Multiple regression analysis was conducted to examine the effect of self-perceived

knowledge, accuracy on a knowledge test and genuine knowledge on knowledge

overestimation. The assumptions of regression, including normal residuals, together
with an absence of multicollinearity, and heteroscedacity, were met. Finally, one
outlying case was deemed to have no undue influence on the model. See Table 1 for
a summary of descriptive statistics and relationships between predictors and
outcome variables.

Table 1. Correlation matrix showing relationships between self-perceived knowledge,

accuracy, genuine knowledge and knowledge overestimation (N = 202)

Mean (±) 1. 2. 3. 4.

1. Knowledge overestimation 0.31 (± 0.23) -

2. Self-perceived Knowledge 4.43 (± 1.17) .48** -

3. Accuracy 0.30 (± 0.21) -.67** .03 -

4. Genuine Knowledge 3.69 (± 1.19) -.04 .32** .39** -

** p < .01

The overall model predicted approximately 71% of the variance in participants’

propensity to overestimate knowledge, F(3, 198) = 162.87, p < .001, ΔR2 = .71. All
predictors were significantly associated with the propensity to overestimate
knowledge [in the expected directions*], see Table 2. Overestimating knowledge was
significantly, and positively predicted by self-perceived knowledge (β = .48, p <. 001),
and genuine knowledge (β = .094, p = .033), but negatively predicted by accuracy (β
= -.72, p < .001. The relationship between the strongest predictors, accuracy in
predicting self-perceived knowledge is illustrated in Figure 1.

Table 2. Self-perceived knowledge, accuracy and genuine knowledge as predictors

of propensity to overestimate knowledge.
 Predictor B SE B β p

Self-perceived knowledge .09 .01 .48 < .001

Accuracy -.79 .05 -.72 < .001

Genuine knowledge (FINRA) .02 .01 .09 = .033

Figure 1. Scatterplot of Accuracy as a predictor of Knowledge Overestimation

Results Section SPSS Output and related commentary
sub-section

Introductor Multiple regression analysis was conducted to examine the This is essentially a restatement of your research hypothesis, which should
y effect of self-perceived knowledge, accuracy on a appear at the end of your introduction section.
paragraph knowledge test and genuine knowledge on knowledge The following box indicates in the Predictors in the variables entered cell and the
1 overestimation. outcome variable (or DV) is indicated below the table. Variables are not always
described in a user-friendly way.

Variables Entered/Removeda
Variables
Model Variables Entered Removed Method
1 genuine knowledge, as . Enter
measured by FINRA,
self_perceived_knowledge,
accuracy (average hit rate
for knowledge questions,
minus false alarms for
bogus terms)b
a. Dependent Variable: porportion of items (out of 3) where
participants claimed knowledge for bogus terms
b. All requested variables entered.

Introductor The assumptions of regression, including normal residuals, Plot a and b speak to the normality of residuals. We are looking here to assess if
y together with an absence of multicollinearity, and residuals look like they confirm to a bell shaped curve in a) and whether residual
paragraph heteroscedacity, were met. observations conform to the diagonal line in b).
2

a) b)
Results Section SPSS Output and related commentary
sub-section

Next in the coefficients table, we need to consider the end two columns, so
Tolerance is this > .2, and is VIF < 2, and. If it is, this would support the idea that
there is no issue with multicollinearity.

Heteroscedacity is indicated by a uniform distribution across c). There may be a

minor issue here to the left hand side, but does not look too problematic.

You will need to actively ask for these options when performing the analysis in
Results Section SPSS Output and related commentary
sub-section

SPSS, refer to the lecture slides.

Finally, one outlying case was deemed to have no undue
Introductor This single outlying observation was identified via casewise diagnostics, see
influence on the model. See Table 1 for a summary of
y case 67 in the below. Note that the Std. Residual is slightly more than 3.
descriptive statistics and relationships between predictors
paragraph
and outcome variables.
3

We can then check the Cook’s Distance ‘COO_1’ that we have requested be
added to the data. Problematic items are > 1. This is not the case here.

Both of these ideas support that no observations have undue influence on the
model.

Table 1. Correlation matrix showing relationships between self-perceived knowledge,

Descriptive N here comes from the final column of the
accuracy, genuine knowledge and knowledge overestimation (N = 202)
statistics ‘Descriptive Statistics’ table
table and
The Mean and SD (±) also come from the
correlation
‘Descriptives Table’ in the output You can
matrix
see these reported in the table.

Selecting this option following the step-by- step instructions means there is no
need to run separate descriptives.

Ignore the correlations that are reported in the Regression results, these are one-
Results Section SPSS Output and related commentary
sub-section

tailed and it is recommend that you run two-tailed Pearson correlations. In the
options, request a correlation table that flags significant correlations, and that
report the lower triangle only. Your output will mirror what is above, and is easier
to work with than the format reported as part of the regression. See below and
compare to the table reported:

Repeating guidance from page 1, you should not directly copy these from SPSS.
The figures and tables should be transformed. This will help you, if you do it well,
to demonstrate LO4, amongst others.

