sample size...2 the problem: determine sample size, n • ensuring a required level of precision...
TRANSCRIPT
![Page 1: Sample Size...2 The Problem: Determine sample size, n • Ensuring a required level of precision Most efficient and largest for a given fixed budget, B Survey budget, B, is fixed,](https://reader034.vdocuments.us/reader034/viewer/2022042213/5eb689e583ec8c4e860651c2/html5/thumbnails/1.jpg)
Second RAP Regional Workshop on Building Training Resources for Improving Agricultural & Rural Statistics Sampling Methods for Agricultural Statistics-Review of Current Practices
SCI, Tehran, Islamic Republic of Iran 10-17 September 2013
1
Sample Size
Dr. A.C. Kulshreshtha U.N. Statistical Institute for Asia and the Pacific (SIAP)
![Page 2: Sample Size...2 The Problem: Determine sample size, n • Ensuring a required level of precision Most efficient and largest for a given fixed budget, B Survey budget, B, is fixed,](https://reader034.vdocuments.us/reader034/viewer/2022042213/5eb689e583ec8c4e860651c2/html5/thumbnails/2.jpg)
2
The Problem:
Determine sample size, n • Ensuring a required level of precision Most efficient and largest for a given fixed budget, B Survey budget, B, is fixed, no matter what the variance is Upper limit to the variance of an estimator is fixed at Vo
whatever B is
![Page 3: Sample Size...2 The Problem: Determine sample size, n • Ensuring a required level of precision Most efficient and largest for a given fixed budget, B Survey budget, B, is fixed,](https://reader034.vdocuments.us/reader034/viewer/2022042213/5eb689e583ec8c4e860651c2/html5/thumbnails/3.jpg)
3
( )θVar( )θVar
Possibilities
• Minimize subject to fixed B • inversely depends on ‘n’, call it V(n) • Thus, we have an optimization problem:
Find ‘n’ such that V(n) is minimum subject to cost function C(n)
![Page 4: Sample Size...2 The Problem: Determine sample size, n • Ensuring a required level of precision Most efficient and largest for a given fixed budget, B Survey budget, B, is fixed,](https://reader034.vdocuments.us/reader034/viewer/2022042213/5eb689e583ec8c4e860651c2/html5/thumbnails/4.jpg)
4
Factors affecting ‘optimal’ n
• Required precision of estimates– higher precision desired, larger sample size needed Variability of characteristic being measured– more
variable, larger sample size Rare characteristic– more rare, larger sample size
• Population size N (sampling fraction)– no effect on sample size if N is large
![Page 5: Sample Size...2 The Problem: Determine sample size, n • Ensuring a required level of precision Most efficient and largest for a given fixed budget, B Survey budget, B, is fixed,](https://reader034.vdocuments.us/reader034/viewer/2022042213/5eb689e583ec8c4e860651c2/html5/thumbnails/5.jpg)
5
Effect of N
• For example, SRSWR:
n)Y()Y(
22 σ
=σ
Precision of sample mean does not depend on population size N
Precision of sample mean depends only on variability of population values
![Page 6: Sample Size...2 The Problem: Determine sample size, n • Ensuring a required level of precision Most efficient and largest for a given fixed budget, B Survey budget, B, is fixed,](https://reader034.vdocuments.us/reader034/viewer/2022042213/5eb689e583ec8c4e860651c2/html5/thumbnails/6.jpg)
6
Effect of N (Contd.)
• For example, SRSWOR:
For large N, (N-n)/(N-1) 1
1NnN*
n)Y()Y(
22
−−σ
=σ
![Page 7: Sample Size...2 The Problem: Determine sample size, n • Ensuring a required level of precision Most efficient and largest for a given fixed budget, B Survey budget, B, is fixed,](https://reader034.vdocuments.us/reader034/viewer/2022042213/5eb689e583ec8c4e860651c2/html5/thumbnails/7.jpg)
7
Factors affecting ‘optimal’ n
• Cost– larger sample size higher cost Example: Simple cost function: C = C0 + n*C1 where C= total cost of survey; C0 is fixed cost; C1 is
cost per sample unit; n is sample size For given total budget C’:
1
0'C
CCn −=
![Page 8: Sample Size...2 The Problem: Determine sample size, n • Ensuring a required level of precision Most efficient and largest for a given fixed budget, B Survey budget, B, is fixed,](https://reader034.vdocuments.us/reader034/viewer/2022042213/5eb689e583ec8c4e860651c2/html5/thumbnails/8.jpg)
8
Factors affecting ‘optimal’ n (Contd.)
