Hierarchical Bayesian Framework Research Articles

Loyalty Currency and Mental Accounting: Do Consumers Treat Points Like Money?

Problem definition: Loyalty programs have greatly expanded in scale and scope, and loyalty points issued by firms serve as a new form of currency alongside the traditional currency of money. In this paper, we study how consumers decide to pay with points or money for a purchase and how these decisions are affected by consumers’ points earning characteristics. Methodology/results: We develop a model of consumers’ payment choices and estimate it on proprietary loyalty program data from a major U.S. airline company using a hierarchical Bayesian framework. Our results demonstrate that mental accounting, the subjective perceived value of points, and the reference exchange rate play important roles in consumers’ payment choices. Moreover, the primary points earning source and the total earning level are jointly associated with consumers’ attitudes toward points and money: Consumers who earn many points and mostly with the focal firm tend to value points more than money, while those who earn few points or mostly through a cobranded credit card tend to value money more than points. To better understand heterogeneity in consumers’ attitudes toward points, we propose a probabilistic segmentation of consumers and identify four behavioral segments with distinctive characteristics. Through counterfactual analysis, we demonstrate how a firm can implement money and point pricing policies to optimally target and influence consumers’ payment choices. Managerial implications: It is important for firms to understand how consumers think about points and decide to pay with points or money as it affects firms’ cash flows, profitability, and consumer loyalty and engagement. We demonstrate that firms can design optimal price targeting policies by utilizing the significant level of heterogeneity in consumers’ attitudes toward points. Furthermore, our results indicate that firms that are expanding their partnership networks and offering more points earning sources should consider the impact that this would have on consumers’ mental accounting of money and points, and consequently their payment choices. Supplemental Material: The online appendix is available at https://doi.org/10.1287/msom.2022.0287 .

The Utility of Socioeconomic and Remoteness Indicators in Understanding the Geographical Variation in the Regional Prevalence of Early Childhood Vulnerability in Australia

The family lives of children and their early childhood development outcomes are attributable to the level of socio-economic disadvantage and relative isolation. This study aims to investigate how the disadvantage of the local area (i.e., socio-economic indexes for areas (SEIFA)) and the remoteness (i.e., accessibility/remoteness index of Australia (ARIA)) contribute to improved prevalence estimates of child development vulnerability in statistical areas level 3 (SA3) and 4 (SA4) across Australia. Data from the 2018 Australian Early Development Census (AEDC) has been used. The study included 308,953 children involved in the AEDC 2018 where one-in-ten of them were considered to be developmentally vulnerable, nationally. We developed models in a hierarchical Bayesian framework at the SA3 level using SEIFA and ARIA indices as covariates to account for spatial and unobserved heterogeneity. The performances of developed models are examined based on the consistency at SA3, SA4, and state level. The results reveal that SEIFA makes a significant contribution to explaining the spatial variation in childhood development vulnerability across small domains in Australia. Further, the inclusion of the ARIA score improves the model performance and provides better accuracy, particularly in remote and very remote regions. In these regions, the spatial model fails to distinguish the remoteness characteristics. The chosen non-spatial model accounting for heterogeneity at higher hierarchies performs best. The utilization of socio-economic disadvantage and geographic remoteness of the finer level domains helps to explain the geographic variation in child development vulnerability, particularly in sparsely populated remote regions in Australia.

A STATISTICAL FRAMEWORK FOR THE MODELING OF THE CONTINUOUS PROGRESSION OF ALZHEIMER'S DISEASE.

Throughout an organism's life, a multitude of biological systems transition through complex biophysical processes. These processes serve as indicators of the underlying biological states. Inferring these latent unobserved states is a key problem in modern biology and neuroscience. Unfortunately, in many experimental setups we can at best obtain snapshots of the system at different times for different individuals, and one major challenge is the one of reconciling those measurements. This formalism is particularly relevant in the study of Alzheimer's Disease (AD) progression, in which we observe in brain donors the aggregation of pathological proteins but the underlying disease state is unknown. The progression of AD can be modeled by assigning a latent score - termed pseudotime - to each pathological state, creating a pseudotemporal ordering of donors based on their pathological burden. This paper proposes a hierarchical Bayesian framework to model AD progression using detailed quantification of multiple AD pathological proteins from the Seattle AD Brain Cell Atlas consortium (SEA-AD). Inspired by biophysical models, we model pathological burden as an exponential process. Theoretical properties of the model are studied, by using linearization to reveal convergence and identifiability properties. We provide Markov chain Monte Carlo estimation algorithms, and show the effectiveness of our approach with multiple simulation studies across data conditions. Applying the methodology to SEA-AD brain data, we infer pseudotime for each donor and order them by pathological burden. Finally, we analyze the information within each pathological feature and utilize it to refine the model by focusing on the most informative pathologies. This lays the groundwork for suggesting future experimental design approaches.

