4.1 Internal reliability
Each of the three dimensions was checked for internal reliability. The knowledge dimension reported a Cronbach’s alpha of .84, while attitude and behaviour had an alpha value of .91 and .77, respectively, which indicated good to excellent levels of internal consistency in the subscales.
4.2 Data Normality
The moderating effects of education level and work exposure on Internet security knowledge, attitude, and behaviour variables were investigated in this study.
The education level was divided into four levels: freshmen (low education level), sophomores and juniors (medium education level), seniors (higher-medium education level), and postgraduates (advanced education level). Meanwhile, work exposure was classified as follows: non-graduation years (low exposure), graduation years (medium exposure), and working graduates (high exposure).
The ISA variables were then examined for normality. Direct hypothesis testing was not recommended because both Kolmogorov-Smirnov and Shapiro-Wilk were designed for smaller sample sizes (n ≤ 50) [47]. Skewness and kurtosis values, on the other hand, were determined by hand. A larger sample with skewness between − 2 and + 2 and kurtosis between − 7 and 7 was considered normally distributed (Bryne, 2010; Hair et al., 2010). Mean knowledge (skewness = − .270, kurtosis = − .393), mean attitude (skewness = − .263, kurtosis = − .783), and mean behaviour (skewness = − .342, kurtosis = .351) were all within the normal range.
4.3 Demographic information
According to the range provided by the questionnaire, grades were segmented into six sequential levels (Year 1, Year 2–3, postgraduate students, young working graduates, experienced working graduates, and final year). Ages (18–25, 26–35, 36–45, 46), disciplines (liberal arts & social sciences, natural sciences, technology & engineering), and genders (male, female) were also used as classification variables.
4.4 Multicollinearity
Next, a multicollinearity test was performed on the data.Correlation tests were conducted for variables such as knowledge, attitude, behaviour, grade, age, discipline, and gender (see Table II for correlation results).Mean knowledge, attitude and behavior were discovered to be significantly positive and highly correlated (p < .001). There was a significant positive correlation between grade and age (p < .001), and they showed significant negative correlations with mean knowledge, attitude, and behaviour (p < .001). Therefore, grade and age as independent variables were considered covariables. To test the effect of grade, age has to be controlled.
Table II. Correlation analysis for variables
Mean knowledge
|
Mean
behaviour
|
Mean
attitude
|
Grades
|
Age
|
Discipline
|
Gender
|
|
Mean knowledge
Mean behaviour
Mean attitude
Grades
Age
Discipline
Gender
|
1.000
|
|
|
|
|
|
|
.780**
|
1.000
|
|
|
|
|
|
.765**
|
.823**
|
1.000
|
|
|
|
|
− .500**
|
− .415**
|
− .443**
|
1.000
|
|
|
|
− .437**
|
− .352**
|
− .422**
|
.636**
|
1.000
|
|
|
− .135**
|
− .138**
|
− .152**
|
.102**
|
.404**
|
1.000
|
|
.097**
|
.154**
|
.160**
|
− .028
|
− .074**
|
− .333**
|
1.000
|
**. Correlation was significant at the .01 level (2-tailed). |
As a result, the regression analysis hypothesis was completely satisfied. Therefore, the data and residuals were normally distributed, with no heteroscedasticity or correlation between regression residuals. The potential collinearity between grade and age was identified, but it could be accommodated in the proposed regression model by controlling for the latter.
4.5 Data Checking
A multivariate regression model was used to examine the effects of different grade levels and levels of work exposure on mean scores of knowledge, attitude, and behaviour, as detailed below according to the corresponding assumptions. For simplicity, the changes were illustrated in Figures II, III, and IV).
H1:
University students of different years of study, postgraduate students and working graduates will display differing cybersecurity knowledge.
H1 was verified. There were significant differences between the knowledge of postgraduate students (M = 3.53, SD = .47) and non-final year university students including Year 1 students (M = 3.72, SD = .78) and Year 2-3students (M = 3.64, SD = .77). Moreover, the knowledge of postgraduate students was found to be different from final-year ones (M = 3.57, SD = .77). At the same time, the knowledge of postgraduate students was also found to be visibly different from working graduates that contained young working graduates (M = 2.63, SD = .47) and experienced working graduates (M = 2.65, SD = .47). For simplicity, the differences in mean values of the three dimensions were illustrated in Table II.
H2a: Postgraduate students will score significantly higher in attitudes than final-year students.
H2a was rejected. The attitude of postgraduate students (M = 3.73, SD = .69) was found to be significantly lower than final-year students (M = 3.80, SD = 1.00).
H2b: Postgraduate students will score significantly higher in attitudes than working graduates.
H2b was verified. The attitude of postgraduate students was found to be significantly higher than young working graduates (M = 2.73, SD = 0.32) and experienced working graduates (M = 2.73, SD = 0.36).
H2c: Postgraduate students will score significantly higher in attitudes than non-final year students.
H2c was rejected. The attitude of postgraduate students was found to be significantly lower than Year 1 students (M = 3.88, SD = 1.00), and Year 2-3students (M = 3.77, SD = 1.02).
