You are on page 1of 22

Analysis NOTES for SPSS

No Type of Analysis Diploma Degree Masters PhD


1. Descriptive Statistics (Nominal / ü ü ü ü
Categorical Data)
2. Descriptive Statistics (Survey x x ü ü
Items)
3. Data Screening and Cleaning x x ü ü
(Data Outliers)
4. Normality Analysis x x ü ü
5. Reliability Analysis x ü ü ü
6. Validity Analysis x ü ü ü
(Optional)
7. Multicollinearity Analysis x x ü ü
8. Cross-Tabulation Analysis ü ü ü ü
(Optional (Optional)
)
9. Correlation Analysis x ü ü ü
10. Number of Continuous Variable 1-2 3-4 5-7 8-15
(IV) / Hypothesis
11. Multiple Regression Analysis x X ü ü
(Optional
)
12. Mediation / Moderation Analysis x x ü ü
(Optional
)

1. DESCRIPTIVE STATISTICS (for nominal data)

a. Basic Reporting

Table 1 : Demographic Profile of the Respondents (n=150)


Profile Frequency Percentage (%)
Gender
Male 82 54.7
Female 68 45.3

Age
Between 20-30 32 21.3
Between 31-40 45 30.0
Between 41-50 37 24.7
50 and above 36 24.0

Qualification
Diploma 11 7.3
Degree 17 11.3
Masters 39 26.0
Doctorate 50 33.3
Others 33 22.0

Sample Explanation
Table 1 shows the descriptive profile of the respondents. A total of 150 respondents
participated in this research. Researchers found that males are the majority of respondents
which consists of 83 (54.7%) while females are 68 (45.3%). While the age group between 31-
40 represents the largest group in this group which is 45 respondents (30%) followed by the
age group between 41-50 which is 37 respondents (24.7%) and the age group 50 and above
which is 36 respondents (24%) ) followed by the age group between 20-30 as many as 32
respondents (21.3%). The age group between 20-30 is the smallest group. In terms of
Qualification, the group with a Doctorate qualification is the largest group found to
participate in this research which is 50 respondents (33.3%) followed by Masters which is 39
respondents (26.0%), first degree 17 respondents (11.3%), other qualifications 33 respondents
(22%) and the diploma of 11 respondents (7.3%). Diploma -educated respondents are the
least group in this research. The ‘others’ category are those with education with a
professional qualification background (5 respondents), vocational (12 respondents) and
certificate level (16 respondents) respectively.

b. Advanced Reporting (Nominal Data)


Table 1 : Demographic Profile of the Respondents (n=150)
Profile Frequency Percentage Min Max Mean Standard
(%) (M) Deviation
(SD)
Gender 1 2 1.45 .45
Male 82 54.7
Female 68 45.3

Age 1 4 2.51 1.10


Between 20-30 32 21.3
Between 31-40 45 30.0
Between 41-50 37 24.7
50 and above 36 24.0

Qualification 1 5 3.51 1.17


Diploma 11 7.3
Degree 17 11.3
Masters 39 26.0
Doctorate 50 33.3
Others 33 22.0

Students’ Notes
Students must explain the basic profile of the respondents and interpret Mean and Standard
Deviation of the nominal data.

Min and Max represent the number of categorical data that you set in the questionnaires.
The mean (average) of a data set is found by adding all numbers in the data set and then
dividing by the number of values in the set.
Standard deviation is a mathematical tool to help you to assess how far the values are
spread above and below the mean. A high standard deviation (more than 1) shows that the
data is widely spread (less reliable) and a low standard deviation (close to zero, not more
than 1) shows that the data are clustered closely around the mean (more reliable).
-1 Min 1
0 Gender Age / Qualification

Strongly Disagree Neutral Agree Strongly


Disagree Agree
1 2 3 4 5

Standard Deviation (SD) in table 1 explain the average amount of variability in your data set.
It tells you, on average, how far each score lies from the mean. If you look at gender, the data
is skewed to one site but still reliable as it is closed to Zero but not more than 1. Whereas, the
data for age and qualification is less reliable because more than 1. Since it is less reliable, we
will use the non-parametric test to access Age and Qualification only.

c. Advanced Reporting (Continuous Data)

