Behavior Change Fast and Slow: Changing Multiple Key Behaviors a Long-Term Proposition?

Lukewarm Findings So Far

The separate midline evaluation (at 20 months) found a substantial proportion of the population reporting having heard the spots and dramas. And compared with the control areas, there were greater improvements in some key reported behaviors including saving money during pregnancy and care seeking/treatment for some conditions (diarrhea and possible pneumonia, two of the three leading causes of child death in Burkina Faso). But notably there were not significant comparative improvements in a variety of other behaviors including habitual behaviors such as hand washing or exclusive breastfeeding. Some support for impact came from a positive association between reported behavior change and exposure to “spot” messages though not to the dramas. And some corroboration of an effect on care seeking came from a marked relative increase in service utilization occurring in government primary care sites. The authors posit that care seeking may be easier to influence than habitual behaviors in this context, perhaps partly because onset of illness is more pressing and demanding of action. Conversely, habitual behaviors may be more entrenched and have cultural and structural limitations. It would be helpful to have some complementary qualitative evidence from the target population to see if their impressions were consistent with this hypothesis.

Might the True Impact Actually Have Been Greater Than Measured?

Very possibly. The intervention area actually showed marked improvements in quite a number of behaviors, including both habitual and care seeking ones, but so too did the control areas (Table). So the evaluation properly compared the changes in both to see where improvements in the intervention area might be greater—a “difference-in-difference” analysis—which produced the more muted results. But actually the randomized allocation may not have ensured comparability of the intervention and control arms. There were only 7 clusters in each arm, and as it turned out there were important differences between the 2 arms. Intervention areas were poorer, had a higher proportion of Muslim population, and were farther from health facilities. While the evaluators adjusted for these measured factors in their analysis, such adjustment does not ensure that there is no remaining confounding, since unmeasured differences may remain. Moreover, it appears likely there could have been some “contamination” of messaging into the control areas; it was later learned that in one area there was overlap of radio coverage. Also, a fairly high proportion of those in the control areas reported they recognized spots when played to them (although it seems very possible this may reflect courtesy bias or confusion with other radio messages). Yet another source potentially attenuating a measured benefit were the numerous health-promoting activities carried out in both intervention and control areas by other health programs during this time, including a very successful malaria bed-net distribution program. Finally, the baseline and midline surveys were carried out in different seasons of the year. All of these factors contributed to “noise” that could possibly have washed out to some extent a true effect of the intervention.

What Else May Have Limited Impact?

First, this was a mass media-only intervention. Behavior change efforts through a combination of methods and channels have been found to be more effective in some settings. And those who are most disadvantaged and could benefit most from behavior change may have had the least exposure to the mass media. Second, addressing multiple behaviors may present some limitations. Clearly each behavior does not receive nearly as much emphasis as it would in an intensive single-behavior approach, and although the intervention changed the theme of the spots weekly, it’s possible multiple, different messages might be confusing to some listeners. Next, in order to have substantial increased utilization of primary health services, care seeking needs to be matched by availability and quality of those services. Although the local primary health system was able to respond to some extent to the increase in demand, its capacity to handle increased demand must have been constrained in this resource-poor environment. Lastly, it may simply be that in this context, some behaviors are inherently resistant to this kind of behavior change effort. Or more time might be needed. Fortunately, the intervention continued for another 15 months, ending in January 2015. The endline analysis should tell whether indeed further shift in behavior occurred, and to answer the ultimate question of whether there was an impact on actual mortality.

The good news though is that in both the intervention and the control areas, it appears that a number of key behaviors have been improving, perhaps in part because of the multiple other health intervention activities carried out in the overall area over time. It is useful to assess the impact of discrete behavior change interventions, but improving behavior over the long haul, probably resulting from multiple sources, seems the most reasonable approach for the most impact toward our global health goals. –Global Health: Science and Practice