Basic Sampling Issues Chapter 11
Basic Sampling Issues
Chapter 11
What is samplingSampling: a way of studying a subset
of the population but still ensuring “generalizability”
(vs. census – study of entire population) – does the study have external validity?
Definitions of Important TermsPopulation or Universe – entire set of
elements to be studiedCensus – all elements that completely
make up the population.Sample – a subset
Unit of Analysis
Level of social life beingstudied – individuals orgroups of individuals
Child Neighborhood
Elements Individual members of thepopulation
Charlie, Lucy, Linus, Patty,Violet, etc.
Midtown, Natomas, Land Park,
Sampling Frame
List of all elements orother units containing theelements; used fordrawing sample
Public school rollsPhone listingsMarketing list ofhouseholds with children
List of neighborhoodsList of cities in Sacramentoregion
Steps in Developing a Sample Plan
Step 1: Define thePopulation of
Interest
Step 2: Choose aData collection Method
Step 3: Choosing a Sampling Frame
Step 5: Sample Size
Step 4: Selecting a Sampling Method
BoundariesOperationalImplementability
Sampling MethodProbability samples: Samples in which
every element of the population has a known, nonzero probability of selection.GeneralizableSampling error Expensive; More time and effort needed
Non-probability samples: Samples that include elements from the population selected in a nonrandom manner.Hidden agendasBiased towards well known members of the population;
Biased against unusual population members
Sampling and Nonsampling ErrorsParameter vs. Statistic (Estimate)Sample statistic: statistic (e.g. mean)
computed from sample data- Population parameter: true value for
statistic (e.g. mean) for population (we don’t know this)
- Sampling error: population parameter – sample statistic (we don’t know this)
- Confidence interval: interval in which we can be confident that true value lies, based on sample statistic and its standard error
Advantages Of Probability Samples
1. Information from a representative cross-section
2. Sampling error can be computed
3. Results are projectable to the total population. Disadvantages Of Probability
Samples
1. More expansive than nonprobabiity samples
2. Take more time to design and execute.
Disadvantages of Nonprobability Samples
1. Sampling error cannot be computed
2. Representativeness of the sample is not known
3. Results cannot be projected to the population. Advantages of Nonprobability Samples
1. Cost less than probability
2. Can be conducted more quickly
3. Produces samples that are reasonably representative
Classification of Sampling Methods
SamplingMethods
ProbabilitySamples
SimpleRandom
Systematic Stratified
Non-probability
Judgment
Convenience Snowball
Cluster
Quota
Sampling ErrorThe error that results when the same sample is not perfectly representative of the population.
Remember?Sampling And Nonsampling Errors
+- sns
+-X =
X = sample mean
= true population mean
s = sampling error
ns = nonsampling error
Sampling ErrorThe error that results when the same sample is not perfectly representative of the population.
• Administrative error: problems in the execution of the sample (can be reduced)
• Random error: due to chance and cannot be avoided; but can be contolled by random sampling and…..estimated!
Measurement or Nonsampling Error
Includes everything other than sampling error that can cause inaccuracy and bias (data entry, biased q’s, bad analysis etc).
Sampling And Nonsampling Errors
Probability Sampling Methods
Simple Random SamplingA probability sample is a sample in which every element
of the population has a known and equal probability of being selected into the sample- EPSEM.
Probability of Selection = Sample Size
Population Size
Probability Sampling Methods
Systematic SamplingProbability sampling in which the entire population is
numbered, and elements are drawn using a skip interval.
Skip Interval = Population Size
Sample Size
Probability Sampling Methods
Stratified SamplesProbability samples that select elements from relevant
population subsets to be more representative.
Cluster SamplesProbability sample of geographic areas
Three steps: In implementing a properly stratified sample:
1. Identify salient demographic or classification factors correlated with the behavior of interest.
2. Determine what proportions of the population fall into various sub subgroups under each stratum.
• proportional allocation
• disproportional or optimal allocation
3. Select separate simple random samples from each stratum
Stratified SamplesProbability samples that select elements from relevant population subsets to be more representative.
Cluster SamplesSampling units are selected in groups.
1. The population of interest is divided into mutually exclusive and exhaustive subsets.
2. A random sample of the subsets is selected.• One-stage cluster—all elements in subset selected• Two-stage cluster—elements selected in some
probabilistic manner from the selected subsets
StratifiedExample
Reasonfor use
StrataDivide city into districts2. Draw random sample ofhouseholds from each district.
To ensure desired number of households in each district.
Cluster1.Divide city into districts (clusters).2.Draw random sample of districts.3.Draw random sample ofhouseholds from each district.
To make it easier to do door-to-doorsurveys.
Handout 1 – Baseball Example1. Ramon Aviles 0.2772. Larry Bowa 0.2673. Pete Rose 0.2824. Mike Schmidt 0.2865. Manny Trillo 0.2926. John Yukovich0.161 Mean = 1.565 / 6 = 0.261
SRS of sample size = 2 Mean Error Aviles, Bowa 0.272 +0.011Aviles, Rose 0.280 +0.019Aviles, Schmidt 0.282 +0.021Aviles, Trillo 0.285 +0.024Aviles, Yukovich 0.219 -
0.042Bowa, Rose 0.275
+0.014Bowa, Schmidt 0.277 +0.016
SRS of sample size = 2Bowa, Trillo 0.280 +0.019Bowa, Yukovich 0.214 -0.047Rose, Schmidt 0.284 +0.023Rose, Trillo 0.287 +0.026Rose, Yukovich 0.222 -0.039Schmidt, Trillo 0.289 +0.028Schmidt, Yukovich 0.224 -0.037Trillo, Yukovich0.227 -0.034
StratificationLet’s divide the sample into two strataOne with Yukovich and another with all
othersStratum 1: YukovichStratum 2: Aviles, Bowa, Rose, Trillo,
Schmidt
Stratified Sampling1. Yukovich, Aviles2. Yukovich, Bowa3. Yukovich, Rose4. Yukovich, Schmidt5. Yukovich, TrilloWeight the sample. Why? For anyone from Stratum 2, multiply
their value by 5
Example – Mean computationYukovich, SchmidtYukovich = 0.161Schmidt = 0.286Therefore, Schmidt’s value is (0.286 * 5)
which is 1.43Yukovich + Schmidt = 0.161 + 1.43 = Mean (Yukovich + Schmidt) = 1.591 / 6 =
0.265
Stratified Sampling1. Yukovich Aviles 0.258 -
0.0032. Yukovich, Bowa 0.249 -0.0123. Yukovich, Rose 0.262
+0.0014. Yukovich, Schmidt 0.265
+0.0045. Yukovich, Trillo 0.270
+0.009What’s happening to errors of estimate?
Nonprobability Sampling Methods
Convenience SamplesNonprobability samples used primarily because they are
easy to collect ; Theory testingJudgment Samples
Nonprobability samples in which the selection criteria are based on personal judgment that the element is representative of the population under study
Nonprobability Sampling Methods
Snowball SamplesNonprobability samples in which selection of additional
respondents is based on referrals from the initial respondents.
Quota SamplesNonprobability samples in which a population subgroup
is classified on the basis of researcher judgment Different from Stratified