Vous êtes sur la page 1sur 5

LAB 14 – HIERARCHICAL REGRESSION

Purpose

One of the primary uses of multiple regression is to examine whether a predictor (or set of predictors)
explains variance in some outcome, over and above another predictor (or another set of predictors).
Researchers often do this by conducting a series of equations – adding one or more predictor at each step (or
“block”, in SPSS lingo) in the series. This “hierarchical” use of regression allows the researcher to ask very
specific questions. In addition, some of the more advanced uses of regression (e.g., mediational analysis and
the analysis or interactions) rely heavily on hierarchical regression procedures.

The purpose of today’s lab is to give you experience conducting and interpreting hierarchical regression
analysis.

Preparation

Interpretation of Hierarchical Regression

Q1. Imagine (again) that a psychologist (say Dr. Jones) was interested in the psychological characteristics
that affect academic achievement. One of her colleagues suggests that the only important determinant of
academic performance is intelligence – once you know someone’s intelligence, nothing else about that
person really matters in affecting his or her academic performance. But Dr Jones disagrees. She believes
that the next most important determinant of GPA is academic motivation – even if two students have the
same exact level of intelligence, the student with greater academic motivation will have greater academic
performance. In addition, she believes that organizational skills matter as well – even if two students have
the same level of intelligence and the same level of motivation , the student who is more conscientious and
well-organized will probably have higher academic performance.

So, Dr. Jones has a set of questions that fits a “hierarchical” analysis. That is, her theory has a clear order to
the importance of the potential predictors.

To begin evaluating her predictions, she surveyed 100 students, and she collected end-of-semester GPA’s
from each. In addition, each student completed an intelligence test (IQ), a measure of academic motivation
(AM), and a measure of Conscientiousness (C).

Here are the correlations among the measures (*** p < .001, * p < .05):
GPA & IQ .66*** IQ & AM .20*
GPA & AM .61*** IQ & C .75***
GPA & C.53*** AM & C .18(ns)

Dr. Jones conducted a three-step hierarchical multiple regression analyses (with standardized coefficients
reported below):

Step 1 – predicting GPA from IQ. βIQ = .66*** R2 = .432

Step 2 – predicting GPA from IQ and AM βIQ = .56*** βAM = .50*** R2 = .671

Step 3 – predicting GPA from IQ, AM, and C. βIQ = .52*** βAM = .50*** βC = .05(ns)
R2 = .672
Please answer the following questions relating to the hierarchical regression:

1. Looking at the correlations, interpret the correlations between the three predictors and the GPA.

2. What do the results of the first regression equation tell us, in terms of the size and direction of the
association? (β and R2)

3. The second equation tells us about the degree to which AM is related to GPA, over and above IQ.
Here, we’re interested in the regression weight for the new predictor (size and significance) and
in the difference in R2 between the first and second equation. Interpret these values, in terms of
how they inform Dr. Jones’s prediction

4. The third equation tells us about the degree to which C is related to GPA, over and above IQ and
AM. Again, we’re initially interested in the regression weight for the new predictor (size and
significance) and in the difference in R 2 between the second and third equation. Interpret these
values, in terms of how they inform Dr. Jones’s prediction

In addition, answer these questions relating to the results in general (these will help prepare you for the
test)

5. Note that the β for IQ is smaller in Equation 1 than it is in Equation 2. Why is this (hint, look at
the correlations)?

6. Look at all of the regression weights in Equation 3 (it has all of the predictors). What do these
regression weights tell us about the strongest unique predictors of academic performance?

7. In equation 3, note that the β for C very small and non-significant. But the correlation between C
and GPA is large and significant at p < .001. Why does this difference exist? Hint, again, look at
some of the other correlations
Q2. Read the following study, available on the class web site:
Wagerman, S.A., & Funder, D.C. (in press). Acquaintance reports of personality and academic
achievement: A case for conscientiousness. Journal of Research in Personality.

Answer the following questions about the article:


Introduction
a) what are the three hypotheses?
Methods
b) Who are the “targets” and who are the “informants”?
c) How was conscientiousness measured?
d) How was academic achievement measured?
Results: correlations
e) To what degree are high school GPA and SAT scores correlated with college academic achievement.
Are there statistically significant correlations?
f) To what degree is self-reported conscientiousness correlated with academic achievement. Are there
statistically significant correlations?
g) To what degree is informant-reported conscientiousness correlated with academic achievement. Are
there statistically significant correlations?
Results: hierarchical regression (see the text and the top part of Table 3).
h) Examine the hierarchical regression analyses for predicting freshman GPA
i) There are three steps in the analysis – how do you know this?
j) Which variable was entered as a predictor in the first step? How much variance in Freshman GPA
did this account for? What is the raw (undstandardized) regression weight for this variable, and why
is it so low? Is it a statistically significant predictor?
a. Which variable was entered in the second step? How much unique variance did this
variable account for? What is the raw (unstandardized) regression weight? Is this
variable a statistically significant predictor
b. Which variable was entered in the third step? Why was this variable entered last? That
is, why was this variable entered only after two other variables had been entered? How
much unique variance did this variable account for, and is it significant?
c. Do you agree with the authors’ conclusion that personality should be considered in the
college admissions process? Is the authors’ research consistent with the conclusion?

