Strictly speaking no. To do statistical inference you need to have a probability sample. Some times you can get an approx. of such a sample but your probability statements are very iffy.
I think that the concern should be whether or not you have represented all parts of the population.
I do not know about your application, but for inference from a sample to a finite population, most applications make use of probability of selection-based designs. Regression modeling is extremely useful in assisting estimation, but not often used alone. In the latter case, independent variable data would have to be known for the entire population, which would be a guide as to whether or not the population is covered well. In any case, stratification of the population into segments which fit together better, say to fit under a given model, sounds like the concern you would also have here: are data predicted under a good model in each case?
The title of your article sounds very appropriate for this question. But I see that it is copyrighted by the Journal of Applied Structural Equation Modeling. I did not see anything saying it was uploaded with permission, so it would be good if you can let us know if that is the case, as I'd think a number of people might want to look it over. Otherwise, could you summarize what it says there which addresses this question best? A little summary/comment would probably be helpful in any case.