I’m currently writing my research paper and I’m new to SmartPLS and SEM in general. I hope some nice people here will be able to give me suggestions. My research is about smartphone brands and loyalty. You have below the model’s latent and measurement variables, as well as paths.
Code: Select all
LATENT VARIABLE: Corresponding measurement variables
SATISFACTION: SAT1 SAT2 SAT3 SAT4 SAT5
RELATIONSHIP COMMITMENT: COM1 COM2 COM3 COM4 COM5
ACTIVE LOYALTY: ALO1 ALO2
PASSIVE LOYALTY: PLO1 PLO2 PLO3
CONSUMER RELATIONSHIP PRONENESS: CRP1 CRP2 CRP3
PRODUCT CATEGORY INVOLVEMENT: PCI1 PCI2 PCI3
EXTRAVERSION: EX1 EX2 EX3 EX4
AGREEABLENESS: AG1 AG2 AG3 AG4
CONSCIENTIOUSNESS: CO1 CO2 CO3 CO4
NEUROTICISM: NE1 NE2 NE3 NE4
OPENNESS: OP1 OP2 OP3 OP4
PATHS
SATISFACTION —> ACTIVE LOYALTY
SATISFACTION —> PASSIVE LOYALTY
SATISFACTION —> RELATIONSHIP COMMITMENT
RELATIONSHIP COMMITMENT —> ACTIVE LOYALTY
RELATIONSHIP COMMITMENT —> PASSIVE LOYALTY
CONSUMER RELATIONSHIP PRONENESS —> RELATIONSHIP COMMITMENT
PRODUCT CATEGORY INVOLVEMENT —> CONSUMER RELATIONSHIP PRONENESS
EXTRAVERSION —> CONSUMER RELATIONSHIP PRONENESS
AGREEABLENESS —> CONSUMER RELATIONSHIP PRONENESS
CONSCIENTIOUSNESS —> CONSUMER RELATIONSHIP PRONENESS
NEUROTICISM —> CONSUMER RELATIONSHIP PRONENESS
OPENNESS —> CONSUMER RELATIONSHIP PRONENESS
- I first looked at the individual reliability of each item, and many indicators of personality traits (e.g. NEUROTICISM, etc.) were below 0.55, so I removed them. As ALL the indicators of EXTRAVERSION were below 0.55, I removed the whole construct.
- Then, I evaluated composite reliability, convergent validity and discriminant validity, and tests were positive.
- I evaluated the structural model with bootstrap (500 samples) and found that 7 hypotheses could be validated (with at least p<0.05).
I decided to split my data in two groups:
- iOS users (146 cases)
- Android users (88 cases)
What I did:
- I looked at the individual reliability of each item, for each group separately. Surprisingly, for each group alone, the loadings were most of the time higher than for the two groups combined. For instance, when considering iOS users alone, I had 2 EXTRAVERSION items (EX1 and EX2) with loading above 0.55. When considering Android users alone … I also had 2 EXTRAVERSION items (EX1 and EX3) with loading above 0.55 !
- I created a new path model for each group, and removed in each path model items with loadings below 0.55 (for instance, for the iOS path model I removed EX3 and EX4, and for the Android path model I removed EX2 and EX4).
- Then, for path model, I evaluated composite reliability, convergent validity and discriminant validity only by looking at the group’s data. Tests were positive.
- I evaluated each structural model with bootstrap (500 samples). Surprisingly again, for the iOS users group, 8 hypotheses could be validated (with at least p<0.05) and for the Android users group, 8 different hypotheses could be validated (with at least p<0.05). I give you the results at the end.
Code: Select all
PATH ALL IOS ANDROID (p-value)
SATISFACTION —> ACTIVE LOYALTY 0,000* 0,000* 0,000*
SATISFACTION —> PASSIVE LOYALTY 0,032* 0,000* 0,043*
SATISFACTION —> RELATIONSHIP COMMITMENT 0,000* 0,000* 0,000*
RELATIONSHIP COMMITMENT —> ACTIVE LOYALTY 0,147 0,378 0,033*
RELATIONSHIP COMMITMENT —> PASSIVE LOYALTY 0,000* 0,000* 0,000*
CONSUMER RELATIONSHIP PRONENESS —> RELATIONSHIP COMMITMENT 0,000* 0,000* 0,000*
PRODUCT CATEGORY INVOLVEMENT —> CONSUMER RELATIONSHIP PRONENESS 0,000* 0,000* 0,052
EXTRAVERSION —> CONSUMER RELATIONSHIP PRONENESS N/A 0,006* 0,056
AGREEABLENESS —> CONSUMER RELATIONSHIP PRONENESS 0,064 0,396 0,001*
CONSCIENTIOUSNESS —> CONSUMER RELATIONSHIP PRONENESS 0,000* 0,003* 0,008*
NEUROTICISM —> CONSUMER RELATIONSHIP PRONENESS 0,153 0,168 0,135
OPENNESS —> CONSUMER RELATIONSHIP PRONENESS 0,243 0,283 0,235
*p<0.05