Sociology 220

1
Sociology 220

• Feb 4, 2009
• Prof. J. Brines MGH 231

2
Sampling

• Sampling issues parallel measurement issues
• Concerns about bias
• Samples include some participants and exclude
  others.
• Do our decisions about inclusion/exclusion
  introduce bias into our findings?

3
Convenience Sampling
4
Vox Populi (Latin voice of the people)

• 1907 Sir Francis Galton publishes a paper in
  Nature that mathematically demonstrates the
  wisdom of crowds.
• The Man-on-the-street interview is a popular
  sampling strategy of the general public used by
  the media.
• Problems?

5
Problems with Vox Pop

• People milling around a city center are not
  representative of the general public.
• Unsystematic picks ? random picks
• The content of interviews is usually edited after
  the fact!

6
Like other facets of the research process

• We want to introduce a procedure, and rules for
  applying it, that minimizes bias and maximizes
  the accuracy of our results.

7
Sampling Some Definitions

• Target Population The population or universe of
  things you wish to make generalizations about

8

• Sampling Frame The tangible list or register of
  elements that one draws a sample from.

9

• Sample The collection of sampled cases that are
  taken from the sampling frame.

10
Probability Sampling

• A class of sampling techniques where
• Every unit in the target population has a known
  probability (chance) of being selected into the
  sample

11
Types of Probability Samples

• Simple Random Sample
• Systematic Sample
• Multistage Cluster Sample

12
Simple Random Sample or SRS

• Defined as a sample where
• every unit in the target population has an equal
  probability of being selected into the sample.

13
Random Numbers Table
14
Systematic Sampling (more efficient than SRS)

• Two steps
• -random entry into the sampling frame
• -selection of every kth unit thereafter, where
• k units in sampling frame
• desired sample size
• E.g. k 33,152 22.1
• 1500
• Therefore, select every 22nd case on the list.

15
Cluster (Multistage) Sample

• Every element in the population has a known (but
  usually not equal) chance of being selected into
  the sample.

16
Cluster Samples

• Used to draw samples from large populations when
  a sampling frame for the target population is not
  available.
• Either it is impossible or too expensive to
  compile an exhaustive list of all the elements in
  the target population.
• Most polls and large-scale surveys are based on
  cluster samples

17
Examples of target populations sampled using
cluster sampling

• Church members in the U.S.
• All undergraduates pursuing science majors in 4
  year colleges/universities
• In short Any meaningful population for which a
  comprehensive list does not exist.

18
Condition for Cluster Sampling

• If the units for a target population are already
  socially grouped in some way, and a list of these
  groupings can be compiled, use cluster sampling.

19

• Cluster samples are used for most public opinion
  surveys and an important Census survey, called
  the Current Population Survey.
• These are usually multistage cluster samples,
  involving 5 or more stages (clusters)

20
Example Two-stage cluster sample

• Cluster sampling is based on the repetition of
  two steps listing and sampling.
• First, a list of primary sampling units (4-yr
  colleges/universities) is drawn up. A random
  sample of those is selected.
• Next, a list of secondary sampling units
  (students who are science majors) is compiled
  from the first-stage sample, and a sample is
  taken from this list.

21
Drawing a Cluster Sample of American Adults

• Step 1
• Determine sample size for final sampling units
  (adults)
• N2000

22

• Step 2. Determine clusters P (selection) if
  random
• 1st cluster State 25/50
• 2nd County 100/752
• (counties in selected states)
• 3rd City/Town 100/501
• (cities in picked counties)
• 4th city block 200/5188
• (blocks in picked cities)
• 5th address/household 2000/6212
• (addresses in blocks)
• 6th adult 2000/3916
• (adults at addresses)

23
How many clusters and how many units within
clusters should be selected?

• General rule
• The sample will give results that are closer to
  the truth for the target population if the number
  of clusters is maximized, and the number of units
  in each cluster is minimized (smaller
  denominators).

24
Desired properties of samples

• Representativeness Does the sample represent the
  target population? Is it unbiased?
• A function of
• 1) completeness of sampling frame(s)
• 2) how the sample was drawn
• (random is ideal)