• Level of detail required More Reporting domains larger sample size needed More subclasses (for analysis) larger sample size
needed
![Page 9: Sample Size...2 The Problem: Determine sample size, n • Ensuring a required level of precision Most efficient and largest for a given fixed budget, B Survey budget, B, is fixed,](https://reader034.vdocuments.us/reader034/viewer/2022042213/5eb689e583ec8c4e860651c2/html5/thumbnails/9.jpg)
9
Basic Steps
![Page 10: Sample Size...2 The Problem: Determine sample size, n • Ensuring a required level of precision Most efficient and largest for a given fixed budget, B Survey budget, B, is fixed,](https://reader034.vdocuments.us/reader034/viewer/2022042213/5eb689e583ec8c4e860651c2/html5/thumbnails/10.jpg)
10
Basic Steps for determining n
• How much precision is desired? Or, how much ‘error’ is tolerable?
• Relate sample size (n) and precision or error requirements (an equation based on sampling theory)
• For this equation, estimate the unknown quantities (usually, variances of population) and solve for value of n n*
![Page 11: Sample Size...2 The Problem: Determine sample size, n • Ensuring a required level of precision Most efficient and largest for a given fixed budget, B Survey budget, B, is fixed,](https://reader034.vdocuments.us/reader034/viewer/2022042213/5eb689e583ec8c4e860651c2/html5/thumbnails/11.jpg)
11
Basic Steps (Contd.)
• Allocate to domains, strata, (subclasses) • Adjust for precision requirements for estimates
for domains, strata n** • Note: Initial computations may start with sample
size requirements for each domain, stratum, etc. • Are there sufficient resources for data collection
on n** units? If not, readjust requirements of precision, reallocate within resource constraints sample size
![Page 12: Sample Size...2 The Problem: Determine sample size, n • Ensuring a required level of precision Most efficient and largest for a given fixed budget, B Survey budget, B, is fixed,](https://reader034.vdocuments.us/reader034/viewer/2022042213/5eb689e583ec8c4e860651c2/html5/thumbnails/12.jpg)
12
Initial Computations
Determine sample size required for SRSWR– n(srs) Adjust n(srs), if N is relatively small:
Nn1
nnSRS
SRS
+≥
Adjust n to allow for a more complex sample design using the deff of the design; n(complex)= n * deff
Adjust n(complex) to take into account expected non-response rates, n(adj) = n(complex) * (1+nonresponse rate)
![Page 13: Sample Size...2 The Problem: Determine sample size, n • Ensuring a required level of precision Most efficient and largest for a given fixed budget, B Survey budget, B, is fixed,](https://reader034.vdocuments.us/reader034/viewer/2022042213/5eb689e583ec8c4e860651c2/html5/thumbnails/13.jpg)
13
Initial Computations- Example
Example of adjustment for cluster sampling: n(srs) = 200 deff(cluster) = 2.0 n(cluster) = 200*2.0 = 400 Expect nonresponse rate = 0.20 n(adj) = 400*(1+.20) = 480
![Page 14: Sample Size...2 The Problem: Determine sample size, n • Ensuring a required level of precision Most efficient and largest for a given fixed budget, B Survey budget, B, is fixed,](https://reader034.vdocuments.us/reader034/viewer/2022042213/5eb689e583ec8c4e860651c2/html5/thumbnails/14.jpg)
14
Determining n(srs)
How much precision do I need? Or, how much error is tolerable?
a. Variance of estimate should not exceed a given value V0 b. Margin of error, e, should be met with a given probability c. Width of confidence interval should not exceed a prescribed
amount, w d. CV (or RSE) should not exceed a given value
![Page 15: Sample Size...2 The Problem: Determine sample size, n • Ensuring a required level of precision Most efficient and largest for a given fixed budget, B Survey budget, B, is fixed,](https://reader034.vdocuments.us/reader034/viewer/2022042213/5eb689e583ec8c4e860651c2/html5/thumbnails/15.jpg)
15
Sample size in SRS n(SRS)– Estimation of Population Mean
0
2
SRS0SRS
2
0 VSnV
nSV)Y(V ≥⇒≤⇒≤
a. Variance of sample mean should not exceed V0
Nn1
nnSRS
SRS
+≥Adjust for small N:
![Page 16: Sample Size...2 The Problem: Determine sample size, n • Ensuring a required level of precision Most efficient and largest for a given fixed budget, B Survey budget, B, is fixed,](https://reader034.vdocuments.us/reader034/viewer/2022042213/5eb689e583ec8c4e860651c2/html5/thumbnails/16.jpg)
16
n(SRS)– Estimation of Population Mean (Contd.)
