Filter by topic area using the buttons below.
On desktop, hover over the community icons for context; on mobile, tap for a legend.
Finding: Federal funding has fallen from 75% to 55% of total R1 R&D expenditure over fifty years with forecasts suggesting stabilization at or below current levels — though current cuts will likely accelerate that trend when 2025 data are available.
Recommendation: State and institutional leaders should use trend analysis, sensitivity analysis, geographic dependency mapping, and volatility rankings together to identify exposure and prioritize diversification and advocacy strategies before cuts accelerate.
Methods: Time series analysis, ARIMA forecasting, and animated geospatial mapping of HERD survey data across 1,400+ institutions, 1974–2024.
Finding: Education is a stronger and more consistent driver of participatory democracy than wealth across two centuries and six world regions.
Recommendation: Development policymakers should prioritize investment in education as a primary democratic institution-building strategy independent of economic growth trajectories.
Methods: Cross-national time series analysis of education, wealth, and democracy indicators using V-Dem and World Bank data, 1800–2000.
Finding: Healthcare worker density positively impacts life expectancy by up to a year, with oral health access representing a systematically neglected driver of mortality risk.
Recommendation: Health policymakers should treat oral health provider access as a mortality risk factor and incorporate dental workforce density into life expectancy models and health equity frameworks.
Methods: Linear regression and regional time series analysis of healthcare worker density and life expectancy across countries, with effect size estimation and identification of 43 primary intervention target countries.
Finding: Market integration improves traditional Hadza male childhood health with no effect on females, revealing a sex-based divergence in the health consequences of economic transition in a remote market economy.
Implication: This is the first study to apply bioarchaeological methods to a living population, providing a longitudinal biological measure of health change alongside conventional human biology approaches.
Methods: Linear mixed-effects modeling and Bayesian confirmation of developmental stress markers across 62 years of Hadza health data, with bibliometric network analysis of the research landscape.
Finding: Modeled on documented NIST performance patterns, FaceNet shows significant performance disparities across race, gender, and age, with accuracy declining most sharply for darker-skinned females and older individuals.
Recommendation: FRT systems should be evaluated on the balance between security benefit and demographic burden — flagging who faces additional scrutiny not because they pose a risk but because the technology failed them.
Methods: Bias audit of FaceNet using synthetic data modeled on documented VGGFace2 training bias patterns across 126 demographic groups and 25,200 comparisons.
App: Interactive Shiny app for exploring NSF OISE award data across four programs — PIRE, IRES, AccelNet, and Global Centers — filterable by year, program, and EPSCoR status.
Highlights: Geographic map, award scatter plot, and funding trend line chart surface program-level disparities in award size, geographic distribution, and EPSCoR representation across two decades of international science investment.
Methods: R Shiny app with reactive filtering, plotly scatter geo map, plotly bubble plot, ggplot2 line chart, and DT searchable table built on NSF Awards Database data.
Finding: The Palmer-Strobeck protocol for fluctuating asymmetry analysis can be fully implemented in modular, reproducible R scripts accessible to researchers without specialist programming expertise.
Implication: Open reproducible implementations of established protocols reduce methodological inconsistency across studies and lower the barrier to entry for developmental stress research.
Methods: Modular R implementation of the full Palmer-Strobeck nine-step FA screening protocol including measurement error detection, outlier screening, mixed-model ANOVA, and FA10 index computation.
Finding: Federal dependency varies dramatically across disciplines and recent federal trends signal a shift away from basic science funding without industry investment filling the gap.
Recommendation: Institutions and funders should prioritize support for high-dependency fields with no nonfederal backstop, particularly physical sciences and geosciences.
Methods: Field-level analysis of 24 S&E disciplines using HERD survey data across five decades, with ARIMA forecasts through 2034.
Finding: Women's representation in governance and educational gender parity are mutually reinforcing across world regions, with egalitarian democracy acting as both cause and consequence of gender equity gains.
Recommendation: Gender equity policymakers should invest simultaneously in educational parity and political representation rather than sequencing one before the other.
