Scribd Logo
Upload

Welcome to Scribd!

  • 81 views

    Creative Commons Attribution-Noncommercial-Sharealike License

    Uploaded by

    Sadick Adriano
    statistics

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd

    Creative Commons Attribution-Noncommercial-Sharealike License

    Uploaded by

    Sadick Adriano
    81 views18 pages
    statistics

    Original Title

    Lecture 6

    Copyright

    © © All Rights Reserved

    Available Formats

    PDF, TXT or read online from Scribd

    Did you find this document useful?

    Is this content inappropriate?

    statistics

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd
    Download as pdf or txt
    81 views18 pages

    Creative Commons Attribution-Noncommercial-Sharealike License

    Uploaded by

    Sadick Adriano
    statistics

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd
    Download as pdf or txt
    of 18

    This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike License.

    Your use of this


    material constitutes acceptance of that license and the conditions of use of materials on this site.

    Copyright 2009, The Johns Hopkins University and Saifuddin Ahmed. All rights reserved. Use of these materials
    permitted only in accordance with license rights granted. Materials provided AS IS; no representations or
    warranties provided. User assumes all responsibility for use, and all liability related thereto, and must independently
    review all materials for accuracy and efficacy. May contain materials owned by others. User is responsible for
    obtaining permissions for use from third parties as needed.

    Methods in Survey Sampling


    Biostat 140.640

    weighting

    Saifuddin Ahmed, PhD


    Dept. of Biostatistics

    Sample Weighting
    The purpose of weighting sample data is to
    improve representativeness of the sample in
    terms of:
    size
    Distribution, and
    characteristics of the study population.

    By introducing sampling weight, the estimation is


    carried out in such a way that the estimates
    reflect the actual distribution/characteristics of
    the population.

    Population total estimate from sample:

    T ( y ) = N * y,
    where y is the estimated mean
    We may rewrite this as:

    N
    T ( y) = * n * y
    n

    By multiplying the inverse


    of sample selection
    probability with a sample
    estimate, we may derive a
    population level estimate.

    An example

    N = 300,000
    n = 300
    Sampling fraction (f) = 300/300,000 = 0.1
    y_bar(mean) = 0.5 (50% of children are
    immunized)

    Total Y = (1/f)*(n*y_bar) = (1/0.1)*300*0.5


    =150,000

    The inverse of the selection probability


    (w=1/f=N/n) is used as a weight to make the
    estimate equal to the population
    Even when the sample is selected with different
    selection probabilities, the same principle could
    be extended, and the weight is inversely
    proportional to the unit selection probabilities.
    The weights as measured from the inverse of
    the selection probabilities are called design
    weights.

    Inverse of the sample selection probability


    is the design weight.
    The sum of sampling weights is equal to
    the population size N

    ta v005
    weight |
    Freq.
    Percent
    Cum.
    ------------+----------------------------------199788 |
    326
    5.15
    5.15
    203687 |
    107
    1.69
    6.84
    342747 |
    91
    1.44
    8.27
    352024 |
    312
    4.93
    13.20
    473247 |
    262
    4.14
    17.33
    571152 |
    248
    3.91
    21.25
    726240 |
    267
    4.21
    25.46
    728423 |
    128
    2.02
    27.48
    792283 |
    81
    1.28
    28.76
    836419 |
    187
    2.95
    31.71
    845095 |
    294
    4.64
    36.35
    851062 |
    138
    2.18
    38.53
    907765 |
    384
    6.06
    44.59
    930111 |
    242
    3.82
    48.41
    967833 |
    134
    2.12
    50.53
    979842 |
    391
    6.17
    56.70
    1005824 |
    348
    5.49
    62.19
    1026552 |
    67
    1.06
    63.25
    1068896 |
    145
    2.29
    65.54
    1083215 |
    496
    7.83
    73.37
    1095179 |
    79
    1.25
    74.62
    1106982 |
    120
    1.89
    76.51
    1224149 |
    417
    6.58
    83.09
    1312089 |
    76
    1.20
    84.29
    1465329 |
    884
    13.95
    98.25
    1608275 |
    111
    1.75
    100.00
    ------------+----------------------------------Total |
    6,335
    100.00

    When weighting is needed?


    Different selection probabilities
    Need of improving the variance estimation
    in case of:
    High non-response and coverage errors
    Departures from representative(probability)
    sampling
    Small sample sizes

    Why do we weight?
    To improve representativeness of the
    sample in terms of size, distribution and
    characteristics of the study population.
    to ensure that the estimates are simple
    unbiased estimates.

    Weighting, however, is not


    without problems.

    Advantages of such self-weighed


    sampling designs
    Weighting increases complexity of the survey operations
    Haphazard weights which are not related to population
    variances increase variance of the survey estimates.
    Weighting may reduce the flexibility and ease with which
    the same sample may be used for diverse purposes and
    different surveys.
    Repeated sampling from the same list is straight forward
    with self-weighting, but the selection probabilities
    become more complex if previous selections from the
    frame were not with equal probabilities.
    Self-weighted samples are more readily understood and
    accepted by the non-statistical and the general public.
    Moderate departures from self-weighting have small
    effect on variances, and weighting is only suggested
    when the estimates involves large departures with self
    weighted samples.

    When to weight?
    For appropriate representativeness of
    smaller domain (e.g., residence,
    geographical territories, race, sex )
    Fixed sample size for different
    geographical areas
    Defect in sampling frame, errors in
    selection, high non-response

    Disadvantages
    Increased complexity
    Inconvenience
    Increased variance with haphazard
    weighting
    Analysis /statistical programming
    Cost
    Higher possibilities of error
    Increased bias with incorrect weight

    Major reasons of weighting in


    practice
    Departures from self-weighting
    Frame defects
    Nonresponse

    Region = A
    N =50,000
    n=500
    P=0.5

    Region = B
    N =15,000
    n =500
    P=0.6

    (# immunized=500*.5
    =250)

    (# immunized=500*6
    =300)

    Region = C
    N =30,000
    n=500
    P=0.7

    Region = D
    N =5,000
    n=500
    P=0.8

    (# immunized=500*.7
    =350)

    (# immunized=500*.8
    =400)

    Then, sample based immunization rate in the country is:


    (250+300+350+400)/(500+500+500+500)= 1300/2000 = 65%

    Nh
    wh =
    nh
    50,000
    wA =
    = 100
    500
    15,000
    wB =
    = 30
    500
    30,000
    wC =
    = 60
    500
    5,000
    wD =
    = 10
    500

    t=

    H nh

    h =1i =1

    wh yih

    t
    y=
    wh
    . di (100*.5+30*.6+60*.7+10*.8)/ (100+30+60+10)
    .59
    . di (100*250+30*300+60*350+10*400)/ (50000+15000+30000+5000)
    .59

    Weighted based unbiased estimate = 59%

    You might also like