α−=
≤− 1eYYobPr
22/
SRS2
2/2
eSzn)Y(Vze
⋅
=⇒⋅=⇒ αα
b. Margin of error, e, should be met with given probability.
![Page 17: Sample Size...2 The Problem: Determine sample size, n • Ensuring a required level of precision Most efficient and largest for a given fixed budget, B Survey budget, B, is fixed,](https://reader034.vdocuments.us/reader034/viewer/2022042213/5eb689e583ec8c4e860651c2/html5/thumbnails/17.jpg)
17
n(SRS)– Estimation of Population Mean (Contd.)
Values of α = 0 (100% confidence level) zα/2 = 3 = 0.05 (95% confidence level) zα/2 = 1.96 = 0.10 (90% confidence level) zα/2 = 1.645
Note: Assumption is that sampling distribution of
sample mean is normal distribution
![Page 18: Sample Size...2 The Problem: Determine sample size, n • Ensuring a required level of precision Most efficient and largest for a given fixed budget, B Survey budget, B, is fixed,](https://reader034.vdocuments.us/reader034/viewer/2022042213/5eb689e583ec8c4e860651c2/html5/thumbnails/18.jpg)
18
n(SRS)– Estimation of Population Mean (Contd.)
α−=+≤≤− αα 1)]Y(SEzYY)Y(SEzY[obPr 2/2/
w)Y(SEz2 2/ ≤⇒ α
22/SRS )
wSz(4n α≥⇒
c. Width of confidence interval should not exceed w
![Page 19: Sample Size...2 The Problem: Determine sample size, n • Ensuring a required level of precision Most efficient and largest for a given fixed budget, B Survey budget, B, is fixed,](https://reader034.vdocuments.us/reader034/viewer/2022042213/5eb689e583ec8c4e860651c2/html5/thumbnails/19.jpg)
19
n(SRS)– Estimation of Population Mean (Contd.)
2
0SRS CV
)Y(CVn
≥⇒
d. CV of sample mean should not exceed CV0
![Page 20: Sample Size...2 The Problem: Determine sample size, n • Ensuring a required level of precision Most efficient and largest for a given fixed budget, B Survey budget, B, is fixed,](https://reader034.vdocuments.us/reader034/viewer/2022042213/5eb689e583ec8c4e860651c2/html5/thumbnails/20.jpg)
20
n(srs)- Estimation of Proportions
0SRS V
)P1(Pn −≥• Specified maximum variance, V0:
• Given margin of error, e: 2
2
SRS e)P1(Pz
n 2/−
≥ α
• Specified maximum CV, CV0: 20
SRS )CV(P)P1(n −
≥
Note: Can use P=0.5 if no information on P
![Page 21: Sample Size...2 The Problem: Determine sample size, n • Ensuring a required level of precision Most efficient and largest for a given fixed budget, B Survey budget, B, is fixed,](https://reader034.vdocuments.us/reader034/viewer/2022042213/5eb689e583ec8c4e860651c2/html5/thumbnails/21.jpg)
21
Sample Size in Stratified Sampling
• Proportional allocation for a specified variance, V0: ∑
∑+
=
h
2hh0
h
2hh
SNV
SNNn
∑∑+
=
h
2hh0
2
hhh
SNV
SNn
• Optimum allocation for a specified variance, V0:
![Page 22: Sample Size...2 The Problem: Determine sample size, n • Ensuring a required level of precision Most efficient and largest for a given fixed budget, B Survey budget, B, is fixed,](https://reader034.vdocuments.us/reader034/viewer/2022042213/5eb689e583ec8c4e860651c2/html5/thumbnails/22.jpg)
22
Sample Size in Stratified Sampling (Contd.)
