Psych 325 1st EditionExam 4 Study Guide: Chapters 7-9Chapter 8: Quasi-Experimental Design - Why do quasi-experiments.- Some questions can’t be studied with experimentso Can’t manipulate some variableso Questions are still interesting- Using best of experimental and non-experimental techniqueso Best internal validity possibleo Also externally valid- Two approaches to finding comparison groups. Advantages and disadvantages of each. - Locate similar groupo Not exposed to treatmento Useful comparisono Problems: Confound: groups may differ Difficult to find- Locate similar peopleo Matching individualso Better, more valid comparisono Problems: Identifying variables to match May introduce a confound Tough to do- One-group pretest-posttest design. - Comparing with self- History, maturation confounds- Posttest-only with non-equivalent groups design. - Compare to other groups- Confound of group differences- Pretest-posttest with non-equivalent groups design. - Uses all of the above- Helps rule out above confounds- Natural treatment design. - Manipulation occurs naturallyo Uncontrolled event (i.e., disaster, government policy change)o Post-event observation- Improves with:o Pre-event observationo Good comparison group- Internal validity problemso Did event cause change?o Are changes due to maturation or history- One of the most popular QE designs (see pages 249-252)- Person by treatment design. - Partly experimental (> = true IV; > = measured IV) o At least one true IV and at least one measured IVo “Person variable interactions” Examine interaction effects How do different people respond to your manipulation?- Selecting extreme groups. - Can select extreme groups Often study people very high and very low on measured variable Maximizes differences Be careful of regression to mean- Natural groups with experimental treatment design. - Experiment conducted with existing groupso Naturally occurring groupso Can’t be changed- Groups get different treatment- Major person confounds: how can you argue causality?o Establish similarity on important variables to studyo Measure potential confoundso Assess pre-test treatment differenceso See pages 253-4- Patched-up design.- Researcher adds (patches) conditions to quasi-designo Establishes size of effecto Rules out confounds- Adding control conditionso Can be done after original data has been collectedo Each patch can improve argument for causalityChapter 7 Multiple Choice Questions - Independent variable. - Hypothesized cause- Experimental group and control group are identical, except for exposure to the IV- Independent of other potential causes- Dependent variable. o Predicted effecto Any observed differences between groups is due to IVo Depends on the influence of the IV- Manipulation. - Changes made by experimenter- Control group. - No manipulation. Used for comparison to manipulation- Random Assignment (vs. matching). Difference between random assignment and random sampling.- Random assignment is besto Participants have an equal chance to be in any conditiono Equates on every dimension Even ones you don’t knowo Not a perfect technique Problem with small samples Concerns?- Measure relevant variables- Replicate the study- Random assignment vs. random samplingo Both maximizes likelihood that two groups will be similar- Strengths of experimental designs. - Eliminates individual differenceso Differences between groups can’t explain differenceso Random assignment to condition accomplishes this- Eliminates other confoundso Laboratory environment helps eliminate confoundso It’s not perfect, though- Placebo effect. - Procedural and operational confounds in experiments. - Proceduralo The procedure manipulates more than one thingo Threatens internal validity- Operationalo The IV isn’t what you think it iso You’ll have trouble interpreting any significant findingso Not internal validity problem- Interaction effect. - Effect of one variable differs depending on levels of another- Tells you when or under what conditions causation occurs- Can identify boundary conditionso When a theory isn’t trueo Qualification approach- Noise. - Extra variables that affect all conditions equally- Artificiality. External validity. - Labs eliminate noise and confounds, but at a cost- Is the unusual environments of the lab an artifact that limits the generalizability of findings?o Observed effects may only occur in labso May not apply to the real world- Mundane realism. - Degree to which your setting looks like the real world- Doesn’t guarantee usefulness- Experimental realism. Ways of enhancing it. - Are you studying a genuine psychological response- Subjects need to be engaged for the study to matter at all- Without it, mundane realism doesn’t matter- There are no explicit ruleso Every situation is differento Use trial and erroro Read prior literatureo Build participant interest Use a good cover story Perhaps use deceptiono Guarantee construct validity Use manipulation checks Pilot test your study- Manipulation check.Chapter 9 Multiple Choice Questions - Experimental design. Three things you need to do to design a study well. - Think things througho Think through every design issue, critically evaluate everything and look for potential problems- Anticipate your resultso Success? What are other interpretations?o Failure? What does it mean?o Adjust design accordingly- Be flexible in the design you chooseo Rules aren’t set in stoneo Every research question is unique- Control groups and the difficulty of creating them. - Control participants should do exactly what experimental subjects doo Same time commitment, info/instructions, activities/similar activities, and interact with same people- Difficultieso Because some IV’s are complex, you may not be able to find similar procedureso If you fail to achieve similarity, serious threat to ability to interpret the findings- Statistical power. - The level of confidence that a significant effect is really significant- Between subjects design: advantages and disadvantages. - Each condition has a different group of people- Advantage: no carry-over effects- Disadvantage: low statistical powero Less efficient than within subjectso Group differences produce noise- Problem: if random assignment fails, you may introduce a confound- How to address these problems: use large sample sizes and measure any subject variable that could be a confound- One-way designs.
View Full Document