AidGrade gathers information from academic studies and synthesizes the results in meta-analyses and systematic reviews. This evidence can inform policy decisions, and it is important to stay up-to-date. A recent study showed that 25% of such reviews go out of date within 2 years, having their former conclusions overturned, but nobody is likely to undertake a new review within that timeframe. That is why we need to move to a model of living
meta-analyses, in which results are constantly updated in real time.
New studies are coming out all the time. ScienceScape, for example, has catalogued 25 million studies in health alone. In this environment, it's very difficult for meta-analyses to stay current so that policymakers are getting the best evidence.
The hardest part of conducting a meta-analysis is gathering the data. We believe that machine learning can help solve this problem. We will use it to try to extract information from academic papers, including details of the study (e.g. where it was done, sample characteristics, whether it was a randomized controlled trial) as well as effect sizes, which are used in meta-analysis. We feel that with recent progress in machine learning, this is well worth the effort. Each piece of information that we extract from papers will be associated with a probability that it is correct. We already have existing data that were manually gathered from papers in international development, including health, economics, and education. We can use these data to train and validate our models.
This fundraising effort cannot cover all the costs of this project, but it will allow us to get started. If we are successful, this project could change the way we learn about what works. Would you like to make history with us?
All donations are tax-deductible in the U.S.