Exploratory Factor analysis
EFA is done through a process called factor analysis and also component analysis which is used to reduce a given set of observed variables or factors a reasonable level that best explains latent variable(s) (Spearman, 1904). Some of the important tests conducted during EFA are the Kaiser-Meyer-Olkin Measure of Sampling Adequacy test (KMO), Bartlett's Test of Sphericity (Approx. Chi-Square, Df. Sig.), Communalities, and Principal Component Analysis. KMO is used to test for sample adequacy and should be above 0.5 for the sample to be adequate (Tabachnick & Fidell, 2001; Yong & Pearce, 2013). On the other hand, Bartlett's Test of Sphericity (Approx. Chi-Square, D.f., Sig.) which is used to test for homogeneity of samples. It ensures that there is similarity in the variances of a group of samples (Bartlett, 1937). Communalities show the variance in a latent variable that is explained by a given observed variable (Costello & Osborne, 2005). The higher the communality, the better that observed variable explains its latent variable (Hatcher, 1994), however, a communality of 0.4 and above is generally considered to be good (Costello & Osborne, 2005). Further, Principal Component Analysis is used to orthogonally transform a set of related observed variables in groups of factors, also known as components (Jérôme, 2014; François, Sébastien & Jérôme, 2009; Jolliffe, 2002).
Data were analyzed using exploratory factor analysis with Extraction Method of Principal Component Analysis and Rotation Method of Varimax with Kaiser Normalization in order to extract the most important factors that measured the study variables. Factors with Eigen values >1 and factor loadings >0.5 were retained in the commonality and rotated component matrix. This validated the questionnaire in terms of convergent validity and discriminant validity (Campell & Fiske, 1959).
For convergent validity, determinant with sig.>0.00, commonalities loadings >0.5 indicated convergence of items in measuring the same variable. For discriminant validity, Rotated Component Matrix distinct factors with loadings of above 0.5 indicated discrimination of factors from each other. In this study, factor analysis was performed on all latent variables as presented in the following section.
EFA for Outcome Expectations
A total of 7 items were listed to measure Outcome Expectations. Item correlation matrix produced a Determinant = .013 meaning that all items converged and were related in measuring Outcome Expectations. The KMO was used to measure sampling adequacy. A KMO =.809 meant that the study sample was adequate. On the other hand, Bartlett's Test of Sphericity was used to measure the significance of the sample. Bartlett's Test of Sphericity Approx. Chi-Square = 2085.746, D.F. =10, Sig=.000 meant that the sample was significant. Table 2 presents KMO and Bartlett's Test results for Outcome Expectations.
Table 2: KMO and Bartlett's Test for Outcome Expectations
Kaiser-Meyer-Olkin Measure of Sampling Adequacy.
|
.809
|
Bartlett's Test of Sphericity
|
Approx. Chi-Square
|
2085.746
|
D.f.
|
10
|
Sig.
|
.000
|
Communalities test for Outcome Expectations
In addition the above descriptive, Communalities and determinant tests were used to examine convergent validity of Outcome Expectations as seen in Table 3.
Table 3: Communalities for Outcome Expectations
|
Initial
|
Extraction
|
Using social media on health related matters makes me a better person
|
1.000
|
.872
|
Using social media on health related matters makes me more acceptable amongst my peers
|
1.000
|
.913
|
My peers will trust me if I use social media on health related matters
|
1.000
|
.638
|
I will not be rejected by my peers if I use social media on health related matters
|
1.000
|
.718
|
I will not be punished by my family if I use social media on health related matters
|
1.000
|
.871
|
Average communality
|
|
0.802
|
Results in Table 3 above reveal that all the items measured Outcome Expectations since they all have factor loadings above 0.40 and determinant of .013. Hence convergent validity was achieved on Outcome Expectations.
Rotated Component Matrix for Outcome Expectations
Data were analyzed using Principal Component Analysis extraction methods with Varimax with Kaiser Normalization rotation method in order to identify the items that most explained Outcome Expectations. The results are presented in Table 4.
Table 4: Component Matrix for Outcome Expectations
|
Component
|
Outcome Expectations
|
Using social media on health related matters makes me more acceptable amongst my peers
|
.956
|
Using social media on health related matters makes me a better person
|
.934
|
I will not be punished by my family if I use social media on health related matters
|
.933
|
I will not be rejected by my peers if I use social media on health related matters
|
.847
|
My peers will trust me if I use social media on health related matters
|
.799
|
Eigen Value
|
4.012
|
Total variance
|
80.237
|
Percentage Total Variance
|
80.237
|
Rotated Component Matrix results in Table 4 show that 5 factors explain Outcome Expectations with (Eigen Value = 4.012, Total Variance = 80.237, Percentage Total Variance = 80.237). these are; Using social media on health related matters makes me more acceptable amongst my peers (Factor loading=.956); Using social media on health related matters makes me a better person (Factor loading=.934); I will not be punished by my family if I use social media on health related matters (Factor loading=.933); I will not be rejected by my peers if I use social media on health related matters (Factor loading=.847); My peers will trust me if I use social media on health related matters (Factor loading=.799).
