1 Descriptive analysis – demographics and variables
Tải bản đầy đủ
28
by less than VND 5M (39.7%), VND 10M – 15M (15.5%), and final is more than VND 15M
(3.7%).
Table 4.1 Respondents’ characteristics
Demographic profile
Age
Gender
Education
Income
Category
From 18 to 25
From 26 to 35
From 36 to 45
Above 45
Total
Male
Female
Total
High school
Bachelor
Graduate degree
Total
Less than 5M
From 5M to
10M
From 10M to
15M
More than 15M
Total
Frequency
106
85
21
7
219
57
162
219
26
172
21
219
87
Percentage (%)
48.4
38.8
9.6
3.2
100.0
26.0
74.0
100.0
11.9
78.5
9.6
100.0
39.7
90
41.1
34
15.5
8
219
3.7
100
4.2 Reliability analysis
It is required to test the Cronbach‟s Alpha of scales for each construct. The
Cronbach‟s Alpha helps remove unstandardized scale or unsuitable scales. Normally, if the
correlation of each specific item with total of the other items in the scale (Corrected ItemTotal Correlation) is moderately high or higher above 0.3 (Burnstein & Nunnally, as cited in
Nguyen, 2011), the item is probably at least moderately correlated with most of the other
item and will make a good component of this summated rating rate. Conversely, if the item –
total correlation is less than 0.3 at any items, such items need to be deleted and it is required
to examine the items for wording problems and conceptual fit.
29
After considering the result of the first reliability testing on all proposed factors, all
variables are accepted and keep in official questionnaire, because they had the Item-total
correlation more than 0.3 and the Cronbach‟s Alpha value of all constructs were acceptable
(greater than 0.6) according to Table 4.2
Table 4.2 Reliability analysis result
Observed
variable
Scale Mean
if Item
Deleted
Scale
Variance if
Item Deleted
Website social presence (WSP) Alpha = 0.777
WSP1
10.4886
5.664
WSP2
10.1826
6.022
WSP3
10.621
5.512
WSP4
10.4612
5.892
Trust (T) Alpha=0.816
T1
10.1005
5.595
T2
10.0183
5.137
T3
9.9315
5.358
T4
9.9635
4.999
Product attitude (PA) Alpha = 0.803
PA 1
6.8128
3.043
PA 2
6.6575
2.777
PA 3
6.7215
2.826
Involvement (I) Alpha = 0.782
I1
9.8904
5.538
I2
9.3425
6.18
I3
9.7717
5.828
I4
9.8995
6.155
Purchase intent (PI) Alpha = 0.741
PI1
10.9406
5.451
PI2
10.8676
4.987
PI3
10.8493
4.853
PI4
10.6986
5.276
Corrected
Item-Total
Correlation
Cronbach's
Alpha if Item
Deleted
0.608
0.616
0.552
0.561
0.710
0.710
0.743
0.734
0.551
0.697
0.621
0.677
0.807
0.740
0.775
0.748
0.598
0.705
0.649
0.784
0.673
0.733
0.589
0.619
0.623
0.529
0.730
0.716
0.710
0.758
0.404
0.545
0.625
0.579
0.757
0.675
0.629
0.660
Summary, as shown in Table 4.2, the results indicated significantly high or very high
internal reliability for most tested item scales including website social presence, trust,
30
involvement, product attitude, and purchase intent with the value of Cronbach‟s Alpha
around 0.8.
The next step, the Exploratory analysis factor was conducted to test the validity of
measurement scales.
4.3 Exploratory factor analysis:
For this study, Exploratory factor analysis (EFA) is conducted to analyze validity. The
main aim of EFA was to investigate a large number of relationships among interval variables
(Leech et al., 2005).
The Rotated Factor Matrix displays the items and factor loading for rotated factors.
For the first EFA analysis, the result in Rotated Component Matrix show the item PI1
(Purchase intent 1) had loading plot lower than 0.5. In this case, the author decided to delete
this item. (See Table 4.3).
Table 4.3 Rotated component matrix (time 1)
Rotated Component Matrixa
Component
1
T2
.787
T3
.763
T4
.757
T1
.713
2
I4
.782
I3
.703
I2
.673
I1
.632
3
PI4
.830
PI3
.699
PI2
.658
.412
PI1
4
WSP1
.804
WSP2
.726
5
31
WSP4
.700
WSP3
.630
PA2
.867
PA3
.859
PA1
.769
Extraction Method: Principal Component Analysis.
Rotation Method: Varimax with Kaiser Normalization.
a. Rotation converged in 6 iterations.
After deleted PI1 item, the scales reduced from 19 items to 18 items and regrouped
into five components. In this case, the EFA analysis was run again and result testing is below
According to Leech et al., (2005) KMO test indicates one whether or not enough
items are predicted by each factor. The Bartlett‟s test shows the significant value lower than
0.05 (Sig. =.000) which indicating that the correlation matrix is significant different from an
identity matrix and the correlation between variables are all zero. Based on the test of
assumptions, the result indicated KMO was 0.832 greater than 0.7 and sig p-value <0.05, this
meant the correlation matrix is unit matrix, items has correlation with each others. As result,
both acceptances for diagnostic tests confirm that the data were suitable for factor
analysis.(see table 4.4)
Table 4.4 KMO and Bartlett’s test result
KMO and Bartlett's Test
Kaiser-Meyer-Olkin Measure of Sampling
Adequacy.
