How the tools work
The research evidence, the exact scoring logic, and which parts are data-driven versus design choices.
Summary
The XR Adoption Readiness Wizard and Predictive Rule Flowchart are exploratory decision-support prototypes informed by data-mining analyses of 161 AAM-accredited museums in the U.S. East North Central region. Institutional indicators were selected based on observed adoption patterns involving educational programming, digital maturity, admission policy, and institutional characteristics. The wizard applies a heuristic weighted scoring system, while the flowchart uses sequential decision rules supplemented by three capacity indicators. Scoring weights, readiness thresholds, and booster conditions were set as provisional design parameters rather than statistically optimized coefficients. The tools therefore provide directional institutional self-assessments and planning recommendations, not validated predictions of future XR adoption.
Dataset
| Sample | 161 museums accredited by the American Alliance of Museums (AAM) |
|---|---|
| Region | East North Central Division: Illinois, Indiana, Michigan, Ohio, Wisconsin |
| Primary source | AAM accreditation directory |
| Collection | Ethical, Octoparse-assisted web scraping of the directory, manual verification, and institutional websites |
| Additional sources | IMLS Museum Universe Data File (2018), institutional records |
| Outcome | XR adoption (yes / no) |
| Variables | Establishment year, state, museum type, affiliation, facility size, annual visitors, paid admission, online and on-site education, website and social media, digitized collections, research department, XR modality |
Adoption and most presence/absence indicators were coded as binary. Establishment year was numeric. State, type, affiliation, size, and visitor counts were categorical.
Findings
The initial exploration used descriptive visualization and decision-tree classification. A later extension added balanced logistic regression with stratified K-fold cross-validation, K-Modes and Ward hierarchical clustering, and Apriori association-rule mining.
Odds ratios (logistic regression)
| Predictor | Odds ratio | Direction |
|---|---|---|
| History museum | 3.52 | Positive |
| Online educational programs | 3.16 | Positive |
| Paid admission | 2.88 | Positive |
| Located in Michigan | 0.46 | Negative |
| Natural history museum | 0.43 | Negative |
| Botanical garden | 0.41 | Negative |
| In-museum-only programs | 0.22 | Negative |
Coefficients and significance
| Predictor | Coefficient | p-value |
|---|---|---|
| Establishment date | 0.016 | 0.026 |
| Paid admission | 1.292 | 0.090 |
| Annual visitors | 0.500 | 0.157 |
| Online education | 0.887 | 0.248 |
| Facility size | 0.391 | 0.448 |
| Digitized collections | −0.243 | 0.737 |
| Research department | 0.193 | 0.758 |
Overall likelihood-ratio p = 0.0004, pseudo-R² = 0.25, balanced accuracy 79%. Mean Cohen's kappa of 0.10 indicates weak generalization. These are associations, not causal effects, and the 79% figure describes the regression model, not the two interactive tools.
Clusters and association rules
Four institutional clusters showed XR adoption rates from 0% to 35.6%. Association-rule mining found 56 XR-related rules. The strongest (lift 5.16) combined Ohio, online education, digitized collections, and a research department.
Wizard scoring
| Question | Weight | Evidence |
|---|---|---|
| Q1 Active education program | 3 | Data-backed |
| Q2 Active website and social media | 3 | Data-backed |
| Q3 Digitized collections | 2 | Weak signal |
| Q4 Digital/IT staff or partner | 2 | Planning factor |
| Q5 Budget or grant pathway | 2 | Planning factor |
| Q6 On-site education | 1 | Weak signal |
| Q7 Online education | 1 | Data-backed |
| Q8 Fragile, demolished, or hard-to-access heritage | 1 | Planning factor |
| Q9 Institution size (Small 0 · Medium 0.5 · Large 1) | 1 | Weak signal |
| Total | 16 |
The weights (3/2/1) and the thresholds (0.45/0.75) were set by hand. They give education and digital presence the most weight, then digitization and resources, then context. No calibration or regression produced these exact values.
Flowchart rules
Main path
Boosters
Q5 on-site education, Q6 digitized collections, Q7 research department.
Two or three Yes answers raise the baseline by exactly one level. One booster changes nothing. Boosters apply once the main path has reached an outcome.
The order of questions follows the strongest observed patterns. The "2+ boosters" rule, the one-level increase, the ~1970 founding cutoff, and the 20-year modernization alternative are heuristic design choices.
Evidence vs. assumptions
| Item | Status |
|---|---|
| 161 museums, 26 adopters | Empirical result |
| Odds ratios and p-values | Statistical estimates |
| 79% balanced accuracy | Regression model only |
| Choice of questions | Research-informed, partly heuristic |
| Flowchart order and branches | Research-informed, manually designed |
| Wizard weights 3/2/1 | Design assumption |
| Thresholds 0.45 and 0.75 | Design assumption |
| 2+ boosters, one-level increase | Design assumption |
| ~1970 and 20-year cutoffs | Design assumption |
| Pilot recommendations | Design guidance, not validated predictions |
| Accuracy of the wizard or flowchart | Not yet established |
Limitations
- The dataset covers AAM-accredited museums in one region. It is not representative of all U.S. museums.
- The regression metrics do not measure the accuracy of the two interactive tools.
- Weights, thresholds, and booster rules have not been validated on held-out data. Future work should examine the confusion matrix, sensitivity, specificity, calibration, and class imbalance before treating the tools as predictive instruments.
- Some analyses report slightly different counts (for example 25 vs. 26 positive cases) due to modeling exclusions and missing data.
Credits
Immersive Technology for Heritage: XR Adoption in U.S. Cultural Museums. Presented at the APT International 2025 Conference, Providence, Rhode Island, session CS3.5.
Sepehr (Sep) Vaez Afshar, PhD student in Informatics, University of Illinois Urbana-Champaign. Faculty advisor: Prof. Laura Shackelford. APT COF mentor: Caroline Alderson.
This is an exploratory, evolving tool. Feedback from museum professionals is welcome.