Sample and Procedures
The current sample comprises 231 adult participants above 21 years old (172 females, 59 males) enrolled in Singapore. Anonymous surveys with same contents were administered to them through either paper-based or online surveys. A convenient sampling was used to recruit 186 adults responded on paper-based questionnaires. The administration was under supervision of research team members, who were responsible to offer further explanations and ensure the confidentiality. Additionally, a snowball sampling was utilized to enrol 45 participants responded online via Qualtrics software. An independent-samples t-test was conducted and no significant difference was found between two survey approaches. English was the language used in all survey forms.
Measures
General Self-efficacy
A shortened version of NGSE developed by Chen, Gully (16) was adapted and used in the current study. It included 7 items and was labelled as NGSE-7. The coefficient alpha of the NGSE-7 was .870 in this sample. Participants responded to all items on a 5-point rating scale from 1 (Not true of me) to 5 (Very true of me).
Theoretically Relevant Variables
Life Satisfaction, Task Stress and Illness Symptoms. The 5-item Life Satisfaction (LS) scale, 9-item Task Stress (TS) scale and 4-item Illness Symptoms (IS) scale have been developed by Pettegrew and Wolf (41) to measure teacher stress. In this study, some items were slightly reworded to apply into general population (e.g., “Trying to provide a good education in an atmosphere of decreasing financial support is very stressful.” into “Trying to get a good work outcome is very stressful.”). The internal consistency reliabilities of the three scales were comparatively high in present sample (α = .788, .835, .732, respectively). To maintain the consistency of whole survey form style, the responding format was simplified from original 6-point to 5-point Likert scale ranging from 1 (Strongly Disagree) to 5 (Strongly Agree).
Perceived Stress. This construct was measured with the 14-item Perceived Stress (PS) scale developed by Cohen, Kamarck (42) to assess PS level of community samples with a junior high school certificate and above. Two of the items were slightly reworded to fit more general situations. The coefficient alpha for the PS scale was .819 in this sample. To align with overall rating method, in this study, items were scored on a 5-point frequency rating scale ranging from 1 (Never) to 5 (Always). The PS scale scores were obtained by summing each item score after reversing scores of seven positive items.
Positive and Negative Emotions. Both positive emotions (PEM) and negative emotions (NEM) were measured by revised scales based on the Scale of Positive and Negative Experience (SPANE) developed by Diener, Wirtz (43). In this study, two items (happy and good) retained from SPANE - P and another added two items (cheerful and excited) composed the PEM scale. The NEM scale consists of six items, including two items (angry and sad) derived from SPANE - N and four reworded items (frustrated, lonely, disrespected and miserable). The coefficient alpha of PEM and NEM scales were .898 and .812, respectively. All items were responded on the same 5-point scale from 1 (Never) to 5 (Always).
Data Analysis and Results
Selection of Single Item
The particular item was selected from NGSE-7 based on following five approaches (32): (1) expert judgement, (2) the highest item-to-total correlations (ITC), (3) the highest factor loading extracted from principal axis factoring (PAF), (4) the highest factor loading extracted from confirmatory factor analysis (CFA) and (5) the highest squared multiple correlations (SMC) in CFA.
Nine scholars with earned doctorates in psychology and with faculty appointments at a research-intensive university provided their comments on each item of the NGSE-7. Experts’ views were not consistent, thus their views had limited value in the selection of the single item. Therefore, four statistical methods were used, and the results are presented in Table 1. Four different statistical methods converged on one particular item, pointing out that item 4 captures the construct of GSE most. Only one factor was extracted in EFA, with an eigenvalue of 3.460, accounting for 49.43% of total variance and a one-factor congeneric CFA model was applied. Item 4 was reported with the highest ITC, factor loadings and SMC among all seven items. Furthermore, Cronbach’s alpha of the whole scale would decrease from .870 to .844 if item 4 deleted, less than the value of any other item. Overall, the results from four statistical methods showed clear and relatively constant ranks for all items and item 4 was elected as the most representative one in NGSE-7 consistently. Therefore, based on the item selections and cross-validations across five different methods, item 4 (“I believe I can succeed at most of any endeavour to which I set my mind”) was finally selected to be the Single-Item General Self-Efficacy Scale (GSE-SI).
