Docstoc

HOUSEHOLD SURVEYS AT STATSSA

Document Sample
HOUSEHOLD SURVEYS AT STATSSA Powered By Docstoc
					     HOUSEHOLD SURVEYS AT
           STATSSA
• OCTOBER HOUSEHOLD SURVEYS (OHS)
  1993/4 – 1999
• HOUSEHOLD INCOME AND EXPENDITURE
  SURVEYS EVERY 5 YEARS: 1990, 1995, 2000 AND
  2005
• SPLIT BETWEEN LABOUR FORCE SURVEYS AND
  GENERAL HOUSEHOLD SURVEYS IN 2000
• FIRST BI-ANNUAL LABOUR FORCE SURVEY (LFS)
  IN FEBRUARY 2000
• FIRST ANNUAL GENERAL HOUSEHOLD SURVEY
  (GHS) IN 2002
    DESIGN OF THE OCTOBER HOUSEHOLD
                SURVEYS

• THE OHS 1994
    1 000 Enumerator areas (EA’S) with 30 households per EA, in
     total thus 30 000 households
    Sampling frame: Initially the 1991 population census set of EA’s,
     later benchmarked to the 1996 population census results iro
     province, gender, race and age group totals
    Explicit stratification variables: Province and ?
    Unsure whether the EA’s were drawn pps or epsem within explicit
     strata and whether number of households (HHs) or number of
     persons was used a measure of size (MOS) if EA’s were drawn pps
    Within each drawn EA the 30 HHs were drawn systematically
   DESIGN OF THE OCTOBER HOUSEHOLD
               SURVEYS

• THE OHS 1995
    Advisory Committee recommendation re the change in the OHS design:
     More EA’s and fewer HHs per EA
    Rough analysis indicates only 8 ot 9 HHs per EA should be drawn
    Decision: 3000 EA’s and 10 HHs per EA, giving again a total sample size
     of 30 000 HHs -- much more expensive survey
    Sampling frame: Initially the 1991 population census set of EA’s. later
     benchmarked to the 1996 population census results iro province, gender,
     race and age group totals
    Explicit stratification variables: Province, race and urban/rural
    EA allocation to explicit strata based on square root of # persons
    Within each explicit stratum the EA’s were drawn pps with the number of
     persons in the EA as MOS
    Within each drawn EA the 10 HHs were drawn systematically
               OHS94                    OHS95

AREA   Unempl SE        DEFF    Unempl SE        DEFF
       Rate                     Rate
RSA    0.326   0.0088   17.37   0.293   0.0045   4.42
WC     0.173   0.014    10.43   0.186   0.009    2.99
EC     0.453   0.023    14.13   0.414   0.012    3.93
NC     0.325   0.025    10.06   0.272   0.019    3.70
FS     0.244   0.021    7.57    0.261   0.010    2.54
KZN    0.322   0.020    17.00   0.331   0.011    4.75
NW     0.366   0.027    10.60   0.328   0.016    4.13
GT     0.288   0.024    21.66   0.209   0.011    4.54
MP     0.364   0.030    12.63   0.334   0.012    2.76
LP     0.470   0.034    16.61   0.410   0.019    5.22
    DESIGN OF THE OCTOBER HOUSEHOLD
                SURVEYS

• THE OHS 1996
    The population census was conducted in October 1996 so that the
     OHS was postponed till November 1996
    Due to financial and time constraints the OHS 1996 was cut to
     1600 EA’s with 10 HHs per EA, i.e. a total sample size of 16 000
    Sampling frame: Initially the 1991 population census set of EA’s
    Explicit stratification variables: Province and Geography type
    For the PES, immediately after the 1996 census, 800 EA’s were
     drawn epsem from explicit strata after disproportional allocation of
     EA’s to explicit strata
    The EA to the east and the EA to the west of the 800 EA’s used in
     the PES were used as the PSU’s in the OHS
    Within each drawn EA 10 HHs were drawn systematically
    DESIGN OF THE OCTOBER HOUSEHOLD
                SURVEYS

• THE OHS 1997
    Sampling frame: The estimated number of persons in the 1996 set
     of EA’s, obtained form the administrative records of 1996
     population census (persons in special institutions excluded)
    The Transitional Metropolitan Councils and District Councils
     were used as explicit stratification variable
    EA allocation to explicit strata based on square root of the number
     of persons in the EA
    Within each explicit stratum the EA’s were drawn pps with the
     number of persons in the EA as MOS
    Within each drawn EA the 10 HHs were drawn systematically
    SE calculations as well as SE curves
     DESIGN OF THE OCTOBER HOUSEHOLD
                 SURVEYS