Methods: Cross-national regression and data visualization across six world regions using V-Dem and World Bank data.
Finding: Odor identification in real-world field conditions differs from laboratory performance, with environmental context shaping olfactory accuracy in ways that controlled studies cannot capture.
Implication: Olfactory research should expand to include field-based methods alongside laboratory protocols to produce ecologically valid findings about human sensory capacity.
Methods: GLMs, quasibinomial models, correspondence analysis, and multinomial logistic regression across forced-choice and free-choice odor identification tasks in field and lab settings.
Finding: Persistent hunter-gatherers in northwest Kyushu were consistent in population health metrics before and after agricultural expansion, suggesting there were cultural buffers against the health costs of the Yayoi transition.
Implication: This is the first application of resilience theory to bioarchaeological analysis, establishing a framework for evaluating population health stability under cultural and environmental pressure that has since been adopted more broadly in the field.
Methods: Enamel hypoplasia rate modeling and fluctuating asymmetry analysis using the full Palmer-Strobeck nine-step protocol on prehistoric skeletal remains.
Finding: A fine-tuned ResNet50 binary classifier successfully distinguishes genuine from morphed face images, with Grad-CAM visualization identifying model attention patterns and failure modes.
Implication: Automated morph attack detection is viable at scale, with implications for border control, identity verification, and any system relying on facial recognition for high-stakes authentication.
Methods: ResNet50 fine-tuned on 500 LFW images and 250 alpha-blended morphs, evaluated on accuracy, precision, recall, and ROC/AUC, with results benchmarked against published morph detection studies.
Finding: Aging demographics will constrain R&D leadership pipelines across all six world regions by 2035, with variation in timing and severity that aggregate global trends obscure.
Recommendation: Research funders and policymakers should prioritize international partnerships and infrastructure investment in demographically young regions before aging economies exhaust their R&D leadership pipelines.
Methods: Time series and regression analysis of aging demographics and R&D productivity across six world regions using V-Dem and World Bank data, 1996–2019.
Finding: A music festival case study reveals that attitudes toward surveillance and policing vary significantly by sex, experience, and sexual orientation, with trust in law enforcement mediating perceptions of safety.
Recommendation: Event organizers and public safety policymakers should adopt differentiated surveillance and policing approaches that account for demographic variation in trust and safety perceptions.
Methods: Logistic regression and Bayesian modeling of survey data from festival-goers on surveillance, policing, and safety perceptions.
Finding: Colonial health disparities in Ryukyu burial populations were spatially structured, with remote sites showing a distinct sex-differentiated health profile and differential biological consequences of political subjugation.
Implication: Colonial health in Japan has received little scholarly attention — this study is among the first to apply bioarchaeological methods to Japanese colonial contexts, revealing that geographic marginalization compounded the biological consequences of political subjugation.
Methods: GLM analysis with site, sex, and site×sex interaction predictors across five skeletal health markers from three colonial-era Ryukyu burial sites.
Finding: Developmental stress in wild Nigerian olive baboons varies by sex and life history stage, with juvenile stress markers revealing biological sensitivity to early environmental conditions.
Implication: Non-human primate developmental stress patterns provide a comparative baseline for understanding how early life adversity shapes long-term biological outcomes in humans.
Methods: Fluctuating asymmetry analysis using FA10b index across ten replicate measurement sets, with Levene's tests for sex differences, life history stage stress, and weaning and reproductive stress signals.
Finding: A serverless pipeline using Claude 3 Haiku via AWS Bedrock successfully automates document validation and auditing with low latency and no persistent infrastructure.
Recommendation: Serverless LLM pipelines should be considered as a cost-effective and scalable approach to document validation tasks traditionally requiring manual review or complex rule-based systems.
Methods: Lambda-based pipeline invoking Claude 3 Haiku via Bedrock across three prompting strategies, evaluated on parse success, response time, and accuracy against known ground truth, with cost modeling across three architectures.
Finding: Research impact varies substantially across countries and fields in ways that raw publication counts obscure, with small high-income nations consistently punching above their weight.
