Home▸Data Science▸Autocorrelated Residuals: Effective Sample Size, Durbin-Watson and Overconfident Intervals

Autocorrelated Residuals: Effective Sample Size, Durbin-Watson and Overconfident Intervals

Generate AR(1) noise and watch a series of hundreds of points collapse into a few dozen effective observations as the autocorrelation grows. Compare the textbook confidence interval with the true one, read the Durbin-Watson statistic and the lag plot, and check the exact variance of the mean, the effective size and the real coverage against Monte Carlo.

Data Science3DModerate60 FPS📱 Mobile-adapted🟠 Blender⇄ 2D version
Autocorrelated Residuals: Effective Sample Size, Durbin-Watson and Overconfident Intervals ↗ Open standalone

Every analytical query engine has to decide which bytes on disk it actually needs to touch, and that decision is dictated by the physical layout underneath the table. This simulator renders an 8-column, 20-block table as a 3D grid and lets you run the same query — a column selection plus a filter — against a row-oriented layout and a column-oriented layout side by side. Predicate pushdown skips whole blocks whose value range can't match the filter in both layouts; column pruning then lets the columnar layout skip whole columns too, while the row-oriented layout is forced to read every column of every matching block because rows are interleaved on disk. Live readouts convert the highlighted blocks into bytes scanned, blocks touched, and an estimated scan time at a fixed disk throughput, with dictionary+run-length compression applied at realistic, layout-specific ratios.

⚙ Under the hood

Generate AR(1) noise and watch a series of hundreds of points collapse into a few dozen effective observations as the autocorrelation grows. Compare the textbook confidence interval with the true one, read the Durbin-Watson statistic and the lag plot, and check the exact variance of the mean, the effective size and the real coverage against Monte Carlo.

data engineeringcolumnar storageOLAPquery optimizationcompressiondata warehouseAutocorrelationDurbin-WatsonEffective Sample SizeAR(1)Confidence Interval Coverage

3D · Three.js / WebGL renderer · 60 FPS target · runs fully client-side, no install

What did you find?

Add reproduction steps (optional)