. The distinction between a cost the analyst wishes to eliminate and a genuine valued characteristic must be defended through evidence, not settled by terminology.

The human-centered insurance-choice study by Handel and Kolstad (2015) demonstrates the relevance of information and hassle costs to welfare assessment. Abaluck and Adams-Prassl's (2021) smart-default comparison further shows that estimated effects depend on which consumers are reassigned and the attention response induced. These findings support counterfactual specificity. An average effect from one enrollment regime is insufficient to validate a universal policy for all lines of insurance.

Equilibrium externalities and selection

Insurance redistributes risk through pooled contracts and actuarial pricing. When additional information or attention induces differential switching by risk type, it can change the loss composition of plans. Handel's health-insurance market analysis (2013) provides a concrete example in which reduced inertia improved certain individual decisions but exacerbated adverse selection in the studied setting. The sign of aggregate effects depends on competition, risk adjustment, subsidy structures, underwriting rules, and premium responses. A purely partial-equilibrium exercise may omit these channels.

The distinction between attention and risk preference is especially important here: a consumer may appear insensitive to an expensive plan because a rival never entered consideration, not because she has high willingness to pay. If an intervention makes the rival visible, enrollment can move suddenly, potentially concentrating claims costs in remaining plans. This is a joint demand-and-supply question. A credible equilibrium study requires insurer cost data, a pricing or competitive model, and assumptions about plan availability after the intervention.

Market power and distribution

Limited consideration can reduce effective competitive pressure by preventing consumers from reacting to attractive rival offers. It may also increase the value of a default, a recommendation, prominent ranking, and intermediary-controlled presentation. Consequently, observed high retention need not establish deep consumer satisfaction, and a small measured cross-price elasticity need not show that insurance buyers attach low importance to prices. Search-friction evidence in U.S. automobile insurance (Honka 2014) illustrates why the search mechanism can be quantitatively consequential for estimates of customer retention.

The distributional consequences of attention interventions require specific investigation. People with lower insurance literacy, fewer digital resources, limited language access, or high opportunity costs of time may face greater comparison burdens, but this is a testable heterogeneity hypothesis rather than a constant across populations. Bhargava, Loewenstein, and Sydnor's health-plan choice evidence (2017) reports that lower-income employees were more likely to select financially dominated options in their sample. Whether that difference reflects attention, knowledge, liquidity, contract access, or several mechanisms is a separate identification question.

An ethically defensible research evaluation should measure which households benefited from salience changes, which were disadvantaged by ranking errors, and whether the intervention created unequal exposure to favorable contract terms. A recommendation system that searches the market for the consumer may reduce decision burden while simultaneously steering the consumer toward products favored by a distributor. Such intermediation changes the operative consideration set and can alter competition even if the underlying set of licensed insurers stays constant.

The design of transparent interfaces

Good interface evaluation includes the actual choices users can make, the decision-relevant information presented, the order of alternatives, the rationale for recommendations, and access to the full eligible menu. Dellaert and coauthors' ranking and partitioning study (2024) supports an important conditional design conclusion: reducing visible complexity can help when rankings are sufficiently reliable, but can harm discovery when the ranking is poor. The result favors empirical auditing of ranking quality and attention displacement rather than assuming every shorter menu is beneficial.

The same logic applies to automated agents, digital insurers, broker comparison tables, and employer enrollment platforms. A consumer can be shown five products and consider only one, or be shown two and compare both thoroughly. “Choice architecture” must be evaluated through meaningful comprehension, observed choice process, quality of selected contracts, distributional consequences, and economic cost, rather than raw click counts or the number of products displayed.

Welfare under incomplete identification

When the data cannot uniquely identify consideration, a robust report should show how welfare varies over plausible attention mechanisms and normative preference assumptions. A policy ranking that holds only when ignored options are assumed to be strictly preferred deserves a different level of confidence from one robust to broad bounds on consideration probabilities and switching costs. If ambiguity is substantial, the proper scientific result can be an identified set or a transparent statement that the sign of welfare is unresolved.

This is not an argument for abandoning structural methods. It is a reason to match the strength of a conclusion to the strength of its identification. The combination of experimental variation, administrative contract records, surveys, and careful counterfactual modeling can substantially narrow relevant uncertainty. At the same time, the effect of an intervention on actual loss experience and market equilibrium may require longitudinal data that short-run choice studies cannot supply.

Open research questions

A mature evidence program has several unresolved priorities. First is the measurement of household-specific feasible menus when underwriting, renewal, and personalized premiums make availability endogenous. Second is identification of correlated attention across automobile coverages, insurer brands, and supplementary products. Third is separating quote acquisition from contract comprehension and from actual inclusion in a consideration set. Fourth is measuring how intermediaries and automated recommendation systems structure attention, including whether compensated presentation changes welfare-relevant shortlists.

