Lecture 11_Qualitative and Quantitative IVs
Visualizing Bivariate Regression When 1 IV is a dichotomy and 1 is Continuous
Suppose that two groups are given training on how to perform a job. One of the groups has been
given the old training method. It is Group 0. The other group has been given a recently purchased new
training program. It is group 1. Suppose also that a test of cognitive ability, probably an ability that is
related to the job, has been given each person. The dependent variable is the final performance after
completion the training program. That final performance depends on both the type of training and
probably also on cognitive ability. In the following, cognitive ability is X and final performance is Y.
.
The data are as follows . . .
CogAbility
54
49
66
45
65
47
50
39
53
29
52
38
32
24
58
49
62
46
35
50
55
55
44
44
56
56
55
67
37
52
GROUP
Performance
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
Number of cases read:
45
60
68
43
61
29
46
49
59
41
58
42
30
30
54
54
77
67
50
58
47
74
47
62
71
54
67
68
58
54
30
Number of cases listed:
Lecture 11_Qualitative And Quantitative IVs - 1
30
7/28/2017
Visualizing the Data – A two-group Scatterplot
A useful way of visualizing the relationship of a dependent variable to a continuous independent
variable and also a dichotomous independent variable is with a plot of the dependent variable vs. the
continuous independent variable using different symbols for the two groups represented by the
dichotomous iv. In this case, we would plot final performance vs. initial ability (Y vs. X) with different
symbols used for Group 0 and Group 1. The following represents such a plot . . .
This plot allows you to see both the relationship of Y to X (generally increasing) but also the
relationship of Y to Group (larger Y's for Group 1).
It also lets you compare the relationship of Y to X within Group 0 with the relationship of Y to X within
Group1. In this case the relationships are similar - generally increasing. Later in the course, when we
study nonlinear relationships, we'll see examples in which the Y-X relationship varies from one group to
the other.
Note that this plot would not work if GROUP had many values, say 20+ values. That’s because there
would be 20 different plotting symbols, and it would be impossible to discern relationships of Y to X
with 20+ different symbols strewn about the graph.
So this only works when the Qualitative independent variable has just a few values.
Lecture 11_Qualitative And Quantitative IVs - 2
7/28/2017
The regression analysis
Note that since the Qualitative factor has only two levels, we do not have to create group coding
variables.
Regression
Va riabl es Entere d/Rem ov e db
Mo del
1
Va riable s
Re move d
Va riable s En tered
X, GRO UP a
Me thod
. En ter
a. All requ ested vari ables ente red.
b. De pend ent V ariab le: Y
Model S ummary
Mo del
1
R
.78 1 a
Std . Erro r of
the Estim ate
8.2 1003
R S quare
Ad justed R S quare
.61 0
.58 1
a. Pre dicto rs: (Consta nt), X , GROUP
ANOVA b
Mo del
1
Re gression
Su m of Squa res
28 44.77 4
df
2
Me an S quare
14 22.38 7
67 .405
Re sidua l
18 19.92 6
27
To tal
46 64.70 0
29
F
21 .102
Sig .
.00 0 a
a. Pre dicto rs: (Consta nt), X , GROUP
b. De pend ent V ariab le: Y
Coeffici ents a
Un stand ardized Co efficients
Mo del
1
(Co nstan t)
B
14 .644
Std . Erro r
7.0 95
9.9 46
3.0 57
.70 7
.14 5
GROUP
X
Sta ndardized
Co effici ents
Be ta
t
Sig .
2.0 64
.04 9
.39 9
3.2 53
.00 3
.59 8
4.8 77
.00 0
a. De pend ent V ariab le: Y
Verbal Interpretation of B Coefficients
BX = .707: Among persons in the same group, a 1-point increase in cognitive ability 0
is associated with a .707 increase in performance. Relationship is statistically significant.
1
BGroup = 9.946: Among persons of equal cognitive ability, a 1-point change in GROUP, i.e., moving
from Group 0 to Group 1 is associated with a 9.946 increase in performance. Difference is significant.
Predicted Y = 14.644 + .707*X + 9.946*GROUP.
When one of the variables is a grouping variable, analysts often write separate equations for each group.
