Respondent Profiles and Measurement Models
The profiles of the respondents are shown in Table 1. There were 439 males (53.3%) and 384 females (46.7%). The majority of the respondents were between 26 and 55 years of age and had either undergraduate/diplomas (56.1%) or masters and/or doctoral degrees (33%). The highest percentage of respondents were government officers (51.9%). In regards to residence, 60.8% of the respondents resided on Java Island. Information on whether respondents had family members older than 65 years with heart and lung disease are also shown in Table 1.
Table 1 Demographic characteristics of the respondents
Characteristics
|
Frequency
|
Percent %
|
Characteristics
|
Frequency
|
Percent %
|
Gender
|
823
|
100
|
Occupation
|
823
|
100
|
Male
|
439
|
53.3
|
Households
|
40
|
4.9
|
Female
|
384
|
46.7
|
Student
|
62
|
7.5
|
Age
|
823
|
100
|
Informal Sector
|
22
|
2.7
|
17-25 years
|
89
|
10.8
|
Private Employment
|
139
|
16.9
|
26-35 years
|
206
|
25
|
Government Officer
|
427
|
51.9
|
36-45 years
|
247
|
30
|
Others
|
133
|
16.2
|
46-55 years
|
221
|
26.9
|
Living areas
|
822
|
100
|
56-65 years
|
59
|
7.2
|
Urban
|
529
|
64.3
|
>65 years
|
1
|
0.1
|
Rural
|
294
|
35.7
|
Educational level
|
823
|
100
|
Family members with high risk
|
-
|
-
|
At least middle school
|
2
|
0.2
|
< 60 years old
|
491
|
60.6
|
High school
|
87
|
10.6
|
Heart disease
|
558
|
69.5
|
Bachelor’s degree
|
462
|
56.1
|
Lung/respiratory disease
|
634
|
78.8
|
Graduate degree
|
272
|
33
|
|
|
|
Note: The responses of having family members with high risk applied more than one possibility answer. The percentage against the total respondents (N=823).
To test the proposed hypotheses, it was essential to evaluate the reliability and validity of the latent variables before examining the structural model. Table 2 shows the results of factor loadings, Cronbach’s alpha, the composite reliability (CR), and the average variance extracted (AVE). Factor loading is preferred equal to or greater than 0.70, however if it is an exploratory research 0.4 or higher is acceptable (39). Factor loading was set at a minimum of 0.6 (40) and the Cronbach’s alpha values were set at a minimum of 0.60 (40). Internal consistency was evaluated and resulted in a composite reliability > 0.7. AVE reflects convergence and divergent validity and it is recommended that the threshold value for AVE should be exceed 0.5 (39). The construct validity test is another discriminant validity, which aims to confirm that certain latent variables differ from others. The AVE square root was calculated and its value should be the highest in comparison with the correlations with other latent variables (41). Table 3 shows the standard of discriminant validity. It can be seen from the Fornell-Larcker criterion table, the number of the square root of AVE appears in the diagonal cells are higher than the number correlations appear below it. Based on these results, the model was considered to be reliable, internally consistent, and with adequate discriminant validity.