EFA for External Locus of Control
Similarly, data were collected and analyzed on a total of 8 items listed under External Locus of Control. Item correlation matrix for External Locus of Control produced a Determinant = .035, meaning that all items converged and were related in measuring the variable. The KMO was used to measure sampling adequacy for this variable. A KMO = .754 was obtained, meaning that the study sample was adequate. On the other hand, Bartlett's Test of Sphericity was used to measure the significance of the sample. Bartlett's Test of Sphericity Approx. Chi-Square = 1190.201, D.F. =10, Sig=.000 meant that the sample was significant. Table 5 presents KMO and Bartlett's Test results for Cognitive Factors.
Table 5: KMO and Bartlett’s Test for External Locus of Control
Kaiser-Meyer-Olkin Measure of Sampling Adequacy.
|
.754
|
Bartlett's Test of Sphericity
|
Approx. Chi-Square
|
1190.201
|
D.f.
|
10
|
Sig.
|
.000
|
Communalities test for External Locus of Control
Further, Communalities and determinant tests were used to examine convergent validity of items under External Locus of Control. Table 6 presents the results.
Table 6: Communalities for External Locus of Control
|
Initial
|
Extraction
|
I am not in control of the consequences of my actions while using social media
|
1.000
|
.707
|
I achieve less by using social media
|
1.000
|
.803
|
I have low morale to learn new things on social media
|
1.000
|
.721
|
I consider myself lucky to be using social media
|
1.000
|
.559
|
I am not responsible for the bad things that happen to me while using social media
|
1.000
|
.643
|
Average communality
|
|
0.687
|
Results in Table 6 reveal that all the items measured External Locus of Control since they all have factor loadings above 0.40 and determinant of .035. This means that convergent validity was achieved on the variable.
Component Matrix for External Locus of Control
Data were analyzed using Principal Component Analysis extraction methods with Varimax with Kaiser Normalization rotation method in order to identify the items that most explained Internal Locus of Control. The results are presented in Table 7.
Table 7: Component External Locus of Control
|
Component
|
External Locus of Control
|
I do not maintain good relations on social media
|
.998
|
I am unable to help myself when faced with challenging situations on social media even if I possess the ability to do so
|
.348
|
I do not think about the consequences of my actions before doing them on social media
|
.223
|
I am not responsible for the bad things that happen to me while using social media
|
.192
|
I consider myself lucky to be using social media
|
.130
|
Eigen Value
|
22.491
|
Total variance
|
70.755
|
Percentage Total Variance
|
70.755
|
Results in Table 7 show that the most important factors explaining External Locus of Control are; I do not maintain good relations on social media (Factor loading =.998), I am unable to help myself when faced with challenging situations on social media even if I possess the ability to do so (Factor loading =.348), I do not think about the consequences of my actions before doing them on social media (Factor loading =.223), I am not responsible for the bad things that happen to me while using social media (Factor loading =.192), I consider myself lucky to be using social media (Factor loading =.130).
EFA for Mental health
A total of 25 items grouped in four constructs including skills, practice, observational learning and moral degeneration were listed to measure mental health. Item correlation matrix produced a Determinant =6.806E-011 meaning that all items converged and were related in measuring Mental health. The KMO was used to measure sampling adequacy. A KMO =.867 meant that the study sample was adequate. On the other hand, Bartlett's Test of Sphericity was used to measure the significance of the sample. Bartlett's Test of Sphericity Approx. Chi-Square = 8197.645, D.F. =153, Sig=.000 meant that the sample was significant. Table 8 presents KMO and Bartlett's Test results for mental health.
Table 8: KMO and Bartlett's Test for Mental health
Kaiser-Meyer-Olkin Measure of Sampling Adequacy.
|
.867
|
Bartlett's Test of Sphericity
|
Approx. Chi-Square
|
8197.645
|
D.f.
|
153
|
Sig.
|
.000
|
Communalities test for mental health
In addition the above descriptive, Communalities and determinant tests were used to examine convergent validity of mental health as seen in Table 9.
