MTH 515a: Inference-II

Assignment No. 7: Uniformly Most Powerful Invariant Tests

1. Let X1 , X2 be i.i.d. with a common Lebesgue p.d.f.

fθ (x) = f (x − θ) , x ∈ R, for some θ ∈ Θ = R,

and let fi (·), i = 0, 1, be two known Lebesgue p.d.f.s on R. Let G = {gc : c ∈ R},
where gc (x1 , x2 ) = (x1 + c, x2 + c), x = (x1 , x2 ) ∈ R2 , c ∈ R. Consider testing
H0 : f ≡ f0 against H1 : f ≡ f1 . If f0 ∼ N(0, 1), f1 ∼ C(0, 1) (standard Cauchy
distribution) and α ∈ (0, 1), find UMPI(α) test under G.

2. Let X1 , . . . , Xn be i.i.d. with a common Lebesgue p.d.f.

1 x
fθ (x) = f , x ∈ R, for some θ ∈ Θ = (0, ∞),
θ θ
and let fi (·), i = 0, 1, be two known Lebesgue p.d.f.s on R. Let G1 = {gr : r 6= 0}
and G2 = {gr : r > 0}, where gr (x) = (rx1 , . . . , rxn ), r ∈ R. Consider testing
H0 : f ≡ f0 against H1 : f ≡ f1 .

(a) If f0 ∼ N(0, 1), f1 ∼ DE(0, 1) and α ∈ (0, 1), find UMPI(α) test under G1 ;
(b) If f0 (x) = 1, 0 < x < 1, f1 (x) = 2x, 0 < x < 1, find UMPI(α) test G2 .

3. Let X1 , . . . , Xn be i.i.d. with a common Lebesgue p.d.f.

 
1 x−µ
fµ,σ (x) = f , x ∈ R, for some (µ, σ) ∈ R × (0, ∞),
σ σ
and let fi (·), i = 0, 1, be two known Lebesgue p.d.f.s on R. Let G1 = {gc,r : c ∈ R, r 6=
0} and G2 = {gc,r : c ∈ R, r > 0}, where gc,r (x) = (rx1 + c, . . . , rxn + c), r, c ∈ R.
i −Xn
Consider testing H0 : f ≡ f0 against H1 : f ≡ f1 . Define Wi = XXn−1 −Xn
,i =
1, 2, . . . , n − 2 and W = (W1 , . . . , Wn−2 ).

(a) Show that W is maximal invariant under G1 . Find a maximal invariant statistic
under G2 ;
(b) If f0 ∼ N(0, 1), f1 ∼ U(− 21 , 12 ) and α ∈ (0, 1), Obtain a UMPI(α) test under
G1 .

4. Let X1 , . . . , Xn be i.i.d. N(µ, σ 2 ) with unknown µ ∈ R and σ > 0. Consider

testing H0 : µ = 0 against H1 : µ 6= 0. Let G = {gc : c ∈ R, c 6= 0}, where
gc (x) = (cx1 , . . . , cxn ), c 6= 0.

(a) Show that the testing problem is invariant under G;

(b) Show that UMPI(α) test coincides with UMPU(α);
(c) Find the UMPI(α) using first principle (i.e., without using the result that an
invariant UMPU(α) is also UMPI(α)).
5. Let X1 and X2 be independent random variables with Xi having the p.d.f.
1 − θx
fθi (x) = e i , x > 0,
θi
θ1
with unknown θi > 0, i = 1, 2. Define θ = θ2
.

(a) For testing H0 : θ ≤ 1 against H1 : θ > 1, show that the testing problem is
invariant under G = {gc : c > 0}, where gc (x1 , x2 ) = (cx1 , cx2 ), c > 0, and the
X2
UMPI(α) test rejects H0 when X 1
> 1−α
α
. Find the UMPU(α) test;
(b) For testing H0 : θ = 1 against H1 : θ 6= 1, show that the testing problem is
invariant under G1 = G ∪ {g0 }, where g0 (x1 , x2 ) = (x2 , x1 ) and G is as defined
X2
in (a). Also show that a UMPI(α) test rejects H0 when X 1
> 2−α
α
or X2
X1
α
< 2−α .
Find the UMPU(α) test.

Generalize the findings in (a) and (b) for sample of size n (≥ 3).

6. Let Xi,1 , . . . , Xi,ni , i = 1, 2, be two independent random samples with Xi,1 ∼

N(µi , σ 2 ), µi ∈ R, i = 1, 2, σ > 0,. Show that the problem of testing the null
hypothesis H0 : µ1 = µ2 against the alternative hypothesis H1 : µ1 6= µ2 is invariant
under the group of transformations G = {ga,b : a 6= 0, b ∈ R}, where

ga,b (x1 , x2 ) = (ax1,1 + b, . . . , ax1,n1 + b, ax2,1 + b, . . . , ax2,n2 + b).

Find the UMPI(α) test.

7. Let Xi,1 , . . . , Xi,ni , i = 1, 2, be two independent random samples with Xi,1 ∼

Exp(µi , σi ), µi ∈ R, σi > 0, i = 1, 2. Let ∆0 be a given positive constant. Show
that the problems of testing the null hypothesis H0 : σσ21 ≥ ∆0 against the alter-
native hypothesis H1 : σσ12 < ∆0 is invariant under the group of transformations
G = {ga,b1 ,b2 : a > 0, bi ∈ R, i = 1, 2}, where

ga,b1 ,b2 (x1 , x2 ) = (ax1,1 + b1 , . . . , ax1,n1 + b1 , ax2,1 + b2 , . . . , ax2,n2 + b2 ).

Find the UMPI(α) test. Show that the UMPI(α) test is also the UMPU(α) test.