Table 2 : Descriptive Statistics of Variables


Variables
N Minimum Maximum Mean Std. Deviation

Technology Acceptance 150 2.33 5.00 4.32 .63


Perceived Confidence 150 2.00 5.00 3.71 .65
Cost Benefits 150 1.00 5.00 2.33 .88
Facilitation 150 2.00 5.00 4.10 .65

Sample Explanation
Table 2 shows the minimum and maximum scores for each continuous variable. Mean
technology acceptance score is the highest which is M = 4.32, SD = .63, followed by
facilitation M = 4.10, SD = .65, perceived confidence M = 3.71, SD = .65 and the lowest is
Cost Benefits M = 2.33 , SD = 88. All standard deviation results show that it is close to the
mean score. This gives an indication that all the variables measured have small distributions.
The mean data for cost benefit indicates that most respondents tend to disagree that cost
benefit is the cause of Technology Acceptance, while the mean facilitation indicates that
facilitation is the main cause of technology acceptance.

2. ASSESSING NORMALITY AND VALIDITY

A. Normality
Assessing the normality of residuals is important in multiple regression analysis for several
reasons:

1. Assumption of ordinary least squares (OLS): OLS regression assumes that the residuals
(i.e., the differences between the observed values and the predicted values) are normally
distributed. Violation of this assumption can lead to biased parameter estimates and
inaccurate hypothesis testing.

2. Validity of hypothesis tests and confidence intervals: Hypothesis tests and confidence
intervals are based on assumptions of normality. If the residuals are not normally distributed,
the p-values and confidence intervals may be inaccurate, leading to incorrect conclusions.

3. Influential outliers: Non-normality in the residuals may indicate the presence of influential
outliers. These outliers can have a disproportionate impact on estimated regression
coefficients, potentially distorting the results. Identifying and addressing influential outliers is
essential for obtaining reliable estimates.

4. Model fit and predictive accuracy: The assumption of normality is also related to model fit
and predictive accuracy. If the residuals are not normally distributed, it may suggest that the
underlying model does not capture the true data-generating process adequately. Additionally,
non-normal residuals can affect the accuracy of predicted values and may lead to biased
predictions.

To assess normality, you can use diagnostic plots such as a histogram or a Q-Q plot to
visually inspect the distribution of the residuals. Alternatively, statistical tests, such as the
Shapiro-Wilk test or the Kolmogorov-Smirnov test, can be used to formally test for
normality. If the residuals deviate substantially from normality, you may need to consider
transformations of the variables or consider alternative regression methods such as robust
regression that can handle non-normal residuals.

Step-by-Step Normality Analysis for continuous variables


Quick Steps
1. Click Analyze -> Descriptive Statistics -> Explore…
2. Move the variable of interest from the left box into the Dependent List box on the right.
3. Click the Plots button, and tick the Normality plots with tests option. Check also data
outliers.
4. Click Continue, and then click OK.
5. Your result will pop up – check out the Tests of Normality section.
Data cleaning and outliers
When you do the normality test, the data will show outliers, if any. The simplest yet most
effective step is to delete the item/s if outliers exist. If few outliers exist for any one variable,
make sure you delete the outlier items simultaneously.

Data outliers will look like below :-

Sample Explanation if using Shapiro-Wilk and Kolmogorov-Smirnov

a. Advanced Reporting
Based on the analysis, the data is not significant p>.05 (n=150). Therefore, researcher
conclude that the data is normally distributed.

b. Advanced Reporting
The data in normality test using Shapiro-Wilk and Kolmogorov-Smirnov fulfil the basic
assumption of performing parametric tests after the outliers are detected and removed. Based
on the analysis, the data is not significant p>.05 (n=150). Therefore, researcher conclude that
the data is normally distributed.
Step-by-Step Validity Analysis
To perform a validity analysis in SPSS 16, you can follow these step-by-step instructions:

Get the initial idea first from this video :


(2) Reliability and Validity (Convergent, Discriminant)
Measures - YouTube

1. Start SPSS and open your dataset: Launch SPSS 16 and open the dataset that contains the
variables you want to analyze for validity.

2. Determine the variables to analyze: Identify the variables or items you want to assess for
validity. These could be Likert-scale items, continuous variables, or any other measurement
scales.

