Docstoc

Experimental and Quasi Experimental Designs (PowerPoint)

Document Sample
Experimental and Quasi Experimental Designs (PowerPoint) Powered By Docstoc
					Experimental and Quasi-
Experimental Designs

       Chapters 9 & 10
Research Design
   It is the outline, plan, or strategy for
    the procedures you will use to address
    your research question.
Research Designs – What are
some limitations of these?
1.       One-group after design
          Treatment     Response measure


2.       One-group before-after design
          Response Measure 1
          Treatment              Compare
          Response Measure 2
Research Designs – What are
some limitation of these?
  Nonequivalent posttest-only design

     Cardiac Bypass &       Cardiac Revascularization
     Revascularization


     Response Measure          Response Measure
    Cognitive Functioning     Cognitive Functioning
    at 6-months follow-up     at 6-months follow-up
What are the requirements of
true research designs?
    Requirements of True
    Research Design
   Design is adequate to answer the research
    question (i.e., test the hypothesis).
   Control for extraneous variables
   Results are generalizable.
Why Pre-testing Your
Participants
   Increased sensitivity of the study
   Ceiling effect
   Initial position
   Initial comparability
   Evidence of change
What are the limitations of
this design?
          True Research Design


                  Randomized to:

            Anxiety              Usual Care
          Management


           Response              Response
            Measure               Measure
       Decisional Regret &   Decisional Regret &
             Distress             Distress
Factorial Design
   Two or more independent variables are
    studied in order to determine their
    independent and interactive effects on
    the dependent variable.
Interaction Effect
   The effect of one factor (independent
    variable) depends on the level of the
    other factor (other independent
    variable).
        Main Effects & Interactions
 High                              High




Low                        High                       High
                                  Low
        Lightness                         Lightness
Seasonal Affect Disorder

Non-Seasonal Affect Disorder
    Advantages of Factorial
    Designs
   More than one hypothesis can be tested.
   Potentially confounding variables can be
    built into the design as factors.
   Enables interaction effects to be tested.
     Within Participant After-Only
     Design
   Same research participants in all
    experimental treatment conditions.
   Repeated measures design.
   Overcomes concerns about creating
    equivalence between groups.
   Requires fewer participants.
   Most serious limitation is the confounding
    influence of a sequence effect.
Combining Within and
Between Participant Designs
   Factorial design based on a mixed
    model.
   Can include as many independent
    variables as is necessary.
      Combining Within and
      Between Participant Designs
                              Before - After Design

                                     Participants
                                      Selected


                                    Pre-treatment
                                      Measures

May want to match before
                                                              Repeated
randomizing                        Randomized to:
                                                              Measure
Between-Subject            Experimental         Control
Factor                        Group             Group


                           Post-treatment    Post-treatment
                             Measure           Measure
       Factorial Design
                           Does stress management work similarly in medicated and
                                        unmedicated hypertensives?

                                            Unmedicated hypertensives
                                                Randomized to:


                    Beta Blocker &     Beta Blocker, no        Placebo &          Placebo, no
                 Stress management   Stress management    Stress management   Stress management


                   Post measure        Post measure         Post measure        Post measure
                    at 6-months         at 6-months          at 6-months         at 6-months



• Main effect for beta blocker
• Main effect for stress management
• Interaction effect for drug and stress management
Main & Interaction Effects
      Drug & Stress Man.      Drug, No Stress Man.            Drug & Stress Man.      Drug, No Stress Man.
      Placebo & Stress Man.   Placebo, No Stress Man.         Placebo & Stress Man.   Placebo, No Stress Man.

112                                                     112
108                                                     108
104                                                     104
100                                                     100
 96                                                      96
 92                                                      92
 88                                                      88
 84                                                      84
 80                                                      80
             Pre                      Post                           Pre                      Post

      Main Effect                                                  Interaction Effect
     Selecting the Appropriate
     Design
   The design must be one that addresses
    your research question.
   What control techniques can and should
    you apply to help you arrive at an
    unambiguous answer.
   Between or within-in design or mixed
    model.
Quasi-Experimental Design
    Quasi-Experimental
   Does not meet all of the requirements
    necessary for controlling the influence of
    extraneous variables.
       Most common criteria not met is random
        assignment.
   While you cannot infer cause and effect,
    well designed quasi-experiments enable
    you to demonstrate that rival
    interpretations are implausible.
Non -Equivalent Control Group
Design: Typical Rival Hypotheses
   Increasing treatment effect I outcome
   A selection-maturation effect
Increasing Treatment Effect I
Outcome
           Experimental Group




              Control Group



Pre-test                        Post-test
Increasing Treatment and
Control Group Outcomes
   Both groups’ scores increase over time
    but one group changes to a greater
    extent than the other group.
       Effect could be due to a treatment effect or
        to a selection-maturation interaction.
                HADS Depression
     14

     12

     10
                                                       Time 1
      8
                                                       Time 2
      6                                                Time 3
      4

      2

      0
              Experimental            Control


ANOVA – time and time by group effect, depression
decreased in both groups but levelled off in control
Group at follow-up while continuing to decrease in experimental.
Increasing Treatment Effect II
Outcome
              Control Group




           Experimental Group


Pre-test                        Post-test
Cross-Over Effects


     Control Group



           Experimental Group

Pre-test                        Post-test
Time Series Analyses
   Useful when you cannot randomize
    participants and where it is possible to
    obtain a series of assessments of the
    dependent variable at pre-treatment
    and post-treatment.
Time Series Analysis
          Treatment
          Applied

                     
                
                   
                 
                 
               
        
     Classic Studies: Effect of Reduced
     Speeding on Traffic Accidents in
     Connecticut
   In 1955 there were a record number of
    traffic accidents ( n = 324) so the Governor
    (Abraham Ribicoff) introduced a law to
    reduce the speed limit.
   In 1956 there were 284 traffic accidents, a
    reduction of 12.3%.
   Governor concluded that his intervention
    worked but the effects could just as easily
    been due to regression to the mean.
     Classic Studies: Effect of Reduced
     Speeding on Traffic Accidents in
     Connecticut
   Campbell and Ross (1968) used interrupted
    time series design to test if the reduced
    trend in traffic accidents was plausible.
   Compared traffic accident trend in
    Connecticut with control States.
     Classic Studies: Effect of Reduced
     Speeding on Traffic Accidents in
     Connecticut
15

14         Control State
                                   
                                                      •
           
13                                                              
     •                    •                       
                                                                                               
                                                                                       
12                                                                  •
                                                                          •
11
            •                           •                                          •
10
                                Connecticut
                                                                                            •
                                                                                                    •
 9
     ‘51        ‘52           ‘53           ‘54           ‘55           ‘56       ‘57   ‘58         ‘59
Class Exercise
   Specify your research question
   Your scientific hypothesis
   Specify your design
   Break into groups of 5

				
DOCUMENT INFO