Independent researcher
Research
Clinical psychology
Clinical psychology
Major Depressive Disorder: Symptomatology, Diagnostic Validity, and Vulnerability-Stress Models with Implications for Prevention and Treatment
Preprint
Abstract
Major Depressive Disorder (MDD) is a leading cause of disability, defined by Major Depressive Episodes (MDEs) comprising at least five of nine symptoms over two weeks with marked distress/impairment. This review outlines symptomatology and epidemiology (U.S. lifetime 15-17%; 12-month 6-7%; female:male ≈2:1; typical onset mid-20s). This review examines the categorical vs dimensional debate: taxometric evidence favors a continuum and supports hybrid practice (categorical decisions supplemented by dimensional severity). This review also critiques the DSM-5-TR on validity (no biomarkers; heterogeneity), reliability (κ ≈ .28), thresholding, and bereavement-exclusion changes. A vulnerability-stress framework integrates psychological (hopelessness, negative schemas) and biological (5-HTTLPR × stress) diatheses with life events. Interventions include CBT (g ≈.79; relapse prevention), ACT (comparable to CBT), RO-DBT (short-term gains), exercise (moderate effects), and pharmacotherapy (SSRIs effective; SNRIs possibly superior in severe cases; ketamine rapid and sustained). Prevention reduces incidence (~22%); maintenance antidepressants roughly halve relapse risk. Personalized, integrated care is recommended.
Read the PDF ↗Work in progress
Role disability after supported psilocybin treatment for major depressive disorder
In progress
Anxiety at follow-up after immersive virtual reality relaxation versus active relaxation
In progress
Hospitality and tourism
Abstract
Persistent labor shortages and elevated voluntary turnover define the hospitality industry. High-intensity service interactions increase workload, emotional labor, and time pressure for non-management employees. The Job Demands-Resources (JD-R) model predicts engagement and retention when job resources offset job demands. Existing research has not specified which operational conditions reliably predict this balance in dynamic service work. This study evaluates whether adapted Agile philosophy serves as a heuristic framework for specifying JD-R constructs in hospitality. The study maps Agile values and principles to validated psychological constructs through Psychological State Matching: Self-Organization to Psychological Empowerment (PE), Regular Reflection to Perceived Team Learning Behavior (PTLB), and Interactions over Processes to Perceived Psychological Safety (PPS). Deviation from Simplicity maps to Hindrance Stressors (HS); deviation from Sustainable Pace maps to Quantitative Workload (QWL). This mapping is subjective and isolated to theory-building; the empirical measurement relies on validated scales. A cross-sectional survey collected data from 386 non-management hospitality employees. Covariance-based structural equation modeling with robust maximum likelihood estimation tested the model. The data supported 5 of 13 hypothesized relationships. PE dominated both pathways: it showed a large positive association with Employee Satisfaction (β = .861) and a large negative association with Burnout (β = −.814). PPS and PTLB showed large bivariate associations with outcomes but contributed no unique variance after controlling for PE; the three resources formed a correlated cluster whose effects concentrated in PE. HS predicted Burnout (β = .158). Employee Satisfaction and Burnout both predicted Turnover Intentions. The model explained 80.2% of variance in Employee Satisfaction, 81.0% in Burnout, and 53.2% in Turnover Intentions. Agile philosophy functions as a valid but imprecise heuristic: it correctly identified the dual-pathway structure but overestimated the independence of team-level constructs relative to individual empowerment.
Read the PDF ↗Hospitality and tourism
Study under peer review
Mixed-methods study using food service reviews
Under peer review. The paper is not listed yet.
Statistical theory and philosophy of science
Mathematical statistics
Common E-Values Across Deterministic Reporting Protocols
Preprint
Abstract
A deterministic reporting protocol partitions an outcome space into reported events. Requiring identical events to receive identical multipliers couples calibration across protocols. For a compact convex positive null class and positive alternative on a finite outcome space, a pair admits a common calibrated score with positive log growth in every model making both protocols informative exactly when the partitions are comparable by refinement or have no identical cell. Every remaining pair admits an obstructing positive null segment. Pairwise compatibility does not suffice for families. Two sufficient classes and a necessary condition leave two structural questions open. On at most three outcomes, one positive rule is valid under every deterministic partition and grows under each informative one; a stable four-outcome example attains the minimum for failure. Finitely many reports on a countable outcome space reduce exactly to a finite model. For randomized coarsening-at-random channels, equal scores on equal truthful subset labels admit a common growing rule on at most two outcomes, while two informative channels can be obstructed on three. Under pre-outcome joint selection of null laws and deterministic protocols, disjoint report hulls characterize existence of an event-history e-process with a uniform positive exponential growth rate under every alternative protocol policy. Pooled history scores grow under alternative protocol changes and remain valid under each fixed null law. The four-outcome certificate gives a changing-null policy under which the pooled score diverges and violates stopped and deterministic-time validity.