3. Assess internal consistency using Cronbach's alpha: If you are analyzing a scale or a set of
related items, you can assess internal consistency using Cronbach's alpha. Go to "Analyze" >
"Scale" > "Reliability Analysis". In the "Reliability Analysis" window, select the items you
want to analyze and move them into the "Items" box. Choose the appropriate settings, such as
the desired method for handling missing data, and click "OK". SPSS will generate the
reliability statistics, including Cronbach's alpha, in the output window.

4. Conduct exploratory factor analysis (EFA): If you want to analyze the underlying factor
structure of your items, you can perform exploratory factor analysis. Go to "Analyze" >
"Data Reduction" > "Factor". In the "Factor" window, select the items you want to analyze
and move them into the "Variables" box. Choose the appropriate settings, such as the
extraction method (e.g., principal components analysis) and rotation method (e.g., varimax),
and click "OK". SPSS will generate the factor analysis output, including factor loadings and
eigenvalues.

For details step-by step and result interpretation for KMO and Bartlett’s Test, refer to
Factor Analysis in SPSS - Reporting and Interpreting Results
(onlinespss.com)

5. Assess convergent (p<0.05) and discriminant validity (between 0.3-0.7 or 0.3 and above):
To assess the convergent validity of your scale, examine the factor loadings of each item on
its corresponding factor in the factor analysis output. Generally, factor loadings above 0.30
are considered acceptable. To assess discriminant validity, compare the item loadings across
factors and ensure they are significantly higher on their own factor compared to other factors.

Use this link to assess validity on Instrument (Convergent and divergent validity) using
Bivariate: How to Test Validity questionnaire Using SPSS - SPSS
Tests; SPSS 04: Convergent and Discriminant Validity of Scale
- YouTube; How to analyze discriminant and convergent validity
of Likert scales - YouTube

6. Calculate scale scores: If you have determined that your items are valid and reliable, you
can calculate scale scores. To create composite scores for each scale, you can use SPSS
syntax or formulas. For example, you can create a new variable representing the sum or
average of individual items for each scale. To do this, go to "Transform" > "Compute
Variable". In the "Compute Variable" window, provide a name for the new variable, enter the
formula (e.g., SUM (item1 to item5) or MEAN (item1 to item5), and click "OK".

7. Perform further analyses: Once you have established the validity of your variables and
calculated scale scores, you can use them in other statistical analyses, such as regression,
ANOVA, or correlation, depending on your research question and objectives.

It is important to interpret the validity analysis results in the context of your research and
consider any specific requirements or guidelines for your study. If you need further guidance
or assistance, consult SPSS 16 Help menus or refer to statistical textbooks for more detailed
instructions and explanations

3. RELIABILITY ANALYSIS
a. Basic Reporting
Table 4 : Cronbach's Alpha
Subscale No of Item Cronbach's Alpha
Technology Acceptance 3 .829
Technology Reliability 3 .762
Perceived Confidence 5 .752
Cost Benefits 3 .840
Facilitation 5 .876

Sample Explanation
According to Hair (2016), a general accepted rule is that, Cronbach's Alpha of 0.6-0.7
indicates an acceptable level of reliability, and 0.8 or greater a very good level. Based on the
presented result in Table 1, the technology acceptance, technology reliability, perceived
confidence, cost benefits and facilitations have a good internal consistency, with a Cronbach
alpha coefficient reported of .829, .762, .752, .840 and .870 respectively. The results indicates
that all items used in the survey questionnaire are reliable.
4. CROSS-TABULATION ANALYSIS
(Non-parametric test – if the distribution of data is not normal)
a.) Basic Reporting
Table 4 (a) : GENDER * QUALIFICATION Cross-tabulation
QUALIFICATION

Diploma Degree Masters Doctorate Others Total

GENDER Male Count 6 11 16 30 19 82

% within GENDER 7.3% 13.4% 19.5% 36.6% 23.2% 100.0%

% within QUALIFICATION 54.5% 64.7% 41.0% 60.0% 57.6% 54.7%

% of Total 4.0% 7.3% 10.7% 20.0% 12.7% 54.7%

Female Count 5 6 23 20 14 68

% within GENDER 7.4% 8.8% 33.8% 29.4% 20.6% 100.0%

% within QUALIFICATION 45.5% 35.3% 59.0% 40.0% 42.4% 45.3%

% of Total 3.3% 4.0% 15.3% 13.3% 9.3% 45.3%

Total Count 11 17 39 50 33 150

% within GENDER 7.3% 11.3% 26.0% 33.3% 22.0% 100.0%

% within QUALIFICATION 100.0% 100.0% 100.0% 100.0% 100.0% 100.0%

% of Total 7.3% 11.3% 26.0% 33.3% 22.0% 100.0%

Sample Explanation
You also need to explain Table 4(a) accordingly. You can support your explanation using
graph which can also be set in SPSS.

b. Advanced Reporting
Table 4(b) Chi-Square Tests
Pearson Chi-Square (Gender) Asymp. Sig df
Technology Acceptance .403 8
Qualification .366 4
a. 1 cells (10.0%) have expected count less than 5. The minimum expected count is 4.99.

