Maikel Yelandi Leyva Vázquez
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Maikel Yelandi Leyva Vázquez has a total of 130 co-authors. Here are the top co-authors by number of shared publications:
- Florentin Smarandache
University of New Mexico, Mathematics, Physics and Natural Science Division, Gallup, NM, USA, USA
20 shared publications - Jesús Rafael Hechavarría Hernández
Research Department, Universidad Ecotec, Samborondón 092302, Ecuador
8 shared publications Florentin Smarandache
6 shared publications- Alexis Matheu Pérez
Centro de Investigación Institucional, Universidad Bernardo O’Higgins, Santiago 8370993, Chile
5 shared publications Jesús Rafael Hechavarría Hernández
University of Guayaquil
5 shared publicationsLorenzo Cevallos-Torres
4 shared publications
Latest Publications
- Open Access
- 13 November 2026
- Published by MDPI AG in Preprints.org
Conflict or Ignorance in the Rashomon Set? Certified Typed Verdicts for Variable Importance
When many models are almost equally accurate, they can disagree about which variables matter. Reports of this disagreement, as a range or a distribution of importance values, rarely say whether it would disappear with more data or would remain with unlimited data, and the two situations call for different actions. We compare the range of a variable’s importance over the population ε-Rashomon set, read as a partially identified quantity, with an importance threshold, and return for each variable one of four verdicts: IMPORTANT and IRRELEVANT (every ε-good model is on the same side of the threshold), STRUCTURAL (ε-good models lie on both sides, a property of the population, the model class and the tolerance that does not change with the sample size) and IGNORANCE (an abstention, with a projection of the sample size that would settle it). The probability of certifying at least one wrong verdict is controlled at level α asymptotically. For linear regression the verdicts have a closed form, and, for coefficients on fixed (population) scales under a correctly specified homoskedastic model, collinearity inflates a coefficient’s identified half-width and its standard deviation by the same factor. For finite model classes, such as all depth-two trees on a split grid, we build outer and inner confidence sets for the Rashomon set from loss differences to a reference fixed on a held-out split, a paired permutation estimator with independent units, and a studentized multiplier bootstrap. In simulations, including designs with an importance endpoint within 0.005 of the threshold or a model about 0.001 from the tolerance boundary, one wrong verdict was certified in more than 2,400 runs, and the empirical family-wise error stayed below α in every setting; bounding raw losses instead of loss differences lowered the share of replicates certifying STRUCTURAL at n = 20,000 from 0.73–0.83 to 0.33–0.67, and adapted general concentration bounds of Fisher et al. (2019), with an empirical loss bound, certified one linear verdict in six at n = 100,000. In a diabetes-screening illustration (BRFSS 2015), the best single tree assigns zero importance to heart disease and walking difficulty, while the procedure certifies both as STRUCTURAL at a one-point error tolerance in two of three sample splits, and reports IGNORANCE, never a contradictory verdict, in the third. Code is released as the Python package typedvi.
- 13 November 2026
- Published by MDPI AG in Preprints.org
Open AccessConflict or Ignorance in the Rashomon Set? Certified Typed Verdicts for Variable Importance
When many models are almost equally accurate, they can disagree about which variables matter. Reports of this disagreement, as a range or a distribution of importance values, rarely say whether it would disappear with more data or would remain with unlimited data, and the two situations call for different actions. We compare the range of a variable’s importance over the population ε-Rashomon set, read as a partially identified quantity, with an importance threshold, and return for each variable one of four verdicts: IMPORTANT and IRRELEVANT (every ε-good model is on the same side of the threshold), STRUCTURAL (ε-good models lie on both sides, a property of the population, the model class and the tolerance that does not change with the sample size) and IGNORANCE (an abstention, with a projection of the sample size that would settle it). The probability of certifying at least one wrong verdict is controlled at level α asymptotically. For linear regression the verdicts have a closed form, and, for coefficients on fixed (population) scales under a correctly specified homoskedastic model, collinearity inflates a coefficient’s identified half-width and its standard deviation by the same factor. For finite model classes, such as all depth-two trees on a split grid, we build outer and inner confidence sets for the Rashomon set from loss differences to a reference fixed on a held-out split, a paired permutation estimator with independent units, and a studentized multiplier bootstrap. In simulations, including designs with an importance endpoint within 0.005 of the threshold or a model about 0.001 from the tolerance boundary, one wrong verdict was certified in more than 2,400 runs, and the empirical family-wise error stayed below α in every setting; bounding raw losses instead of loss differences lowered the share of replicates certifying STRUCTURAL at n = 20,000 from 0.73–0.83 to 0.33–0.67, and adapted general concentration bounds of Fisher et al. (2019), with an empirical loss bound, certified one linear verdict in six at n = 100,000. In a diabetes-screening illustration (BRFSS 2015), the best single tree assigns zero importance to heart disease and walking difficulty, while the procedure certifies both as STRUCTURAL at a one-point error tolerance in two of three sample splits, and reports IGNORANCE, never a contradictory verdict, in the third. Code is released as the Python package typedvi.