For people in Group 1, Predicted Y = 14.644 + .707*X + 9.946*1 = 24.590 + .707*X
For people in Group 0, Predicted Y = 14.644 + .707*X + 9.946*0 = 14.644 + .707*X
Lecture 11_Qualitative And Quantitative IVs - 3
7/28/2017
Showing the regression parameters on the graph
The separate group regression lines are drawn on the graph.
Group 1: Predicted Y = 24.590 + .707*X
Group 0: Predicted Y = 14.644 + .707*X
80
Y-hat = 24.644 + .707X
70
Y-hat = 14.590 + .707X
60
50
40
GROUP
30
TP=1
Y
20
TP=0
20
30
40
50
60
70
X
This plot precisely represents the results of the MR analysis. Specifically . . .
1. The common slope represents the partial regression coefficient for X - .707. It's the expected
change in Y when X increases by 1 among persons in the same group. As shown above there are two
lines – one for Group 0 and the other for Group 1. Both have the same slope: 0.707.
2. The difference in heights of the two lines represents the partial regression coefficient for
Group. It's the expected change in Y when Group changes by 1 unit (going from 0 to 1 or vice versa)
among persons with the same X value.
Lecture 11_Qualitative And Quantitative IVs - 4
7/28/2017
More on the partial regression coefficient for the dichotomous variable, Group in this case.
80
70
60
50
40
GROUP
Partial regression
coefficient forTP=1
the
Group (9.946).
30
Y
20
TP=0
20
30
40
50
60
70
Cognitive
INITABILAbility
The partial regression coefficient for the dichotomous predictor (Group) is the difference in height of the
two lines.
Among persons equal on X (that is, among person at a particular position on the X-axis a one -unit
difference in group (going from one group to the other) is associated with a 9.946 difference in Y.
80
Think of the people equal on X
as those within a thin vertical
slice of the scatterplot, as
illustrated here.
70
60
50
40
GROUP
30
TP=1
Y
20
20
TP=0
30
40
50
INITABIL
Cognitive
Ability
60
70
A bunch of people with the
same X value.
X
Lecture 11_Qualitative And Quantitative IVs - 5
7/28/2017
The Y vs. X Graph with group-specific lines through each group of points.
80
70
60
50
40
GROUP
30
Y
1.00
20
.00
20
30
40
50
60
70
X
This plot is different from that on the previous page in that the lines through each group’s points have
been allowed to have slopes appropriate to the group through which they’re drawn. In the previous
graph, the slopes of the lines were constrained to be equal to 0.707, the partial regression slope for X.
The regression analysis conducted for these data assumed equal slopes in the two populations. The
graph here provides a visual test of that assumption. To my eyes the two slopes are pretty nearly the
same, but not precisely parallel.
There is a formal test of the inequality of the slopes. If that test is significant, then a more complicated
analysis must be performed. More on that in the lecture on moderation and in the lecture on analysis of
covariance.
Lecture 11_Qualitative And Quantitative IVs - 6
7/28/2017
Regression with One 3-Category Qualitative and One Quantitative Independent
Variable
(Data are RandomizedBlocks in P595B HA folder. N is 240, too large to allow the data to be displayed
here.)
A company is considering switching from the current training program to one of two alternative
programs. The current program has been in effect for many years. Two alternative programs have
recently become available. The first is a video / computer based method, involving presenting key
information on DVDs and then using a stand-alone computer program to present drill-and-practice on
the material presented. The 2nd is a completely web-based training method, in which the information
is presented over the web and interactive drill-and-practice are also presented over the web. Before the
company decides to make any kind of switch, it must determine whether there are any significant
differences in learning after having gone through the three programs.
Two hundred forty participants are randomly assigned to one of three groups of trainees with 80
participants in each group. One group receives the current training, the 2nd group receives the
video/computer training, and the third receives the web-based training. A measure of cognitive
ability (X), known to predict performance after training, is taken prior to beginning training. The data
are as follows. X is the cognitive ability measure. Y is the performance after training. TRTMENT
is the training program: 1=standard; 2=video/computer, and 3=web-based.
So, for this problem, we’re comparing TRTMENT means controlling for cognitive ability.
Summarize
De scriptiv e S tatis tics
TRTME N
1
T
2
X
N
80
Me an
49. 68
110 .06
Me dian
50. 50
111 .00
Std .
De viatio n
10. 45
15. 51
N
80
80
Me an
49. 54
111 .95
Me dian
50. 00
114 .00
9.6 5
13. 96
Std .
