Does Lavaan use standardized scores?
2,309 views
Skip to first unread message
Alexandre F. Souza
Mar 7, 2013, 1:34:25 PM3/7/13
to lav...@googlegroups.com
Hello,
I am new to both SEM and LAVAAN, and work with plant ecology.
I am studying Grace's book Structural Equation Modeling and Natural Systems.
There he states that input data can be either raw (if normally distributed) or standardized through z scores. As far as I got it, the standardized form is preferable, and in lavaan it is carried out setting the option std.ov = TRUE in the sem function.
Is that understanding ok? I am not so sure, because through the lavaan documentation examples keep std.ov as FALSE.
Thanks for any comments,
Alexandre
I am new to both SEM and LAVAAN, and work with plant ecology.
I am studying Grace's book Structural Equation Modeling and Natural Systems.
There he states that input data can be either raw (if normally distributed) or standardized through z scores. As far as I got it, the standardized form is preferable, and in lavaan it is carried out setting the option std.ov = TRUE in the sem function.
Is that understanding ok? I am not so sure, because through the lavaan documentation examples keep std.ov as FALSE.
Thanks for any comments,
Alexandre
Alexander Beaujean
Mar 8, 2013, 12:00:22 AM3/8/13
to lav...@googlegroups.com
It's probably better to use the data in its natural metric for input and then in the summary() or parameterEstimates() function, use the standardized=TRUE argument to get fully standardized (i.e, latent and observed variables) results. They will be in the std.all column.
As far as standardized results being preferable, that is not always the case. As with regression, it depends on your variables and what questions you are wanting to answer. Typically, it's good to examine unstandardized, standardized and (if applicable) structure coefficients.
As far as standardized results being preferable, that is not always the case. As with regression, it depends on your variables and what questions you are wanting to answer. Typically, it's good to examine unstandardized, standardized and (if applicable) structure coefficients.
Best,
Alex
Ruben Arslan
Mar 8, 2013, 10:47:12 AM3/8/13
to lav...@googlegroups.com
Alexandre,
if you don't care about the natural metric, you could standardise your data, e.g.
using
df.s <- plyr::numcolwise(scale)(dataframe)
before using df.s in lavaan.
If natural metrics differ greatly (standard deviation differences in orders of magnitude) , they may cause problems with convergence.
Also, if your data isn't normally distributed, a standardization wouldn't change that, but various transforms would.
In plant ecology you might deal with non-normal count data a lot? Count data can be handled (without transforms) by Mplus, I think lavaan doesn't do that (for now..?)
Best
Ruben
--
You received this message because you are subscribed to the Google Groups "lavaan" group.
To unsubscribe from this group and stop receiving emails from it, send an email to lavaan+un...@googlegroups.com.
To post to this group, send email to lav...@googlegroups.com.
Visit this group at http://groups.google.com/group/lavaan?hl=en.
For more options, visit https://groups.google.com/groups/opt_out.
Alexandre F. Souza
Mar 9, 2013, 8:02:03 PM3/9/13
to lav...@googlegroups.com
Hello,
I would like to thank Alex and Ruben for their answers to my question.
In ecology variables are seldom normally distributed and we usually log- or sqrt-transform them before parametric analyses. In some cases no transformation works.
I changed my mind and understood that it is better to use raw or transformed variables, but not standardized variables as input to sem.
All the best,
Alexandre
I would like to thank Alex and Ruben for their answers to my question.
In ecology variables are seldom normally distributed and we usually log- or sqrt-transform them before parametric analyses. In some cases no transformation works.
I changed my mind and understood that it is better to use raw or transformed variables, but not standardized variables as input to sem.
All the best,
Alexandre
Camille Williams
Apr 11, 2019, 2:51:33 AM4/11/19
to lavaan
Hi,
Questions
I am using log transformed data and have the following two issues. Since things 'look better' when I scale my values before entering them in the model I want to do this, but it apparently gives me inaccurate SE and does not allow me to test metric or scalar invariance properly... What can I do?
