Machine Learning: Making Its Unique Capabilities Work for Behavioral Programming

Summary:

The ACCELERATE project uses machine learning to assist health programmers in developing cross-cutting, behaviorally-focused strategies. Machine learning (ML) is the scientific study of algorithms and statistical models that computer systems use to perform a specific task without using explicit instructions, relying on patterns and inference instead. It is seen as a subset of artificial intelligence. So how can ML help behavioral programming? Imagine eight to twelve behaviors with 64 to 96 different critical factors affecting adoption of these behaviors; some factors in common, some factors unique to a particular behavior. It is not only time consuming, but difficult to synthesize this large set of factors for cross-cutting behavioral programming. The ACCELERATE project has leveraged ML to summarize research done on factors influencing a program's priority behaviors to identify patterns and commonalities, ultimately assisting programmers with strategy development. The ML Summarize Tool allows health programmers to quickly deal with several behaviors at once and synthesize qualitative data on factors, supporting actors and strategies influencing these behaviors to develop a strong behaviorally-focused strategy. The ML Summarize Tool has been used successfully to create both a broad development strategy as well as more focused project strategy.

Background/Objectives

To pursue a behaviorally-focused strategy requires analyzing factors influencing priority behaviors using qualitative data and then analyzing the factors often across multiple behaviors to find similarities and differences important to draw out in the strategy. ML optimizes time, and resources necessary for this analysis. It allows health programmers to focus on the area where their expertise is most needed, on the art of behavior change - interpreting sets of data in ways that make sense in their context. ML allows programmers to look at set of behaviors together in different ways quickly assessing the most effective way to address them.

Description Of Intervention And/or Methods/Design

ACCELERATE has created an online ML-based tool, the Summarize Tool, that is coded to pull in qualitative data from previously analyzed priority behaviors. This previous analysis is organized by element looking at factors, supporting actors and possible strategies to encourage adoption for each priority behavior. Each element is sorted by the machine according to a specific behavioral-typology on which the machine has been trained. For example, factors might be sorted and grouped first by accessibility, then by cost as an aspect of accessibility. The ML-based Summarize Tool can pull the information from any number of analyzed behaviors and group factor, supporting actor and strategy information according to the behavioral typology algorithms on which it has been trained to use. As the tool gathers more information, it learns to improve its sorting and grouping ability. The tool also allows for merging of information once appropriate groupings have been determined.

Results/Lessons Learned

ACCELERATE's ML-based Summarize Tool has been used by all technical offices in the USAID Ghana Mission, the USAID West Africa Regional Health Office (WARHO), and the WASH for Health (W4H) project in Ghana.. With the USAID Ghana Mission, use of this tool culminated in a fully-integrated, behavior-focused CDCS Results Framework with all teams aligned to common behavioral goals across all technical areas. With WARHO, use of the tool allowed the team to identify factors that cut across priority health, development, and WARHO team behaviors in less than half a day and apply this thinking to a behaviorally-focused five-year sub-strategy. With W4H, use of the tool allowed the team to easily add new behaviors to an existing strategy late in the project.

Discussion/Implications For The Field

Using an ML tool saves time. It allows programmers to process large sets of qualitative data critical to ensuring behaviors are adequately represented in less than one minute, instead of the several hours it would take to accomplish this manually. But more importantly, ML provides the space and time for critical thinking; critical thinking essential to a program that will achieve behavioral outcomes. ML should be used to take behavioral programming to the next level by simplifying the complex process of analyzing multiple behaviors, often across multiple sectors, into one cohesive strategy.

Abstract submitted by:

Lynne Cogswell - The Manoff Group

Christina Wakefield - The Manoff Group

Kristen LaFleche - The Manoff Group