Transcript Correlational Research - University of Massachusetts Amherst

Correlational Research
Goal: Description and Prediction
• Look at relationships amongst naturally
occurring variables. There are no “IVs”.
• Not a “true” experiment. No basis for
cause/effect type conclusions.
Examples
• Survey/questionnaire research (Chapter 5)
• Observational research (Chapter 4)
• Quasi-experimental research (Chapter 10)
• Single-subject/small-n research (Chapter 9)
• Mostly non-statistical, some use of correlation
Today’s Lecture:
Survey/Questionnaire Research
• Some terminology used in Survey Research:
• Survey- to examine or look at closely
• Census- a census is a survey of a population
where EVERY member of the population is
included in the survey data
• Questionnaire- a questionnaire is an instrument
used to measure in a survey. It is a set of
predetermined questions asked of all respondents
• In order to conduct a survey, you would select
a sample from your population
• measure that sample
• and then draw inferences about the population
based on the sample
• This is true in ALL research actually
• How do you select an appropriate sample that
will give you a good chance of answering your
research questions?
Sampling: a three step process
1) Define your population of interest
(A population is the set of all cases of interest)
2) Develop a sampling frame-a physical list of
all members of the population
3) Choose a technique for selecting a subset of
members from the frame that will be
“representative” of your population. You want
an “unbiased” sample .
Population, Sampling Frame, Sample,
Element
Example:
Attitudes toward domestic violence
• Target population: Enrolled UMASS students.
• Sampling frame: a list of all currently enrolled
UMASS students. Where would you get this
list? Would the list be complete?
• Sample: select a subset of students from the
list.
• What would your “sampling frame” be if your
target population were all college students in
the New England?
• What would it be if your target population were
all college students in the US?
• There are many different methods for selecting
members from a population to use as a survey
sample
• Some are better than others
• Some serve a particular purpose
• Two general categories of sampling techniques:
Non-probability and Probability
• We will talk about Non-probability samples
first
Non-Probability Samples
• A sampling procedure in which there is no
way to estimate the probability of any specific
element’s inclusion in the sample.
• We will cover two non-probability samples:
Convenience sample and Snowball sample
The Convenience Sample
• The most common form of non-probability
sample is the convenience sample
• convenience sample: elements are selected
based on availability and willingness to
participate
• Example: Shere Hite (1987), Women and
Love: A Cultural Revolution in Progress
• Mailed 100,000 surveys to various women’s
organizations in 43 states
• Part of survey asked about women’s sexual fantasies.
• People objected to the idea that the information in the
book was representative of typical American woman’s
sexual fantasy
• Shere Hite maintained that the results were based on a
large number (4500) surveys
• 4500 is a lot of surveys but actual response
rate was only 4.5%
• Could those women who did respond be very
different from those who did not respond???
• In general, a 50% response rate is considered
“adequate”, 60%= good, 70%= very good.
• Typical response rate= 30%
Snowball Sample
• a non-probability sampling technique that is
especially useful when the target population is
not one where you can easily get a sampling
frame
• Find one or two possible participants and then
use these to locate others
Example
• A survey of the safe-sex practices of
prostitutes in Springfield
• How (where) would you find a complete list of
Springfield prostitutes?
• Unlikely to find such a list
• Locate one or two prostitutes in Springfield
willing to participate in the survey
• Ask them for contact information on others
who might participate
Probability Sampling
• The probability of any particular element being
included in the sample can be determined
(calculated)
• This does NOT mean each element has an
EQUAL chance of inclusion
• We will cover four Probability Samples:
Simple Random, Systematic,
Stratified Random, and Stratified
Proportionate
Simple Random Sample
• Not so simple
• A probability sample where every element of
the population has an equal chance of
inclusion
• Must use a formal random method for
selecting elements from the sampling frame
• random # table, coin, dice, random # generator
Example #1
• Go to library at 11 AM on Thursday and hand
a survey to every person passing by
• Is this a truly random sample of UMASS
students?
• Why/why not?
Example #2
• Use a list of all students currently enrolled at
UMASS from the registrar
• Randomly select 50 students, contact via
phone
• Ask if they would be willing to respond to a
short, 5-min phone survey
• Is this a simple random sample?
• Why/Why not?
