effective sample size

Estimation Quality

Accuracy of a statistical inference depends on the diversity of the underlying points. When effective sample size is markedly lower than the raw count, the variance of the estimate exceeds the theoretical variance. This occurs often in time series where consecutive days look nearly identical.

Sparse data with high independence often outperforms dense data with high dependency.

Particle Diversity

Computational filters rely on a broad set of hypotheses to track a moving target. If the effective sample size drops too low, the filter collapses onto a single explanation, ignoring other possibilities. This phenomenon is known as weight collapse.

Maintaining a broad variety of samples prevents this failure.

Convergence Limit

Mathematical bounds on the speed of learning depend on the accumulation of unique information. A founder tracking a metric observes that effective sample size grows slowly when users are acquired from the same narrow demographic. Broadening the source of information accelerates the stabilization of the estimate.

Precise measurement of this growth rate informs the budget for data acquisition. It defines the point where further data collection yields diminishing returns for decision making.