The Research Context Ethical Treatment of Participants

The Research Context
• Research is not done in a vacuum
– Build on prior findings
– Build on and contribute to theory
• Good design is crucial
– Statistics cannot overcome flawed design
• Ethical issues
Treatment of participants
Treatment of data
Treatment of manuscript – as author or reviewer
Comply with laws
Ethical Treatment of Participants
• Institutional Review Board (IRB) established
and operated as matter of Federal law
• Research ethics became a focus after WWII
– National Research Act became law July 12, 1974
– Belmont Report issued some 4 years later
• Became the basis for codifying ethical treatment of
human participants
– APA adopted a code of research ethics in 1982
and has amended it on occasion since then
• Belmont Report
– Respect for Persons
• people are autonomous agents and those with
diminished autonomy require extra protection
– Beneficence
• Maximize benefit and minimize harm to individuals
– Justice
• Except when it can be specifically justified, research
participation and benefits should be fairly distributed
• Operationalizing these principles
Obtain informed consent
Assess risks/benefits
Select participants fairly
Freedom to discontinue
Assure privacy and confidentiality
Types of Research
• Experimental
– Involves random assignment of individuals to
conditions (or to combinations of conditions)
– For example,
• Two group design: experimental vs. control
• Three group: group 1 vs. group 2 vs. group 3
• Factorial: Cross 3 levels of A with 3 levels of B
– Absolutely necessary for inferring cause-effect
Naturalistic observation
Group comparisons
Correlational studies
Combinations of the above
1 2 3 4 5 6
• Observational
1 2 3 4 5 6 7 8 9 10
• Does not involve random assignment
• Done outside or inside the lab
• Can never support cause-effect inferences
Types of Variables
• Experimental research
– Independent – manipulated by experimenter
– Dependent – variable(s) that is (are) measured or
• Observational research
– Sometimes terms above are used
– Sometimes predictor and outcome (or criterion)
– Sometimes just X and Y
• Quantitative vs. qualitative
• Measured vs. counted
• Discrete vs. continuous
Additional Terms
• Population
– The complete set of scores of interest
• Can be finite or infinite
• Numerical characteristics are called parameters
• Sample
– A subset of the population
• Is finite
• Numerical characteristics are called statistics
• Descriptive statistics
– Characterize properties of a sample
• Inferential statistics
– Draw probabilistic inferences about populations
Sample Description to Population Inference
• If every element in the population is observed,
no inferences are required
– Data can simply be summarized and displayed
• Otherwise
– Gather, display, and summarize data with respect
to a randomly selected sample
– Assume a probability model of the population
– Use this model along with the sample statistics to
draw probabilistic conclusions about the
• Simple random sampling is the ideal for most
(but not all) research
– Each observation (or set of N observations) is
equally likely to be drawn
– This is virtually impossible in practice
• Depends on definition of population
• May be prohibitively expensive in time or money
• Some types of sampling schemes
– Convenience
– Stratified (sometimes used to intentionally oversample specific groups)
Within-Subject, Between-Subject, and
Mixed Designs
• Within-subject
– Also called repeated measures
– Each participant receives each treatment
– Advantages and disadvantages
• Each participant is his/her own control
• Observations are not independent (often control order)
• Between-subject
– Participants randomly assigned to conditions
– Advantages and disadvantages
• Often less sensitive than within-S design
• Independence of observations is assured
• Mixed designs (e.g., time series)
Measurement Scales
• Measurement is the assignment of numbers to
objects such that certain numerical relations
represent certain empirical relations within
the construct being scaled.
– An important often misunderstood issue
• The type of scale is determined by the specific
empirical relations that are represented
– Affects the meaningful statements that can be
made about the underlying construct
• Consider temperature to measure heat or weight to
measure mass
– Our interested is in constructs, not measurements
• Nominal scales
– Classify observations into mutually exclusive
and exhaustive categories (e.g., male/female)
– Any 1:1 transformation is allowable
• Ordinal scales
– Rank order observations (e.g., scratch scale
of hardness; degree of learning by number
of errors)
– Any ordinal transformation is allowable
• Interval scales
– Information about ratios of differences is
meaningful (rare to have this in psychology – IQ is
not an interval-scale measure of intelligence)
– Linear transforms, y’=a+by (a any constant, b>0),
are allowable
• Ratio scales
– 0 truly means the absence of the variable
– Information about ratios is meaningful (even more
rare in psychology – time is not a ratio-scale
measure of learning)
– Ratio transforms, y’=by (b>0), are allowable