Nigeria - General Household Survey 2006, Third edition

The GHS was implemented as a NISH module. Six replicates were studied per State while three replicates were studied in the FCT, Abuja. With a fixed-take of 10 HUs systematically selected per EA, 600 HUs thus were selected for interview per State and 300 for FCT, Abuja. Hence, nationally, a total of 21,900 HUs drawn from the 2,190 cut across the rural and urban sectors.

Introduction:
The sample design for the survey derives from the National Integrated Survey of Household (NISH) developed by National Bureau of Statistics (NBS). The NISH design employed a replicated sampling design that is technique by which many sample (replicates) were selected independently from a population such that each replicate sample represents the population.

Essentially, the NISH sample design is a 2-stage replicated and rotated cluster sample design with Enumeration Areas (EAs) as first stage sampling unit or Primary Sampling Unit (PSU) and Housing Units the second stage sampling units (secondary sampling units). Generally, for each state of the Federation, the NISH Master Sample is made up of 120 EAs drawn in12 replicates. A replicate consists of 10 EAs.

Selection Procedures:
The EAs demarcated by the National Population Commission (NpopC) for the 1991 Population Census served as the primary Sample Frame for the design.

First Stage Selection:
Sixty EAs were selected with equal probability from the list of EAs in each state of the federation and 30 EAs for FCT, Abuja. The selected EAs cuts across rural and urban sectors. The study EAs for the collaborative survey were drawn from replicates 7,8,9,10,11 and 12 of the master sample of each state.

Second Stage Selection:
In each selected EA, a listing of housing units was carried out. The result provided the frame for the second stage selection. Ten housing units were selected systematically in each EA after the completion of the listing exercise. Thereafter, all the households within the selected HUs were interviewed using GHS questionnaire.
Out of the expected 2,190 EAs, 1,883 were studied.
Out of the 21,900 housing units expected to be covered, 18,826 were canvassed.


Variance Estimate (Jackknife Method)
Estimating variances using the Jackknife method will require forming replicate from the full sample by randomly eliminating one sample cluster [Enumeration Area (EA) at a time from a state containing k EAs, k replicated estimates are formed by eliminating one of these, at a time, and increasing the weight of the remaining (k-1) EAs by a factor of k/(k-1). This process is repeated for each EA.

For a given state or reporting domain, the estimate of the variance of a rate, r, is given by
k
Var(r ) = (Se)2 = 1 S (ri - r)2
k(k-1) i=1

where (Se) is the standard error,
k is the number of EAs in the state or reporting domain.

r is the weighted estimate calculated from the entire sample of EAs in the state or reporting domain.
ri = kr - (k - 1)r(i), where

r(i) is the re-weighted estimate calculated from the reduced sample of k-1 EAs.

To obtain an estimate of the variance at a higher level, say, at the national level, the process is repeated over all states, with k redefined to refer to the total number of EAs (as opposed to the number in the states).


On National basis, 85.98 percent response rate was acheived at EA level while 85.96 percent was acheived at housing units level.


The variable (Hweight ) Household weight is computed and attached to the data file.

The formula adopted in calculating the design weights for the survey data (sample results) were as follows:
(i) The probability of selecting an EA within a state was obtained by dividing the total number of EAs sampled in a state by total number of EAs in that particular state. Let this be represented by fj. That is,
fj = Total Number of EAs sampled in a state
Total Number of EAs in that particular State

(ii) Likewise, the probability of selecting an housing unit (HU) within an EA was obtained by dividing the total number of housing units selected in an EA by the total number of housing units (HUs) listed in that particular EA. Let this be represented by fk. That is,
fk = Total Number of HUs selected in an EA
Total Number of HUs listed in that particular EA
Mathematically,
Design weight = Total number of EAs in a state
Total number of EAs sampled in that particular state
X
Total Number of HUs listed in an EA
Total Number of HUs selected in that particular EA

Estimation Procedures:
Let the probability of selecting the EA be fj and the probability of selecting the housing unit be fk. Then the product f = fjfk = 1 where fj = n and fk = h
Wj k N H.

Household Weight (HHWeight )


n h
Ys = N ? H ? X sj k
n j=1 h k=1

n h
= N H ? ? X sj k
n h j=1 k=1

n h
= W s j k ? ? X sj k (Note: W s j k = N . H )
j = 1 k=1 n h
Where:
ˆ
Ys = Estimate for states
N = Total Number of EAs in states
n = Selected number of EAs in states
H = Total number of Housing Units listed in the jth EA
h = Selected number of Housing Units in the jth EA.
Xsj k = Value of the element in the kth housing unit of jth EA in states.
Wsjk = Weight of the element in kth housing unit of the jth EA in states.