correlational research
establishes whether naturally occurring variables are statistically related
Correlational research characteristics
�No attempt to explain the relationship between the variables �No attempt to manipulate variables �No attempt to control threats to internal validity �Cannot establish a cause and effect relationship
Conducting Correlational Research
1. Researcher measures 2 or more variables
2.Research identifies the direction, form and strength of the relationship
Scatter plot or Correlation coefficient calculated to conduct hypothesis test
Direction of Relationship: Negative
�Two variables tend to move in opposite directions �Higher scores on X are associated with lower scores on Y �Lower scores on X are associated with higher scores on Y �Envision two people on a see-saw
Direction of Relationship: Positive
together �Higher scores on X are associated with higher scores on Y �Lower scores on X are associated with lower scores on Y �Envision two people in an elevator
correlational strength of relationship
-1 0 1
Anything farthest from zero on either side is stronger.
Is the relationship between two variables weak? Moderate? Strong?
Weak .10-.29
Moderate .30-.49
Strong >.50
third variable problem
a third unidentified variable may be controlling the two variables and producing the observed relation
directionality problem
correlational strategy cannot determine which variable is influencing the other
simple linear regression (regression analysis)
scores on X can be used to predict scores on Y assuming a meaningful relationship (r) has been established between X and Y in past research
Linear Regression
E.g., Scores on a job interview (X) can be used to predict job performance (Y) �X is the predictor; Y is the criterion �Interview scores plugged into regression equation and hiring decisions made based on results �This is an illustration of criterion vali
Multiple Regression
�Multiple predictors are used to predict a criterion measure �Strive for as little overlap as possible between predictors (i.e., want to account for unique variance in criterion)
How do we rule out all plausible third variables (confounds) using correlational research designs?
We can't... only the control afforded by rigorous experimentation provides strong tests of causation.
Strategies to Reduce Causal Ambiguity
1. Statistical approaches � Measure and statistically control for (i.e., partial out) a third variable
2.Research design approaches � When possible, conduct longitudinal studies
3 criteria that need to be met to make inferences about cause and effect
1. Covariation of X and Y. As X changes, Y changes.
2. Temporal order. Changes in X occur before changes in Y.
3. Absence of plausible alternative causes.
benefits of correlational research
test validation, venturing where experiments cannot tread, prediction in daily life
range restriction
occurs when the range of scores obtained for a variable has been artificially limited in some way; can lead to erroneous conclusions about strength, direction, and nature of relation between variables
Interpretations of correlation coefficients
�Statistical Significance of a Relationship �Probability of obtaining a specific value of r given a true null hypothesis. �Directly Affected by sample size �Weak correlation coefficient can be significant with a large sample size
Spearman's rank-order correlational coefficient
�Spearman's rho �One or both variables measured on ordinal scale
�Pearson product-moment correlation coefficient
�Pearson's r �Variables measured on interval or ratio scale
reducing bidirectionally problem
researchers conduct longitudinal research or statistically reduce influence of third variables
correlation
statistical association between variables
correlational research
involves examining potential associations between naturally curing variables by measuring those variables and determining whether they are statistically related
positive correlation
means that higher scores or levels of one variable tend to be associate with higher scores or levels of another variable
negative correlation
means that higher scores or levels of one variable tend to be associated with lower scores of levels of another variable
scatterplot
graph in which data points portray the intersection of X and Y values
bidirectionality problem
ambiguity about whether X has caused Y or Y has caused X.
key properties of a correlational coefficient
When it is computed, the way in which the measurement scales have been coded affects whether the statistical analysis yields a plus or minus sign for that coefficient.
The way a researcher conceptualizes a variable affects whether a correlation emerges as
partial correlation
correlation between X and Y is computed with statistically controlling for their individual correlations with a third variable, Z.
prospective design
variable X is measured at an earlier point in time than Y
cross sectional research design
each person participates on one occasion, and all variables are measured at that time.
cross lagged panel design
1. measure X and Y at Time 1
2. measured X and Y again at time 2
3. examine patterns of correlation
criterion variable
the variable we are trying to estimate or predict
predictor variable
a variable whose scores are used to estimate scores of a criterion variable
Correlation and predicting outcomes
Correlation enables prediction even when no causal relation between two variables is assumed. The stronger the correlation, the more accuracy we gain in predicting one from the other.
case study
in depth analysis of an individual, social unit, or event.
characteristics of case study
direct observation, talking with family or friends, physiological measures, intelligence tests, etc.
Advantages of case studies
offer unique window into nature of subject, provide insight into possible causes of behavior, provide evidence to support or contradict theories
disadvantages of case studies
difficulty of drawing clear causal conclusions, generalizability of findings, the potential for observer bias
observer bias
occurs when researchers have expectations or other predispositions that distort their observation
Well known case studies?
