SPSS GUIDELINE

FOR FINAL YEAR PROJECT -

RESEARCH SUPERVISION
FACULTY OF BUSINESS, ACCOUNTANCY AND MANAGEMENT
Prepared by :
Dineswary Nadarajan Noor Ain Zeni
 Data Analysis: Step by Step Approach
• 1. Pilot Study – 20-30 data- Reliability Analysis (method 1 only)
• 2. Sampling size – Rules of 5
• 3. Reliability Analysis for all data (method 1 & 2)
• 4. Compute mean for each IVs & DV
• 5. Normality Test (DV only)- outlier
• 6. Linearity Test (Each IVs)-R2 & slope
• 7. Homogeneity Test (Levene Test & Residual vs Predicted scatter plot)
• 8. Independence of error (Durbin- Watson)
• 9. Descriptive Analysis (Tables & pie/bar Charts for demographic questions, line charts for 2nd data)
• 10. Correlation Analysis (2 tailed test, Sig, positive/negative direction of rship, strength – only for DV)
• 11. Regression Analysis (model summary-R2 & Durbin Watson, anova table – Sig & no need F stat,
coefficient table-VIF, B & Beta,)
• 12. Model equation
 Data Entry (Variable View Tab)
Step 1: Coding all the answers of the questionnaire with numbers
Step 2: Defining each variable in the variable view tab in SPSS
Eg : Demographic variables, Independent variables, Dependent variables
• Name : Name each question by section (unique, no space or symbol)
• Type : Numeric
• Width, decimal, missing value and align: Default
• Label : The question labels
• Value : Meaning of each values
• Measure :
• Nominal : Categorical Data
• Exp: Gender, Marriage status, Nationality
• Ordinal : Order (Rank)
• Exp : Age, Winning position, Education level
• Scale : Likert Scale & Figures answer
• Exp: Strongly agree, agree, neutral, disagree and strongly disagree, weight in kg, height in metres.
Variable View in SPSS
Data View in SPSS
 Sample Size

Rule of 5
• Total items in questionnaire x 5= p
• Sample size > p
 Reliability Analysis
Reliability analysis using ‘Cronbach’s alpha (α)’
• Expressed on scale of 0–1
• Where 1 is most reliable outcome
• Optimum reliability usually at least 0.70
• More than 0.70 consider as the data is reliable

SPSS Instruction :
Analyze – Scale – Reliability Analysis – Move all questions into the item
box (except demographic questions)– Click OK.
 Two Step to analyze reliability
• Step 1– overall data reliability
Analyze all the variable in the questionnaire (excluding
demographic factor)
• Step 2– to justify the unreliability of the data
Independent variable (Each variable Separately)
Dependent variable (Separately)

Since the value is more than 0.7

Then the data is reliable for further
analysis.
 Research Tips:
• In the pilot study, the reliability analysis is being done to justify the
efficiency and consistency of the data from the questionnaires.

• If the Cronbach Alpha value falls below 0.7, Method Two must be executed
to identify which variable that is not reliable for questions’
amendment(possible redundancy, leading etc) and recollecting new data.

• Hence, when the Cronbach Alpha value reach above 0.7, a thorough and
complete data collection for the rest of the sample respondents now has to
be done before further analysis.
 Compute Mean
• Click Transform 
Take Compute
Variable

Step 1:
Name the target
variable as unique as
naming each variable
such in variable view
tab.
Step 2: Counter Clockwise
Selection Panel
 Select StatisticalMean 
click UP arrow button 
select all the questions that
contain in the variable one
by one  click LEFT arrow
button
Eg :
MEAN(B3.1,B3.2,B3.3,B3.4)
 Click OK
Step 3: Repeat this step for all
the independent and
dependent variable.
The new COMPUTED variable
will be in data view set.
 Normal Distribution Analysis
• There are a number of ways that we can do this :
• Statistically
• Significant value of Kolmogorov–Smirnov and Shapiro–Wilk test
• Z-score test of Skewness and Kurtosis (Second approach for normality
assessment)

• Graphically
• Box plot (to exclude the outliers when data is not normal)
• Q-Q plots
• Histogram with normal curve

• We will only examine the normality for the DEPENDENT VARIABLE.

Statistical Approach

Explore the rules of statistical approach for normality test.

