My research is regarding the impact of the information consumed through social media (SMC) on creation of information in Social Media (SCMr). Based on literature, a five point likert scale (anchored from -2 to 2) was used to record 9 variables. Using Principal Axis Factoring and Promax rotation, the variables were nicely loaded as 2 factors as required. After checking AVE, CR, and other Tests, the Convergent, Discriminant validity and the construct validity also supports the CONSTRUCT validity. The Factor Analysis Scores were generated using Regression Method.
However, I am interested to perform Regression Analysis to see the dependency of Social Media Consumption on Social Media Creation. When I tried to do the same, although R-square value and Significant values looks great. The Unstandardized Coefficient value reports a value like "0.693E-71". Hence, I am unable to reach any conclusion.
PLEASE pardon my limited knowledge regarding the analysis.
Yes, you certainly can do that, but the only reason to use factor scores is if there is substantial variation in your factor loadings. Otherwise, it is just as effective to simply sum the items. The reason is that calculating factor scores is a matter of assigning weights to the variables, according to how important they are in the factor analysis. But if all the loadings are similar, then there is no need for the weights.
The reason for the small size of your coefficient may be due to the range of your computed scale. Run descriptive statistics on the scales that you created to check this. FYI, if the coefficient is significant, then this isn't really a problem, since the standardized coefficient will be more interpretable that they so-called "metric" coefficient.
On another point, using an orthogonal rotation is seldom realistic with social science data. I suggest that you try the factor analysis again, using a oblique rotation (i.e., correlated factors). To see the point of this argument, consider the names that you have given your factors, and then ask if those concepts would be uncorrelated in the real world.
Dear David, Thank you for your valuable suggestions. I have followed your advice and performed descriptive and regression. The problem still exists. If you could kindly spare me some time and overview at the files one showing the descriptive of the two factors and another regression value. I would be grateful for your suggestions.
Your regression accounts for over a third of the variance, so that is highly significant.
Your scales have been standardized so that they, effectively, have a mean of zero and and std. dev. of one. This means that the associated regression coefficients will tell you how much the dependent variable changes for every one std. dev. increase in that scale.
Dear David,
Thank you for your help. I just want to know that while sending the findings for publication. Do I need to report the unstandardized coefficient as you already know the value is terribly low?
- Badges
- Science topic
- Similar topics
- Social Science
- Media
- Social Media