Quantitative Expression of Latent Disease Factors in Individuals Associated with Psychopathology Dimensions and Treatment Response.

Psychiatric comorbidity is common in symptom-based diagnoses like autism spectrum disorder (ASD), attention/deficit hyper-activity disorder (ADHD), and obsessive-compulsive disorder (OCD). However, these co-occurring symptoms mediated by shared and/or distinct neural mechanisms are difficult to profile at the individual level. Capitalizing on unsupervised machine learning with a hierarchical Bayesian framework, we derived latent disease factors from resting-state functional connectivity data in a hybrid cohort of ASD and ADHD and delineated individual associations with dimensional symptoms based on canonical correlation analysis. Models based on the same factors generalized to previously unseen individuals in a subclinical cohort and one local OCD database with a subset of patients undergoing neurosurgical intervention. Four factors, identified as variably co-expressed in each patient, were significantly correlated with distinct symptom domains (r = -0.26-0.53, P < 0.05): behavioral regulation (Factor-1), communication (Factor-2), anxiety (Factor-3), adaptive behaviors (Factor-4). Moreover, we demonstrated Factor-1 expressed in patients with OCD and Factor-3 expressed in participants with anxiety, at the degree to which factor expression was significantly predictive of individual symptom scores (r = 0.18-0.5, P < 0.01). Importantly, peri-intervention changes in Factor-1 of OCD were associated with variable treatment outcomes (r = 0.39, P < 0.05). Our results indicate that these data-derived latent disease factors quantify individual factor expression to inform dimensional symptom and treatment outcomes across cohorts, which may promote quantitative psychiatric diagnosis and personalized intervention.

Interactions between runoff volume, timing, and annual temperatures shape migration phenology of a threatened adfluvial sucker

AbstractMigratory fish species have been particularly impacted by changes to the hydrologic and climatic cues to which migration and spawning behaviours have been adapted across generations. While conservation and recovery programs increasingly implement flow management actions to promote successful migration and spawning, uncertainty regarding how spawning migration phenology responds to changing environmental conditions can limit the ability to effectively target such recovery actions. Here, we use a Bayesian hierarchical modelling framework to analyse spawning migration phenology of individually tagged June suckers (Chasmistes liorus) – a federally threatened, long‐lived, iteroparous, adfluvial species endemic to Utah Lake (Utah, USA) and its tributaries. We then examine how annual hydrologic and thermal conditions relate to different components of annual migration phenology, including peak migration date, in‐stream residence time, and among‐individual variation in migration timing. Peak migration date occurred earlier in years with warmer spring air temperatures (a proxy for water temperatures), though this effect interacted with peak runoff timing. Both residence time and among‐individual variation in migration timing were greater in years with larger spring discharge and later peak flows. Residence time was also longer in warmer years. These results highlight how natural and anthropogenic changes to river flow and thermal regimes are likely to impact June sucker migration timing and duration, and our approach can be applied to other migratory species to identify the external drivers of the different components of migration phenology.

Uncertainty analysis for forest height inversion using L / P band PolInSAR datasets and RVoG model over kryclan forest site

Forest height, as a measure of the quantity and quality of forest resources, plays a significant role in the study of the ecological functions performed by forests. Although the polarimetric synthetic aperture radar interferometry (PolInSAR) technique has evolved as a potent method for forest height inversion, uncertainties still exist in the process of estimating forest height, and the uncertainties in predicted forest height directly lead into the uncertainty of terrestrial carbon stock calculation results. In this study, we took the Random Volume over Ground (RVoG) model as likelihood function and constructed a hierarchical Bayesian framework to calculate and reduce the uncertainty of forest height inversion using L / P band PolInSAR airborne data via RVoG model. Uncertainties resulted from five canopy types and three forest densities were analyzed, respectively. The results showed that among the five different canopy types, L band has the highest prediction accuracy in pure coniferous canopy with Acc. = 0.90. The uncertainty is extremely low for pure forest, with the ratio of uncertainty values of 0.09 for L band and 0.15 for the P band in pure coniferous canopy, and uncertainty values of 0.16 for L band and 0.11 for P band in pure broadleaf canopy, respectively. Furthermore, when the forest density is between 300 and 600 stems/ha, the ratio of uncertainties for the L band is 0.27, whereas the P band is 0.24. As forest density increases, the uncertainty in forest height estimates decreases for both bands. The changes in canopy types and forest density affect forest height estimation uncertainties obviously, the effects are different at each frequency. The forest height inversion accuracy of the L band in pure coniferous canopy surpasses that in other canopy types, with the lowest uncertainty. P band performed well in broadleaf canopy forest height inversion. The inversion uncertainties at both frequencies decrease with increase of forest densities.

