CONTINUOUS
Global healthy life expectancy
177 map units · 2000–2020
From data to interpretable evidence.
A practical guide to BSTVC Desktop. Follow a complete workflow for continuous, binary, or count outcomes.
- 1Prepare dataStructure your data and check requirements.
- 2Check spatial orderEnsure the spatial units match your data.
- 3Fit & exportRun models and export results for interpretation.
BINARY
COVID-19 occurrence
31 provinces · 12 months
COUNT
Acute hepatitis B
5 states · 2012–2023
Choose your starting point
New to BSTVC? Read Install & prepare, then follow one case from start to finish. Already have a result workbook? Go to Understand results or R visualization. If an import or model run fails, use the troubleshooting checklist.
Each case includes the original data, all required map components, an English companion README, and links back to this guide. Binary and count packages also contain ready-to-import panel CSVs, reshaped without imputation, transformation, or changing measured values. The full conversion procedure remains available for learning and checking.
What BSTVC helps you examine
BSTVC Desktop is a graphical environment for Bayesian analysis of relationships that vary across space and time. It combines local coefficients, global spatiotemporal contribution, and prediction within a common workflow. Standard desktop model fitting does not require R programming; R is optional for customized figures.
| Research task | Question | Outputs |
|---|---|---|
| Local interpretation | How do associations vary by location or period? | Spatial coefficients (SCs), temporal coefficients (TCs), credible intervals |
| Global contribution | Which modeled components contribute to spatiotemporal variation? | Spatiotemporal Variance Partitioning Index (STVPI) |
| Dynamic prediction | What values does the model estimate for observed or missing outcomes? | Predictions, smoothing, imputation, and uncertainty |
The methodological framework
BSTVC stands for Bayesian Spatiotemporally Varying Coefficients. The model represents relationships that can change across locations and periods within one Bayesian hierarchical framework. The source manual calls this a whole-map framework: local estimates are fitted together, with spatial and temporal structure, rather than obtained by fitting an unrelated regression separately to every location.
Two ideas guide the analysis. Autocorrelation represents dependence among nearby places or neighboring periods. Nonstationarity allows the relationship between a predictor and the response to differ across those places or periods. Accounting for dependence and estimating variation in relationships answer different questions; a spatially patterned outcome alone does not demonstrate a spatially varying predictor effect.
| Approach | Scope | Practical distinction |
|---|---|---|
| BSTVC | Repeated observations across space and time | Use a location–time panel and interpret temporal and spatial coefficient outputs. |
| BSVC | Spatial observations from one cross-section | Use the spatial modeling module for a single time point; repeated years should not be treated as independent map units. |
| STVPI | Contribution based on the fitted model | Summarize relative spatial, temporal, and combined contributions, with posterior uncertainty. |
The methods cited in the Chinese manual discuss the STVI, STIVI, STVC, and STIVC model family. These are methodological variants, not interchangeable labels for the three response types in this tutorial. The walkthroughs use the documented desktop controls; consult the method references before selecting or extending a statistical specification.
The framework can support interpretation, smoothing, and prediction, including model-based estimates for missing outcomes. Whether it improves prediction or yields useful evidence depends on the data, model assumptions, and validation. Posterior uncertainty describes uncertainty under the fitted model; it does not remove bias from incorrect data alignment, confounding, or an unsuitable neighborhood.
For methodological details and programmable extensions, use BSTVC-R. To read the desktop outputs, continue to Understand results.
About this guide
This English edition consolidates the supplied July 2026 desktop manual with the fuller Chinese online manual, reorganizes their workflows, and checks case filenames, identifiers, dimensions, and missingness against the accompanying data. It targets the English Windows desktop release v2026.8.1. Interface illustrations originate from the supplied manual; historical figures illustrate how to read outputs and are not a new reproduction of model fitting.
The spatiotemporal interpretability research organization and BSTVC development team maintains the BSTVC ecosystem. For programmable workflows, see BSTVC-R. Guide revised: 2 October 2026.