From brainstorms to analyses

PSYC 11: Laboratory in Psychological Science

Jeremy R. Manning
Dartmouth College
Spring 2026

Where we are

  • Examined your sleuthing dataset
  • Brainstormed how to answer each of the 5 questions
  • Actually run the analyses, make the figures, and interpret what you find
  • Write it up

Sharing time!

  • What's been surprising about your group's dataset?
  • What's been easier than you expected?
  • What's been harder?
  • Any questions you'd like to bring to the class?

Handling the "impossible" questions

A few of the "impossible" questions asks about causation, when you only have correlational data. (Other versions: missing comparison group, no time-ordering, no manipulation, etc.)

  • Should you just give up?
  • What can you still say that's interesting?
  • How can you be honest about the limits of your data, while still making a contribution?
  • How might you turn an "impossible" question into a figure?

When you can't answer the question, answer a different (related) one

  • Describe what you can see: report the correlation, the group difference, the trend — but be explicit that it's not causal
  • Show what would be needed: e.g., "to address this we would need a randomized experiment with X manipulation"
  • Probe alternative explanations: are there confounds in the data? Can you rule any of them out?
  • Reframe the question: turn "Does X cause Y?" into "Among people who experienced X, what's their Y?" — that's a real, answerable question

Acknowledging a limit is not the same as throwing in the towel. A clear-eyed "here's what we can say, and here's what we can't" is often the most valuable contribution.

Suggested workflow: Colab + your AI of choice

  1. Open a Colab notebook and share with your group on Slack
  2. Upload or provide a Google Sheets link to your dataset
  3. Pick your favorite GenAI: Gemini, Claude, ChatGPT, chat.dartmouth
  4. Give it: (a) the dataset description document, (b) a sample of the data, (c) the analysis plan you brainstormed yesterday
  5. Ask it to implement — not invent — the analyses and figures you already designed

GenAI is designed to sound confident even when it's wrong. Before you run the analyses, think about (a) what you expect to see, and (b) how you could check that the output is correct.

Questions? Want to chat more?

📧 Email me
💬 Join our Slack
💁 Come to office hours