- Open Access
- 29 October 2026
- Published by Universidad Bernardo O'Higgins in Autoctonia Revista de Ciencias Sociales e Historia
- Vol. 10 (Especial)
Modèle de gestion et innovation institutionnelle comme déterminants de la performance universitaire : le cas de l’Université Bernardo O’Higgins (2008–2026)
Cet article a pour objectif d’analyser l’articulation entre le modèle de gestion universitaire et l’innovation institutionnelle comme facteur explicatif de la performance universitaire dans un contexte périphérique, à partir du cas de l’Université Bernardo O’Higgins entre 2008 et 2026. La méthode repose sur une étude de cas unique à caractère longitudinal, fondée sur l’analyse documentaire, des données quantitatives descriptives et une classification interprétative des innovations selon le Manuel d’Oslo. Les résultats montrent que la combinaison entre gouvernance stratégique, discipline financière, innovation institutionnelle et suivi par indicateurs s’associe à des améliorations soutenues de la dotation académique, de la production scientifique, de la captation de financements externes, de la rétention en première année, de l’accréditation institutionnelle et de la soutenabilité financière. L’analyse comparative met également en évidence une efficacité relative dans la production de connaissances et une forte couverture de gratuité étudiante. En conclusion, le modèle de gestion et l’innovation institutionnelle configurent un système cohérent de capacités dynamiques, capable de médiatiser les restrictions structurelles et les résultats institutionnels.
- 29 October 2026
- Published by Universidad Bernardo O'Higgins in Autoctonia Revista de Ciencias Sociales e Historia
- Vol. 10 (Especial)
Open AccessModèle de gestion et innovation institutionnelle comme déterminants de la performance universitaire : le cas de l’Université Bernardo O’Higgins (2008–2026)
Cet article a pour objectif d’analyser l’articulation entre le modèle de gestion universitaire et l’innovation institutionnelle comme facteur explicatif de la performance universitaire dans un contexte périphérique, à partir du cas de l’Université Bernardo O’Higgins entre 2008 et 2026. La méthode repose sur une étude de cas unique à caractère longitudinal, fondée sur l’analyse documentaire, des données quantitatives descriptives et une classification interprétative des innovations selon le Manuel d’Oslo. Les résultats montrent que la combinaison entre gouvernance stratégique, discipline financière, innovation institutionnelle et suivi par indicateurs s’associe à des améliorations soutenues de la dotation académique, de la production scientifique, de la captation de financements externes, de la rétention en première année, de l’accréditation institutionnelle et de la soutenabilité financière. L’analyse comparative met également en évidence une efficacité relative dans la production de connaissances et une forte couverture de gratuité étudiante. En conclusion, le modèle de gestion et l’innovation institutionnelle configurent un système cohérent de capacités dynamiques, capable de médiatiser les restrictions structurelles et les résultats institutionnels.