Read the PDF ↗Abstract
For a known countable family of stationary binary laws, I characterize when one fixed predictor’s conditional forecasts admit uniformly safe expanding horizons. Each dependent candidate copies one remote bit with a prescribed bias. For every coherent predictor that converges at each fixed horizon under each candidate, safe expansion is possible exactly when the biases tend to zero. The positive rule controls the probability of any inaccurate approval and expands almost surely under every candidate. With the specified likelihood weights, biases bℓ = ℓ^(−γ)/8, and strong merging under independence, the ideal rule’s horizon has almost-sure order n^(2γ) for 0 < γ < 1/2 and n/log n for γ ≥ 1/2. For the chronological Brier aggregate, a current-weight certificate preserves these guarantees and rates without evaluating future continuations. Its implementation uses O(n²) arithmetic operations per date and has polynomial bit cost for rationally supplied amplitudes and parameters. An explicit finite-date bound controls approval under independence. Every fixed-date-safe selector obeys an asymptotic approval-probability cap on the same scales, with sharp error level. A same-predictor comparison separates conditional approval from stationary-marginal approval.
Read the PDF ↗Philosophy of statistics
When Fit and Test Summaries Do Not Support Later Estimation
Preprint
Abstract
When a retained archive fails a reliability specification, an inquiry can change its procedures, target, tolerance, state class, weights, or resources. This paper assesses which repairs answer a declared question at a required uniform reliability. A contributory point specification has disjoint success intervals whenever live states require different answers. More generally, sufficiently frequent decisive successes, which lie in no success interval for a different answer, certify reliable decoding. This certificate is sufficient: its failure leaves open decoding that uses the known procedure’s report distribution. An elementary obstruction gives negative judgments directly. When two live states with different answers give the archive the same probability law, no decoder answers correctly with probability above one-half at both. Total-variation bounds quantify the repair and exclusion limits for unequal archive laws. In a hypothetical Gaussian drift archive that omits the fitted drift’s sign, two class restrictions are each attainable yet issue opposite answers on almost every report. The archive cannot discriminate between the paired states to justify either restriction. Unequal weights change the guarantee even when justified; they cannot recover the original uniform requirement. A coarser tolerance and a different target are attainable but do not recover that requirement. Retaining the sign supports a relaxation of a tolerance that no procedure can meet on the protocol’s class. A hypothetical pooled-count archive gives a second application. These results concern reliable answering, leave broader inquiry continuation and intermediate evidence open, and supply no empirical justification for the hypothetical aims or parameter bounds.
Read the PDF ↗Philosophy of science
Arbitrary Switches and Pool Exclusion: A Reply to Schurz
Preprint
Abstract
Schurz proposes excluding methods with arbitrary parameters from a meta-inductive comparison pool because they lack a regularity-based or causal route to predictive success. Arbitrarily scheduled exceptions, including infinitely recurring exceptions of zero density, can preserve an admitted learner’s limiting-success event under every common outcome law. This preservation refutes the unrestricted rationale without warranting exceptional forecasts. A shared bound on departures instead permits omitting these variants as redundant performance comparators at an explicit score cost. That bound does not select a warranted next action. The distinction separates three grounds for pool decisions: sustained performance, current predictive warrant, and limits on tracking.
Read the PDF ↗Abstract
Positive necessity requires a uniform margin below contour one on an assertion’s complement. Containing every contour-one parameter does not by itself provide that margin. For inclusive relative-likelihood contours on a finite observation space, I derive a positive gap from the likelihood-envelope masses. For normalized contours the gap bounds every positive necessity value from below, without compactness of the parameter frame or attainment of complementary suprema. The same envelope gives a uniform lower bound on repeated-sampling coverage of the contour-one set, called the core. Both bounds are sharp when the fixed envelope has total mass above one. The gap bounds necessity after a fixed observation, while the coverage bound concerns the sampling probability that the core contains the true parameter. Countably infinite observations permit both zero necessity of the core and zero core coverage, even with attained likelihood maxima and a summable envelope. A finite partition into blocks with globally identical ranks gives sharp grouped-envelope bounds, including for countable observations. For n binomial successes, the core is [cₙ, 1], where cₙ = [1 + n⁻¹(1 − 1/n)ⁿ⁻¹]⁻¹. Its width is asymptotic to e⁻¹/n, while its necessity converges to exp(−1 − e⁻¹). The necessity exceeds that limit for every n, whereas the envelope certificates for this observation are at most 1/(n + 1). A proof compares every rank interval and identifies the exact necessity for each n. The application clarifies an incidental endpoint statement in a recent possibilistic account of induction. An appendix records the separate domain condition for extending parameter assessments by entailment to additional unknowns.
Read the PDF ↗Applied statistics
Abstract
Knuth et al. report a mean acceleration near 550g₀ for their visual-departure calculation. Their summary table uses the most probable value near 150g₀. This note examines the meaning and reproducibility of the reported mean. Reconstructing their Figure 4D acceleration histogram gives 548.5g₀. The analysis interprets the stated Gaussian values as parameters before truncation, adds an angular bound that keeps distance positive, and assumes angle-duration independence. The resulting model has an infinite unrestricted mean because its duration density approaches a positive value at zero and acceleration depends on inverse-square duration. A finite simulation run produces a finite average almost surely; conditioning on a bounded acceleration range gives a finite mean. Applying these choices to the reported inputs and restricting acceleration to 2000g₀ gives a mean of 595.5g₀ and excludes 27.4% of the model’s probability. A separate fit to their Figure 4C timing bars gives a restricted mean of 549.9g₀, close to the reported number, without identifying their averaging procedure. The analysis distinguishes finite interpretations of the reported number and identifies the information needed to reproduce it. Duration models with identical means and variances can imply different acceleration expectations; the Bethune calculation shares the inverse-square duration dependence. The analysis leaves their modal summary and separate encounter estimates unchanged.
Read the PDF ↗