Chapter 8 Experimental Design Causality Review

Remember from Chapter 4 that in order to make a causal inference, you need to satisfy three requirements: 1. Covariation 2. Temporal order 3. Eliminate alternative explanations (3rd variable problem)

• Third variables which are not controlled for are called confounding variables, or simply confounds. Experimental Design Basic • As we already know, experiments address confounding variables through: – Experimental control – • Experiments establish temporal order by manipulating the independent variable and measuring its effects on the dependent variable. • Thus, the most basic will have only two variables, one IV and one DV, each one with 2 levels. Basic Experiments

Example: What is the effect of whining on the sympathy of a romantic partner?

Whining (yes/no) Sympathy (1-10)

To test this hypothesis in an experiment, there are three steps: 1. Assign participants to conditions 2. Conduct manipulation 3. Measure the dependent variable Step 1: Assigning Participants • Independent Groups (Between subjects) – Assign different participants to each experimental condition • Repeated Measures (Within groups) – Every participant is assigned to every condition of the independent variable Repeated Measures design

Advantages – Fewer participants needed – Extremely sensitive to effects of the independent variable • Why? Repeated Measures design

Disadvantages Order effects 1. Practice effect 2. Fatigue effect 3. effect Developmental Designs

• Developmental designs explore how individuals change as a function of age. – Cross-sectional – Longitudinal – Sequential Chapter 9: Conducting the Experiment Step 2: Conducting the Manipulation • There are 2 types of manipulation: • Straightforward manipulation – Independent variable is manipulated through written, verbal, or visual materials • Staged manipulation – Event is artificially created by the researcher to: 1. Create a psychological state in participants 2. Increase fidelity to real world Strength of Manipulation

• Within ethical and financial limits, a stronger manipulation is generally better. • Remember, if the manipulation is too weak, then it isn’t possible to detect any differences between groups. Step 3: Measuring the DV • Types of Measures: – Self-report – Behavioral – Physiological Measuring the DV Back to our example:

Whining (yes/no) Sympathy (1-10)

• This design is called a posttest-only design because the participants were only tested once, after the manipulation (hence, posttest). • Contrast this design with the pretest-posttest design (Cozby, p. 141) Pretest-Posttest Design • However, sometimes it is helpful to measure the participants on the DV both before and after the manipulation: – Small sample size (less than ~25 participants per cell) – Participant selection – Risk of mortality • Unfortunately, pretesting is more time consuming and costly. Plus, pretesting risks sensitizing the participant. Interpreting Experimental Results

• Once you have measured the participants on the DV, you can compare the scores of the two groups:

9 8 7 6 5 Sympathy 4 3 2 1 0 Whining Control Expanding the Experiment... Expanding the experiment

• Psychological phenomena are way too complicated to ever be explained with this basic design. Usually, experiments will increase: – the number of levels of the IV – the number of IVs. Increasing # of Levels of the IV

Increasing the # of levels allows us to: 1. Detect curvilinear relationships: Whining Sympathy (1-10) (control, low, high)

2. Compare more than two groups:

Persuasion (none, whining, screaming, Sympathy (1-10) begging, asking) Increasing the number of IVs

• Designs with more than one IV are called factorial designs. • Suppose we want to know how the frustration level of the partner interacts with different persuasion techniques Persuasion (whining, screaming, asking) Sympathy (1-10) Partner’s Frustration (low, high) Factorial designs

Terminology • Each IV is called a factor, and of course each will have at least 2 levels • An experimental design can be described by the number of IVs and the number of levels of each: Number of levels (IV1) X Number of levels (IV2)... Factorial designs

• Thus, our study would be a 3 X 2 • This creates 6 conditions, or cells, that participants will be assigned to. Each cell of a design requires ~25 participants. Interpreting Factorial designs

• Factorial designs yield two kinds of info: 1. Main effects Information about each of the IVs taken by itself. 2. Interactions Information about each of the IVs across levels of the other IV(s). Main Effects Remember the results from our first, basic design:

9 8 7 6 5 4 3 2 1 0 Whining Control For all practical purposes, this is a main effect for whining Main Effects Now we can combine persuasion with frustration:

12 10 8 6 4 2 Hi F rus tra tio n Lo F rus tra tio n

0 g g n ng in i ki in s h am e A r W Pe rs ua s io n Sc te c hni que Main Effects & Interactions

• Three ways to present information: 1. Bar graph 2. Line graph 3. Table

See p. 185 Expanding some more...

• Longitudinal designs are a special kind of expansion, in which the same measures are repeated at least twice, in order to assess how people change over time. • Thus, Time can be thought of as an additional IV, and the number of repetitions (Waves) are the levels of the variable. Questions?