Page 126 -

112 CHAPTER 5 Surveys

USE OF SAMPLING FRAMES IN STUDYING INTERNET ACCESS—CONT’D
Since the survey is used to make national population estimates, the research
approach used must be highly structured and controlled. Data was more
recently collected using a web-based survey, with a paper letter mailed to public
libraries to inform them about the existence of the survey. The letter included an
identification code so that the survey data collected was identified to a specific
library system or branch. The 2008 survey included 16,457 library outlets, a
6984 sample frame, and 5488 library responses (78.6%). The 2011–12 survey
(with data published in 2012) included 16,776 library outlets, an 8790 sample
frame (stratified by state and proportional by state and metropolitan status), and
7260 responses (82.5% response rate). See http://www.plinternetsurvey.org/ for
more information on methodological issues related to this survey.

STRATIFICATION
Shklovski et al. (2008) were interested in studying how technology influences
the maintaining of friendships after a residential move. A sample of 6000
individuals was chosen from the US Postal Service's Change of Address
Database. These were all individuals who had moved in the previous few
months. The sample was stratified so that 1/3 of those selected had local moves,
of 50 miles or less, while the other 2/3 selected had longer distance moves, of
50 miles or more. This stratification was done because the researchers were
interested in studying long-distance moves. However, it is implied in their
write-up that a majority of moves are local moves. Of the 6000 people selected
from the database, 1779 (32%) responded to the survey. Two follow-up surveys
were sent to the 1779 individuals who responded to the first survey. The second
survey received 1156 responses, and a third survey received 910 responses. This
research provides an example of stratification.

5.4.2 RESPONSE SIZE

If it is feasible for random sampling to be used in the research, this is preferable.
However, the next question that comes up most often is, “how many responses do
I need?” The statistics on this are not as clear as in, say, statistics in experimen-
tal design, where there is a clear threshold of significance or nonsignificance. In
probabilistic sampling, the number of responses required depends on what level of
confidence and margin of error are considered acceptable. For instance, for a simple
random sample, a sample size of 384 may lead to a 95% confidence level with a ±5%
margin of error (Sue and Ritter, 2007). That means that “if the survey were conducted
100 times, the true percentage would be within 5 percentage points of the sample
percentage in about 95 of the 100 surveys” (Sue and Ritter, 2007, p. 30). To change
the margin of error to ±4%, 600 responses are needed; for ±3% margin of error, 1067