Non-Gaussian normative modelling with hierarchical Bayesian regression

Abstract Normative modelling is an emerging technique for parsing heterogeneity in clinical cohorts. This can be implemented in practice using hierarchical Bayesian regression, which provides an elegant probabilistic solution to handle site variation in a federated learning framework. However, applications of this method to date have employed a Gaussian assumption, which may be restrictive in some applications. We have extended the hierarchical Bayesian regression framework to flexibly model non-Gaussian data with heteroskdastic skewness and kurtosis. To this end, we employ a flexible distribution from the sinh-arcsinh (SHASH) family, and introduce a novel reparameterisation and a Markov chain Monte Carlo sampling approach to perform inference in this model. Using a large neuroimaging dataset collected at 82 different sites, we show that the results achieved with this extension are equivalent or better than a warped Bayesian linear regression baseline model on most datasets, while providing better control over the parameters governing the shape of distributions that the approach is able to model. We also demonstrate that the attained flexibility is useful for accurately modelling highly nonlinear relationships between aging and imaging derived phenotypes, which shows that the extension is important for pushing the field of normative modelling forward. All methods described here are available in the open-source pcntoolkit.

Assessing the E-bike trends and impact on sustainable mobility: A national-level study in the Netherlands

Over the past decade, e-bikes have become increasingly popular, sparking interest in their potential replacement for car use and benefit for the environment. However, many studies on e-bike development and their substitution effects exhibit limitations. These include a lack of modeling on e-bike trend development, inadequate assessments of their impact on national-level mobility, a predominant focus on commuting, and a lack of foresight into future e-bike substitution effects. Our research introduces an innovative approach to model e-bike development, employing a multilevel Richards growth curve model fitted within a hierarchical Bayesian framework using the Hamiltonian Monte Carlo (HMC) method. Further, we incorporate an intention-based method to delve into the potential of e-bikes in stimulating sustainable mobility in the Netherlands. Our findings highlight an ongoing increase in e-bike distance share, with marked gender and generational differences in growth patterns. Notably, women have higher e-bike usage than men, and this gap is narrowing for older age groups while widening among younger demographics, suggesting that younger people may adopt e-bike usage differently than older generation. E-bike ownership strongly reduces the conventional bicycle use and, to a lesser extent, car and public transport use, especially for commuting. This study provides insight into whether and to what extent e-bikes substitute for car use and other modes of transportation, and how the expected growth in e-bike use in coming years may impact national mobility in the Netherlands.

Improved hierarchical Bayesian modeling framework with arbitrary polynomial chaos for probabilistic model updating

Bayesian finite element model updating techniques have found widespread application in the structural health monitoring. The conventional Bayesian modeling framework (CBMF) identifies the posterior distribution of structural parameters using data from a single experiment. However, structural parameters exhibit variability due to changes in environmental or experimental conditions, an aspect overlooked by CBMF. In recent years, the hierarchical Bayesian modeling framework (HBMF) has gradually attracted attention. This framework incorporates an additional layer of hyperparameters and utilizes multiple sets of experimental data to describe the variability in structural parameters caused by changes in experimental conditions. Unfortunately, the increase in hyperparameters and data size leads the HBMF to face the dilemma of exceedingly high computational costs. To address this issue, this study proposes an improved hierarchical Bayesian modeling framework (IHBMF). To enhance the computational efficiency of IHBMF, the importance sampling method is employed to simplify the computation of the likelihood function for hyperparameters. Subsequently, a surrogate model based on arbitrary polynomial chaos expansion is introduced to further reduce the computational cost of IHBMF. The framework's accuracy and efficiency were validated through analysis of a simply supported beam and a steel pedestrian bridge. The results demonstrate that IHBMF not only delivers a more accurate quantification of structural parameter uncertainty than CBMF but also exhibits higher computational efficiency compared to HBMF, showcasing its superior capability in structural health monitoring applications.

