Exam 2 Flashcards
Design classification degree of experimental control
- in a true experimental design, subjects are RANDOMLY assigned to at least 2 COMPARISON groups
- experiment enables control over most threats to INTERNAL VALIDITY and provides the strongest evidence for CAUSAL relationships
- randomized control trial (RCT) is the gold standard of true experimental design
Are quasi-experimental designs true experiments?
- NO
- because they lack randomization and comparison groups
Types of group assignment for design classifications
- completely randomized design
- randomized block design
- repeated-measures design
Completely randomized design group assignment
- between subject design
- subjects assigned to groups based on a randomization process
Randomized block design group assignment
- subjects classified according to an attribute (blocking variable) (i.e. males vs females)
- then randomized to treatment groups (i.e. males get control and random group as well as females)
Repeated-measures design group assignment
- within-subjects design
- subjects act as own control
Variation with number of independent variables/factors
- single-factor designs have one independent variable
- multi-factor designs have 2+ independent variables
Single-factor design (one-way design) for independent groups
- 1 independent variable is investigated
- 1 or more dependent variables
Pretest-Posttest control groups design
- RCT with 2 groups based on random assignment
- independent groups = treatment arms
- testing pre- and post-treatment
- changes in experimental group are attributable to the treatment
- establishes cause-and-effect relationship
2-group pretest-posttest design
- comparison group receives a second form of the intervention
- 2 experimental groups formed by random assignment
- control group is not feasible or ethical
- compares new treatment with standard care
- can quantify the difference between pre and post by “delta” or the change
For GroupXTime interaction
- looks at if there is any change between the time (pre vs post) and comparing the groups’ changes to each other
- can also use a 2-way mixed design
- main effects: groups, time
- interaction: groupsXtime
Multi-group pretest-posttest control group design
- multiple intervention groups
- includes a control group
- conclude that treatment 1 is better than treatment 2 or vice versa AND that it is or is not better than no treatment
Internal validity with pretest and post-test designs
- strong internal validity
- initial EQUIVALENCE of groups can be established by pretest scores (important for inferring causality)
- SELECTION BIAS controlled because of random assignments
- HISTORY, MATURATION, TESTING, INSTRUMENTATION EFFECTS SHOULD AFFECT ALL GROUPS EQUALLY
Analysis of pretest-posttest designs
- often analyzed using CHANGE scores (diff between posttest and pretest)
- also can use analysis of covariance (ANCOVA) to compare posttest scores (using pretest scores as covariates)
Posttest only control group design
- same as pretest-posttest control group design, EXCEPT NO PRE-TEST
- used when dependent variables can only be assessed following treatment (i.e. length of stay in hospital)
- used when pretest is impractical or detrimental
- is an experimental design involving randomization and comparison groups (STRONG INTERNAL VALIDITY)
- assumes groups are equivalent prior to treatment (works best with large samples to increase probability of equivalency)
Multi-factor design for independent groups
- single factor designs have 1 independent variable (with 1+ levels), and do not account for interactions of severable variables
- multi-factor designs have 2+ independent variables
Factorial Design
- incorporates 2+ independent variables, with subjects randomly assigned to various combinations of levels of the two variables
- two-way (two-factor) design has 2 independent variables
- three-way (three-factor) design has 3 independent variables
Repeated measures Design
- up to now considered 2 independent GROUPS
- experimental and control groups created by RANDOM ASSIGNMENT and by BLOCKING
- can also use repeated measures design where one group of subjects is tested under ALL CONDITIONS, each subject acting as their OWN CONTROL (aka within-subject desgin)
Advantage of repeated measures design
- subject differences are controlled
- differences between experimental and control groups are nullified because no groups used
- physiological and other factors remain CONSTANT throughout experiment
- subjects acting as their own controls provides most equivalent “Comparison group” possible
Disadvantages of repeated measures designs
- LEARNING/PRACTICE effects when one person repeats measurements over and over
- CARRYOVER effects when exposed to multiple treatment conditions (must allow enough time for dissipation of previous effects)
- may NOT be TRUE EXPERIMENTS because NO RANDOMIZED COMPARISON GROUPS
- however, if they incorporate randomization of the order of repeated treatments/interventions then can be considered experiment
Single-factor designs for repeated measures
- one-way repeated measures design
- one group of subjects is exposed to all levels of one independent variable
- has element of looking like an experiment because randomized order of who gets what experiment in which order
Solution to problem of order effects
- randomize order of conditions/interventions for each subject so there is no bias in choosing order of testing
two-way design with 2 REPEATED MEASURES for multi-factor designs
- 2 repeated measures (=2 independent variables….i.e. type of lift and orthosis)
- each person exposed to 4 test conditions (2-way design…2X2 design)
Mixed Design for multi-factor repeated measures
- 2 independent variables (i.e. exercise is IND factor (experimental and contorl), and time is REPEATED factor (3 time periods during tests))
- 2 way design or 2X3 design
Multi-factor designs
- two-way design with 2 repeated measures
- mixed design
Group variable
- independent factor/variable
- this is because has 2 levels independently
Time variable
- is a repeated factor
- because measures at time 1, time 2, time 3, etc
Two-way factorial design
- incorporates TWO INDEPENDENT VARIABLES
- effect of INTENSITY (vigorous/moderate) on exercise behavior
- effect of LOCATION (home/community center) on exercise behavior
- 2X2 design means (2 independent variables and 2 levels of each independent variable = 4 groups total)
Main Effects of a twoo-way factorial design
- is there an effect of moderate versus vigorus exercise
- is there an effect of exercising at home or in community?
