David has started a discussion that I find intrinsically interesting and one that well-designed impact evaluations can help clarify: why don’t more people adopt low-cost efficacious health technologies? We may be able to think of examples in our own lives – i.e. “why don’t I take vitamins more regularly?” or “why, if diabetic, don’t I self-test my blood sugar more frequently?” These same questions also resonate for large-scale health programs in many settings. For example, the take-up of freely provided malaria bed-nets is often high but not universal. Adherence to life-saving drug regimens can be less than optimal (which in the case of tuberculosis has spurred the Directly Observed Treatment programs, or DOTS, advocated by WHO).
The underlying causes of this behavior can be complex and can vary by setting and technology. Surely the perceived net benefits of the health technology play a role in the final calculus of whether to adopt or not. But there may also be important behavioral components at play such as inconsistent time preferences. And of course even if the cost of adoption is low, liquidity constraints may prevent purchase/usage.
Preconceived notions of the use cost of the technology – for example fears of a medicine’s bad taste or other side-effects – will clearly also have an effect on the decision of whether to adopt. For example, one of the biggest deterrents around the world to more widespread use of malaria bed nets is the belief that sleeping under the net is uncomfortable and hot. I mention bed nets because it is germane to a recent study by Pascaline Dupas that explores determinants of usage of Long-Lasting Insecticide Treated Nets (LLINs), the latest technology in malaria prevention.
Pascaline conducted an LLIN adoption experiment in Western Kenya that was rolled out in two phases. In the first phase households were offered vouchers of randomly determined value to redeem for LLINs. The value of the voucher ranged from 100% (i.e. a free net) to 40% of the LLIN market price. This first phase essentially duplicates an earlier study by Dupas and Jessica Cohen and finds the same results – initial net adoption is quite high when nets are free but drops precipitously even at relative low prices.
In a second study phase, the researchers returned to the same households 12 months later and offered every household an identically valued voucher giving a moderate discount over the LLIN market price. In the second phase, the study found that the willingness to pay for an LLIN increased after the first phase subsidy program, especially for households that received the larger subsidies. Thus a temporary subsidy, by inducing households to adopt and experiment with the LLIN, increased the average willingness to pay one year later.
Of course the declared willingness to pay may not be a reliable indicator if the household perceives social pressure to tell the interviewer what “they want to hear”. But in this case the actual behavior of the households supports these stated preferences – second phase LLIN take-up was higher in households that were given the first-phase LLIN for free than in those households charged a first phase price, also suggesting a learning- by-doing effect (since usage rates were higher for this group).
This experiment suggests that, pre-usage, either the perceived costs of LLIN usage were too high or the perceived benefits were too low. Once households had some experience with the technology, they were more likely to purchase subsequent additional nets (even at the subsequent higher price).
From the policy perspective we learn that, given experimentation with the product can be an important determinant of ultimate use, the initial low subsidies enticed households with prior beliefs of high usage costs to adopt and learn that these costs weren’t as high as feared. This is encouraging for the adoption of constant-use prevention technologies like LLINs. But how applicable are these results to other technologies, such as curative technologies, and other settings?
One other technology, deworming medicine, witnessed a decline with usage after short-run subsidies. Perhaps, as Pascaline speculates, the ex-ante perceived costs of adoption in this setting were low – hence initial high adoption – but increased after side-effects of the medicine became known. Surely perceived net benefits are an important factor in health technology adoption and indicate the critical role not only of information, but of programs managing expectations of new health technologies.
Many questions remain regarding the relative importance of other behavioral factors such as inconsistent time preferences as well as how the nature of the setting and the nature of the technology interact to determine the salient barriers to adoption. Key questions for the conduct of health programs include:
- How does a program identify the constraints to adoption and usage in the particular setting at hand?
- How does a program successfully communicate the (true) benefits and costs of adoption and are there more efficient venues than self-experimentation and/or social learning?
- How does a program sustainably finance encouragements to adoption such as subsidies or commitment mechanisms?
Further developing these questions will surely involve more work in behavioral theory as well as a healthy dose of impact evaluation. I am sure that my fellow-bloggers and I will return to this issue repeatedly in the future as we have much more to learn…