correlational research
focuses on examining the relationships among variables
allows for study of individual differences, but you cannot make
conclusions about causality
correlation does not = causation!!!
what is the difference between correlational research and
experimental research
in experimental research, there is some kind of manipulation
in correlational research you are only measuring 2 or more
characteristics in the same individual and finding a correlation
what is the main concern in correlational research
investigating the relationships between naturally occurring variables
and with studying individual differences
two variables that are related to each other "in some fashion"...
are said to be correlated
direction of correlation
positive correlation or negative correlation
positive (direct) correlation
high score on one variable is associated with high score on a second variable
ex: height positively correlated with weight; study time and GPA
negative (inverse) correlation
high score on one variable is associated with a low score on a second
variable, or vice versa
ex: # weeks training negatively correlated with race time;
time goofing off and GPA
strength of correlation
correlation coefficient - pearson's r
ranges from -1 to 1
-1 - strong negative correlation
0 - no correlation
1 - strong positive correlation
as strength of correlation weakens...
points on the scatterplot move further away from a perfect diagonal line
threats to detection of linear relations between variables
non-linearity, restricted range
non-linearity
threat to detection of linear relations between variables
pearson's r describes the direction and magnitude of linear
association between variables
if the variables are associated in a non-linear fashion,
pearson's r cannot identify the nature of the relationship
scatterplot
provides a visual representation of the relationship shown by a correlation
example of curvilinear relationship in psych
yerkes-dodson law (inverted U)
arousal and performance - medium arousal means highest
performance, low or high arousal means low performance
pearson's r would not work for this
restricted range
it is important that you have a sample with a spread out range
otherwise you will get a weak correlation, or won't find a correlation
at all
ex: sample of SAT scores ranging from 400 to 1200 - strong correlation
vs. just a sample of kids who scored high (900-1200)
r^2
r^2 = coefficient of determination - will always be positive b/c it's squared
r^2 = portion of variability in one of the variables in the
correlation that can be accounted for by variability in the second variable
outlier
a score that is dramatically different from the remaining scores in a
data set
interpreting r^2
ex: SAT and GPA
if r=.5, r^2=.25 - here we can say that 25% of the
variability in GPA scores can be associated with SAT scores, the
other 75% is due to other factors
effect of outliers
gives a false representation of what the real correlation coefficient is
pulls the mean up or down
regression analysis
making predictions based on correlational research
if X and Y are strongly correlated, knowing score on X
allows you to predict score on Y
Y = a + bX
a = y intercept, b = slope
correlation does not equal causation
a correlation between two variables does not allow you to conclude
that one variable causes the other to occur
ex: research that is misinterpreted in the press
this is different from experimental research because there
you have manipulated a variable and controlled for other factors, so
you can say the DV is caused by the IV
directionality problem
causal relationship can occur in any direction
is A causing B or is B causing A?
ex: correlation between amount of TV and aggressive
behavior...could watching TV make kids aggressive, or do
aggressive kids like to watch TV?
the most you can say is that there is an association between
TV watching and aggression
to talk about A causing B you need to show...
1. A and B occur together2. A precedes B in time3. A
causing B makes sense in relation to theory4. other explanations
for co-occurrence can be ruled out
cross-lagged panel correlation
think about the box with the X inside, each corner has a variable,
and you find the correlation between each variable in all directions
.3-.4 is a strong correlation, anything higher is not
important, anything lower is not correlated
third variables and problems with causality
correlational research does not attempt to control extraneous variables
- but extraneous variables could account for the association
between 1 and 2
"variable 3 causes both 1 and 2
partial correlation
ruling out the other variable to see a correlation, attempts to
control for 3rd variables statistically
imagine mickey mouse's head (ears and face) - unless the
face is there, the ears wont touch and therefore don't interact - 1
and 2 work THROUGH 3
examining the influence of a third variable by looking at a partial correlation
if partial correlation is a lot smaller than correlation, this means
that the third variable is indeed accounting for the association
between the variables
if partial correlation is about the same as the correlation,
this means that that third variable doesn't account for the
association between the variables
personality research
correlations between different personality dimensions
ex: pessimism and depression - r=.56
therapy reduces depression but magnitude of correlation
stays the same..why? - as depression went down, so did pessimism
the nature-nurture issue
are certain characteristics more highly correlated among identical
versus fraternal twins?
