research-pipeline

Research Pipeline

Execute a rigorous research analysis workflow with pre-specified hypotheses, assumption verification, and publication-quality reporting.

Steps

1. Define hypotheses. Before touching data, articulate the null and alternative hypotheses in precise terms. State the primary outcome variable, the predictor or grouping variable, and the expected direction of the effect.

2. Assess data fitness. Call describe_database and get_data_quality_report with the database name from $ARGUMENTS. Evaluate:

   • Sample size relative to power requirements
   • Missing data patterns and their potential impact
   • Measurement quality (appropriate scales, reasonable ranges)

3. Power analysis. Call execute_query to compute the observed effect variability. Estimate whether the available sample provides at least 80% power to detect the expected effect size at alpha = 0.05. If underpowered, report the minimum sample needed.

4. Verify assumptions. For the planned statistical test:

   • Normality: compute skewness and kurtosis via execute_query; run Shapiro-Wilk via analyze_hypothesis_test if supported
   • Homoscedasticity: compare group variances via execute_query
   • Independence: assess based on study design (not testable from data alone, but note clustering)
   • Multicollinearity (for regression): compute VIF or correlation matrix via execute_query

5. Run primary analysis. Execute the pre-specified test:

   • Group comparisons: analyze_hypothesis_test or analyze_anova
   • Regression: analyze_regression with evaluate_model_performance Let assumption checks guide method selection (parametric vs. non-parametric).

6. Compute effect sizes. Call analyze_effect_sizes. Report with confidence intervals. Classify magnitude using standard conventions (Cohen's d: small=0.2, medium=0.5, large=0.8).

7. Sensitivity analysis. Re-run the primary analysis under alternative conditions:

   • Excluding outliers (identified via IQR or z-score)
   • Using the alternative test (parametric if non-parametric was primary, or vice versa)
   • Adjusting for covariates if available Report whether conclusions are robust to these changes.