2.3. The Chi-Square Distribution One of the most important special cases of the gamma distribution is the chi-square distribution because the sum of the squares of independent normal random variables with mean zero and standard deviation one has a chi-square distribution. This section collects some basic properties of chi-square random variables, all of which are well known; see Hogg and Tanis [6]. A random variable X has a chi-square distribution with n degrees of freedom if it is a gamma random variable with parameters m = n/2 and = 2, i.e X ~ (n/2,2). Therefore, its probability density function (pdf) has the form 2 (n/2) 0 t(n/2)-1e-t/2 n/2 (1) f(t) = f(t; n) = if t > 0 if t < 0 In this case we shall say X is a chi-square random variable with n degrees of freedom and write X ~ (n). Usually n is assumed to be an integer, but we only assume n > 0. Proposition 1. If X has a gamma distribution with parameters m and then 2X/ has a chi-square distribution with 2m degrees of freedom. Proof. By Proposition 5 in section 2.2 the random variable X has a gamma distribution with parameters m and 2, i.e X ~ (m,2) = ((2m)/2,2). The proposition follows from this. Proposition 2. If X has a chi-square distribution with n degrees of freedom, then the mean of X is X = E(X) = n. If Y/ has a chi-square distribution with n degrees of freedom, then the mean of Y is Y = E(Y) = n. 2.3 - 1 Proof. Since X ~ (n/2,2) it follows from Proposition 2 of section 2.2 that X = (n/2)(2) = n. One has Y/ = X where X has a chi-square distribution with n degrees of freedom. Therefore E(Y) = E(X) = E(X) = n. Proposition 3. If X has a chi-square distribution with n degrees of freedom, then the variance of X is X2 = E((X - X)2) = 2n. If Y/ has a chi-square distribution with n degrees of freedom, then the variance of Y is Y2 = 2n2. Proof. Since X ~ (n/2,2) it follows from Proposition 3 of section 2.2 that X2 = (n/2)(22) = 2n. One has Y/ = X where X has a chi-square distribution with n degrees of freedom. Therefore Y2 = 2X2 = 2n2. Proposition 4. If f(t) is given by (1) then for t > 0 one has f(t) has a single local maximum at t = n - 2 if m > 2. f(t) is strictly decreasing for t > 0 if m 2 Proof. Since X ~ (n/2,2) this follows from Proposition 4 of section 2.2. Proposition 5. If X and Y are independent chi-square random variables with n and p degrees of freedom respectively, then X + Y is a chi-square random variable with n + p degrees of freedom. Proof. Since X ~ (n/2,2) and Y ~ (p/2,2), it follows from Proposition 5 of section 2.2 that X + Y ~ ((n+p)/2,2). The proposition follows from this. Proposition 6. If X has a chi-square distribution with n degrees of freedom, then the Laplace transform L(s) and moment generating function M(r) of X are given by 2.3 - 2 1 L(s) = (1 + 2s)n/2 1 M(r) = (1 - 2r)n/2 Proof. Since X ~ (n/2,2) this follows from Proposition 6 of section 2.2. Proposition 7. Let Z1, …, Zn be independent normal random variables with mean zero and standard deviation one and let S = Z12 + … + Zn2. Then S has a chi square distribution with n degrees of freedom. Proof. First consider the case n = 1, i.e. S =Z2 where Z is a normal random variable with mean zero and standard deviation one. Let F(s) = Pr{S s} be the distribution function of S. Then for s > 0 one has s F(s) = Pr{U2 s} = Pr{ - s U s } = g(u) du - s where g(u) is the density function of Z. Therefore, the density function of S is (19) dF e-s/2 s1/2-1e-s/2 1 -1 1 f(s) = ds = g( s) - g( s) = g( s) = = 1/2 2 (1/2) 2 s 2 s s 2s This is a gamma random variable with parameters m = 1/2 and = 2.so the result is true for n = 1. The case of general n follows from Proposition 5. Corollary 8. Let V1, …, Vn be independent normal random variables with mean zero and standard deviation and let W = V12 + … + Vn2. Then W/2 has a chi-square distribution with n degrees of freedom and W is a gamma random variable with parameters n/2 and 22. Proof. Vj = Zj where the Zj are independent normal random variables with mean zero and standard deviation 1. So W = 2S where S = Z12 + … + Zn2. By Proposition 7, S has a chi-square distribution with n degrees of freedom and the result follows. 2.3 - 3 Corollary 9. Let U1, …, Un be independent normal random variables with mean and standard deviation and let W = (U1 - )2 + … + (Un - )2. Then W/2 has a chi-square distribution with n degrees of freedom. Proof. W = V12 + … + Vn2 where Vj = Uj - . The Vj are independent normal random variables with mean zero and standard deviation . So the result follows from Corollary 8. Theorem 10. Let U1, …, Un be independent normal random variables with the same _ mean and standard deviation and let U = (U1 + … + Un)/n and _ _ W = (U1 - U)2 + … + (Un - U)2. Then W has a chi square distribution with n – 1 degrees of freedom and W is a gamma random variable with parameters (n-1)/2 and 22. The proof of this is more involved; see Rao [11, p. 147]. Let X have a chi-square distribution with n degrees of freedom and let t t (n/2)-1 -s/2 s e G(t) = G(t;n) = ds f(s;n) ds = 2n/2 (n/2) 0 0 be its cummulative distribution function and let m-1 -s m(t) = s e ds t be the upper incomplete gamma function and t sm-1e-s ds m(t) = 0 be the lower incomplete gamma function. 2.3 - 4 Theorem 11. (t/2) G(t;m,) = n/2(n/2) Proof. Since X ~ (n/2,2) this follows from Proposition 7 of section 2.2. 2.3 - 5