Sampling Distribution Lab

Repeat independent samples. Compare their means, unbiased variances, and—for dice—their proportions of odd faces.

Quick sizes:

Statistic across replicationsObserved meanTheoretical mean / referenceObserved varianceTheoretical variance

Observed variance across replications uses R − 1. Population μ = 3.5; σ² = 35/12 ≈ 2.91667; σ = √(35/12) ≈ 1.70783 in both modes. For the s row, σ is a population reference, not E[s].

Sample mean

Run simulation after changing width.

Unbiased sample variance S²

Run simulation after changing width.

Proportion of odd faces

Why does the variance look jagged?

Dice are discrete: small samples can produce only particular variance values. More replications stabilize their frequencies, but do not make the distribution continuous. Changing bin width changes the picture, not the underlying randomness.

For independent normal observations, (n − 1)S²/σ² follows χ² with n − 1 degrees of freedom. Thus S² follows a scaled chi-square distribution. This relationship is not exact for dice, even though the two populations share a mean and variance.

With automatic spacing, dice-mean bars show exact possible values separated by 1/n (0.2 when n = 5). Choose a mean interval width to group values instead; widths smaller than 1/n can reveal empty intervals. Fixed mean intervals start at multiples of the selected width, including negative values in normal mode. Variance interval widths can be selected above the chart; dice variances do not generally have the same spacing as dice means. Other histogram bars show probability in equal-width intervals; the final interval includes its right edge. All simulated values are counted. In normal mode, the comparison shows exact theoretical probability in each variance bin, computed from the chi-square CDF. This avoids the infinite density at zero when n = 2. The theoretical distribution extends beyond the displayed sample range; its omitted tail probability is reported.

Reference: normal sample variance and chi-square