do genetics or the environment play a bigger role in shaping
a person
multiple regression
examining the relations among more than 2 variables - multiple
predictors of a single criterion variable
ex: do SAT score, motivation and high school GPA predict
college GPA?
mediation
where the 3rd variable is necessary for a correlation to occur
think mickey, if the head is removed, there is no more correlation
necessary for a correlation
moderation
there is an X-Y correlation, and it doesn't depend on the 3rd
variable, but correlation changes depending on the level of the 3rd variable
ex: michelle is 3rd variable - tells you everything that is
wrong and changes the strength of relationship
changes what is already there
split-half reliability
split the test you are giving in half (ex: odd and even numbers) and
correlate the two halves.
someone scoring high on one half should score high on the
other half
test-retest reliability
the relationship between two separate administrations of the test.
a reliable test yields consistent results from one to the
other, so the reliabilities should be high on both
how to test for reliability
split-half reliability
test-retest reliability
a multiple regression study has...
one criterion variable and a minimum of two predictor variables
this allows you to determine that the two variables predict
some criterion, and the relative strengths of the predictors
multivariate analysis
examines the relationships among more than 2 variables
bivariate analysis
investigates the relationships between any two variabels
when would a multivariate analysis be used
when there are multiple factors that could lead to some outcome
ex: success in college cannot simply be measured by high
school GPA alone, it could be ACT/SAT score, extracurriculars, etc.
factor analysis
a large number of variables are measured and correlated with each
other, then groups of variables are clustered together to form factors
ex: giving children tests of different tasks (vocabulary,
geometry, puzzle) and finding the correlation between each test and
one other
quasi-experimental design
has all the characteristics of a true experiment, but there is
something missing, usually no manipulation of IV
cannot make causal conclusions because you did not have
complete control over all the variables
why is P x E is a quasi-experiment
P = person (subject) variable, can't be manipulated
cannot make causal statement because you can't rule out other variables
non-equivalent control groups design
used in order to evaluate effectiveness of the treatment
did the groups start/end in the same place?
if they started at the same place and ended at different
places, there might be some kind of correlation, if they started at
different places and ended at the same place, there might be some
kind of correlation
2 ways of interpreting different end scores for nonequivalent control
groups design
1. the IV worked
2. the groups were different to start with
possible problems with nonequivalent control groups
ceiling effect could explain why some groups didn't change at all -
they had no where to go to but down to begin with
must make sure there is room for movement (up or down) when
designing a study
when pretest measures are really low or high, regression to
the mean may occur
matching samples when the actual populations are different to
start with (ex: head start example - groups were different at start)
regression effect in nonequivalent control groups
if the sample is a biased measure of the population (too high or too
low) the effect of IV will go one way, but the true population mean
will pull them another way
think about head start study - if you sample only the top
performing head start students, their improvement wants to go up but
the real mean for the group is lower and pulls them down - they
cancel out and show no effectiveness
example of nonequivalent control groups design
little league study - one team coach given effectiveness training,
other not - self-esteem score of the players measured pre and post season
used two different leagues so there was a 50% chance win rate
langer and rodin study
study with nursing home patients given the option to do things on
their own vs. have a nurse do things for them - happiness, death rate,
nurse score measured
not randomly assigned because they had to allow the patient to
choose if they wanted to be more in control or not
hollingworth coca-cola study
used multiple kinds of procedures (counterbalancing, placebo/control
group vs. different caffeine dosage group, and double blind) which
gave them a range of results, as well as enough money to travel and go
to grad school
combined several techniques for a strong study
applied reseach
psychological studies that produce results that can be applied to the
real world for some kind of benefit
problems with applied research
ethics - consent and privacy, non-proper debriefing,
employees believe their job status depends on their participation
internal vs external validity - high external
validity because it models real life situations, low internal validity
because of possible confounds
between subjects design problem - cant use random
assignment, so you have to compare nonequivalent groups so reduced
internal validity
within subjects design problem-cant always properly
counterbalance so this creates order effects, attrition is also a
problem for long studies
what do you usually compare in nonequivalent control groups design?