Bartlett's Test of
Sphericity
Approx. Chi-Square
.832
1.582E3
Df
153
Sig.
.000
By doing EFA (principal components analysis, rotation method: Varimax) extracted
four factors from 18 items measuring. The cumulative of the five factors accounted for
66.743 percent of variance.
32
The determination of the number of factors is usually done by considering only
factors with Eigenvalues greater than 1.0, since each variable is expected to have a variance
of 1.0. The proposed model has five factors. However, only the first five factors that had the
eigenvalue more than 1, which is a common criterion for a factor to be useful. With these five
factors were extracted, it could explain 66.743% of variance of the original variables. Total
variance explained of factor analysis is demonstrated in Appendix D
There is some little changes in factors after run EFA again (See table 4.5). Only PI1
should delete. The rest factors almost have no change. This means the perception of
interviewees about some variables is the same from the hypotheses of the study and some
previous theories.
Component 1: “Trust” (T1, T2, T3, T4)
Component 2: “Involvement” (I1, I2, I3, I4)
Component 3: “Website social presence” (WSP1, WSP2, WSP3, WSP4)
Component 4: “Product attitude” (PA1, PA2, PA3)
Component 5: “Purchase intention” (PI2, PI3, PI4)
Table 4.5 Rotated Component Matrix (times 2)
Rotated Component Matrixa
Component
1
T2
.794
T4
.761
T3
.760
T1
.717
2
I4
.781
I3
.719
I2
.679
I1
.646
WSP1
3
.808
4
5
33
WSP2
.737
WSP4
.699
WSP3
.625
PA2
.869
PA3
.859
PA1
.768
PI4
.823
PI3
.708
PI2
.656
Extraction Method: Principal Component Analysis.
Rotation Method: Varimax with Kaiser Normalization.
a. Rotation converged in 6 iterations.
After checking the result of the reliability of each variable and validity of the
measurement scale, the model has been still kept and was used to run regression.
4.4 Regression analysis
4.4.1 Testing assumptions of multiple regressions
The sample size of this research is 219 samples was much larger than the minimum
requirement for multiple regression analysis (100 samples).
The purpose of the Standard Multiple Regression is used for testing the hypotheses,
studying the correlation, and measuring the level of impact of each independent factor on the
dependent factor. However, before conducting the hypotheses testing, it required the
variables to satisfy some crucial assumptions. As recommended by Leech et al. (2005), there
were some main assumptions:
Assumption 1: The residuals were independent
Assumption 2: The linear relationship between independent variables and
dependent variable occurred
Assumption 3: The residual was distributed normally
34
Assumption 4: No multicollinearity among independent variables
As a result, all assumptions are satisfied and details are summarized below.
Results of testing assumptions
Assumption 1
To test this assumption, it was necessary to consider the Durbin-Waston value. If
Durbin-Waston was between 0 and 4, the best was around 2, it could be concluded that the
residuals were independent. As presented in Appendix D, the value of Durbin-Waston was
1.923 so assumption 1 was satisfactory.
Assumption 2
With regard to assumption 1, the shape of overall regression plot could help to test
this assumption. If the overall regression plot made a curvilinear shape, it indicated that
predictors did not linear relate to dependent variable. With the plot shown in Appendix E, this
assumption was well supported.
Assumption 3
This assumption could be confirmed by drawing the residual scatterplot chart. If the
dots in chart were scattered, it meant that the data met the assumption of residuals being
normally distributed. In this research, this assumption was confirmed (see Appendix E).The
scatterplot of regression for Purchase intent indicated that all the values were roughly
distributed symmetrically around the central point zero, therefore the homoscedasticity was
not occurred. This result pointed out the sample data was appropriated to run multiple linear
regressions for testing.
Assumption 4
The assumption about multicollinearity was the most important one that needed to be
satisfied before running multiple regression. This problem can lead to misleading or impact
negatively to the significance of data analysis results. Multicollinearity happened in case
35
there were the high intercorrelations among some composite of the independent variables
(Leech et al., 2005). Stated differently, multicollinearity would appear when there was the
overlapping of information between two or more predictors. To investigate this issue,
correlation matrix would be quite useful.
Table 4.6 Correlations matrix
Correlations
Trust
Involvement Product
attitude
Description
1
Trust
.357**
Involvement
1
**
.264
.184**
Product attitude
1
Website social
.461**
.441**
.160**
presence
.411**
.573**
.109**
Purchase intent
**. Correlation is significant at the 0.01 level (2-tailed).
Website
social
presence
1
.464**
Purchase
intent
1
According to the Correlations matrix, the Pearson test indicated the value lower than
0.8 and had significant value (Sig <5%) among predictors. This matrix proved that there
would be a low possibility of multicollinearity. Although the Correlation matrix was helpful
in identifying multicollinearity, sometimes it failed in testing such problem. For more strictly
investigation, therefore, it needed to consider more the VIF value in the table of Coefficients
(see Table 4.9) for withdrawing the best conclusion. The result showed that multicollinearity
happens if Tolerance <0, 1 or VIF>10. All factors have their VIF < 10 and Tolerance >0.1.
