Econ 120A Final Study Guide

In this study guide, I will go over basic ideas we covered this quarter

and provides general solutions on different

questions which

probably shown on final exam. Good Luck on Finals!

In Final Exam, we will have:

2 questions on confidence interval (Chapter 8)

1 question on hypothesis test (Chapter 9.1 &9.2)

3 questions on rest (Chapter 2,4,5,6)

Confidence (Chapter 8)
1. Single Mean Case

Interval

Question

General Question:

Suppose each observation Xi has a normal distribution with

unknown mean and variance . Recall this implies that if you

sample N observations, then Xbar is exactly a normal distribution

with mean and variance . Suppose you sample N observations

and find that Xbar= D

(a) Construct a 95% confidence interval for using a normal

approximation

General Answer:

Za/2 /N

(b) What is the probability that is in the confidence interval? Is that

exact?

General Answer: Since

the

confidence

level is

95%, the

probability that is in the confidence interval is 0.95. The

probability is exact since in this case Xbar is exactly normally

distributed and hence there are not approximations involved.

(c) Someone told you that since only have few observations you

should not use a normal approximation but a t-distribution

instead. Is he right or not?

General Answer: He is incorrect. The confidence interval has

exact confidence level 0.95. Using a t-distribution would lead to a

larger confidence interval with confidence level higher than the

desired 0.95.

Note: If in question variance is unknown and Xi is normally

distributed, we have sample variance Sx, t-distribution will provide

exact confidence level but larger than 0.95. However, if Xi is not

normally

distributed,

both

normal

approximation

and

t-

approximation will give us approximations and it is not necessarily

to choose t-approximation or normal approximation.

Related Questions: Question 2, 8, 15 on Final problems

1. Difference in Two Means, Independent Samples

General Question:

Given 2 variables Xi, Yi. Var(X) and Var (Y). Sample Nx and Ny

observations.

(a) Build a 95% confidence interval for E[X]-E[Y]

General solution: Use formulas to calculate SP and for the normal

approximation xbar-ybarZa/2 Sp1Nx+1Ny

(b)Given your answer to (a), does the elements different between Xi

and Yi effect Xi and Yi?

General solution: See if zero is contained in confidence interval. If it

is, then there may not have effect.

(c) What is length of the confidence interval?

Solution: (xbar-ybar+Za/2 Sp1Nx+1Ny)-( xbar-ybar-Za/2 Sp1Nx+1Ny)

(d) Can you make the length in (c) arbitrarily small by sampling more

observations in Xi (but the same number from Yi)? Explain the

intuition for your answer.

Answer: In general, the length of the confidence interval cannot be

made small just by letting Nx be large as we also need Ny be large.

The intuition is that to learn E[X]-E[Y] we need to learn both E[X](so

Nx must be large) and E[Y](so Ny must be large).

Related Questions: Question 14,27

1. Difference in two means, matched Samples

General Question:

X1 is the level before the treatment, and X2 the level after the

treatment. You are interested in E[X1]-E[X2]=1-2=E[D]

(a) What is Dbar and SD?

Solution: use formula to solve them.

(b) Find confidence interval

Solution: proceed as in the Single Mean Case by working with

Dbar and SD

Related Questions: 9, 20, 35

1. Proportions

General Question: Let P(X=1)= and P(X=0)=1-. We are interested

in

(a) Suppose you poll N people and Find Xbar. Xi=1 is yes. X2=0 if

No. Build a 95% confidence interval for using a normal

approximation.

Solution: XbarZ a/21-N

(b) What is the probability that is in the interval you found in (a)? Is

that exact?

Answer: The probability is approximately 0.95. It is not exact

because in constructing the confidence interval we pretended

Xbar is exactly normal, while in reality it is only approximately so.

(c) Build a 95% confidence interval for using a t-distribution

approximation.

Solution: use same formula as (a) only change Z to t and Talbe IV

to Table V.

(d) Someone claims something like prefer t-approximation since the

interval is larger and hence more likely to cover . Is this a valid

reason? Why or why not?

Solution: It is indeed more likely to cover . However, this is not

necessarily a good thing, as any confidence interval that is larger

will satisfy this. What we ultimately care about is whether the

probability of being there is close to 0.95 or not. Unfortunately,

we do not know which one of the two approximations is more

accurate.

Related Questions: 21

Hypothesis Test Question (Chapter 9.1,9.2)

General QuestionSomeone claims that mean =c and variance .

You sample Xi of N observations and are interested in testing the

claim.

(a) One sided test: State your null hypothesis H0 and the alternative

H1

Answer: H0: =c, H1: >c or <c

(b) Suppose you find Xbar=a, what is the p-value for this test?

Answer: If H1: >c, this corresponds to P(Xbar>a); If <c, then

we use P(Xbar<a) For example, P(Xbar<a)=P((Xbar-)/ (/N)<(a-

)/ (/N))P(Z<(a-)/ (/N))

(c) Would you reject the null hypothesis at v% error level?

Solution: If P-value > error level, we would not reject. If P-value< error

level, reject.

Related Questions: 4, 23

Questions from Chapter 2,4,5,6

Related Questions: 1, 7, 13, 19, 25, 31

(a) What covariance indicates?

If cov(x,y)<0, it suggests that higher x leads to lower y or they

move opposite directions.

If cov(x,y)>0, it means x and y move in same directions.

If Cov(x,y)=0, it means x and y are independent to each other.

1.

Binomial question

Distribution

Question;

Proportion

Related question: 11, 17, 30, 33

General question: Calculate P(Xbar>C) exactly

Answer: use binomial distribution formulas

2. Sample Mean and Variance

General question: Given E[X], E[Y], VAR(X), VAR(Y)

(a) What is VAR(Xbar)? What is Var(Ybar)? What is E[Xbar]? What

is E[Ybar]?

Answer: Var(Xbar)=VAR(X)/Nx, VAR(Ybar)=VAR(Y)/Ny

(b)Linear Transformation question

Note: When you sample huge amount. N becomes lager and

E[Xbar] would close to E[X] and centered in E[X].