lavaan() - Mediation model: RMSEA p-value = NA and mod indices 'rows with length 0'

Anna

unread,

May 27, 2017, 2:11:44 AM5/27/17

to lavaan

Dear all,

I am a little desperate in trying to fit a - I thought it would be - simple mediation model in lavaan.

My problem is that in the output RMSEA is 0.00 (which is actually good since RMSEA < .05 indicates a good fit, if I understood it correctly) but the p-value 'NA'. How do I need to interpret the 'NA' ?

When asking for the modifications indices R tells me ' <0 rows> (or 0-length row.names) ', which I don't understand.

I only found one post in this forum related to my problem, advising to either rescale the variables (because they were measured on different scale) or to check whether R processes missing values wrongly . However, I already did the first by using only standardized variables and checked for the second, which did not apply to my data. I would be very happy to get any further help/ advices/ suggestions here, regarding what I could do to solve this problem.

Thanks in advance,

Anna

MODEL:

modelLSPCl <- '

Clarity ~ c*LSP

Interference ~ a*LSP

Clarity ~ b*Interference

#indirect effects

# interference mediates LSP to Clarity

m := a*b

#total effects

total := c + (a*b)

#variances

Clarity ~~ Clarity

LSP ~~ LSP

Interference ~~ Interference

#Intercepts

Clarity ~ 1

LSP ~ 1

Interference ~ 1'

modelLSPCl_fit <- lavaan(modelLSPCl, fixed.x =FALSE, data = EEMT)

summary (modelLSPCl_fit, fit.measures=T, standardized = T, rsq=T, modindices = T)

OUTPUT:

lavaan (0.5-23.1097) converged normally after 12 iterations

Number of observations 127

Number of missing patterns 1

Estimator ML

Minimum Function Test Statistic 0.000

Degrees of freedom 0

Model test baseline model:

Minimum Function Test Statistic 11.639

Degrees of freedom 3

P-value 0.009

User model versus baseline model:

Comparative Fit Index (CFI) 1.000

Tucker-Lewis Index (TLI) 1.000

Loglikelihood and Information Criteria:

Loglikelihood user model (H0) -533.290

Loglikelihood unrestricted model (H1) -533.290

Number of free parameters 9

Akaike (AIC) 1084.580

Bayesian (BIC) 1110.178

Sample-size adjusted Bayesian (BIC) 1081.716

Root Mean Square Error of Approximation:

RMSEA 0.000

90 Percent Confidence Interval 0.000 0.000

P-value RMSEA <= 0.05 NA

Standardized Root Mean Square Residual:

SRMR 0.000

Parameter Estimates:

Information Observed

Standard Errors Standard

Regressions:

Estimate Std.Err z-value P(>|z|) Std.lv Std.all

Clarity ~

LPS (c) 0.205 0.086 2.390 0.017 0.205 0.205

Interference ~

LPS (a) -0.108 0.088 -1.225 0.221 -0.108 -0.108

Clarity ~

Interference (b) -0.166 0.086 -1.932 0.053 -0.166 -0.166

Intercepts:

Estimate Std.Err z-value P(>|z|) Std.lv Std.all

.Clarity -0.000 0.085 -0.000 1.000 -0.000 -0.000

LPS 0.000 0.088 0.000 1.000 0.000 0.000

.Interference 0.000 0.088 0.000 1.000 0.000 0.000

Variances:

Estimate Std.Err z-value P(>|z|) Std.lv Std.all

.Clarity 0.916 0.115 7.969 0.000 0.916 0.923

LPS 0.992 0.125 7.969 0.000 0.992 1.000

.Interference 0.981 0.123 7.969 0.000 0.981 0.988

R-Square:

Estimate

Clarity 0.077

Interference 0.012

Defined Parameters:

Estimate Std.Err z-value P(>|z|) Std.lv Std.all

m 0.018 0.017 1.035 0.301 0.018 0.018

total 0.223 0.087 2.576 0.010 0.223 0.223

Modification Indices:

[1] lhs op rhs mi epc sepc.lv sepc.all sepc.nox

<0 rows> (or 0-length row.names)

Yves Rosseel

unread,

May 27, 2017, 4:01:55 AM5/27/17

to 'Anna' via lavaan

On 05/27/2017 08:11 AM, 'Anna' via lavaan wrote:
> My problem is that in the output RMSEA is 0.00 (which is actually good
> since RMSEA < .05 indicates a good fit, if I understood it correctly)
> but the p-value 'NA'. How do I need to interpret the 'NA' ?
> When asking for the modifications indices R tells me ' <0 rows> (or
> 0-length row.names) ', which I don't understand.

All of this has one simple explanation: you have df=0, because you have
fitted a saturated model: there are as many free parameters as sample
statistics.

There is nothing wrong with this. But as your fit is already perfect,
fit indices are useless, and modification indices (suggesting how to
improve your fit) can not be computed.

Yves.

Anna

unread,

May 27, 2017, 5:33:52 AM5/27/17

to lavaan

Ohh I did not expect this.. Thats actually pretty nice, thanks a lot, Yves !!

Edward Rigdon

unread,

May 27, 2017, 12:49:13 PM5/27/17

to lav...@googlegroups.com

Your model is saturated. DF = 0, no fit assessment possible.

--
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+unsubscribe@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.

Message has been deleted

Maria Usacheva

unread,

Dec 19, 2017, 7:36:30 PM12/19/17

to lavaan

Hi,

I see that it's an old thread, but I'll try my luck. I am also running a mediation model in lavaan with bootsrtapped SE, and have the exact same problem: p-value = NA, but my model is not saturated, and I'm also only using z-scores, so it's not the scaling problem either. "mompa" and "momabhx2" are dummy-coded binaries ("mompa" is declared as ordinal), is that the problem?. What would be the right solution here? Thank you!!

set.seed(444)

model5 <- 'zcceadiff ~ b1*mompa + zcdpea + c*momabhx2

mompa ~ zpeadiff + mommhvz + a1*momabhx2 + zcdpea

zpeadiff ~ momabhx2

zcdpea ~ momabhx2

mommhvz ~ momabhx2

zpeadiff ~~ zcdpea

direct := c

indirect1 := a1 * b1

total := c + (a1 * b1)'

fit5 <- sem(model = model5, data = dat, estimator= "ULS", optimizer = "objectiveGLS",

ordered = c("mompa"), missing = "default", se = "boot")

summary(fit5, standardized = TRUE, fit = TRUE, ci = T)

lavaan (0.5-23.1097) converged normally after 13 iterations

Used Total

Number of observations 216 217

Estimator ULS

Minimum Function Test Statistic 21.798

Degrees of freedom 4

P-value (Unknown) NA

Model test baseline model:

Minimum Function Test Statistic 84.750

Degrees of freedom 15

P-value NA

User model versus baseline model:

Comparative Fit Index (CFI) 0.745

Tucker-Lewis Index (TLI) 0.043

Root Mean Square Error of Approximation:

RMSEA 0.144

90 Percent Confidence Interval 0.088 0.206

P-value RMSEA <= 0.05 0.004

Standardized Root Mean Square Residual:

SRMR 0.085

Weighted Root Mean Square Residual:

WRMR 0.953

Parameter Estimates:

Information Observed

Standard Errors Bootstrap

Number of requested bootstrap draws 1000

Number of successful bootstrap draws 1000

Regressions:

Estimate Std.Err z-value P(>|z|) ci.lower ci.upper Std.lv Std.all

zcceadiff ~

mompa (b1) 0.249 0.093 2.695 0.007 0.060 0.431 0.249 0.256

zcdpea 0.174 0.103 1.691 0.091 -0.022 0.389 0.174 0.182

momabhx2 (c) -0.090 0.091 -0.988 0.323 -0.278 0.080 -0.090 -0.076

mompa ~

zpeadiff 0.237 0.097 2.446 0.014 -0.003 0.392 0.237 0.228

mommhvz -0.186 0.108 -1.721 0.085 -0.408 0.006 -0.186 -0.150

momabhx2 (a1) 0.337 0.107 3.138 0.002 0.123 0.551 0.337 0.275

zcdpea -0.117 0.093 -1.255 0.209 -0.285 0.069 -0.117 -0.119

zpeadiff ~

momabhx2 -0.094 0.072 -1.306 0.192 -0.238 0.051 -0.094 -0.080

zcdpea ~

momabhx2 0.071 0.083 0.856 0.392 -0.081 0.241 0.071 0.057

mommhvz ~

momabhx2 0.226 0.063 3.567 0.000 0.101 0.347 0.226 0.230

Covariances:

Estimate Std.Err z-value P(>|z|) ci.lower ci.upper Std.lv Std.all

.zpeadiff ~~

.zcdpea 0.192 0.079 2.431 0.015 0.032 0.331 0.192 0.187

Intercepts:

Estimate Std.Err z-value P(>|z|) ci.lower ci.upper Std.lv Std.all

.zcceadiff 0.011 0.083 0.137 0.891 -0.154 0.175 0.011 0.011

.mompa 0.000 0.000 0.000 0.000 0.000

.zpeadiff 0.052 0.091 0.574 0.566 -0.122 0.244 0.052 0.053

.zcdpea -0.091 0.093 -0.976 0.329 -0.279 0.091 -0.091 -0.087

.mommhvz -0.134 0.065 -2.057 0.040 -0.263 0.002 -0.134 -0.162

Thresholds:

Estimate Std.Err z-value P(>|z|) ci.lower ci.upper Std.lv Std.all

mompa|t1 0.661 0.115 5.777 0.000 0.458 0.901 0.661 0.646

Variances:

Estimate Std.Err z-value P(>|z|) ci.lower ci.upper Std.lv Std.all

.zcceadiff 0.906 0.137 6.624 0.000 0.639 1.158 0.906 0.911

.mompa 0.919 0.919 0.919 0.919 0.877

.zpeadiff 0.964 0.137 7.047 0.000 0.707 1.246 0.964 0.994

.zcdpea 1.085 0.178 6.099 0.000 0.751 1.445 1.085 0.997

.mommhvz 0.645 0.063 10.209 0.000 0.522 0.768 0.645 0.947

Scales y*:

Estimate Std.Err z-value P(>|z|) ci.lower ci.upper Std.lv Std.all

mompa 1.000 1.000 1.000 1.000 1.000

Defined Parameters:

Estimate Std.Err z-value P(>|z|) ci.lower ci.upper Std.lv Std.all

direct -0.090 0.091 -0.990 0.322 -0.278 0.080 -0.090 -0.076

indirect1 0.084 0.043 1.933 0.053 0.012 0.183 0.084 0.071

total -0.006 0.085 -0.074 0.941 -0.182 0.165 -0.006 -0.005

Maria Usacheva

unread,

Dec 19, 2017, 7:46:32 PM12/19/17

to lavaan

I narrowed it down to using "ULS" estimator for bootstrapping, as opposed to using ULSMV estimator without bootstrapping. I wonder why such difference, and what would be appropriate to report or how to interpret p = NA with estimator ULS?

On Friday, May 26, 2017 at 11:11:44 PM UTC-7, Anna wrote:

Yves Rosseel

unread,

Dec 20, 2017, 3:25:03 AM12/20/17

to lav...@googlegroups.com

When estimator is ULS, the test statistic does not follow a chi-square
distribution. Not even asymptotically. Therefore, we have p = NA.

You can use ULSM or ULSMV.

Yves.

> [1] lhs op rhs mi epc sepc.lv <http://sepc.lv>

> sepc.all sepc.nox
> <0 rows> (or 0-length row.names)
>

> --
> 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
> <mailto:lavaan+un...@googlegroups.com>.

> To post to this group, send email to lav...@googlegroups.com

> <mailto:lav...@googlegroups.com>.

Reply all

Reply to author

Forward