Correlation measures the linear relationship between variables: positive values move together, while negative values move in opposite directions. It can reveal patterns, but association alone does not establish causation. section 2.6.3 section 4.3
Use correlation as an investigative clue rather than an automatic explanation. Trends, seasonality, autocorrelation, outliers, and external shocks can create misleading or temporary relationships, especially in time-series data. section 2.6.3 section 4.3
Interpret correlation alongside visual inspection and the business context:
- Inspect the direction and strength of relationships in a scatterplot matrix. section 4.3
- Investigate outliers because data errors, unusual cases, or genuine extremes may influence the apparent relationship. section 4.3
- Question correlations between time-ordered variables when shared trends may explain their movement. section 2.6.3
Review timestamps
A tempting check compares review_answer_timestamp with review_creation_date. Because both values relate to orders placed over time, their correlation may reflect the shared passage of time rather than a meaningful relationship. A more useful investigation calculates the time difference, summarizes it, and examines its histogram and box plot. section 2.6.3
