Technical Analysis

Tourism Dependence and the 2020 Shock: Full Results

Author

Bryan C. Johns

Published

November 5, 2026

Candidate Universe and Sample Definitions

Table A1. Candidate universe: 58 economies across three SIDS regions. Authority: UN-OHRLLS list of SIDS, retrieved November 2026.
sids_region OHRLLS-listed SIDS Associate Member SIDS Other island economy
AIMS 8 0 0
Caribbean 16 13 1
Pacific 15 5 0

Three fixed samples

Sample membership is defined by 2017–2019 data availability and is fixed — it does not change when the baseline window is varied. Changes in N across specifications reflect only baseline-window coverage within a fixed set of countries.

Table A2. Sample definitions and sizes.
Sample Criterion N
Candidate universe All 58 economies 58
Tourism-data sample At least one tourism indicator, 2017-2019 48
OWID shock sample OWID vspend (2017-19) + 2020 GDP growth 29
WDI shock sample WDI receipts (2017-19) + 2020 GDP growth 35
Common sample Both indicators (2017-19) + 2020 GDP growth 24
Recovery sample OWID shock sample + real GDP (2019 + later year) 29

OWID-only economies (in OWID shock sample but not common sample): BRB, CUB, PRI, TUV, VIR.

WDI-only economies (in WDI shock sample but not common sample): DMA, DOM, GRD, HTI, KNA, LCA, NRU, PLW, SGP, STP, TLS.

Analytical panel note. The candidate universe contains 58 economies; the analytical panel contains data for 57. BES (Bonaire, Sint Eustatius and Saba) is the one economy present in the candidate universe but absent from the panel: it has no coverage in any OWID dataset and no World Bank API coverage, so the full join that constructs the panel produces no rows for it. The remaining six WDI-absent economies (AIA, GLP, MTQ, MSR, COK, NIU) have at least partial OWID data and therefore appear in the panel with WDI columns missing. None of the seven WDI-absent economies qualifies for any shock or recovery sample.

Tourism Indicator Coverage

Coverage matrix

Figure 1: Data coverage matrix across the candidate universe. Green = all requested years available; yellow = partial (one year); grey = absent.

WDI receipts: end of series in this retrieval

WDI ST.INT.RCPT.XP.ZS has no observations beyond 2020 in the retrieved data. This does not appear to be a simple publication lag; it likely reflects a change in UN Tourism’s data dissemination approach following the pandemic. The pre-pandemic coverage is also uneven: 14 of 35 WDI sample members have no 2019 receipts observation, meaning their 2017–2019 baseline draws on 2017 and 2018 only. The OWID visitor spending series, by contrast, is available through 2024 with near-complete coverage.

Table A3. Key coverage differences between tourism indicators.
Indicator Latest year in retrieval 2019 available? Notes
OWID visitor spending (% GDP) 2024 Yes (all sample members) Updated 2026
WDI receipts (% exports) 2020 21 of 35 only Series ends 2020; 14/35 lack 2019 data
WDI real GDP (const. 2015 USD) 2024–2025 Yes (all sample members) 2025 preliminary for 22/29

Baseline coverage distributions

Table A4. Annual observation counts within each baseline window, restricted to fixed sample members. WDI receipts 2017-19 window has many economies with only 2 observations (2017+2018) because 2019 data is missing from the WDI series.
Sample Indicator Window 0 obs 1 obs 2 obs 3 obs 4 obs 5 obs
OWID (N=29) OWID vspend 2015-19 0 1 0 1 1 26
OWID (N=29) OWID vspend 2017-19 0 1 0 28 0 0
OWID (N=29) WDI receipts 2015-19 4 1 1 2 2 19
OWID (N=29) WDI receipts 2017-19 5 1 2 21 0 0
WDI (N=35) OWID vspend 2015-19 10 2 0 0 1 22
WDI (N=35) OWID vspend 2017-19 11 1 0 23 0 0
WDI (N=35) WDI receipts 2015-19 0 1 0 3 12 19
WDI (N=35) WDI receipts 2017-19 0 2 12 21 0 0
Common (N=24) OWID vspend 2015-19 0 1 0 0 1 22
Common (N=24) OWID vspend 2017-19 0 1 0 23 0 0
Common (N=24) WDI receipts 2015-19 0 1 0 2 2 19
Common (N=24) WDI receipts 2017-19 0 1 2 21 0 0

Indicator Comparison

Figure 2: OWID visitor spending (% GDP) vs. WDI tourism receipts (% exports), 2015–2019 averages. The two indicators are correlated but not interchangeable — they divide by different denominators and measure related but distinct economic concepts.

The Pearson correlation between the two indicators (across economies with both) is approximately 0.77, and the Spearman rank correlation is 0.86. They produce similar rankings but material differences for some economies. The WDI receipts/exports denominator amplifies the apparent dependence of export-intensive small economies, while the OWID visitor spending/GDP denominator captures the share of overall economic activity attributable to visitors.

