Hypothesis Generation and Research Design

From Literature Gaps to Research Questions

🔄 Quick Recall: In the previous lesson, you used AI tools to search 200M+ papers, map field structures, extract data, and evaluate evidence. Now you’ll use those findings to generate and refine research hypotheses.

The literature review tells you what’s been done. The hypothesis generation phase asks: what should be done next? AI helps you explore this question systematically.

Identifying Research Gaps

After completing your literature review, AI can analyze the findings for gaps:

Based on my literature review on [topic], identify research gaps:

What I found:
- [Key finding 1 with source]
- [Key finding 2 with source]
- [Key finding 3 with source]
- [Methodological approaches commonly used]
- [Populations commonly studied]

Identify:
1. Questions that the literature raises but doesn't answer
2. Populations or contexts that haven't been studied
3. Methodological limitations that a new study could address
4. Contradictions between studies that need resolution
5. Cross-disciplinary connections that haven't been explored

What makes a gap worth filling:

✅ Quick Check: Why is the “worth pursuing” column critical? Because the goal isn’t to find any gap — it’s to find a gap whose filling advances knowledge. AI can identify dozens of gaps. Your expertise determines which ones matter.

Generating Hypotheses

Once you’ve identified promising gaps, AI helps generate specific, testable hypotheses:

Help me develop testable hypotheses for this research gap:

Gap: [describe the specific gap you identified]
Field: [your discipline]
Theoretical framework: [theory or model guiding your thinking]
Available methods: [what methods you can realistically use]
Available data: [what data you can access or collect]

Generate 5-10 hypotheses that:
1. Are specific and falsifiable
2. Follow logically from existing evidence
3. Are testable with available methods and data
4. Would contribute meaningfully to the field if confirmed or rejected
5. Vary in scope (some narrow and immediately testable, some broader)

Evaluating AI-generated hypotheses:

For each hypothesis, ask:

• Is this truly falsifiable? (Could evidence disprove it?)
• Does existing theory support this prediction? (Or is it arbitrary?)
• Can I operationalize the key variables? (How would I measure them?)
• What would finding support (or refutation) mean for the field?
• Is this achievable within my resources and timeline?

Research Design with AI

Once you’ve selected your hypothesis, AI assists with study design:

Help me design a study to test this hypothesis:

Hypothesis: [your selected hypothesis]
Field and context: [discipline, population, setting]
Constraints: [budget, timeline, access to participants/data, ethical limits]
My methodological expertise: [methods I'm trained in]

Propose:
1. Study design (experimental, quasi-experimental, observational, etc.)
2. Sample size estimation with justification
3. Data collection methods
4. Primary and secondary outcome measures
5. Analysis plan (statistical tests or qualitative approach)
6. Potential threats to validity and how to address them
7. Ethical considerations

The simplicity principle: When AI suggests a complex design, always ask: “Is there a simpler design that answers my question equally well?” Complex designs increase the chance of errors, require more resources, and take longer. The best design is the simplest one that rigorously addresses your hypothesis.

✅ Quick Check: Why is the simplicity principle important? Because every additional element in a study design — extra conditions, additional measures, mixed methods — introduces opportunities for error, increases cost, and extends the timeline. A well-designed simple study produces cleaner evidence than a poorly executed complex one. AI often suggests comprehensive designs; your job is to trim to the essential.

Power Analysis and Sample Planning

Calculate the required sample size for my study:

Study design: [design type]
Primary outcome measure: [what you're measuring]
Expected effect size: [small/medium/large, or specific value if known]
Significance level: [typically 0.05]
Desired power: [typically 0.80 or higher]
Number of groups/conditions: [if applicable]
Expected attrition: [percentage likely to drop out]

Explain the reasoning behind the recommendation and note any assumptions.

Key Takeaways

• AI identifies gaps in the literature systematically — but domain expertise determines which gaps are worth filling
• Generate 5-10 hypotheses per gap, then evaluate each for falsifiability, theoretical grounding, and feasibility
• Research design should follow the simplicity principle: choose the simplest design that rigorously answers your question
• AI tends to suggest comprehensive designs; your job is to trim to the essential
• Always include power analysis and sample size estimation before committing to a study

Up Next: You’ll learn AI-assisted data analysis — from natural language statistical interfaces to code generation and visualization.