Table 2. Mean, SD, Factor Loading, Cronbach’s Alpha, Composite Reliability and AVE
Construct and Items
|
Mean (1-7)
|
SD
|
Factor Loading
|
Cronbach’s a
|
CR
|
AVE
|
Media Use
|
|
|
|
0.856
|
0.887
|
0.755
|
Print
|
2.962
|
1.880
|
0.717
|
|
|
|
Radio
|
2.544
|
1.703
|
0.627
|
|
|
|
TV
|
4.400
|
2.120
|
0.814
|
|
|
|
WhatsApp
|
5.146
|
1.907
|
0.775
|
|
|
|
Social Media
|
4.925
|
2.020
|
0.749
|
|
|
|
Domestic Web
|
5.094
|
1.882
|
0.771
|
|
|
|
Overseas Web
|
3.547
|
2.068
|
0.629
|
|
|
|
Risk Perception
|
|
|
|
1.000
|
1.000
|
1.000
|
Covid-19 is deadly
|
5.887
|
1.455
|
1.000
|
|
|
|
Attitude
|
|
|
|
0.676
|
0.861
|
0.755
|
Stay-at-home
|
6.219
|
1.244
|
0.868
|
|
|
|
Social distancing
|
6.495
|
0.685
|
0.870
|
|
|
|
Subjective Norm
|
|
|
|
0.779
|
0.871
|
0.694
|
Family agrees
|
6.465
|
0.669
|
0.828
|
|
|
|
Family support
|
6.417
|
1.016
|
0.880
|
|
|
|
Religious leaders agree
|
6.340
|
1.029
|
0.789
|
|
|
|
Perceived Behavior Control
|
|
|
0.751
|
|
0.663
|
Able to stay-at-home
|
5.360
|
1.656
|
0.769
|
|
855
|
|
I can control
|
6.070
|
1.230
|
0.867
|
|
|
|
Easy to stay at home
|
5.583
|
1.573
|
0.803
|
|
|
|
Intention
|
|
|
|
0.864
|
0.937
|
0.881
|
Stay-at-home
|
5.614
|
1.609
|
0.939
|
|
|
|
Socially distance
|
5.807
|
1.477
|
0.938
|
|
|
|
Note: SD: Standard Deviation, CR: Composite Reliability, AVE: Average Variance Extracted, 1-7: Measurement scales
Table 3. Discriminant validity tests results (Fomell-Larcker Criterion)
Construct
|
ATT
|
Intention
|
Media
|
PBC
|
RP
|
SN
|
ATT
|
0.869*
|
|
|
|
|
|
Intention
|
0.260
|
0.938*
|
|
|
|
|
Media
|
0.107
|
0.009
|
0.729*
|
|
|
|
PBC
|
0.418
|
0.367
|
0.041
|
0.814*
|
|
|
RP
|
0.217
|
0.108
|
0.116
|
0.218
|
Single Item
|
|
SN
|
0.628
|
0.382
|
0.104
|
0.511
|
0.285
|
0.833*
|
Note: ATT: Attitude, PBC: Perceived behavior control, RP: Risk perception, SN: Social norm. Significance level: * p < 0.001, ** p < 0.05.
Structural Model and Hypotheses Testing
To test the structural model and hypotheses, a bootstrapping procedure with 5,000 iterations and 823 subsamples was used (42). Figure 1 shows the results of the explained variance or adjusted R-square (R2) and path-values of the model. These results show that the R2 value of intention was 0.183, attitude was 0.052, social norms was 0.084, PBC was 0.046, and risk perception had value of 0.012. The R2 of intention was higher than the 0.10 threshold (43) indicating that 18.3 % of the variance in social distancing intention can be explained by the components of the extended TPB.
Table 4 shows the assessment of the significance of the path coefficients, t-statistics and p-values among the components of the extended TPB model. Intention to socially distance was determined by social norms (β = 0.265, t-value = 4.575, p-value < 0.000) and PBC (β = 0.234, t-value = 4.625, p < 0.000). Risk perception significantly influenced attitudes (β = 0.207, t-value = 4.717, p < 0.000), social norms (β = 0.276, t-value = 6.333, p < 0.000), and PBC (β = 0.216, t-value = 5.418, p < 0.000). In addition, media use affected risk perception (β = 0.116, t-value = 3.177, p < 0.001), attitudes (β = 0.083, t-value = 2.323, p < 0.020), and social norms (β = 0.072, t-value = 2.158, p < 0.031). Therefore, hypotheses H2 through H9 were supported. Hypotheses H1 and H10 were not supported, as there were no significant causal relationships between attitudes and intentions, and media to PBC.