1. I get the following warning in my model. From the lavInspect (fit, "optim.gradient), I see that I have most values at 0.000 and some at - 0.004 or -0.005. This wanring is also obtained when I do not include any covariance or set the TBV intercepts to different values.
My model
TBV0.model <- 'TBV =~ NA*CorticalWhiteMatterVol_log + Brain_Stem_log + Left_VentralDC_log + Right_VentralDC_log + lhCortexVol_log + rhCortexVol_log+ Left_Accumbens_area_log + Right_Accumbens_area_log+ Left_Cerebellum_Cortex_log + Right_Cerebellum_Cortex_log + Left_Caudate_log + Right_Caudate_log + Left_Hippocampus_log + Right_Hippocampus_log+ Left_Pallidum_log + Right_Pallidum_log + Left_Putamen_log + Right_Putamen_log + Right_Thalamus_Proper_log + Left_Thalamus_Proper_log+ Left_Amygdala_log + Right_Amygdala_logTBV ~~ 1*TBV TBV ~ c(0, 0.031)* 1lhCortexVol_log ~~ rhCortexVol_logLeft_Caudate_log ~~ Right_Caudate_logLeft_Putamen_log ~~ Right_Putamen_logLeft_Cerebellum_Cortex_log ~~ Right_Cerebellum_Cortex_logRight_Thalamus_Proper_log ~~ Left_Thalamus_Proper_logLeft_Accumbens_area_log ~~ Right_Accumbens_area_logLeft_VentralDC_log ~~ Right_VentralDC_logLeft_Amygdala_log ~~ Right_Amygdala_logLeft_Hippocampus_log ~~ Right_Hippocampus_logLeft_Pallidum_log ~~ Right_Pallidum_log
Right_Putamen_log ~~ Right_Amygdala_logLeft_Putamen_log ~~ Left_Amygdala_logRight_Putamen_log ~~ Left_Amygdala_logLeft_Putamen_log ~~ Right_Amygdala_log'Ex_fit_TBV<- cfa(TBV0.model, data=Abide_MGCFA, group = "DX_GROUP")Warning message:In lavaan::lavaan(model = TBV0.model, data = Abide_MGCFA, group = "DX_GROUP", : lavaan WARNING: not all elements of the gradient are (near) zero; the optimizer may not have found a local solution; use lavInspect(fit, "optim.gradient") to investigate
2. I also get very high std.all = intercepts: CFA with log values WITHOUT scaling
> summary(Ex_fit_TBV, standardize = T )lavaan 0.6-3 ended normally after 978 iterations
Optimization method NLMINB Number of free parameters 160
Number of observations per group Control 352 ASD 302
Estimator ML Model Fit Test Statistic 2400.863 Degrees of freedom 390 P-value (Chi-square) 0.000
Chi-square for each group:
Control 1180.281 ASD 1220.582
Parameter Estimates:
Information Expected Information saturated (h1) model Structured Standard Errors Standard
Group 1 [Control]:
Latent Variables: Estimate Std.Err z-value P(>|z|) Std.lv Std.all TBV =~ CrtclWhtMttrV_ 0.045 0.002 18.113 0.000 0.045 0.812 Brain_Stem_log 0.043 0.002 17.286 0.000 0.043 0.787 Lft_VntrlDC_lg 0.038 0.002 15.823 0.000 0.038 0.741 Rght_VntrlDC_l 0.036 0.002 15.565 0.000 0.036 0.732 lhCortexVol_lg 0.033 0.003 12.868 0.000 0.033 0.634 rhCortexVol_lg 0.034 0.003 12.998 0.000 0.034 0.639 Lft_Accmbns_r_ 0.043 0.005 7.897 0.000 0.043 0.417 Rght_Accmbns__ 0.041 0.005 7.977 0.000 0.041 0.421 Lft_Crbllm_Cr_ 0.027 0.003 9.424 0.000 0.027 0.489 Rght_Crbllm_C_ 0.028 0.003 8.855 0.000 0.028 0.463 Left_Caudat_lg 0.044 0.003 12.734 0.000 0.044 0.629 Right_Caudt_lg 0.048 0.004 13.046 0.000 0.048 0.641 Lft_Hppcmps_lg 0.037 0.003 13.934 0.000 0.037 0.674 Rght_Hppcmps_l 0.038 0.003 13.379 0.000 0.038 0.654 Left_Palldm_lg 0.040 0.003 13.678 0.000 0.040 0.665 Right_Plldm_lg 0.039 0.003 13.429 0.000 0.039 0.656 Left_Putamn_lg 0.035 0.003 11.913 0.000 0.035 0.596 Right_Putmn_lg 0.035 0.003 12.266 0.000 0.035 0.611 Rght_Thlms_Pr_ 0.036 0.002 17.732 0.000 0.036 0.801 Lft_Thlms_Prp_ 0.034 0.002 16.772 0.000 0.034 0.772 Left_Amygdl_lg 0.041 0.003 11.877 0.000 0.041 0.595 Rght_Amygdl_lg 0.042 0.004 11.511 0.000 0.042 0.580
Covariances: Estimate Std.Err z-value P(>|z|) Std.lv Std.all .lhCortexVol_log ~~ .rhCortexVol_lg 0.002 0.000 12.285 0.000 0.002 0.966 .Left_Caudate_log ~~ .Right_Caudt_lg 0.003 0.000 11.876 0.000 0.003 0.909 .Left_Putamen_log ~~ .Right_Putmn_lg 0.002 0.000 11.587 0.000 0.002 0.860 .Left_Cerebellum_Cortex_log ~~ .Rght_Crbllm_C_ 0.002 0.000 9.772 0.000 0.002 0.635 .Right_Thalamus_Proper_log ~~ .Lft_Thlms_Prp_ 0.001 0.000 8.897 0.000 0.001 0.682 .Left_Accumbens_area_log ~~ .Rght_Accmbns__ 0.006 0.001 11.009 0.000 0.006 0.747 .Left_VentralDC_log ~~ .Rght_VntrlDC_l 0.001 0.000 8.513 0.000 0.001 0.600 .Left_Amygdala_log ~~ .Rght_Amygdl_lg 0.002 0.000 9.043 0.000 0.002 0.592 .Left_Hippocampus_log ~~ .Rght_Hppcmps_l 0.001 0.000 8.121 0.000 0.001 0.533 .Left_Pallidum_log ~~ .Right_Plldm_lg 0.001 0.000 8.721 0.000 0.001 0.583 .Right_Putamen_log ~~ .Rght_Amygdl_lg 0.001 0.000 4.924 0.000 0.001 0.290 .Left_Putamen_log ~~ .Left_Amygdl_lg 0.001 0.000 3.738 0.000 0.001 0.216 .Right_Putamen_log ~~ .Left_Amygdl_lg 0.000 0.000 3.436 0.001 0.000 0.198 .Left_Putamen_log ~~ .Rght_Amygdl_lg 0.000 0.000 2.809 0.005 0.000 0.160