Further questions concern dynamic learning after claims; the effects of changing legal minimum limits; attention to exclusions, insurer solvency, and basis risk; the impact of repeated default renewal; the measurement of demand among households who never purchase insurance; and external validity across voluntary private insurance, mandated participation, and public insurance systems. Each question requires research designs capable of distinguishing the decision mechanism from associated preferences, beliefs, and true eligibility.

Original analytical synthesis

The literature supports an integrated reading of latent consideration in insurance without requiring an invented theoretical system. Observed purchase has three distinct empirical layers: the feasible contract opportunity; the generated subset that becomes decision-relevant; and the contingent valuation of contracts inside that subset. Unobserved attention can induce bias in estimates of risk preferences and prices; unmeasured constraints can masquerade as inattention; and interventions affecting shortlists can induce selection and premium changes. These propositions follow from the identified theoretical models and substantive empirical findings reviewed above, with their applicable conditions retained.

The most defensible general conclusion is that attention is part of the data-generating process of insurance demand. Its importance cannot be inferred from an unusual observed selection alone, and it cannot be excluded merely because a standard preference model fits purchase shares. The appropriate standard is comparative identification: articulate competing mechanisms, use observed menus and process measures when available, test valid restrictions, and report welfare under the resulting uncertainty. The analytical value of consideration sets lies in improving the integrity of these inferences, not in replacing every other determinant of insurance choice.

Conclusion

Latent consideration sets describe a missing but economically consequential stage between an individual's feasible insurance options and the policy actually selected. Consideration varies with information, contractual complexity, experience, defaults, search technology, intermediary behavior, and the properties of the insurance market. Because preferences and attention may be correlated, insurance demand cannot automatically be interpreted as direct revelation of the valuation of every feasible contract.

The theoretical record supplies formal models of evaluation costs, revealed attention, stochastic consideration, and structural identification. Empirical evidence from U.S. auto and property insurance, employer health plans, and Medicare Part D shows that search, informational frictions, limited comparison, and the operation of defaults can materially affect observed choices and estimated welfare. The 2026 research on liability limits and deductibles is an important contemporary extension connecting limited consideration to choices involving very different stakes, while remaining a dated working-paper estimate subject to its data and specifications. The monograph has identified no defensible basis for a claim that a single mechanism or parameter is universal across these settings.

Five scientific conclusions follow. First, the observed purchase must be analyzed against the contract menu actually feasible for the household. Second, awareness, mere exposure, shortlist inclusion, and terminal choice require separate definitions and, where feasible, separate measurements. Third, identification of attention and risk preferences depends on explicit economic restrictions and sufficient variation; purchase shares alone generally do not settle their allocation. Fourth, policy and platform interventions must be assessed for their effects on what becomes considered, on the quality of comparison, and on insurer-market equilibrium. Fifth, welfare conclusions should follow the empirically supported scope of each estimated mechanism and disclose uncertainty whenever the considered set remains incompletely identified.

For Insurance by Bogdanova, the research object is thus established as a specific scientific problem in behavioral insurance and insurance econometrics. Its solution is cumulative: careful source criticism, credible measurement of options and attention, identified structural or experimental methods, and explicit welfare counterfactuals. The original contribution of this dossier is the comparative analytical synthesis and source-grounded clarification of these requirements. No independent dataset was analyzed and no new named theory, model, doctrine, or principle is claimed.

References

  • Abaluck, Jason, and Abi Adams-Prassl. 2021. “What Do Consumers Consider Before They Choose? Identification from Asymmetric Demand Responses.” The Quarterly Journal of Economics 136 (3): 1611–1663. https://doi.org/10.1093/qje/qjab008.

  • Abaluck, Jason, and Jonathan Gruber. 2011. “Choice Inconsistencies among the Elderly: Evidence from Plan Choice in the Medicare Part D Program.” American Economic Review 101 (4): 1180–1210. https://doi.org/10.1257/aer.101.4.1180.

  • Barseghyan, Levon, Isaac Cohen, and Yujie Feng. 2026. “Insuring Risks Large and Small: The Role of Preferences and Limited Consideration.” Working paper, March 23, 2026. Cornell University author-hosted version; Johns Hopkins School of Government and Policy research account. https://barseghyan.economics.cornell.edu/docs/barseghyan_et_al_2026.pdf. Research summary: https://sgp.jhu.edu/research/insuring-risks-large-and-small-the-role-of-preferences-and-limited-consideration/.

