Specialty
The proposed AL-LPAL method combines central concentration with very heavy tails and showed robustness in contaminated tests, while interval coverage remained imperfect.
Scientific publication5 min read
The proposed corrections lowered simulation error, while an observational schooling application produced estimates that shifted with the controls used.
Scientific publication4 min read
A methods study reports that standard error estimates can become unreliable in small randomized trials with many covariates, even when G-computation is less sensitive to bias in some settings.
Scientific publication4 min read
A mathematical preprint classifies covariance denominators that can support exact Gaussian ratio laws and shows why other formulas lose stability or interval precision near weak identification.
Scientific publication5 min read
A preprint reports that a partially collapsed particle-learning sampler mixed well in synthetic models and showed better autocorrelation-based mixing than JAGS in one Norway hospitalization analysis, with the authors warning against broad generalization.
Scientific publication4 min read
In one 10-variable quarterly exercise, non-centred stochastic volatility scored higher on density forecasts than centred stochastic volatility, but point accuracy was nearly unchanged.
Scientific publication4 min read
A theoretical analysis finds that exact calibration does not make two-sided p-values unique. Symmetry forces several methods to agree, while numerical examples show large differences in asymmetric models.
Scientific publication4 min read
SCAN is designed to detect multiple distributional shifts in long, serially dependent time series. Its strongest reported results came on longer simulations, while ARFIMA long-memory scenarios produced less favorable and more threshold-sensitive results.
Scientific publication4 min read
A methods study found higher reported sampling efficiency for BPS and Boomerang in RI-CLPM simulations, but OMRF comparisons were mixed and one posterior gap remained unresolved.
Scientific publication5 min read
A new methods preprint describes a model that groups observations, estimates cluster-specific relationships and flags unusual cases while handling missing responses and covariates.
Scientific publication4 min read
A statistical preprint adds spatial information to factor analysis and tests the approach on simulations, one brain-tissue section and African vegetation records.
Scientific publication5 min read
A statistical methods preprint reports a one-step log-quantile calculation, low variability and nearly no bias in selected simulations, with model-specific results in Google returns and hurricane-loss data.
Scientific publication5 min read
An arXiv preprint reports that BERGM Elastic Net had the lowest mean squared errors among the methods compared in a 50-replication simulation. It also had the lowest false-positive and false-discovery rates, but a lower true-positive rate at the selected reporting rule.
Scientific publication5 min read
A mathematical preprint reports a positive sampling-noise baseline and suggests that widespread small shifts may be easier to detect than low-rank changes in modeled settings.
Scientific publication5 min read
A methods preprint introduces a multiple-quantile test for structural changes in multivariate volatility. In simulations it stayed near its 5% target under the null, while an application to weekly Fama-French factor returns rejected covariance stability across four periods.
Scientific publication3 min read
A mathematical preprint gives finite-sample Gaussian and bootstrap bounds for U-statistic norms, with dimension-explicit results and a minimax rate for dense Kendall’s tau alternatives.
Scientific publication4 min read
A statistical methods preprint reports a theoretical link between observation frequency and prediction rates in scalar-on-function and function-on-function linear regression. It pairs a pooling-ridge estimator with matching bounds and tests the approach in simulations, wheat spectra and child-growth data.
Scientific publication5 min read
A penalised model selected a sparse cross-firm volatility network and had lower out-of-sample errors than preset networks in a 20-firm UK sample.
Scientific publication3 min read
A mathematical preprint derives upper bounds for gradient-trained, over-parametrized neural-network regression on exponentially β-mixing data and replaces ambient dimension with manifold dimension under an exact support condition.
Scientific publication4 min read
A preprint models Opportunity Zone designation and neighborhood spillovers in California, reporting positive estimated contrasts for designated tracts but uncertain results as policy expands.
Scientific publication4 min read
A new method combines independent frequency simulation with subsampling to estimate dependence that the first stage misses. Its full consistency is theoretical and conditional on stated assumptions, while simulations found close agreement with reported exact values in three synthetic models.
Scientific publication4 min read
A modeling paper combining theory, simulations and regional onset reconstructions says pooled mobility can retain aggregate invasion signals while making transport-mode and route attribution harder.
Scientific publication5 min read
A mathematical study finds distinct limiting behavior for rearranged density and copula estimates, with a separate boundary result for histograms.
Scientific publication3 min read
A theoretical preprint maps uncertainty about discrete outcomes into uncertainty about partially identified model parameters, then tests the approach in simulation.
Scientific publication4 min read
A new econometrics preprint reports that K-averaging can deliver the fastest theoretical convergence, while coefficient differences and weak trends limit the advantage in applications.
Scientific publication4 min read
An arXiv preprint presents a closed-form random-rotation test for autocorrelation. Its strongest reported results came at lag 3 and across selected lags, while moving-average and heavy-tail tests exposed important limits.
Scientific publication4 min read
A proposed deep-neural-network method explicitly models observed covariates, improving recovery of image-related effects in simulations and restoring much of an age-confounded MRI model’s performance.
Scientific publication4 min read
A preprint reports that SPM stayed unbiased in an out-of-phase simulation, with efficiency comparable to a correctly specified spline method.
Scientific publication5 min read
A mathematical analysis of Laplace-noised observations finds that privacy and recovery pull in opposite directions, with a sufficient low-noise consistency condition and an impossibility result at square-root-of-sample-size noise.
Scientific publication4 min read
The framework aims to turn quantitative estimates into cautious qualitative conclusions, with an explicit inconclusive outcome when the evidence does not separate the available possibilities.
Scientific publication5 min read
A methods study finds that standard Wald tests can miscalibrate false-positive rates in some generalized linear models, while an adjusted approach restores the intended null behavior in theory and selected simulations.
Scientific publication4 min read
A computational study found that PosCoSeA’s covariance-based influence scores closely matched exact leave-one-out changes in GLMM simulations, while agreement was weaker in N-mixture models.
Scientific publication4 min read
A statistical preprint proposes a model for incomplete compositional data and mild outliers, combining simulations with an application to daily activity durations.
Scientific publication3 min read
A theoretical study says a randomly pruned neural network can approach the minimax regression rate for smooth functions, with a dimension-reduced rate under a manifold assumption; its only empirical illustration is synthetic.
Scientific publication3 min read
The DExtrI method was designed to extrapolate interaction effects from one-factor-at-a-time training data to tests where several factors change together, with performance varying across datasets.
Scientific publication3 min read
In synthetic experiments, two proposed learners were more resilient to simulated unmeasured confounding, but their performance fell when instruments carried little information or treatment choices outnumbered instrument levels.
Scientific publication3 min read
A proposed framework separates concurrent and historical terms in curve-based regression; simulations found much lower coefficient errors for the combined model, while two gait applications illustrated the method without establishing causality.
Scientific publication4 min read
A statistical model reproduced controlled battery-ageing profiles closely in the data used to fit it, while its performance on new conditions remains untested.
Scientific publication5 min read
A mathematical note argues that the K − 1 factor in fixed-confidence best-arm identification is relocated rather than eliminated: it can count true nulls in one setup and reappear inside a composite test in another.
Scientific publication3 min read
A theoretical preprint proposes matching privacy spending to the exploration an online experiment needs, then extending that budget across a firm's portfolio.
Scientific publication4 min read