Regression The overall model predicted approximately 71% of the First, you want the model summary box from the output, and the ‘Adjusted R
– overall variance in participants’ propensity to overestimate Square’ is the conservative effect estimate of proportion of variance, i.e. 71% that
2
model knowledge, F(3, 198) = 162.87, p < .001, ΔR = .71. All we should report.
predictors were significantly associated with the propensity
to overestimate knowledge [in the expected directions*],
Model Summaryb
Results Section SPSS Output and related commentary
sub-section

see Table 2.

Adjusted R Std. Error of the

Model R R Square Square Estimate Durbin-Watson
a
1 .844 .712 .707 .12565 1.978

a. Predictors: (Constant), genuine knowledge, as measured by FINRA,

self_perceived_knowledge, accuracy (average hit rate for knowledge questions, minus
false alarms for bogus terms)

b. Dependent Variable: porportion of items (out of 3) where participants claimed

knowledge for bogus terms

Next the ANOVA box enables us to report the statistical syntax for the regression
model ANOVA summary, starting with F. Remember that we have already dealt
with the ΔR2. This is repeated

F(3, 198) = 162.87, p < .001, ΔR2 = .71

ANOVAa

Model Sum of Squares df Mean Square F Sig.

1 Regression 7.714 3 2.571 162.867 <.001b

Residual 3.126 198 .016

Total 10.840 201

a. Dependent Variable: porportion of items (out of 3) where participants claimed knowledge for bogus terms

b. Predictors: (Constant), genuine knowledge, as measured by FINRA, self_perceived_knowledge, accuracy

(average hit rate for knowledge questions, minus false alarms for bogus terms)

n.b. the sum of the degrees of freedom i.e. 201, should equal the number of participants entered into the
Results Section SPSS Output and related commentary
sub-section

regression model N = 202, less one. Which is does. You do not need to report this but acts as an additional

check.

Regression All predictors were significantly associated with the We next need to use the coefficients table from the SPSS output.
predictors – propensity to overestimate knowledge [in the expected
Coefficientsa
Alternative directions*], see Table 2. Unstandardized Standardized

reporting Table 2. Self-perceived knowledge, accuracy and genuine knowledge as Coefficients Coefficients Collinearity Statistics

approach predictors of propensity to overestimate knowledge. Std.

1. Model B Error Beta t Sig. Tolerance VIF

Predictor B SE B β p
1 (Constant) .058 .039 1.474 .142

Self-perceived knowledge .09 .01 .48 < .001 self_perceived_knowledge .094 .008 .475 11.732 <.001 .889 1.125

accuracy (average hit rate for -.793 .046 -.724 -17.374 <.001 .838 1.193
Accuracy -.79 .05 -.72 < .001
knowledge questions, minus

Genuine knowledge (FINRA) .02 .01 .09 = .033 false alarms for bogus terms)

genuine knowledge, as .018 .009 .094 2.142 .033 .754 1.326

measured by FINRA

The relationship between the strongest predictors, accuracy a. Dependent Variable: porportion of items (out of 3) where participants claimed knowledge for bogus terms

in predicting self-perceived knowledge is illustrated in Each of the columns, along with the relevant rows are highlighted. Ignore all
Figure 1. other data here, including the constant row.

n.b. in the column to the left I have made the font size smaller, to fit to the format. However, do try
to maintain consistent font sizes.

Also only use [in the expected directions*] if a direction is specified, which is unlikely.

Regression Overestimating knowledge was significantly, and positively The colour coded data here, again can be seen in the ‘Coefficients’ table from the
predictors – predicted by self-perceived knowledge (β = .48, p <. 001), regression output in SPSS.
Results Section SPSS Output and related commentary
sub-section

Alternative and genuine knowledge (β = .094, p = .033), but negatively I have reported the positive predictors first, followed by the negative, as it allows
reporting predicted by accuracy (β = -.72, p < .001. The relationship as more concise narrative.
approach between the strongest predictors, accuracy in predicting
If there are non-significant predictor (where the sig value in the relevant predictor
1. self-perceived knowledge is illustrated in Figure 1.
row is non significant, you could report x, y and z were not-significant preditors of
a, but then also include the β and p values as described.

If using the tabular format you might preface the table by saying only x and y
were significant predictors of a, but y demonstrated no significant association.

Figure 1 Figure 1. Scatterplot of Accuracy as a predictor of Knowledge Typically regression results sections in published works do not include a figure,
Overestimation
indicating the assocaitions. For the purpose of the assessment, including a figure
of the strongest association (or predictor) in the regression model would be a
useful demonstration of your understanding. This is something you will need to
run separately, please see the Unit 6 Prepare (Part II) content for more
infomation.

Accuracy is the strongest association in this regression model, despite it being a

negative predictor, as the absolute value i.e. ignoring the minus sign the
predictors in order of standardized beta weight magnitude are -.72, .48 and .09,
as can be seen in the two rows above.