Degrees of freedom

Degrees of freedom are the number of independent values that can vary in a statistical calculation after constraints are applied. They determine how uncertainty is estimated and which reference distributions apply.

Sample variance: A measure of how widely sample observations vary around their sample mean. Estimating the mean uses one constraint, so variance calculations commonly divide by sample size minus one.
Student's t-distribution: A family of symmetric probability distributions used for inference about a mean when variance is estimated. Its shape depends on degrees of freedom, especially in small-sample inference.
Effective degrees of freedom: A measure of model complexity that estimates how many parameters a fitting procedure effectively uses. For flexible or regularized models, complexity need not equal the raw parameter count.
Statistical sample: A collection of observations selected or recorded for statistical analysis. Sample size sets the starting count from which many degrees-of-freedom calculations proceed.
Bessel's correction: The use of n−1 rather than n when estimating a population variance from a sample. It compensates for estimating the mean from the same observations and corresponds to one lost degree of freedom.
Chi-squared distribution: A probability distribution formed by summing squares of independent standard normal variables. Its degrees of freedom equal the number of independent squared components.
Fractional degrees of freedom: Non-integer degrees-of-freedom values used in approximations for statistical reference distributions. Approximations such as Welch's test can yield degrees of freedom that are not whole numbers.
Independent and identically distributed random variables: Random variables that are mutually independent and share the same probability distribution. Many standard degrees-of-freedom formulas assume observations with this structure.
Linear constraints: Equations that restrict variables to combinations satisfying specified linear relationships. Each independent constraint removes a possible direction of variation.
F-distribution: A probability distribution for ratios of scaled independent chi-squared variables. Its numerator and denominator each have their own degrees-of-freedom count.

Linked from 11 pages