mplus 8.1 and lavaan mismatch

[Dai Duong]

I run a model on lavaan (0.6-3) and Mplus 8.1 but results are very much different:

Lavaan (0.6-3)	Mplus (8.1)
library(haven) library(lavaan) x180108_lkh <- read.csv("//180914-lkh-with names.csv",header=TRUE, sep=",")   ctrl.labor <- " control_labor =~ income001+wh0001+skill+mwh21+njob income001 ~~   njob income001 ~~ wh0001 wh0001 ~~  njob income001 ~~ skill " fit.l <- sem(ctrl.labor, data=x180108_lkh,std.lv = TRUE,ordered=c("njob","skill")) summary(fit.l, fit.measures=TRUE,standardized=TRUE,rsquare=TRUE)	TITLE:  AoL;   DATA:   FILE = "\ \180914-lkh.csv";   VARIABLE:   NAMES ARE income wh0001 mwh21 skill njob;   USEVARIABLES ARE income wh0001 mwh21 skill njob;   CATEGORICAL ARE skill njob; MODEL: ctrll BY income* wh0001 mwh21 skill njob; ctrll@1; income WITH njob; income WITH wh0001; wh0001 WITH njob; income WITH skill; OUTPUT:  SAMPSTAT !RESIDUAL STANDARDIZED CINTERVAL FSCOEFFICIENT FSDETERMINACY TECH1 TECH2 TECH3 TECH4 TECH5 TECH6

Which result is reliable?

Both lavaan and Mplus use DWLS (because two indicators are categorical)

Thank you very much,

[Yves Rosseel]

On 10/1/18 7:51 PM, Dai Duong wrote:
> I run a model on lavaan (0.6-3) and Mplus 8.1 but results are very much

Indeed. I wouldn't trust either. The results are unstable, due to some
'extreme' values for some of the continuous variables.

Mplus 8.1 and lavaan have different strategies to handle these. I can
not tell you what Mplus 8.1 is doing. What I can tell you, is that older
version of Mplus would not run for this data. Mplus 6.1 says:

COMPUTATIONAL PROBLEMS ESTIMATING THE CORRELATION FOR SKILL AND INCOME.
INCREASING THE ITERATION OR CONVERGENCE OPTIONS MAY RESOLVE THIS
PROBLEM.
COMPUTATIONAL PROBLEMS ESTIMATING THE CORRELATION FOR SKILL AND MWH21.
INCREASING THE ITERATION OR CONVERGENCE OPTIONS MAY RESOLVE THIS
PROBLEM.
COMPUTATIONAL PROBLEMS ESTIMATING THE CORRELATION FOR NJOB AND MWH21.
INCREASING THE ITERATION OR CONVERGENCE OPTIONS MAY RESOLVE THIS
PROBLEM.


While Mplus 3.1 is even more explicit (and gives the problematic cases):

SERIOUS COMPUTATIONAL PROBLEMS OCCURRED IN THE BIVARIATE ESTIMATION OF
THE CORRELATION FOR VARIABLES SKILL AND INCOME. CHECK YOUR DATA.
IF THE PROGRAM RECOVERS FOR THIS PAIR OF VARIABLES (SEE TECHNICAL 6
OUTPUT), THE ESTIMATES ARE VALID. THE PROBLEM OCCURRED FOR THE
FOLLOWING
OBSERVATION(S):
OBSERVATION 5387
OBSERVATION 5387
COMPUTATIONAL PROBLEMS ESTIMATING THE CORRELATION FOR SKILL AND INCOME


SERIOUS COMPUTATIONAL PROBLEMS OCCURRED IN THE BIVARIATE ESTIMATION OF
THE CORRELATION FOR VARIABLES SKILL AND MWH21. CHECK YOUR DATA.
IF THE PROGRAM RECOVERS FOR THIS PAIR OF VARIABLES (SEE TECHNICAL 6
OUTPUT), THE ESTIMATES ARE VALID. THE PROBLEM OCCURRED FOR THE
FOLLOWING
OBSERVATION(S):
OBSERVATION 3120
OBSERVATION 3120
COMPUTATIONAL PROBLEMS ESTIMATING THE CORRELATION FOR SKILL AND MWH21


SERIOUS COMPUTATIONAL PROBLEMS OCCURRED IN THE BIVARIATE ESTIMATION OF
THE CORRELATION FOR VARIABLES NJOB AND MWH21. CHECK YOUR DATA.
IF THE PROGRAM RECOVERS FOR THIS PAIR OF VARIABLES (SEE TECHNICAL 6
OUTPUT), THE ESTIMATES ARE VALID. THE PROBLEM OCCURRED FOR THE
FOLLOWING
OBSERVATION(S):
OBSERVATION 2289
OBSERVATION 2289
COMPUTATIONAL PROBLEMS ESTIMATING THE CORRELATION FOR NJOB AND MWH21



I would remove these cases, and try again (in both Mplus and lavaan). If
the solution is stable, the results should be the same.

One last thing, in lavaan, you need to use parameterization = "theta",
in order to match the Mplus results exactly.

Yves.