Eight-Model Shock Analysis

Outcome: 2020 annual GDP growth (%, WDI NY.GDP.MKTP.KD.ZG).
Note: The 2015–2019 primary window was evaluated after the 2017–2019 results were already observed. No window was designated a priori. All eight results are reported.

Table A5. Complete eight-model results. Holm correction applied across all 8 raw p-values. No result is suppressed.
Model Indicator Window Sample N Slope 95% CI R2 Pearson_r Spearman_rho p_raw p_holm
1 OWID vspend 2015-19 OWID (N=29) 29 -0.308 [-0.453, -0.163] 0.414 -0.643 -0.550 0.000 0.001
2 WDI receipts 2015-19 WDI (N=35) 35 -0.120 [-0.248, 0.008] 0.100 -0.316 -0.443 0.064 0.129
3 OWID vspend 2015-19 Common (N=24) 24 -0.331 [-0.481, -0.181] 0.489 -0.699 -0.655 0.000 0.001
4 WDI receipts 2015-19 Common (N=24) 24 -0.158 [-0.285, -0.031] 0.233 -0.483 -0.480 0.017 0.053
5 OWID vspend 2017-19 OWID (N=29) 29 -0.320 [-0.465, -0.174] 0.428 -0.654 -0.564 0.000 0.001
6 WDI receipts 2017-19 WDI (N=35) 35 -0.116 [-0.242, 0.01] 0.097 -0.311 -0.432 0.069 0.129
7 OWID vspend 2017-19 Common (N=24) 24 -0.339 [-0.491, -0.187] 0.493 -0.702 -0.660 0.000 0.001
8 WDI receipts 2017-19 Common (N=24) 24 -0.160 [-0.284, -0.037] 0.248 -0.498 -0.491 0.013 0.053

Four-panel figure (2015–2019 baseline)

Figure 3: Four model specifications, 2015–2019 baseline. Two indicators × two samples. OLS line with 95% CI. Open circles = thin baseline (Kiribati, 1 obs).

Common-sample comparison

The common sample (N = 24) holds the country set constant and varies only the indicator. Any difference in results between the two panels below reflects the indicator choice, not sample composition.

Figure 4: Common sample (N=24): OWID visitor spending vs. WDI tourism receipts, same 24 economies, 2015–2019 baseline. The OWID vspend indicator produces a substantially stronger association.

Influential observations

Table A6. Most influential observation per model (Cook’s D threshold = 4/N). MDV = Maldives; TLS = Timor-Leste. The OWID association survives MDV exclusion; the WDI/TLS exclusion actually strengthens M2 and M6.
Model Most influential Cook’s D Slope (orig.) Slope (excl.) R² (orig.) R² (excl.) p (excl.)
M1 OWID/2015-19/OWID (N=29) MDV 0.392 -0.308 -0.251 0.414 0.309 0.002
M2 WDI/2015-19/WDI (N=35) TLS 0.287 -0.120 -0.141 0.100 0.224 0.005
M3 OWID/2015-19/Common (N=24) MDV 0.363 -0.331 -0.276 0.489 0.389 0.002
M4 WDI/2015-19/Common (N=24) MDV 0.339 -0.158 -0.116 0.233 0.157 0.062
M5 OWID/2017-19/OWID (N=29) MDV 0.384 -0.320 -0.263 0.428 0.331 0.001
M6 WDI/2017-19/WDI (N=35) TLS 0.324 -0.116 -0.142 0.097 0.234 0.004
M7 OWID/2017-19/Common (N=24) MDV 0.362 -0.339 -0.284 0.493 0.402 0.001
M8 WDI/2017-19/Common (N=24) MDV 0.334 -0.160 -0.120 0.248 0.173 0.048

Why Timor-Leste influences M2 and M6: Timor-Leste has moderately high WDI tourism receipts as a share of exports but experienced a modest 2020 contraction, making it a structural outlier in the WDI receipts models. Excluding it makes those models stronger, not weaker — a counter-intuitive diagnostic result worth noting.

Coverage-restricted sensitivity (≥3 baseline observations)

Table A7. Sensitivity analysis restricted to economies with ≥3 annual observations in the baseline window. S5 (WDI/2017-19/WDI) drops 14 economies (N drops from 35 to 21) because WDI 2019 receipts data is missing for many SIDS.
Sensitivity Window Indicator Sample N Slope R2 Spearman p_raw Excluded
S1 2015-19 OWID vspend OWID 28 -0.302 0.397 -0.522 0.0003 KIR
S2 2015-19 WDI receipts WDI 34 -0.116 0.092 -0.432 0.0820 KIR
S3 2015-19 OWID vspend Common 23 -0.325 0.470 -0.637 0.0003 KIR
S4 2017-19 OWID vspend OWID 28 -0.314 0.412 -0.537 0.0002 KIR
S5 2017-19 WDI receipts WDI 21 -0.143 0.180 -0.422 0.0554 DMA,DOM,GRD,HTI,KIR,KNA,LCA,MHL,NRU,PLW,PNG,SGP,STP,TLS
S6 2017-19 OWID vspend Common 23 -0.333 0.474 -0.643 0.0003 KIR

