Success in research: Difference between revisions

Lumo (talk | contribs)
Make the steps for a successful research project easier to follow and add relevant wiki links
Lumo (talk | contribs)
Show how to plan and combine several studies using an AI and automation example
 
Line 22: Line 22:


Then state what the evidence supports. If you found three floods in the 1960s, give their dates and sources. If the archive is missing several years, say which ones and leave open how often the street flooded. Share the timeline and enough of your method for someone else to check or extend it. This is one way to put [[Open research|open research]] into practice.
Then state what the evidence supports. If you found three floods in the 1960s, give their dates and sources. If the archive is missing several years, say which ones and leave open how often the street flooded. Share the timeline and enough of your method for someone else to check or extend it. This is one way to put [[Open research|open research]] into practice.
== When a big question needs several studies ==
How could [[Artificial intelligence|AI]] and automation make goods and services more affordable while limiting surveillance and the concentration of wealth and power? Keep that question at the top of a shared plan. Start with a service you can compare across local governments, such as building permits.
# '''Decide what progress would look like.''' Track fees, wait times, and errors. Check separately for residents who need help applying or who apply offline, since an average can hide who is worse off. Look at which data the agency and vendor can collect or reuse, who can reverse a decision, and whether the agency can switch vendors and keep its records.
# '''Divide the work and agree on what to record.''' One team can compare service records across offices. Another can read public contracts and data policies. A third can interview residents and staff about appeals and how the rules work in practice. Use the same offices, time periods, and definitions. If records are unavailable, ask for summary figures or mark the gap in the shared log.
# '''Check what else might explain a change.''' Measure outcomes before and after adoption. If possible, compare the changes with those in a similar office that kept its previous process. Check whether the offices had similar trends before adoption and record changes in staffing or application volume. If those factors cannot be sorted out, report what changed and how uncertain its cause is.
# '''Bring the results back to the big question.''' Make a table with one row for each arrangement studied. Record who benefited, who gained control over data or decisions, and how strong the evidence is. A shorter wait alongside broader vendor access to residents' data is a mixed result. Look for offices that gained speed under tighter data rules. Where the sources disagree, record what would help resolve the difference.
A permit study can show which arrangements are worth testing elsewhere. To study abundance on a larger scale, repeat the approach where automation might change the cost of essentials or workers' pay. Explain which results appear across studies and which depend on the rules or conditions in one place.


== When working with others helps ==
== When working with others helps ==
Line 48: Line 59:


* [https://www.nationalacademies.org/read/25303/chapter/3 National Academies, Reproducibility and Replicability in Science]. See the summary for ways to check an analysis and repeat a study with new data.
* [https://www.nationalacademies.org/read/25303/chapter/3 National Academies, Reproducibility and Replicability in Science]. See the summary for ways to check an analysis and repeat a study with new data.
* [https://www.gov.uk/government/publications/the-magenta-book/guidance-on-the-impact-evaluation-of-ai-interventions-html HM Treasury, Guidance on the Impact Evaluation of AI Interventions]. See the sections on choosing a comparison and combining different sources of evidence.


== See also ==
== See also ==