•   THE OHS 1998
      Sampling frame: Based on the 1996 population census (special institutions
       excluded)
      Due to budget constraints: Only 2000 EA’s with 10 HHs per EA
      Explicit stratification variables: Province and 7 EA types
      EA allocation to explicit strata based on square root of the number of
       HHs in the EA
      Within each explicit stratum the EA’s were drawn pps with the number of
       HHs in the EA as MOS
      Drawn EA’s with less than 50 HHs were pooled with consecutive EA’s
       using the Kish pooling method to obtain a minimum of 100 HHs
      Within each drawn (combined) EA the 10 HHs were drawn systematically
      SE calculations as well as SE curves provided
             MASTER SAMPLE DESIGN 1998

•   Based on GIS based improved version of 1996 Census dataset
•   Explicit stratification variables: Province and Urban/rural
•   Comparison between Neyman’s optimal allocation and square root of #EA’s
    formulae has shown little difference on # EA’s to be allocated to explicit
    strata. MOS was # HHs in EA.
•   3000 EA’s allocated to strata based on (MOS)
•   EA’s drawn pps within explicit strata
•   Drawn EA’s with less than 100 HHs were pooled with neighbouring EA’s
    using Kish’s pooling method (after drawing of the EA’s) to form PSUs
•   Dwelling units or “visiting points” were used as listing units in PSUs
•   Representative clusters of 10 listing units each formed in each PSU
•   Clusters in PSU randomly numbered of which one was to be selected
•   Different clusters to be used in different surveys – principle of rotation
           ALLOCATION OF EA’s IN MS
Province               Area Type
              Urban    Rural       TOTAL
WC            250      87          337
EC            174      216         390
NC            90       60          150
FS            162      120         282
KZN           231      219         450
NW            130      174         304
GT            465      15          480
MP            126      144         270
LP            94       243         337
TOTAL         1772     1278        3000
     DESIGN OF THE OCTOBER HOUSEHOLD
                 SURVEYS
•   THE OHS 1999
      Based on the 1998 Master Sample (MS). One cluster of 10 dwelling units
       was randomly selected within each of the 3000 EA’s
      All OHSe from 1995 till 1999 were re-weighted once the results of the
       2001 population census became available. First midyear estimates were
       recalculated for the period 1995 till 1999 and these midyear estimated
       population figures i.r.o. province, gender, population group and age group
       used for benchmarking the OHSe
      OHS 1999 was the last OHS in the series of OHSe. From 2000 onwards
       the OHS was split into bi-annual Labour Force surveys (LFS) and General
       Household surveys (GHS). The first GHS was conducted in 2002.
      The samples for the LFSe and GHSe were initially obtained from the first
       MS and later on from the second MS (based on the 2001 population
       census)
    DESIGN OF THE LABOUR FORCE SURVEYS

•   THE LFS 2000
      Based on the 1998 Master Sample (MS). In February a sub-sample of
       1000 EA’s was drawn from the MS 3000 EA’s. In September all 3000
       EA’s were used. One cluster of 10 dwelling units was randomly selected
       within each of the EA’s
      Rotation scheme: Replace in 20% of the EA’s the selected cluster of 10
       dwelling units with another cluster in the EA. This will result into a 80%
       overlap in dwelling units between consecutive LFSe. Such a rotating panel
       will facilitate the study of the change in persons employment status over
       time.
      Rotation principle created identification problems of the dwelling units on
       the ground, especially in the first two years
      After the results of the 2001 population census became available, the
       record weights of all LFS datasets were benchmarked to align them with
       the midyear estimates based on the 2001 census results iro provinces,
       gender, race and age group
    PANEL STUDY COMPARING LFS4 TO LFS8

