Amos - a ML model with bootstrap for non non normal distrinution ?

Hi,

my problem is that I try to construct an Amos model for my my master's thesis results in psychology and all the variables of my study are non-normal. I worked on univariate normality to check skewness and Kutosis with transforming variables using a natural log function. I screened multivariate outliers with detecting with cook's distance, I have deleted a few ones.

Now, even i worked on normality, i must decide what method i'm going to use between Maximum Likelihood (ML) or Asyntoticaly Distribution Free (ADF).

I know i must use a bootstrap procedure because i'm going to study indirect efffecs with mediations. However, I doubt on the method. I've read that ML required a good normality and the second one did not but it needs a lot of subjects (1000-2000).

I've read that in spite of the non normality, ML could be used with a bootstrap and that it needed to screen bootstrap distribution et results (to evaluate the normality after).

I did a ML model with bootstrap and this one seems to be adjusted. However, the mardia's coefficient is 5,925 (c.r. 3,390).

Chi-square 25.22
Degrees of freedom 13
Sig. .02
Normed chi-square (chi-square/df) 1.94
Tucker-Lewis index (TLI) .92
Comparative fit index (CFI) .96
Root mean squared error (RMSEA) .08
< 90% confidence int erval .03
> 90 % 90% confidence interval .12
Standardized root mean square residual .07

Bootstrap were fixed at 1000 with a confidence at 0.05

My question is : is my model is ok ? what about the non normality when we screen mardia's coefficient, bootstrap distribution and Observations farthest from the centroid (Mahalanobis distance) that i have some problems to interpret (see tables under).