Strong outflows and inefficient star formation in the reionization-era ultrafaint dwarf galaxy Eridanus ii

ABSTRACT We present novel constraints on the underlying galaxy formation physics (e.g. mass-loading factor, star formation history, and metal retention) at z ≳ 7 for the low-mass (M* ∼ 105 M⊙) Local Group ultrafaint dwarf galaxy (UFD) Eridanus ii (Eri ii). Using a hierarchical Bayesian framework, we apply a one-zone chemical evolution model to Eri ii’s CaHK-based photometric metallicity distribution function (MDF; [Fe/H]) and find that the evolution of Eri ii is well characterized by a short, exponentially declining star formation history ($\tau _\text{SFH}=0.39\pm _{0.13}^{0.18}$ Gyr), a low star formation efficiency ($\tau _\text{SFE}=27.56\pm _{12.92}^{25.14}$ Gyr), and a large mass-loading factor ($\eta =194.53\pm _{42.67}^{33.37}$). Our results are consistent with Eri ii forming the majority of its stars before the end of reionization. The large mass-loading factor implies strong outflows in the early history of Eri ii and is in good agreement with theoretical predictions for the mass scaling of galactic winds. It also results in the ejection of &amp;gt;90 per cent of the metals produced in Eri ii. We make predictions for the distribution of [Mg/Fe]–[Fe/H] in Eri ii as well as the prevalence of ultra metal-poor stars, both of which can be tested by future chemical abundance measurements. Spectroscopic follow-up of the highest metallicity stars in Eri ii ([Fe/H] &amp;gt; −2) will greatly improve model constraints. Our new framework can readily be applied to all UFDs throughout the Local Group, providing new insights into the underlying physics governing the evolution of the faintest galaxies in the reionization era.

Grouped Spherical Data Modeling Through Hierarchical Nonparametric Bayesian Models and Its Application to fMRI Data Analysis.

Recently, spherical data (i.e., L2 normalized vectors) modeling has become a promising research topic in various real-world applications (such as gene expression data analysis, document categorization, and gesture recognition). In this work, we propose a hierarchical nonparametric Bayesian model based on von Mises-Fisher (VMF) distributions for modeling spherical data that involve multiple groups, where each observation within a group is sampled from a VMF mixture model with an infinite number of components allowing them to be shared across groups. Our model is formulated by employing a hierarchical nonparametric Bayesian framework known as the hierarchical Pitman-Yor (HPY) process mixture model, which possesses a power-law nature over the distribution of the components and is particularly useful for data distributions with heavy tails and skewness. To learn the proposed HPY process mixture model with VMF distributions, we systematically develop a closed-form optimization algorithm based on variational Bayes (VB). The merits of the proposed hierarchical Bayesian nonparametric model for modeling grouped spherical data are demonstrated through experiments on both synthetic data and a real-world application about resting-state functional magnetic resonance imaging (fMRI) data analysis.

Characterisation of Plasmodium vivax lactate dehydrogenase dynamics in P. vivax infections

Plasmodium vivax lactate dehydrogenase (PvLDH) is an essential enzyme in the glycolytic pathway of P. vivax. It is widely used as a diagnostic biomarker and a measure of total-body parasite biomass in vivax malaria. However, the dynamics of PvLDH remains poorly understood. Here, we developed mathematical models that capture parasite and matrix PvLDH dynamics in ex vivo culture and the human host. We estimated key biological parameters characterising in vivo PvLDH dynamics based on longitudinal data of parasitemia and PvLDH concentration collected from P. vivax-infected humans, with the estimates informed by the ex vivo data as prior knowledge in a Bayesian hierarchical framework. We found that the in vivo accumulation rate of intraerythrocytic PvLDH peaks at 10–20 h post-invasion (late ring stage) with a median estimate of intraerythrocytic PvLDH mass at the end of the life cycle to be 9.4 × 10−3ng. We also found that the median estimate of in vivo PvLDH half-life was approximately 21.9 h. Our findings provide a foundation with which to advance our quantitative understanding of P. vivax biology and will facilitate the improvement of PvLDH-based diagnostic tools.

Bayesian Forecasting of Mortality Rates for Small Areas Using Spatiotemporal Models.

