Institutt for datavitenskap og analyse

Mathematics and Computers in Simulation 250 s. 152-177 Doi: https://doi.org/10.1016/j.matcom.2026.06.024

We study a stochastic model for the installation of renewable energy capacity under demand uncertainty and jump driven dynamics. The system is governed by a multidimensional Ornstein–Uhlenbeck (OU) process driven by a subordinator, capturing abrupt variations in renewable generation and electricity load. Installation decisions are modeled through control actions that increase capacity in response to environmental and economic conditions. We consider two distinct solution approaches. First, we implement a structured threshold based control rule, where capacity is increased proportionally when the stochastic capacity factor falls below a fixed level. This formulation leads to a nonlinear partial integro-differential equation (PIDE), which we solve by reformulating it as a backward stochastic differential equation with jumps. We extend the DBDP solver in Huré et al. (2020) to the pure jump setting, employing a dual neural network architecture to approximate both the value function and the jump sensitivity. Second, to benchmark the performance of the first approach, we propose a deep control algorithm that directly learns a state and time dependent feedback threshold policy by minimizing the expected cost functional using neural networks. This approach preserves the threshold based structure while allowing the threshold to adapt dynamically to the evolving system state, enabling more flexible and responsive interventions. Numerical experiments on calibrated data highlight the strengths of both methods. While the threshold based BSDE approach offers interpretability and tractability, the deep control strategy achieves improved performance through flexibility in capacity allocation. Together, these tools provide a robust framework for decision support in long term renewable energy expansion under uncertainty.

Journal of Infection and Public Health 19(9) s. 1-12 Doi: https://doi.org/10.1016/j.jiph.2026.103328

Background: Urban areas often act as early epicentres of pandemics, yet transmission dynamics across regions with different demographic profiles remains poorly understood. We aimed to characterise COVID-19 transmission in Norway and assess how it interacted with infection control measures across urban and rural settings. Methods: We analysed daily COVID-19 cases across three pandemic waves (Alpha, Delta, Omicron). For local trends, 36 sites in the Greater Oslo Region were grouped by infection trajectories using group-based trajectory modelling. For national trends, we examined the cities of Oslo, Bergen, Trondheim, Stavanger, Tromsø, and Akershus county. Infection trends were analysed using linear mixed models and lagged regression analyses and related to demographic characteristics. Results: In the Greater Oslo Region, three distinct site groups-high, moderate, and low infection-were identified. The high infection group had the highest population density, largest proportion of crowded housing and immigrant population and preceded the moderate and low groups by 5-13 days during the Alpha wave. Nationally, Oslo's infection rates peaked 11-24 days before other cities during the Alpha wave, with shorter lags during the Delta and Omicron waves. Despite similar control measures, infection rates differed substantially, with the high infection group and Oslo city consistently bearing the greatest burden. Conclusions: Densely populated areas with crowded housing act as early warning signals, with infection growth preceding that in other areas by days to weeks. Systematic monitoring of such high-burden urban areas could provide valuable lead time for policymakers and support more targeted, timely responses in future pandemics.

Psychometrika s. 29-29 Doi: https://doi.org/10.1017/psy.2026.10139

Due to the indeterminacy problem in factor analysis, factors are not uniquely determined by observed data. We investigate the consequences of this problem for the interpretation of simulation studies and for linear structural equation models. For simulation procedures, factor indeterminacy implies that as long as the simulated data have the correct covariance matrix, there exists a possibly large space of factors and error terms that can generate the simulated data via the model equations. For linear structural equation models, the linearity of the structural part of the model is only partly testable by the distribution of the data. Practical recommendations for conducting simulation studies are provided. We give numerical illustrations, including a case where the observed data are normal and fit a linear structural equation model perfectly, yet are generated by a highly non-linear structural equation model.

Psychometrika s. 32-32 Doi: https://doi.org/10.1017/psy.2026.10135

Mean and sum scores are widely used in psychometric practice, but their behavior under weak assumptions is not fully understood. We study two basic questions under a latent variable framework with increasing-item asymptotics: when does the observed mean score approximate the true score of the administered form, and when is a true score insensitive to removing a vanishing fraction of items? For the first question, we introduce martingale factor models, which, for polytomous items, weaken conditional independence to a conditional-mean restriction, yet still yield concentration and asymptotic normality results for mean scores around their true scores. For the second question, we show that this stability is equivalent to the convergence of the true scores when the items are bounded, and we identify conditions under which this convergence takes place. These results clarify when mean scores can be interpreted as stable proxies for latent variables, and yield psychometrically relevant consequences. Several theoretical and numerical illustrations are given, including a new interpretation of large-form mean-score correlations and a person-specific confidence interval procedure for the true score.

Structural Equation Modeling s. 5-5 Doi: https://doi.org/10.1080/10705511.2026.2651100

Muda and Yangxuan recommended a ridge-calibrated test statistic for controlling Type I error inflationin confirmatory factor analysis under non-normality. Although improving finite-sample inference is animportant goal, their empirical evaluation is undermined by incorrectly implemented benchmark statis-tics, misattributed comparison methods, and a simulation design restricted to asymptotically robustconditions. Using corrected code, we reanalyze the original conditions and replace the misattributedbenchmarks with the penalized eigenvalue procedures pEBA4RLS and pOLS2RLS: Under the originalconditions, the best-performing procedures are the ridge-calibrated TCsFCr statistic and the penalizedeigenvalue methods. However, under additional non-asymptotically robust conditions, including high-dimensional models, pEBA4RLS provides the most reliable Type I error control and outperforms theridge-calibrated competitors.