De viatio n
3
To tal
Y
80
N
80
80
Me an
49. 21
117 .94
Me dian
51. 00
120 .00
Std .
De viatio n
10. 91
15. 28
N
240
240
Me an
49. 47
113 .32
Me dian
50. 00
114 .00
Std .
De viatio n
10. 31
15. 25
Lecture 11_Qualitative And Quantitative IVs - 7
7/28/2017
The questions we’re asking are the following:
1) Among persons equal in cognitive ability (X), are there mean differences between the groups in
performance after training? This is the main question.
2) Among persons within the same group, is there a relationship between performance after training
and cognitive ability? This is a question that we assume will be answered positively, otherwise we
wouldn’t control for cognitive ability. But we will check it anyway.
The graphical representation with plot of Y vs. X with different plotting symbols for each group.
(Individual group regression lines were added because they’re so easy to get.)
1) Note that the Treatment 3 line – the green one - is above the Treatment 2 line (the red one) which is
(usually) above the Treatment 1 line (the dotted one). This hints at differences between average
performance between groups.
2) Note the generally positive relationship of Y to X, overall and within each group. Looks like the
dependent variable is related to the cognitive ability.
The analysis using the Regression Procedure
The data originally consists of just 3 columns - the Y column, the TRTMENT column, and the X
column.
Lecture 11_Qualitative And Quantitative IVs - 8
7/28/2017
Creation of Group Coding Variables
To analyze the problem using the REGRESSION procedure, we must create group coding variables for
the TRTMENT variable.
Since one of the groups is a natural control group, we’ll use dummy coding, using TRTMENT=1 as the
reference group. So the coding will be
TRTMENT
1
2
3
DCODE1
0
1
0
DCODE2
0
0
1
So, after creating group coding variables and using them to represent the groups, the data editor might
look like the following
The hypotheses being tested.
1. We want to assess the significance of differences in mean performance between groups,
controlling for X.
The first test is assessed by computing the increase in R2 due to addition of the group coding
variables to the equation.
We do a regression without the group coding variables, then adding the group coding variables, as a set,
to the equation.
Specifically, first, we enter X
Second, we enter the two group-coding variables representing TRTMENT.
We assess the significance of R2 change in Step 2.
2. We want to assess the significance of the relationship of Y to X, controlling for group
differences.
Since X is a single variable, we can assess its significance by simply examining the t value (and its pvalue) in the Coefficients box.
Lecture 11_Qualitative And Quantitative IVs - 9
7/28/2017
Regression
A two-step regression analysis is conducted.
Two steps are needed because one of the tests involves a set of independent variables.
Step 1: Enter continuous predictor, X.
Step 2: Add the set of group-coding variables, DCODE1, DCODE2.
[DataSet1] F:\MdbT\P595B\HAs\RandomizedBlocks.sav
De scriptiv e Statis tics
x
Me an Std . Deviatio n
49 .48
10 .307
y
N
24 0
11 3.32
15 .247
24 0
dco de1
.33
.47 2
24 0
dco de2
.33
.47 2
24 0
Group
Standard
Video
Web
DCODE1
0
1
0
DCODE2
0
0
1
Correla tions
Pe arson Correlatio n x
x
1.0 00
y
.43 0
dco de1
.00 4
dco de2
-.0 18
y
.43 0
1.0 00
-.0 64
.21 5
dco de1
.00 4
-.0 64
1.0 00
-.5 00
dco de2
-.0 18
.21 5
-.5 00
1.0 00
Va riable s Entered/Rem ov e db
Mo del
1
Va riable s En tered
xa
2
dco de1, dcod e2 a
Va riable s
Re move d Me thod
. En ter
Significance of the increase in
R2 (from 0) when X
(cognitive ability) was added
to the equation, p < .001.
. En ter
a. All requ ested varia bles entered.
b. De pend ent V ariab le: y
Model S umm ary
Mo del
1
Ch ange Stat istics
Std . Erro r of
R
R S quare Ad justed R S quare the Estim ate R S quare Ch ange F Chang e
df1
df2
Sig . F Chang e
.43 0 a
.18 5
.18 1
13 .798
.18 5
53 .847
1
23 8
.00 0
2
.48 7 b
.23 7
.22 7
13 .404
.05 2
8.0 92
2
23 6
.00 0
a. Pre dicto rs: (Consta nt), x
b. Pre dicto rs: (Consta nt), x, dco de1, dcod e2
Significance of the increase in
R2 when the group-coding
variables were added to the
equation, p < .001.
