i have run a SEM model which has number of explanatory variables. in order to improve the GOF on my model i remove those variables which has low factor loading. but this affects the factor loading of remaining variables. what is the technical explanation of this?
Generally speaking, factor loading values for a set of indicators shouldn't move around too much with the removal of a limited number of indicators; they are interpreted as estimates of the true population loading values of those indicators for that particular latent variable. Thus, as long as the latent variable you are modeling stays the same, so too (to a large extent) should the estimated loading values for its indicators. If you are seeing substantive shifts in the value of remaining factor loadings, it may be the case that you are removing so many indicators that you are actually changing the substance of the latent variable you are modeling.
A rough conceptual example of this in action might be if you had a set of five indicators of Negative Emotion (e.g., "Nervous", "Worried", "Sad", "Depressed", "Concerned"). If, however, you removed "Sad" and "Depressed", the substance of you latent variable might actually be closer to representing Anxiety, specifically (i.e., "Nervous", "Worried", "Concerned"), than the more general Negative Emotion construct.
So, my intuition is that you are simply removing so many indicator variables that you are no longer measuring the exact latent construct you were initially modeling, hence the changed loading values. Then again, perhaps the amount of change you're seeing in loading values isn't that large, in which case, I wouldn't worry too much about it.