I am completely new to SEM. I have tried reducing the co-variances by checking the modification indices. Please help me out with this problem.
Attaching the screenshot of my model for reference.
Your SEM model is totally wrong. That is why you do not get the output. Among the mistakes are:
1. Sub Construct namely Learner and Affordability has no residual
2. The dependent construct namely Perception has no residual
3. Stakeholders is not a latent construct since it has only one item
4. Correlating the measurement error for items across is wrong
5. The structural model should be drawn from left to right.
Your SEM model is totally wrong. That is why you do not get the output. Among the mistakes are:
1. Sub Construct namely Learner and Affordability has no residual
2. The dependent construct namely Perception has no residual
3. Stakeholders is not a latent construct since it has only one item
4. Correlating the measurement error for items across is wrong
5. The structural model should be drawn from left to right.
Thanks for your response Sir,
I have incorporated your comments.
I actually removed few variables to get a model fit, therefore a sub-construct had only one manifest variable.
When I added error terms, I am unable to execute.
I am getting the following error
Dear @Zainudin Awang
Is this the right model. I am able to run the model now.
1. Do not have a constraint parameter between perception and acceptance. Unless, you are doing for the multi-group CFA
2. Your acceptance construct has one variable (Q20). If you are doing the second order for acceptance construct, then, you must have more than one variable
3. Q20 should be not has correlation with Q22
I suspect that the error term Q18 correlating with the error term for Q22 is a problem. There are two other error terms for indicators (Q2 and Q11) of independent variables correlated with an error term for an indicator (Q20) of a dependent variable.
It seems to me that there maybe content overlap in the measures of your IVs and DVs.
What would happen to the model if you deleted Q11, Q18, and Q1?
Deleting Q11 leaves you with one item for the indicator variable Learner. But you would handle that the same way you handled having one question (Q20) for the indicator "stakeholders."
It can be okay to have a one indicator for a construct provided you have an estimate for the reliability (e.g., an alpha coefficient) for that indicator. There are ways to do it. Wanous estimated the reliability of one item that assesses job satisfaction. The alpha coefficient would reflect the influence of latent variable on the indicator. One minus alpha would reflect the impact of measurement error on the indicator.
I'm an old time Lisrel user. I would look at the modification indices, to see what they tell you about improving the fit of the model.
Your construct namely Acceptance has no items (as if no data for that construct).
One more thing - the residual of Q20 and the measurement error of Q22 should not be constrained as correlated.
I got it.
I use the hypothesized data to your model and get the output. See the attached file.
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques
- Structural Equation Modeling