PSYC 2017

• Multivariate Correlational Research


  Research that examines the relationships between multiple variables
• Which of the three causal criteria is unmet by bivariate (a correlational design that has 2 measured variables) correlational designs?
• Three causal criteria


  • Covariance
  • Temporal precedence
  • Internal validity
• Bivariate correlational designs do not meet the causal criterion of temporal precedence
• Longitudinal designs


  Can provide evidence for temporal precedence by measuring the same variable in the same people at different times
• Practical examples of longitudinal designs


  • Developmental psychologist – change in personality in adolescents over time
  • Organizational psychologists – change in commitment to organization and supervisor support over time
  • Clinical psychologist – change in depression over the course of treatment
• Example longitudinal study


  • Overvaluation and narcissism (Brummelman et al. 2015)
  • Researchers gathered data from 565 children and their parents in Netherlands
  • Children provided self-report of their narcissism personality
  • Parents indicated their engagement in overvaluation of their children
  • Both children and parents provided responses on the two variables every 6 months over 24 months
• Types of correlations in longitudinal designs


  • Cross-sectional correlations
  • Autocorrelations
  • Cross-lag correlations
• Cross-sectional correlation

  The correlation between two variables measured at the same time
• Autocorrelations


  The correlation (relationship) between multiple measures of the same variable
• Cross-lag correlations


  The correlation between variables measured at different time points
• The previous finding does not allow us to conclude that child narcissism increases parent overvaluation
• Causal criteria fulfilled by longitudinal designs


  • Covariance
  • Temporal precedence
  • Internal validity
• Longitudinal designs are not sufficient to rule out third-variable problems (spurious correlations)
• Possible explanations for the positive correlation between exposure to sexual content on TV and pregnancy risk include third-variable problems
• Statistical control


  Evaluating the association between two variables while holding a third variable constant
• Regression analysis


  • Allows us to test the linear associations between multiple variables
  • Predictor variables - variables used to predict another variable
  • Criterion variables - variable being predicted (dependent variable)
• Types of regression analysis


  • Simple Regression - ACT score predicting college GPA
  • Multiple Regression - Multiple predictor variables (ACT score, SES, gender, etc.)
• Beta


  The beta value is similar to the correlation coefficient, it indicates the effect size (strength of the association)
• Interpretation of previous regression results: even when age is considered (controlled), exposure to sex on TV still predicted risk of pregnancy
• Regression in popular media articles may use terms like "controlled for", "adjusting for", "considering"
• Multiple regression is not a foolproof way to rule out all kinds of third variables
• Mediation


  Examines the "why" - the mechanism by which a predictor variable influences a criterion variable
• Moderation


  Examines the "when" or "for whom" - the conditions under which a predictor-criterion relationship holds
• The association between ACT Math score and GPA is moderated by major - ACT positively predicted GPA for STEM majors but not for humanities majors
• Key concepts


  • Mediation
  • Moderation
  • Main effects
  • Confounds
  • Longitudinal designs
  • Multiple regression
• Bivariate correlations provide little insight into causal mechanisms, while longitudinal analyses and multiple regression can shed light on complex causal relationships
• Association claims


  Claims that can be made from a correlational research (not causal)
• Effect size


  In a correlational research context, it provides insights into the size of the relationship beyond significance. Interpreted based on magnitude and direction: ~0.2=small, ~0.5=medium, ~0.8=large
• Categorical variable


  Variable assigned to discrete categories
• Continuous variable


  Variable that can take any value in a range
• Testing association claims statistically


  1. t-test, which describes the difference in a continuous variable between groups of a categorical variable

  2. T-Test results can be transformed into a Pearson's correlation coefficient
• Association claims


  Use verbs such as link, associate, correlate, predict, tie to, and be at risk for (in contrast with causal claims which use verbs like cause, enhance, affect, decrease, and change)
• Statistical significance


  If the P value is less than the pre-specified alpha, meaning the results did not come from chance
• Factors affecting statistical validity


  • Statistically significant (p < 0.05)
  • Effect size (strength of association, does not tell if effect is significant)
  • Outliers (determine if due to error or legitimate, analyze with and without)
  • Restriction of range (observed correlation is attenuated)
• Correlational studies for establishing causality


  • Covariance (measures total variation of two random variables from their expected value)
  • Temporal precedence (establishing cause occurs before effect)
  • Internal validity (check for confounds and third-variable problems)
• Correlation does not equal causation
• Longitudinal design


  • Researchers do not manipulate variables or interfere with the environment, they simply conduct observations on the same group of subjects over time
• Cross-Sectional Correlations


  Two variables measured at the same point in time, provide insights into concurrent relationships but don't indicate causality or directionality
• Autocorrelations


  One variable measured at two different points in time (correlation with itself), help assess the stability of individual differences over time