Marianne SB posted on Tuesday, January 20, 2015 - 11:58 pm
I have constructed a model with acceptable overall fit, with some significant and some non-significant paths between the constructs. Should the non-significant paths be constricted to zero? This results in a model that have overall poorer fit than the first model (probably because quite a few of the non-significant paths is pretty close to being significant).
Should I keep the first model, and report the nonsignificant paths in the model? How to interpret these paths when the estimate of a non-significant path is higher than the estimate of a significant path?
Marianne SB posted on Wednesday, January 21, 2015 - 12:09 am
Addition: In the last paragraph I refer to the standardised estimates. Thus, the variances of my constructs are quite different - most of the constructs have means around 2.0 and variances between 0.3 and 0.8, whereas the mean and variance of my main variable of interest is 1.5 and 0.06, respectively
Perhaps I should report the means, variances, keep the first model, and report unstandardised estimates?