Table 4(b) indicates the results of the square test (Pearson Chi Square) in the table above (c₂
= .403, df = 4, p> .05) showed no difference between Gender and Technology Acceptance
and between gender and qualification (c₂ = .366, df = 4, p> .05). This means that the
differences between the genders of men and women do not directly affect the qualification.
Students’ Notes
c₂ à This symbol is used for Chi-Square
5. CORRELATION ANALYSIS

Table 4.1 : Linear Relationship


Correlations
Perceived
Facilitation Confidence Cost Benefits
Technology Pearson’s r .469** .425** -.274**
Acceptance p-value .000 .000 .001
**. Correlation is significant at the 0.01 level (2-tailed).

Sample Explanation
a. Basic Reporting - Bivariate Correlation (Pearson Correlation)

Results of the Pearson correlation indicates that there is a moderate significant relationship
between technology acceptance and facilitation r=.469, p<.05 (n=150). It is also found that
the relationship between technology acceptance and perceived confidence is moderately
significant r=.425, p<.05 (n=150). However, it is found that there is a weak negative
relationship between technology acceptance and cost benefits r = -.247, p<.05 (n=150) which
indicate that an increase in the cost benefits will decrease technology acceptance.
b. Advanced Reporting
Correlation analysis is used to describe the strength and direction of the linear relationship
between two variables. Cohen (1988) suggests the following guidelines to determine the
strength of the relationship between variables as follows:

r=.10 to .29 or r=–.10 to –.29 small/weak


r=.30 to .49 or r=–.30 to –.4.9 medium/moderate
r=.50 to 1.0 or r=–.50 to –1.0 large.

Results of the Pearson correlation indicates that there is a moderate significant relationship
between technology acceptance and facilitation r=.469, p<.05 (n=150). It is also found that
the relationship between technology acceptance and perceived confidence is moderately
significant r=.425, p<.05 (n=150). However, it is found that there is a weak negative
relationship between technology acceptance and cost benefits r = -.247, p<.05 (n=150) which
indicate that an increase in technology acceptance will decrease the cost benefits.

6. MULTIPLE REGRESSION ANALYSIS

Students’ Notes
Below are the table generated from SPSS. You need to transform the table into more precise
data to explain the statistical results.

Correlations

TECHACCEPT PERCON COSTBEN FACL

Pearson Correlation TECHACCEPT 1.000 .425 -.274 .469

PERCON .425 1.000 -.406 .290

COSTBEN -.274 -.406 1.000 -.373

FACL .469 .290 -.373 1.000

Sig. (1-tailed) TECHACCEPT . .000 .000 .000

PERCON .000 . .000 .000

COSTBEN .000 .000 . .000

FACL .000 .000 .000 .

N TECHACCEPT 150 150 150 150

PERCON 150 150 150 150

COSTBEN 150 150 150 150

FACL 150 150 150 150


Coefficientsa

Standardize

Unstandardized d 95% Confidence Collinearity

Coefficients Coefficients Interval for B Correlations Statistics

Std. Lower Upper Zero-

Model B Error Beta t Sig. Bound Bound order Partial Part Tolerance VIF

1 (Constant) 1.697 .453 3.747 .000 .802 2.592

PERCON .305 .074 .313 4.116 .000 .159 .451 .425 .322 .283 .813 1.230

COSTBEN -.005 .057 -.007 -.086 .931 -.117 .107 -.274 -.007 -.006 .764 1.309

FACL .366 .073 .376 5.012 .000 .222 .510 .469 .383 .344 .838 1.194

a. Dependent Variable:

TECHACCEPT

he effect of
les (IV) on
(DV).
significant
and DV.

ions show the


chnology
st Benefit
tionship with

e larger than
en.