Lab Session

Bring your computer to lab.

Q& A about lecture

Go over the homework assignments

Work on an exercise in conducting and interpreting hierarchical regression analysis.


Lab Exercise - Conducting and Interpreting Basic Hierarchical Regression Analysis

The “Interpersonal Circumplex” approach to personality and behavior emphasizes two key dimensions of
social behavioral styles. One is dominance, and the other is friendliness/warmth. In this approach, many
other traits or styles of behavior are blends of these two basic dimensions. This approach is based on early
theorizing by Timothy Leary (the famous LSD-taking Harvard Psychologist from the 1960’s), and it is often
represented in a circular figure:

One issue in the circumplex approach is known as


“complementarity.” The idea is that in our interactions with each
other, our behavioral styles complement each other. Specifically,
dominance in one person should elicit submissiveness in his/her
partner. But warmth/friendliness in one person should elicit
warmth/friendliness in his or her partner.

A related issue in personality psychology is the degree to which our


behavior is influenced by our own personality and by the
environment around us.

Today, you’ll be analyzing some real data to examine an issue related to interpersonal circumplex and to self
vs environmental determination of behavior. Specifically, we’ll use SPSS “Blocks” to conduct a hierarchical
regression analysis of sets of predictors. This will give us significance tests of the sets of predictors.

Download the “hierregpersbeh2.sav” data set from the class web site.

This file includes real data on about 180-190 “target” people. For this study, each target person completed a
set of self-report personality questionnaires, including measures of assertiveness and warmth. In addition,
each target person was partnered with an opposite-sex stranger for a 5-minute “getting to know you”
conversation which was videotaped (with the participants’ and partners’ full knowledge). From these
videotapes, a group of independent research assistants observed each target participant and coded the
participant’s behaviors into set of behaviors, which included “Dominates the interaction” and “Expresses
warmth”. Finally, the partners all completed self-report personality questionnaires, which included measures
of their assertiveness and warmth.

So, the key variables:


 Subassert personality: target subjects’ self-reported assertiveness
 Subwarmth personality: target subject’s self-reported warmth
 Partassert personality: partners’ self-reported assertiveness
 Partwarmth personality: partner’s self-reported warmth
 BDominance behavior: target subject’s dominance in the interaction
 BWarmth behavior: target subject’s warmth in the interaction

We’ll do analysis to examine:


1. To what degree are self-reported personality traits associated with actual behavior?
2. Do the associations make sense, with regard to circumplex theory?
3. To what degree is one’s behavior affected by a partners’ personality? That is, after controlling for the
association between “your own personality” and “your own behavior”, do partner’s personality seem
to have an effect on your behavior?
4. Do these association make sense, with regard to circumplex theory?
First, lets examine the dominance behaviors.

We’ll treat the four personality variables as two sets (or blocks) of predictors. The first block will include the
targets’ two self-report personality ratings. The second block will include the partners’ two self-report
personality ratings. We’re going to tell SPSS to analyze two equations (or models) – the first model will
include only the first block of predictors, and the second model will include the first block and the second
block of predictors. The difference between the two models will provide information about the effect of
partner’s personality on behavior, over and above the target’s own personality.

 Analyze  Regression  Linear


 Put BDominates in the Dependent box (this is the outcome to be predicted)
 Put subassert and subwarmth in the Independent box (these are the first block of predictors)
 Click the “Next” button. This will tell SPSS that we want to have more than one block of predictors.
 Put partassert and partwarmth in the Independent box (these are the second block of predictors)
 Click the “Statistics” button and make sure that the “R squared change” option is marked. This tells
SPSS to give us information about the change in R2 when we go from the first to the second model.
Then click the “Continue” button
 Click the “OK” button.

Interpret the output. There are two main things that we’re interested in.

1. Model summary.
 Note that there’s one row of results for each of the two models.
 Interpret the R2 for each model.
 In the “Change Statistics” section, interpret the “R square change” for model 2 and the “Sig F
Change” result.
 What do these results say about the conceptual issue regarding the degree to which our behavior is
influenced by our own traits and by the traits of those around us?

2. Coefficients

 In Model 1 (target personality), which trait is most strongly associated with dominant behavior? Are
they each statistically significant? Do the results make sense?
 The results for Model 2 present the results of the 4-predictor analysis (the two target personality traits
and the two partner personality traits). We’ll focus only on the coefficients for the partner trait
variables. Which (if any) of the partner traits is associated with target behavior, over and above target
traits).
 What do these results suggest about the complementarity issue?

Re-do the analysis, using the same four predictors, but using Behavioral warmth as the outcome
variable.

Vous aimerez peut-être aussi