H3a: Postgraduate students will score significantly higher in behaviours than final-year students.
H3a was verified. The behaviour of postgraduate students (M = 3.54, SD = .47) was found to be significantly higher than final-year ones (M = 3.39, SD = .82).
H3b: Postgraduate students will score significantly higher in behaviours than working graduates.
H3b was verified. The behaviour of postgraduate students was found to be significantly higher than working graduates containing young (M = 2.83, SD = .28) and experienced (M = 2.88, SD = .28).
H3c: Postgraduate students will score significantly higher in behaviours than non-final year students.
H3c was partially verified. The behaviour of postgraduate students was found to be significantly lower than Year 1 students (M = 3.62, SD = .82), but significantly higher than that of Year 2-3students (M = 3.50, SD = .85).
In terms of knowledge, as students advanced to higher levels, their cybersecurity knowledge deteriorated. Their cybersecurity knowledge dropped significantly at work.
As graduates gained more work experience, their knowledge seemed to improve.
In terms of behaviour, postgraduate students performed similarly to Year 2–3 students but not as well as Year-1 students. Working graduates performed similarly poorly, but slight improvements could be seen after gaining more work experience. In terms of attitude, as students advanced to higher levels, became postgraduate students, and went to work, their cybersecurity attitudes deteriorated. Even inexperienced employees showed no improvement.
4.6 Gender difference
Many studies have hypothesised that personal factors such as gender influence behavioural decisions [36, 48]. The analysis revealed that the hypothesis was statistically supported. Females' knowledge (M = 3.44, SD = .78) was found to be significantly higher than males' (M = 3.27, SD = .78), as was their attitude (M = 3.66, SD = .94 vs M = 3.32, SD = .99). The same was true for female and male behaviour (M = 3.44, SD = .70 vs M = 3.19, SD = .81). As a result, females had higher ISA scores than males in all three dimensions.
4.7 Significant Predictors
A multivariate regression model was also run to see if age, discipline, grade, or gender were significant predictors of ISA. Previous research suggested that age was a significant predictor of ISA (e.g., [49], but a recent study found that age was not a significant predictor [7]. Since then, the most stable, whose value was always positive, Pillai's Trace, to test the sum of a matrix's eigenvalues, and Roy's Largest Root statistic, to test the maximum value in a matrix's eigenvalues, have been used. And the greater the magnitude of these values, the greater the contribution of this effect to the model. Wilk's Lambda was also used, with a value between 0 and 1. In Wilk's Lambda, the smaller the value, the greater the contribution. As a result of the above analysis, the results showed that, overall, grade and gender were significant predictors of ISA, whereas age and discipline were not.
4.8 Significant Correlations
In addition, tests of between-subjects effects were examined further, with mean knowledge, mean attitude, and mean behaviour as the dependent variables.
Significant correlations were discovered between grades, discipline, and gender.
According to the Sum of Squares, the grade had the greatest impact on knowledge (112.96), attitude (123.35), and behaviour (70.33), in that order.
(See Table III.)
Table III: Tests of Between-Subjects Effects
Source
|
Dependent Variable
|
Type III Sum of Squares
|
Mean Square
|
F
|
Sig.
|
Partial Eta Squared
|
Grades
|
mean_knowledge
|
122.955
|
30.739
|
67.265
|
< .001
|
.127
|
mean_behaviour
|
70.330
|
17.582
|
38.507
|
< .001
|
.077
|
mean_attitude
|
123.345
|
30.836
|
44.605
|
< .001
|
.088
|
Age
|
mean_knowledge
|
.195
|
.097
|
.213
|
.808
|
.000
|
mean_behaviour
|
.175
|
.087
|
.191
|
.826
|
.000
|
mean_attitude
|
1.108
|
.554
|
.802
|
.449
|
.001
|
discipline
|
mean_knowledge
|
4.237
|
2.119
|
4.636
|
.010
|
.005
|
mean_behaviour
|
2.948
|
1.474
|
3.229
|
.040
|
.003
|
mean_attitude
|
2.856
|
1.428
|
2.065
|
.127
|
.002
|
gender
|
mean_knowledge
|
.038
|
.038
|
.083
|
.773
|
.000
|
mean_behaviour
|
4.352
|
4.352
|
9.531
|
.002
|
.005
|
mean_attitude
|
5.386
|
5.386
|
7.791
|
.005
|
.004
|
4.9 Discipline Differences
To the best of the authors' knowledge, ISA research in higher education rarely focuses on discipline. Thus, Fisher's LSD was used in this study for inter-group multiple comparisons to compare disciplines. The analysis found no difference between the disciplines of natural sciences and technology and engineering, but there was a difference between the above two disciplines and the discipline of liberal arts and social sciences. That was because, in Chinese universities, the disciplines of natural sciences, technology, and engineering had at least two programme design courses and one professional computer application course, whereas students majoring in liberal arts and social sciences only took the common course of Fundamentals of College Computing in the first year, according to the Computer Basic Curriculum System and the actual investigation [50, 51].