Model Summaryb

Change Statistics

R Adjusted R Std. Error of R Square F Sig. F


Model R Square Square the Estimate Change Change df1 df2 Change

1 .558a .312 .297 .53222 .312 22.030 3 146 .000

a. Predictors: (Constant), FACL, PERCON,


COSTBEN

b. Dependent Variable: TECHACCEPT

on
8 indicates a
relation

.7% of the variance in


e predicted from
ANOVAb

Model Sum of Squares df Mean Square F Sig.

1 Regression 18.720 3 6.240 22.030 .000a

Residual 41.355 146 .283

Total 60.075 149

a. Predictors: (Constant), FACL, PERCON, COSTBEN

b. Dependent Variable: TECHACCEPT

The P-value <0.05, indicates the equation is a good fit


(F(3, 146)= 22.03, p=0.000)

Student Notes
The above tables are the data copied from SPSS output. We must choose the number that we
need only to report on the thesis/assignment. Based on the report above, we need to follow
the steps as listed below.
Step 1 : Report on Multicollinearity
Step 2 : Evaluating the Model including normality data using PP Plot
In multiple regression analysis, PP Plot can also tell you whether your data is normal or not.
A normal distribution of data indicates your model is fit.

Why is Model Fitting Important? As previously mentioned, a well-fit model does not match
every data point given but follows the overall curves. This shows that the model is neither
underfit nor overfit. If a model is not properly fit, it will produce incorrect insights and should
not be used for making decisions.

Step 3 : Summary of the Result


Sample Reporting / Explanation (Only advance explanation needed)
a.) Report on Multicollinearity

Table 4.1 Multicollinearity


Model Pearson Tolerance VIF
Correlation Value
Perceived
.425 .813 1.230
Confidence
Cost Benefits -.274 .764 1.309
Facilitation .469 .838 1.194
Dependent Variable: Technology Acceptance
Test for multicollinearity is conducted to check the relationship among independent variables.
According to Pallant (2016), multicollinearity exists if the independent variables are highly
correlated or when r=.9 and above, which indicates problems in statistical significance of an
independent variables and the data may not be reliable. Based on the statistical result,
Perceived Confidence, Cost Benefits and Facilitation correlated with Technology Acceptance
(.425, -.274 and .469 respectively). Whereas, the coefficient correlation also shows that all
variables are less that .7. Tabachnick and Fidell (2001, p. 84) suggest to the researcher to
think carefully before including two variables with a bivariate correlation of .7 or more in the
same analysis. In this case, therefore, all variables will be retained. VIF (Variance inflation
factor) also shows all variables are below 10 which indicates no multicollinearity problem
exist. Tolerance value (.813, .764 & .838) is larger than 1-R 2 (1-0.28 = 0.72) which confirms
there is no multicollinearity problem in the model.
Notes to student
1. Tolerance value should be larger than 1-R2.
2. R2 values is depicted from Adjusted R Square in the Model Summary Table.

b.) Model Evaluation (Step by Step)

Notes to student

A linear regression analysis is used to predict changes in the dependent variable based on the
value of independent variables. Below is the standard APA format of reporting for linear
regression.
A linear regression was calculated to predict [dependent variable] based on [independent
variable] . A significant regression equation was found (F(_,__)= __.___, p < .___).

(F(Regression df value, Residual df value )= F Value, p < ANOVA Significance Value)

Sample Advanced Explanation


A simple linear regression was calculated to predict Technology Acceptance based on
Perceived Confidence, Cost Benefits and Facilitation. Multiple correlation coefficient,
R=.558 indicates a high degree of correlation. Adjusted R2 indicates that 29.7% of the
variance in the dependent variable can be predicted from the independent variable. A
significant regression equation was found (F(3, 146)= 22.03, p=0.000). The p-value or
p<0.05, indicates the equation is a good fit.
c.) Summary of the Result
Table 4.2 summarizes the hypothesis that test the relationship between independent variable
and dependent variable as follows :-
Table 4.2 : Result of the Hypothesis Testing
No Hypothesis Testing Result
H₁ There is a relationship between Perceived Confidence and Supported
Technology Acceptance
H₂ There is a relationship between Cost Benefits and Technology Not Supported
Acceptance
H₃ There is a relationship between Facilitation and Technology Supported
Acceptance
6. MULTIPLE REGRESSION ANALYSIS WITH MEDIATING VARIABLES

