I have run 2 path analysis models, where I constrained a few parameters to 0 in the second model. As expected, the chi-square and SRMR worsened slightly, but the CFI and RMSEA actually improved slightly when adding constraints. Is this possible or reflective of a common error? I have pasted both models and some corresponding fit index results below (where the only differences between the 2 models are the constraints are added to relationships when predicting Y2 and Y4). Thank you!
Model 1 - Fewer constraints
MODEL: Y1 ON X1 X2; Y2 ON X1 X2; Y3 ON X3 X4; Y4 ON X3 X4; X1 WITH X3@0X4@0; X3 WITH X2@0;
Loglikelihood H0 Value -2324.189 H1 Value -2318.040
Chi-Square Test of Model Fit 12.298 RMSEA 0.028 CFI 0.981 TLI 0.962 SRMR 0.062