Statistics Assignment Sample 2 With Solutions
Statistics Assignment Sample 2 With Solutions
Statistics Assignment Sample 2 With Solutions
Go To Answer Directly
info@statisticsassignmenthelper.com
Problem
Instructor: Set #5
Konrad Menzel
Question One
The convolution formula is a useful trick when we are interested in the sum or average
of independent random variables. In the last problem set, we dealt with the random
variable X, below.
e-" forx > 0
f (4= 0 f o r x < O
Now, suppose that X = XI = X2 = . . . = Xk are independent, identically distributed
random variables.
+
1. Using the convolution formula, determine the PDF of fi = ;(x1 X 2 ) Hint:
+
Define Zl = X1 and Z2 = XI X2 and then use the transformation method to
get back to Y2 from 2 2 .
2. Compute its expectation: E[Y2].
+ +
3. Using the convolution formula, determine the PDF for & = ;(XI X 2 X3).
+ +
Hint: Use the hint from part 1 to define Z3 = X1 X2 X3 and perform a
convolution with X3 and Z2 to transform the problem into Z2 and Z3.
4. Compute its expectation: IE[Y3].
5. Using the convolution formula, determine the PDF for Yk= :(x1 X2 . - + + +
Xk) = :ck
Z=I Xk. Hint:
Try to determine a pattern from part 1 and part 3
using the methods from their hints.
7. What does this tell us about the mean of a sample of size Ic? Is this property
specific to the exponential distribution? Explain.
https://www.statisticsassignmenthelper.com/
Question Two
(BainIEngelhardt , p. 228)
Suppose that XI,X2,. . . ,Xk are independent random variables and let Y , = ui (Xi)
for i = 1 , 2 , . . . , k. Show that Yl, Y2, . . . , Yk are independent. Consider only the case
where Xi is continuous and yi = ui(xi) is one-to-one. Hint: If xi = wi(yi) is the
inverse transformation, then the Jacobian has the form
Question Three
Moved to a later problem set.
Question Four
Order statistics are very useful tools for analyzing the properties of samples.
1. Write down the general formula of the pdf and cdf for the kth order statistic of
a sample of size n of a random variable X with CDF F'(x).
Question Five
(BainIEngelhardt p. 229)
Let X1 and X2 be a random sample of size n = 2 from a continuous distribution
with pdf of the form f (x) = 2x if 0 < x < 1 and zero otherwise.
1. Find the marginal pdfs of the smallest and largest order statistics, Yl and Y2.
5. Compute the expectation, IE[X],of a single draw X from f (x). Does the integral
diverge? What does that say about the existence of E[Y,]? Explain.
6. Derive the pdf of the sample range, R = Y, - Yl, for n = 2. Hint: Use par-
tial fractions, searching o n Yahoo for "QuickMathn will help you get the partial
fractions via computer:
http://72.3.253.76:8080/webMathematica5'/quickmath/page.jsp ?sl =algebra&s2=partialfractions&s3= basic
+
(n 1)/2 E N.Express the pdf as a function of just r and y, (eliminate all n's
and k7s).
Question One
The convolution formula is a useful trick when we are interested in the sum or average of
independent random variables. In the last problem set, we dealt with the random variable
X, below.
e-" forx > 0
f (4= 0 f o r x < O
Now, suppose that X = X1 = X2 = - .- = XI, are independent, identically distributed
random variables.
+
1. Using the convolution formula, determine the PDF of & = (xl X2). Hint: Defilze
+
Zl = X1 and Z2 = X1 X2 and then use the transformation method to get back to Y2
from Z2.
Solution to 1: To use the convolution formula, we need the joint PDF of X1 and
X2 and x2 as a function of y2 and xl. The function is x2 = 2y2 - xl. Also, since
they are independent, we can just construct the joint PDF by multiplying the two
marginals, fxl(xl) and fx2(x2). This gives a joint P D F of f (xl, x2) = e-(x1+x2)
for xl > 0 and x2 > 0 and zero otherwise. The convolution formula adapted to
this problem (taking into account the limits) is (where we take into account the
change in variables formula for x2 as a function of xl and y2 using the Jacobian)
3. Using the convolution formula, determine the PDF for & = ;(XI + X2 + X3). Hint:
+ +
Use the hint from part 1 to define Z3 = XI X2 X3 and perform a convolution with
X3 and Z2 to transform the problem into Z2 and Z3.