For further learning, you can also refer to V14.16 - Testing Mediation in SPSS - Example 1 -
YouTube and Introduction to Mediation Analysis | University of Virginia Library Research
Data Services + Sciences.
M

a b

X c Y

Step 1 : Estimate the total effect between X (IV) and Y (DV) using regression;
Step 2 : Estimate direct effect between X and M using bivariate regressions;
Step 3a: Estimate direct effect between X and Y using multiple regression with X and
M as predictors and Y as dependent variable.
Step 3b: Estimate direct effect between M and Y using multiple regression with X and
M as predictors and Y as dependent variable. This test is done to estimate co-
efficient and standard errors to estimate indirect effect for statistical
significance.
Step 4 : Estimate and test indirect effect for statistical significance using SOBEL test.

Based on data provided, a researcher hypothesizes that the effect between Self-Efficacy and
Intention may be mediated by Level of Integrity.

Hence;
X = Self Efficacy
Y = Intention to Engage
M = Level of Integrity
Level of Integrity

a b

X c Y

Step 5 : Estimate direct effect = c Intention


Self-efficacy

Estimate indirect effect = a*b


Report the mediation (supported or not supported)

Using SPSS, estimate the effect;


Step 1 : Estimate Total Effect Using Regression

X Y

1. In SPSS, choose Analyzeà Regression à Linear


2. Select Self-Efficacy into IV and Intention into DV box
3. Select Statistic, check box of Estimate, Confidence Interval, Covariance Matrix,
Model Fit, R squared Change
4. Click OK. Observe Coefficient Table.

Coefficientsa

Unstandardized Standardized 95% Confidence Interval


Coefficients Coefficients for B

Lower Upper
Model B Std. Error Beta t Sig. Bound Bound

1 (Constant) .260 .231 1.123 .266 -.203 .723

MEANSelfEfficacy .945 .055 .910 17.228 .000 .835 1.055

a. Dependent Variable: MEANIntention

Statistically Significant, therefore we


can proceed with further analysis

Is b1 significant? We want X to affect Y. If there is no relationship between X and Y,


there is nothing to mediate.

Step 2 : Direct Effect X and M (this is not our interest)


Level of Integrity
M

Self-efficacy

1. In SPSS, choose Analyzeà Regression à Linear


2. Select Self-Efficacy into IV and Level of Integrity into DV box
3. Select Statistic, check box of Estimate, Confidence Interval, Covariance Matrix,
Model Fit, R squared Change
4. Click OK. Observe Coefficient Table.
Coefficientsa

Unstandardized Standardized 95% Confidence Interval


Coefficients Coefficients for B

Lower Upper
Model B Std. Error Beta t Sig. Bound Bound

1 (Constant) .787 .346 2.276 .026 .096 1.478

MEANSelfEfficacy .737 .082 .752 8.985 .000 .573 .901

a. Dependent Variable: MEANIntegrity

Statistically Significant, therefore we can


proceed with further analysis

Mediation result

a=.737 M
(0.082)
b

c
X Y

Step 3 : Direct Effect X and M and Y

1. In SPSS, choose Analyzeà Regression à Linear


2. Select Self-Efficacy (IV) into IV and change Level of Integrity (M) into IV box and
select Intention (DV) (Y) into dependent box.
3. Select Statistic, check box of Estimate, Confidence Interval, Covariance Matrix,
Model Fit, R squared Change
4. Click OK. Observe Coefficient Table.

Coefficientsa

Unstandardized Standardized 95% Confidence Interval


Coefficients Coefficients for B

Lower Upper
Model B Std. Error Beta t Sig. Bound Bound

1 (Constant) .081 .228 .356 .723 -.376 .538

MEANSelfEfficacy .778 .079 .749 9.854 .000 .620 .936

MEANIntegrity .227 .081 .214 2.816 .007 .066 .388

a. Dependent Variable: MEANIntention


Level of Integrity

a=.737 M
(.082)
b=.227
(.081)

X Y
c =.778
(.079)
Self-efficacy (IV) Intention (DV)

So, now we have all the figure to estimate indirect effect à a*b=Indirect effect. We will test
indirect effect for statistically significant.

It is also observed that direct effect between Self-Efficacy is statistically significant with
Integrity where p<0.05.

Step 4 : Use SOBEL test in order to test indirect effect for statistical significant.

