+ All Categories
Home > Documents > Review of Some Basic Statistical Concepts and the Horvitz ...faculty.nps.edu/rdfricke/OA4109/Lecture...

Review of Some Basic Statistical Concepts and the Horvitz ...faculty.nps.edu/rdfricke/OA4109/Lecture...

Date post: 04-Aug-2020
Category:
Upload: others
View: 1 times
Download: 0 times
Share this document with a friend
24
Review of Some Basic Statistical Concepts and the Horvitz- Thompson Estimator Professor Ron Fricker Naval Postgraduate School Monterey, California 2/1/13 1 Reading Assignment: Scheaffer, Mendenhall, Ott, & Gerow, Chapter 3
Transcript
Page 1: Review of Some Basic Statistical Concepts and the Horvitz ...faculty.nps.edu/rdfricke/OA4109/Lecture 6-2 -- Some... · Review of Some Basic Statistical Concepts and the Horvitz-Thompson

Review of Some Basic Statistical Concepts and the Horvitz-

Thompson Estimator!Professor Ron Fricker!

Naval Postgraduate School!Monterey, California!

2/1/13 1

Reading Assignment:!Scheaffer, Mendenhall, Ott, & Gerow,!

Chapter 3!

Page 2: Review of Some Basic Statistical Concepts and the Horvitz ...faculty.nps.edu/rdfricke/OA4109/Lecture 6-2 -- Some... · Review of Some Basic Statistical Concepts and the Horvitz-Thompson

Goals for this Lecture!

•  Compare and contrast classical statistical assumptions to survey data requirements!–  Infinite vs. finite populations!

•  Estimators for infinite and finite populations!–  Particularly Horvitz-Thompson estimator!

•  Review:!–  Sampling distributions!–  Central Limit Theorem!–  Margin of error!

2/1/13 2

Page 3: Review of Some Basic Statistical Concepts and the Horvitz ...faculty.nps.edu/rdfricke/OA4109/Lecture 6-2 -- Some... · Review of Some Basic Statistical Concepts and the Horvitz-Thompson

Purpose of Survey Analysis:Statistical Inference!

•  Values calculated from survey data (i.e., means and standard deviations) are statistics !

•  Statistics are estimates of the true values of population values (or parameters)!–  They’re unlikely to correspond exactly to the

values had the entire population been surveyed!•  Whole point of a survey is to use the sample

data to infer back to the entire population!ü Can be relatively easy to very complicated

depending on sampling design!

2/1/13 3

Page 4: Review of Some Basic Statistical Concepts and the Horvitz ...faculty.nps.edu/rdfricke/OA4109/Lecture 6-2 -- Some... · Review of Some Basic Statistical Concepts and the Horvitz-Thompson

Classical Statistical Assumptions vs. Survey Practice / Requirements!

•  Classic statistical methods assume:!–  Population is of infinite size (or so large as to be

essentially infinite)!–  Sample size is a small fraction of the population!–  Sample is drawn from the population via SRS!

•  In surveys:!–  Population always finite (though may be huge)!–  Sample could be sizeable fraction of the

population!•  “Sizeable” is roughly > 5%!

–  Sampling may be complex!

2/1/13 4

Page 5: Review of Some Basic Statistical Concepts and the Horvitz ...faculty.nps.edu/rdfricke/OA4109/Lecture 6-2 -- Some... · Review of Some Basic Statistical Concepts and the Horvitz-Thompson

Summarizing Population Information: Infinite Population Case!

•  Consider a population of integers that are equally likely to be 0, 1,…, 8, 9!

•  That is, !

•  The distribution (probability mass function) can be depicted as:!

2/1/13 5

p(0) = p(1) == p(8) = p(9) = 1

10

0 1 2 3 4 5 6 7 8 9

Prob

abilit

y! 0.1!

0.0!

Page 6: Review of Some Basic Statistical Concepts and the Horvitz ...faculty.nps.edu/rdfricke/OA4109/Lecture 6-2 -- Some... · Review of Some Basic Statistical Concepts and the Horvitz-Thompson

Summarizing Population Information: Infinite Population Case!

•  Summarize the population using the expected value (“mean”) and the variance:!

•  For the example, the mean is!!!