Intercepts: Estimate Std.Err z-value P(>|z|) Std.lv Std.all TBV 0.000 0.000 0.000 .CrtclWhtMttrV_ 5.670 0.003 1928.156 0.000 5.670 102.771 .Brain_Stem_log 4.305 0.003 1495.363 0.000 4.305 79.703 .Lft_VntrlDC_lg 3.613 0.003 1308.749 0.000 3.613 69.757 .Rght_VntrlDC_l 3.617 0.003 1370.427 0.000 3.617 73.044 .lhCortexVol_lg 5.429 0.003 1933.187 0.000 5.429 103.039 .rhCortexVol_lg 5.430 0.003 1934.671 0.000 5.430 103.118 .Lft_Accmbns_r_ 2.883 0.005 527.384 0.000 2.883 28.110 .Rght_Accmbns__ 2.898 0.005 554.175 0.000 2.898 29.538 .Lft_Crbllm_Cr_ 4.733 0.003 1628.268 0.000 4.733 86.787 .Rght_Crbllm_C_ 4.737 0.003 1474.385 0.000 4.737 78.585 .Left_Caudat_lg 3.607 0.004 962.971 0.000 3.607 51.327 .Right_Caudt_lg 3.619 0.004 908.733 0.000 3.619 48.436 .Lft_Hppcmps_lg 3.610 0.003 1237.236 0.000 3.610 65.945 .Rght_Hppcmps_l 3.614 0.003 1160.374 0.000 3.614 61.848 .Left_Palldm_lg 3.294 0.003 1038.040 0.000 3.294 55.328 .Right_Plldm_lg 3.249 0.003 1017.033 0.000 3.249 54.208 .Left_Putamn_lg 3.828 0.003 1236.407 0.000 3.828 65.901 .Right_Putmn_lg 3.809 0.003 1256.364 0.000 3.809 66.964 .Rght_Thlms_Pr_ 3.874 0.002 1602.477 0.000 3.874 85.412 .Lft_Thlms_Prp_ 3.866 0.002 1629.038 0.000 3.866 86.828 .Left_Amygdl_lg 3.221 0.004 883.101 0.000 3.221 47.069 .Rght_Amygdl_lg 3.239 0.004 841.303 0.000 3.239 44.842
Variances: Estimate Std.Err z-value P(>|z|) Std.lv Std.all TBV 1.000 1.000 1.000 .CrtclWhtMttrV_ 0.001 0.000 11.112 0.000 0.001 0.340 .Brain_Stem_log 0.001 0.000 11.457 0.000 0.001 0.380 .Lft_VntrlDC_lg 0.001 0.000 11.888 0.000 0.001 0.451 .Rght_VntrlDC_l 0.001 0.000 11.951 0.000 0.001 0.464 .lhCortexVol_lg 0.002 0.000 12.523 0.000 0.002 0.598 .rhCortexVol_lg 0.002 0.000 12.505 0.000 0.002 0.592 .Lft_Accmbns_r_ 0.009 0.001 13.033 0.000 0.009 0.826 .Rght_Accmbns__ 0.008 0.001 13.028 0.000 0.008 0.823 .Lft_Crbllm_Cr_ 0.002 0.000 12.919 0.000 0.002 0.761 .Rght_Crbllm_C_ 0.003 0.000 12.964 0.000 0.003 0.786 .Left_Caudat_lg 0.003 0.000 12.541 0.000 0.003 0.605 .Right_Caudt_lg 0.003 0.000 12.496 0.000 0.003 0.590 .Lft_Hppcmps_lg 0.002 0.000 12.328 0.000 0.002 0.545 .Rght_Hppcmps_l 0.002 0.000 12.421 0.000 0.002 0.573 .Left_Palldm_lg 0.002 0.000 12.377 0.000 0.002 0.558 .Right_Plldm_lg 0.002 0.000 12.418 0.000 0.002 0.570 .Left_Putamn_lg 0.002 0.000 12.632 0.000 0.002 0.644 .Right_Putmn_lg 0.002 0.000 12.585 0.000 0.002 0.627 .Rght_Thlms_Pr_ 0.001 0.000 11.248 0.000 0.001 0.358 .Lft_Thlms_Prp_ 0.001 0.000 11.562 0.000 0.001 0.404 .Left_Amygdl_lg 0.003 0.000 12.636 0.000 0.003 0.646 .Rght_Amygdl_lg 0.003 0.000 12.682 0.000 0.003 0.664
Group 2 [ASD]:
Latent Variables: Estimate Std.Err z-value P(>|z|) Std.lv Std.all TBV =~ CrtclWhtMttrV_ 0.053 0.003 18.279 0.000 0.053 0.855 Brain_Stem_log 0.050 0.003 18.161 0.000 0.050 0.851 Lft_VntrlDC_lg 0.038 0.003 15.004 0.000 0.038 0.750 Rght_VntrlDC_l 0.036 0.003 14.030 0.000 0.036 0.715 lhCortexVol_lg 0.042 0.003 13.999 0.000 0.042 0.713 rhCortexVol_lg 0.043 0.003 14.044 0.000 0.043 0.715 Lft_Accmbns_r_ 0.040 0.007 6.045 0.000 0.040 0.348 Rght_Accmbns__ 0.039 0.006 6.990 0.000 0.039 0.398 Lft_Crbllm_Cr_ 0.033 0.003 10.792 0.000 0.033 0.582 Rght_Crbllm_C_ 0.037 0.003 11.864 0.000 0.037 0.628 Left_Caudat_lg 0.043 0.003 12.689 0.000 0.043 0.663 Right_Caudt_lg 0.047 0.004 13.321 0.000 0.047 0.687 Lft_Hppcmps_lg 0.044 0.004 11.130 0.000 0.044 0.598 Rght_Hppcmps_l 0.047 0.003 14.047 0.000 0.047 0.715 Left_Palldm_lg 0.042 0.003 12.554 0.000 0.042 0.657 Right_Plldm_lg 0.040 0.003 12.325 0.000 0.040 0.648 Left_Putamn_lg 0.030 0.003 10.100 0.000 0.030 0.552 Right_Putmn_lg 0.036 0.003 10.711 0.000 0.036 0.579 Rght_Thlms_Pr_ 0.040 0.002 16.329 0.000 0.040 0.795 Lft_Thlms_Prp_ 0.038 0.003 14.691 0.000 0.038 0.739 Left_Amygdl_lg 0.058 0.005 11.602 0.000 0.058 0.618 Rght_Amygdl_lg 0.055 0.005 10.317 0.000 0.055 0.561
Covariances: Estimate Std.Err z-value P(>|z|) Std.lv Std.all .lhCortexVol_log ~~ .rhCortexVol_lg 0.002 0.000 11.130 0.000 0.002 0.957 .Left_Caudate_log ~~ .Right_Caudt_lg 0.002 0.000 10.518 0.000 0.002 0.846 .Left_Putamen_log ~~ .Right_Putmn_lg 0.002 0.000 9.942 0.000 0.002 0.740 .Left_Cerebellum_Cortex_log ~~ .Rght_Crbllm_C_ 0.002 0.000 11.118 0.000 0.002 0.898 .Right_Thalamus_Proper_log ~~ .Lft_Thlms_Prp_ 0.001 0.000 9.338 0.000 0.001 0.762 .Left_Accumbens_area_log ~~ .Rght_Accmbns__ 0.007 0.001 9.681 0.000 0.007 0.684 .Left_VentralDC_log ~~ .Rght_VntrlDC_l 0.001 0.000 7.873 0.000 0.001 0.580 .Left_Amygdala_log ~~ .Rght_Amygdl_lg 0.004 0.000 8.954 0.000 0.004 0.640 .Left_Hippocampus_log ~~ .Rght_Hppcmps_l 0.001 0.000 7.282 0.000 0.001 0.502 .Left_Pallidum_log ~~ .Right_Plldm_lg 0.001 0.000 8.256 0.000 0.001 0.587 .Right_Putamen_log ~~ .Rght_Amygdl_lg 0.001 0.000 5.678 0.000 0.001 0.363 .Left_Putamen_log ~~ .Left_Amygdl_lg 0.001 0.000 2.830 0.005 0.001 0.173 .Right_Putamen_log ~~ .Left_Amygdl_lg 0.001 0.000 2.420 0.016 0.001 0.147 .Left_Putamen_log ~~ .Rght_Amygdl_lg 0.001 0.000 4.016 0.000 0.001 0.247