Genie,Kitty Genovese
case study vs single case design
Single subject designs focus purely on single individuals and investigate the effectiveness of an intervention. Case studies are non-experimental observations that are going to happen, or have happened due to natural or economic or personal causes.
observational research
encompasses different types of non experimental studies in which behavior is systematically watched and recorded
Naturalistic Observational Research
Behavior examined in "ecologically valid" (i.e., real life) conditions �But research design lacks control and some data may be overlooked �Reactivityoccurs when behavior is altered through the process of being observed �Must remain mindful of APA Ethics C
reactivity
occurs when behavior is altered through the process of being observed
Participant Observational Research
�Researchers embed themselves in the phenomena of interest�Disguised vs. undisguised distinction still applies -participants may not know researchers are among them
ethnographic approaches
are qualitative and incorporate interviews to develop a narrative of the research topic
Structured Observational Research
Researchers "tweak" the research setting, influencing what happens when �Used when the behavior of interest is a rare event and/or it is unlikely to occur during a naturalistic observation�Affords more efficiency and control compared to other forms of obs
recording observation
�Narrative records -extensive description of behavior as it unfolds �Field notes -less comprehensive records of behavior�Behavioral coding systems -categorize behaviors into mutually exclusive categories
Sampling Behavior
focus on one person at a time �Scan sampling -Observe everyone for a short period of time at predetermined intervals �Situation sampling -Observe behavior across multiple settings �Time sampling -Conduct observations over representative set of time period
inter rater reliability
extent to which observers agree
Overcoming Observer Bias
�Well-developed coding system�Observer training�Blind observation �Verify reliability of observer practices
archival research
use of data recorded in the past by other individuals for other purposes
disguised observation
an extreme form of unobtrusive observation
disguised vs undisguised research
depends on whether the subject knows they are being observed
ethnography
qualitative research approach that often combines participant observation with interviews to gain an integrative description of social groups
behavioral coding system
involve classifying participants' responses into mutually exclusive categories
observer ranking and rating scales
are used to evaluate participants behavior or other characteristics
diary
asks participants to record their behaviors or experiences for defined periods of time or whenever certain events take place
focal sampling
is used to select a particular member who will be observed at any given time
scan sampling
at preselected items the observer rapidly scans each member of a group so that the entire group is observed within a relatively short period
situation sampling
used to establish diverse settings in which behavior is observed
time sampling
used to select a representative set of time periods during which observation will occur
blind observation
observers should be kept unaware of all hypotheses being tested and any key information about participants that relates to those hypotheses
habituation
decrease in the strength of a response, over time, to a repeated stimulus
unobtrusive measure
assesses behavior without making people aware the behavior is being measured
physical trace measure
unobtrusively examines traces of behavior that people create or leave behind
archival records
previously existing documents or other data that were produced independently of the current research
sample size
larger the sample, the more likely it is to represent the population
probability sampling
�Every member of the population has chance of being sampled �Probability of selection can be specified
nonprobability sampling
Probability sampling conditions do not apply
simple random sampling
�Build a sampling frame containing all population members
stratified random sampling
�Sampling frame divided into groups (based on demographic characteristics) �Random sampling applied to each group
cluster sampling
Units (e.g., schools) containing population members are identified �Essentially, this step creates the sampling frame �These "clusters" are then randomly sampled �May not represent the entire population
convenience sampling
�"Grab whomever you can" �Likely to generate a nonrepresentative sample
quota sampling
�Sample designed to mirror population characteristics (e.g., % of females) �Uses convenience sampling to create sample within each quota group (e.g., males and females)
self selected samples
�Participants elect to participate (as opposed to being sought out by researcher) �A form of convenience sampling �Likely to generate a large sample size, but keep in mind that representativeness matters more than sample size!
purposive sampling
�Sample created in line with study goals (e.g., focus only on students in Top 10 graduate programs in research on the work habits of successful graduate students) �Two common strategies �Expert sampling �Snowball sampling - participants recruit others to
�Sampling variability
captures how sample characteristics fluctuate
sampling error
acknowledges that our population estimates vary depending on the sample
margin of sampling error
a range of values within which the true population value falls
confidence level
�Keeping in mind that we can never be 100% certain in our results, we also report confidence levels (typically 95%)
Six steps in developing a questionnaire
1. Research goals and list specific topics
2.Identify variables of interest within each topic.
3.Consider practical limitations of the survey.
4.Develop questions, decide on order, and get feedback from mentors or colleagues.
5.Pretest your questionnaire.
closed ended questions
provide specific response options
open ended questions
allowing participants to answer in whatever form they choose
forced choice
closed ended question even if neither statement is really correct
rating scales
likert scales
putting the survey together
�Place open-ended questions before closed-ended questions �Move from more general to more specific questions �Place personally sensitive questions in middle or near the end
�Face-to-face (in-person) interviews
�Facilitate establishment of rapport �Enable standardized approach �Interviewer can clarify any participant confusion �But, they're pricey!
high response rate
ensures that the results are representative of the population.
nonresponse bias
occurs when participants who declined to participate would have responded differently than participants did �Introduces more error into population estimates
sample
subset of cases or observations from the population
sampling frame
list of names, phone numbers, addresses, or other units from which a sample will be selected
representative sample
reflects important characteristics of the population
non-representative sample
does not reflect important characteristics of the population
population
refers to all the cases or observations of interest to us
social desirability bias
tendency to respond in a way that a person feels is socially appropriate, rather than as he or she truly feels
survey
uses questionnaires and interviews to gather information
sugging
sell or attempt to sell a product under the guise of conducting market research.
frugging
fund-raising under the guise of research
pugging
politicking under the guise of research
limitations of surveys
generally not well suited for examining cause-effect relations; there are many ways to select samples; validity of survey depends on peoples' willingness to treat it seriously
law of large numbers
a principle of probability according to which the frequencies of events with the same likelihood of occurrence even out, given enough trials or instances. As the number of experiments increases, the actual ratio of outcomes will converge on the theoretica
factors that affect sample size
margin of error, confidence level, proportion