• What would you call these samples?
• In Psychology, if you work with people, you
will always have a convenience sample
because all people must be asked for their
participation and have the right to refuse…thus
they must be willing
• We only use willing participants
Systematic Sample
• A probability sample that is similar to a simple
random sample
• rather than using a formal randomization
process to select elements, you use a
systematic process such a “choose every nth
element” from the list.
• This is much faster and generates a sample
that, while not truly random, is unlikely to be
biased
• You do need to make sure “n” is set so that you
will span the entire list
• Suppose you want a sample of 100 students
and your list is 1,000 students long
• If you used n=3, then you would only cover the
first 300 people on the list
• If the list is ordered systematically in anyway
(by year of graduation, for example) you could
end up with a very biased sample.
• If you used n=10, you would cover the entire
list and have lower risk of a biased sample
Stratified Samples
• A probability sample where you break the
sampling frame into “strata” (layers or
categories) and then sample from each strata
• Two types: Stratified Random and Stratified
Proportionate
Stratified Random Sample
• Break frame into strata and randomly sample
an equal number of participants from each
strata
• Useful when you do inferential statistical tests
between (amongst) some particular types of
participants where some categories are
underrepresented relative to others
• Creates approximately equal group sizes so
that you can use inferential statistics as a
tool to analyze results
Example:
Attitudes toward Euthanasia at a college campus
• Interested in the relationship of religion and
attitudes toward euthanasia
• want a sample of 200 students
• want to contrast attitudes of those of Christian
religions (90% of population) and Atheists
(10% of population)
• In order to use inferential stats, you need fairly
equal group sizes
• If you use random sampling you would likely
end up with about 90% (180) Christian and
10% (20) Atheist
• Split sampling frame into a list of Christian and
a list of Atheists
• Sample an equal number (50%) of participants
from each list (100 Christian and 100 Atheist)
Stratified Proportionate Sample
• Break frame into strata and randomly sample
from each strata such that the proportions in
the population are the same as the proportions
in the sample.
• This sampling technique would be best if you
want a truly representative sample in order to
describe the “typical” attitude
Example: Attitudes toward night-time
safety on campus
• Want a sample of 200 students
• Know that this attitude differs quite a bit for
men versus women
• Want to describe “typical” UMASS students’
attitude, you are NOT interested in contrasting
male and female attitudes.
• Population is 80 % women and 20% men
• Select a sample such that it reflects these
proportions, 80% women (160 women) and
20% men (40 men)
Three Survey Designs
Cross-sectional design
• Used for descriptive purposes, describe
characteristics of the population or differences
between two or more populations
• One sample drawn from the population at one
point in time
• Example: the surveys done in Psych 241 labs
Successive Independent Samples Design
• Same population but different samples surveyed
multiple times over some time period
• Can look for changes in the population
• Example:
• do a survey on attitudes toward drug use in 1970,
1980, 1990, 2010 using same survey and same
population (UMASS students) but different
sample (people) each time
Second example: Sax et al (2006)
• Survey done every year since 1966
• 350,000 each year from 700 colleges and
universities
• Largest, longest empirical study of higher
education in US
• Over 10 million students so far asked about 40
questions
Data from Sax et al
Longitudinal Design:
• Same population, same survey, and same
sample (same people) surveyed repeatedly over
time
• Very labor intensive, costly and difficult but
can provide some information about why
populations change.
Example
• Longitudinal study of eating habits and
disorders
• Mostly studied in adolescents
• What happens as people become young adults?
Middle-aged?
• Followed 625 women and 276 men beginning in college.
Surveyed in 1982, 1992, 2002
• 1992: women’s eating-disorder symptoms, chronic
dieting, and body dissatisfaction decreased. Men, who
rarely reported problems in college, saw an increase in
weight gain, dieting, and symptoms of disordered eating 10
years after college.
• 2002: Across the time period, women demonstrated the
most overall body dissatisfaction, most dieting, most
disordered eating. Body weight increased for both men and
women across the 20 years. Men’s dieting and weight
dissatisfaction was greatest in the 1992 survey, women’s
greatest dissatisfaction was in the 1982 survey and
decreased across the time period to be at its lowest in 2002
(less dieting, less disordered eating, less dissatisfaction).