Why should clustering be accounted for in statistical analysis?

• A. It increases sample size and reduces p-values erroneously
• B. It violates independence and variances must be adjusted
• C. It guarantees normal distribution of errors
• D. It has no impact on analysis

Answer: (B)

Clustering matters because observations within the same cluster are more similar to each other than to observations from different clusters, creating intra-cluster correlation. This means the usual assumption of independence underlying many statistical methods is violated. When independence is assumed but not present, standard errors tend to be too small, which makes p-values overly optimistic and can lead to false claims of significance. To get valid inferences, you need to account for clustering by either modeling the correlation (for example, with random effects or mixed models) or by adjusting the standard errors (such as cluster-robust standard errors or generalized estimating equations). The impact grows with larger cluster sizes and higher intra-cluster correlation, summarized by the design effect, which inflates the variance and reduces effective sample size.