Lecture 11_Qualitative And Quantitative IVs - 10
7/28/2017
ANOVAc
Mo del
1
2
Re gressi on
Su m of Squa res
102 50.9 65
df
Me an S quare
102 50.9 65
1
Re sidua l
453 08.9 68
238
To tal
555 59.9 33
239
Re gressi on
131 58.5 65
3
438 6.18 8
Re sidua l
424 01.3 69
236
179 .667
To tal
555 59.9 33
239
F
53. 847
Sig .
.00 0 a
24. 413
.00 0 b
Performance is related to the
whole collection of IVs in each
model.
For Mode1 1, X is the only
predictor.
For Model 2, the X and the
two group coding variables
make up the collection of IVs.
190 .374
a. Pre dicto rs: (Consta nt), x
b. Pre dicto rs: (Consta nt), x, dcod e1, d code 2
c. De pend ent V ariabl e: y
The significance of each individual variable can be obtained from the Coefficients table below.
You’ll probably want the significance of X, the continuous predictor.
You may also want the significance of the dummy coding variables.
Coeffici ents a
Un stand ardized
Co effici ents
Mo del
1
(Co nstan t)
x
B
Std . Erro r
81 .880
4.3 76
Sta ndardized
Co effici ents
Be ta
Co rrelat ions
t
18 .712
.63 5
.08 7
.43 0
78 .182
4.4 40
.64 2
.08 4
.43 4
dco de1
1.9 76
2.1 19
dco de2
8.1 72
2.1 20
Sig .
Ze ro-ord er
.00 0
7.3 38
.00 0
17 .609
.00 0
7.6 28
.06 1
.93 2
.25 3
3.8 55
Pa rtial
Pa rt
.43 0
.43 0
.43 0
.00 0
.43 0
.44 5
.43 4
.35 2
-.0 64
.06 1
.05 3
.00 0
.21 5
.24 3
.21 9
dco de1
dco de2
2
(Co nstan t)
x
a. De pend ent V ariab le: y
Ex clude d Va riabl es b
Co llinea rity
Sta tistics
Mo del
1
dco de1
Be ta In
-.0 65 a
t
-1. 117
dco de2
.22 3 a
3.9 14
Sig .
Pa rtial Corre lation
.26 5
-.0 72
.00 0
.24 6
To leran ce
1.0 00
1.0 00
a. Pre dicto rs in the M ode l: (Co nstan t), x
b. De pend ent V ariab le: y
Mean of Group 3
was significantly
larger than the
mean of the
standard group
among people of
equal X.
I’ve never used this
table.
Lecture 11_Qualitative And Quantitative IVs - 11
7/28/2017
The analysis with GLM
The data, again.
Specifying the analysis . . .
Putting the name of a
variable in the Fixed
Factor(s) field tells
GLM that the variable
needs group coding
variables. GLM will
automatically create
them.
Don’t put the name of a
quantitative variable in
the Fixed Factor(s)
field. GLM will create
many many many group
coding variables, then
perhaps terminate with
an error message.
Lecture 11_Qualitative And Quantitative IVs - 12
7/28/2017
Plots . . .
Post hocs . . .
Alas, Post Hocs are not available when you have a continuous Covariate.
So, we can’t, for example, compare the group means with the means of the control group. This is a
problem with the use of GLM and one of the reasons it may pay to know how to do this analysis using
the REGRESSION procedure .
Options . . .
Lecture 11_Qualitative And Quantitative IVs - 13
7/28/2017
Results . . .
Univariate Analysis of Variance
[DataSet1] G:\MdbT\P595B\HAs\RandomizedBlocks.sav
Between-Subjects Factors
N
trtment
1
80
2
80
3
80
Descriptive Statistics
Dependent Variable:y
trtment
Mean
Std. Deviation
N
1
110.06
15.514
80
2
111.95
13.965
80
3
117.94
15.276
80
Total
113.32
15.247
240
Levene's Test of Equality of Error Variancesa
Dependent Variable:y
F
df1
.064
df2
2
Sig.