Solution to 3: Apply the same method to the results from part I, where we
obtained fY2(y2).We write the joint PDF of XI, X2, and X3 as
We'll take a slightly different approach for this problem, in order to help us solve
part 5 later. Define Z2 = X I + X 2 and Z3 = Xl +X2 +X3. Now, we can ignore any
Jacobian transformation and directly apply the convolution formula. We can first
solve for 22's distribution by only using the convolution on the joint distribution
of XI and X2.
Now, we use the convolution on the joint distribution of Z2 and X3 to obtain the
distribution of Z3:
But, we were interested in Y3, not Z3. SO,we perform a simple continuous inverse
transformation of random variables using Z3 = 3Y3:
5. Using the convolution formula, determine the PDF for Yk = i(xl + X2+ .. + X k ) =
X k . Hint: Trg to determine a pattern from part 1 and part 3 using the methods
from their hints.
Solution to 5: We use the convolution formula using the sequence determined from
+
part 3. We define more partial sums Z1 = XI, Z2 = X1 X2, Z3 = X1 X2 +X3, +
... , Zk = c:=, Xi. This allows us to easily construct the inverse functions for
X ' s as function of 2 ' s as they are just the difference between adjacent 2's. This
gives the following Jacobian:
which, conveniently has a determinant of 1. So, we can just ignore the Jacobian
component. We now write the joint distribution:
Since the joint distribution of all x's is just a function of zk, we would say that
xk is a sufficient statistic for the distribution of the sum of k independent x's.
Now, all we need to do is integrate out each of xl through xk-1 in order to get the
marginal of zk. F'rom part 3, we learned that the convolution will yield:
and
which, if repeated, will yield the general formula for Zk:
2.5
-
- Normal approx. k=10
Normal approx. k=50
h
Y
1.5
2
0.5
0 I
I I \ ?
\%F
- ---
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
7. What does this tell us about the mean of a sample of size k? Is this property specific
t o the exponential distribution? Explain.
Question Two
(BainIEngelhardt , p. 228)
Suppose that X I , X2,. . . , Xk are independent random variables and let Y , = ui(Xi) for
i = 1 , 2 , . . . , k. Show that Yl, Y2,.. . , Yk are independent. Consider only the case where Xi is
continuous and yi = ui(xi) is one-to-one. Hint: If xi = wi(yi) is the inverse transformation,
then the Jacobian has the form
k .
Solution: We can write Xi = wi (Yl, Y2,..., Yk) where wi is the inverse function of
( 9 )
ui(.), i.e. wi(.) = u i l ( - ) . Since ui is only a function of Xi, the inverse transform
will only be a function of Y,, or Xi = wi(Y,). With this intuition, we can show the
independence by showing that pdfs can be factored:
where
since each wi is only a function of yi and not y-i. So, all of the off diagonal derivatives
are zero. Thus, the Jacobian can be easily computed since we have a diagonal matrix:
k d
J = n i = l &wi (yi)-
With this fact, we can now substitute in all of the pieces:
where we use the independence of the X's and the .Hint about the Jacobian in the first
line, the
The absolute value of a product is equal to the product of the absolute values in
the second line, commutative and associative properties of multiplication in the third
line, and the definition of the transformation of a single random variable in the last
line. So, Yl, ..., Yk are independent and we are done.
Question Three
Moved to a later problem set.
Question Four
Order statistics are very useful tools for analyzing the properties of samples.
1. Write down the general formula of the pdf and cdf for the kth order statistic of a sample
of size n of a random variable X with CDF F x ( x ) .
Solution to 1: The general formula for the kth order statistic, Yk,with pdf f ( x )
( f ( x ) > 0 on a < x < b and zero otherwise) and cdf F ( x ) is:
n!
gk(Yk) =
(k - I ) ! ( n- k)! [ F ( Y ~ ) ] " '[ I - ~ ( y k ) ] ~f -('y n )
if a < yk < b and zero otherwise. The marginal cdf can be written as
Question Five
(BainIEngelhardt p. 229)
Let X 1 and X 2 be a random sample of size n = 2 from a continuous distribution with
pdf of the form f ( x ) = 2x if 0 < x < 1 and zero otherwise.