the change scores between observation 1 and observation 2, after the
treatment in the experimental group, and before and after in the
control group
standard for presenting nonequivalent control groups design (stanley
and campbell)
O1 T O2 - experimental group
O1 O2 - control group
pittsburgh vs. cleveland plant example
studied 2 different plants producing pans, one implemented a new
"flexible" worktime schedule, while the other scheduled
people as normal. both made workers work 40 hrs/week. measured
productivity in the plants before and after the new schedule in one plant.
attempt to reduce the nonequivalency in groups - matching
this creates problems because you may be taking a sample that is too
high or too low, so the mean will pull them in one direction while the
effectiveness will pull them in another
this leads to no change - think head start study
interrupted time series design
taking multiple observations over a period of time before the
treatment, then imposing the treatment, then taking multiple
observations periodically after the treatment
O1 O2 O3 T O4 O5 O6
allows researcher to rule out alternative explanations of an
apparent change from pre to post test
variations on time series design
inclusion of control can help with interpretation - seeing a control
groups scores compared to the experimental groups scores over time,
before and after the treatment
switching replications - vary the time at which the
"event" or treatment occurs, can the change be directly tied
to the event?
measure multiple dependent variables - select some where you
expect change and others where you dont expect change, (ex: 3 strikes
law in california, misdemeanors vs felonies)
advantage of interrupted time series design
allows researcher to observe trends - consistent patterns of events
over time
rule out alternative explanations of an apparent change from
pre to post test
example of interrupted time series design
implementing a worker incentive program and observing the worker
productivity for a period of time before and after the new program was
implemented. this allowed researchers to compare the productivity pre
and post to see if the program worked.
example of a variation in a time series design - california crime index
implemented a "three strikes" program where after 3
felonies resulted in jail time. they compared this to misdemeanor
charges, which were not affected by the new program. they measured the
amount of felony charges before and after the program, and saw that
the amount of felonies decreased significantly while misdemeanor
charges remained relatively constant
interrupted time series design with switching replications
implementing the treatment or program in two locations at two
different times. if the same outcome can be seen in both locations
following the treatment, this is a good indication of the program working
archival research
going through existing records (medical files, census data, court
reports), data that you are not collecting, someone else already did
benefits of archival research
very convenient, ready to go, you don't have to collect anything more
medical research - can look at changes in cohorts
don't have to worry about reactivity - the subject has nothing
to react to, they don't have to worry about being observed
drawbacks of archival research
you're at the discretion of whoever collected it, limited to what has
already been collected
there might be one question you wanted to know that was not recorded
problems with consent - people may not have wanted their private
information released
experimenter bias - you know what you're looking for and may
ignore anything else that is important
quasi experimental design using archival research - ulrich
study using archival hospital records where researchers collected
data on patients after surgery in a room with a view of trees or a
brick wall. length of stay, nurses notes, minor complications after
surgery, request for medication was recorded - people in rooms with
the view of trees had an advantage and spent a shorter amount of time
in recovery
program evaluation
applied research that seeks to asses the effectiveness and value of a
public policy or specially designed program
4 purposes of program evaluation
needs analysis - determine community and individual needs for programs
formative evaluative - asses whether program is being run as
planned and if not, implement change
summative evaluation - evaluate program outcomes and get rid of
them if they are pointless
cost-effective analysis
example of program evaluation in connecticut
governor recognized that there was a record number of traffic
fatalities so he implemented a crackdown on speeding, the next year
there were less deaths but was it because of the crackdown? record
showed that even before the crackdown there were less deaths in
previous years
qualitative analysis
along with numbers, assessing the non-numerical data as well like
interviews and surveys
ethical problems with evaluation research
informed consent - participants believe that health services will be
cut off if they do not participate
maintaining confidentiality - it is sometimes necessary for the
researcher to know who the participant is
perceived injustice - some people object to being in a control
group because they believe that they are missing out on some kind of
beneficial treatment
avoiding conflict with stakeholders - make sure that the research
from program evaluators will not interfere with stakeholders who have
a vested interest in the program
benefits of small N designs
better individual subject validity - not just the average as a
representation of a single individual
you lose info that would have been found on a closer investigation
(ex: learning in children overcoming "learning curve")
when may it be hard to find enough participants for your study
when it has a large N and you are studying something specific or rare
clinical psychology (diseases), counseling psychology, invasive
animal studies
goal of small N design
to show change in behavior as a result of treatment being applied
steps of small n design
1. operationally define behavior of interest
2. establish baseline level of responding over set period of time (=A)
3. implement treatment and record change in responding (=B)
withdrawal design
how tightly coupled are the behavior changes and the treatment?