• Consider the hypothesis, we want a NON-SIGNIFICANT outcome! This
indicates data is normally distributed.
• Significant is when sig-value is less than 0.05, reject Ho.
• Non-significant is when sig-value is more than 0.05, accept Ho.
Using SPSS for normality
• Select Analyze  Descriptive Statistics  Explore  (in
new window) transfer the ‘COMPUTED MEAN’ of
DEPENDENT VARIABLE into Dependent List box  click on
Statistics  tick Outlier click on Plots  tick Normality
plots with tests, Stem and Leaf, Histogram  click Continue
 click OK

• Kolmogorov–Smirnov Test
- can be used when N > 50
• Shapiro–Wilk Test
- for smaller samples when N < 50
 Output for Normality Test
Kolmogorov–Smirnov Shapiro–Wilk
- should use when is N > 50 - smaller samples when N < 50

Since Sig. value is greater than 0.05,

Then failed to reject Ho.
So, the data is NORMALLY DISTRIBUTED
 Z-score Test for Skewness and Kurtosis
• If one do not passed the normality test using Kolmogorov-Smirnov or
Shapiro-Wilk test (where Sig < 0.05)  Use Z-score!
• Z-score test is dividing the statistic and the standard error of the SKEWNESS
from the descriptives output in SPSS.
• Here are some guidelines for Z-score cut-off points:

Sample size z -score cut-off

< 50 ±1.96
51–100 ± 2 . 58
> 100 ± 3 . 29
Z-score Output – Descriptive table
 Mood Score Statistic Std. Error
Male Skewness .909 .378
Kurtosis .908 .741
Z-scores
Divide the statistics (highlighted in blue) by standard error
(green) of the skewness.
𝑠𝑡𝑎𝑡𝑖𝑠𝑡𝑖𝑐 0.909
𝑧 − 𝑠𝑐𝑜𝑟𝑒 = = = 2.4047
𝑠𝑡𝑑. 𝑒𝑟𝑟𝑜𝑟 0.378
Therefore, Sample size z-score Conclusion
< 50 NOT NORMAL
±1.96
Check outlier
51-100 ±2.58 NORMAL
>100 ±3.29 NORMAL
 Check For The Outliers
• When approaching statistical approaches for normality of data and
results found to be NOT NORMAL, check for the outliers.

• HOW? - Based on the box plot in the output, delete the

respondents/sample number that appear outside of the box plot. This
has to do simultaneously.

• RUN THE NORMALITY ANALYSIS AGAIN. If problem persist,

increase the number of sample size/transform the data.
Box-Whisker Plot

• This are outliers.

• They must be
deleted
simultaneously in
the data view tab.
Graphical Approach for
Normality
• Normal Q-Q plot
• Histogram with normal curve
Normal Q-Q Plot
• Right click - Copy
Comment 1:
R-square = 0.186
means, there is 18.6%
of the total variation
in DV (CS) can be
explained by IV (ivv4).

Comment 2:
Slope = 0.4 means, for
every 1 unit increase
in IV (ivv4), there will
be an increase by 0.4
in DV (CS).
 Homoscedasticity

• There are a number of ways that we can do this :

• Statistically
• Levene Statistic

• Graphically
• Plot Z residual and Z predicted
Levene Statistic
Homogeneity – Levene Statistic
• Significant value > 0.05 – Homogeneity exist
• If Significant value < 0.5 – Homogeneity does not exist

Homogeneity does not exist

Homoscedasticity

• We are hoping that

the data is
distributed
rectangular (-3 to 3)

• No clustering or no
systematic
patterns in plotting
indicates that
variance are
constant.

• Homocedasticity
(homogeneity) exist in
the data model.
 Independence of error

• Independence of error means that the distribution of errors is

random and not influenced by or correlated to the errors in prior
observations. The opposite of independence is
called autocorrelation.

• The independent/autocorrelation is a more quantitative measure

which involves calculating the Durbin-Watson Statistic.

A Durbin-Watson value between 1-3 indicates there is NO

AUTOCORRELATION problem among the residuals (error). Therefore,
the data are Independence of error.
DESCRIPTIVE STATISTICS
 DESCRIPTIVE STATISTICS

Interpretation : The average value (mean) and the

standard deviation value (smallest value is good variable)
CHARTS
Charts
 CORRELATION
• Relationship assessed in terms of a ‘correlation coefficient’
• Positive correlation
• Values change in same direction
• Temperature   sale of ice creams 
• Negative correlation
• Values change in opposite directions
• Temperature   sale of overcoats 
• No relationship between temperature and sale of hamsters
Scatterplots
 Negative correlation –
line slopes downwards from
left to right

 Positive correlation – line

slopes upwards from left to
right

 No correlation – no
pattern to cluster of data

Scale of correlation coefficient Value

0 < 𝒓 ≤ 0.19 Very Low Correlation

0.2 ≤ 𝒓 ≤ 0.39 Low Correlation

0.4 ≤ 𝒓 ≤ 0.59 Moderate Correlation

0.6≤ 𝒓 ≤ 0.79 High Correlation

0.8 ≤ 𝒓 < 1.0 Very High Correlation

Using SPSS for correlation

Select Analyze  Correlate  Bivariate  transfer the all independent

variable and dependent variable to Variable box  click OK

• Use the computed

mean of each
variable (mean of
independent
variable’s and
dependent variable)
Output of Correlation result
 • Sig value should be ≤0.05  Non-Significant because ≥0.05

COMMENTS:
1) SIGNIFICANT CORRELATION!
2) STRENGTH (low/moderate/high) and SIGN (positive or negative relationship)
of the Pearson correlation (r) value.