Estimation and prediction of subnational mortality rates for small areas are essential planning tools for studying health inequalities. Standard methods do not perform well when data are noisy, a typical behavior of subnational datasets. Thus, reliable estimates are difficult to obtain. I present a Bayesian hierarchical model framework for prediction of mortality rates at a small or subnational level. By combining ideas from demography and epidemiology, the classical mortality modeling framework is extended to include an additional spatial component capturing regional heterogeneity. Information is pooled across neighboring regions and smoothed over time and age. To make predictions more robust and address the issue of model selection, a Bayesian version of stacking is considered using leave-future-out validation. I apply this method to forecast mortality rates for 96 regions in Bavaria, Germany, disaggregated by age and sex. Uncertainty surrounding the forecasts is provided in terms of prediction intervals. Using posterior predictive checks, I show that the models capture the essential features and are suitable to forecast the data at hand. On held-out data, my predictions outperform those of standard models lacking a regional component.

Dynamic pricing using flexible heterogeneous sales response models

We combine nonparametric price response modeling and dynamic pricing. In particular, we model sales response for fast-moving consumer goods sold by a physical retailer using a Bayesian semiparametric approach and incorporate the price of the previous period as well as further time-dependent covariates. All nonlinear effects including the one-period lagged price dynamics are modeled via P-splines, and embedding the semiparametric model into a Hierarchical Bayesian framework enables the estimation of nonlinear heterogeneous (i.e., store-specific) immediate and lagged price effects. The nonlinear heterogeneous model specification is used for price optimization and allows the derivation of optimal price paths of brands for individual stores of retailers. In an empirical study, we demonstrate that our proposed model can provide higher expected profits compared to competing benchmark models, while at the same time not seriously suffering from boundary problems for optimized prices and sales quantities. Optimal price policies for brands are determined by a discrete dynamic programming algorithm.

Evaluating the Influence of Musical and Monetary Rewards on Decision Making through Computational Modelling.

A central question in behavioural neuroscience is how different rewards modulate learning. While the role of monetary rewards is well-studied in decision-making research, the influence of abstract rewards like music remains poorly understood. This study investigated the dissociable effects of these two reward types on decision making. Forty participants completed two decision-making tasks, each characterised by probabilistic associations between stimuli and rewards, with probabilities changing over time to reflect environmental volatility. In each task, choices were reinforced either by monetary outcomes (win/lose) or by the endings of musical melodies (consonant/dissonant). We applied the Hierarchical Gaussian Filter, a validated hierarchical Bayesian framework, to model learning under these two conditions. Bayesian statistics provided evidence for similar learning patterns across both reward types, suggesting individuals' similar adaptability. However, within the musical task, individual preferences for consonance over dissonance explained some aspects of learning. Specifically, correlation analyses indicated that participants more tolerant of dissonance behaved more stochastically in their belief-to-response mappings and were less likely to choose the response associated with the current prediction for a consonant ending, driven by higher volatility estimates. By contrast, participants averse to dissonance showed increased tonic volatility, leading to larger updates in reward tendency beliefs.

Shared Differential Clustering across Single-cell RNA Sequencing Datasets with the Hierarchical Dirichlet Process

Single-cell RNA sequencing (scRNA-seq) is a powerful technology that allows researchers to understand gene expression patterns at the single-cell level and uncover the heterogeneous nature of cells. Clustering is an important tool in scRNA-seq analysis to discover groups of cells with similar gene expression patterns and identify potential cell types. Integration of multiple scRNA-seq datasets is a pressing challenge, and in this direction, a novel model is developed to extend clustering methods to appropriately combine inference across multiple datasets. The model simultaneously addresses normalization to deal with the inherent noise and uncertainty in scRNA-seq, infers cell types, and integrates multiple datasets for shared clustering in principled manner through a hierarchical Bayesian framework. A Gibbs sampler is developed that copes with the high-dimensionality of scRNA-seq through consensus clustering. The methodological developments are driven by experimental data from embryonic cells, with the aim of understanding the role of PAX6 in prenatal development, and more specifically how cell-subtypes and their proportions change when knocking out this factor.

Land use drives differential resource selection by African elephants in the Greater Mara Ecosystem, Kenya

