Prioritizing among impact questions

1. Prioritizing among impact questions Module 3.3

2. How important will the results be? • Is the program (or similar programs) large or expanding? • If the program works will it be cheap and easy to scale—i.e. high potential impact? • How much existing evidence is there already? • Is the question more general (theory based) or more tied to one context?

3. Can the question be answered well? • Must have good way to measure outcomes • When evaluating a HIV education program it is critical to get measures that go beyond self reported attitudes and behavior or results may mislead • Need sufficient sample size to be able to pick up a reasonable effect (more on this later) • Study that finds treatment is not statistically different from zero, but cannot rule out a reasonably big impact may be worse than useless • may be misread as proving no effect

4. Is the program representative? • Representative program design • Testing the impact of a standard microfinance program provides a benchmark against which variants can be measured • Representative geography and cultural context • Our results are more likely to generalize to a similar context, so testing in a context that looks like many others is a plus • Representative partner • If implementation is of higher quality in the evaluation than normal we may worry the results do not generalize to other partners • But doing impact evaluation of a program that is implemented very poorly is a waste: a process evaluation can tell you it wont work • Also, partner must be committed to evaluation or evaluation will not work

5. When being unrepresentative is good? • Proof of concept evaluations can be very informative • These test: does this type of program work when implemented very well • If proof of concept evaluation shows positive results, can then try cheaper, more scalable versions of the program • If proof of concept evaluation finds no impact, cheaper or more scalable versions done by less good partners unlikely to work

6. Is the implementing partner right? • Implementation partner commitment is essential, both at local and headquarter level • Q: Why is implementer commitment so important? How could lack of commitment disrupt an evaluation? • Potential tradeoff between having a partner committed to evaluation and a representative partner • One approach is work with committed partner but make sure program does not use highly specific and hard to replicate skills • Eg does not rely on very committed staff

7. Cost-effective evaluation • Rigorous impact evaluations vary enormously in cost, important to balance cost and benefits • What makes an evaluation expensive? • Collecting new data vs using administrative data • Randomizing by group (rather than individual) • Rural and remote locations drive up transport costs • When skilled labor is expensive (much of Africa) • Multiyear evaluations • Many treatment arms • Nonsurvey outcomes (like biological tests) • Large expensive evaluations can still be cost-effective by improving the effectiveness of spending

8. Developing an evaluation strategy • If an organization wants to develop a plan of where to invest its evaluation budget, start with key questions facing the organization • Determine whichcan be answered from existing evidence and which from process evaluation • Select top priority questions for impact evaluation • Find best context for answering them: where is representative and where are there opportunities to evaluate?

9. From impact back to process • With key questions answered from impact evaluations, process evaluation can be used to testoverall impact • Example of bednets and malaria • Impact evaluation shows hung bed nets reduce malaria • Impact evaluation shows bed nets give free get used • Needs assessment finds areas where malaria high, bednets low • Process evaluation monitors how many bed nets give out • Possible to impute lives saved from numbers given out