Selecting Your Sample

1
Selecting Your Sample
2
Determining the Sample Size

• To calculate the specific sample size required
  for your site you need the following information
• Level of Confidence
• Acceptable margin of error
• Determine the starting or baseline levels of the
  indicators (diabetes if doing Step 3)
• Design effect

3
Level of Confidence

• The level of confidence for cross sectional
  studies is 95.
• For example, if you have a prevalence of 12.0
  and a 95 CI of (11.0 13.0), then you would be
  able to say that in 95 out of every 100 studies,
  the prevalence of factor X would be between 11.0
  and 13.0.
• 95 CI translates into 1.96 for calculating
  sample size.

4
Acceptable Margin of Error

• The margin of error is the difference between the
  population proportion and the sample proportion.
• The standard choices range from 1 and 5 of the
  population.

5
Determining the Baseline Level for Indicators

• Select the lowest prevalence of the indicators
  that you are surveying.
• If you are doing Step 3, this is most likely
  going to be diabetes.

6
Design Effect

• Sample size is commonly calculated assuming a
  simple random sample.
• Provides a correction for the loss of sampling
  efficiency from using cluster sampling.
• If no information on previous surveys is
  available, assume a DE of 2.

7
Create Table of Information
8
Equation for Calculating
Sample Size
n
9
Equation for Calculating
Z Level of confidence P Baseline level for
indicator E Margin of error
n
10
Create Table of Information
Z 1.96 E 0.05 P 0.10
11
Equation for Calculating
Z1.96 E0.05 P 0.1
n
12
Equation for Calculating
138.3
Sample Size
13
Equation for Calculating
138.1
Number will only change if the sample size
exceeds 5 of the total population.
14
Including Age-Sex Stratum
Sample size 138.1

• Take into account
• that we want to make age-sex comparisons
• cluster sampling

Sample size (number age-sex stratum) Design
effect
Sample size 138.1 8 2 2,211
15
Adjust Sample Size for Non Response

• Sample Size as calculated (2,211) adjusts for
  cluster sample design.
• Assume 80 response rate and inflate sample size
  and accordingly.
• (2,211.20) 442
• Final Sample size 2,211 442 2,653

16
STEPS Sampling Spreadsheet
17
Add Districts to PSU Spreadsheet

• Add names of districts and the estimated size of
  each district.

18
Select Number of Clusters

• Type in the number
• of clusters to select.
• Type in the random number.

19
Selected Districts
20
Population Distribution
21
Sample Distribution
Sample Total sample size Proportion of
population
Total Castries 2,653 0.5511
22
Determine Cluster Size

• Assume 50 participants per cluster

23
Clustering SSU Spreadsheet

• Allows selection of settlements by PPS sampling
• Duplicate Spreadsheet until have five
  spreadsheets (number of districts selected)

24
Record Names of Selected Districts

• Will see spreadsheets at the bottom of the
  workbook.
• Rename the spreadsheets to reflect the names of
  the selected districts.

25
Select Settlements for Each District

• Select Anse La Raye settlements.
• Need to select three (3) clusters.

26
Type in Settlements and Estimated Population Size

• Label the settlements with either numbers or
  names.
• Type in the estimated size of the sampling units.

27
Determine the Number of Clusters

• Type in the number
• of clusters to select.
• Type in the random number.

28
Anse La Raye

• Cluster 3 is associated with Au Tabor Hill.
• Au Tabor Hill has less than 50 households, so
  combine it with settlement 4 and select 50
  households at random from the combined list.

29
Anse La Raye

• Cluster 15 is associated with Millet Caico.
• Millet Caico has 64 households.

30
Select the Settlements for the Remaining Districts

• Enter the settlement information for each
  district into the corresponding spreadsheet.
• Use the table below to determine the number of
  clusters to select

31
Select Households from Each Settlement

• Use the Rand Hhold spreadsheet to select
  households to sample randomly.
• Determine the number of households in each
  settlement
• If you have an address list with each household
  by address.
• Otherwise, when you are visiting a settlement,
  list all the households by location.

32
Duplicate Rand Hhold for Number of Settlements
Selected

• There are 52 clusters in the sample.
• Duplicate the Rand Hhold spreadsheet 51 times
  using the Duplicate Spreadsheet button.

33
Rename Rand Hhold for Selected Settlements

• You will see spreadsheets at the bottom of the
  workbook
• Rename the spreadsheets to reflect the names of
  the selected districts

34
Select Households for each Settlement

• Determine how many households there are in each
  settlement.
• List all the households on the Rand Hhold
  spreadsheet and randomly select 50 households.
• One participant will be selected from each
  household.

35
Millet Caico Rand Hhold

• Millet Caico has 64 households, of which 50
  will be selected.

36
Sampling Options

• This example has shown only one possibility.
  There are many other ways to draw and divide the
  same population.
• You could divide the sample into constituencies
  instead of districts.
• You could select all the districts to sample and
  then select a smaller, more representative group
  from all the districts.

37
Determining Your Sample

• Your sample will be a balance between
• What information is available (sampling frame)
• The scope of the survey
• The budget (finance and human resources)

38
Statistical Assistance

• You should always have someone review your
  sampling methodology.
• If you need assistance, contact CAREC or PAHO or
  WHO-Geneva.