• Cost-optimum allocation for a specified variance, V0, and given cost where Ch is average variable cost per sample unit in stratum h :
∑∑∑
+
=
h
2hh0
hhhh
hhhh
SNV
C/1(SNCSN
n
![Page 23: Sample Size...2 The Problem: Determine sample size, n • Ensuring a required level of precision Most efficient and largest for a given fixed budget, B Survey budget, B, is fixed,](https://reader034.vdocuments.us/reader034/viewer/2022042213/5eb689e583ec8c4e860651c2/html5/thumbnails/23.jpg)
23
Sample Size in Cluster Sampling
Effect of clustering on variance • More similar the elements within each
cluster, the larger the deff of cluster sample; i.e., cluster sampling is less efficient compared to srs
• Sample size needed for a clustered sample for same precision as n(srs) is:
• n(cluster) = n(srs) * deff
![Page 24: Sample Size...2 The Problem: Determine sample size, n • Ensuring a required level of precision Most efficient and largest for a given fixed budget, B Survey budget, B, is fixed,](https://reader034.vdocuments.us/reader034/viewer/2022042213/5eb689e583ec8c4e860651c2/html5/thumbnails/24.jpg)
24
Sample Size in Cluster Sampling(Contd.)
In cluster sampling and two-stage sampling, need to determine: Size of PSU Number of SSUs to be sampled in each
sample PSU Number of PSUs to be sampled
![Page 25: Sample Size...2 The Problem: Determine sample size, n • Ensuring a required level of precision Most efficient and largest for a given fixed budget, B Survey budget, B, is fixed,](https://reader034.vdocuments.us/reader034/viewer/2022042213/5eb689e583ec8c4e860651c2/html5/thumbnails/25.jpg)
25
Sample Size in Cluster Sampling (Contd.)
Size of PSU Larger PSUs, smaller ρ and smaller deff Too large PSUs, loose cost savings of cluster
sampling Subsampling rate
In general, balance costs for sampling PSU and SSU and precision requirements
![Page 26: Sample Size...2 The Problem: Determine sample size, n • Ensuring a required level of precision Most efficient and largest for a given fixed budget, B Survey budget, B, is fixed,](https://reader034.vdocuments.us/reader034/viewer/2022042213/5eb689e583ec8c4e860651c2/html5/thumbnails/26.jpg)
26
Approximate Exact
2
212
1
221
2
21
2
)1(zz
yV
NVNV
n
yy
y
y
σ
ε
=
−+=
2
21
2zε
yVn ≅
221
1ˆ yy sN
N −=σ ∑
=
−−
=n
iiy yy
ns
1
2clu
2 )(1
1
Sample size for One-stage Cluster Sampling
![Page 27: Sample Size...2 The Problem: Determine sample size, n • Ensuring a required level of precision Most efficient and largest for a given fixed budget, B Survey budget, B, is fixed,](https://reader034.vdocuments.us/reader034/viewer/2022042213/5eb689e583ec8c4e860651c2/html5/thumbnails/27.jpg)
27
2
222
22
212
1 ,Lety
Vy
V yy
yy
σ=
σ=
units listing average=M
212
2
22
21
11
z
)1(1
y
yy
VN
VMm
mMVN
N
n
−+
−
−+
−=
ε
Sample Size for Two Stage Cluster Sampling
mSuppose is known (later we show how to estimate ) = number of listing units sampled from each cluster
m m
![Page 28: Sample Size...2 The Problem: Determine sample size, n • Ensuring a required level of precision Most efficient and largest for a given fixed budget, B Survey budget, B, is fixed,](https://reader034.vdocuments.us/reader034/viewer/2022042213/5eb689e583ec8c4e860651c2/html5/thumbnails/28.jpg)
28
Need to know the relative costs of first and second stage sampling
It also depends on variance of ‘y’ between first-stage units, i.e. σ2
1y and variance of ‘y’ within second-stage units, i.e. σ2
2y.
