TABLE 3.
Summary of computational modeling methods employed overall and by model type1 across included studies using complex systems models to study retail food environments for population health
| Overall (n = 56) | SD (n = 4) | DES (n = 2) | Networks (n = 4) | ABM (n = 13) | Hybrid (n = 10) | Microsimulation (n = 23) | |
|---|---|---|---|---|---|---|---|
| Characteristic | N (%) | N (%) | N (%) | N (%) | N (%) | N (%) | N (%) |
| Model development methods2 | |||||||
| Literature review/data sources | 53 (95) | 4 (100) | 2 (100) | 1 (25) | 13 (100) | 10 (100) | 23 (100) |
| Primary data collection | 5 (9) | 1 (25) | 0 (0) | 3 (75) | 1 (8) | 0 (0) | 0 (0) |
| Consulted experts/advisory group | 10 (18) | 2 (50) | 0 (0) | 0 (0) | 0 (0) | 2 (20) | 6 (26) |
| Engaged stakeholders in modeling | 1 (2) | 0 (0) | 0 (0) | 0 (0) | 1 (8) | 0 (0) | 0 (0) |
| Additional methods2 | |||||||
| Calibration | 20 (36) | 3 (75) | 0 (0) | 0 (0) | 5 (39) | 5 (50) | 7 (30) |
| Verification/external validation | 37 (66) | 3 (75) | 0 (0) | 1 (25) | 9 (69) | 8 (80) | 16 (70) |
| Sensitivity analyses | 40 (71) | 4 (100) | 2 (100) | 0 (0) | 9 (69) | 6 (60) | 19 (83) |
| Empirical anchoring3 | |||||||
| High | 40 (71) | 1 (25) | 2 (100) | 4 (100) | 7 (54) | 4 (40) | 22 (96) |
| Medium | 13 (23) | 3 (75) | 0 (0) | 0 (0) | 4 (31) | 5 (50) | 1 (4) |
| Low | 3 (5) | 0 (0) | 0 (0) | (0) | 2 (15) | 1 (10) | 0 (0) |
| Transparency appraisal4 | |||||||
| High | 24 (43) | 2 (50) | 0 (0) | 0 (0) | 6 (46) | 1 (10) | 15 (65) |
| Adequate | 22 (39) | 1 (25) | 1 (50) | 3 (75) | 5 (39) | 5 (50) | 7 (30) |
| Low | 10 (18) | 1 (25) | 1 (50) | 1 (25) | 2 (15) | 4 (40) | 1 (4) |
Abbreviated model types include: ABM, agent-based model; DES, discrete event simulation; Hybrid, >2 model approaches used; SD, system dynamic model.
Categories are not mutually exclusive. Multiple methods may be employed by authors in a single study; studies using >1 method are represented within each applicable category. Definitions for each method type are available in Supplemental Methods C.
Assessment based on Langellier et al. (34) definition of the degree of anchoring a model has to empirical data. Low consists of models that are primarily stylized or simplistic and do not directly link to empirical data; medium involves models that have some but not all features (e.g., parameters) linked to empirical data; and high consists of models that are entirely linked to empirical data sources. Categories are mutually exclusive.
10 items inspired by Jalali et al. (58, 59) (See Supplemental Table 1). The number of “Yes” answers was divided by the number of applicable criteria to derive a percentage score qualified as low (<70%), adequate (70–89%), or high transparency (≥90%). Categories are mutually exclusive.