Example:
There is a SIGNIFICANT relationship between
• Gender inequality and female employee performance
• Sexual inequality and female employee performance
• Inflexible working hours and female employee performance
There is a no SIGNIFICANT relationship between
• Low wages and female employee performance
 Multiple Linear Regression
 Model has several regression lines
 Each with their own gradient
 But ‘constant’ will change each time another predictor added
 Regression equation now a little more complex
 𝒀𝒊 = 𝜷𝟎 + 𝜷𝟏𝑿𝟏 + 𝜷𝟐𝑿𝟐 + ⋯ 𝜷𝒏𝑿𝒏 + 𝜺𝒊
 Gradient for each predictor
 β1X1 for first predictor, β2X2 for second predictor…

• Using the SPSS for regression

• Select Analyze  Regression  Linear…  (in new window) dependent to Dependent  transfer
independent to Independent(s)  click Statistics… (in new window) tick boxes for Estimates,
Model fit, Collinearity diagnostic and durbin-watson  click Continue  click Plots  Move
ZRESID to Y box and ZPRED to X box  tick Histogram and Normal Probability Plot  click
Continue  click OK
Homoscedasticity
Homoscedasticity

• No clustering or no
systematic patterns in
plotting indicates that
variance are constant.

• Homocedasticity
(homogeneity) exist in
the data model.

• Acceptable - Assumption
satisfied!
 Output Interpretation
1)Examine the regression assumptions:
• Normality analysis
• Homocedasticity (scatter plot of residual vs predicted value)
• Durbin – Watson test of autocorrelation(independence of residual)
• Multicollinearity assessment tools (Tol & VIF)

2)Interpret the Multiple Regression Analysis results:

• Model summary table : Interpret R-square
• ANOVA table : Interpret Sig. value
• Coefficient table : Interpret Sig. values and Beta/B coefficients,
R-square – 0.18518.5% of the total variation of dependent variable in this model can be explained
by the independent variables. (Please list down the names according to your variables)

A Durbin-Watson value between 1-3 indicates there is NO AUTOCORRELATION problem among the
residuals (error).

ANOVA table is referred at significant value :

Sig value ≤ 𝟎. 𝟎𝟓 means the model is fit to use for further analysis.
 3) Collinearity (TOL & VIF)
1) Significant 2) Unstandardized & Standardized Coefficients TOL 𝒗𝒂𝒍𝒖𝒆 ≤ 𝟏𝟎
 Beta 𝐯𝐚𝐥𝐮𝐞 (𝐬𝐭𝐚𝐧𝐝𝐚𝐫𝐝𝐢𝐳𝐞𝐝 𝐜𝐨𝐞𝐟𝐟𝐢𝐜𝐢𝐞𝐧𝐭𝐬 − VIF value close to 1
Sig value ≤ 𝟎. 𝟎𝟓 𝐬𝐞𝐜𝐨𝐧𝐝𝐚𝐫𝐲 𝐝𝐚𝐭𝐚)
 1 = not correlated.
(check each variable and if  B value (unstandardized coefficients – primary data)  Between 1 and 5 = moderately
significant reject null  comment on the value and sign; correlated.
hypothesis, accept your  Greater than 5 = highly correlated.
• Highest value means most crucial factor
proposed hypothesis) • Positive : When x increase then y will increase too
then there is no multicollinearity
• Negative : When x increase then y will decrease problem
 Model Equation
• Overall regression model equation :
𝒀𝒊
= −𝟎. 𝟎𝟐𝟒 𝑰𝒏𝒇𝒍𝒆𝒙𝒊𝒃𝒍𝒆 𝒉𝒐𝒖𝒓𝒔 + 𝟎. 𝟑𝟖𝟔 𝒈𝒆𝒏𝒅𝒆𝒓 𝒊𝒏𝒆𝒒𝒖𝒂𝒍𝒊𝒕𝒚
− 𝟎. 𝟏𝟐𝟓 𝒔𝒆𝒙𝒖𝒂𝒍 𝒉𝒂𝒓𝒂𝒔𝒔𝒎𝒆𝒏𝒕) + 𝟎. 𝟏𝟓𝟐(𝒍𝒐𝒘 𝒘𝒂𝒈𝒆𝒔

• Regression model equation with significant predictors :

𝒀𝒊 = 𝟎. 𝟑𝟖𝟔 𝒈𝒆𝒏𝒅𝒆𝒓 𝒊𝒏𝒆𝒒𝒖𝒂𝒍𝒊𝒕𝒚 + 𝟎. 𝟏𝟓𝟐 (𝒍𝒐𝒘 𝒘𝒂𝒈𝒆𝒔)