Whose definition of violence is the dataset using?

Rhoda, I think this is an important measurement issue: standardization should not come at the expense of cultural validity, but adaptation can also affect comparability if it changes the construct being measured.

My preference would be to retain a validated core set of behaviour-specific items so that the primary estimates remain comparable with established datasets, while using qualitative work, cognitive interviewing and community input to identify locally meaningful forms of harm that the core instrument may miss. WHO guidance on VAW research supports this kind of contextual adaptation and pre-testing while recognizing the importance of methodological consistency for comparability.

For harms such as economic coercion, spiritual abuse, control involving children or community ostracism, I would not automatically collapse them into an “other” category. If they are conceptually relevant to the study's definition of violence, they should ideally be documented explicitly, even if they are analyzed separately from the standardized core outcome. That preserves the woman's experience rather than forcing it into a category that may not fit.

The methodological challenge is then to distinguish between the standardized outcome used for cross-study comparison and locally identified forms of harm used to improve contextual understanding. If new items are incorporated into the primary definition, their development, cognitive testing, translation/adaptation and psychometric properties should be documented, and the effect on comparability should be acknowledged.

So perhaps the question is not simply “Do we adapt the instrument or preserve comparability?” but rather: “How can we preserve a comparable core measure while creating sufficient space for women to describe forms of harm that the core instrument does not capture?” That seems to offer both methodological rigour and respect for women's own accounts.

Dr. Magoba's core-plus-contextual-layer solution
is methodologically sound, and I'd push on one implication of it: a "core
comparable measure" is never actually neutral — it's a frozen snapshot of
whichever context first validated it. Treating it as the stable anchor and
everything else as "additional context" quietly ranks whose
definition of violence counts as the default and whose counts as local color.
That's a defensible methodological choice, but it's a choice, not a neutral
baseline, and it's worth naming as one rather than treating comparability as
self-evidently the senior priority.

The forms Rhoda names — economic coercion,
spiritual abuse, control through children, community ostracism — aren't edge
cases the instrument happened to miss. In many of the settings I've worked in,
they're the primary mode through which control operates; physical
violence is often the exception that gets recorded precisely because it's the
only form the instrument was built to see. If those harms consistently get
folded into "other" or dropped, the dataset isn't measuring less
violence in that setting — it's measuring a narrower slice of it and reporting
the slice as the whole.

Where I'd differ slightly from Dr. Magoba's
framing: rather than asking how to preserve the comparable core while making
room for the rest, I'd ask what we lose by treating comparability as the fixed
point at all. Cross-context comparison is genuinely valuable — but a dataset
that's comparable and wrong about what's happening on the ground isn't a
stronger dataset, it's a more exportable one. Sometimes the honest answer is
that full comparability with an instrument built elsewhere isn't achievable
without also inheriting that instrument's blind spots.

@Rhoda Nakhosi When women describe context-specific forms of violence such as spiritual abuse, economic coercion, or community ostracism standardized Violence Against Women and Girls (VAWG) instruments (like the WHO Multi-country Study or DHS domestic violence modules) routinely fail to capture them. In practice, these accounts usually follow one of three paths: they are lost entirely because closed-ended survey screeners filter them out before they reach an enumerator; they are absorbed into generic "other" or psychological abuse categories, which strips them of their distinct structural dynamics; or they are captured purely as qualitative notes that get excluded from primary quantitative reporting.

When programs and researchers actively adapt their operational definitions to reflect locally named harms, it creates a direct tension between contextual validity and cross-dataset comparability:

• The Impact on Data Comparability: Adapting a definition by introducing local sub-scales or expanded behavioral taxonomies makes direct, 1-to-1 statistical comparisons with global datasets (like SDG 5.2 indicators or standardized DHS estimates) impossible if the denominator or overall prevalence threshold shifts. A setting that counts spiritual abuse or community-level ostracism as IPV or non-partner violence will report higher prevalence rates than a neighboring country using a strict physical/sexual/psychological tri-part model.

• The "Module Add-On" Compromise: To preserve global comparability without erasing local realities, researchers often employ a dual-structure approach. They retain the core validated cross-cultural items untouched to generate a globally comparable baseline, while appending a locally co-designed "contextual module" developed through participatory qualitative work. This allows the dataset to report both a globally benchmarked prevalence rate and a locally meaningful severity/harm profile.

• What Adapting Definitions Does to Service Delivery: Beyond the data, adapting definitions fundamentally changes who gets served. When tools explicitly operationalize behaviors like control exercised through children or land/property deprivation, referral pathways shift. Case managers and intake workers stop viewing these disclosures as "family disputes" or "economic hardship" and begin recognizing them as actionable protection risks that require legal aid, shelter access, or economic empowerment interventions.

Ultimately, prioritizing rigid global comparability over locally named harm risks creating sanitized datasets that measure what the instrument wants to see rather than the actual mechanisms of violence controlling women's lives.