Intercepts: Estimate Std.Err z-value P(>|z|) Std.lv Std.all TBV 0.031 0.031 0.031 .CrtclWhtMttrV_ 5.666 0.004 1591.272 0.000 5.666 91.597 .Brain_Stem_log 4.302 0.003 1274.897 0.000 4.302 73.385 .Lft_VntrlDC_lg 3.614 0.003 1254.754 0.000 3.614 72.229 .Rght_VntrlDC_l 3.619 0.003 1242.944 0.000 3.619 71.550 .lhCortexVol_lg 5.429 0.003 1594.381 0.000 5.429 91.781 .rhCortexVol_lg 5.428 0.003 1562.458 0.000 5.428 89.943 .Lft_Accmbns_r_ 2.891 0.007 434.827 0.000 2.891 25.033 .Rght_Accmbns__ 2.900 0.006 520.078 0.000 2.900 29.941 .Lft_Crbllm_Cr_ 4.729 0.003 1457.027 0.000 4.729 83.878 .Rght_Crbllm_C_ 4.740 0.003 1409.608 0.000 4.740 81.147 .Left_Caudat_lg 3.612 0.004 962.462 0.000 3.612 55.405 .Right_Caudt_lg 3.621 0.004 923.668 0.000 3.621 53.172 .Lft_Hppcmps_lg 3.611 0.004 855.776 0.000 3.611 49.265 .Rght_Hppcmps_l 3.618 0.004 957.005 0.000 3.618 55.090 .Left_Palldm_lg 3.300 0.004 905.177 0.000 3.300 52.108 .Right_Plldm_lg 3.256 0.004 918.531 0.000 3.256 52.877 .Left_Putamn_lg 3.833 0.003 1228.972 0.000 3.833 70.750 .Right_Putmn_lg 3.813 0.004 1067.099 0.000 3.813 61.431 .Rght_Thlms_Pr_ 3.875 0.003 1325.588 0.000 3.875 76.305 .Lft_Thlms_Prp_ 3.866 0.003 1289.980 0.000 3.866 74.257 .Left_Amygdl_lg 3.222 0.005 599.307 0.000 3.222 34.500 .Rght_Amygdl_lg 3.238 0.006 574.278 0.000 3.238 33.060
Variances: Estimate Std.Err z-value P(>|z|) Std.lv Std.all TBV 1.000 1.000 1.000 .CrtclWhtMttrV_ 0.001 0.000 9.954 0.000 0.001 0.269 .Brain_Stem_log 0.001 0.000 10.022 0.000 0.001 0.275 .Lft_VntrlDC_lg 0.001 0.000 11.173 0.000 0.001 0.438 .Rght_VntrlDC_l 0.001 0.000 11.367 0.000 0.001 0.489 .lhCortexVol_lg 0.002 0.000 11.402 0.000 0.002 0.491 .rhCortexVol_lg 0.002 0.000 11.394 0.000 0.002 0.489 .Lft_Accmbns_r_ 0.012 0.001 12.170 0.000 0.012 0.879 .Rght_Accmbns__ 0.008 0.001 12.128 0.000 0.008 0.841 .Lft_Crbllm_Cr_ 0.002 0.000 11.847 0.000 0.002 0.661 .Rght_Crbllm_C_ 0.002 0.000 11.732 0.000 0.002 0.605 .Left_Caudat_lg 0.002 0.000 11.614 0.000 0.002 0.561 .Right_Caudt_lg 0.002 0.000 11.522 0.000 0.002 0.527 .Lft_Hppcmps_lg 0.003 0.000 11.795 0.000 0.003 0.643 .Rght_Hppcmps_l 0.002 0.000 11.390 0.000 0.002 0.489 .Left_Palldm_lg 0.002 0.000 11.632 0.000 0.002 0.568 .Right_Plldm_lg 0.002 0.000 11.662 0.000 0.002 0.580 .Left_Putamn_lg 0.002 0.000 11.901 0.000 0.002 0.696 .Right_Putmn_lg 0.003 0.000 11.843 0.000 0.003 0.665 .Rght_Thlms_Pr_ 0.001 0.000 10.816 0.000 0.001 0.368 .Lft_Thlms_Prp_ 0.001 0.000 11.219 0.000 0.001 0.454 .Left_Amygdl_lg 0.005 0.000 11.748 0.000 0.005 0.618 .Rght_Amygdl_lg 0.007 0.001 11.881 0.000 0.007 0.685If I scale my log values before inputing them in the model, I
1. No longer have this warning
2. I have the same std.all values for my loadings and covariances but not for the intercept. Why is this?
CFA with log values WITH scaling
> summary(Ex_fit_TBV, standardize = T )lavaan 0.6-3 ended normally after 99 iterations
Optimization method NLMINB Number of free parameters 160
Number of observations per group Control 352 ASD 302
Estimator ML Model Fit Test Statistic 2400.863 Degrees of freedom 390 P-value (Chi-square) 0.000
Chi-square for each group:
Control 1180.281 ASD 1220.582
Parameter Estimates:
Information Expected Information saturated (h1) model Structured Standard Errors Standard
Group 1 [Control]:
Latent Variables: Estimate Std.Err z-value P(>|z|) Std.lv Std.all TBV =~ CrtclWhtMttrV_ 0.767 0.042 18.113 0.000 0.767 0.812 Brain_Stem_log 0.756 0.044 17.286 0.000 0.756 0.787 Lft_VntrlDC_lg 0.752 0.048 15.823 0.000 0.752 0.741 Rght_VntrlDC_l 0.724 0.047 15.565 0.000 0.724 0.732 lhCortexVol_lg 0.598 0.046 12.868 0.000 0.598 0.634 rhCortexVol_lg 0.596 0.046 12.998 0.000 0.596 0.639 Lft_Accmbns_r_ 0.393 0.050 7.897 0.000 0.393 0.417 Rght_Accmbns__ 0.423 0.053 7.977 0.000 0.423 0.421 Lft_Crbllm_Cr_ 0.481 0.051 9.424 0.000 0.481 0.489 Rght_Crbllm_C_ 0.469 0.053 8.855 0.000 0.469 0.463 Left_Caudat_lg 0.649 0.051 12.734 0.000 0.649 0.629 Right_Caudt_lg 0.666 0.051 13.046 0.000 0.666 0.641 Lft_Hppcmps_lg 0.576 0.041 13.934 0.000 0.576 0.674 Rght_Hppcmps_l 0.616 0.046 13.379 0.000 0.616 0.654 Left_Palldm_lg 0.644 0.047 13.678 0.000 0.644 0.665 Right_Plldm_lg 0.645 0.048 13.429 0.000 0.645 0.656 Left_Putamn_lg 0.614 0.052 11.913 0.000 0.614 0.596 Right_Putmn_lg 0.584 0.048 12.266 0.000 0.584 0.611 Rght_Thlms_Pr_ 0.757 0.043 17.732 0.000 0.757 0.801 Lft_Thlms_Prp_ 0.713 0.043 16.772 0.000 0.713 0.772 Left_Amygdl_lg 0.503 0.042 11.877 0.000 0.503 0.595 Rght_Amygdl_lg 0.492 0.043 11.511 0.000 0.492 0.580
