Scott’s Rule
A normal-reference histogram bin-width rule that scales standard deviation by the inverse cube root of sample size.
Scott's rule supplies a bin width for a histogram treated as a density estimator. A common normal-reference form derived from David W. Scott's 1979 analysis is:
h ≈ 3.49 × σ / n^(1/3)σ denotes standard deviation, n sample size and h the proposed width. More precisely, the multiplier comes from σ × (24 × √π / n)^(1/3). The rule estimates width in the units of the data; it does not directly prescribe an integer count of bins.
At n=1000 and σ=10, the cube root of n is 10 and the width is approximately 3.49 units. A data range of 70 units would imply about twenty equal-width bins, subject to edge placement and rounding.
Sensitivity to extremes
An extreme measurement can inflate standard deviation and consequently produce wider bins. Important detail near the center may then disappear. Freedman–Diaconis, based on IQR, provides a more resistant comparator. With zero variance, Scott's expression also fails to produce a positive width.
Scott's rule in a KDE is different
SciPy's gaussian_kde also uses a Scott bandwidth factor, but its dimensional scaling is n^(−1/(d+4)), where d is the dimension. That factor belongs to kernel density estimation, not to the histogram formula with n^(−1/3). A KDE and a histogram are different density estimators.