PSYCH291 Lecture Notes - Lecture 15: Linear Regression, Internal Validity, Multilinear Map
16 Dec 2018
School
Department
Course
Professor
Bivariate correlational research:
● research that examines an association between two measured variables (neither
variable is manipulated)
● support association claim - not causal.
e.g. violent media, linked to children aggression, must be combated
● causal claim
○ preference of violent tv or not vs. level of aggression.
○ positive association
○ fairly strong - data points are close together and not spread out too much
● three claims to satisfy causal claims:
○ covariance: two correlated variables
○ temporal precedence: one comes before the other
○ internal validity: no alternate explanation
● what causal criteria is satisfied in this claim?
○ two variables: aggression and tv preference
○ temporal precedence: violent tv comes before child aggression
○ internal validity is not satisfied.
Research methods that help satisfy causal claims
1. experimental design: manipulate one variable and measure the other
2. multivariate correlational research:
○ a longitudinal design: supports temporal precedence
○ multiple linear regression: a statistical approach to ruling out confounds.
Longitudinal Designs:
addressing causal claim with correlational research:
● claim: tv violence cause aggression?
● longitudinal design:
○ measuring variables at different times for the same participants
○ preference for violent tv and aggression (peer rated) in grade 3
■ e.g. how mean is this person?
○ preference for violent tv and aggression (peer rated) in grade 13
○ is there a strong correlation?
Autocorrelations(Longitudinal design)
● correlations between the same variable at different times
● can be considered different variables because it is measured at two different times.
● tv correlation: weak
● aggression: moderate
○ aggression is more stable over time than tv violence.
○ this may have low construct validity between violent tv and aggression
Cross-sectional Correlations (Longitudinal Design)
● association between the two variables at two different time points
○ 3rd grade: weak correlation
○ 13th grade: weak, almost non-existent correlations.
Cross-lag correlations(longitudinal design)
● temporal precedence is supported (violent tv came first)
● violent tv in third grade leads to more aggression
○ r = .31
○ moderate strength
● aggression in third grade does not lead to violent tv preference in 13th grade.
○ r = .01
○ very weak, almost non-existent correlation.
● can a confound still account for the association between watching violent tv and
aggression?
○ yes
○ internal validity isn’t satisfied.
Gender could be a confound:
● hypothetical example
○ boys could have more of a preference for violent tv than girls
○ boys on average are more aggressive than girls
■ girls: r = 0
■ boys: r = .4
● gender is more of a moderating variable than a third-variable.
Document Summary
Bivariate correlational research: research that examines an association between two measured variables (neither variable is manipulated) Support association claim - not causal. e. g. violent media, linked to children aggression, must be combated. Preference of violent tv or not vs. level of aggression. Positive association fairly strong - data points are close together and not spread out too much three claims to satisfy causal claims: Covariance: two correlated variables temporal precedence: one comes before the other internal validity: no alternate explanation. What causal criteria is satisfied in this claim? two variables: aggression and tv preference temporal precedence: violent tv comes before child aggression internal validity is not satisfied. Research methods that help satisfy causal claims: experimental design: manipulate one variable and measure the other, multivariate correlational research: Multiple linear regression: a statistical approach to ruling out confounds. addressing causal claim with correlational research: Claim: tv violence cause aggression? longitudinal design: Measuring variables at different times for the same participants.
