Overview of Exam 2 A 40 to 45 Multiple Choice Questions B Should take roughly 50 minutes to complete but you have the full 1hr15min and I ll stay longer too time is not an issue C Roughly 65 about research design concepts and scenarios 35 about statistics interpreting SPSS output Stats questions will be very similar to what was asked of you in lab D Questions will be based primarily on information in the lecture slides in class lecture For studying purposes the readings are probably best used as a reference when something from your notes or the slides is unclear If a topic or term is not on the outline below it will NOT be on the exam E F Types of Questions will include a Questions about the definition of key concepts b c Identifying internal validity threats for a given experimental scenario Identifying the type of design i e between subjects within subjects 2x2 2x3 longitudinal etc for a given experimental scenario Identifying the appropriate statistical test i e t test anova for a given experimental scenario d e Choosing the appropriate interpretation of SPSS output using p values graphs and means tables f Reasoning about the different elements of null hypothesis testing alpha random error type 1 type 2 errors p values etc g Each of our design choices has trade offs Some questions will assess your understanding of the trade off issues confronting each type of experimental design Topic Outline for Exam 2 Intro to Experimentation Ch 9 10 Lecture Thurs 6 13 Simple logic of Experimentation attempts to create a situation in which two groups are perfectly equal at baseline Then introduce a single treatment a single change a single manipulation to one of the groups and take a measurement If the two groups differ on the measurement afterward we can conclude that the manipulation cause this change Difficulties when conducting experiments with humans Groups are NEVER perfectly equal at baseline and It is difficult to manipulate ONLY one thing at a time Independent Variable The variable we manipulate the one that potentially causes the DV to Dependent Variable The variable we measure the one that is potentially affected by the IV change Experimental Group Control Group Operationalize Manipulate the IV Researchers always need to make trade offs between practicality and experimental realism when operationalizing Independent Variables Internal Validity Refers to how confidently one can conclude that the group differences were produced solely by the independent variable and not extraneous ones Internal Validity is concerned with the question Was it really the treatment that caused the difference between groups Systematic Error a k a Confounds Groups differ on a dimension that makes them unequal at baseline that is unrelated to the IV AND that may influence DV scores Examples include having two conditions always run in two separate rooms or by two separate experimenters or at two times of day Selection bias Our two groups may be predetermined by a characteristic other than our IV i e personalities of those who sit up front vs sit in back or rushing to class heart rate Random Assignment can negate this validity threat by giving each participant an equal chance of being in any condition flipping a coin Maturation effects Sometimes participants change over time and it has nothing to do with your manipulation boredom and fatigue are possible maturation effects Running between group studies and Counterbalancing an experiment can negate this validity threat Example used in class involved the task in different lightings History effects Sometimes participants change over time and it has nothing to do with your manipulation This can be due to historical circumstances that systematically affect only one group history threats Control groups are needed in this study for comparison preferably one that experiences the same shared history Attrition Losing subjects over time in a way that may be systematically related to the IV or DV Most problematic in longitudinal research as well as experiments where participants must return later Prevention careful planning getting permission to search public databases for contact info statistical techniques called imputation Instrumentation Experimenter Effects If the testing apparatus changes over the course of the experiment it introduces error unrelated to your DV The Experimenter Effects threat is why the gold standard of experiments is a Randomized Double Blind Placebo control experiment Double blind means even the experimenter doesn t know what condition the participant is in Prevention scripted protocols awareness of issue track who ran who Random Error Aspects of the testing environment that affect both groups equally and that create noise in our data Examples include flickering light temperature of room how hungry a participant may be etc These factors may cause an experimenter to miss detecting an effect but is less of a problem if both groups are equally subjected to it Random error is minimized statistically with null hypothesis testing and by running more controlled experiments Between Groups All participants experience only one of the experimental conditions Example placebo group imitation of a real drug vs drug group Within Groups Designs Repeated Measures All participants experience all experimental conditions Example Everyone is measured under the placebo drug drug and high drug dosages These designs are less prone to random error because participants act as their own baseline control Null Hypothesis Testing T Tests Supplemental Reading Lectures Tues 6 18 Thurs 6 20 The Logic of Null Hypothesis Testing Indirect Proof Random chance random error Dr Null s Argument two random groups won t be perfectly equal at baseline The Effect of the IV The drug made groups more different than we would expect just by chance alone even taking into account the random error we can t control for Alpha Level of significance generally a 05 P values The value that is measured and used to determine whether an experiment should either Reject or Fail to Reject the null hypothesis With Large N s even very small differences may be statistically significant However the effect may not be practically significant Aka its too small to have much real world consequence Type 1 error or false positive Rejecting the null hypothesis when in fact it is true This type of error is much more worrisome Its like saying a drug works when really it doesn t Reduce the risk of error by Replication
View Full Document
Unlocking...