- Open Access
- 29 October 2026
- Published by FapUNIFESP (SciELO) in SciELO Preprints
Provenance-Aware PCR6 for Conflicting Evidence in Retrieval-Augmented Generation
Retrieval-augmented generation (RAG) must often combine passages that are ambiguous, redundant, or mutually contradictory. This study asks whether proportional conflict redistribution from the Dezert-Smarandache theory (DSmT) provides a measurable advantage over Dempster-Shafer (DS) combination when both rules receive the same evidence and the same control for source dependence. Eight fusion methods were evaluated on 500 RAMDocs (Retrieval with Ambiguity and Misinformation in Documents) questions in three conditions. Under duplicated misinformation, provenance-aware PCR6 (Proportional Conflict Redistribution rule #6) reached 80.6% exact-set accuracy and 92.31% set F1, against 78.2% and 90.98% for provenance-matched DS; the exact-set difference was 2.4 percentage points (95% bootstrap interval [1.2, 3.8]) and the F1 difference was 1.33 percentage points (95% bootstrap interval [0.60, 2.33]). With NLI (natural language inference)-derived masses, PCR6 and DS selected identical answer sets, but PCR6 reduced Brier error. A preregistered 1,000-item multi-source experiment using singletons, proper unions, and total ignorance was followed by external validation on 428 RAMDocs items. In controlled partial-ignorance conflict, PCR6+ reduced BetP (betting probability transformation) Brier error from 0.3185 for PCR6 to 0.2733 and remained exactly neutral to an added vacuous BBA (basic belief assignment), although the DS rule achieved the lowest BetP Brier error of the three rules in that same controlled evaluation (0.1878). On the prespecified 260-item RAMDocs applicability subset, ordinary PCR6 had lower BetP Brier error than PCR6+ (0.2354 versus 0.2629), while both achieved set F1 = 0.7718. Exploratory decision sensitivity showed that maximum belief gave PCR6+ its best hard-decision result (exact set 0.5077; set F1 0.7859), whereas DSmP (Dezert-Smarandache probability transformation) sharpened rankings but worsened the RAMDocs Brier score. The evidence therefore supports a bounded claim: PCR6+ corrects a structural multi-source weakness of PCR6 under broad or vacuous focal evidence, but the best fusion and decision rules depend on the downstream loss and the mapping from text to BBAs.
- 29 October 2026
- Published by FapUNIFESP (SciELO) in SciELO Preprints
Open AccessProvenance-Aware PCR6 for Conflicting Evidence in Retrieval-Augmented Generation
Retrieval-augmented generation (RAG) must often combine passages that are ambiguous, redundant, or mutually contradictory. This study asks whether proportional conflict redistribution from the Dezert-Smarandache theory (DSmT) provides a measurable advantage over Dempster-Shafer (DS) combination when both rules receive the same evidence and the same control for source dependence. Eight fusion methods were evaluated on 500 RAMDocs (Retrieval with Ambiguity and Misinformation in Documents) questions in three conditions. Under duplicated misinformation, provenance-aware PCR6 (Proportional Conflict Redistribution rule #6) reached 80.6% exact-set accuracy and 92.31% set F1, against 78.2% and 90.98% for provenance-matched DS; the exact-set difference was 2.4 percentage points (95% bootstrap interval [1.2, 3.8]) and the F1 difference was 1.33 percentage points (95% bootstrap interval [0.60, 2.33]). With NLI (natural language inference)-derived masses, PCR6 and DS selected identical answer sets, but PCR6 reduced Brier error. A preregistered 1,000-item multi-source experiment using singletons, proper unions, and total ignorance was followed by external validation on 428 RAMDocs items. In controlled partial-ignorance conflict, PCR6+ reduced BetP (betting probability transformation) Brier error from 0.3185 for PCR6 to 0.2733 and remained exactly neutral to an added vacuous BBA (basic belief assignment), although the DS rule achieved the lowest BetP Brier error of the three rules in that same controlled evaluation (0.1878). On the prespecified 260-item RAMDocs applicability subset, ordinary PCR6 had lower BetP Brier error than PCR6+ (0.2354 versus 0.2629), while both achieved set F1 = 0.7718. Exploratory decision sensitivity showed that maximum belief gave PCR6+ its best hard-decision result (exact set 0.5077; set F1 0.7859), whereas DSmP (Dezert-Smarandache probability transformation) sharpened rankings but worsened the RAMDocs Brier score. The evidence therefore supports a bounded claim: PCR6+ corrects a structural multi-source weakness of PCR6 under broad or vacuous focal evidence, but the best fusion and decision rules depend on the downstream loss and the mapping from text to BBAs.
