28 · Moving forward

End of semester · date to be announced

Materials for this session are not published yet. They appear here before class.

Objectives

By the end of this session you can:

  • Name the parts of a research project that stay with you when tools do the coding and the labelling.
  • Say what a given measurement instrument cannot see, and write that limit into a methods section.
  • Name the independent source you would compare a variable in your own research with, and what that comparison would show a referee.
  • List what you read recently outside economics, and what each piece made you want to measure.
  • Lay out how you would learn a new method on your own.

What we cover

  • Many advances in economics came from measuring something nobody had measured before. Two survey articles, papers that summarise a field’s methods and findings for people new to it, trace this for economic activity seen from space and for sentiment read from text:
  • What the tools do not replace. A model will answer a badly posed question, and a pipeline will run on a variable that means nothing. Choosing the question and judging whether the variable measures what you claim stay with you, and they decide the quality of the work.
  • Know what your instrument cannot see. A measurement instrument (the device or data source that records a quantity, such as a satellite sensor or a survey) sees part of what you care about, so the variable you build from it is often a proxy, an observable stand-in for something you cannot observe, such as lights for income. A vegetation index tracks greenness, which is only partly yield. Transcripts drop what the room knew. A classifier’s label is a prediction with an error rate (the share of cases it labels wrongly, measured on examples whose answers are known), and it belongs in the analysis as one. Measurement error, the gap between the recorded and the true value, has a cost: random errors in an explanatory variable shrink its estimated effect toward zero (attenuation bias). Write the limit into your methods section, the part of a paper that explains how the data were built and analysed, in your own words.
  • To verify a variable is to compare it with an independent source that measured the same thing, ideally ground truth (values observed directly, such as a harvest weighed on the field). For a variable in your own research, name that source and the comparison a referee, the expert a journal asks to assess a submitted paper, would need to see.
  • Read outside economics. Questions come from noticing things, and you notice more when you read widely: history, geography, the other social sciences, good reporting, and the domain literature, meaning research from the field that studies your subject directly, such as hydrology for water use. Come with three things you read recently that were not economics, and what each one made you want to measure.
  • Where to take this next. How to learn a new method on your own. Start from a survey article and a replication package (the data and code a paper’s authors release so others can reproduce its results), then apply the method to data where you already know the answer before you trust it on new data.