237
.938
Tests the null hypothesis that the error variance of
the dependent variable is equal across groups.
a. Design: Intercept + x + trtment
Lecture 11_Qualitative And Quantitative IVs - 14
7/28/2017
Tests of Between-Subjects Effects
Dependent Variable:y
Partial
Type III Sum of
Source
Squares
df
Mean Square
F
Sig.
Eta
Noncent.
Observed
Squared
Parameter
Powerb
13158.565a
3
4386.188
24.413
.000
.237
73.239
1.000
Intercept
66125.624
1
66125.624
368.046
.000
.609
368.046
1.000
x
10453.806
1
10453.806
58.184
.000
.198
58.184
1.000
2907.600
2
1453.800
8.092
.000
.064
16.183
.956
Error
42401.369
236
179.667
Total
3137320.000
240
55559.933
239
Corrected Model
trtment
Corrected Total
a. R Squared = .237 (Adjusted R Squared = .227)
b. Computed using alpha = .05
Corrected Model: Same information as in the REGRESSION Anova box.
Intercept: Test of the hypothesis that the population Y-intercept (Constant in REGRESSION output) is
zero.
X: Test of the hypothesis that in the population, controlling for trtment, the slope relating DV to X is
zero.
Conclusion: Among persons equal on trtment (i.e., in the same group) there is a significant
relationship of Y to X (cognitive ability)
Trtment: Test of the hypothesis that the population means of the 3 conditions are equal when
controlling for differences in X.
Conclusion: Among persons equal on X (cognitive ability) there are significant differences in the means
of the three groups.
Lecture 11_Qualitative And Quantitative IVs - 15
7/28/2017
New Topic - Creating Scale scores – Start here on 4/9/13
Questions like: “Does Conscientiousness predict Test Performance?” are answered by computing scale
scores.
A scale score is computed from a collection of conscientiousness items. This scale score represents
Conscientiousness.
A scale score may be computed from the items of a measure of performance. That scale score would
represent Performance.
Finally, the correlation coefficient between the two scale scores is computed.
Procedure for computing a scale score
Data: Biderman, Nguyen, & Sebren, 2008.
GET FILE='G:\MdbR\1Sebren\SebrenDataFiles\SebrenCombined070726NOMISS2EQ1.sav'.
The data typically are entered in the order in which they appear on questionnaire data sheets.
In this case, 50 columns contain the responses to the 50-item IPIP Big Five exactly as they appear on the
questionnaire.
The next 20 or so columns contain the reverse coded responses to the negatively-worded items. I
created them using the SPSS RECODE command.
1. Reverse score the negatively-worded items.
q2 q4 q6 q8 q10 q12 q14 q16 q18 q20 q22 q24 q26 q28 q29 q30 q32 q34 q36 q38 q39 q44 q46 q49
Here’s syntax to perform the recode:
recode q2 q4 q6 q8 q10 q12 q14 q16 q18 q20 q22 q24 q26 q28 q29 q30 q32 q34 q36 q38 q39 q44
q46 q49 (1=7)(2=6)(3=5)(4=4)(5=3)(6=2)(7=1) into
q2r q4r q6r q8r q10r q12r q14r q16r q18r q20r q22r q24r q26r q28r q29r q30r q32r q34r q36r
q38r q39r q44r q46r q49r.
You don’t need to do this using syntax. It can be done using pull-down menus or by hand. But it must
be done.
Lecture 11_Qualitative And Quantitative IVs - 16
7/28/2017
2. Define Missing Values Tell SPSS if specific values are to be treated as missing.
This is very important. A fairly recent thesis student lost several days because the student
created scale scores without declaring missing values. MISSING VALUES MUST BE DECLARED
FOR ALL ITEMS.
3. Determine which items belong to which scale?
The IPIP items are distributed as follows: E A C S O E A C S O E A C S O . . .
That is the 1st, 6th, 11th, 16th, 21st, 26th, 31st, 36th, 41st, and 46th items are E items.
The 2nd, 7th, 12th, 17th, etc are A. And so forth.
4. Compute Scale scores.
4a. In syntax
To compute the E scale score,
E = (q1 + q6r + q11 + q16r + q21 + q26 + q31 + q36r + q41 + q46r) / 10.
Manual Arithmetic.
In syntax, that would be
compute e = (q1+q6r+q11+q16r+q21+q26+q31+q36r+q41+q46r)/10.
If it’s computed this way, the result for any case with a missing value will be treated as missing.