- Open Access
- 29 October 2026
- Published by MDPI AG in Preprints.org
Why Escalate? Typed Escalation for LLM-Judge Panels with Per-Type Conformal Guarantees
LLM judges are increasingly deployed with a reject option. Recent methods certify the error rate of the verdicts these judges accept, but items that are escalated carry no reason, even though conflicting evidence, missing evidence and a false premise call for different remedies. We ask whether a panel of LLM judges can say why it escalates, with a separate statistical guarantee for each reason. Learn-then-Test acceptance is kept, and each escalated item receives a set of escalation types calibrated by per-type split conformal prediction. For each type separately, an escalated item that carries the type includes it with probability at least 0.9 (marginally, for the constructed labels, assuming exchangeable items); for the whole set, only the weaker union bound holds. Typed items were built on SQuAD 2.0 without annotators (conflict, scarcity, both, false premise), and the panel had three judges from different model families (4,801 items). In the preregistered analysis (protocol amended after a pilot that selected the scores), mean per-type coverage over 200 article-level re-partitions was 93.8% (conflict), 91.2% (scarcity) and 94.0% (false premise), and accepted verdicts had a selective risk of 7.5%. Because items are clustered in articles, we added an exploratory article-level variant that is exactly valid when articles are exchangeable. Per-type coverage stayed above 0.9, but sets grew from 1.74 to 1.91 types, and exact article-level acceptance could not be certified with 74 calibration articles. Typed sets contained every true reason for 91.2% of typed escalations; that is higher coverage of reasons, not a measured operational gain. In further exploratory analyses, counting accepted items as misses, end-to-end detection was 42.1%, 69.2% and 35.9%, and on an external false-premise set it was 4.0% and stayed low after recalibration. A supervised score on the same panel outputs gave smaller sets. In an executed retrieval pilot, retrieval resolved most escalations that truly lacked evidence but few others; because the flags put scarcity in most sets, typed routing saved little over retrieving for everything. An automated check by two LLMs flagged 12.9% to 28.5% of conflict labels as doubtful; a human audit is pending. The glut of the existential triple (Dempster’s conflict) was far worse at detecting conflict than asking the judges directly.
- 29 October 2026
- Published by MDPI AG in Preprints.org
Open AccessWhy Escalate? Typed Escalation for LLM-Judge Panels with Per-Type Conformal Guarantees
LLM judges are increasingly deployed with a reject option. Recent methods certify the error rate of the verdicts these judges accept, but items that are escalated carry no reason, even though conflicting evidence, missing evidence and a false premise call for different remedies. We ask whether a panel of LLM judges can say why it escalates, with a separate statistical guarantee for each reason. Learn-then-Test acceptance is kept, and each escalated item receives a set of escalation types calibrated by per-type split conformal prediction. For each type separately, an escalated item that carries the type includes it with probability at least 0.9 (marginally, for the constructed labels, assuming exchangeable items); for the whole set, only the weaker union bound holds. Typed items were built on SQuAD 2.0 without annotators (conflict, scarcity, both, false premise), and the panel had three judges from different model families (4,801 items). In the preregistered analysis (protocol amended after a pilot that selected the scores), mean per-type coverage over 200 article-level re-partitions was 93.8% (conflict), 91.2% (scarcity) and 94.0% (false premise), and accepted verdicts had a selective risk of 7.5%. Because items are clustered in articles, we added an exploratory article-level variant that is exactly valid when articles are exchangeable. Per-type coverage stayed above 0.9, but sets grew from 1.74 to 1.91 types, and exact article-level acceptance could not be certified with 74 calibration articles. Typed sets contained every true reason for 91.2% of typed escalations; that is higher coverage of reasons, not a measured operational gain. In further exploratory analyses, counting accepted items as misses, end-to-end detection was 42.1%, 69.2% and 35.9%, and on an external false-premise set it was 4.0% and stayed low after recalibration. A supervised score on the same panel outputs gave smaller sets. In an executed retrieval pilot, retrieval resolved most escalations that truly lacked evidence but few others; because the flags put scarcity in most sets, typed routing saved little over retrieving for everything. An automated check by two LLMs flagged 12.9% to 28.5% of conflict labels as doubtful; a human audit is pending. The glut of the existential triple (Dempster’s conflict) was far worse at detecting conflict than asking the judges directly.
