Commentary on the research industry, and short briefs from studies we ran ourselves. Posts drafted with Surdatics AI are labelled, and a person on our team reviews each one before it appears here.
A study of an online intimate-partner violence survey suggests that respondent authenticity can affect what a measure appears to show, not just which records get removed. Research teams need to test whether their conclusions survive uncertain classifications.
A broad training opt-out may not help someone whose conversation becomes sensitive midway through. Research teams need to design choices around what people share and when those choices can still make a difference.
Online survey platforms showed markedly different rates of failure on tests designed to detect AI agents. Research teams should treat those results as a reason to examine recruitment and validation, not as a count of bots.
A new consumer survey shows why price sensitivity, quality preferences and shopping channels need to be studied together. The useful question for research teams is not which single trend wins, but when and where each behaviour appears.
A UK regulatory review of foundation-model developers makes training-data provenance a practical accountability question. Research teams should ask not only whether data can be used, but whether its origins and the rights of the people concerned remain clear.
Evidence from Ghana shows why implementation monitoring should examine what teachers do in classrooms, not just whether they received training. A checklist improved a teaching practice, but its effect on learning remains a separate question.
Ipsos finds no clear national majority ahead of South Africa’s local elections. The most useful reading of its survey starts with who was asked, how voting intention was measured, and where uncertainty grows.
A new development-effectiveness report points to a persistent gap between commitments to country ownership and the information used to manage aid. Research teams should treat that gap as a question about who can inspect, challenge and use the evidence.
56.8% of respondents who answered the storage-access question said they lacked access to proper storage or cold rooms. Respondents also frequently selected bad roads as a transport challenge and better storage as a way to reduce waste.
Afrobarometer’s survey puts everyday service and livelihood concerns beside the agendas of development-cooperation experts. The useful lesson is not to discard long-term reform, but to test whether it answers the problems citizens actually name.
46.4% of respondents who answered the training question said they received safety compliance training regularly. Lack of resources was the most frequently selected barrier among those who answered the barriers question.
A model given information about real panelists still produced a different picture of public opinion. Research teams should distinguish useful questionnaire experiments from evidence gathered directly from people.
ODK Collect works offline by design. The interesting engineering is not the collection — it is what happens when a week of interviews finally reaches a server.
Most survey platforms pay out of revenue. We fund rewards into escrow before a survey goes live, and the difference matters most on the day something goes wrong.
Scraped data is getting more expensive by the lawsuit. The alternative is not better scraping — it is asking, and being able to prove you asked.