Understanding drivers of space use by African elephants is critical to their conservation and management, particularly given their large home-ranges, extensive resource requirements, ecological role as ecosystem engineers, involvement in human-elephant conflict and as a target species for ivory poaching. In this study we investigated resource selection by elephants inhabiting the Greater Mara Ecosystem in Southwestern Kenya in relation to three distinct but spatially contiguous management zones: (i) the government protected Maasai Mara National Reserve (ii) community-owned wildlife conservancies, and (iii) elephant range outside any formal wildlife protected area. We combined GPS tracking data from 49 elephants with spatial covariate information to compare elephant selection across these management zones using a hierarchical Bayesian framework, providing insight regarding how human activities structure elephant spatial behavior. We also contrasted differences in selection by zone across several data strata: sex, season and time-of-day. Our results showed that the strongest selection by elephants was for closed-canopy forest and the strongest avoidance was for open-cover, but that selection behavior varied significantly by management zone and selection for cover was accentuated in human-dominated areas. When contrasting selection parameters according to strata, variability in selection parameter values reduced along a protection gradient whereby elephants tended to behave more similarly (limited plasticity) in the human dominated, unprotected zone and more variably (greater plasticity) in the protected reserve. However, avoidance of slope was consistent across all zones. Differences in selection behavior was greatest between sexes, followed by time-of-day, then management zone and finally season (where seasonal selection showed the least differentiation of the contrasts assessed). By contrasting selection coefficients across strata, our analysis quantifies behavioural switching related to human presence and impact displayed by a cognitively advanced megaherbivore. Our study broadens the knowledge base about the movement ecology of African elephants and builds our capacity for both management and conservation.

Nurse-led self-care interventions for older adults with multiple chronic conditions: A protocol for a systematic review and network meta-analysis.

The importance of self-care interventions is progressively recognized, marked by a rapidly aging population that results in growing demands on costly tertiary and institutional care services, placing substantial pressure on health and social care systems. Against this background, in this study, we will provide evidence for the impact of self-care interventions on health-related quality of life by focusing on interventions led by nurses who can employ integrated approaches. Several databases, including Ovid MEDLINE, Cochrane, CINAHL, Embase, and PubMed, will be searched along with gray literature to prevent biased results. There will be no time restrictions, and all literature with English abstracts will be included. Using the Template for Intervention Description and Replication framework, we will evaluate intervention characteristics. The primary outcome will be health-related quality of life, and the secondary outcomes will be symptom burden, physical function, and cost-effectiveness. Pairwise and network meta-analyses will be performed, and network geometry and the surface under the cumulative ranking curve will be used to determine which type of nurse-led self-care intervention is best for health-related quality of life for older adults with multiple chronic conditions. We will secure study quality through PRISMA, AMSTAR 2, RoB 2.0, and the GRADE checklist. To ensure the quality of network meta-analysis, similarity will be checked based on the PICO framework. The consistency of the network meta-analysis results will be checked to ensure transitivity by using the Bayesian hierarchical framework. The study protocol is registered with PROSPERO (CRD42022365467).

Bayesian Fusion of Degradation and Failure Time Data for Reliability Assessment of Industrial Equipment Considering Individual Differences

In the field of industrial equipment reliability assessment, dependency on either degradation or failure time data is common. However, practical applications often reveal that single-type reliability data for certain industrial equipment are insufficient for a comprehensive assessment. This paper introduces a Bayesian-fusion-based methodology to enhance the reliability assessment of industrial equipment. Operating within the hierarchical Bayesian framework, the method innovatively combines the Wiener process with available degradation and failure time data. It further integrates a random effects model to capture individual differences among equipment units. The robustness and applicability of this proposed method are substantiated through an in-depth case study analysis.

Drought-related wildfire accounts for one-third of the forest wildfires in subtropical China

The threat of wildfires to human safety and ecosystem services is widely known to increase due to climate change characterized by drought. In subtropical China, seasonal drought during the non-monsoon period contributes to increase of fire risk. However, both the relationship between drought and wildfire and the extent of drought effect are poorly understood. Based on satellite active fire detections and monthly drought metrics, we quantified the impact of drought on forest wildfire frequency in subtropical China during 2001–2020, using a Bayesian hierarchical modelling framework. We observed excess wildfire, defined here as above normal wildfire activity level, is significantly correlated to drought across subtropical China, and drought-wildfire relationship showed threshold behavior that the probability of excess wildfire increased significantly when the vapor pressure deficit (VPD) crossed 0.24 standard deviations of its long-term average. From 2001 to 2020, excess wildfire caused by drought accounted for approximately 31 % of the total number of forest fire points during the fire season (November to May). The drought-related excess wildfire was concentrated in 2007–2010 and 2014, and their spatial distribution showed a dipole between the western and eastern subtropical China in individual years. Moreover, we evaluated the spatiotemporal sensitivity of wildfire to drought by Bayesian model coefficients. There was higher sensitivity than average level of the entire region in 54 % of subtropical China, especially in northwestern Yunnan. Our results indicate that drought plays a dominant role on wildfire regime at the regional scale in subtropical China. As the drying trend intensifies in the future, wildfires will show different patterns due to the large differences in the sensitivity of wildfire to drought in subtropical China.