Multi-Factor Experiments

Multi-Factor Experiments

• Crossing levels of independent variables creates experiment conditions.
• Interaction effects can be identified in experiments.

Introduction to Factorial Design

• Single-factor experiments, like the "Daphne" study from chapter 12, aim to reduce anxiety by testing different treatments (chew toys/clothes) across varying situations (trip length).
• Treatment effectiveness depends on the length of the outing.
• Factorial designs present results in a structured manner

Factorial Designs

• Most experiments utilize factorial designs and contain more than one independent variable.
• Factorial designs efficiently test the effects of multiple independent variables
• They help researchers examine the combined effects of independent variables.
• More information about causal relationships is gained through factorial designs.
• Marginal means are calculated from the independent variable, regardless of other variables
• Levels of independent variables depend on the situation
• Factorial designs test effects of each independent variable on the dependent variable.
• Main effects are tested using ANOVA.
• Means of each condition are assessed.
• The average means for each level are examined.
• Comparing marginal means determines the effect of an independent variable.
• Interaction effects explore relationships between independent variables
• Compares differences between levels of independent variables

2x2 Factorial Design

• As a guide, this involves two independent variables, each with two levels
• Examples of independent variables: type and length of therapy.
• Examples of levels: individuals vs groups, 1 week vs 6 weeks.
• Possible main effects from these examples are therapy type and length.
• Interaction can be visually represented through a line graph of condition means.
• Possible results can show how independent variables may react

Types of Interactions

• Independent variables may interact differently with individual versus group therapy.
• Independent variables can interact depending on therapy length.
• Nonparallel lines in a graph can indicate interactions.
• Simple effects tests can follow up on graphs.
• Interactions can highlight interesting effects of variables.
• An ice cream company aiming to boost sales considers adding more chocolate chips or using real vanilla beans.
• Two independent variables can exist for chocolate chip flavor.
• Ice cream data is rated on a scale of 1-7.
• Real vanilla, chips or conditions received a mean rating of 4.5/3
• Recommendation is to implement both proposals for chocolate chip flavor to increase ratings.

Experiment Examples: Cognitive

• Shaw and Porter's 2015 study explored false memories.
• Participants had no prior contact with police and had not committed crimes.
• Participants experienced emotional events from ages 11 to 14.
• Three interviews, one week apart, were conducted:
• One real and one false event were told.
• False events were categorized as crimes or emotional experiences.
• Participants were asked to recall as much as possible.
• At first interview, participants could not recall false event
• Additional events produced rated anxiety.
• 83.3% of participants believed the false event had actually happened.
• 70% classified criminal false events as a false memory.
• Main effects of memory type related to confidence and interaction may be present for anxiety.

Experiment Examples: Biological

• Silvers et al. studied brain activity response to food in children and adults.
• The hypothesis was food cravings differ with age

Food and Brain Activity Study Conditions

• Participants assessed food in two conditions:
• Close: thinking about taste and smell
• Far: focusing on color and shape.
• Foods were shown and participants assessed according to the given instructions.
• Then they rated how much they wanted to eat the food.
• Adults had lower craving ratings, and lower ratings on "far" trials
• Age and trial type did not interact.
• Adults showed different brain activity than children.
• Age had effects on craving and brain activity.

Social Example: Exclusion and Emotion

• Wesselmann et al. studied the effects of exclusion on emotion.
• Participants watched a game of Cyberball
• Inclusion: all three players included.
• Exclusion: one player was excluded by the other two.
• Participants took perspective of one of the players whether they received intrusions
• Participant mood was measured using a questionnaire
• Observation showed the exclusion game produced lower scores.
• Interaction occurred between type of game and perspective taking.

Developmental Example: Object Location Memory

• Hund and Plumert assessed memory of object location and influencing factors.
• They tested 7, 9, and 11-year-olds, as well as adults.
• Participants learned the location of objects in a house.
• In each group, objects were either placed by similarity or randomly.
• Location errors were recorded.
• Older participants made fewer errors, but all age groups placed similar objects closer together.
• Age group differences cannot be interpreted as causal effects.

Quasi-Experiments

• Quasi-experiments explain the difference between a true independent and a quasi-independent variable
• Quasi-experiments identify sources of bias.

Quasi-Experiment Introduction

• A study by Benenson et al. (2009 - study 1) examined roommate satisfaction by gender.
• Concluded that males were more tolerant than females regarding roommates.
• Gender could not be confirmed as the definite cause of tolerance, due to lack of true IV and random assignment.

Types of Quasi-Experiments

• Lack of random assignment is a defining characteristic.
• Used when random assignment is not possible.