1. Find the marginal pdfs of the smallest and largest order statistics, Yl and Y2.
Solution to 1: The marginal pdfs of the smallest and largest order statistics from
a sample of size n = 2 can be derived by using the marginal pdf formula, but we
first need the CDF of X:
and
So, the expectation of the first order statistic is . For the second (maximal)
order statistic, we get
Solution to 3: The joint pdf of Yl and Y2 can be obtained from the pdf f (x) (the
joint pdf can be written as g(yl, ..., y,) = n!f (yl). . . f (9,) for n independently
sampled observations from the distribution defined by f (x)) taking into account
the permutations of yl and y2:
Solution to 4: The pdf of the sample range is just a transformation of the joint
PDF, similar to all of the convolutions that we've been doing. Consider the inverse
+
transforms: Yl = Yl and Y2 = R Yl. Then we can use the transform methods:
where J =
we get:
1 1
1 0
= 1 So, applying the formula for the joint pdf from part 3;
~ Y I , R ( Y ~ ,= 8 ~ 1+( ~~ 11 1 )
which we now use to integrate out yl to get the marginal of R. We proceed,
paying careful attention to the change in bounds (draw a picture to convince
yourself-first with Yl and fi and then with R and Yl)
It is worth noting that the expectation for the sample range is equal to the differ-
ence between the expected first and last order statistics (minimal and maximal
order statistics).
Question Six
(BainIEngelhardt p. 229)
Consider a random sample of size n from a distribution with pdf f (x) = -$if 1 5 x < m;
zero otherwise.
1. Give the joint pdf of the order statistics.
a Solution to 2: The pdf of the smallest order statistic is the PDF when we integrate
out all of the other order statistics. We already have a formula for this, so we'll
use it, rather than deriving it all over again. We use k = 1 and n:
n!
gk ( y k ) =
( k - I ) ! ( n - k ) ! [F(Y~)I"' [l - F ( Y ~ ) ] "f -( ~
yx)
Solution to 3: Again, we have another integration to practice the formula for the
expectation of a random variable using our answer from part 2:
5 . Compute the expectation, IE[X], of a single draw X from f (x). Does the integral
diverge? What does that say about the existence of E[Yn]? Explain.
Solution to 5:
= [logXIS"
= 00-0
E[X] = co,
or, in other words, the integral diverges. This suggests that at least one order
statistic cannot have an expectation, even though the smallest order statistic did
since the expectation of X could be constructed from the order statistics as well
by computing the expectation of each and then average over all n order statistics.
By a bounding argument, we can say that since the smallest order statistic has
an expectation, and the total expectation does not exist for X, then the largest
order statistic must not have a finite expectation. While it would be possible
for multiple order statistics to have nonexistent expectation,. I do not determine
which order statistic is the dealbreaker. Feel free to tinker with the formulas to
figure it out for yourself.
6. Derive the pdf of the sample range, R = Yn - Y', for n = 2. Hint: Use partial fractions,
We use the transformation formula, noting that the Jacobian has a determinant
which we can solve very painfully via partial fractions, unless you use a computer.*
like this great online service I found: http: //72.3.253.76:8080/webMathematica3/quickmath/pa
This is the partial fraction decomposition:
which we can now integrate. The solution we obtain for the pdf is
where r > 0. The PDF of the sample range and the first and second order
statistics in a sample of size n = 2 are plotted below, and compared to the kernel
estimates from 1 million samples of size n = 2 to benchmark the theory.
Kernel and Analytic PDFs
Solution t o 7: The expectation of the range cannot exist as the expectation of the
maximal order statistic does not exist, and E[R] = E[&] - EIYl]. Even though
IE[K]exists, we need both to exist to compute it.
8. Give the pdf of the sample median, Y,, assuming that n is odd so that r = ( n + 1 ) / 2 E N.
Express the pdf as a function of just r and y, (eliminate all n's and k's).
-
- (2r -
[ ( r- I ) ! ]
1" (y. ypl
- l).-l
dyr
-
- ( 2- 1 )
[ ( r- I ) ! ]
[( ;-- (Yir- 1 Y ) "
2~-1 1
-
La-72r - 1 (yr -
Yr2' dY.1
= [[,
(27- - I ) !
- l)!I2 (-G
1 (y. -
.:3
03
+
r
1" Y - 1) ( (27- - 1 ( y - 1)-l
[ ( r - 1)!12 3:.
)
A few numerical checks will yield that this is, in fact, the correct formula: for
r = 2 , we get a median of 3 and for r = 3, we get a median of 2.5. For r = 5, the
median is 2.25.