Data Analysis

What This Skill Does

The Data Analysis skill transforms raw inputs into actionable intelligence, prioritizing statistical rigor and methodological transparency. It bridges the gap between arithmetic and decision-making by forcing a focus on the 'so what?' of every query. Designed for analysts, product managers, and researchers, this tool provides a structured framework for hypothesis testing, cohort analysis, and pattern discovery while actively flagging cognitive biases and statistical errors that often lead to faulty conclusions.

Installation

You can install this skill directly via the command line within the OpenClaw environment: clawhub install openclaw/skills/skills/ivangdavila/data-analysis

Use Cases

• A/B Testing: Evaluate whether a UI change truly improved conversion rates by analyzing p-values, effect sizes, and confidence intervals.
• Product Churn: Perform cohort analysis to determine if recent updates or marketing campaigns effectively moved the needle on long-term user retention.
• Performance Metrics: Normalize time-series data to compare trends across periods of different lengths (e.g., leap years or varying month lengths) to avoid common pitfalls.
• Statistical Validation: Use it as a sanity check before presenting executive summaries to ensure you aren't falling for Simpson's Paradox or over-interpreting correlation as causation.

Example Prompts

1. "We ran an A/B test on our landing page. The variant B conversion rate is 5.2% versus 4.8% for the control, but the sample size is only 200 users per group. Is this result statistically significant and should we roll it out?"
2. "I'm looking at our retention data. Our overall retention seems to be increasing, but I suspect this is Simpson's Paradox. Can you help me segment this by signup cohort to see if the trend holds?"
3. "Our customer churn spiked in March. How should I account for the fact that February was a shorter month when comparing the churn rate across these two periods?"

Tips & Limitations

To maximize the effectiveness of this skill, always start by defining the decision you are trying to make. Analysis without a clear decision is merely arithmetic. Remember that correlation does not equal causation; always check for confounding variables. Be wary of 'p-hacking'—testing your data until you find a 'significant' result is not valid methodology. Finally, always quantify uncertainty by using ranges rather than point estimates. This skill is a decision-support tool, not a substitute for domain expertise, and it may not always be able to control for unobserved variables in your raw datasets.