Producing Data: Sampling

1
Chapter 8

• Producing Data Sampling

2
From Exploration to Inference
3
Types of Studies

• Observational studies ? individuals are studied
  without an experimental intervention (e.g., most
  surveys)
• Experimental studies ? individuals receive an
  experimental intervention to determine its effect
  (e.g., a study of a drug effectiveness)

4
Example of an Observational Study (Weight Gain
CHD)

• Purpose understand relationship between weight
  gain and coronary heart disease (CHD)
• 115,818 women, 30 to 55 years of age, recruited
  in 1976
• Measure weight and height at age 18 and at
  recruitment, record weight gain
• Followed individuals for 14 years
• Record fatal and nonfatal CHD outcomes (1292
  cases)
• Adjusted results for lurking variables such as
  smoking and family history of CHD

Source JAMA 1995273(6)461-5
5
Illustrative Example Results

• Compared to subjects who gained less than 11
  pounds
• Subjects who gains 11 to 17 lbs 25 more likely
  to develop CHD
• 17 to 24 lbs gained 64 more likely
• 24 to 44 lbs gained 92 more likely
• 44 lbs gained 165 more likely

6
Illustrative Example (Questions)

• What is the population in this study?
• What is the sample?
• What makes this study observational?
• Can we say that weight gain caused CHD?
• Can we say weight gain is associated with CHD?

7
Sample Quality

• Poor quality samples favor a certain outcome ?
  misleading results ? sampling bias
• Examples
• Voluntary response sampling Allows individuals
  to choose to be in the study, e.g., call-in polls
  (pp. 1789 in text)
• Convenience sampling individuals that are
  easiest to reach are selected, e.g., Interviewing
  at the mall (p. 179)

8
Voluntary Response Bias

• To prepare for her book Women and Love, Shere
  Hite sent questionnaires to 100,000 women asking
  about love and sexual relationships
• Only 4.5 responded
• Respondents were fed up with men and eager to
  fight them
• Selection bias angry women were more likely
  to respond ? sampling bias

9
Convenience Sample

• A lab study was conducted to see if a drug
  affected physical activity in lab animals
• The lab assistant reached into the cage to select
  the mice for study
• The less active mice were chosen ? made it seem
  like the drug decreased physical activity ?
  sampling bias

10
Simple Random Sample (SRS)

• To avoid sampling biases, use chance (random)
  mechanisms to select subjects
• The most basic random sampling mechanism ? Simple
  Random Sample (SRS)
• SRSs ? every conceivable subset has the same
  chance to be studied

11
Selecting a SRS

• Methods we can pick them from a hat, use a
  random number generator, or use a table of random
  digits (Table B) to derive our sample
• We will use Table B
• Each digit 0 to 9 is equally likely
• Entries are independent (knowledge of one entry
  gives no information about any other entries)

12
Choosing a Simple Random Sample (SRS)

• STEP 1 Label each individual in the population
  with a identification number
• STEP 2 Use Table B to select numbers at random
  (enter table at a different location each time it
  is used)

13
Selecting a SRS (Illustration)

• Population of N 30 individuals
• Labeled the individuals 01 30
• Select a row in table at random
• Enter table at different random location each
  time (e.g., to illustrate, enter at row 106)
• Row 106 with lines to indicate pairs 68417
  35013 15529
• First two individuals relevant entries are 13 and
  15

14
Remainder of Chapter

• Not responsible for the sampling designs
  discussed on pp. 200201
• Are responsible for the cautions (pp. 201202)
• Undercoverage some population groups left out of
  sampling process ? sampling bias
• Nonresponse bias some individuals do not respond
  or refuse to participate ? sampling bias
• Even good quality samples may not be a perfect
  reflection of the population due to random
  sampling error ? unavoidable dealt with in
  future chapters