- Open Access
- 20 October 2026
- Published by MDPI AG in Preprints.org
Finite GE-Algebras of Order at Most 6
Up to isomorphism there are 1, 3, 21, 479 and 40,791 GE-algebras of orders 2 to 6. Two separately written programs give the counts, checked with an SMT solver up to order 5. We give the counts for six named subclasses (transitive, commutative, left exchangeable, belligerent, antisymmetric and Hilbert; two of the columns were known) and, for every non-inclusion between them, the least order of a separating example together with the number of minimal examples. The nine patterns of properties that occur are exactly those allowed by the published inclusions, which therefore axiomatize the Boolean theory of the six properties. Finally, a GE-algebra fails the exchange identity exactly when two of its rows are different constant maps on a two-element set; hence every GE-algebra with at most four elements is left exchangeable, and the bound is sharp.
- 20 October 2026
- Published by MDPI AG in Preprints.org
Open AccessFinite GE-Algebras of Order at Most 6
Up to isomorphism there are 1, 3, 21, 479 and 40,791 GE-algebras of orders 2 to 6. Two separately written programs give the counts, checked with an SMT solver up to order 5. We give the counts for six named subclasses (transitive, commutative, left exchangeable, belligerent, antisymmetric and Hilbert; two of the columns were known) and, for every non-inclusion between them, the least order of a separating example together with the number of minimal examples. The nine patterns of properties that occur are exactly those allowed by the published inclusions, which therefore axiomatize the Boolean theory of the six properties. Finally, a GE-algebra fails the exchange identity exactly when two of its rows are different constant maps on a two-element set; hence every GE-algebra with at most four elements is left exchangeable, and the bound is sharp.
- Open Access
- 20 October 2026
- Published by MDPI AG in Preprints.org
Extension of GE‐Algebras to NeutroGE‐Algebras
GE-algebras, introduced by Bandaru, Borumand Saeid and Jun as a generalization of Hilbert algebras, are implication-type algebras (X, ∗, 1) defined by three short axioms. This paper presents them in order of increasing difficulty and extends them to neutrosophic and NeutroAlgebraic settings. We start from the smallest examples (the two-element Boolean implication, the three-element chain, the Gödel implication on [0, 1] and a five-element algebra in which the induced relation is not antisymmetric), show by an explicit counterexample that the Łukasiewicz implication on [0, 1] is not a GE-algebra, and prove that a residuated implication gives a GE-algebra exactly when its conjunction is idempotent. We then prove a transport theorem and use it to build neutrosophic GE-algebras on triples (T, I, F), refined neutrosophic GE-algebras on (T₁, …, Tₚ; I₁, …, Iᵣ; N₁, …, N_q; F₁, …, Fₛ), and coupled versions in which the indeterminacy and falsity of an implication depend on the truth values; their orders are bounded distributive lattices. Finally, in the setting of Smarandache's NeutroAlgebra, we show that GE3 is never an AntiAxiom, so every groupoid satisfying GE1 and GE2 is either a GE-algebra or a NeutroGE-algebra; we bound the share of tuples satisfying each subclass identity (sharply for commutativity and antisymmetry), and we classify the instances of GE3 in partial Cayley tables as necessarily true, indeterminate or necessarily false, which yields a neutrosophic triple (T, I, F) for every partial table.
- 20 October 2026
- Published by MDPI AG in Preprints.org
Open AccessExtension of GE‐Algebras to NeutroGE‐Algebras
GE-algebras, introduced by Bandaru, Borumand Saeid and Jun as a generalization of Hilbert algebras, are implication-type algebras (X, ∗, 1) defined by three short axioms. This paper presents them in order of increasing difficulty and extends them to neutrosophic and NeutroAlgebraic settings. We start from the smallest examples (the two-element Boolean implication, the three-element chain, the Gödel implication on [0, 1] and a five-element algebra in which the induced relation is not antisymmetric), show by an explicit counterexample that the Łukasiewicz implication on [0, 1] is not a GE-algebra, and prove that a residuated implication gives a GE-algebra exactly when its conjunction is idempotent. We then prove a transport theorem and use it to build neutrosophic GE-algebras on triples (T, I, F), refined neutrosophic GE-algebras on (T₁, …, Tₚ; I₁, …, Iᵣ; N₁, …, N_q; F₁, …, Fₛ), and coupled versions in which the indeterminacy and falsity of an implication depend on the truth values; their orders are bounded distributive lattices. Finally, in the setting of Smarandache's NeutroAlgebra, we show that GE3 is never an AntiAxiom, so every groupoid satisfying GE1 and GE2 is either a GE-algebra or a NeutroGE-algebra; we bound the share of tuples satisfying each subclass identity (sharply for commutativity and antisymmetry), and we classify the instances of GE3 in partial Cayley tables as necessarily true, indeterminate or necessarily false, which yields a neutrosophic triple (T, I, F) for every partial table.
