As my data has multivariate non-normality even after reduce the sample size based on M.D. (from 402 to 348), I did bootstrapping using Bollen-styne with 2000 iterations. However, the p-value is 0.002 which indicates the model is incorrect. However, based on the various model fit indices (RMSEA, GFI, CFI,..) the model has a good fit. The biases in standard errors are also very small. In this case, can I accept the results of the bootstrapping?
Hi Balu Ramoo, First thing, as per (Klein 2011), You should understand that higher the number of sample size lower the chances to get chi-square p-value insignificant. It mean you have no need to worry. Due to this limitation of chi-square test, researchers used other model fit indices to check model fit. If your model indices showing perfect fit, then in my opinion, you should accept results with bootstrapping.
Hi Adil,
Thank you for your reply. However, the p-value that I referred in my question was for the Bollen-Stine Bootstrap. The Amos output for the default model produced the following 'Testing the null hypothesis that the model is correct, Bollen-Stine Bootstrap p=0.002'. In this case, do I need to conclude that the model is incorrect?
hi Balu,
i think your model is fit enough and you dont need to worry, if all fit model standard result passed the minimum requirements, and only the Bollen-Stine Bootstrap test result didnt pass the minimum requirement, just shows all result and explains those condition.
Dear Balu Ramoo , Bollen-Stine Bootstrap p=0.002 means your model is fit but poor fit. But you still can proceeds to check the results. For detail kindly check out the following sources:
Using the Bollen-Stine Bootstrapping Method for Evaluating Approximate Fit Indices by Hanjoe Kim and Roger Millsap
Applying the Bollen-Stine Bootstrap for Goodness-of-Fit Measures to Structural Equation Models with Missing Data by Craig K. Enders
Calculation of Bollen-Stine significance value in AMOS at http://www-01.ibm.com/support/docview.wss?uid=swg21592992
https://www.youtube.com/watch?v=DktTdTnFkCI
If the p value is below 0.05 and does not register at at least above 0.000 (eg. p value = 0.002), This indicates the Χ2/DF ratio is outside the 1-3 range and so the model is not acceptable. Hence, only if there is a lowish p value -> we bootstrap. Then if the Bollen-stine p is still below 0.05 at 200(0) iterations -> this means the model is not acceptable.
If the data is non-normal, you can try transformations eg SQRT, but you lose some meaning across the model.
Maybe you should check the construct's measurement items (indicators) for interaction effects, and this may improve the model (i.e. x2/DF & p value)
Also, you have plenty of data, and outliers can affect model - so try to eliminate them one-by-one -> the worst one first, then rerun the model (with the original data set less one outlier) and check for improvement. This can be repeated again, and again until the mahalanobis d2 dist comes down to the break point -> then stop and check if any of these smaller data sets work, and check is there a general trend improvement?
p-value is 0.002 represents that ur model reject H0. meaning thats a good model rather than poor. the rationale is that ML is used to run H0, therefore, once p values from Bollen stine less than .005 which means good model.
Hi, I came across this interesting topic.
My measurement model on AMOS shows poor fit indices. It gets chi2= 0.00, GFI= 0.77, AGFI=0.73, CFI=0.83, RMSEA=0.93, and measure
I should mentioned that all of the observed variables are skewed (around -1 to -2) with high kurtosis (ranging between 3 and 7).
However, when I run Bollen-Stein bootstrapping, it says p=.111 and the model is correct.
How I can interpret this results?
Any backup reference would be highly appreciated.
Thanks
- Badges
- Science topic