Nigeria - Nigeria Living Standard Survey 2003, First round

SAMPLE DESIGN
The sampling designs for the NLSS was meant to give estimates at National, Zonal and State levels. The first stage was a duster of housing units called Enumeration Area (EA), while the second stage was the housing unit.

SAMPLE SIZE
One hundred and twenty (120 EAs) were selected and sensitized in each state while sixty enumeration areas were selected at the Federal Capital Territory (FCT). Ten E.As with five housing units were studied per month. This meant that fifty housing units were canvassed per month in each state and twenty-five housing units in Abuja.

One hundred and twenty (120) EAs were selected in 12 replicates in each State from the NISH master sample frame in replicates (4-15). However, 60 EAs were selected in the Federal Capital Territory. Five (5) housing units (HUs) were scientifically selected in each of the selected EAs. One replicate consisting of 10 EAs in the State and 5 EAs in the Federal Capital Territory were covered every month. Fifty (50) HUs were covered in each State and 25 HUs in the Federal Capital Territory per month. This implied that the survey had an anticipated national sample size of twenty-one thousand and nine hundred (21,900) HUs for the country for the 12-month survey period. Each State had a sample size of 600 HUs, while the Federal Capital Territory had a sample size of 300. The sample size is robust enough to provide reasonable estimates at national and sub-national (State) levels.
ESTIMATION PROCEDURE The following statistical notations were used:
N = the number of EAs in each State
ni = Size of replicates rth
r = number of replicates in a State
H = number of housing units listed in the ith selected EA.
Xhj = number of housing units selected from ith selected EA.
Wrij = weight of the replicate =????????nhijNxH
Yrij = total value of variable from the ith HU of ith selected EA.
Replicate Estimate (Monthly Estimate) ()??=yWyi
Annual State Estimate ???
NOTE
See page 91 and 92 of the report


Sampling Error (Variance) Estimate The Jacknife indefinite method of variance estimation was used for the survey because the method required replication and clustering. An estimate of State variance was first obtained. Cluster estimate is ()ywijiji???= Mean Estimate rnrz??= Therefore
mean variance is ()rSnrV2=?
where ()()221-?-?=?rnrSr
NOTE
See page 93 of the report


The response rate was very high


The NLSS, like most household surveys, is based on NISH frame. The NISH design is a two-stage design with EA's as first stage units and households as second stage units. Ten enumeration areas (EAs) were randomly selected each month and five household were systematically selected from the household listing of each selected EAs. Population level estimates are made by multiplying the data for each household by two factors, one equal to the inverse of the probability of selecting that household from the total list of households in its EA, and one equal to the inverse of the probability of selecting that EA from the list of EAs in its state. The selections can be done by treating every unit as the same and using simple random selection or, if the data is available, a more efficient sample can be selected using some size variable known for every unit of the population thought to be correlated with the variables of interest for measurement. So the weighting factor is at the EA level in each state:


where
Nh = the total number of EAs in state h.
nh = the number of sampled EAs in state h.
Mhi = the number of listed households in ith EA of state h.
nhi = the number of sampled households in ith EA of state h.
Xhij = the number of persons in the jth household in ith EA of state h.
Phij = the poverty score for the jth household in ith EA of state h.

So the above will apply to all the individual members in order to give the population. However, the above weighting factor will be multiplied by average household size, when there is need to take the household aggregates to the population.