- Open Access
- 16 October 2026
- Published by MDPI AG in Preprints.org
Four-Verdict Statistical Inference: Extending Royall’s Evidential Paradigm to Contradiction Between Sources
When several sources bear on two hypotheses, Royall’s three evidential outcomes (strong for one, strong for the other, weak) leave no room for a fourth state: some sources are strong evidence for one hypothesis and others for the other. We integrate four verdicts (supported, refuted, undetermined, contradicted) in one likelihood-based report and state which error probability of each verdict is bounded in finite samples, which vanishes only asymptotically, and when “contradicted” stops describing a real conflict between source parameters. Spurious contradiction has probability at most \( m/c \) for \( m \) sources and per-source threshold \( c \), under any dependence, for any correctly specified family. With a truth midway between two simple hypotheses the contradiction probability tends to \( 1-2^{1-m} \); a guarded rule keeps it at most \( m\Phi (-\sqrt{2\ln\,c}) \) for marginally Gaussian sources with any common mean and any dependence, and at most \( m(m-1)\Phi (-2\sqrt{\ln\,c}) \) under pairwise joint normality with non-negative covariances. For composite hypotheses an integrated-over-maximised likelihood ratio gives the bound \( m/c \) for heterogeneous parameters within one hypothesis. A preregistered simulation confirmed the bounds. In the Bacillus Calmette–Guérin (BCG) vaccine trials the verdict for “risk ratio at most 0.5” against “at least 1” is “contradicted” under normal, binomial-profile and conditional likelihoods; because these hypotheses leave an indifference zone, this does not establish that the trials’ true effects conflict.- 16 October 2026
- Published by MDPI AG in Preprints.org
Open AccessFour-Verdict Statistical Inference: Extending Royall’s Evidential Paradigm to Contradiction Between Sources
When several sources bear on two hypotheses, Royall’s three evidential outcomes (strong for one, strong for the other, weak) leave no room for a fourth state: some sources are strong evidence for one hypothesis and others for the other. We integrate four verdicts (supported, refuted, undetermined, contradicted) in one likelihood-based report and state which error probability of each verdict is bounded in finite samples, which vanishes only asymptotically, and when “contradicted” stops describing a real conflict between source parameters. Spurious contradiction has probability at most \( m/c \) for \( m \) sources and per-source threshold \( c \), under any dependence, for any correctly specified family. With a truth midway between two simple hypotheses the contradiction probability tends to \( 1-2^{1-m} \); a guarded rule keeps it at most \( m\Phi (-\sqrt{2\ln\,c}) \) for marginally Gaussian sources with any common mean and any dependence, and at most \( m(m-1)\Phi (-2\sqrt{\ln\,c}) \) under pairwise joint normality with non-negative covariances. For composite hypotheses an integrated-over-maximised likelihood ratio gives the bound \( m/c \) for heterogeneous parameters within one hypothesis. A preregistered simulation confirmed the bounds. In the Bacillus Calmette–Guérin (BCG) vaccine trials the verdict for “risk ratio at most 0.5” against “at least 1” is “contradicted” under normal, binomial-profile and conditional likelihoods; because these hypotheses leave an indifference zone, this does not establish that the trials’ true effects conflict. - Open Access
- 12 October 2026
- Published by Center for Open Science in SocArXiv
Agreement under abstention: an agreement–abstention–conflict triple and sharp bounds for Fleiss' kappa
Raters in agreement studies are often allowed to abstain, for example by answering "can't tell" or "unsure". Abstentions are usually deleted or recoded, and each treatment carries an untested assumption about why raters abstained. We show how to report Fleiss' kappa without such an assumption. Assigning the abstentions to categories changes both the observed agreement and the category totals that determine chance agreement, so the extremes of kappa require optimising the two jointly. A dynamic programme gives the exact extremes over all such completions for two or three categories, with completions that attain and thereby certify them; closed-form bounds enclose them for more categories. A budget interval and tipping points grade conclusions by how far the missing ratings must depart from a reference predictive model. The shares of agreeing, abstaining and conflicting rater pairs, (pA, pU, pC), serve as an intuitive summary. Under independent sampling of units the sample extremes converge to the population identified set, and an exploratory unit bootstrap supports a three-outcome decision: adequate, inadequate or undetermined. In simulations at 2–10% abstention, estimator errors depended on the mechanism; the budget interval contained the complete-data kappa in 97–100% of replicates under random abstention but 44–51% under value-dependent abstention, and population decisions were undetermined in 43–76% of replicates. In crowdsourced weather-sentiment labels, listwise deletion declared agreement adequate where the three-outcome decision was undetermined; in safety ratings of chatbot conversations, listwise estimates fell on both sides of the interval.
