COMM3710 - Fall 2011 - Section 001 - Lecture Notes - 10/27


*Please enter 3 lines between the last entry and put in your username using heading 3 then type in your notes.*



riosjosh

Research shows that strangers are better at detecting lies than people with intimate relationships. This is because those not invested are able to use their resources for lie detection instead of relational establishing.

Threats to Internal Validity (what can screw up an experiment)
1. Researcher related threats
Experimenter effects – researchers influence results
Observer bias – observers need to be blind to the study as well
Researcher attribute effect – depending on the question who the person asking the question may influence the answer
2. Participant related threats
Hawthorne effect - people change their outcomes when they know they are being watched
Testing effect – if you measure someone twice then there is the possibility that the first will influence the second
Maturation – people change even over the course of a short survey or experiment
Experimental mortality - subjects die if not literally then in the experiment (i.e. they move away or stop participating)
Selection bias – stopped if you randomly select; (i.e. one study showed that people who watched bill ni the science guy did better later on in science and math, but the thing is that they were more interested in those to begin with)
Inter-subject bias – subjects talking to each other about the experiment before, after, and during
Compensatory rivalry – the control group (if they know they are) might try harder
Demoralization – the control group (if they know they are) might give up
3. Procedure related threats
History – Event that occur outside of the experiment
Instrumentation – reliability of your measuring instruments and procedures
Treatment confound – you have to make sure the independent variable is causing the change and not something else
Statistical regression – the mean is the mean is the mean; eventually things will focus back to the mean maybe not because of your treatment
Compensation – you give the control group a little something more (in the design)

Pre-experimental design – these can’t show causation (No pretest; no control group; no random selection)
One-shot case study: treatment-observation; you don’t know the before
One-group pretest-posttest design: observation-treatment-observation
Static-group comparison: treatment-observation 2nd group observation only

Quasi-experimental design
• May Lack control group
• May lack equivalent conditions
• May lack random assignment
• Often in field settings enabling better external validity
• Often use time series designs

Time-series design - Things can change