Supplemental Material


Table 1: Dispositif STOP Web Traffic Data for 2025
Week (2025) Number of visitors
1 32
2 67
3 77
4 63
5 96
6 83
7 132
8 163
9 230
10 149
11 176
12 220
13 192
14 215
15 195
16 169
17 162
18 125
19 190
20 144
Table 2: Impact of VPN-Mediated Traffic on Experimental Cohort Assignment and Effect Estimation. Because sampling is based on Query-Level Geolocation (IP Address), a user’s physical location (Origin) may differ from their analyzed location (VPN Exit).
Physical Origin VPN Exit (IP) Analysis Group Intervention Received Likely Impact on Effect Size
Treated Region\(_{A}\)
(e.g., USA)
Treated Region\(_{B}\)
(e.g., UK)
Treatment New Onebox\(_{B}\) Neutral / Underestimation. User is analyzed as treated originating from treated Region B. However, users may not interact with the Onebox because it is localized for Region B (e.g., wrong phone numbers or translations), potentially depressing help-seeking rates despite successful deterrence.
Treated Region\(_{A}\)
(e.g., USA)
Convenience Comparison Region
(e.g., Norway)
Convenience Comparison Legacy / None Neutral / Dilution. Target population is attributed to the Convenience Comparison group. They receive no intervention; their continued searching is expected and does not affect the Treatment stats.
Convenience Comparison Region
(e.g., Norway)
Treated Region\(_{A}\)
(e.g., USA)
Treatment New Onebox\(_{A}\) Neutral / Underestimation. A convenience comparison group user is attributed to the treatment dataset. If they receive the warning, the therapeutic resources offered are for the wrong region (e.g., US hotlines). If the user desists after encountering the new Onebox, they do not inflate the success rate. If this lack of relevant local help reduces the user’s ability to desist, it inflates the treatment group’s failure rate.
Convenience Comparison Region\(_{A}\)
(e.g., Norway)
Convenience Comparison Region\(_{B}\)
(e.g., Sweden)
Convenience Comparison Legacy / None Neutral. User is correctly analyzed as convenience comparison.