Experiments in social science Seeking causal explanations.
-
Upload
bryan-hudson -
Category
Documents
-
view
212 -
download
0
Transcript of Experiments in social science Seeking causal explanations.
Experiments in social science
Seeking causal explanations
Causality Telecommunications managers, like social
scientists, would like to be able to make causal statements--even if they only point to partial causality
“X causes Y” Incomplete or imperfect causality
• Multiple causality• Partial causality (increased likelihood)• Necessary and sufficient conditions
Three conditions for establishing a causal relationship between two concepts
1. Covariation
2. Time order
3. Elimination of alternative explanations
Experiments
The experiment is a method where the researcher manipulates one variable (independent variable) and observes its effect on another (dependent variable) under controlled conditions
Experiments Example: A researcher may expose a group
of students to a movie with one ending and a second group to the same movie with a different ending (both under laboratory conditions), then measure their emotional response to the movie
Features of the experiment
Independent variable Dependent variable Subjects Control
Independent variable
The independent variable is the ‘cause’ or ‘causal variable’ in the hypothesis to be tested
The researcher manipulates the independent variable and subsequently measures subjects on the dependent variable
A factor in an experiment is an independent variable whose levels are set by the experimenter • http://www.stat.yale.edu/Courses/1997-98/101/expdes.htm
Independent variable
The levels of the factor that are introduced into the experiment are called the ‘treatments’
If a group is measured on the dependent variable but is not exposed to a non-zero treatment, it is called a ‘control group’ • Some consider this a zero-level treatment,
others say it is not a treatment
Independent variable
The factor in an experiment must be represented by at least two treatments—experimental and control treatments or two experimental treatments
Stronger experiments include multiple treatment levels
Examples: • exposure to a video game v. non-exposure • exposure to different executions of a creative
idea • having one half of a class use a website as part
of the course and the other half not use it• Exposing one group of subjects to one hour of
rap music, another to two hours, another to three, and another to four hours
Dependent variable (effect) The outcome of interest in the study All groups are measured on the DV Represents the ‘effect’ Examples:
• liking for a show or a television personality• recall of information from a website • time spent at a website• purchase of cell phones • political activity
Subjects
People who are assigned to experimental conditions and measured on the dependent variable
They should be members of the target market/audience
They are often a ‘convenience sample’—especially students in lower-level psychology classes• Nonrandom sampling
Control
Any procedures used to see that the only thing that varies for the subject groups is the independent variable• The goal is to isolate impact of the independent
variable on the dependent variable
Forms of control
Control of the environment• Minimize distraction from noise, lights, action
other than exposure to the independent variable• Keep the environment the same across groups
Random assignment of subjects to treatments (randomization)• Trying to make subject groups equivalent in
terms of personalities, experiences, demographics
Forms of control (continued)
Identical presentation of treatment and measure of dependent variable among groups• Placebo• Timing
Statistical control• Statistically remove the influence of
demographics, prior experience, etc.• Requires measuring all variables you will use
as controls
Forms of control (cont’d)
Experimental Design• Blocking
Basic experimental design
R X O
R O
Goal
The ultimate goal of the research is to determine whether the independent variable ‘causes’ the dependent variable under specified conditions
This sounds simpler than it is
Strengths of the experimental method Strong claims to ‘internal’ validity Strongest ability to infer causality Relatively low cost Straightforward interpretation Scientific aura
Weaknesses of the experimental method Troubles with ‘external’ validity
• Artificial setting• Demand characteristics• Hawthorne effect• Forced exposure• Multiple influences controlled
• Non-representative samples• Measures divorced from concepts
• Kicking a Bobo doll