∑=
−==N
iiy YY
N 1
221 )(1 PSUsbetween Varianceσ
∑∑= =
−==N
i
M
jiijy YY
N 1 1
222 )(1SSUs within Y of Varianceσ
Two stage sample size, m
![Page 29: Sample Size...2 The Problem: Determine sample size, n • Ensuring a required level of precision Most efficient and largest for a given fixed budget, B Survey budget, B, is fixed,](https://reader034.vdocuments.us/reader034/viewer/2022042213/5eb689e583ec8c4e860651c2/html5/thumbnails/29.jpg)
29
mnCnCC *2
*1 +=
Cost of sampling a unit at first-stage
Cost of sampling a unit at second-stage
for ‘n’ PSU’s and ‘ ’ SSU’s within each PSU, cost function is:
m
Cost Function Two Stage Sampling
![Page 30: Sample Size...2 The Problem: Determine sample size, n • Ensuring a required level of precision Most efficient and largest for a given fixed budget, B Survey budget, B, is fixed,](https://reader034.vdocuments.us/reader034/viewer/2022042213/5eb689e583ec8c4e860651c2/html5/thumbnails/30.jpg)
30
• Cost of interview or measurement for a sampling unit
• Cost of traveling to each sample cluster
• Listing SSU’s, and cost of selecting a sample units from each cluster
• Going back to cluster for interview or measurement
Cost of sampling a unit at second-stage
Cost of sampling a unit at first-stage
Mm
Costs in Two Stage Sampling
![Page 31: Sample Size...2 The Problem: Determine sample size, n • Ensuring a required level of precision Most efficient and largest for a given fixed budget, B Survey budget, B, is fixed,](https://reader034.vdocuments.us/reader034/viewer/2022042213/5eb689e583ec8c4e860651c2/html5/thumbnails/31.jpg)
31
• It costs ‘0.5’ person-hour to travel to each sample cluster
• It costs ‘1.0’ person-hour to list the ‘20’ SSU within the cluster and then select a random sample
• It costs ‘0.5’ person-hour to return to clusters
2.000.51.000.5*1 =++=C
• It costs ‘0.25’ person-hour to interview or measure a sampling unit:
mn.n.CThusC
250002:
25.0*2
+=
=
Then:
Example--
![Page 32: Sample Size...2 The Problem: Determine sample size, n • Ensuring a required level of precision Most efficient and largest for a given fixed budget, B Survey budget, B, is fixed,](https://reader034.vdocuments.us/reader034/viewer/2022042213/5eb689e583ec8c4e860651c2/html5/thumbnails/32.jpg)
32
m• For values of , use the previous formula to estimate ‘n’:
)1(
11
z
111
212
2
22
21
y
yy
VN
VmM
mMVN
N
n
−+
−
−+
−=
ε
This meets the accuracy and confidence condition for a given .
• For this specific solution, compute:
m
)2(*2
*1 mnCnCC +=
Two Stage Sample Size
![Page 33: Sample Size...2 The Problem: Determine sample size, n • Ensuring a required level of precision Most efficient and largest for a given fixed budget, B Survey budget, B, is fixed,](https://reader034.vdocuments.us/reader034/viewer/2022042213/5eb689e583ec8c4e860651c2/html5/thumbnails/33.jpg)
33
• Repeat this calculation for all possible combinations of ‘ ’ and ‘n’.
• Eliminate those combinations that do not meet the accuracy specification, using (1).
• Make a table of , n, and cost.
• Identify the pair ( , n) with lowest cost.
m
mm
Two Stage Sample Size (cont’d)
![Page 34: Sample Size...2 The Problem: Determine sample size, n • Ensuring a required level of precision Most efficient and largest for a given fixed budget, B Survey budget, B, is fixed,](https://reader034.vdocuments.us/reader034/viewer/2022042213/5eb689e583ec8c4e860651c2/html5/thumbnails/34.jpg)
34
6.75 7.0 1 20 6.75 1 19
3 9 16.0 4 8 15.0 4 7 17.5 5 6
Minimum Cost
Field cost from equation (2)
‘n’ from equation (1) Selected m
Example
![Page 35: Sample Size...2 The Problem: Determine sample size, n • Ensuring a required level of precision Most efficient and largest for a given fixed budget, B Survey budget, B, is fixed,](https://reader034.vdocuments.us/reader034/viewer/2022042213/5eb689e583ec8c4e860651c2/html5/thumbnails/35.jpg)
35
Optimum sampling and sub-sampling fractions
C=c1n+c2nm
222
2 11)(1)( bww
bts SN
SmnM
SSn
yV −+−=
2
12
2 }{c
c
MSS
Smw
b
wopt
−=
![Page 36: Sample Size...2 The Problem: Determine sample size, n • Ensuring a required level of precision Most efficient and largest for a given fixed budget, B Survey budget, B, is fixed,](https://reader034.vdocuments.us/reader034/viewer/2022042213/5eb689e583ec8c4e860651c2/html5/thumbnails/36.jpg)
36
Optimum sampling and sub-sampling fractions (Contd.)