  • Barseghyan, Levon, Francesca Molinari, Ted O'Donoghue, and Joshua C. Teitelbaum. 2013. “The Nature of Risk Preferences: Evidence from Insurance Choices.” American Economic Review 103 (6): 2499–2529. https://doi.org/10.1257/aer.103.6.2499.

  • Barseghyan, Levon, Francesca Molinari, and Matthew Thirkettle. 2021. “Discrete Choice under Risk with Limited Consideration.” American Economic Review 111 (6): 1972–2006. https://doi.org/10.1257/aer.20190253.

  • Barseghyan, Levon, Jeffrey Prince, and Joshua C. Teitelbaum. 2011. “Are Risk Preferences Stable across Contexts? Evidence from Insurance Data.” American Economic Review 101 (2): 591–631. https://doi.org/10.1257/aer.101.2.591.

  • Bhargava, Saurabh, George Loewenstein, and Justin Sydnor. 2017. “Choose to Lose: Health Plan Choices from a Menu with Dominated Option.” The Quarterly Journal of Economics 132 (3): 1319–1372. https://doi.org/10.1093/qje/qjx011.

  • Dellaert, Benedict G. C., Eric J. Johnson, Shannon Duncan, and Tom Baker. 2024. “Choice Architecture for Healthier Insurance Decisions: Ordering and Partitioning Together Can Improve Consumer Choice.” Journal of Marketing 88 (1): 15–30. https://doi.org/10.1177/00222429221119086.

  • Edenbrandt, Anna Kristina, Carl-Johan Lagerkvist, Malte Lüken, and Jacob L. Orquin. 2022. “Seen but Not Considered? Awareness and Consideration in Choice Analysis.” Journal of Choice Modelling 45: 100375. https://doi.org/10.1016/j.jocm.2022.100375.

  • Handel, Benjamin R. 2013. “Adverse Selection and Inertia in Health Insurance Markets: When Nudging Hurts.” American Economic Review 103 (7): 2643–2682. https://doi.org/10.1257/aer.103.7.2643.

  • Handel, Benjamin R., and Jonathan T. Kolstad. 2015. “Health Insurance for ‘Humans’: Information Frictions, Plan Choice, and Consumer Welfare.” American Economic Review 105 (8): 2449–2500. https://doi.org/10.1257/aer.20131126.

  • Hauser, John R., and Birger Wernerfelt. 1990. “An Evaluation Cost Model of Consideration Sets.” Journal of Consumer Research 16 (4): 393–408. https://doi.org/10.1086/209225.

  • Honka, Elisabeth. 2014. “Quantifying Search and Switching Costs in the US Auto Insurance Industry.” RAND Journal of Economics 45 (4): 847–884. https://doi.org/10.1111/1756-2171.12073.

  • Honka, Elisabeth, and Pradeep Chintagunta. 2017. “Simultaneous or Sequential? Search Strategies in the U.S. Auto Insurance Industry.” Marketing Science 36 (1): 21–42. https://doi.org/10.1287/mksc.2016.0995.

  • Honka, Elisabeth, Ali Hortaçsu, and Matthijs Wildenbeest. 2019. “Empirical Search and Consideration Sets.” In Handbook of the Economics of Marketing, vol. 1, 193–257. Elsevier. https://doi.org/10.1016/bs.hem.2019.05.002.

  • Kling, Jeffrey R., Sendhil Mullainathan, Eldar Shafir, Lee C. Vermeulen, and Marian V. Wrobel. 2012. “Comparison Friction: Experimental Evidence from Medicare Drug Plans.” The Quarterly Journal of Economics 127 (1): 199–235. https://doi.org/10.1093/qje/qjr055.

  • Lleras, Juan Sebastián, Yusufcan Masatlioglu, Daisuke Nakajima, and Erkut Y. Ozbay. 2017. “When More Is Less: Limited Consideration.” Journal of Economic Theory 170: 70–85. https://doi.org/10.1016/j.jet.2017.04.004.

  • Manski, Charles F. 1977. “The Structure of Random Utility Models.” Theory and Decision 8: 229–254. https://doi.org/10.1007/BF00133443.

  • Manzini, Paola, and Marco Mariotti. 2014. “Stochastic Choice and Consideration Sets.” Econometrica 82 (3): 1153–1176. https://doi.org/10.3982/ECTA10575.

  • Masatlioglu, Yusufcan, Daisuke Nakajima, and Erkut Y. Ozbay. 2012. “Revealed Attention.” American Economic Review 102 (5): 2183–2205. https://doi.org/10.1257/aer.102.5.2183.

  • Quiroga Gutierrez, Ana Cecilia, and Stefan Boes. 2024. “Bridging the Gap: Experimental Evidence on Information Provision and Health Insurance Choices.” Health Economics 33 (6): 1368–1386. https://doi.org/10.1002/hec.4820.