Table 4. Results of the proposed hypotheses test
Hypothesis
|
Path Coefficient (β-value)
|
t-statistic
|
p-value
|
Results
|
H1: ATT → INT
|
-0.004
|
0.068
|
0.945
|
Not Supported
|
H2: SN → INT
|
0.265
|
4.575
|
0.000*
|
Supported
|
H3: PBC → INT
|
0.234
|
4.625
|
0.000*
|
Supported
|
H4: RP → ATT
|
0.207
|
4.717
|
0.000*
|
Supported
|
H5: RP → SN
|
0.276
|
6.333
|
0.000*
|
Supported
|
H6: RP → PBC
|
0.216
|
5.418
|
0.000*
|
Supported
|
H7: Media → RP
|
0.116
|
3.177
|
0.001*
|
Supported
|
H8: Media → ATT
|
0.083
|
2.324
|
0.020**
|
Supported
|
H9: Media → SN
|
0.072
|
2.158
|
0.031**
|
Supported
|
H10: Media→ PBC
|
0.016
|
0.405
|
0.685
|
Not Supported
|
Note: ATT: Attitude, INT: Intention, PBC: Perceived behavior control, RP: Risk perception, SN: Social norm. Significance level: * p < 0.001, ** p < 0.05.
Multigroup Analysis
As this study involved a wide range of respondent backgrounds, a multigroup analysis was used to explore whether media use, risk perception, and the TPB components were different across genders (male vs. female), ages (elder vs. younger), and residential areas (urban vs. rural). A two-step procedure was used to examine the statistical differences between the groups under investigation: bootstrapping and Multi-Group Analysis (MGA). The bootstrapping procedure was used to assess the path coefficients and p-values of each group, as well as the mean, STDEV and t-values from the results. The PLS-MGA test assessed the differences in path coefficients and significant p-values between each group. A PLS-MGA test indicates significance differences between groups if the p-value is lower than 0.05 or larger than 0.95 for the differences between group-specific path coefficients (39).
Table 5 shows the result of the path coefficient for each group. H1 hypothesized that attitude positively influences the intention to socially distance, which was not supported across all of the groups. However, H2, which proposed that subjective norms influence the intention to socially distance, was supported across all of the groups: gender (male: β = 0.226, p < 0.05; female: β = 0.299, p <0.001), age group (elder: β = 0.281, p < 0.05; younger: β = 0.263, p < 0.001), and residential area (urban: β = 0.262, p < 0.001; rural: β = 0.282, p < 0.05,). The third hypothesis, the influence of PBC on intention to socially distance (H3), was also confirmed for all of the groups: gender (male: β 0.310, p < 0.001; female: β = 0.153, p < 0.05), age group (elder: β = 0.357, p < 0.001; younger: β = 0.176 p < 0.001) and residential area (urban: β = 0.240, p < 0.001; rural: β = 0.229, p < 0.05). Similar to H2 and H3, the fourth hypothesis (H4), which suggested that risk perception influences attitudes towards social distancing, and H5, which suggested that risk perception influences subjective norms, was supported across all of the groups: gender (male: β = 0.263, p < 0.001 and β = 0.320, p < 0.001; female: β = 0.154, p < 0.05 and β = 0.227, p <0.05), age group (elder: β = 0.191, p < 0.05 and β = 0.230, p < 0.05; younger: β = 0.221, p < 0.001 and β = 0.308, p < 0.001), and living area (urban: β = 0.164, p < 0.05 and β = 0.215, p < 0.001; rural: β = 0.299, p < 0.001). H7, which proposed that media use influenced risk perception, was supported in all of the groups: gender only for females (β = 0.127, p <0.05), ages for the younger group (β = 0.127, p < 0.05) and residential areas for rural populations (β = 0.234, p < 0.05). The effect of media use on attitude, as hypothesized in H8, was only confirmed for the younger age group (β = 0.111, p < 0.05). H9 which suggested media use influences subjective norms, was supported only for females (β = 0.123, P < 0.05), while H10, which proposed media use positively influences PBC, was not supported in any of the groups.