Covariances: Estimate Std.Err z-value P(>|z|) Std.lv Std.all .lhCortexVol_log ~~ .rhCortexVol_lg 0.507 0.041 12.285 0.000 0.507 0.966 .Left_Caudate_log ~~ .Right_Caudt_lg 0.583 0.049 11.876 0.000 0.583 0.909 .Left_Putamen_log ~~ .Right_Putmn_lg 0.539 0.046 11.587 0.000 0.539 0.860 .Left_Cerebellum_Cortex_log ~~ .Rght_Crbllm_C_ 0.489 0.050 9.772 0.000 0.489 0.635 .Right_Thalamus_Proper_log ~~ .Lft_Thlms_Prp_ 0.227 0.025 8.897 0.000 0.227 0.682 .Left_Accumbens_area_log ~~ .Rght_Accmbns__ 0.583 0.053 11.009 0.000 0.583 0.747 .Left_VentralDC_log ~~ .Rght_VntrlDC_l 0.275 0.032 8.513 0.000 0.275 0.600 .Left_Amygdala_log ~~ .Rght_Amygdl_lg 0.278 0.031 9.043 0.000 0.278 0.592 .Left_Hippocampus_log ~~ .Rght_Hppcmps_l 0.240 0.030 8.121 0.000 0.240 0.533 .Left_Pallidum_log ~~ .Right_Plldm_lg 0.313 0.036 8.721 0.000 0.313 0.583 .Right_Putamen_log ~~ .Rght_Amygdl_lg 0.152 0.031 4.924 0.000 0.152 0.290 .Left_Putamen_log ~~ .Left_Amygdl_lg 0.121 0.032 3.738 0.000 0.121 0.216 .Right_Putamen_log ~~ .Left_Amygdl_lg 0.102 0.030 3.436 0.001 0.102 0.198 .Left_Putamen_log ~~ .Rght_Amygdl_lg 0.091 0.033 2.809 0.005 0.091 0.160
Intercepts: Estimate Std.Err z-value P(>|z|) Std.lv Std.all TBV 0.000 0.000 0.000 .CrtclWhtMttrV_ 0.021 0.050 0.423 0.672 0.021 0.023 .Brain_Stem_log 0.008 0.051 0.160 0.873 0.008 0.009 .Lft_VntrlDC_lg -0.017 0.054 -0.320 0.749 -0.017 -0.017 .Rght_VntrlDC_l -0.034 0.053 -0.641 0.521 -0.034 -0.034 .lhCortexVol_lg -0.007 0.050 -0.132 0.895 -0.007 -0.007 .rhCortexVol_lg 0.000 0.050 0.009 0.992 0.000 0.001 .Lft_Accmbns_r_ -0.039 0.050 -0.773 0.439 -0.039 -0.041 .Rght_Accmbns__ -0.013 0.054 -0.239 0.811 -0.013 -0.013 .Lft_Crbllm_Cr_ 0.026 0.052 0.505 0.614 0.026 0.027 .Rght_Crbllm_C_ -0.031 0.054 -0.571 0.568 -0.031 -0.030 .Left_Caudat_lg -0.043 0.055 -0.785 0.433 -0.043 -0.042 .Right_Caudt_lg -0.026 0.055 -0.463 0.643 -0.026 -0.025 .Lft_Hppcmps_lg -0.023 0.046 -0.495 0.621 -0.023 -0.026 .Rght_Hppcmps_l -0.047 0.050 -0.939 0.348 -0.047 -0.050 .Left_Palldm_lg -0.055 0.052 -1.064 0.287 -0.055 -0.057 .Right_Plldm_lg -0.064 0.052 -1.228 0.219 -0.064 -0.065 .Left_Putamn_lg -0.051 0.055 -0.933 0.351 -0.051 -0.050 .Right_Putmn_lg -0.037 0.051 -0.725 0.468 -0.037 -0.039 .Rght_Thlms_Pr_ -0.019 0.050 -0.385 0.700 -0.019 -0.021 .Lft_Thlms_Prp_ -0.003 0.049 -0.068 0.946 -0.003 -0.004 .Left_Amygdl_lg -0.019 0.045 -0.421 0.673 -0.019 -0.022 .Rght_Amygdl_lg -0.001 0.045 -0.020 0.984 -0.001 -0.001
Variances: Estimate Std.Err z-value P(>|z|) Std.lv Std.all TBV 1.000 1.000 1.000 .CrtclWhtMttrV_ 0.304 0.027 11.112 0.000 0.304 0.340 .Brain_Stem_log 0.351 0.031 11.457 0.000 0.351 0.380 .Lft_VntrlDC_lg 0.465 0.039 11.888 0.000 0.465 0.451 .Rght_VntrlDC_l 0.454 0.038 11.951 0.000 0.454 0.464 .lhCortexVol_lg 0.533 0.043 12.523 0.000 0.533 0.598 .rhCortexVol_lg 0.516 0.041 12.505 0.000 0.516 0.592 .Lft_Accmbns_r_ 0.732 0.056 13.033 0.000 0.732 0.826 .Rght_Accmbns__ 0.831 0.064 13.028 0.000 0.831 0.823 .Lft_Crbllm_Cr_ 0.736 0.057 12.919 0.000 0.736 0.761 .Rght_Crbllm_C_ 0.807 0.062 12.964 0.000 0.807 0.786 .Left_Caudat_lg 0.644 0.051 12.541 0.000 0.644 0.605 .Right_Caudt_lg 0.638 0.051 12.496 0.000 0.638 0.590 .Lft_Hppcmps_lg 0.398 0.032 12.329 0.000 0.398 0.545 .Rght_Hppcmps_l 0.508 0.041 12.421 0.000 0.508 0.573 .Left_Palldm_lg 0.523 0.042 12.377 0.000 0.523 0.558 .Right_Plldm_lg 0.552 0.044 12.418 0.000 0.552 0.570 .Left_Putamn_lg 0.682 0.054 12.632 0.000 0.682 0.644 .Right_Putmn_lg 0.575 0.046 12.585 0.000 0.575 0.627 .Rght_Thlms_Pr_ 0.320 0.028 11.248 0.000 0.320 0.358 .Lft_Thlms_Prp_ 0.345 0.030 11.562 0.000 0.345 0.404 .Left_Amygdl_lg 0.461 0.036 12.636 0.000 0.461 0.646 .Rght_Amygdl_lg 0.478 0.038 12.682 0.000 0.478 0.664
Group 2 [ASD]:
Latent Variables: Estimate Std.Err z-value P(>|z|) Std.lv Std.all TBV =~ CrtclWhtMttrV_ 0.905 0.050 18.279 0.000 0.905 0.855 Brain_Stem_log 0.888 0.049 18.161 0.000 0.888 0.851 Lft_VntrlDC_lg 0.735 0.049 15.004 0.000 0.735 0.750 Rght_VntrlDC_l 0.722 0.051 14.030 0.000 0.722 0.715 lhCortexVol_lg 0.756 0.054 13.999 0.000 0.756 0.713 rhCortexVol_lg 0.765 0.054 14.044 0.000 0.765 0.715 Lft_Accmbns_r_ 0.369 0.061 6.045 0.000 0.369 0.348 Rght_Accmbns__ 0.395 0.057 6.991 0.000 0.395 0.398 Lft_Crbllm_Cr_ 0.592 0.055 10.792 0.000 0.592 0.582 Rght_Crbllm_C_ 0.617 0.052 11.864 0.000 0.617 0.628 Left_Caudat_lg 0.634 0.050 12.689 0.000 0.634 0.663 Right_Caudt_lg 0.652 0.049 13.321 0.000 0.652 0.687 Lft_Hppcmps_lg 0.684 0.061 11.130 0.000 0.684 0.598 Rght_Hppcmps_l 0.757 0.054 14.047 0.000 0.757 0.715 Left_Palldm_lg 0.677 0.054 12.554 0.000 0.677 0.657 Right_Plldm_lg 0.655 0.053 12.325 0.000 0.655 0.648 Left_Putamn_lg 0.529 0.052 10.100 0.000 0.529 0.552 Right_Putmn_lg 0.605 0.056 10.711 0.000 0.605 0.579 Rght_Thlms_Pr_ 0.841 0.052 16.329 0.000 0.841 0.795 Lft_Thlms_Prp_ 0.798 0.054 14.691 0.000 0.798 0.739 Left_Amygdl_lg 0.712 0.061 11.602 0.000 0.712 0.618 Rght_Amygdl_lg 0.646 0.063 10.317 0.000 0.646 0.561
Covariances: Estimate Std.Err z-value P(>|z|) Std.lv Std.all .lhCortexVol_log ~~ .rhCortexVol_lg 0.532 0.048 11.130 0.000 0.532 0.957 .Left_Caudate_log ~~ .Right_Caudt_lg 0.418 0.040 10.518 0.000 0.418 0.846 .Left_Putamen_log ~~ .Right_Putmn_lg 0.504 0.051 9.942 0.000 0.504 0.740 .Left_Cerebellum_Cortex_log ~~ .Rght_Crbllm_C_ 0.567 0.051 11.118 0.000 0.567 0.898 .Right_Thalamus_Proper_log ~~ .Lft_Thlms_Prp_ 0.356 0.038 9.338 0.000 0.356 0.762 .Left_Accumbens_area_log ~~ .Rght_Accmbns__ 0.618 0.064 9.681 0.000 0.618 0.684 .Left_VentralDC_log ~~ .Rght_VntrlDC_l 0.266 0.034 7.873 0.000 0.266 0.580 .Left_Amygdala_log ~~ .Rght_Amygdl_lg 0.553 0.062 8.954 0.000 0.553 0.640 .Left_Hippocampus_log ~~ .Rght_Hppcmps_l 0.341 0.047 7.282 0.000 0.341 0.502 .Left_Pallidum_log ~~ .Right_Plldm_lg 0.351 0.043 8.256 0.000 0.351 0.587 .Right_Putamen_log ~~ .Rght_Amygdl_lg 0.294 0.052 5.678 0.000 0.294 0.363 .Left_Putamen_log ~~ .Left_Amygdl_lg 0.125 0.044 2.830 0.005 0.125 0.173 .Right_Putamen_log ~~ .Left_Amygdl_lg 0.114 0.047 2.420 0.016 0.114 0.147 .Left_Putamen_log ~~ .Rght_Amygdl_lg 0.189 0.047 4.016 0.000 0.189 0.247
Intercepts: Estimate Std.Err z-value P(>|z|) Std.lv Std.all TBV 0.031 0.031 0.031 .CrtclWhtMttrV_ -0.053 0.061 -0.867 0.386 -0.053 -0.050 .Brain_Stem_log -0.037 0.060 -0.618 0.537 -0.037 -0.036 .Lft_VntrlDC_lg -0.003 0.056 -0.046 0.963 -0.003 -0.003 .Rght_VntrlDC_l 0.017 0.058 0.293 0.770 0.017 0.017 .lhCortexVol_lg -0.016 0.061 -0.257 0.797 -0.016 -0.015 .rhCortexVol_lg -0.024 0.062 -0.394 0.694 -0.024 -0.023 .Lft_Accmbns_r_ 0.034 0.061 0.553 0.580 0.034 0.032 .Rght_Accmbns__ 0.003 0.057 0.047 0.962 0.003 0.003 .Lft_Crbllm_Cr_ -0.049 0.059 -0.841 0.400 -0.049 -0.048 .Rght_Crbllm_C_ 0.017 0.057 0.298 0.766 0.017 0.017 .Left_Caudat_lg 0.031 0.055 0.556 0.578 0.031 0.032 .Right_Caudt_lg 0.010 0.055 0.178 0.859 0.010 0.010 .Lft_Hppcmps_lg 0.005 0.066 0.077 0.938 0.005 0.004 .Rght_Hppcmps_l 0.031 0.061 0.516 0.606 0.031 0.030 .Left_Palldm_lg 0.043 0.059 0.725 0.468 0.043 0.042 .Right_Plldm_lg 0.055 0.058 0.941 0.347 0.055 0.054 .Left_Putamn_lg 0.043 0.055 0.782 0.434 0.043 0.045 .Right_Putmn_lg 0.024 0.060 0.405 0.685 0.024 0.023 .Rght_Thlms_Pr_ -0.003 0.061 -0.057 0.955 -0.003 -0.003 .Lft_Thlms_Prp_ -0.021 0.062 -0.335 0.737 -0.021 -0.019 .Left_Amygdl_lg 0.000 0.066 0.001 1.000 0.000 0.000 .Rght_Amygdl_lg -0.019 0.066 -0.287 0.774 -0.019 -0.017
Variances: Estimate Std.Err z-value P(>|z|) Std.lv Std.all TBV 1.000 1.000 1.000 .CrtclWhtMttrV_ 0.302 0.030 9.954 0.000 0.302 0.269 .Brain_Stem_log 0.299 0.030 10.022 0.000 0.299 0.275 .Lft_VntrlDC_lg 0.421 0.038 11.173 0.000 0.421 0.438 .Rght_VntrlDC_l 0.499 0.044 11.367 0.000 0.499 0.489 .lhCortexVol_lg 0.552 0.048 11.402 0.000 0.552 0.491 .rhCortexVol_lg 0.560 0.049 11.394 0.000 0.560 0.489 .Lft_Accmbns_r_ 0.988 0.081 12.170 0.000 0.988 0.879 .Rght_Accmbns__ 0.828 0.068 12.128 0.000 0.828 0.841 .Lft_Crbllm_Cr_ 0.683 0.058 11.847 0.000 0.683 0.661 .Rght_Crbllm_C_ 0.583 0.050 11.732 0.000 0.583 0.605 .Left_Caudat_lg 0.514 0.044 11.614 0.000 0.514 0.561 .Right_Caudt_lg 0.474 0.041 11.522 0.000 0.474 0.527 .Lft_Hppcmps_lg 0.842 0.071 11.795 0.000 0.842 0.643 .Rght_Hppcmps_l 0.548 0.048 11.390 0.000 0.548 0.489 .Left_Palldm_lg 0.603 0.052 11.632 0.000 0.603 0.568 .Right_Plldm_lg 0.593 0.051 11.662 0.000 0.593 0.580 .Left_Putamn_lg 0.641 0.054 11.901 0.000 0.641 0.696 .Right_Putmn_lg 0.725 0.061 11.843 0.000 0.725 0.665 .Rght_Thlms_Pr_ 0.413 0.038 10.816 0.000 0.413 0.368 .Lft_Thlms_Prp_ 0.529 0.047 11.219 0.000 0.529 0.454 .Left_Amygdl_lg 0.822 0.070 11.748 0.000 0.822 0.618 .Rght_Amygdl_lg 0.906 0.076 11.881 0.000 0.906 0.685Questions
(1) why my std.all values are so high when using log transformed data without scaling prior to its input in the model ?