Table 1 Item Selection Based on Four Statistical Methods (n =231)
Items
|
ITC
|
Factor loading (PAF)
|
Factor loading (CFA)
|
SMC
|
1.I will be able to achieve most of the goals that I have set for myself.
|
.615
|
.670
|
.679
|
.460
|
2. When facing difficult tasks, I am certain that I will accomplish them.
|
.665
|
.724
|
.723
|
.522
|
3. In general, I think that I can obtain outcomes that are important to me.
|
.696
|
.759
|
.766
|
.587
|
4. I believe I can succeed at most of any endeavour to which I set my mind.
|
.701
|
.770
|
.779
|
.607
|
5. I will be able to successfully overcome many challenges.
|
.644
|
.696
|
.696
|
.485
|
6. I am confident that I can perform effectively on many different tasks.
|
.669
|
.711
|
.694
|
.481
|
7. Compared to other people, I can do most tasks very well.
|
.537
|
.573
|
.561
|
.315
|
Note. ITC: item-to-total correlation; PAF: principal axis factoring; CFA: confirmatory factor analysis; SMC: squared multiple correlation
Reliability
The results obtained from three different reliability tests (i.e. 1) correction for attenuation formula, 2) PAF in EFA and 3) maximum likelihood in EFA) were listed in Table 2. Since the two variables in current study come from the same conceptual domain (GSE), the estimated reliability of GSE-SI was .726 based on the following formula (44):
Based on the communality obtained from EFA (45), the reliability of GSE-SI was at least .644 and .607, respectively. Overall, the reliability values ranged from .594 to .726, with a mean of .642, indicating a meritorious reliability of GSE-SI.
Table 2 Reliability Based on Three Methods (n = 231)
|
CAF
|
Communalities
(PAF)
|
Communalities
(ML)
|
I believe I can succeed at most of any endeavour to which I set my mind.
|
.726
|
.594
|
.607
|
Note. CAF: correction for attenuation formula; PAF: principal axis factoring; ML: maximum likelihood
Validity
Criterion-Related and Construct-Related Validity
Concurrent validity was assessed by calculating the Pearson’s correlation between the GSE-SI and NGSE-7. The results showed a strong positive correlation of .795 at .01 level (two-tailed), indicating that the responses to the GSE-SI are highly correlated with those to the NGSE-7.
Construct-related validity were evaluated by comparing the Pearson’s correlations between the two GSE measures and six theoretically related variables. As Table 3a shown, the patterns of correlations with other six measures were exactly same across two scales (p < .01, two-tailed). Positive emotions were positively correlated with both GSE-SI and NGSE-7. Similarly, positive correlations between life satisfaction and GSE-SI and NGSE-7 were obtained with comparatively lower but still significant coefficients. Negative emotions and illness symptoms were negatively correlated with both two GSE measures with limited difference. Besides, perceived stress showed strongest negative correlations with two GSE measures and task stress showed the same correlation pattern with slightly lower coefficients.
To further test the construct validity of GSE-SI, the variance reductions were calculated based on the correlations for two GSE measures. When for the GSE-SI were subtracted from those for the NGSE-7, the average reduction was .055, ranging from .022 to .115 (Table 3b). Therefore, single-item and multiple-item GSE scales did not show a significant difference in correlations with other variables.
Adjusting with Reliability: More Valid Use
To gain more confidence, the correlation coefficients between GSE-SI and six theoretically relevant variables were corrected with its reliability based on the correction for attenuation formula (39) with three estimated and the mean reliability of GSE-SI applied (Table 3a). Based on the corrected correlations, the variance reductions decreased in all four correction scenarios, comparing to uncorrected results (Table 3b). Specifically, the average variance reductions calculated by three estimated reliabilities were .013, -.010 and -.008, respectively. Notably, when applying the mean reliability, the reduction in the average variance was only -.001. In addition to the means, standard deviations based on four reliabilities were almost the same (.025) and lower than original values. Overall, the slight differences in correlations indicated that the GSE-SI and the NGSE-7 shared a very similar correlation with the six theoretically relevant constructs.
Table 3a Correlations with Related Variables: Single-item versus Multiple-item General Self-efficacy Scale
|
/
|
Corrected /
|
|
GSE-SI
|
NGSE-7
|
|
|
|
|
Positive emotions
|
.256**/.066
|
.368**/.135
|
.317/.100
|
.351/.123
|
.347/.120
|
.337/.114
|
Life satisfaction
|
.192**/.037
|
.311**/.097
|
.254/.065
|
.281/.079
|
.278/.077
|
.270/.073
|
Negative emotions
|
-.179**/.032
|
-.245**/.060
|
-.233/.054
|
-.258/.066
|
-.255/.065
|
-.248/.061
|
Task stress
|
-.290**/.084
|
-.342**/.117
|
-.372/.138
|
-.412/.170
|
-.407/.166
|
-.396/.157
|
Perceived stress
|
-.343**/.118
|
-.483**/.233
|
-.445/.198
|
-.492/.242
|
-.486/.237
|
-.473/.224
|
Illness symptoms
|
-.188**/.035
|
-.239**/.057
|
-.258/.067
|
-.285/.081
|
-.282/.080
|
-.274/.075
|
*p < .01.
The reliability estimated by correction for attenuation formula (.726) was applied.
The reliability estimated by principal axis factoring (.594) was applied.
The reliability estimated by maximum likelihood (.607) was applied.
The mean reliability (.642) was applied.