- 12 October 2026
- Published by Center for Open Science in SocArXiv
Open AccessAgreement under abstention: an agreement–abstention–conflict triple and sharp bounds for Fleiss' kappa
Raters in agreement studies are often allowed to abstain, for example by answering "can't tell" or "unsure". Abstentions are usually deleted or recoded, and each treatment carries an untested assumption about why raters abstained. We show how to report Fleiss' kappa without such an assumption. Assigning the abstentions to categories changes both the observed agreement and the category totals that determine chance agreement, so the extremes of kappa require optimising the two jointly. A dynamic programme gives the exact extremes over all such completions for two or three categories, with completions that attain and thereby certify them; closed-form bounds enclose them for more categories. A budget interval and tipping points grade conclusions by how far the missing ratings must depart from a reference predictive model. The shares of agreeing, abstaining and conflicting rater pairs, (pA, pU, pC), serve as an intuitive summary. Under independent sampling of units the sample extremes converge to the population identified set, and an exploratory unit bootstrap supports a three-outcome decision: adequate, inadequate or undetermined. In simulations at 2–10% abstention, estimator errors depended on the mechanism; the budget interval contained the complete-data kappa in 97–100% of replicates under random abstention but 44–51% under value-dependent abstention, and population decisions were undetermined in 43–76% of replicates. In crowdsourced weather-sentiment labels, listwise deletion declared agreement adequate where the three-outcome decision was undetermined; in safety ratings of chatbot conversations, listwise estimates fell on both sides of the interval.
- Open Access
- 8 October 2026
- Published by Center for Open Science in SocArXiv
What QCA Cannot See: Identification Bounds for Configurational Solutions with Missing Set Memberships
Qualitative comparative analysis (QCA) needs a membership score for every case in every set, and applied studies commonly remove cases whose scores are missing. In a truth table, removal can turn a configuration that exists but was not scored into a logical remainder available as a simplifying assumption. We propose partial-identification QCA (PI-QCA). Each missing score is represented by an interval, every completion is considered, and the analysis reports exact bounds for the consistency and coverage of any term, the sure solution (the terms present in the solution of every completion), the possible region, bracketed by an inner bound from sampled completions and an outer bound from enumeration, and the identification gap. A solution outside the outer bound is produced by no completion. By construction, the sure solution cannot contain a term absent from the complete-data solution. In 900 simulated data sets its sampled approximation also satisfied this in every replicate, whereas case deletion and mean imputation returned false terms about as often as the complete-data analysis on average but up to 22 percentage points more often under some missingness mechanisms. The price is an empty sure solution in 71–95% of replicates when a fifth of the scores were missing. In an illustration with 131 economies, deletion returns a solution that no completion produces when the outcome is kept at its recorded values. Reanalyses of two published studies find a certified empty sure solution for one outcome and, for another, published solutions attained by none of 208 sampled completions.