Recommendation: Institutions and funders should adopt context-sensitive benchmarking frameworks that account for country size, field composition, and resource constraints rather than applying uniform impact standards.
Methods: Bibliometric framework for benchmarking research impact across countries and fields using Google BigQuery and network analysis.
Finding: Terrorism attacks within ape habitat ranges in sub-Saharan Africa cluster in the poorest and least democratic regions, creating a compounding threat to human and non-human ape populations.
Recommendation: Conservation policymakers should integrate political stability and poverty reduction strategies into ape conservation frameworks rather than treating security and ecology as separate policy domains.
Methods: QGIS mapping of terrorism attacks overlaid on primate habitat data across sub-Saharan Africa.
Finding: Functional odorant receptor testing revealed that extinct hominins and ancient human populations had a more limited olfactory repertoire than contemporary humans.
Implication: Greater olfactory receptor diversity in contemporary humans may reflect functional ecological adaptation, suggesting that sensory capacity continued to evolve after the divergence of archaic and modern lineages.
Methods: Ancient DNA extraction and variant calling using VCFtools and HPC pipelines, phylogenetic reconstruction, chimeric PCR with site-directed mutagenesis, and luciferase reporter assays across seven odorants.
Finding: Occipital bone morphology varies systematically across hunter-gatherer, horticultural, and agricultural skeletal collections, reflecting differences in biomechanical loading by subsistence type and sex.
Implication: By comprehensively synthesizing scattered studies through multivariate classification, this study determines that the use of occipital morphology for demographic classification falls far below forensic evidentiary standards while establishing value as a bioarchaeological tool for reconstructing subsistence history — an avenue not previously explored.
Methods: Two-way ANOVA, discriminant function analysis, and PCA of ectocranial variables across subsistence groups, with sex estimation models using foramen magnum morphology.
Finding: Few-shot prompt engineering via AWS Bedrock successfully classifies biometric sensor data, with performance varying substantially across Temperature, Top-K, and Top-P parameter combinations.
Implication: Parameter optimization is critical for LLM-based classification tasks — default settings produce substantially suboptimal results compared to tuned configurations.
Methods: Few-shot prompt engineering with systematic parameter sweep across Temperature, Top-K, and Top-P, with stochastic consistency testing and cross-model comparison via AWS Bedrock.
Finding: DEI barriers in international research organizations are structural rather than individual, requiring evidence-based policy frameworks rather than awareness initiatives.
Recommendation: International science organizations should move beyond awareness initiatives toward structural policy reform guided by mixed-methods evidence.
Methods: Survey analytics, Slack network analysis, and bibliometric review across peer international science organizations.
Finding: Political violence in China is geographically tied to the Uyghur ethnic minority, an indigenous people under active state suppression, suggesting that state-labeled terrorism may reflect political activism rather than conventional terrorism.
Recommendation: Researchers and policymakers should critically examine state-defined terrorism databases as potential evidence of political repression, particularly in authoritarian contexts where dissent and terrorism are conflated.
Methods: QGIS mapping of Global Terrorism Database data, China 2010–2018.
Finding: A small subset of storm and flood event types account for the majority of fatalities, injuries, and property damage across the US from 1996–2011.
Recommendation: Emergency management agencies should prioritize preparedness and mitigation resources toward the highest-impact event types identified by VSL-weighted impact analysis and spatial damage mapping.
Methods: VSL-weighted combined impact indexing, animated time series of top event types, and interactive choropleth mapping of state-level damage normalized by population using NOAA storm data.
Finding: Random forest and gradient boosting models applied to wearable accelerometer data successfully classify bicep curl form into correct and common error categories with high accuracy.
Implication: Exercise form is as important as exercise volume for fitness outcomes — wearable sensor data can be tuned to measure technique quality without human supervision or specialized equipment.
Methods: Four models trained and evaluated using caret with 5-fold cross-validation, variable importance extraction, and optimal predictor subset identification across decision tree, bagged tree, GBM, and random forest approaches.