1. Google SOBEL Test Calculator; Interactive Mediation Tests (quantpsy.org)


2. Insert figure a (.737) and b (.227) in the box and standard error for a (.082) and
standard error for b (.081).
3. Press Calculate.

4. Observe Sobel Test result. We look at p-value for indirect effect is 0.00746341 or
p<0.05. Result of Sobel Test shows indirect effect of M (Integrity) is statistically
significant. Therefore, we conclude that, Integrity (MV) indirectly mediate the
relationship between Self-Efficacy (IV) and Intention (DV).
5. Summary of the Statistical Result – This table should be presented all in the last
subheading for all variables.

Table 4.3 : Result of the Hypothesis Testing


No Hypothesis Testing Result
H₁ There is a relationship between Self Efficacy (IV) and Supported
Intention (DV)
H₂ Integrity mediates the relationship between Self-Efficacy and Supported
Intention.
Sample Advanced Reporting for Mediation Analysis using Baron & Kenny (1986)
approach

Level of Integrity

a=.737 M
(.082)
b=.227
(.081)

X Y
c =.778
(.079)
Self-efficacy Intention

Steps
i. Observe the output table from SPSS :-
ii. Direct X àY (Model 1)
iii. Indirect Effect X àMàY (Model 2)
iv. Get all the figures in the table
v. Explain in APA format

You can refer to this picture for guide (enlarge it for clearer view)):-
How to report multiple regression with mediating variables? Please refer to Table 1 for
guidance. You also must open the SPSS output for figure reference (File name - Sample
Data with Mediation uum.spv).

Table 1 : Regression Analysis for Mediation of Level of Integrity between Self-Efficacy and
Intention.

Variable B 95%CI SE B BETA R2 Δ R2


(β)
Model 1 .83 .83***
Constant .260 [-.20, .72] .23
Self-Efficacy (IV) .945 [.84,1.01] .06 .91**
Model 2 .85 .85***
Constant .081 [-.38,.54] .228
Self-Efficacy (IV) .778 [.62, .94] .079 .75
Level of Integrity (M) .227 [.07,.39] .214 .21
Note : CI = Confidence Interval
***p<.001

Table 1 shows the impact of Self-Efficacy and Level of Integrity on Intention for
academicians in Private Universities in Klang Valley. In model 1, the R2 value of .83 reveals
that the Self-Efficacy explained 83% variance of intention with F(1,62)=296.8, p<0.001). The
findings revealed that Self-Efficacy positively predicted intention (β = .91, p<001. In model
2, the R2 of .85 reveals that the Self-Efficacy and Level of Integrity explained 85% variance
in the Intention with F (2, 61) = 169, p<0.001). The finding reveals that self-efficacy (β = .75,
p<001) and Level of Integrity positively predicted Intention (β = .21). The ΔR2 value of .85
reveals 85% in the variance of model 1 and model 2 with ΔF (2, 61)=168.9, p<0.001. The
regression weight for self-efficacy subsequently reduced from mode 1 to model 2 (.91 to .75)
but still remained significant which confirmed the partial mediation. More specifically, Self-
Efficacy has direct as well as indirect effect on intention.

Notes for students on the symbols used


1. Uppercase delta (Δ) at most times means “change” or “the change” in maths.
2. B & β - In regression, the difference between B (or b) and β relates to whether the
coefficients are standardised or unstandardised. B (or b) generally refers to the
unstandardised coefficient. This means that the regression coefficient in in the
original measurement units.
For example, imaging I am trying to predict a final exam score based on the number
of hours spent studying. If I get a B = 2, this tells me that for every of 1 hours
study time, I predict an increase in the final exam score of 2. This relationship is in
the original units (hours studying, and exam score). This is useful for predicting
things in the real world, but it is difficult to compare different predictors. Predictors
might have large B values just because they are measured on a larger scale
(compare minutes to hours in the above example).

Standardised regression coefficients do a similar thing, but in a standardised way.


The β refers to the number of standard deviation changes we would expect in the
outcome variable for a 1 standard deviation change in the predictor variable. For
example, if I got β= .5 for hours of study, this would tell me that for every 1 standard
deviation increase in hours of study, I can expect a .5 standard deviation increase in
the exam score. Because this is standardised, β make it easier to compare different
predictors to see which is more important.

You might also like