•   PANEL STUDY
      Two LFSe were and are conducted per year, viz. in February and in
       September. Five clusters of 10 dwellings units each were reserved in the
       MS EA’s to facilitate for a period of 5 years a cluster (of 10 dwellings
       units each) replacement in 20% of the EA’s in consecutive surveys
      Theoretically the same dwelling units should have been visited in 80% of
       the cases in two consecutive LFSe. In the vast majority of these cases the
       same HH should have been visited. Furthermore, a HH should stay for 5
       consecutive LFSe in the sample
      The rotating design allow thus a study of a person’s movement into or out
       of the labour market
      Estimation of variances of a function of variable values between two LFSe
       within a 5-year period includes pair-wise as well as independent values.
       This required special adjustments of the variance estimation program to
       calculate also covariances between variable values in the two surveys
      Ongoing project
     DESIGN OF THE GENERAL HOUSEHOLD
                 SURVEYS
•   THE GHS 2002
      A cluster of 10 dwelling units, different from the 5 clusters reserved for
       the consecutive LFSe, in the 1998 MS PSU’s was used. The sample thus
       consisted of 30 000 dwelling units within 3 000 PSU’s, 10 dwelling units
       per PSU.
      Square root (of # HHs in EA) based allocation of EA’s to the 18 explicit
       strata formed by the variables province and urban/rural. A drawn EA with
       less than 100 HHs was pooled with adjacent EA’s using Kish’s pooling
       method to from a PSU with al least 100 HHs
      Realised dataset was benchmarked to align it with the midyear estimates,
       based on the 2001 population census results, iro province, gender, race
       and age group
      Standard errors were calculated for main variables and a general SE-curve
       was created from which rough estimates of the SE of variables for other
       variables and for sub-classes could be obtained.
SE-graph of HH facilities
SE-graph of number of HHs
     DESIGN OF THE GENERAL HOUSEHOLD
                 SURVEYS

•   USING THE SE-GRAPH
      Take as an example the SE-graph for HH facilities. Let the estimated
        proportion of HHs with a specific facility (such as “with house
        ownership”) in a sub class be = 0.15. Then from the appropriate graph it
        follows the the CV (coefficient of variation) of the estimate = 0.03,
        approximately.
       Now CV = SE (facility proportion)/facility proportion, or
       SE(facility proportion) = CV * facility proportion
                               = 0.03*0.15 = 0.0045
      Similarly for number of HHs. If say, # HHs = 150 000, then
       from appropriate graph (# HHS) = 0.047 , which gives
       SE(#HHs) = CV(#HHs) * #HHs = 0.047*150000
                                         = 7050
    DESIGN OF THE GENERAL HOUSEHOLD
                SURVEYS

• THE DESIGN OF GHS2003 TO GHS2006
    HHS2003 and GHS2004 were designed similarly to GHS2002,
     only different clusters of dwelling units were used in the 3000
     PSUs obtained from the 1998 Master Sample
    A new Master Sample was constructed in 2003, based on the 2001
     population census results -- cf. next slide. This new MS2 was used
     for the GHS2005 and GHS2006, using an independent cluster of
     dwelling units of size 10
    Realised datasets were benchmarked to align them with the
     midyear estimates, based on the 2001 population census results, iro
     province, gender, race and age group
    StatsSA unfortunately discontinued the calculation of SE-graphs
    DESIGN OF THE SECOND MASTER SAMPLE

•   THE DESIGN OF SECOND MASTER SAMPLE
      In 2003 a new Master Sample was constructed, based on the 2001
       population census results
      Based on the September 2002 LFS, extensive relative precision (RP)
       calculations were made to see what RP results can be obtained (1) per
       district council and (2) per municipality for various sub-sample sizes
      Different allocation methods to determine sample sizes per (1) district
       council and (2) municipality, such as Neyman optimum allocation and
       power root method were compared
      Different cost models for adding an EA in stratum and adding a HH in
       drawn EA, i.e. to investigate possible changes in design: 3000 EA’s with
       10 HH’s per EA, taking into account effect on RP
      Finally, it was decided to use district council as explicit stratification
       variable and the 0.5 power allocation rule. For a RP<=0.25, minimum
       allocation to DC becomes 46 PSU’s
    DESIGN OF THE SECOND MASTER SAMPLE

•   THE DESIGN OF SECOND MASTER SAMPLE: Construction of PSU’s
      In 2001 census GIS at StatsSA used a new method of numbering EA’s.
       Consequence: Consecutive EA numbers jumped in and out sub places
       (suburbs, villages, etc.) Thus Kish’s method of pooling consecutive EA
       numbers after the drawing of the EA’s, if a drawn EA was too small,
       could no longer be used.
      A small EA had thus to be grouped with another similar EA’s into a larger
       PSU before the drawing of the PSU’s. This was established by using
       Google Earth on which the EA boundaries were over-laid
      Minimum number of dwelling units (DU’s) for an EA to qualify as PSU
       was set at 100.
      In the PSU 2 clusters of 12 DU’s and 5 clusters of 10 DU’s were formed,
       each cluster being representative of the EA. The larger clusters were to be
       used for the HIES and the smaller clusters for the LFSe and GHSe
      Realised datasets were benchmarked to align them with the midyear
       estimates, based on the 2001 population census results, iro province,
       gender, race and age group

				
DOCUMENT INFO
Shared By:
Categories:
Tags:
Stats:
views:8
posted:6/10/2011
language:English
pages:19