provided Values of n is found by solving either the cost
equation or the variance equation
MSS w
b
22 >
![Page 37: Sample Size...2 The Problem: Determine sample size, n • Ensuring a required level of precision Most efficient and largest for a given fixed budget, B Survey budget, B, is fixed,](https://reader034.vdocuments.us/reader034/viewer/2022042213/5eb689e583ec8c4e860651c2/html5/thumbnails/37.jpg)
37
Sample Allocation to Domains
For example: Whole country
Region 1 Region 2 Region 3
Districts
EAs
![Page 38: Sample Size...2 The Problem: Determine sample size, n • Ensuring a required level of precision Most efficient and largest for a given fixed budget, B Survey budget, B, is fixed,](https://reader034.vdocuments.us/reader034/viewer/2022042213/5eb689e583ec8c4e860651c2/html5/thumbnails/38.jpg)
38
Sample Allocation to Domains
One approach Calculate sample size requirements for each
domain Add up the individual sample size requirements to
get total sample size Adjust depending on resource constraints
![Page 39: Sample Size...2 The Problem: Determine sample size, n • Ensuring a required level of precision Most efficient and largest for a given fixed budget, B Survey budget, B, is fixed,](https://reader034.vdocuments.us/reader034/viewer/2022042213/5eb689e583ec8c4e860651c2/html5/thumbnails/39.jpg)
39
Sample Allocation to Domains - Strata
*nNNn h
h ⋅=Proportional allocation:
∑=
⋅= H
1hhh
hhh
SN
SN*nnOptimum or Neyman allocation:
∑=
⋅= H
1hhhh
hhhh
)C(1/SN
)C(1/SN*nnCost-optimum allocation:
Given n, allocation into strata
![Page 40: Sample Size...2 The Problem: Determine sample size, n • Ensuring a required level of precision Most efficient and largest for a given fixed budget, B Survey budget, B, is fixed,](https://reader034.vdocuments.us/reader034/viewer/2022042213/5eb689e583ec8c4e860651c2/html5/thumbnails/40.jpg)
40
Sample Allocation to Domains (Contd.)
Some considerations: Need for minimum and maximum sample sizes Domains may differ in importance– may require
more precise estimates for some domains Some domains may be more heterogeneous than
others with greater underlying variability of study variables
Survey costs may differ among domains
![Page 41: Sample Size...2 The Problem: Determine sample size, n • Ensuring a required level of precision Most efficient and largest for a given fixed budget, B Survey budget, B, is fixed,](https://reader034.vdocuments.us/reader034/viewer/2022042213/5eb689e583ec8c4e860651c2/html5/thumbnails/41.jpg)
41
Sample Allocation to Domains (Contd.)
• Note: Need at least two sampling units (minimum) per cell. For cells with many establishments, specify maximum number. Typically, all large establishments are selected. Allocate remaining sample size to the cells.
Small Medium LargeTotal Manufacturing
Food and beveragesWearing apparelWood productsPlastic productsOther manufacturing
SIZESector/Subsector
![Page 42: Sample Size...2 The Problem: Determine sample size, n • Ensuring a required level of precision Most efficient and largest for a given fixed budget, B Survey budget, B, is fixed,](https://reader034.vdocuments.us/reader034/viewer/2022042213/5eb689e583ec8c4e860651c2/html5/thumbnails/42.jpg)
42
Sample Allocation to Domains (Contd.)
Optimum allocation (or in many cases, proportional allocation) gives required precision for whole population (e.g., whole country; total trade establishments) but may not give required precision for all domains (e.g., regions; trade subsectors)
Equal allocation is ideal for comparison of domain estimates but may not be “representative” at the population level
![Page 43: Sample Size...2 The Problem: Determine sample size, n • Ensuring a required level of precision Most efficient and largest for a given fixed budget, B Survey budget, B, is fixed,](https://reader034.vdocuments.us/reader034/viewer/2022042213/5eb689e583ec8c4e860651c2/html5/thumbnails/43.jpg)
43
Sample Allocation to Domains (Contd.)
Compromise between equal allocation for each domain and optimum allocation
h
hh
h MM
nn ⋅=∑
• For example, allocate sample size to domain h proportional to square root of its size:
![Page 44: Sample Size...2 The Problem: Determine sample size, n • Ensuring a required level of precision Most efficient and largest for a given fixed budget, B Survey budget, B, is fixed,](https://reader034.vdocuments.us/reader034/viewer/2022042213/5eb689e583ec8c4e860651c2/html5/thumbnails/44.jpg)
44
Sample size: Other Issues
Different survey variables may have different sample size requirements for a given desired precision Prioritise and select the critical study variables Compute required sample size for each Adopt the largest sample size required
Finally, Sample size determination and allocation is an iterative process.
![Page 45: Sample Size...2 The Problem: Determine sample size, n • Ensuring a required level of precision Most efficient and largest for a given fixed budget, B Survey budget, B, is fixed,](https://reader034.vdocuments.us/reader034/viewer/2022042213/5eb689e583ec8c4e860651c2/html5/thumbnails/45.jpg)
45
THANK YOU