2/1/13 6

( ) ( )y

E Y yp yµ = =∑2 2 2Var( ) ( ) ( ) ( )

yY E Y y p yσ µ µ= = − = −∑

9 9

0 0

1 1( ) (45) 4.510 10y y

y p y yµ= =

= = = =∑ ∑

Page 7: Review of Some Basic Statistical Concepts and the Horvitz ...faculty.nps.edu/rdfricke/OA4109/Lecture 6-2 -- Some... · Review of Some Basic Statistical Concepts and the Horvitz-Thompson

Summarizing Population Information: Infinite Population Case!

•  And!

•  So,!2/1/13 7

σ 2 = ( y − µ)2 p( y)y=0

9

= 110

( y − 4.5)2

y=0

9

= 110

0− 4.5( )2++ 9− 4.5( )2⎡

⎣⎢⎤⎦⎥

= 110

82.5⎡⎣ ⎤⎦ = 8.25

8.25 2.9σ = =

Page 8: Review of Some Basic Statistical Concepts and the Horvitz ...faculty.nps.edu/rdfricke/OA4109/Lecture 6-2 -- Some... · Review of Some Basic Statistical Concepts and the Horvitz-Thompson

Estimating Population Information: Infinite Population Case!

•  But, we only observe a sample from the population:!

•  Estimate with and with!

•  Why these? They have good statistical properties, such as they’re unbiased: and !

2/1/13 8

1,..., ny y

1

1 n

ii

y yn =

= ∑( )22

1

11

n

ii

s y yn =

= −− ∑

( )E Y µ= ( )2 2E S σ=

µ σ2

Page 9: Review of Some Basic Statistical Concepts and the Horvitz ...faculty.nps.edu/rdfricke/OA4109/Lecture 6-2 -- Some... · Review of Some Basic Statistical Concepts and the Horvitz-Thompson

Estimating Population Information: Infinite Population Case!

•  Also we can derive the standard error of the mean:!

•  And we can estimate the standard error of the mean with!

•  These are important quantities for inference!

2/1/13 9

s.e. Y( ) = Var Y( )

n= σ 2

n= σ

n

s.e. Y( ) = Var Y( )

n= s2

n= s

n

Page 10: Review of Some Basic Statistical Concepts and the Horvitz ...faculty.nps.edu/rdfricke/OA4109/Lecture 6-2 -- Some... · Review of Some Basic Statistical Concepts and the Horvitz-Thompson

Key Idea!

•  Probability distributions are models of reality!–  They assume that the population is so large, and

the sample is so small with respect to the population, that each draw of an observation into the sample has no effect on the probability of drawing the next and future observations!

–  So we can ignore issues like whether the observations are drawn with or without replacement!

•  When the population is finite and sampling is without replacement then this is no longer true!

2/1/13 10

Page 11: Review of Some Basic Statistical Concepts and the Horvitz ...faculty.nps.edu/rdfricke/OA4109/Lecture 6-2 -- Some... · Review of Some Basic Statistical Concepts and the Horvitz-Thompson

Summarizing Population Information: Finite Population Case!

•  In your statistics classes, everything was based on the infinite population case!

•  In surveys, populations can be finite:!•  Consider the situation where you will choose

n elements out of the N with probabilities , perhaps different on each draw!

•  How to estimate the population total ?!

•  An unbiased estimator is!

2/1/13 11

{ }1,..., Nu u

{ }1,..., nδ δ

1

N

iiuτ

=

=∑

1

1ˆn

i

i i

yn

τδ=

= ∑

Page 12: Review of Some Basic Statistical Concepts and the Horvitz ...faculty.nps.edu/rdfricke/OA4109/Lecture 6-2 -- Some... · Review of Some Basic Statistical Concepts and the Horvitz-Thompson

Summarizing Population Information: Finite Population Case!

•  To illustrate, imagine you know all the y values (all positive), and thus the total !–  Choose any n items each with probability !

•  This is probability sampling according to size!–  Then!

–  And every estimate is perfect!!•  But there’s no point in sampling and

estimating if you already know all the values!–  So, optimal sampling probabilities not possible!

2/1/13 12

τ̂ = 1n

yiδ ii=1

n

∑ = 1n

yiyi / τi=1

n

∑ = τ

δ i = yi ττ

Page 13: Review of Some Basic Statistical Concepts and the Horvitz ...faculty.nps.edu/rdfricke/OA4109/Lecture 6-2 -- Some... · Review of Some Basic Statistical Concepts and the Horvitz-Thompson

Summarizing Population Information: Finite Population Case!

•  Now, a sampling with replacement case (as an example – it’s not what you’d really do)!–  Choose n items with probability each!–  Then!

