Chapter 9: Non-Experimental Designs

Correlation and Regression: The Basics Finding the relationship between two variables without being able to infer causal relationships Correlation is a statistical technique used to determine the degree to which two variables are related Three types of [linear] correlations: Positive correlation Negative correlation No correlation

Correlation and Regression: The Basics Positive correlation Higher scores on one variable associated with higher scores on a second variable

Correlation and Regression: The Basics Negative correlation Higher scores on one variable associated with lower scores on a second variable

Correlation and Regression: The Basics Correlation coefficient Pearson’s r Statistical tests include: Pearson’s r, Spearman’s rho Ranges from –1.00 to +1.00 Numerical value = strength of correlation Closer to -1.00 or +1.00, the stronger the correlation Sign = direction of correlation Positive or Negative

Correlation and Regression: The Basics Scatterplots Graphic representations of data from your two variables One variable on X- axis, one on Y-axis Examples:

Correlation and Regression: The Basics Scatterplots Creating a scatterplot from data Each point represents an individual subject

Correlation and Regression: The Basics Scatterplots from the hypothetical GPA data for positive (top) and negative (bottom) correlations

Correlation and Regression: The Basics Scatterplots Correlation assumes a linear relationship, but scatterplot may show otherwise Curvilinear  correlation coefficient will be close to zero Left half  strong positive Right half  strong negative

Correlation and Regression: The Basics Coefficient of determination Equals value of Pearson’s r2 Proportion of variability in one variable that can be accounted for (or explained) by variability in the other variable The remaining proportion can be explained by factors other than your variables r = .60  r2 = .36 36% of the variability of one variable can be explained by the other variable 64% of the variability can be explained by other factors

Correlation and Regression: The Basics Regression Analysis – Making Predictions The process of predicting individual scores AND estimating the accuracy of those predictions Regression line – straight line on a scatterplot that best summarizes a correlation Y = bX + a Y = dependent variable—the variable that is being predicted Predicting GPA from study hours  Y = GPA X = independent variable—the variable doing the predicting Predicting GPA from study hours  X = study hours a = point where regression line crosses Y axis b = the slope of the line Use the independent variable (X) to predict the dependent variable (Y)

Correlation and Regression: The Basics Regression lines for the GPA scatterplots Study time (X) of 40 predicts GPA (Y) of 3.5 Goof-off time (X) of 40 predicts GPA (Y) of 2.1

Interpreting Correlations Correlations and causality Directionality problem Given correlation between A and B, A could cause B, or B could cause A Third variable problem Given correlation between A and B uncontrolled third variable could cause both A and B to occur Partial correlations “partial out” possible third variables

Interpreting Correlations Caution: correlational statistics vs. correlational research Not identical Correlational research could involve t tests Experimental research could examine relationship between IV and DV Using correlations The need for correlational research Some IVs cannot be manipulated Subject variables Practical/ethical reasons e.g., brain damage

Combining Correlational and Experimental Research Research example 27: Loneliness and anthropomorphism Study 1: correlation between loneliness and tendency to anthropomorphize r = .53 Studies 2 & 3: manipulated loneliness to tests its effects on likelihood to anthropomorphize IVstudy1 = [false] personality feedback (will be lonely, will have many connections with others) DVstudy1 = degree of belief in supernatural beings (e.g., God, Devil, ghosts) IVstudy2 = induce feeling of connection or disconnection DVstudy1 = anthropomorphic ratings of own pets and others’ pets Results  feelings of disconnection (loneliness) increased likelihood to anthropomorphize

Multivariate Analysis Bivariate vs. multivariate analyses Multiple regression One dependent variable More than one independent variable Relative influence of each predictor variable can be weighted Examples: predicting school success (GPA) from (a) SAT scores and (b) high school grades predicting susceptibility to colds from (a) negative life events, (b) perceived stress, and (c) negative affect

Multivariate Analysis Factor analysis After correlating all possible scores, factor analysis identifies clusters of intercorrelated scores First cluster  factor could be called verbal fluency Second cluster  factor could be called spatial skill Often used in psychological test development

Quasi-Experimental Designs no causal conclusions, less than complete control, no random assignment From prior chapters: Single-factor nonequivalent groups designs Nonequivalent groups factorial designs subject x manipulated variable factorial designs All the correlational research

Quasi-Experimental Designs Nonequivalent control group designs Typically (but not necessarily) include pretests and posttests Experimental  O1 T O2 Nonequiv control  O1 O2 Random assignment to groups not possible for practical reasons Two groups may initially be different at O1 Thus, nonequivalent groups and inevitably confounded with Treatment/No Treatment Pre-test  Flextime  Post-test Pittsburgh Pre-test  Nothing  Post-test Cleveland

Quasi-Experimental Designs Nonequivalent control group designs Research example 29 IV  whether coaches given “coach effectiveness” training Nonequivalent groups – coaches from two different leagues DV  player self-esteem (preseason and postseason)

Quasi-Experimental Designs Nonequivalent control group designs Research example: NO PRE-TEST????? IV  living distance from SF earthquake Experimental  California Nonequivalent control  Arizona DV  nightmare frequency Results California > Arizona Ruled out alternative explanation that those in California would always have more earthquake nightmares Beebe, B. (2011) Introduction to mothers, infants, and young children of September 11, 2001: A primary prevention project. Journal of Infant, Child, and Adolescent Psychotherapy, 10(2-3), 145-155.

Quasi-Experimental Designs Interrupted time series designs

Quasi-Experimental Designs Interrupted time series designs Useful for evaluating overall trends Basic design  O1 O2 O3 O4 O5 T O6 O7 O8 O9 O10 Outcomes: Best outcome  d (lower right)

Quasi-Experimental Designs Interrupted time series designs Research example 31 Effect of incentive plan on productivity Ruled out effects of history, instrumentation, and subject selection

Quasi-Experimental Designs Interrupted time series designs Variations on the basic time series design Add a control group O1 O2 O3 O4 O5 T O6 O7 O8 O9 O10 O1 O2 O3 O4 O5 O6 O7 O8 O9 O10 Add a “switching” replication Second treatment, but at a different time O1 O2 O3 T O4 O5 O6 O7 O8 O9 O10 O1 O2 O3 O4 O5 O6 O7 T O8 O9 O10

Quasi-Experimental Designs Research using archival data Data previously collected for some other purpose Often undergoes content analysis Susceptible to missing data and bias, but no reactivity Research example 32 IV  patient recovering room Experimental  pleasant view of park Nonequivalent control  brick wall DV  recovery & other factors (better for room with a view)