Background Patients increasingly ask large language model (LLM) assistants which doctor to see. Existing audits observe real, correlated physician profiles, so they cannot isolate what drives these recommendations. Objective To estimate how reputation signals, fees, list position, and names signaling gender and ethnicity affect LLM physician recommendations in a controlled task. Methods Prespecified randomized choice-based conjoint audit. Seven LLMs (six small open-weight models and GPT-4o-mini) each chose one of five synthetic family-medicine physician cards with independently randomized attributes (3024 choice sets; three personas; nine prompts; eight secondary arms). Gender and ethnicity were signaled only by names. Average marginal component effects were estimated with choice-set–clustered linear probability models, per model and pooled. Results Across 40,068 responses, a 3.9→4.7 rating raised choice probability by 31.4 percentage points (pp; 95% CI 30.5–32.3) and a $90→$190 fee lowered it by 20.0 pp (19.1–21.0). Pooled, female-signaled names were chosen 2.5 pp (1.8–3.2) more often than male-signaled names, and Hispanic- and South-Asian–signaled names were chosen 2.8–2.9 pp more often than White-signaled names; these held with names as the units of inference, whereas a Black-signaled estimate (1.3 pp) did not. Specific names within categories differed (placebo P<.001), and name-cue effects varied significantly between models (female-signaled: −0.5 to +7.4 pp; gpt-4o-mini +0.8, CI −0.3 to 1.8). The last-listed card lost 6.2 pp. Gender or ethnicity appeared in ≤0.03% of stated reasons. Conclusions In this panel, recommendations ran mainly on ratings and fees but also responded, in model-specific ways, to name cues and list position that the models’ explanations did not mention. These controlled-task differences are not estimates of real-world access; they call for behavioral audits of each model version. The materials will be released for repeat audits.
Managing municipal solid waste in rapidly urbanizing Sub-Saharan Africa remains challenging due to dispersed informal dumping and limited high-resolution datasets for spatial monitoring. We present an open-access deep learning model for automated detection of openly dumped dispersed MSW (ODD-MSW) via crowdsourced UAV imagery, trained and evaluated across 29 regions in 10 Sub-Saharan African countries, encompassing diverse environmental contexts. A deep learning model trained on manually annotated image tiles achieved an overall accuracy of 92.87% (F1 = 0.927) in detecting ODD-MSW across all study regions. Applied without any site-specific retraining to 25 independent OpenAerialMap scenes from 21 cities in 19 countries on five continents, the frozen model retained an F1 score of 0.867, providing direct evidence of transferability beyond the training domain. Predicted distributions reveal heterogeneous accumulation patterns, ranging from localized hotspots - often along waterways, where waste can exacerbate flood and public health risks - to more dispersed litter across urban areas. Waste accumulation is most strongly associated with population density (Spearman rho = 0.674, p < 0.001 ) and indicators of lack of local infrastructure access ( rho = 0.700, p < 0.001 ), whereas its relationship with broader measures of regional development is weaker and not statistically significant ( rho = 0.215, p = 0.26 ), highlighting the importance of fine-scale data for understanding localized waste dynamics. By releasing the model, this study provides a ready-to-use tool for UAV imagery collected by municipalities and local mapping communities, enabling ODD-MSW monitoring without extensive technical expertise. This approach empowers local practitioners to convert UAV imagery into actionable insights, supporting targeted interventions and improved municipal solid waste management across Sub-Saharan Africa.
Reducing CO2 emissions from residential space heating is critical to achieve carbon neutrality, particularly in urban areas. However, effective climate action is often hindered by the lack of precise location-specific emissions data. This study introduces the first web tool for estimating direct (i.e., scope 1) carbon dioxide emissions resulting from space heating of residential buildings at a sub-city scale (100-m resolution). The web tool employs a bottom-up approach entirely based on publicly available data, including buildings’ age of construction and energy source, average living space per capita, and population counts. Validation against official emission inventories at the national level and for four cities in Germany showed that the tool’s CO2 estimates deviated by less than 2% Visualized spatial emission patterns reveal their spatial heterogeneity and apparent associations with urban morphology, highlighting areas where action could be prioritized. A web-based dashboard is available at HeiGIT’s Climate Action Navigator, and the code of the residential heating emission tool available on GitLab.
We present a fully variational, model-independent formulation of the Cut Finite Element Method (CutFEM) for finite-strain elasticity. The discrete problem is derived from a single augmented energy functional consisting of the bulk hyperelastic energy, the Nitsche terms that impose the boundary conditions weakly, and the ghost-penalty stabilisation. At each nonlinear iterate, the residual is the exact first variation of this functional with the adaptive Nitsche weight frozen, while the correction uses a symmetric, coercive approximation of its Hessian. Automatic differentiation (AD) generates the first Piola–Kirchhoff stress tensor and the elasticity tensor directly from the scalar energy density, avoiding manual re-derivation when exchanging hyperelastic models. To our knowledge, this is the first unfitted finite-strain scheme combining an energy-only, model-independent construction with AD and an accuracy analysis at unfitted boundaries.Analysis of the linearised problem solved at each quasi-Newton step establishes cut-independent coercivity, continuity, and an O(h−2) condition number bound, yielding a quasi-optimal convergence theorem for regular solutions through the Brezzi–Rappaz–Raviart framework. Numerically, the method attains optimal h-convergence for quadratic and cubic elements on a smooth test case. Furthermore, we quantify the method’s accuracy limit at mixed Dirichlet–Neumann junctions using the Kolosov–Muskhelishvili characteristic equation. The exact solution’s corner singularity caps the convergence rate identically for fitted and unfitted methods. Local mesh refinement removes this bound: we verify the recovery of optimal rates numerically for first-order elements, and prove that the unfitted discretisation inherits the rate the underlying refinement attains, with cut-independent constants.
Despite major gains in primary school enrollment across low- and middle-income countries, secondary school completion remains low. We conducted a randomized controlled trial in Burkina Faso to evaluate the effectiveness of three interventions designed to reduce early secondary school leaving: (i) information on the economic and health returns to education, (ii) tailored tutoring, and (iii) conditional cash incentives. The cash intervention significantly increased school enrollment, any school attendance, and regular attendance, and reduced the total number of school days missed. Tutoring significantly increased school enrollment, while its effects on regular attendance and academic progression were imprecisely estimated. The information intervention generated limited evidence of impacts, with no statistically significant effects on the primary outcome or academic progression and only marginal evidence of an increase in any school attendance. Financial and academic support can help adolescents who left school early re-enroll, but larger studies are needed to identify which interventions most effectively sustain regular attendance.