- 8 October 2026
- Published by Center for Open Science in SocArXiv
Open AccessWhat QCA Cannot See: Identification Bounds for Configurational Solutions with Missing Set Memberships
Qualitative comparative analysis (QCA) needs a membership score for every case in every set, and applied studies commonly remove cases whose scores are missing. In a truth table, removal can turn a configuration that exists but was not scored into a logical remainder available as a simplifying assumption. We propose partial-identification QCA (PI-QCA). Each missing score is represented by an interval, every completion is considered, and the analysis reports exact bounds for the consistency and coverage of any term, the sure solution (the terms present in the solution of every completion), the possible region, bracketed by an inner bound from sampled completions and an outer bound from enumeration, and the identification gap. A solution outside the outer bound is produced by no completion. By construction, the sure solution cannot contain a term absent from the complete-data solution. In 900 simulated data sets its sampled approximation also satisfied this in every replicate, whereas case deletion and mean imputation returned false terms about as often as the complete-data analysis on average but up to 22 percentage points more often under some missingness mechanisms. The price is an empty sure solution in 71–95% of replicates when a fifth of the scores were missing. In an illustration with 131 economies, deletion returns a solution that no completion produces when the outcome is kept at its recorded values. Reanalyses of two published studies find a certified empty sure solution for one outcome and, for another, published solutions attained by none of 208 sampled completions.
- Open Access
- 8 October 2026
- Published by MDPI AG in Preprints.org
Which Comparisons Does a University Ranking Support? Typed Indeterminacy in the QS World University Rankings: Latin America & The Caribbean 2027 Under Imprecise Weights and Data
University rankings publish positions and bands that readers may interpret as precise comparisons. We examine which comparisons the QS World University Rankings: Latin America & The Caribbean 2027 supports under imprecise weights and indicator data, and assess change using both the 2026 and 2027 editions. Nominal weights centre an epsilon-contamination credal set with an imprecise Dirichlet interpretation. Missing indicators enter as observed-range intervals, and one-decimal scores reflect rounding uncertainty. Closed-form necessary orders and sequential typing distinguish conflict, ignorance and resolution among 517 institutions. At ε=0.10, only 4.0% of 5115 near pairs (reference ranks at most 10 positions apart) have a necessary order, against 61.2% of all pairs. Conflict represents 99.3% of undecided neighbouring pairs; ignorance represents 55.6% overall. An eight-coalition factorial analysis and symmetric Shapley attribution allocate 70.1% of additional neighbouring indeterminacy to weights and 57.3% of additional overall indeterminacy to missing indicators. For 494 matched institutions, mixed-integer optimisation closes 1976 marginal rank-endpoint problems under edition-specific admissible weights. Median change-hull width falls from 365.5 positions under outer bounds to 294 under attainable endpoints; exact rational witnesses support both a rise and a fall for 493 institutions. These are model-based possibilities, rather than evidence of stability or institutional quality. The analysis combines necessary comparisons, attainable rank and change bounds, and complementary sequential and symmetric descriptions of indeterminacy. Its findings support reporting comparison precision and underlying indicator coverage alongside ranking positions.
- 8 October 2026
- Published by MDPI AG in Preprints.org
Open AccessWhich Comparisons Does a University Ranking Support? Typed Indeterminacy in the QS World University Rankings: Latin America & The Caribbean 2027 Under Imprecise Weights and Data
University rankings publish positions and bands that readers may interpret as precise comparisons. We examine which comparisons the QS World University Rankings: Latin America & The Caribbean 2027 supports under imprecise weights and indicator data, and assess change using both the 2026 and 2027 editions. Nominal weights centre an epsilon-contamination credal set with an imprecise Dirichlet interpretation. Missing indicators enter as observed-range intervals, and one-decimal scores reflect rounding uncertainty. Closed-form necessary orders and sequential typing distinguish conflict, ignorance and resolution among 517 institutions. At ε=0.10, only 4.0% of 5115 near pairs (reference ranks at most 10 positions apart) have a necessary order, against 61.2% of all pairs. Conflict represents 99.3% of undecided neighbouring pairs; ignorance represents 55.6% overall. An eight-coalition factorial analysis and symmetric Shapley attribution allocate 70.1% of additional neighbouring indeterminacy to weights and 57.3% of additional overall indeterminacy to missing indicators. For 494 matched institutions, mixed-integer optimisation closes 1976 marginal rank-endpoint problems under edition-specific admissible weights. Median change-hull width falls from 365.5 positions under outer bounds to 294 under attainable endpoints; exact rational witnesses support both a rise and a fall for 493 institutions. These are model-based possibilities, rather than evidence of stability or institutional quality. The analysis combines necessary comparisons, attainable rank and change bounds, and complementary sequential and symmetric descriptions of indeterminacy. Its findings support reporting comparison precision and underlying indicator coverage alongside ranking positions.