2/1/13 13

( ) ( )1 1 1

1

1 1 1

1 1 1ˆ1/ 1/

11 1 1

1/

n n nii

i i ii

N

jn n Nj

ji i j

E yyE En n N n N

uN

u nn N n n

µτδ

τ τ

= = =

=

= = =

⎛ ⎞= = =⎜ ⎟

⎝ ⎠

= = = × × =

∑ ∑ ∑

∑∑ ∑∑

δ i = 1 N

Page 14: Review of Some Basic Statistical Concepts and the Horvitz ...faculty.nps.edu/rdfricke/OA4109/Lecture 6-2 -- Some... · Review of Some Basic Statistical Concepts and the Horvitz-Thompson

Summarizing Population Information: Finite Population Case!

•  For example, consider the population where!–  Pr(pick 1) = 0.1 = !–  Pr(pick 2) = 0.1 =!–  Pr(pick 3) = 0.4 =!–  Pr(pick 4) = 0.4 =!

•  Note the population total is =1+2+3+4=10!•  Now, imagine we are going to randomly

choose two elements from the population again with replacement!

2/1/13 14

{ } { }1 2 3 4, , , 1,2,3,4u u u u = δ1

δ 2

δ3

δ 4

τ

Page 15: Review of Some Basic Statistical Concepts and the Horvitz ...faculty.nps.edu/rdfricke/OA4109/Lecture 6-2 -- Some... · Review of Some Basic Statistical Concepts and the Horvitz-Thompson

Summarizing Population Information: Finite Population Case!

•  Then if we happen to choose a 1 and a 2, our estimate of the total is!

•  Also, the variance of the total is estimated as !

2/1/13 15

1

1 1 1 2 1ˆ (10 20) 152 0.1 0.1 2

ni

i i

yn

τδ=

⎛ ⎞= = + = + =⎜ ⎟⎝ ⎠∑

Var τ̂( ) = 1n

i1

n−1yi

δ i

− τ̂⎛

⎝⎜⎞

⎠⎟

2

i=1

n

∑ = 12

i1

2−1yi

δ i

− τ̂⎛

⎝⎜⎞

⎠⎟

2

i=1

2

= 12

10.1

−15⎛⎝⎜

⎞⎠⎟

2

+ 20.1

−15⎛⎝⎜

⎞⎠⎟

2⎡

⎣⎢⎢

⎦⎥⎥= 1

225+ 25⎡⎣ ⎤⎦ = 25

Page 16: Review of Some Basic Statistical Concepts and the Horvitz ...faculty.nps.edu/rdfricke/OA4109/Lecture 6-2 -- Some... · Review of Some Basic Statistical Concepts and the Horvitz-Thompson

Summarizing Population Information: Finite Population Case!

•  Table gives all possible outcomes!

•  From this, we see !

–  Unbiased!!•  Also,!

!2/1/13 16

E τ̂( ) = 15(0.02)+ 354

(0.08)

++10(0.16) = 10

Var τ̂( ) = 15−10( )2

(0.02)+ 354−10

⎛⎝⎜

⎞⎠⎟

2

(0.08)++ (10−10)2(0.16) = 6.25

.02

ü  Under sampling with replacement, the estimator is unbiased for any choice of s!δ

Page 17: Review of Some Basic Statistical Concepts and the Horvitz ...faculty.nps.edu/rdfricke/OA4109/Lecture 6-2 -- Some... · Review of Some Basic Statistical Concepts and the Horvitz-Thompson

Summarizing Population Information: Finite Population Case!

•  What about sampling without replacement?!–  That’s what most surveys do!

•  Define the as the average probability the ith observation is selected: !

•  Often it’s expressed as a weight, , so!

2/1/13 17

1 1 1

1 1ˆ/

n n ni i i

i i ii i i

y y yn n n

τδ π π= = =

= = =∑ ∑ ∑

1

ˆn

i iiw yτ

=

=∑ wi = 1 π i

δ i

δ i = π i n

Page 18: Review of Some Basic Statistical Concepts and the Horvitz ...faculty.nps.edu/rdfricke/OA4109/Lecture 6-2 -- Some... · Review of Some Basic Statistical Concepts and the Horvitz-Thompson

Estimator Still Unbiased!