- this examines the MAIN EFFECT of each independent variable
Interactions of Two-way factorial design
- can examine INTERACTION EFFECTS between 2 independent variables
- effect of 1 variable varies at different LEVELS of the second variable
- i.e. maybe moderate exercise is more effective in changing exercise behavior but only when performed at a community center
Randomized Block Design
- used when there is a concern that an extraneous factor such as GENDER might INFLUENCE DIFFERENCES BETWEEN GROUPS
- build the variable into the design as an independent variable
Quai-experimental designs
- similar to experimental designs but lack random assignment, comparison group, or both
- may involve non-equivalent groups
- may be a reasonable alternative to RCT
- conclusions drawn must take into account biases of the sample
One-group designs pretest-posttest
- effect of treatment is determined by change in pre and post scores
- pretest –> intervention –> posttest
- vulnerable to threats to internal validity because no control group (i.e. history, maturation, testing)
One-way repeated measure design over time
- effect of treatment over time
- pretest –> intervention –> postest 1–> postest 2
- no control group so internal validity threatened
Mulit-group design pretest-posttest
- non-equivalent pretest-posttest control group design
- similar to pretest-posttest experimental design EXCEPT subjects not assigned to groups randomly (i.e. volunteers self-select groups)
- EXP: pretest-intervention-posttest
- CONTROL: pretest-nointervention-posttest
Multi-group design posttest only control group design
- static group comparison
- EXP: intervention–postest
- CON: no intervention–postest
- NOTICE NO PRETEST
Single subject designs
- draw conclusion on treatment effects based on 1 patient’s response
- controlled experimental approach
- independent variable is treatment
- dependent variable is target behavior (outcome)
- also called (N of 1 study, or time series designs)
Structure of a single subject study
- Repeated Measurements: Each session; observe trends
At least 2 testing phases: Baseline and Intervention - Target behavior is measured across both phases on multiple occasions
- Baseline phase: state of target behavior over time in the absence of treatment (control conditions)
How do single subject designs differ from traditional experimental designs?
-Multiple assessments in baseline and intervention phases
Ethical issues regarding baseline conditions
- withholding treatment
When treatment starts, any change from baseline to intervention phase is attributed to what?
- the intervention
Baseline data
- comparison for evaluating potential cause and effect relationship between intervention and target behavior
- Baseline period = A
- Intervention period = B
- A-B design
Baseline characteristics
- 2 baseline data characteristics are important for interpreting clinical outcomes
- Stability (consistency of response over time): stable or variable
- Trend: accelerating or decelerating
Length of phases
- best to have equal phase length
- often 1 week per phase (take daily measurements, minimum of 3-4 per phase)
- greater number of data points easier it will be to identify trends
- often measures can be taken more frequently than daily if behavior changes rapidly
- More than a single session
Target Behaviors
- Choose clinically relevant outcomes measures for a particular patient
- i.e. Strength, ROM, Gait speed, Balance measures, Pain
Limitations of A-B design
- Experiments can control for threats to internal validity
- To do this in the A-B single subject design is more challenging
- Other treatments/events (history effects)
- What other evidence can we include to strengthen design control ?
- to increase confidence that treatment caused the changes in target behavior
Additional control for A-B design
- Replication of effects
- Repeat phases
- withdrawal designs—treatment: no treatment
- Withdrawing and reinstating baseline and treatment conditions
- Withdraw intervention and show that target behavior occurs only in presence of treatment
- 2nd baseline period (A-B-A design)
- Could also include a 2nd intervention phase (A-B-A-B design—see over)
Visual Data Analysis
- Level (last data point of a phase to first data point of next phase)
- Trend (direction of change in a phase)
- Accelerating or decelerating
- Slope of a trend (rate of change in the data)
Single Subject Data Generalization
- Single subject research can provide data for clinical decision making
- Not enough to show effect during intervention period on a single patient
- Must also be able to show changes in the target behavior will occur in other individuals
- Generalization: external validity for the single case
- Assume treatment will be effective in others with similar characteristics
Observational Designs
- no manipulation of variables as in experimental designs
- exploratory or descriptive