Table 2. Overview of enhanced and novel insight generation methods as part of the CUBES toolkit.

Method	Primary insight gain	Most testable CUBES components	Method type	Advantages	Disadvantages
Descriptive
Journey mapping	Tracking experiences and influencers over time	Stages of change, beliefs, emotions, influencers and channels	Qualitative	Mapping experience, influencers drivers over time	Self-report
Observation ■   Time-and-motion ■   Infrastructure audits ■   ‘Structured immersive observation’ (SIO)	Systematically tracking practices/ duration (time and motion), infrastructure and supplies (audits), and CUBES- structured contextual drivers (SIO)	Contextual drivers: structural, systems and processes; behaviors observed	Quantitative	Behaviors and contextual drivers and barriers can be measured in their natural environment, in a standardized and replicable way	Observed participants and researchers are prone to behavioral or recording biases, respectively.
Enhanced surveys
■   Driver-structured surveys	Using CUBES as checklist aids systematic capture of enablers, barriers, influencers, and stages of change	All	Quantitative	Holistic overview of all potential drivers possible in one dataset per respondent	Not all drivers are equally well captured by self-report
■   Informal confidential voting interview (ICVI), polling-booth surveys	Adding anonymized components encourages responses on sensitive issues	Social norms, beliefs	Quantitative, ICVI also qualitative	Greater disclosure on sensitive issues	Yes/no response format leaves no room to explore; anonymous data can only be analyzed in aggregate
■   Standardized scales	Testing perceptual drivers with validated, standardized tools (e.g. self- efficacy, risk propensity, personality)	Beliefs, personality	Quantitative	Ready-made aids to assessing perceptual drivers and barriers	Prone to self-report bias
Standardized patients	Tracking behaviors, context, and interactions through simulated ‘patients’ with a set of standardized characteristics	Contextual drivers: structural, systems and processes; behaviors observed	Quantitative, qualitative components	Standardization allows for comparability, realistic setting and covert data collection for realism	On its own, is mostly limited to ‘what’ data and cannot explore drivers for practices.
Social network analysis	Revealing direction and strength of relationships in a system	Influencers, social norms	Qualitative or quantitative	Versatile (qualitative or quantitative), useable for networks of any size and type, unique method of identifying influential targets for potential intervention	Network modeling can only investigate limited drivers and barriers in one network.
Leveraging ‘passive’ datasets	Generating insights from sensor, mobile phone, satellite, GPS, social media, and other databases, with no direct customer interaction	Different, depending on dataset	Quantitative	Large-scale existing datasets can be tapped and integrated with other research methods, ‘bird’s- eye’ view of context possible	Passive nature means no opportunity to probe; existing datasets may not focus on key customer groups
’ In vitro’ experimental
Discrete choice experiments	Participants make repeated choices between a set of options whose attributes are systematically varied, in order to uncover which attributes are most important	All, least useful for biases	Quantitative	Quick to develop, test, and analyze. Participants do not have to explain ‘reasons why’, which are inferred from choices	Correlation of hypothetical with real- world choices is difficult to predict. Providing response options that clearly represent distinct drivers and barriers is not trivial
Decision games	Gamified, social experiment version of a discrete choice experiment	All, least useful for biases	Quantitative and/or qualitative	Gamification increases engagement, asking about what other participants select instead of own choices circumvents some respondent biases	Same as discrete choice experiments; qualitative approach is difficult to interpret
Implicit attitude tests	Using reaction time in response to tasks and other measurements to determine whether participant sees concepts as related or not	Biases	Quantitative	Unique method to assess strong biases inaccessible to self-report or observation	Method not well tested in low-resource settings, correlation of output and behavior not obvious, each test can only test a limited number of associations
Simulated
‘What-if’ simulations	Modeling simulated decision-making or outcomes in response to changing parameters in complex systems	All can be simulated	n/a	Unlimited permutations of changes (‘what-if scenarios’) in a complex system can be modelled	Any model will only be as good as the input data (which does require field-level input), highly specialized skills required