2/1/13 18

E τ̂ e( ) = (6+8+10+10+12+14) / 6 = 10

E τ̂ u( ) = 0.0222×12.2748++ 0.5333× 9.2652 = 10

Page 19: Review of Some Basic Statistical Concepts and the Horvitz ...faculty.nps.edu/rdfricke/OA4109/Lecture 6-2 -- Some... · Review of Some Basic Statistical Concepts and the Horvitz-Thompson

Horvitz-Thompson Estimators!

•  Generally referred to as the Horvitz-Thompson estimator!–  To estimate mean:!

•  Estimator is particularly useful in complex sampling where is the probability a respondent is selected from sampling frame!–  Probability can vary by each respondent

depending on the sampling scheme!–  Probability can also incorporate the probability of

nonresponse!–  wi has a nice interpretation we will discuss later!

2/1/13 19

1 1

ˆ 1 1 1ˆn n

i i ii i i

w y yN N Nτµ

π= =

= = =∑ ∑

π i

Page 20: Review of Some Basic Statistical Concepts and the Horvitz ...faculty.nps.edu/rdfricke/OA4109/Lecture 6-2 -- Some... · Review of Some Basic Statistical Concepts and the Horvitz-Thompson

Other Important Concepts:Remember Sampling Distributions!

•  Abstract from people and surveys to random variables and their distributions!

•  Sampling distribution is the probability distribution of a sample statistic!

2/1/13 20

Distribution individual obs with standard

deviation

0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9

Individual Mean of n Sampling

distribution of means of n obs

Standard error:

X nσ σ=

σ

Page 21: Review of Some Basic Statistical Concepts and the Horvitz ...faculty.nps.edu/rdfricke/OA4109/Lecture 6-2 -- Some... · Review of Some Basic Statistical Concepts and the Horvitz-Thompson

Remember: Central Limit Theorem (CLT) !

•  Let X1, X2, …, Xn be a random sample from any distribution with mean and standard deviation !

•  For large sample size n, the distribution of the sample mean has approximately a normal distribution !–  with mean , and!–  standard error!

•  The larger the value of n, the better the approximation!

2/1/13 21

σ n

µσ

µ

Page 22: Review of Some Basic Statistical Concepts and the Horvitz ...faculty.nps.edu/rdfricke/OA4109/Lecture 6-2 -- Some... · Review of Some Basic Statistical Concepts and the Horvitz-Thompson

Example: Sums of Dice Rolls!

2/1/13 22

Roll of a Single Die

0

20

40

60

80

100

120

1 2 3 4 5 6

Outcome

Freq

uenc

y

Sum of Two Dice

0

100

200

300

400

500

600

700

2 3 4 5 6 7 8 9 10 11 12Sum

Freq

uenc

y

Sum of 5 Dice

0

10

20

30

40

50

60

70

1 3 5 7 9

11

13

15

17

19

21

23

25

Sum

Fre

qu

en

cy

Sum of 10 Dice

0

50

100

150

200

250

300

350

5 8 11 14 17 20 23 26 29 32 35 38 41 44 47 50 53 56 59

Sum

Freq

uenc

y

One roll

Sum of 2 rolls

Sum of 5 rolls

Sum of 10 rolls

Why do we care about dice?!Translate this into the

probability of response on a six-point Likert scale…!

Page 23: Review of Some Basic Statistical Concepts and the Horvitz ...faculty.nps.edu/rdfricke/OA4109/Lecture 6-2 -- Some... · Review of Some Basic Statistical Concepts and the Horvitz-Thompson

What Does “Margin of Error” Mean?!

•  Margin of error is just half the width of a 95 percent confidence interval!

•  Common survey terminology!–  Convention is a 3% margin of error!–  Means a 95% CI is the survey result +/- 3%!

•  To achieve a desired margin of error, must have the right sample size (n)!–  Power calculations are done by statisticians to

figure out the required sample size to achieve a particular margin of error!

2/1/13 23

Page 24: Review of Some Basic Statistical Concepts and the Horvitz ...faculty.nps.edu/rdfricke/OA4109/Lecture 6-2 -- Some... · Review of Some Basic Statistical Concepts and the Horvitz-Thompson

What We Have Just Covered!

•  Compared and contrasted classical statistical assumptions to survey data requirements!–  Infinite vs. finite populations!

•  Estimators for infinite and finite populations!–  Particularly Horvitz-Thompson estimator!

•  Briefly reviewed:!–  Sampling distributions!–  Central Limit Theorem!–  Margin of error!

2/1/13 24


Recommended