Following the primaries? Maybe you’ve come across The Needle, the odometer-like data visualization The New York Times uses to display live election results. This seemingly simple graphic does a big job, analyzing initial returns and other sources to produce easy-to-consume forecasts. It also encapsulates The Times’ ambitious goal of providing real-time election predictions to information-hungry voters.
In the Pay for Success (PFS) world, we grapple with similar demands. Like voters feverishly anticipating their candidates’ delegate counts, PFS project stakeholders crave to measure their impact as quickly as possible. In fact, the desire for data-based validation is even more intense in our space. PFS initiatives typically unfold over 3 to 7 years to account for service delivery and participant observation. But few funders want to wait that long to know whether the millions in catalytic capital they’ve contributed are making an impact on society. Consequently, the temptation to peek under the hood mid-project can be overwhelming. However, looking at the preliminary evaluation results of an intervention and using that data to make funding or continuation decisions is risky.
The dangers of early analysis
If you’ve been keeping an eye on The Needle this primary season, you probably understand why. During early the hours of the New Hampshire Democratic primary on Feb. 11, it reflected the initial assumption that Vermont Senator Bernie Sanders had a 70% chance to win. But as initial returns rolled in, former South Bend, Indiana, Mayor Pete Buttigieg gained momentum and The Needle shifted — Sen. Sanders’ chances of victory dropped and Mayor Buttigieg’s increased sharply. However, several waves of fresh voting data pushed the senator back into the lead, stabilized The Needle, and allowed it form one final conclusion: 100% for Sen. Sanders.
PFS project evaluation results tend to fluctuate just like this. So, if stakeholders review intervention data too early, they might draw improper inferences about final outcomes much like you might’ve assumed Mayor Buttigieg was going to win New Hampshire after glancing at The Needle at 8:30 p.m. Actions like this can naturally lead to premature corrective action that pushes a successful program off track or results in the termination of an initiative on the verge of gaining traction.
Strategies for going under the hood
That said, in a PFS initiative, or for that matter, any pilot, demonstration, or project with an evaluation-based learning agenda, reviewing preliminary evaluation results is sometimes necessary. So if you must check on your evaluation cake before it’s fully baked, consider adopting the following principles to help avoid making improper inferences:
- Write a short list of questions you want to answer before you look at the data. The more analyses you conduct, the greater likelihood you’re going to find skewed data. So, carefully think about why you’re looking into evaluation data early and limit your analysis to the most relevant queries.
- Plan ahead. Create a plan for what actions you will take based on the various possible outcomes you could see. If you do this before looking at the data, you’re more likely to make decisions with a clear head and avoid reactionary responses.
- Follow The Needle. The Needle measures and communicates the likelihood of a given electoral outcome, which is why a candidate might look like a primary winner at 6 p.m. but end up giving a concession speech at 11 p.m. Keep this in mind when reviewing your preliminary results and really consider whether an outcome you’ve observed is likely to remain true. You and your stakeholders might make a different decision if you’re 80% confident the outcomes you see are going to hold versus if you’re only 30% sure. Because at the end of the day, the last thing anyone wants to do is make an error akin to “Dewey defeats Truman” when the future of a key social program is at stake.