bivariate populations

25
Lecture 5 1 Econ 140 Econ 140 Bivariate Populations Lecture 5

Upload: upton

Post on 04-Feb-2016

44 views

Category:

Documents


0 download

DESCRIPTION

Bivariate Populations. Lecture 5. Today’s Plan. Bivariate populations and conditional probabilities Joint and marginal probabilities Bayes Theorem. A Simple E.C.P Example. Introduce Bivariate probability with an example of empirical classical probability (ecp). - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: Bivariate Populations

Lecture 5 1

Econ 140Econ 140

Bivariate Populations

Lecture 5

Page 2: Bivariate Populations

Lecture 5 2

Econ 140Econ 140Today’s Plan

• Bivariate populations and conditional probabilities

• Joint and marginal probabilities

• Bayes Theorem

Page 3: Bivariate Populations

Lecture 5 3

Econ 140Econ 140A Simple E.C.P Example

• Introduce Bivariate probability with an example of empirical classical probability (ecp).

• Consider a fictitious computer company. We might ask the following questions:– What is the probability that consumers will actually buy

a new computer?– What is the probability that consumers are planning to

buy a new computer?– What is the probability that consumers are planning to

buy and actually will buy a new computer?– Given that a consumer is planning to buy, what is the

probability of a purchase?

Page 4: Bivariate Populations

Lecture 5 4

Econ 140Econ 140

certainnull

A Simple E.C.P Example(2)

• Think of probability as relating to the outcome of a random event (recap)

• All probabilities fall between 0 and 1:

1)(0 AP

• Probability of any event A is:

naaaaAnm

AP ...,, with )( 321

Where m is the number of events A and n is the number of possible events

Page 5: Bivariate Populations

Lecture 5 5

Econ 140Econ 140A Simple E.C.P Example(3)

• The cumulative frequency is: 1)( iaP

• The sample space (of a 1000 obs) looks like this:Actually PurchaseYes (b1) No (b2) Total

Plan Yes (a1) 200 50 250to Purchase No (a2) 100 650 750

Total 300 700 1000

• Before we move on we’ll look at some simple definitions

Page 6: Bivariate Populations

Lecture 5 6

Econ 140Econ 140A Simple E.C.P Example(4)• If we have an event A there will be a compliment to A which we’ll call A’ or B • We’ll start computing marginal probabilities

– Event A consists of two outcomes, a1 and a2:

21,aaA – The compliment B consists also of two outcomes, b1 and b2:

21,bbB – two events are mutually exclusive if both events cannot occur– A set of events is collectively exhaustive if one of the events must occur

Page 7: Bivariate Populations

Lecture 5 7

Econ 140Econ 140A Simple E.C.P Example(5)

• Computing marginal probabilities

kBAPBAPBAPA ...)Pr( 21

Where k is some arbitrary large number

• If A = planned to purchase and B=actually purchased:P(planned to buy) = P(planned & did) + P(planned & did not)=

25.01000250

100050

1000200

Actually PurchaseYes (b1) No (b2) Total

Plan Yes (a1) 200 50 250to Purchase No (a2) 100 650 750

Total 300 700 1000

Page 8: Bivariate Populations

Lecture 5 8

Econ 140Econ 140A Simple E.C.P Example(6)

• If the two events, A and B, are mutually exclusive, then

)()()( BPAPAorBP – General rule written as:

– Example: Probability that you draw a heart or spade from a deck of cards

• They’re mutually exclusive eventsP(Heart or Spade) = P(Heart) + P(Spade) – P(Heart + Spade)=

50.021

5226

05213

5213

)()()()( BAPBPAPAorBP

Page 9: Bivariate Populations

Lecture 5 9

Econ 140Econ 140A Simple E.C.P Example(6)

• Probability that someone planned to buy or actually did buy: use the general addition rule:

)()()()( BAPBPAPAorBP • If A is planning to purchase, and B is actually purchasing, we can plug

in the marginal probabilities to find

35.01000350

1000200

1000300

1000250

Actually PurchaseYes (b1) No (b2) Total

Plan Yes (a1) 200 50 250to Purchase No (a2) 100 650 750

Total 300 700 1000

Joint Probability: P(A and B): Planned and Actually Purchased

Page 10: Bivariate Populations

Lecture 5 10

Econ 140Econ 140Conditional Probabilities

• Lets leave the example for a while and consider conditional probabilities.

• Conditional probabilities are represented as P(Y|X)

• This looks similar to the conditional mean function:

Xn

Y

• We’ll use this to lead into regression line inference, and then we’ll look at Bayes theorem

Page 11: Bivariate Populations

Lecture 5 11

Econ 140Econ 140Conditional Probabilities (2)

• Probabilities will be defined as

Kk

Jj

kYYjXXPjkp

,...1

,...1

),(

• If we sum over j and k, we will get 1, or:

j k jkp 1

• We define the conditional probability as f (X|Y)– This is read “a function of X given Y”– We can define this as:

YXf |Y ofy probabilit MarginalY& X ofy probabilitJoint

Page 12: Bivariate Populations

Lecture 5 12

Econ 140Econ 140Conditional Probabilities (3)

• Similarly we can define f (Y|X):

• Looking at our example spreadsheet, we have a sample of weekly earnings and years of education: L5_1.XLS.

• There are two statements on the spreadsheet that will clarify the difference between a joint and conditional probabilities

XYf |X ofy probabilit MarginalX& Y ofy probabilitJoint

Page 13: Bivariate Populations

Lecture 5 13

Econ 140Econ 140Conditional Probabilities (4)

• The joint probability is a relative frequency and it asks: – How many people earn between $600 and $799 and have 10 years of education?

