POLI 205
12/20/2018
1
Chapter 13: Correlation and Linear Regression
Introduction to the Concept of Correlation
• Throughout the course so far, we’ve tested differences among groups
– Categorical IV(s) with continuous DV
• In this chapter, we discuss research designs where both the IV and DV are continuous
Howard T. Tokunaga, Fundamental Statistics for the Social and Behavioral Sciences © SAGE Publications, 2016
12/20/2018
2
• Correlation – Relationship between two variables such that changes in the values of one variable are accompanied by changes in the values of the other variable
Howard T. Tokunaga, Fundamental Statistics for the Social and Behavioral Sciences © SAGE Publications, 2016
Introduction to the Concept of Correlation
• Describing the relationship between variables – Nature: Linear vs. nonlinear relationships
Introduction to the Concept of Correlation
Howard T. Tokunaga, Fundamental Statistics for the Social and Behavioral Sciences © SAGE Publications, 2016
12/20/2018
3
• Describing the relationship between variables
– Direction: positive vs. negative relationship
Howard T. Tokunaga, Fundamental Statistics for the Social and Behavioral Sciences © SAGE Publications, 2016
Introduction to the Concept of Correlation
• Describing the relationship between variables – Strength of the relationship
Howard T. Tokunaga, Fundamental Statistics for the Social and Behavioral Sciences © SAGE Publications, 2016
Introduction to the Concept of Correlation
12/20/2018
4
• Measuring the relationship between variables: Pearson correlation coefficient (r)
– Measures linear relationship
– Sign (+ or ‐) indicates direction of relationship
– Numeric value (ranging from ‐1.00 to +1.00) indicates strength of relationship
Howard T. Tokunaga, Fundamental Statistics for the Social and Behavioral Sciences © SAGE Publications, 2016
Introduction to the Concept of Correlation
• Pearson correlation coefficient (r)
Howard T. Tokunaga, Fundamental Statistics for the Social and Behavioral Sciences © SAGE Publications, 2016
Introduction to the Concept of Correlation
12/20/2018
5
• Pearson correlation coefficient (r) – Calculation based on variance and covariance: See chapters 11 & 12.
– Variance: Differences in scores for each variable
– Covariance: Extent to which two variables vary together
• Height and weight have variance (individuals differ in how tall they are and how much they weigh) and also covariance (ie, people who are taller than average also weigh more than average)
Howard T. Tokunaga, Fundamental Statistics for the Social and Behavioral Sciences © SAGE Publications, 2016
Introduction to the Concept of Correlation
Predicting One Variable from Another: Linear Regression
• Regression: using the relationship between variables to predict values of one variable from values of the other
• You might remember the following equation from previous coursework:
Y = mX + b
Howard T. Tokunaga, Fundamental Statistics for the Social and Behavioral Sciences © SAGE Publications, 2016
12/20/2018
6
Guessing Game
With no information:
Age of a person
Guessing Game
With no information: Guess the measure of central tendency (mean, median, or mode).
CIA Factbook: 38.5 years
I will give you one piece of information about me in order to guess my wife’s age.
12/20/2018
7
OLS History
A.M. Legendre (1805) – comet orbits
C.F. Gauss claims he developed it in 1794.
Legendre wins: “Méthode des moindres carrés”
Method of least squares
See: Stephen M. Stigler, "Gauss and the Invention of Least Squares," Ann. Statist., 9(3), 1981: 465– 474.
12/20/2018
8
The Math
y = 0 + 1x1 + 0 = y‐intercept (constant, a.k.a. alpha) 1 = Beta (coefficient) = error (residuals)
= predicted value of y ( = 0 + 1x1) = mean value of y
Beta ()
is the amount that y increases as x increases by 1 unit.
The sign of the indicates the direction of the slope of the prediction line.
Works best when y is a continuous variable.
12/20/2018
9
Explained & Unexplained
12/20/2018
10
Correlational Statistics vs. Correlational Research
• Correlational research are designs that examine the relationship between variables, but without the ability to infer causality
– Variables are measured rather than manipulated by researchers
– Causation can only be established when IVs are isolated and manipulated
Howard T. Tokunaga, Fundamental Statistics for the Social and Behavioral Sciences © SAGE Publications, 2016