Research Design Flashcards
Types of Design
ANOVAs
Alpha level
The chance of making a type I error
Type 1 and 2 Errors
Type I Error: Detect a significant difference when there is not one, chances of making a type I error = the alpha level
Type II Error: Fail to detect a significant difference when there is one
Measures of Central Tendency
Mean: average
Median: response at the center of the distribution
Mode: most frequent response
Effect Size
- Cohen’s D
- Calculation: d = (M1 – M2) / pooled SD
- Measures whether the independent variable has an effect on the dependent variable
- Other stats related to effect size: Correlations, Standardized Mean Differences, Odds ratio
Reliability
See if test measures accurately
Cronbach’s alpha (Reliability coefficient - internal consistency)
Validity
See if test measures what it is supposed to measure
Double-Blind Study
- Multiple participant groups assigned
- Neither researcher nor participants know which group they are in
Distribution Curve
Standard Error of Estimate and Measurement
Percentile Ranks
Sample Size
- Sample needs to represent the larger population
- Sample needs to be large enough to be high-powered but not so large that it is wasting resources that are not necessary to find a meaningful difference
Statistical Power
- The probability of avoiding a type II error (failure to detect a meaningful difference in the sample)
- Typically set at 80%
- Too much power = tests are too sensitive to true effects
Instrumentation
- Threat to internal validity
- E.g., you are relying on an expert rater’s ratings for your dependent variable and find that the rater’s accuracy changes over time
Difference between a true experiment and a quasi experiment
ability to randomly assign participants to treatment and comparison conditions
Testing threat to internal validity
- E.g., There are differences on the post-intervention test between the participants who did and did not take the pretest
Independent Variable
Variable that you manipulate to see if a change occurs
Dependent Variable
Variable that you measure to see if it changes in response to the manipulation of the independent variable
Efficacy
- Internal validity
- Assesses a causal link between the treatment and a change in the clinical condition
Effectiveness
- External Validity
Reliable Clinical Change Index (RCI)
- Calculates the effect size of change for an individual client over the course of treatment
- Demonstrates that after the intervention, the client is functioning in the non-clinical range
Results from a single-subject research study are evaluated using
Visual inspection
ABA Design
Assesses behavior changes when the treatment is introduced and how it changes again when the treatment is withdrawn
Compliance (in randomized control trials)
The degree to which study subjects adhere to an assigned treatment protocol
Michael Lambert, the author of the Outcome Questionnaire – 45 (OQ45) advocates for what?
Tracking the therapy progress especially for clients who are not progressing or who are deteriorating
ITT (intention to treat) group
- Important because it reduces the risk of sample bias that might result from attrition
- Still take into account data from participants who did not complete the study
Which of the following research designs is considered the most trustworthy, according to SAMSHA?
Meta-analysis
Stratified random sampling
- Creates a probabilistic sample
Sampling Frame
All of the units or cases that a researcher has accessible to study
Random Sampling
Sample members are selected so each member of the population has an equal probability of being selected
“Positivist approach” said the scientific method could be applied to
Observable and repeatable behaviors
Construct
The theoretical idea or concept that is being described or analyzed in the process of research
Construct Validity
Is your measure assessing the construct of interest?
Do the items capture the construct?
Convergent Validity
Two test that are supposed to be related are in fact related
Discriminant Validity
Two tests that are not supposed to be related are in fact unrelated