Table 9: Communalities for Mental health
|
Initial
|
Extraction
|
I have acquired health skills via social media
|
1.000
|
.819
|
I have learned how to treat diseases via social media
|
1.000
|
.721
|
I have the desire to do the health issues I see other influential people in society doing via social media
|
1.000
|
.802
|
I train myself on doing the health related things that I see and like on social media
|
1.000
|
.847
|
I try to do the health issues as I am told to do via social media
|
1.000
|
.805
|
I seek sexual pleasures via social media
|
1.000
|
.767
|
I have learned how to smoke by observing other people’s smoking images or videos via social media
|
1.000
|
.856
|
I have learned how to consume alcohol by observing other people’s images or videos drinking it via social media
|
1.000
|
.787
|
I have learned how to access sexual partners using social media because observing other people doing it
|
1.000
|
.585
|
I have learned how to make money by giving sexual pleasures via social media through observing others
|
1.000
|
.831
|
I smoke because of the information I have consumed over time via social media
|
1.000
|
.831
|
I use drugs because of the information I have consumed over time via social media
|
1.000
|
.826
|
I drink alcohol because of the information I have consumed over time via social media
|
1.000
|
.744
|
I use pornography because of the information I have consumed over time via social media
|
1.000
|
.643
|
I am gay because of the information I have consumed over time via social media
|
1.000
|
.869
|
I have multiple sex partners because of the information I consume via social media
|
1.000
|
.966
|
I know of someone who obtained sex via social media
|
1.000
|
.818
|
I know of someone who engages in commercial sex via social media
|
1.000
|
.967
|
Average communality
|
|
0.805
|
Results in Table 9 above reveal that all the items measured mental health since they all have factor loadings above 0.40 and determinant of 6.806E-011. Hence convergent validity was achieved on mental health.
Rotated Component Matrix for Mental health
Rotated Component Matrix shows that all the four components explained Mental health namely; observational learning (Percentage Total Variance=37.207), Moral Degeneration (Percentage Total Variance=55.107), Practice (Percentage Total Variance=70.440) and Skills (Percentage Total Variance=80.477). Hence discriminant validity was achieved. Table 10 presents the results.
Table 10: Rotated Component Matrix for Mental health
|
Component
|
Observational learning
|
Moral degeneration
|
Practice
|
Skills
|
I have learned how to smoke by observing other people’s smoking images or videos via social media
|
.869
|
|
|
|
I use drugs because of the information I have consumed over time via social media
|
.865
|
|
|
|
I have learned how to make money by giving sexual pleasures via social media through observing others
|
.836
|
|
|
|
I smoke because of the information I have consumed over time via social media
|
.829
|
|
|
|
I have learned how to consume alcohol by observing other people’s images or videos drinking it via social media
|
.815
|
|
|
|
I seek sexual pleasures via social media
|
.739
|
|
|
|
I use pornography because of the information I have consumed over time via social media
|
.710
|
|
|
|
I drink alcohol because of the information I have consumed over time via social media
|
.704
|
|
|
|
I know of someone who obtained sex via social media
|
.692
|
|
|
|
I have learned how to access sexual partners using social media because observing other people doing it
|
.632
|
|
|
|
I know of someone who engages in commercial sex via social media
|
|
.886
|
|
|
I have multiple sex partners because of the information I consume via social media
|
|
.884
|
|
|
I am gay because of the information I have consumed over time via social media
|
|
.721
|
|
|
I train myself on doing the health related things that I see and like on social media
|
|
|
.852
|
|
I have the desire to do the health issues I see other influential people in society doing via social media
|
|
|
.832
|
|
I try to do the health issues as I am told to do via social media
|
|
|
.788
|
|
I have acquired health skills via social media
|
|
|
|
.901
|
I have learned how to treat diseases via social media
|
|
|
|
.820
|
Eigen Value
|
6.697
|
3.222
|
2.760
|
1.807
|
Total variance
|
37.207
|
17.900
|
15.333
|
10.037
|
Percentage Total Variance
|
37.207
|
55.107
|
70.440
|
80.477
|
The effect of Outcome Expectations and External Locus of Control on the mental health of social media users in Sub-Sahara Africa
Multiple Hierarchical Regression analysis was used to determine the predicting power of outcome expectation and External Locus of Control on health. Gender, Age, Level of education, Marital Status, and Country of Residence were treated as extraneous or control variables. Table 11 presents the results.
