SAMPLE DESIGN: WHO WILL BE IN THE SAMPLE ? (CONTINUED)

1
SAMPLE DESIGN WHO WILL BE IN THE SAMPLE ?
(CONTINUED)

• Lu Ann Aday, Ph.D.
• The University of Texas
• School of Public Health

2
COMBINED DESIGNS

• A. Area probability sample design (example
  PPS)
• B. Random digit dialing (RDD)
• C. List sample

3
PROBABILITY PROPORTIONATE TO SIZE (PPS) EXAMPLE
(Aday Cornelius, 2006, Table 6.2)
STEPS EXAMPLE
1. Estimate the desired sample size (n). 77
2. Fix the desired cluster size (nc). 7
3. Calculate the number of clusters (c ) needed to achieve the desired sample size n/nc. 77/7 11
4. Estimate the total number of units in the universe from which the sample will be drawn (N). 2200 Col. B, Table
5. Calculate the cumulative total of the number of units across all clusters in the universe. Col. C, Table
6. Calculate the sampling interval (k) for selecting clusters for the universe N/c. 2200/11200
7. Pick a random starting point (r ) to select clusters within the designated sampling interval (Step 6), using a random numbers table. 50
8. Calculate the selection numbers (HU ) for the blocks to be sampled by entering the random starting point, adding the sampling interval, and then repeat the process to identify sampled blocks. Col. D, Table
9. Assign cluster numbers to each designated block. Col. E, Table
10. Confirm in strata for sample agree with in universe. Col. B, E (), Table
4
RANDOM DIGIT DIALING WAKSBERG-MITOFSKY DESIGN

• Target population residents of the State of
  California
• STAGE 1
• 1. Implement random or systematic selection of
  area code-central office code combinations
  for area (xxx) xxx.
• 2. Add two random digits to each area
  code-central office code combination.
• 3. Prepare list of possible 8 digit numbers,
  which become PSUs with clusters of 100
  numbers each (xxx) xxx-xx00 thru xx99.
• 4. Assign last two digits of the number
  randomly, such as (xxx) xxx-xx24.
• 5. Dial the resulting number.
• 6a. Eligible household numbercomplete the
  interview. Retain
• PSU of 100 numbers.
• 6b. Ineligible household numberterminate the
  interview.
• Eliminate the PSU of 100 numbers from
  further calls.

5
RANDOM DIGIT DIALING WAKSBERG-MITOFSKY DESIGN
(cont.)

• Target population residents of the State of
  California
• STAGE 2
• 7. Randomly assign two new digits to end of
  cluster of
• numbers for same or new PSU (as
  appropriate).
• 8. Repeat process until desired sample size is
  reached.

6
LIST SAMPLE

• Target population U.S. dentists in active
  practice
• STAGE 1 Identify eligible dentists.
• STAGE 2
• 1. Determine sampling fraction.
• 2. Draw systematic random sample of eligible
• dentists.

7
SAMPLING RARE POPULATIONS

• Screening- Ask respondents whether they/household
  have the attribute of interest and drop those
  from sample that do not.
• Disproportionate sampling- Assign a higher
  sampling fraction to stratum that has attribute
  of interest.
• Network sampling- Ask the respondents if they
  know others in family network (defined in certain
  way) who have attribute of interest.
• Dual frame sampling- Use a second sampling frame
  containing elements with attribute of interest to
  supplement original frame.

8
PROCEDURES FOR SELECTING THE RESPONDENT
SELECTION TABLES

• Kish tables- Ask about all potentially eligible
  individuals in the household, list them and then
  use Kish tables.
• Troldahl-Carter-Bryant (TCB)- Ask how many
  persons live in the household, how many of them
  are women, and then use TCB selection charts.

9
PROCEDURES FOR SELECTING THE RESPONDENT
RESPONDENT CHARACTERISTICS

• Hagan and Collier Method- Ask to speak with one
  of four types of age-sex individuals and if no
  one of that gender, ask for counterpart of
  opposite gender (youngest adult male/youngest
  adult female/oldest adult male/oldest adult
  female).
• Last/Next Birthday Method- Ask to speak with the
  person who had a birthday last or who will have
  one next.

10
SURVEY ERRORS Deciding Who Will Be in the
Sample
Systematic Errors Systematic Errors Variable Errors
Noncoverage bias (frame bias) Noncoverage bias (respondent selection bias) Design effects
Solutions to errors Match the sample frame to the target population. Use multiple sample frames, if needed, to more fully capture the target population of interest. Employ methods for randomly selecting the study respondents. Try to balance the complexity (especially the cluster nature) of the sample design needed to address the study objectives in relationship to survey costs.
11
REFERENCES

• Bennett, S., Woods, T., Liyanage, W.M., Smith,
  D.L. (1991). A simplified general method for
  cluster-sample surveys of health in developing
  countries. World Health Statistics Quarterly 44
  98-106.