Evidence
What we are learning
Participants describe changes in how they understand accusations and respond to them. Surveys completed before and after the course also show measurable changes in their answers.
Following the same 303 people
In four delivery groups, we compared each participant's written answers before and after the course. Fewer said sanguma was real, fewer described the person accused as dangerous, and fewer proposed violence.[18]
Preliminary findings from the 29 September 2026 analysis.
Says sanguma is real
% of the same 303 people
Describes the person accused as dangerous
% of the same 303 people
Proposes violence
% of the same 303 people
Read the figures as a table
| Written response | Before | After |
|---|---|---|
| Says sanguma is real | 121 (39.9%) | 92 (30.4%) |
| Describes the person accused as dangerous | 39 (12.9%) | 23 (7.6%) |
| Proposes violence | 8 (2.6%) | 2 (0.7%) |
Most already proposed action to protect the person accused or stop accusations and violence before the course; that broad measure changed little (74% to 75%). The clearest changes in these headline measures concern belief, perceived danger and proposed violence.[18]
What change means to participants
Noreen Thomas describes refusing an accusation after her daughter's death. In endline surveys, participants write about rejecting stories they once accepted, taking responsibility as leaders and standing with people accused of sorcery. Others report intervening when an accusation threatened to become an attack.[9, 18, 21]
These accounts give a human meaning to the changes in survey responses. They show how participants connect what they learned with decisions, responsibilities and action in their own lives.
A promising direction
Community requests for further rounds, participants' accounts and the matched survey findings together offer promising early evidence for this approach: train and support local leaders to work among their own people, helping communities question harmful beliefs and take responsibility together.[18, 21]
About the evidence
Participants answered short written questions before the first session and after the last, mostly in Tok Pisin. Answers were transcribed, reviewed and coded using written rules, with AI assistance and human checks. Names and sites were used to link each person's two forms. The main findings above concern 303 matched participants from four delivery groups with near-complete coding: 226 men, 57 women and 20 whose sex was not recorded.[18]
These are preliminary changes in written responses. Without a comparison group or longer-term follow-up, they do not establish how much change the programme caused or whether violence fell. Participant accounts are reported experiences, not independently verified outcomes.
Prepared for the SVRI Forum 2026 from the 29 September analysis. Please contact the programme team before citing these preliminary figures.
More about the analysis
The four groups are SCAA and Maunkui in Simbu, and Foursquare Lower Asaro and the Catholic Diocese of Goroka in Eastern Highlands. The analysis includes people whose before-and-after forms could be matched and coded; it does not represent everyone who attended. Coding used the original Tok Pisin answers or the writer's own English, without showing the coding model names, survey round or site.[18]
Exact paired tests gave p = 0.012 for belief, p = 0.029 for perceived danger and p = 0.031 for proposed violence. These p-values were not adjusted for multiple comparisons. The violence comparison involves only eight people at baseline and two at endline. A separate coding of English translations found the same direction of change for belief and danger; it used the same underlying surveys, rather than a new sample.[18, 28]
Consent and privacy
The documented survey procedure calls for an oral consent script in Tok Pisin explaining voluntary participation, the right to skip questions or stop, and consent through completing the form. Names were used to link the two surveys and are omitted from published survey quotations. Named contributors agreed to be named. AI models used for coding received answers without names.[18, 27]
About the patterns shown on the poster
The poster also presents an exploratory analysis that groups answers into seven patterns and estimates movement between them. These are statistical profiles, not fixed kinds of people, and not every answer in a group states everything in its label. In particular, the poster label containing “I endorse violence” does not mean that everyone in that group proposed violence. The smaller groups were less stable when the analysis was repeated. The direct before-and-after comparisons above are the main findings presented on this site.[18]