Uniform Distribution — $ \mathrm{Unif}(a,b) $

Definition

Parameters:

  • Lower bound $ a \in \mathbb{R} $
  • Upper bound $ b > a $

Support:
\(a \le x \le b\)

PDF:
\(f(x \mid a,b) = \frac{1}{b-a}\, \mathbb{1}_{[a,b]}(x)\)

CDF:
\(F(x) = \begin{cases} 0, & x < a \\ \frac{x-a}{b-a}, & a \le x \le b \\ 1, & x > b \end{cases}\)


Moment Generating Function (MGF) and Characteristic Function

MGF:
\(M_X(t) = \frac{e^{tb}-e^{ta}}{t(b-a)}, \quad t \ne 0\)

Characteristic Function:
\(\varphi_X(t) = \frac{e^{itb}-e^{ita}}{it(b-a)}\)


Moments

Mean and Variance

\(\mathbb{E}[X] = \frac{a+b}{2}\)

\[\operatorname{Var}(X) = \frac{(b-a)^2}{12}\]

Raw Moments

\(\mathbb{E}[X^k] = \frac{b^{k+1}-a^{k+1}}{(k+1)(b-a)}\)


Location–Scale Family

If $ X \sim \mathrm{Unif}(0,1) $, then \(a+(b-a)X \sim \mathrm{Unif}(a,b)\)

Thus Uniform is a location–scale family.


Order Statistics

Let $ X_1,\dots,X_n \stackrel{iid}{\sim} \mathrm{Unif}(a,b) $.

Minimum

\(X_{(1)} \sim a + (b-a)\,\mathrm{Beta}(1,n)\)

Maximum

\(X_{(n)} \sim a + (b-a)\,\mathrm{Beta}(n,1)\)

General Order Statistic

\(\frac{X_{(k)}-a}{b-a} \sim \mathrm{Beta}(k,n-k+1)\)


Sufficient Statistics

For $ \mathrm{Unif}(a,b) $ with both parameters unknown:

  • Minimal sufficient statistic: \((X_{(1)}, X_{(n)})\)

  • Not complete

This is a classic nonregular family.


Likelihood and Estimation

Likelihood

\(L(a,b) = (b-a)^{-n} \mathbb{1}\{a \le X_{(1)},\, X_{(n)} \le b\}\)

MLEs

\(\hat{a} = X_{(1)}, \quad \hat{b} = X_{(n)}\)

Bias:
\(\mathbb{E}[X_{(1)}] = a + \frac{b-a}{n+1}\) \(\mathbb{E}[X_{(n)}] = b - \frac{b-a}{n+1}\)


UMVU Estimators

Using order statistics:

\[\tilde{a} = X_{(1)} - \frac{X_{(n)}-X_{(1)}}{n-1}\] \[\tilde{b} = X_{(n)} + \frac{X_{(n)}-X_{(1)}}{n-1}\]

Exist despite lack of completeness.


Nonregularity

  • Support depends on parameters
  • Fisher information not well-defined
  • CRLB does not apply

Exam favorite: counterexample to standard theory assumptions.


Transformations

Probability Integral Transform

If $ X \sim \mathrm{Unif}(a,b) $, then \(\frac{X-a}{b-a} \sim \mathrm{Unif}(0,1)\)

Conversely, if $ U \sim \mathrm{Unif}(0,1) $, then \(F^{-1}(U) \sim F\)


Bayesian Notes

  • No natural conjugate prior for $ (a,b) $
  • Often handled via hierarchical or improper priors

Key Theorems and Facts (Prelim-Relevant)

  • Order statistic distributions
  • Minimal but not complete sufficiency
  • Nonregular family
  • Failure of CRLB
  • Probability integral transform

Exam Takeaways

  • Always reduce to $ \mathrm{Unif}(0,1) $
  • Order statistics carry all information
  • Expect boundary-based estimators
  • Uniform breaks many “nice” theorems

Comments