(2) why std.all loadings and variances are the same when scaling and not scaling my data before inputing the values in my model?
Since I am only interested in seeing if there are group differences in loadings and intercepts, can I use the scaled log values as input? If so, do I report the std.lv, since my observed variables have already been scaled?
Thank you very much for your time.
Terrence Jorgensen
Apr 12, 2019, 4:37:04 AM4/12/19
to lavaan
I am using log transformed data
Why? If your data are not normally distributed, you can just use a robust estimator, so that your parameters are still interpretable.
cfa(..., estimator = "MLR")Since things 'look better'
What does that mean, by what criteria? You can obtain standardized estimates without transforming your data, as you were already told in another thread.
What can I do?
Analyze the original data, using a robust estimator, and compare models that test your null hypotheses of invariance.
Your model is having trouble converging. Perhaps analyzing log-transformed data is making estimation difficult.
(1) why my std.all values are so high when using log transformed data without scaling prior to its input in the model ?
Because the means of z scores are 0 by definition, whereas your log-transformed variables clearly have higher means.
(2) why std.all loadings and variances are the same when scaling and not scaling my data before inputing the values in my model?
That is why it is pointless to standardize your data before analysis. The standardized estimates already provide that information. (The fact that the raw estimates do not match the std.all estimates means that you weren't fitting your model to z scores, so I'm not sure what you mean by "scaling".)
Since I am only interested in seeing if there are group differences in loadings and intercepts, can I use the scaled log values as input?
Standardized estimates are not the data-generating parameters, so they are not the values relevant to testing measurement invariance. You need to analyze your raw data, using a robust estimator if variables are non-normal (but continuous), and compare models with and without equality constraints on certain parameters (e.g., test equality of loadings, then equality of intercepts, etc.).
http://lavaan.ugent.be/resources/books.html (numerous books describe the process with lavaan)
Terrence D. Jorgensen
Assistant Professor, Methods and Statistics
Research Institute for Child Development and Education, the University of Amsterdam
Camille Williams
Apr 12, 2019, 8:18:22 AM4/12/19
to lavaan
Hi,
Thank you very much for time and patience with me. I am discovering this on my own and appreciate your help.
I want to be analyzing log transformed data, because I am interested in the group differences in slope and intercept of these volumes once they have been log transformed. . This is why I insist on using log transformations on my volumes. The slope obtained following log-log transformation differs from that of the linear model.
Can I use an estimator on my log transformed data in the cfa ?
What I did to "scale" my variables is first apply a log transformation and then the scale function. for example:
Right_Cerebellum_Cortex_log = scale(log10(`Right-Cerebellum-Cortex`))
thank you again,
Camille
maryam nasseri
Apr 12, 2019, 10:47:59 AM4/12/19
to lav...@googlegroups.com
Hi Terrence and everyone else,
I'm having a similar issue. You and others recommended me to use the raw original data and a robust estimator (and not scaling/z-score my data as input); when I do this I get multiple warning messages and the summary() function returns an Error of subscript out of bound! I am printing the exact code I used below; any thoughts? Many thanks for any help/info:
mydata<- read.csv("data.csv") #this is my raw original datalex.modelB<- 'group1=~ mtld + hdd + ndwesz + ndwerz + vocd + mattr + msttr
+ group2=~ logttr + rttr + uber + maas
+ group3=~ lv + nv + vv1 + vv2 +cvv1 + adjv + ld
+ group4=~ ls1 + ls2 + vs2 + cvs1'
fitB<- sem(lex.modelB, mydata, sample.nobs=210, estimator = "MLM") # using a robust estimator
Warning messages:
1: In lav_data_full(data = data, group = group, cluster = cluster, :
lavaan WARNING: some observed variances are (at least) a factor 1000 times larger than others; use varTable(fit) to investigate
2: In lav_model_vcov(lavmodel = lavmodel, lavsamplestats = lavsamplestats, :
lavaan WARNING:
Could not compute standard errors! The information matrix could
not be inverted. This may be a symptom that the model is not
identified.
3: In lav_test_satorra_bentler(lavobject = NULL, lavsamplestats = lavsamplestats, :
lavaan WARNING: could not invert information matrix
4: In lav_object_post_check(object) :
lavaan WARNING: some estimated ov variances are negative
5: In lavaan::lavaan(model = lex.modelB, data = lexdata, sample.nobs = 210, :
lavaan WARNING: not all elements of the gradient are (near) zero;
the optimizer may not have found a local solution;
use lavInspect(fit, "optim.gradient") to investigate
summary(fitB, ci = TRUE, fit.measures=TRUE, standardized = TRUE)
lavaan 0.6-3 ended normally after 599 iterations
Optimization method NLMINB
Number of free parameters 50
Number of observations 210
Estimator ML
Model Fit Test Statistic 2597.331
Degrees of freedom 203
P-value (Chi-square) 0.000
Error in TEST[[2]] : subscript out of bounds
--
You received this message because you are subscribed to the Google Groups "lavaan" group.
To unsubscribe from this group and stop receiving emails from it, send an email to lavaan+un...@googlegroups.com.
To post to this group, send email to lav...@googlegroups.com.
Visit this group at https://groups.google.com/group/lavaan.