if you stop the treatment will behavior go back to baseline or stay
the same?
how would this pattern affect your interpretation of the
effectiveness of treatment
withdrawal design = reversal design
simplest version is A-B-A design
could also use an A-B-A-B design to strengthen interpretation
further? - if scores during all treatment sessions are high and all
baseline times are opposite, this is a strong treatment
multiple baseline designs
establish baseline measures and implement treatments at different times
withdrawal designs aren't always possible - if you're trying to
teach new behavior then you may not be able to reverse it - if nothing
changes after treatment is removed, maybe it was a permanent treatment
ex: 1. baseline for same behavior in 2 or more individuals2.
baseline for 2 or more behaviors in same individual3. baseline
for same behavior in same individual but in different settings (home
vs school)
changing criterion designs
based on operant conditioning paradigm of shaping
when target behavior is too complex to acquire all at once - break
down into steps or increments, sequentially change criterion making it
more stringent until target behavior is learned
applied to health-related behaviors, sever developmental disabilities
applied behavior analysis
uses behavior or operant principles to solve real life behavioral problems
examples of multiple baseline designs
decreasing stuttering in 6-10 year olds, goal was to change same
behavior in 2 or more individuals - took baseline readings and
implemented treatments in each child at different times and observed
the outcome
changing football performance, changing 3 behaviors within single
participant - establish baseline for the players and observe
perfomance in each player after "public posting" for improvement
example of changing criterion design
changing lifestyle of obese children, goal was to increase stationary
bike use in obese and non-obese 11 y/o boys
recorded baseline with 8 sessions saying to "exercise as long
as you like"
1st criterion - reward after 15% increase in use2nd criterion
- reward after 30% increase in bike usewithdrawal phase -
reinforcement removed
both the obese and non-obese showed the same trends
case study research
an in-depth analysis of a single person or case, most often clinical
cases - want to describe the developmental causes and consequences of
an event
"the ultimate small N
classic example of case study in psychology
fineas gage - got part of his brain taken out by a rod and his
emotions completely changed from nice to mean afterwards. researchers
were able to study him because that already happened, they could never
impose that on someone
strengths/weaknesses of case study research
strengths - very detailed analysis not usually found in research, can
show what happens in extreme cases that can be casued over a lifetime
if not taken care of (head injuries in boxer), support for a theory,
suggest hypothesis for future research
weaknesses - limited generalizability because they are extreme
cases, subject recall/eyewitness testimony can be iffy
two types of observational research
naturalistic observation and participant observation
vary based on the level of involvement of the researcher with the participant
naturalistic observation
as pure as you can get, you are not changing the system at all
no researcher involvement with participants (ex: set up post and
observe passer-bys)
study of people in their natural environments
downsides/benefits to naturalistic observation
bad - can't impose or change anything, cant listen to people, don't
know what happens after the observation
good - for initial research to eventually lead to a hypothesis and
study on behavior
lab based "naturalistic" observations
where you are in a lab but it is made as "natural" as
possible (ex: make the lab into a person's living room)
downside - ecological validity - how well does this apply to the
real world?