It can also be computed as
compute e = mean(q1,q6r,q11,q16r,q21,q26,q31,q36r,q41,q46r).
If it’s computed this way, if a response is missing, the mean will be taken across the remaining
nonmissing items.
So, after all negatively worded items have been recoded, the syntax to compute all of the Big Five scale
scores would be
compute
compute
compute
compute
compute
e
a
c
s
o
=
=
=
=
=
mean(q1,q6r,q11,q16r,q21,q26r,q31,q36r,q41,q46r).
mean(q2r,q7,q12r,q17,q22r,q27,q32r,q37,q42,q47).
mean(q3,q8r,q13,q18r,q23,q28r,q33,q38r,q43,q48).
mean(q4r,q9,q14r,q19,q24r,q29r,q34r,q39r,q44r,q49r).
mean(q5,q10r,q15,q20r,q25,q30r,q35,q40,q45,q50).
Cut this page and the previous page out and paste it on your wall for when you analyze your thesis
data.
Lecture 11_Qualitative And Quantitative IVs - 17
7/28/2017
4b. Computing a scale score using the TRANSFORM menu . .
Repeat the above for each scale, substituting appropriate item names.
Lecture 11_Qualitative And Quantitative IVs - 18
7/28/2017
5. Run FREQUENCIES on scale scores.
Negatively
skewed.
Lecture 11_Qualitative And Quantitative IVs - 19
7/28/2017
6. Run Reliabilities of each scale
Reliability Statistics
Cronbach's
N of Items
e
Alpha Based on
Cronbach's
Standardized
Alpha
Items
.792
.794
10
Reliability Statistics
a
Cronbach's
Alpha Based on
Cronbach's
Standardized
Alpha
Items
.833
N of Items
.832
10
Reliability Statistics
c
Cronbach's
Alpha Based on
Cronbach's
Standardized
Alpha
Items
.799
N of Items
.799
10
s
Cronbach's
Alpha Based on
Cronbach's
Standardized
Alpha
Items
.825
N of Items
.836
10
Reliability Statistics
Cronbach's
o
Alpha Based on
Cronbach's
Standardized
Alpha
Items
.848
N of Items
.849
10
Lecture 11_Qualitative And Quantitative IVs - 20
7/28/2017
7. Compute correlations between scale scores.
Correlations
hcon C
summated scale
scores from IPIP
50-item Big Five
hext
hext
Pearson Correlation
hagr
1.000
Sig. (2-tailed)
scale
hsta
hopn
.254
.155
.194
.241
.003
.074
.024
.005
N
135
135
135
135
135
Pearson Correlation
.254
1.000
.421
.224
.426
Sig. (2-tailed)
.003
.000
.009
.000
N
135
135
135
135
135
Pearson Correlation
.155
.421
1.000
.277
.238
scores from IPIP 50-item Big Sig. (2-tailed)
Five scale
N
.074
.000
.001
.005
135
135
135
135
135
hsta
Pearson Correlation
.194
.224
.277
1.000
.226
Sig. (2-tailed)
.024
.009
.001
N
135
135
135
135
135
Pearson Correlation
.241
.426
.238
.226
1.000
Sig. (2-tailed)
.005
.000
.005
.008
N
135
135
135
135
hagr
hcon C summated scale
hopn
.008
135
The mean of the correlations between scale scores is
(.254+.155+.194+.241+.421+.224+.426+.277+.238+.226)/10 = .265.
Wait!! Aren’t the Big Five dimensions supposed to be independent dimensions of personality? If so,
why are they generally positive correlated?
This question is leading to a ton of research right now. Key phrases: higher order factors of the Big
Five; general factor of personality.
Lecture 11_Qualitative And Quantitative IVs - 21
7/28/2017
8. Compute correlations of scale scores with variables your theory says they should correlate
with.
Correlations
hcon C
summated scale
scores from IPIP
50-item Big Five
scale
hcon C summated scale
scores from IPIP 50-item Big
Pearson Correlation
test
1.000
Sig. (2-tailed)
.086
.320
Five scale
test
N
135
135
Pearson Correlation
.086
1.000
Sig. (2-tailed)
.320
N
135
135
In this case, the correlation is not significant. After two years of thinking about it, we hit upon the idea
that perhaps the correlation was suppressed by method bias. That turned out to be a viable hypothesis.
Lecture 11_Qualitative And Quantitative IVs - 22
7/28/2017