Pretest-Posttest Designs

• Behaviors are measured before and after a condition is implemented.
• Researchers compared scores, searching for a change.
• If a change occurs, conditions cannot be automatically induced.
• For example, in smoking cessation programs:
• An alternative explanation may exist if cigarette consumption decreases post-test.
• Control groups can be created to address alternate explanations.
• If the pretest-posttest design with a control group is not randomly assigned, it has nonequivalent groups.
• A study on student achievement in low-income elementary students:
• Issue that schools funding correlate with low-scoring students
• Shadish et al. recommended variations to pretest-posttest designs.
• Being tested twice can lead to bias known as testing effects.
• Solomon four-group designs help evaluate testing effects in pretest-posttest designs.

Time Series Designs

• Scores may fluctuate.
• They account for fluctuations by measuring several times before and after treatment.
• Score patterns are compared to see if a difference exists.
• Do Americans get more depressed depending on the president?
• Researchers look at patterns of depression over the year.
• If there is no difference, patterns would be similar.
• Interrupted time series designs involve independent events, which researchers can control for.
• Example: the effect of stay-at-home orders and protests on the number of arrests
• Non-interrupted time series measure behavior before and after controlled treatment.
• For example: effect of a new therapy on depression
• Depression symptoms are measured for a period before and after therapy.
• Researchers must be cautious when interpreting relationships.
• Random assignment is more likely in time series designs.
• Traditional data analysis is often inadequate.

Threats to Validity: History

• Historical events can be experienced by large groups or personally.
• Examples include a declining economy, or the loss of a loved one.
• Researchers can minimize history effects by using a control group.

Threats to Validity: Maturation

• A source of bias that occurs when people naturally change over the course of a study.
• This includes actual maturation or other types of changes.
• As an example, participants can improve in depression study
• Control groups that do not receive treatment can minimize maturation bias.

Threats to Validity: Attrition

• Also called mortality and can be when participants drop out of a study.
• More likely in long-term studies.
• Results in data getting deleted.
• Those who drop out may differ from those who remain.
• For example, some participants in a depression study might not return for the posttest.
• Results are more difficult to interpret without knowing the cause of attrition.
• A control group can best handle it.

Specialized Designs

• This identifies different developmental designs
• This identifies different types of small-n designs.

Developmental Designs: Longitudinal

• Treat age as a within-subjects variable
• Can be an experiment, if an independent variable is included.
• Examples include studies on social withdrawal of children over time.
• Disadvantages:
• Takes participants a long time to age.
• Subject to attrition/mortality
• Testing effects can occur with multiple testings.

Developmental Designs: Cross-Sectional

• Compare different age groups of participants.
• For example: assessing student's knowledge of a topic
• Disadvantages:
• Generation (cohort) effects, cohort experiences can vary.

Developmental Designs: Cohort-Sequential

• Treats age as both a between-subjects and within-subjects factor.
• Begins with separate samples of age groups that are tested over time during development.
• For example: decline in motivation for math skills in high schoolers.
• Allows for quick age comparison.
• Generation effects may be less problematic.

Small-n Designs

• Baseline measurement determines the rate of behavior before intervention.
• Sometimes called single-case designs but often include +1 participation.
• The goal is to describe/change an individual's behavior.
• Research generally tests a theory about how behaviors work in most people.
• A treatment plan for problematic behavior can be tested
• Accomplished through repeated measurement:
• Discrete trials designs involve one or a few participants complete a lot of trials

Baseline Designs

• Compares repeated measurement of baseline behaviors to behavior after treatment.
• Advantages:
• Observations can reduce error in the data.
• Easier to control for external factors with a small group of participants.
• Disadvantages:
• Results cannot always be generalized.
• Not appropriate for some types of behaviors
• Carryover effects can occur when participant experiences affect behavior.

Discrete Trial Designs

• Some of the earliest psychology studies used discrete trials.
• For example: Ebbinghaus studied memory on himself.
• Helps researchers describe/understand the basic behavioral process.
• Can be done with just a few participants who complete a large number of trials.
• Allows for good testing of casual relationships
• Usually used to study basic processes of psychophysics and learning.
• Examples are Magood and Critchfield studying positive and negative reinforcement.
• Insight acquired includes negative and positive reinforcement schedules.
• For example: Gazzaniga et al. studied patients that had undergone a corpus callosotomy.
• Gained a better understanding of hemisphere specialization

More on Baseline Designs

• Comparison of baseline behavior and behavior with treatment.
• Aim to determine if the treatment created or desired change.
• Commonly used by school psychologists.
• ABA or reverse designs test to see if baseline behavior appears again.
• It is common for the effect of treatment to carry over to the second baseline measure.
• A common variant is the A-B-A-B design where treatment is implemented a second time.
• Tasky et al investigated the effects of treatment to improve patients tasks after brain injuries.

Data Analysis in Small-n Designs

• Data is often presented for individual participants.
• This means inferential stats can only be used if a large number of observations are recorded.
• Researchers commonly report mathematical descriptions of behavior.