Key Takeaways: Lies, Damn Lies, and Statistics: Ethical Data Practice

One-Sentence Summary

Ethical data practice requires recognizing that statistics can mislead without fabrication (through cherry-picking, Simpson's paradox, and the ecological fallacy), that research integrity depends on distinguishing confirmatory from exploratory analysis (preventing p-hacking and HARKing), that data privacy and informed consent are harder to achieve than most people think, and that every data-driven decision embeds value judgments about whose welfare matters — judgments that demand transparency, accountability, and input from affected communities.

Core Concepts at a Glance

Simpson's Paradox

Questionable Research Practices

Ethical Frameworks for Data Practice

Research Ethics Timeline

Data Privacy

The "Lying with True Statistics" Checklist

Ask yourself before every analysis:

Perspective-Taking Framework

For any data-driven decision, consider:

Personal Code of Statistical Ethics (Template)

Key Python Code

Simpson's Paradox Detector

def check_simpsons_paradox(df, outcome, group, stratify_by):
    """
    Compare aggregate and stratified trends to detect
    Simpson's paradox.

    Returns dict with aggregate and per-stratum results,
    plus a paradox_detected flag.
    """
    agg = df.groupby(group)[outcome].mean()
    groups = sorted(agg.index)
    agg_diff = agg[groups[1]] - agg[groups[0]]

    reversal_count = 0
    for stratum in df[stratify_by].unique():
        subset = df[df[stratify_by] == stratum]
        strat_means = subset.groupby(group)[outcome].mean()
        strat_diff = strat_means[groups[1]] - strat_means[groups[0]]
        if (agg_diff > 0 and strat_diff < 0) or \
           (agg_diff < 0 and strat_diff > 0):
            reversal_count += 1

    return reversal_count > len(df[stratify_by].unique()) / 2

Common Mistakes

Connections