• The conditional probability asks: – How many people earn between $600 and $799 given they have 10 years of education?

• On the spreadsheet I’ve outlined the cells that contain the highest probability in each completed years of education– There’s a pattern you should notice

Page 14: Bivariate Populations

Lecture 5 14

Econ 140Econ 140Conditional Probabilities (5)

• We can use the same data to graph the conditional mean function– the graph shows the same pattern we saw in the outlined cells– The conditional probability table gives us a small distribution

around each year of education

Page 15: Bivariate Populations

Lecture 5 15

Econ 140Econ 140Conditional Probabilities (6)

• To summarize, conditional probabilities can be written as

)|(X ofy probabilit Marginal

Y& X ofy probabilitJoint )(

)&(YXf

XPYXP

– This is read as “The probability of X given Y”– For example: The probability that someone earns between $200 and $300, given that he/she has completed 10 years of

education• Joint probabilities are written as P(X&Y)

– This is read as “the probability of X and Y”– For example: The probability that someone earns between $200 and $300 and has 10 years of education

Page 16: Bivariate Populations

Lecture 5 16

Econ 140Econ 140A Marketing Example

• Now we’ll look at joint probabilities again using the marketing example from earlier in the lecture.

• We will look at:– Marginal probabilities P(A) or P(B)– Joint probabilities P(A&B)– Conditional probabilities

)()&(

BPBAP

Page 17: Bivariate Populations

Lecture 5 17

Econ 140Econ 140Marketing Example(2)

• Here’s the matrix

Actually PurchaseYes No Total

Plan Yes 200 50 250to Purchase No 100 650 750

Total 300 700 1000

• Let’s look at the probability you purchased a computer given that you planned to purchase:

%808.250200purchase) toplanned | purchased P(actually•

• The joint probability that you purchased and planned to purchase: 200/1000 = .2 = 20%

Page 18: Bivariate Populations

Lecture 5 18

Econ 140Econ 140Marketing Example (3)

• We can also represent this in a decision tree

Actually

Purchased

Actually

Purchased

P(AandB)= 200/1000

P(A’andB)= 100/1000

P(A’andB’)= 650/1000

P(AandB’)= 50/1000

Marginal Probabilities

Joint Probabilities

B

A

B

Page 19: Bivariate Populations

Lecture 5 19

Econ 140Econ 140Statistical Independence

• Two events exhibit statistical independence if

P(A|B) = P(A)

• We can change our marketing matrix to create a situation of statistical independence:

Actually PurchaseYes No Total

Plan Yes 75 175 250to Purchase No 225 525 750

Total 300 700 1000

Note: all we did was change the joint probabilities

30.0025.

075.

1000250

100075

)|( BAP 30.1000

300)( AP

30.0)()|( APBAP

Page 20: Bivariate Populations

Lecture 5 20

Econ 140Econ 140Sampling w/ and w/o Replacement

• How would sampling with and without replacement change our probabilities?

• If we have 20 markers (14 blue and 6 red)

– What’s the probability that we pick a red pen?

P(BR)=6/20

– If we replace the pen after every draw, what’s the probability that we pick red twice in a row?

(6/20)(6/20)=36/400 = .09 = 9%

– What’s the probability of drawing two reds in a row if we don’t replace after each draw?

(6/20)(5/19) = 30/380 =.079 = 7.9%

Page 21: Bivariate Populations

Lecture 5 21

Econ 140Econ 140Bayes Theorem

• With decision trees we had to know the probabilities of each event beforehand

• Using Bayes we can update using complement probabilities

• Consider the multiplication of independent events:

)()|(

)()()&(

APBAP

BPAPBAP

• The marginal probability rule says:)(...)2()1()( kBAPBAPBAPAP

Page 22: Bivariate Populations

Lecture 5 22

Econ 140Econ 140Bayes Theorem (2)

• Because of independence we can write P(A) another way)()|()2()2|()1()1|()( kBPkBAPBPBAPBPBAPAP

• We can now write our conditional probability function as:

)(

)()|()|(

AP

BPBAPBAP

)()|()()|()()|()()|(

2211 kk BPBAPBPBAPBPBAPBPBAP

• Plugging in our expression for P(A) gives us Bayes Theorem:

Page 23: Bivariate Populations

Lecture 5 23

Econ 140Econ 140Bayes Theorem (3)

• Think of the Bayes Theorem as probability in reverse

– You can update your probabilities in light of new information

• Suppose you have a product with a known probability of success

P(success) = P(S) = 0.4

P(failure) = P(S’) = 0.6

• We also know that a consumer group will write either a favorable or unfavorable report on the product

P(F|S) = 0.8 P(F|S’) = 0.3

Page 24: Bivariate Populations

Lecture 5 24

Econ 140Econ 140Bayes Theorem (4)

• Given our information, we want to find the probability that the product will be successful given a favorable report

P(S|F)

• In this case, Bayes says

)'()'|()()|(

)()|()|(

SPSFPSPSFP

SPSFPFSP

• We can plug values into the above equation to find

%6464.50.

32.

)6.0)(3.0()4.0)(8.0(

)4.0)(8.0()|(

FSP

• We can use the theorem to update the probability of a successful product given that the product gets a favorable report

Page 25: Bivariate Populations

Lecture 5 25

Econ 140Econ 140Recap

• We’ve seen how we can calculate marginal, joint, and conditional probabilities

– Computer company example

– Spreadsheet: L5_1.XLS

• We talked about statistical independence

• We’ve seen how Bayes Theorem allows us to update our priors