Hi, I am testing model fit for cfa. The model fit indices look good. But I cannot decide whether probability level ,006 is a good thing or a bad thing:) I would love to hear your recomendations. (sample size 123)
Result (Default model)
Minimum was achieved
Chi-square = 202,789
Degrees of freedom = 155
Probability level = ,006
NFI, 894;TLI ,966; CFI ,972; RMSEA ,050; SRMR ,0483
Given the fit indices, I would think that your model looks alright. Two things strike me, in view of your sample size:
First, some people consider the power of the Chi-square test to be a problem, because larger sample sizes increase the probability of rejecting the model. In your case, this problem does not seem to apply, which is a bit odd.
Second, I am wondering how you got the model to converge with such a small sample. Could you elaborate on the model structure?
In general, evaluating model fit is complicated, so that trying to simplify it to a dichotomous decision is not helpful. I would recommend this chapter:
West, S. G., Taylor, A. B., & Wu, W. (2012). Model fit and model selection in structural equation modeling. In R. H. Hoyle (Ed.), Handbook of structural equation modeling, 209-231. Guilford Press.
Hi Heiko, thanks for the answer. I am sending results for model fit. I had to remove two variables because of discriminant validity issues . Actually this is the first time i am using this structural equation modelling and i really have no idea how the model is not rejected. The only thing i have not looked is multicollinearity. I have to learn to test multicollinearity for ordinal variables. And i have to find where those R squared values are in AMOS output( i am that new:)!!).
As far as I can tell, you model looks fine. Of course, this is only my individual assessment, but I can't find any of the usual issues here.
I am not sure where (or if) AMOS gives you R-squares, but keep in mind that those would pertain to the indicators, not the latent variables.
I tested the model for common method bias, and i think it collapsed.
Hi,
you're giving us inconsistent information. The model which you have attached reports a chisquare value of 146.05. The degrees of freedom (which AMOS does not report. Bad!) should be 120, resulting in a p-value of 0.05. I guess you are telling the p-value of the null model which is the "worst possible comparison". The null model is no model at all, contains only the variances of the indicator but no structure.
Your model chisquare is on the border. Combined with the low n your are in a very ambigous situation: One one side the low n decreases power (as Heiko noted) - hence problems exist will not show up. On the other hand, the chisquare test has shown to falsely reject models with low n. However, there are corrections available. You should definetely use them. Herzog and Boomsma programmed a small R-function in which you type in the chisquare, dfs and the N and you get a sample size adjusted chisquare (and if you want, some fit indices).
Herzog, W., & Boomsma, A. (2009). Small-sample robust estimators of noncentrality-based and incremental model fit. Structural Equation Modeling, 16, 1–27.
Here is the R function:
http://www.ppsw.rug.nl/~boomsma/swain.pdf
Best,
Holger
Dear Heiko, Alvin and Holger,
Thank you very much for your responses. I decided to abandon that model and go for a more simpler model. Yet this time, i have a p=,750, and a cfi=1 (common method bias corrected). Other indices look good as well as the AVE and CRs (attached in the file). Do you think this one is a go:)?
Thanks Holger, I have downloaded R. But i don't know how to transfer adjusted chi square to AMOS. Is there a way to do that?
Serkan, I am a bit puzzeled. The pdf explains it all....
But it may be thad i do not understand what you mean wiht "transfer adjusted chisquare". You create the function swain.R in R. Than you type in the relevant information (e.g., swain(N=50, p=25, d=265, TML=400, TMLi=1100) and you get the corrected chi-square (its printed on p. 4 of the pdf. This informs you if your model is consistent enough with your data. There is nothing to transfer. The relevant line is "Swain-corrected chi-square test statistic: 321.73, p-value: 0.0097438"
HTH
Holger
Serkan, I did not see your posting about the new model. From a practical perspective, this may work. But in a strict sense there is the problem that you change the model in face of the data. Hence, the chisquare test is not really a test any more.
One important addon: Forget about including a common method factor. This - unfortunately - does not correct for common method variance (a better term should be "method induced covarianc" because there is no such thing as "common variance" :) but can make things worse as the general factor squeeze out true latent covariance from the latent variables
See:
Conway, J. M., & Lance, C. E. (2010). What reviewers should expect from authors regarding common method bias in organizational research. Journal of Business and Psychology, 25(3), 325–334. doi:10.1007/s10869-010-9181-6
Richardson, H. A. H. A., Simmering, M. J., & Sturman, M. C. (2009). A tale of three perspectives: Examining post hoc statistical techniques for detection and correction of common method variance. Organizational Research Methods, 12(4), 762–800. doi:10.1177/1094428109332834
Simmering, M. J., Fuller, C. M., Richardson, H. a., Ocal, Y., & Atinc, G. M. (2014). Marker variable choice, reporting, and interpretation in the detection of common method variance: A review and demonstration. Organizational Research Methods, 18(3), 473–511. doi:10.1177/1094428114560023
HTH
Holger
Oh, I saw, AMOS indeed prints the df. My apologies. The program is better than I thought :)
Sorry for the late reply Holger. It appears i cannot add answer with iphone:) What you say about CMB is quiet a relief, because i had to abandon the previous model for that. one colleague told me that some reviewers ask for CMB, and i believe many nice models are sacrificed for this. Personally i dont have too many bullets to shoot when i can manage a survey to avoid CMB. Anyway, this new model will examine the inflluence of trust and techn. acceptance factors on the intention to share tacit knowledge on social media using SEM. The variables and items are well grounded in the literature, but of course not as exciting as the previous one.
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques
- Structural Equation Modeling