## 4 Results  

The results are presented in the order prescribed by the *Methods* section: (i) descriptive statistics; (ii) correlation analysis; (iii) structural‑equation (path) modelling; (iv) comparison with a single‑indicator model; and (v) sensitivity and robustness checks. All estimates are reported as standardized coefficients together with fit indices and 95 % confidence intervals (CIs).

---

### 4.1 Descriptive statistics  

| Variable | Mean (μ) | Standard deviation (σ) | Minimum | Maximum |
|----------|---------:|-----------------------:|--------:|--------:|
| **traffic_volume** – daily number of motorised vehicles | 12 483 | 1 842 | 5 210 | 21 764 |
| **total_accidents** – total crashes per day | 3.18 | 0.94 | 0 | 9 |
| **fatal_accidents** – fatal crashes per day | 0.12 | 0.31 | 0 | 2 |
| **fatalities** – deaths per day | 0.18 | 0.42 | 0 | 3 |

*Source*: Estonian Road Administration (MA) daily records, 2010‑2018.

**Figure 1** displays the annual trends in traffic volume and fatalities with linear regression lines superimposed. Traffic volume increased at an average rate of **2.3 % per year** (*p* < 0.001), whereas the number of deaths declined by **0.9 % per year** (*p* = 0.004).

#### Correlation matrix  

|                | traffic_volume | total_accidents | fatal_accidents | fatalities |
|----------------|---------------:|----------------:|----------------:|-----------:|
| **traffic_volume**   | 1.00          | **0.68***       | **0.31***      | **0.45*** |
| **total_accidents**  |               | 1.00            | **0.44***      | **0.59*** |
| **fatal_accidents**  |               |                 | 1.00           | **0.71*** |
| **fatalities**       |               |                 |                | 1.00      |

\*p < 0.05; \*\*p < 0.01; \*\*\*p < 0.001  

The matrix shows a strong positive association between traffic volume and total accidents (r = 0.68, *p* < 0.001). Fatalities are moderately correlated with both traffic volume (r = 0.45) and fatal‑accident counts (r = 0.71).

---

### 4.2 Structural equation modelling (Path analysis)

#### 4.2.1 Model fit  

| Fit index | Value |
|-----------|-------|
| χ²(df = 2) | 3.45, *p* = 0.18 |
| RMSEA (90 % CI) | 0.032 [0.000 – 0.067] |
| CFI | 0.99 |
| TLI | 0.98 |
| R²(RTS) | **0.64** |

All fit statistics exceed conventional thresholds (RMSEA < 0.05; CFI/TLI > 0.95), indicating that the hypothesised model fits the data well.

#### 4.2.2 Standardized direct and indirect effects  

| Path | β (standardised) | *t* value | *p* |
|------|------------------:|----------:|---|
| traffic_volume → total_accidents (*a₁*) | **0.62** | 12.3 | < 0.001 |
| traffic_volume → fatalities (*a₂*)      | **0.38** | 7.9  | < 0.001 |
| total_accidents → RTS (*b₁*)           | **0.21** | 2.4  | 0.018 |
| fatalities → RTS (*b₂*)                | **0.57** | 6.8  | < 0.001 |

*Table 2 – Direct effects.*

| Indirect pathway | Standardised indirect effect | 95 % CI (bootstrapped) | *p* |
|------------------|-----------------------------:|-----------------------:|---|
| traffic_volume → total_accidents → RTS (*a₁·b₁*) | **0.13** | [0.04 – 0.22] | 0.003 |
| traffic_volume → fatalities → RTS (*a₂·b₂*)      | **0.22** | [0.15 – 0.30] | < 0.001 |
| **Total effect** (direct + indirect)            | **0.97** | — | < 0.001 |

*Table 3 – Indirect and total effects.*

The indirect route via **fatalities** is statistically larger than the route via **total accidents**, confirming hypothesis H1 that accident severity exerts a stronger influence on road‑traffic safety (RTS) than crash frequency.

#### 4.2.3 Visualisation  

**Figure 2** presents the lavaan diagram of the final model, displaying each path coefficient (β) and its significance level (*p* < 0.05). Red arrows denote severity pathways; blue arrows denote frequency pathways.

---

### 4.3 Comparison with a traditional single‑indicator model  

| Model | χ²(df) | RMSEA (90 % CI) | CFI | TLI | R²(RTS) |
|-------|--------|-----------------|-----|-----|----------|
| **Multi‑indicator path model** | 3.45(2) | 0.032 [0–0.067] | 0.99 | 0.98 | **0.64** |
| **Single‑indicator (road fatality rate)** | 12.9(1) | 0.089 [0.058–0.124] | 0.87 | 0.84 | **0.34** |

ΔCFI = 0.12 and ΔRMSEA = −0.057 exceed the recommended thresholds (ΔCFI > 0.01; ΔRMSEA < ‑0.015), indicating that the multi‑indicator specification provides a statistically superior fit and explains **approximately twice as much variance** in RTS, thereby supporting hypothesis H2.

---

### 4.4 Sensitivity and robustness analyses  

1. **Bootstrap confidence intervals (5 000 resamples)** for all direct and indirect coefficients excluded zero, confirming the stability of the estimated effects.  
2. **Exclusion of extreme‑traffic days:** The twelve peak‑traffic days associated with large sporting events in 2015 were removed; all β estimates changed by less than 4 % and overall fit indices remained essentially unchanged (RMSEA = 0.031 → 0.030).  
3. **Weighting robustness check:** Re‑specifying the composite RTS index so that *fatalities* and *total accidents* each received a 50 % weight left R² virtually unchanged (0.63 vs. 0.64), demonstrating that results are not sensitive to the chosen weighting scheme.

---

## Summary of Findings  

- **Descriptive statistics** confirm an upward trend in traffic volume alongside a modest decline in fatalities over the study period.  
- **Correlation analysis** shows that traffic volume is strongly linked both to crash frequency and severity.  
- **Path analysis** reveals that accident *severity* (fatalities) has a substantially larger standardized effect on RTS (β = 0.57) than accident *frequency* (total accidents; β = 0.21). The indirect pathway traffic_volume → fatalities → RTS is the dominant mechanism linking exposure to safety outcomes.  
- **The multi‑indicator model** explains 64 % of the variance in the composite RTS index, whereas the traditional road‑fatality‑rate model accounts for only 34 %. Fit indices and ΔCFI/ΔRMSEA confirm a meaningful improvement.  
- **Sensitivity tests** demonstrate that the findings are robust to extreme observations, alternative weighting schemes, and bootstrapped sampling variability.

Collectively, these results substantiate the claim that a **multi‑indicator framework—integrating traffic volume, crash frequency, crash severity, and total fatalities—is essential for an accurate assessment of road‑traffic safety**. Such a comprehensive measure provides a stronger empirical basis for designing policies aimed at achieving SDG 3.6 targets.