Table 11: Regression for Mental health
|
Model 1
|
Model 2
|
Model 3
|
Variable
|
B
|
Beta
|
B
|
Beta
|
B
|
Beta
|
(Constant)
|
4.037**
|
|
3.16**
|
|
1.70**
|
|
Gender
|
-0.33**
|
-0.22**
|
-0.44**
|
-0.30**
|
-0.30**
|
-0.20**
|
Age
|
0.02
|
0.02
|
0.11*
|
0.11*
|
0.09
|
0.09
|
Education
|
0.03
|
0.08
|
0.06*
|
0.16*
|
0.03
|
0.08
|
Marital Status
|
0.02
|
0.02
|
0.04
|
0.03
|
-0.04
|
-0.03
|
Country of Residence
|
-0.08
|
-0.07
|
-0.05
|
-0.04
|
-0.18**
|
-0.18**
|
Outcome expectation
|
|
|
0.21**
|
0.39**
|
0.25**
|
0.47**
|
External Locus of Control
|
|
|
|
|
0.43**
|
0.46**
|
R
|
.240
|
.434
|
.594
|
R2
|
.058
|
.188
|
.352
|
Adj R2
|
.044
|
.175
|
.339
|
R2 Change
|
.058
|
.131
|
.164
|
F Change
|
4.303
|
56.560
|
88.619
|
Sig. F
|
.001
|
.000
|
.000
|
F
|
4.303
|
13.579
|
27.204
|
Sig.
|
.001
|
.000
|
.000
|
|
**.Significant at 0.01
|
|
*. Significant at 0.05
|
As seen in Table 11, results in model 1 show that Control variables including Gender, Age, Education, Marital status, and Country of residence predict 4.4% of Mental health (Adj R2 =0.044). The relationship between Gender and Mental health is significant (Beta=-0.22**, P<.01). The relationship between Age and Mental health is not significant (Beta=0.02). The relationship between level of education and Mental health is not significant (Beta=0.08). The relationship between, Marital Status and Mental health is not significant (Beta=0.02). The relationship between Country of Residence and Mental health is not significant (Beta=-0.07).
Results in model 2 reveal that control variables together with Outcome Expectations predict 17.5% of Mental health (Adj R2=.175) while Outcome Expectation alone predicts 13.1% of Mental health (R2 Change = .131). Further, the relationship between Outcome Expectation and Mental health is significant at 99% confidence level (Beta=0.39**).
Results in model 3 reveal that control variables, Outcome expectation and External Locus of Control combined predict 33.9% of Mental health (Adj R2=.339). However, External Locus of Control alone predicts 16.4% of Mental health (R2 Change=.164). The results also show that External Locus of Control has a significant relationship with Mental health at 99% confidence level (Beta=0.46**).
Given the above, we see that Outcome Expectations and External Locus of Control together with control variables contributed 33.9% of the changes in mental health of social media users in Sub-Sahara Africa.
Confirmatory analysis
Table 12: Social media and mental health Model Fit Summary
X2
|
DF
|
P
|
X2/DF
|
GFI
|
AGFI
|
NFI
|
RFI
|
IFI
|
TLI
|
CFI
|
RMSEA
|
8.653
|
6
|
.194
|
1.442
|
.993
|
.968
|
.983
|
.940
|
.995
|
.981
|
.994
|
.035
|
|
|
|
Estimate
|
S.E.
|
C.R.
|
Beta
|
P
|
Hypothesis
|
External Locus of Control
|
<---
|
Outcome Expectations
|
-.230
|
.031
|
-7.511
|
-.363
|
***
|
H1 -rejected
|
Mental health
|
<---
|
External Locus of Control
|
.411
|
.049
|
8.336
|
.385
|
***
|
H2 –accepted
|
Testing of research hypotheses using SEM
H1: Outcome Expectations have a positive effect on the External Locus of Control of social media users in Sub-Sahara Africa.
Results in Table 12 reveal that the relationship between Outcome Expectations and External Locus of Control was significant and negative at 1% level of significance (Beta=-.363, P<0.001). This implies that an increase in the Outcome Expectations of social media users will reduce their External Locus of Control. In other words, if the expected outcome from learning new behaviors that affect mental health via social media are high then the reliance on others to learn the behavior reduces. This relationship could probably be attributed to the confidential nature of health related information which most people do not want to share easily via social media. Therefore H1 that stated that Outcome Expectations have a positive effect on the External Locus of Control of social media users in Sub-Sahara Africa was not supported.
H2: External Locus of Control positively affects the mental health of social media users in Sub-Sahara Africa.
As seen in Table 12, the relationship between External Locus of Control and Mental health was also found to be positive and significant at 99% confidence level (Beta=.385, P<0.001). This indicated that there is a high certainty of the existence of a relationship between External Locus of Control and Mental health. More to the relationship between External Locus of Control and Mental health, it suffices to mention that, social media users who are highly influenced by external factors such as social influence from friends and family are more likely to learn new behaviors that affect mental health from social media platforms. This finding is in support of the hypothesis H2 that External Locus of Control positively affects the mental health of social media users in Sub-Sahara Africa.