upside - you can change the stimuli
ex: kids in a lab with other kids and toys
participant observation
researcher actually joins the group being observed
participation provides first hand insight into group dynamics
ex: psychologists join religious cult to learn about it
problems with observational research
absence of control - variability in degree of control, implications
for interpreting results
observer bias - researcher unintentionally acts in a way that
influences participants
ethics - invading privacy, no consent
problem with participant observation
because the researcher is actually joining the group, the group is
now a new group - they might act different because there is a
"newcomer" or "outsider"
ex: religious cult study - researchers acted weird and the cult
started to believe that the researchers were sent there by their aliens
problem with ethics and consent!
observer bias
experimenter unconsciously acts in a way that can make participants
behave in an "expected"way
preconceived ideas about what will happen color one's observations
ways to minimize observer bias
develop clear, precise operational definitions for behaviors
generate behavior checklists (coding for behavior)
train observers in identifying target behaviors
have 2 researchers observe and correlate outcome/see if the they match
use different sampling procedures to reduce the amount of data (time
sampling-observations at specific times, event sampling-observe only a
specific set of events)
subject reactivity in observational research
people's behavior changes when they know they are being observed
minimize this by: direct unobtrusive measures (2 way mirror,
hidden video)indirect unobtrusive measures - record events that
are assumed to result from behavior of interest (ex: study trash to
see eating habits)
when dont we need informed consent?
when behavior is studied in public
if people are not interfered with in any way
if confidentiality is maintained
survey research
a structured set of questions or statements given to a group of
people in order to measure their attitudes, beliefs, values or
tendencies to act
advantage of survey research
can collect a lot of data with minimal effort, but
must be sure that sample observed is representative of the population
you want to generalize your results to
probability sampling
used when the goal is to describe features about an identifiable
group of individuals
population = group of individuals
often not possible to study the entire population of
interestsample=subgroup of population of interest
sample needs to be unbiased and representative
random sample! each person has an equal chance of being selected
self-selection problem in probability sampling
people who respond to the survey or questionaire are the ones who
feel an extreme need to respond
ex: good or bad experience with customer service are more likely to respond
random sampling
stratified sampling - use when there's a systematic feature of the
population you want reflected in sample
ex: college campus with 80/20 girls/boys but want to ensure the same
gender distribution in sample - proportions of important subgroups in
the population are represented precisely in sample
cluster sampling
random selection of a cluster of peope all having some feature in common
use when population of interest is huge
ex: survey of on-campus living experiences - high rise building
(6/10, floors 3-8)
used by national polling organizations
face to face interviews
good - you know who you are talking to, can do a follow up
bad - social desirability bias (don't want to release all info and
sound weird), people actually have to take their time to go into see
the researcher
phone interviews
good - easy, don't have to be there, more people
bad - you don't exactly know who is on the other end
written surveys
open-ended vs closed-ended questions - sometimes people don't want to
have to think up a response and write it out, they want to fill in a bubble
likert scale to measure degree of agree/disagree
electronic surveys
good - access to anyone with a computer, once the program is made it
will run itself
bad - don't know who is on the other end of if they are who they say
they are, personal info is vulnerable, no follow up quesitons
tips for written and electronic survey compliance
make it simple, mostly closed-ended questions with optional open ended
start with interesting questions
make it look professional
designing a survey
rely mostly on closed-ended questions
use likert scale for agree/disagree, easier to quantify and make
conclusions, use same scale on all questions, reverse order some questions
use careful and not confusing wording - complete sentences, no
abbreviations or slang, dont phrase questions negatively (hard to process/understand)
issues to consider when interpreting the results of a survey
sampling biases - how well can results be generalized to the population?
response bias - social desirability bias, are people responding how
they really feel or how they think they should respond
jane goodall observational study
spent a lot of time observing the apes from varying distances and
they became so habituated to her that they acted completely normal
with her only 10 feet away
festinger study
researchers joined a religious cult and observed them to see how they
reacted, and coped (cognitive dissonance) after their theory of the
world ending was wrong