Table 3b Variance Reduction in Related Variables: Single-item versus Multiple-item General Self-efficacy Scale
|
Variance reduction compared with NGSE-7
|
|
|
|
|
|
|
Positive emotions
|
.069
|
.035
|
.012
|
.015
|
.021
|
Life satisfaction
|
.060
|
.032
|
.018
|
.020
|
.024
|
Negative emotions
|
.028
|
.006
|
-.006
|
-.005
|
-.001
|
Task stress
|
.033
|
-.021
|
-.053
|
-.049
|
-.040
|
Perceived stress
|
.115
|
.035
|
-.009
|
-.004
|
.009
|
Illness symptoms
|
.022
|
-.010
|
-.024
|
-.023
|
-.018
|
M
|
.055
|
.013
|
-.010
|
-.008
|
-.001
|
SD
|
.035
|
.025
|
.026
|
.025
|
.025
|
The reliability estimated by correction for attenuation formula (.726) was applied.
The reliability estimated by principal axis factoring (.594) was applied.
The reliability estimated by maximum likelihood (.607) was applied.
The mean reliability (.642) was applied.
Discriminative Power
Two steps were conducted to classify the participants based on six theoretically relevant constructs. First, a hierarchical cluster analysis was utilized to explore the optimal number of clusters. In this step, Ward’s method and the squared Euclidean distance were opted, meanwhile, participants were allocated into different clusters. The results showed that the first huge gap in the coefficient values from the agglomeration schedule indicated that the significant clusters occurred at stage 228, suggesting three clusters as the optimum solution. Second, a k-means cluster analysis was conducted using the means of three previous clusters as the initial cluster centres. Subsequently, the participants were relocated into the final three clusters. Importantly, a cross validation procedure was repeated five times to confirm the validity of this 3-cluster solution. The sample was randomly divided into two parts (115 and 116 participants, respectively). The similar two steps were conducted on two parts except that the initial centres for two parts in k-means cluster analysis both used the cluster means based on one part. Subsequently, the Cohen’s κ values were calculated to test the consistency between new and original clusters. The mean Cohen’s κ was .824 across five times cross validations, verifying the stable 3-cluster solution (46). Descriptive statistics and a one-way ANOVA was used to investigate the differences across clusters in six related constructs. The results were presented in Table 4. Homogeneity of variance could be assumed only in illness symptoms, in which the Bonferroni post hoc tests were conducted. Where homogeneity could not be assumed, the Brown-Forsythe F ratio was reported and Games-Howell post hoc tests were used. The ANOVA results revealed significant differences between three clusters across all six relevant constructs (p < .001). A post-hoc test showed that all comparisons were significant (p < .01). Thus, based on the score patterns, the three clusters were named as negative (high on four negative constructs, low on two positive constructs), balanced (no obvious tendency) and positive (high on two positive constructs, low on four negative constructs).
A one-way ANOVA was conducted to assess the discriminative power of the GSE-SI across three clusters. As expected, GSE measured by two GSE scales both increased from negative group to positive group. As Table 5 shown, both GSE-SI and NGSE-7 had significantly discriminated between three clusters in an extremely similar pattern (F (2, 228) = 9.127 and 21.122, respectively, p < .001).
Table 4 Discriminations on Related Variables across Clusters
|
Negative
(n = 60)
|
Balanced
(n = 115)
|
Positive
(n = 56)
|
F ratio
|
p-Value
|
|
M
|
SD
|
M
|
SD
|
M
|
SD
|
|
|
Positive emotions
|
2.858
|
.632
|
3.387
|
.557
|
4.013
|
.579
|
54.686
|
< .001
|
Life satisfaction
|
3.250
|
.590
|
3.784
|
.529
|
4.329
|
.471
|
59.531
|
< .001
|
Negative emotions
|
2.883
|
.586
|
2.216
|
.467
|
1.804
|
.490
|
64.083
|
< .001
|
Task stress
|
3.643
|
.476
|
3.024
|
.427
|
2.264
|
.487
|
125.662
|
< .001
|
Perceived stress
|
3.258
|
.283
|
2.727
|
.208
|
2.305
|
.260
|
201.278
|
< .001
|
Illness symptoms
|
3.121
|
.812
|
2.554
|
.605
|
1.786
|
.581
|
59.732
|
< .001
|
Table 5
Discriminations across Clusters: Single-item versus Multiple-item General Self-efficacy Scale.
|
Negative
(n = 60)
|
Balanced
(n = 115)
|
Positive
(n = 56)
|
F ratio
|
p-Value
|
|
M
|
SD
|
M
|
SD
|
M
|
SD
|
|
|
GSE-SI
|
3.870
|
.853
|
4.170
|
.652
|
4.430
|
.657
|
9.127
|
< .001
|
NGSE-7
|
3.700
|
.594
|
3.993
|
.453
|
4.289
|
.424
|
21.122
|
< .001
|