Table 5. Multigroup Analysis Statistical Tests
Construct
|
H1
|
H2
|
H3
|
H4
|
H5
|
H6
|
H7
|
H8
|
H9
|
H10
|
Gender
|
|
|
|
|
|
|
|
|
|
|
Male (N=439)
|
0.021
|
0.226**
|
0.310*
|
0.263*
|
0.320*
|
0.263*
|
0.103
|
0.100
|
0.052
|
0.046
|
Female (N=384)
|
-0.034
|
0.299*
|
0.153**
|
0.154**
|
0.227**
|
0.161**
|
0.127**
|
0.082
|
0.123**
|
-0.011
|
Diff.
|
0.055
|
-0.073
|
0.157
|
0.110
|
0.093
|
0.102
|
-0.024
|
0.018
|
-0.071
|
0.567
|
PLS-MGA p-value
|
0.635
|
0.505
|
0.085
|
0.215
|
0.296
|
0.195
|
0.737
|
0.806
|
0.317
|
0.527
|
Ages
|
|
|
|
|
|
|
|
|
|
|
Elder/ > 46 (N=281)
|
-0.013
|
0.281**
|
0.357*
|
0.191**
|
0.230**
|
0.104
|
0.117
|
0.036
|
0.088
|
0.655
|
Younger/< 45 (N=542)
|
-0.002
|
0.263*
|
0.176*
|
0.221*
|
0.308*
|
0.283*
|
0.127**
|
0.111**
|
0.063
|
-0.121
|
Diff.
|
-0.010
|
0.018
|
0.181
|
-0.029
|
-0.077
|
-0.178
|
-0.009
|
-0.074
|
0.025
|
0.077
|
PLS-MGA p-value
|
0.949
|
0.872
|
0.056
|
0.724
|
0.375
|
0.020
|
0.995
|
0.459
|
0.668
|
0.390
|
Residential areas
|
|
|
|
|
|
|
|
|
|
|
Urban (N=529)
|
-0.033
|
0.262*
|
0.240*
|
0.164**
|
0.215*
|
0.177**
|
0.059
|
0.084
|
0.082
|
0.060
|
Rural (N=294)
|
0.043
|
0.282**
|
0.229**
|
0.299*
|
0.404*
|
0.305*
|
0.234**
|
0.066
|
0.043
|
-0.087
|
Diff.
|
-0.076
|
-0.019
|
0.010
|
-0.136
|
-0.188
|
-0.127
|
-0.174
|
0.017
|
0.038
|
0.147
|
PLS-MGA p-value
|
0.623
|
0.858
|
0.909
|
0.135
|
0.028
|
0.108
|
0.014
|
0.819
|
0.562
|
0.082
|
Note:
H1…H10: Hypotheses refer (same as) to the structural model of the extended TPB
* and **: Path coefficient each sub-group with significance level, * p < 0.001, ** p < 0.05.
Diff: Path Coefficient Differences between sub-group in the PLS-MGA.
Bold font: PLS-MGA p-value below 5% and above 95% indicate a significant difference.
The PLS-MGA p-values indicated that there were no significant differences between gender. Multigroup analysis also indicated that there were small differences between the elder and younger age groups. Among the 10 hypotheses, only H6 and H7 were supported. H6 suggests that the effect of risk perception on PBC for young people (equal to and under 45 years old) was stronger than for elder people (difference = -0.178, p<0.05). H7 indicated that media use had more of an impact on risk perception for young people compare to elder people (difference = -0.009, p > 0.095).
The differences between rural and urban residents applied to hypotheses H5 and H7. Hypothesis 5 (H5) described the effect of risk perception on social norms and was supported in rural populations (difference = -0.188, p < 0.05). Similarly, H7 explained that the media’s effect on risk perception was also more effective for people living in rural areas (difference = -0.174, p > 0.05).