Recovery Analysis

Coverage

Table A8. Annual count of primary shock sample economies (N ≤ 29) with real GDP data. 2025 data are preliminary estimates.
year N with real GDP
2015 29
2016 29
2017 29
2018 29
2019 29
2020 29
2021 29
2022 29
2023 28
2024 28
2025 22

Trajectories

Figure 5: Real GDP recovery trajectories, 2015–2025 (index: 2019=100). Line colour = SIDS region. Bottom numbers = annual sample size.

Recovery status by 2024

Table A9. Aggregate and per-capita real GDP recovery indices (2019=100). Recovery means an index >= 100 in 2024. VIR has no 2024 data. 21/28 recovered aggregate GDP; 20/28 recovered per-capita GDP. Four economies recovered aggregate but not per-capita GDP (ATG, PNG, SYC, VUT); three recovered per-capita but not aggregate GDP (CUW, MHL, TUV).
ISO3 Economy Region 2020
(shock)		 GDP idx
2024		 GDP/cap
idx 2024		 Agg.
recovered		 Per-capita
recovered
COM Comoros AIMS 99.8 111.2 100.9 Yes Yes
CPV CapeVerde AIMS 79.2 110.1 107.8 Yes Yes
GNB Guinea-B. AIMS 102.9 124.5 111.3 Yes Yes
MDV Maldives AIMS 67.1 114.1 105.4 Yes Yes
MUS Mauritius AIMS 85.8 105.6 107.3 Yes Yes
SYC Seychelles AIMS 88.3 106.0 85.3 Yes No
ABW Aruba Caribbean 76.1 113.0 114.3 Yes Yes
ATG Antigua Caribbean 81.1 101.4 98.8 Yes No
BHS Bahamas Caribbean 79.9 111.0 109.2 Yes Yes
BMU Bermuda Caribbean 93.0 110.1 109.2 Yes Yes
BRB Barbados Caribbean 82.6 108.9 108.4 Yes Yes
CUB Cuba Caribbean 89.1 89.0 90.9 No No
CUW Curacao Caribbean 81.6 99.4 100.4 No Yes
CYM Cayman Is. Caribbean 95.0 115.6 104.5 Yes Yes
PRI Puerto Rico Caribbean 95.8 102.9 106.1 Yes Yes
SUR Suriname Caribbean 84.0 87.5 83.5 No No
SXM Sint Maarten Caribbean 86.7 110.4 102.8 Yes Yes
TTO Trinidad Caribbean 91.2 95.0 94.7 No No
VCT St.Vincent Caribbean 95.7 110.5 114.4 Yes Yes
VIR US VI Caribbean 98.4 NA NA NA NA
FJI Fiji Pacific 82.8 105.7 104.1 Yes Yes
KIR Kiribati Pacific 98.7 121.5 111.9 Yes Yes
MHL Marshall Is. Pacific 97.2 96.4 113.0 No Yes
PNG Papua NG Pacific 96.8 109.8 99.8 Yes No
SLB Solomon Is. Pacific 96.6 99.5 88.4 No No
TON Tonga Pacific 101.8 103.2 104.7 Yes Yes
TUV Tuvalu Pacific 97.0 92.4 101.4 No Yes
VUT Vanuatu Pacific 94.3 108.0 96.2 Yes No
WSM Samoa Pacific 96.9 113.6 109.3 Yes Yes

Recovery vs. tourism dependence

Figure 6: Pre-pandemic visitor spending vs. 2024 real GDP index (2019=100). No statistically significant association (Pearson r = 0.21, p = 0.27, N = 28).

The absence of a significant association between pre-pandemic tourism dependence and the 2024 recovery index (Figure 6) does not mean tourism dependence was harmless or that the initial shock was inconsequential for these economies. Rather, the relationship between pre-pandemic exposure and longer-term economic recovery is not straightforward. Although tourism-dependent economies experienced particularly severe contractions in 2020, their subsequent recovery trajectories varied. By 2024, the degree of pre-pandemic tourism dependence alone did not explain which economies had regained their 2019 aggregate GDP levels.

Per-capita recovery provides a more nuanced picture. By 2024, 20 of 28 economies had recovered their 2019 real GDP per capita, compared with 21 of 28 for aggregate GDP. The difference illustrates why aggregate output alone can overstate the extent of economic recovery: an economy may regain its total output while remaining below its pre-pandemic level on a per-person basis. Population growth is one possible contributor to this divergence, although this analysis does not isolate its effects.

See Data Dictionary for source documentation and Methodology for research design.