Hii...everyone...Need your suggestions on the following problem.
The latent measurement model consists of one independent variable (IV) and two dependent variables (DVa and DVb). The causal relation between IV and DVa & DVb is mediated by a mediating variable (M).
Data (n=367) has been cross-checked (missing data, normality, multicollinearity, reliability and validity of constructs etc.) before running the structural equation modeling (SEM) in AMOS. The data seems to be free from any discrepancies.
However, on considering the model :
IV (3 indicators)-->M (4 indicators)--> DVa (5 indicators)/ DVb (4 indicators) as mediation/indirect effect and,
IV (3 indicators)--> DVa (5 indicators)/ DVb (4 indicators) as direct effect.
Following ambiguous results have obtained:
1) Path coefficient between IV-->DVa is greater than 1 (i.e.,std. beta value=2.0)
2) Path coefficient between M-->DVa is also greater than 1 ( std.beta value =-1.65)
Both the paths are significant at p
Him Sayoni,
maybe you see the unstandardized coefficients (B and not beta)? Try to select standardized coefficients.
Otherwise look this video: https://www.youtube.com/watch?v=Vx24KFf-rAo
I think it is a similar issue and the author proposes an explanation. I can also recommend askin your question in comments there. Gaskin is answering comments fastly and very competent.
Sincerely,
Eugene
@Eugene...Thanks for your reply...actually, I saw the standardized beta coefficients only and they appeared to be greater than 1. That was my concern. I emailed professor Gaskin and he gave a prompt reply. Nevertheless, the problem was resolved after I did some simple modifications in the model. Thanks again..!! :-)
@Hamid: Thanks :-)
@Sayoni Hi, could you please share how you resolved the problem ? TIA
@Chandrama:
1. Check for the multicollinearity or negative error variance within the variables, it could be one of the reasons for std beta values greater than 1.
If multicollinearity exists, try to substitute one of the variables with any other conceptually related variable. In case of negative error variance, fix the value with a small positive value like 0.005.
2. Opt for cross-loadings of all the items in model constructs/variables (say, for example, your regression model has 3 variables, X, M and Y each consisting of 5, 5 and 5 items respectively. Perform PCA in SPSS (Varimax rotation) fixed with 3 factors (as you have 3 variables (X, M, Y) (Note: igen value should be >1). If any of the items cross load with other variables (say the item of M also loads on Y) remove those items. This approach may reduce the likelihood of getting std beta >1.
Hope these facts work for your model.
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques
- Structural Equation Modeling