Hi

The CFA model only check the covariance between the constructs, not the relation between them. It means that CFA shows how much the constructs in the model are correlated. In running the SEM, the regression weight and the in fact the quality of relationship between the construct is checked," how much and in which direction they have effect on DV" , but in CFA there is no DV or IV. More important is to check the significance of obtained weight in SEM.  

thank you Maryam kh for your valuable responde. 

i agree that there is a difference in the nature of relationship tested in correlation and regression. but, correlation also indicates the direction of relationship between the two variables(positive or negative). 

in simple my question is, if correlation between DV and IV is positive, but the regression coefficient of IV has negative sign, how to interpret it. 

thanks in advance

Are your IVs highly correlated with each other?

In multiple regression, where IVs are assumed to be independent of each other, multicollinearity in the Independent Variables means that information contained in one variable is also contained in the other variable (or variables, or combination of variables) That makes your coefficient estimates unreliable - they can vary a great deal, shifting from highly significant to non-significant, and even to changing sign, with small changes to the data.

In SEM we can partly account for correlation among the IVs by drawing correlations among them, but this does not completely remove the problems of multicollinearity with respect to the reliability of coefficients.

Bootstrapping will give you some ideas about the range of values that your coefficients could have.

thank you Prof. Hume F. Winzar

your explanation is very much useful to me in understanding the implications of multicollinerity and also how the SEM handles this issue. 

thanks a lot. 

i did bootstrapping in AMOS. but i could not understand how to calculate confidence intervals for the coefficient values. AMOS output does not display the confidence intervals for the regression estimates.  

please help me in this regard

Bootstrapping results should appear with the Estimates output.

The attached PDF will help with interpretation. It refers to an earlier version of AMOS, but it should be still appropriate.

Did you check the factor loadings for the constructs.  They could be negative in one scenario (CFA) and positive in another (SEM).  This would lead to differences in the relationships between the constructs since high scores on each would mean different things in each scenario