AI Skill Report Card
Conducting Scientific Research
Scientific Research Methodology
Quick Start10 / 15
Python# Research pipeline example research_pipeline = { "question": "How does X affect Y?", "hypothesis": "X increases Y through mechanism Z", "methods": ["experimental_design", "data_collection", "statistical_analysis"], "analysis": "statistical_tests + visualizations", "conclusion": "evidence_based_findings" }
Recommendation▾
Make Quick Start more actionable with a specific research problem and complete workflow example rather than just a generic pipeline structure
Workflow15 / 15
Progress:
- Literature Review: Search databases (PubMed, Google Scholar, arXiv)
- Research Question: Formulate specific, testable hypothesis
- Methodology Design: Choose appropriate experimental/analytical methods
- Data Collection: Gather primary/secondary data systematically
- Data Analysis: Apply statistical methods and create visualizations
- Results Interpretation: Draw evidence-based conclusions
- Paper Writing: Structure findings in academic format
- Peer Review: Validate methodology and conclusions
Literature Review Process:
- Define search terms and databases
- Screen abstracts for relevance
- Full-text review of selected papers
- Extract key findings and methodologies
- Identify research gaps
Experimental Design:
- Define variables (independent, dependent, control)
- Determine sample size using power analysis
- Select randomization and blinding strategies
- Plan data collection protocols
- Establish statistical analysis plan
Recommendation▾
Reduce length by removing verbose explanations (e.g., the detailed Literature Review and Experimental Design sub-sections could be condensed)
Examples15 / 20
Example 1: Hypothesis Formation Input: "Climate change effects on crop yields" Output:
- Research Question: "How do increasing temperatures affect wheat yield in temperate regions?"
- Hypothesis: "Rising temperatures above optimal range (15-20°C) will decrease wheat yields by 10-15% per degree increase"
- Variables: Temperature (independent), wheat yield (dependent), rainfall/soil (controls)
Example 2: Data Analysis Input: Experimental dataset with treatment groups Output:
Python# Statistical analysis framework import scipy.stats as stats import pandas as pd # Descriptive statistics summary_stats = data.groupby('treatment').agg({ 'outcome': ['mean', 'std', 'count'] }) # Hypothesis testing t_stat, p_value = stats.ttest_ind(control_group, treatment_group) effect_size = (treatment_mean - control_mean) / pooled_std
Example 3: Research Paper Structure Input: Completed analysis results Output:
1. Abstract (250 words): Objective, methods, results, conclusion
2. Introduction: Background, literature gap, research question
3. Methods: Study design, participants, procedures, analysis
4. Results: Descriptive stats, main findings, figures/tables
5. Discussion: Interpretation, limitations, implications
6. References: APA/journal format
Recommendation▾
Provide more concrete input/output examples - Example 1 shows good specificity but Examples 2-3 are too template-like rather than showing actual research scenarios
Best Practices
Research Design:
- Use PICOT framework (Population, Intervention, Comparison, Outcome, Time)
- Pre-register study protocols to prevent bias
- Calculate required sample sizes before data collection
- Include positive and negative controls where appropriate
Data Management:
- Use version control for data and analysis code
- Document all preprocessing steps
- Implement data validation checks
- Store raw data separately from processed data
Statistical Analysis:
- Check assumptions before applying tests
- Use appropriate multiple comparison corrections
- Report effect sizes alongside p-values
- Include confidence intervals for estimates
Writing:
- Follow journal-specific formatting guidelines
- Use active voice and clear, concise language
- Include all necessary details for reproducibility
- Cite recent and relevant literature
Common Pitfalls
Study Design Issues:
- Correlation vs Causation: Don't imply causation from observational data
- Sample Bias: Ensure representative sampling for target population
- Confounding Variables: Control for alternative explanations
- Underpowered Studies: Calculate sample size requirements early
Analysis Errors:
- P-hacking: Don't fish for significant results by trying multiple tests
- Cherry-picking: Report all analyses performed, not just significant ones
- Inappropriate Tests: Match statistical tests to data type and distribution
- Missing Data: Handle systematically, don't just delete cases
Reporting Problems:
- Selective Reporting: Include non-significant results
- Overgeneralization: Stay within bounds of your sample and methods
- Incomplete Methods: Provide sufficient detail for replication
- Unsupported Claims: Base conclusions strictly on presented evidence
Reproducibility:
- Code Documentation: Comment analysis code thoroughly
- Data Availability: Share data when possible per journal policies
- Method Transparency: Report all steps, including failed approaches
- Version Control: Track changes to analysis and manuscripts