Therefore, assumption 4 is satisfied, no multicollinearity.
In summary, the data almost met all the required assumptions. Therefore, all
predictors were qualified enough for multiple regression analysis.
4.4.2. Result of multiple regression analysis
The follows part of the output is the coefficient section, after checking for the model,
the identification of relationship between predictors and dependent factor are fit the most
36
important. The standardized coefficients (β) are the coefficients of the estimated regression
model. In addition, testing histogram and normal probability plot showed that they are
satisfied the regression analysis
The multiple regression result of 4 independent variables and 1 dependent variable
with Enter method outputted the following table 4.7
Table 4.7: Model Summary of multiple regression analysis
Model Summaryb
Model
R
R Square
Adjusted R
Square
Std. Error of the
Estimate
1
.638a
.407
.396
1.81447
a. Predictors: (Constant), Website social presence, Product attitude,
Involvement, Trust
b. Dependent Variable: Purchase intent
The Model Summary table presented the multiple correlation coefficient (R) was
0.638, R Square was equal 0.407 and adjusted R Square was 0.396, showing that 39,6 % of
the variance purchase intent of online shopping in Facebook could be predicted from four
independent variables. In other words, the model of this study was fit.
Table 4.8 ANOVA of multiple regression analysis
Sum of
Model
1
Squares
df
Mean Square
F
Sig
Regression
483.673
4
120.918
36.727
.000a
Residual
704.555
214
3.292
1188.2 8
218
Total
a. Predictors: (Constant), Website social presence, Product attitude, Involvement,
Trust
b. Dependent Variable: Purchase intent
37
The table 4.8 ANOVA showed that the value of F was 36.727 and significance value
less than 0.05. It indicates that the combination of these variables significantly predict the
dependent variable. The model is fit.
4.5 Hypotheses testing result and discussions of the findings
The initial research model developed from the previous studies and literature review
to predict that Trust, product attitude, involvement and website social presence had an effect
on Facebookers‟ purchase intent in Hochiminh city, Vietnam, and then conducted tests on the
hypotheses through a variance analysis via SPSS statistical software to find out what
independent variables are valid. There are some different from the result of data analysis. The
summary of hypotheses testing results were indicated in table 4.10. Below part will present
the comparison between data analysis result and original theoretical model research. And
then it can provide us the practical insight in Facebook shopping, case in Hochiminh city
market.
Table 4.9 Coefficients of multiple regression analysis
Unstandardized
Coefficients
Model
(Constant)
B
Std. Error
3.296
.779
.140
.048
-.048
Involvement
Website social
presence
Trust
Product attitude
Standardized
Coefficients
Beta
Collinearity
Statistics
t
Sig.
Tolerance
VIF
4.234
.000
.178
2.891
.004
.729
1.372
.053
-.050
-.907
.365
.921
1.086
.320
.045
.430
7.166
.000
.769
1.300
.151
.048
.200
3.180
.002
.700
1.429
a. Dependent Variable:
Purchase intent
The result being show in Table 4.10 can answer research questions mentioned in
chapter I. The empirical results suggest three key factors including “involvement”, “website
38
social presence”, and “trust” have a significant impact on purchase intent. Only one variable
“product attitude” has not impact on purchase intent. Further details are described, as follows
H1: The perception of social presence within an online retailers’ website will positively
influence consumers’ purchase intent
See Table 4.9, value is significant at five percent level. Website social presence were
found to have significant effect on Purchase intent (β= 0.2, t = 3.180, p = 0.002<0.05). Hence,
H1is supported.
According the relationship between “Website social presence” and “purchase intent”,
the ANOVA testing and Coefficients table showed that the hypothesis 1“The perception of
social presence within an online retailers‟ website will positively influence consumers‟
purchase intent” is supported. In the other previous research, this result is the same. Gefen
and Straub (2003) showed that the perception of social presence has an effect on online
consumers‟ intention to purchase from a commercial e-service website. In this research, the
result show that the website social presence appears to have a positive significant effect on
purchase intent (β=0.20). Some remarked that the website lacked a personal touch, lacked the
human aspect of displaying and general unappealing, the purchase intent is not promoted.
H2: Trust has a positive impact on purchase intent.
According to Table 4.9, Trust had significant effect on Purchase intent (β=0.178, t =
2.891, p = 0.004 < 0.05). Thus, Hypothesis H2 is supported.
Previous studies suggested that the trust of consumers while shopping online and past
online experiences might influence their future intentions. Especially, the Beta values of trust
are also quite small, around 0.178. The important message from this finding is that actually
customers have not put totally their trust on online service like shopping in retailer shops in