This preview shows page 1. Sign up to view the full content.
Unformatted text preview: s √ Quality of a sample affects the estimation of parameters
√ To find out if a sample is any good… Just look at how it
was chosen. Any selection bias? √ When there is a high nonresponse rate, look out for
nonresponse bias. NUS/FOS/DSAP 7 GEM2900 Semester 1, 2009/1010 More on Bias in Sampling…
Example: The Year the Polls Elected Dewey...
Before the US presidential election in 1948, 3 major polls
covered election campaign, Crossley, Gallup and Roper, had
declared Thomas Dewey a winner over Harry Truman with
a lead of 5 percentage points.
The result of the election: Dewey 45%; Truman 50%...
Truman won! All 3 major polls made the wrong prediction.
What went wrong? NUS/FOS/DSAP 8 GEM2900 Semester 1, 2009/1010 More on Bias in Sampling…
Example: The Year the Polls Elected Dewey...
◙ Find out how the sample was chosen…
The method used to chose sample is called quota sampling.
… A fixed number of individuals (quota) with certain key
characteristics were predetermined to resemble the
population. … Within assigned quota, the interviewers of the poll were
free to choose anyone they like.
… Any human choice is, more often than not, subject to
bias. NUS/FOS/DSAP 9 GEM2900 Semester 1, 2009/1010 Using Chance in Survey
To minimize the bias in sampling, an objective and
impartial probability method should be used to draw
samples.
Some probability methods in sampling… … Simple random sampling √ Each observation in the population has an equal
probability of being included in the sample
√ It makes statistical sense, but hard to execute in
reality… Not practical
… Cluster sampling
… Stratified sampling NUS/FOS/DSAP 10 GEM2900 Semester 1, 2009/1010 Using Chance in Survey
Cluster Sampling:
√ Divide population into several subgroups (clusters)
√ Each cluster is representative of the population √ Randomly select 1 or few clusters √ Include all observations from selected clusters or
simple random samples from selected clusters to
form a sample
Population
divided into
few clusters NUS/FOS/DSAP Randomly selected
one or few clusters 11 GEM2900 Semester 1, 2009/1010 Using Chance in Survey
Stratified Sampling: Strata √ Divide observations to mutually exclusive subgroups
√ Observations within the same subgroup have some
common properties
√ Observations from different subgroups are
different with respect to certain properties
√ Use simple random sampling to choose
observations from each subgroup to form sample NUS/FOS/DSAP 12...
View Full
Document
 Fall '10
 CHEN

Click to edit the document details