Transcript Sampling-Fundamentals

Sampling Fundamentals
1
Basic Concepts

Population: the entire group under study (or of interest)
Exercise: Define population for a study seeking to assess
SUU student attitudes towards a) program quality and
delivery, b) program content, and c) social environment.

Sample: subset of the population
Used to represent the population

Sample unit (elements): basic unit investigated (choose
sampling units/elements when sampling)
Individuals, households, etc.

Census: data collected from EVERYONE in population
2
Basic Concepts (continued)

AGAIN: total error = sampling error + nonsampling error
Sampling error: error due to taking a sample (+/-zs)
Nonsampling error: everything else (measurement, data
analysis, etc.)
Sample frame: list from which the sample is selected
 Sample frame error: Pop’n members not in frame, and
members in frame not in pop’n of interest

3
Reasons for Sampling
 Cost
 Too much information to handle
 Sampling can be more accurate
Nonsampling errors can overwhelm reduction in
sampling errors
– Sampling work behaviors example
– Census Bureau
 Time problem
4
Developing a sampling plana
 1. Define the population of interest.
 2. Choose a data-collection method (mail,
telephone, Internet, intercept, etc.).
 3. Identify a sampling frame.
 4. Select sampling method
 5. Determine sample size.
 6. Develop operational procedures for selecting
sampling elements/units.
 7. Execute the operational sampling plan.
5
PROBABILITY SAMPLING METHODS
Each member of population has a ‘known’
probability of being selected
 Simple Random Sampling: Each member
has an equal probability of being selected
Blind Draw Method
Table of Random Numbers
Useful for small samples, when Random Digit
Dialing (or +1) is appropriate, and computerized lists
6
PROBABILITY METHODS (Cont’d)
 Stratified Sampling: Population is
segmented (stratified), and then samples are
chosen from each strata using some other
method
Can be more efficient (smaller sampling error)
– Homogeneous within, heterogeneous without
Useful when interested in different strata (e.g., small
numbers, etc)
Disproportionate versus proportionate
7
PROBABILITY METHODS (Cont’d)
 Cluster Sampling: Population is divided into
groups, or clusters, and then clusters are
randomly chosen.
Homogenous without, heterogeneous within
Every unit in cluster examined, OR
A Random (or systematic) sample is taken from
chosen cluster (2-stage or 2-step approach)
Careful with the probabilities!
8
PROBABILITY METHODS (Cont’d)
 Systematic Sampling: Randomly choosing a
starting point and then choosing every nth member.
Example: Need 52 data points (daily sales) for a
year
– Skip interval = 365/52=7.01
– Randomly choose 1 day out of first 7, then choose every
7th one after that.
Variation: Choose every nth visitor
9
NONPROBABILITY SAMPLING
METHODS
Probability of selection not known, and hence
representativeness cannot be assessed
Technically, confidence intervals, H0 tests, etc. not
appropriate
 Convenience Samples:
Shopping mall intercepts, classes asked to fill out
questionnaires, etc.
 Judgment Samples: Someone puts together what is
believed to be a relatively representative sample
Ex.: Test markets
10
Nonprobability Sampling (Cont’d)
 Referral (or Snowball) Samples
 Quota Samples
EXAMPLE: Choose sampling units so their
representation equals their frequency in the pop’n
(e.g., 52% females, 48% males)
11
Identifying the Target Population
Determining the Sampling Frame
Reconciling the
Population, Sampling
Frame Differences
Selecting a Sampling Frame
Probability
Sampling
Non-Probability
Sampling
Determining the Relevant Sample Size
The Sampling
Process
Execute Sampling
Data Collection From Respondents
Handling the NonResponse Problem
Information for Decision-Making
12
Nonresponse Bias

Reason for nonresponse:
Refusal
Lack of ability to respond
Not at home
Inaccessible

Handling nonresponse
Improve research design
Call-backs
Estimate effects
– Sample nonrespondents; trends
13