For more options, visit https://groups.google.com/d/optout.
Edward Rigdon
Apr 12, 2019, 10:58:25 AM4/12/19
to lav...@googlegroups.com
Dividing by a constant to reduce differences in variances, is not the same as standardization using estimates from the data, like standard deviations. If one observed variable has a variance in single digits and abother has a variance in teh range of 10,000, dividing the latter variable by 100 changes nothing statistically (you can easily undo the rescaling after the fact), but eliminates a potential problem. With the warning about a large difference in variances, that is the first thing I would do, before worrying about the otehr warning / error messages.
maryam nasseri
Apr 13, 2019, 11:13:24 AM4/13/19
to lav...@googlegroups.com
Thanks Edward, this explanation makes sense. I'm going to re-do the whole thing with this method instead of standardisation, to see if the fit also improves.
Just a question, if SEs are not calculated for a model (e.g., NA), is it valid to report the results of fit indices still? Are we supposed to report standard errors for a publication also? How do we interpret SEs together with fit indices?
Many thanks to whoever could help with this
Best
Edward Rigdon
Apr 13, 2019, 11:24:43 AM4/13/19
to lav...@googlegroups.com
If the software fails to estimate standard errors, I would be suspicious of the values reported for fit indices and parameter estimates. The model may not be identified, meaning that the reported parameter estimates may not be optimal--or there might be another problem that casts doubt on the model results. Moreover, without standard errors, you don't know what range of other values for the parameters would also be consistent with your model and data.
Message has been deleted
Camille Williams
Apr 15, 2019, 4:26:59 AM4/15/19
to lavaan
Hi,
Since I am interested in the loading and intercept obtained from the log transformation (to study allometry), I applied the log10 transformation to my data with the estimator = "MLR" since applying the log transformation made my data non-normal.
However, my model is having trouble converging because analyzing log-transformed data is making estimation difficult. I get this warning not for the configural, but for the metric and scalar invariance models:
lavaan WARNING: not all elements of the gradient are (near) zero; the optimizer may not have found a local solution; use lavInspect(fit, "optim.gradient") to investigateWhen I examine my optim.gradient, I see that I have 2 (2/139) of my residual correlations that are at - 0.001 in the scalar invariance cfa and 2 (2/139) other residual correlations that are at - 0.001 in the metric invariance model.
I have so many correlations (139) because I have one latent variable total brain vol which is inferred from 22 local brain volumes, which are obviously correlated with each other and must be taken into account to establish appropriate fit.
I do not get this error when I do not include any residual correlations but my fit is terrible based on the CFI TLI RMSEA & SRMR. When I add in theoretically supported residual correlations that have on the modindices between 105 and 300 in my model, I still get this warning:
> TBV0.model <- 'TBV =~ NA*CorticalWhiteMatterVol_log + + Brain_Stem_log + + Left_VentralDC_log + Right_VentralDC_log + + lhCortexVol_log + rhCortexVol_log+ + Left_Accumbens_area_log + Right_Accumbens_area_log+ + Left_Cerebellum_Cortex_log + Right_Cerebellum_Cortex_log + + Left_Caudate_log + Right_Caudate_log + + Left_Hippocampus_log + Right_Hippocampus_log+ + Left_Pallidum_log + Right_Pallidum_log + + Left_Putamen_log + Right_Putamen_log + + Right_Thalamus_Proper_log + Left_Thalamus_Proper_log+ + Left_Amygdala_log + Right_Amygdala_log+ TBV ~ c(0, 0.031)* 1+ TBV ~~ 1*TBV + lhCortexVol_log ~~ rhCortexVol_log+ Left_Caudate_log ~~ Right_Caudate_log+ Left_Putamen_log ~~ Right_Putamen_log+ Left_Cerebellum_Cortex_log ~~ Right_Cerebellum_Cortex_log+ Right_Thalamus_Proper_log ~~ Left_Thalamus_Proper_log+ Left_Accumbens_area_log ~~ Right_Accumbens_area_log+ Left_VentralDC_log ~~ Right_VentralDC_log+ Left_Amygdala_log ~~ Right_Amygdala_log+ Left_Hippocampus_log ~~ Right_Hippocampus_log+ Left_Pallidum_log ~~ Right_Pallidum_log + '> > fit_TBV<- cfa(TBV0.model, data=Abide_MGCFA, group = "DX_GROUP",estimator = "MLR")Warning message:In lavaan::lavaan(model = TBV0.model, data = Abide_MGCFA, group = "DX_GROUP", : lavaan WARNING: not all elements of the gradient are (near) zero; the optimizer may not have found a local solution; use lavInspect(fit, "optim.gradient") to investigateI'm starting to think that lavaan may not be adapted for me to investigate group differences in allometric scaling (log transformed loadings) and intercepts with this method.
Thank you very much for your patience and help,
Camille
On Friday, April 12, 2019 at 10:37:04 AM UTC+2, Terrence Jorgensen wrote:
Terrence Jorgensen
Apr 16, 2019, 4:43:48 AM4/16/19
to lavaan
Can I use an estimator on my log transformed data in the cfa ?
If you want to interpret results on the log scale, then that is fine, I was just checking that this was intentional (most researchers follow textbook advice to transform data to make it more normal, which is quite outdated advice in my opinion). I still would not recommend scaling your log-transformed variables, since standardized estimates are already available after estimation.
Camille Williams
Nov 13, 2019, 3:38:27 AM11/13/19
to lavaan
Hi Terrence,
Refresher: I am looking at group differences in the allometric scaling (power) relationship between total brain volume and each local volume (e.g. hippocampus, amygdala...). To look at this non-linear (power) relationship: Local Volume = Intercept + TBV^a, we use a log transformation to obtain a linear equation: log10(Local Volume) = Intercept + a * log10(TBV).
I want to compare my MGCFA and LMEMs results.
Question on Latent Variable Interpretation with the log10 transformation of observed variables
In my LMEMs, the log10(TBV) is the log10 transformation of my sum of local volumes (grey & white matter in the brain).
In my MGCFA, my observed variables are log10 transformed local volumes and they all load on my single latent variable. My colleagues and I were wondering if the latent variable corresponds to the log10(sum(local volumes)) or the sum(log10(local volumes)). To be coherent with the LMEMs, it should be the former.
Thank you very much for your valuable time and input,
Camille
Camille Williams
Nov 13, 2019, 3:43:00 AM11/13/19
to lavaan
Please ignore my last message. I found the answer here: https://groups.google.com/forum/#!searchin/lavaan/log$20transformation|sort:date/lavaan/lHPjOa_kYx4/GKSnNRPkAwAJ
Reply all
Reply to author
Forward
0 new messages