Skip to first unread message
Emmanuel Banda
Aug 7, 2018, 10:57:39 PM8/7/18
to lavaan
I have tried to make a few adjustments to this model but its failing give me the good fit even thought its able to give estimates. The first message its gives is some variances are negative. See the commands below and further the output.
semMed2 <- '
###Measuremement models
###Measuremement models
Underlying_causes =~bord+ m_educa+ mo_age
Basic_causes =~ HH_members+N_under5+ w_index
Basic_causes =~ HH_members+N_under5+ w_index
# direct effect
Haz ~ c*Basic_causes
# mediator
Underlying_causes ~ b*Basic_causes
Haz ~ a*Underlying_causes
# indirect effect
ab := a*b
# total effect
total := c + (a*b)
Underlying_causes ~ b*Basic_causes
Haz ~ a*Underlying_causes
# indirect effect
ab := a*b
# total effect
total := c + (a*b)
###Residual correlation
bord~~mo_age
mo_age~~HH_members
'
bord~~mo_age
mo_age~~HH_members
'
fitsem2 <- sem (semMed2, data=data4R, estimator = "MLR", test = "bootstrap")
summary (fitsem2, fit.measures=TRUE, standardized=TRUE)
Warning message: In lav_object_post_check(object) : lavaan WARNING: some estimated ov variances are negative > summary (fitsem2, fit.measures=TRUE, standardized=TRUE) lavaan 0.6-2 ended normally after 94 iterations Optimization method NLMINB Number of free parameters 18 Number of observations 3047 Estimator ML Robust Model Fit Test Statistic 815.270 820.407 Degrees of freedom 10 10 P-value (Chi-square) 0.000 0.000 Scaling correction factor 0.994 for the Yuan-Bentler correction (Mplus variant) Model test baseline model: Minimum Function Test Statistic 6493.250 5470.808 Degrees of freedom 21 21 P-value 0.000 0.000 User model versus baseline model: Comparative Fit Index (CFI) 0.876 0.851 Tucker-Lewis Index (TLI) 0.739 0.688 Robust Comparative Fit Index (CFI) 0.875 Robust Tucker-Lewis Index (TLI) 0.739 Loglikelihood and Information Criteria: Loglikelihood user model (H0) -37374.708 -37374.708 Scaling correction factor 1.428 for the MLR correction Loglikelihood unrestricted model (H1) -36967.072 -36967.072 Scaling correction factor 1.273 for the MLR correction Number of free parameters 18 18 Akaike (AIC) 74785.415 74785.415 Bayesian (BIC) 74893.810 74893.810 Sample-size adjusted Bayesian (BIC) 74836.616 74836.616 Root Mean Square Error of Approximation: RMSEA 0.163 0.163 90 Percent Confidence Interval 0.153 0.172 0.154 0.173 P-value RMSEA <= 0.05 0.000 0.000 Robust RMSEA 0.163 90 Percent Confidence Interval 0.153 0.172 Standardized Root Mean Square Residual: SRMR 0.091 0.091 Parameter Estimates: Information Observed Observed information based on Hessian Standard Errors Robust.huber.white Latent Variables: Estimate Std.Err z-value P(>|z|) Std.lv Std.all Underlying_causes =~ bord 1.000 2.833 1.408 m_educa -0.046 0.012 -3.825 0.000 -0.130 -0.211 mo_age 1.338 0.129 10.401 0.000 3.791 0.571 Basic_causes =~ HH_members 1.000 2.349 1.121 N_under5 0.135 0.014 9.461 0.000 0.317 0.436 w_index 0.084 0.008 10.503 0.000 0.197 0.141 Regressions: Estimate Std.Err z-value P(>|z|) Std.lv Std.all Haz ~ Basic_cass (c) 0.017 0.010 1.670 0.095 0.041 0.027 Underlying_causes ~ Basic_cass (b) 0.334 0.045 7.396 0.000 0.277 0.277 Haz ~ Undrlyng_c (a) 0.005 0.005 1.015 0.310 0.015 0.010 Covariances: Estimate Std.Err z-value P(>|z|) Std.lv Std.all .bord ~~ .mo_age 0.366 2.802 0.131 0.896 0.366 0.034 .mo_age ~~ .HH_members 2.573 0.222 11.583 0.000 2.573 0.445 Variances: Estimate Std.Err z-value P(>|z|) Std.lv Std.all .bord -3.978 1.991 -1.998 0.046 -3.978 -0.982 .m_educa 0.360 0.011 32.998 0.000 0.360 0.955 .mo_age 29.635 4.407 6.725 0.000 29.635 0.673 .HH_members -1.127 0.610 -1.847 0.065 -1.127 -0.257 .N_under5 0.428 0.020 21.104 0.000 0.428 0.810 .w_index 1.924 0.031 61.725 0.000 1.924 0.980 .Haz 2.269 0.090 25.268 0.000 2.269 0.999 .Underlying_css 7.412 1.988 3.728 0.000 0.923 0.923 Basic_causes 5.516 0.669 8.243 0.000 1.000 1.000 Defined Parameters: Estimate Std.Err z-value P(>|z|) Std.lv Std.all ab 0.002 0.002 1.034 0.301 0.004 0.003 total 0.019 0.010 1.886 0.059 0.045 0.030
Terrence Jorgensen
Aug 8, 2018, 6:50:18 AM8/8/18
to lavaan
Judging from your factor loadings, the measurement model does not look reasonable. I'm guessing it is this model misspecification that is causing the error variances of your reference indicators to be negative.
I can't tell exactly what all these indicators are, but some of the names seem like demographics (age, education), and your syntax labels them as sets of different types of causes. Are you sure reflective measurement is appropriate? Perhaps a formative "construct" is what you are looking for:
Terrence D. Jorgensen
Postdoctoral Researcher, Methods and Statistics
Research Institute for Child Development and Education, the University of Amsterdam
UvA web page: http://www.uva.nl/profile/t.d.jorgensen
Reply all
Reply to author
Forward
0 new messages