Syntax

Syntax Guide

Statistical Programs
Statistical Primers

Interfaces with SPSS, Excel, Ascii, SAS and STATA Formats

This page helps you with the basics of Mplus and lavaan syntax. For a brief summary of Mplus syntax rules see this file and this link. For the Mplus Users guide, click here and for its technical appendix, click here. For an introduction to Mplus by UCLA, click here, here and here. For videos on uses of Mplus, click here. For videos that walk you through Mplus syntax click here, here and here. For an introduction to R, click here and here and for more advanced material on R, click here. For video introductions to R, click here. For an overview and syntax summary of lavaan, see this file and this link. For a free on-line course on lavann, click here and here. For two programs that provide additional lavaan-compatible tools, see semPlot and semTools.

My book features Mplus because I believe it has advantages over lavaan. However, lavaan is free, it integrates with a powerful stat package (R), and it has many strengths, so I provide some support for it here.

Program Evaluation and Randomized Explanatory Trials

M-10

The example study examined a weight loss program for overweight individuals represented by a variable called TREAT (0 = control group, 1 = treatment group). Weight loss was measured in pounds in the variable called WLOSS. It is the number of pounds lost per month. Positive numbers equal pounds lost; negative numbers indicate pounds gained. There were two targeted program mediators, self regulation (higher tendencies to self regulate oneself leads to more weight loss) and self-efficacy for weight loss (higher confidence in one's ability to lose weight leads to more weight loss). Self regulation is measured by two indicators (SR1 and SR2), as is efficacy (EFF1 and EFF2). The indicators were the average response to multi-item inventories where each item was responded to on a 5 point scale (-2 = strongly disagree, -1 = moderately disagree, 0 = neither, 1 = moderately agree, 2 = strongly agree). Higher scores mean more SR and more EFF.

There are two latent variables, one for self regulation and one for self-efficacy. I label latent variables with the letter L. The e represent measurement error because the indicators are assumed to be interchangeable (see my book). The d are disturbances that reflect unmeasured variables assumed to impact the variables they point to. To keep things simple, I do not include any covariates, but normally these would be in the model. The influence diagram is to the left.

This page shows you annotated syntax for a contrived example. I use this example because it is amenable to explaining syntax without getting caught up in substantive details. I provide syntax with detailed commentary lines (! in Mplus and # in lavaan). Obviously, syntax under normal programming conditions would not include the extensive comments.

! All lines have a 90 character limit. If you have more, hit
! carriage return before reaching 90 and continue typing.

! Mplus uses commands and subcommands of those commands.

! Termination of a command is indicated by a colon
! Termination of a subcommand is indicated by a semi-colon

! The below line tells Mplus you are going to provide a title

TITLE:

Web Example  ;  ! This is the title I gave

! The command below tells Mplus you are going to provide
! information about the data file, such as its location
  
DATA:

! The FILE subcommand tells Mplus where to find the data
! file. If no folder path is given, Mplus assumes the data
! file is in the same folder as the Mplus syntax. See my
! book for how the data file must be structured.

FILE IS webexample.dat ;

! The below command tells Mplus you are going to provide
! information about the variables in the analysis

VARIABLE:

! The subcommand below assigns names to the variables in the
! data file in the order they appear in that file. Names
! are limited to 8 characters. They are not case sensitive
 
   NAMES ARE ID WLOSS SR1 SR2 EFF1 EFF2 TREAT;

! The subcommand below identifies the subset of variables 
! the NAMES subcommand that will be used in the analysis

USEVARIABLES ARE WLOSS SR1 SR2 EFF1 EFF2 TREAT;

! The subcommand below tells Mplus that if it encounters a  
! value of -9999 on any variable, to treat it as missing
! data. The ALL word says to apply this rule to all variables

MISSING ARE ALL (-9999) ;

! The below command tells Mplus you are going to specify
! the type of analysis you want

ANALYSIS:

! The subcommand below identifies the type of estimator to  
! use. In this case, the robust maximum likelihood estimator
! with Huber-White adjustments is used

ESTIMATOR = MLR ;

! The below command tells Mplus you are going to describe
! the model   
    
MODEL:

! The BY keyword on a subcommand signfies the latent variable
! named to the left of BY has as indicators the variables 
! named to the right of BY. The first variable on the right 
! is treated as the reference indicator. In this case, EFF1
! and SR1 are the reference indicators. You create the name
! of the latent variable. It has an eight character limit
 
   LEFF BY EFF1 EFF2 ; 
   LSREG BY SR1 SR2 ;

! The ON keyword specifies a regression equation. The outcome
! variable is listed to the right of ON and the predictors are
! listed to the right of ON. Here, I regress the two latent
! variables onto the treatment condition and the weight loss
! variable onto the two mediators. The order of the predictor 
! listing is irrelevant

LEFF ON TREAT ;
   LSREG ON TREAT ;
   WLOSS ON LSREG LEFF ;

! The below command tells Mplus you want it to perform a
! detailed analysis of mediational chains. 
  
MODEL INDIRECT:

! The below subcommand identifies which mediational chains 
! you want the detailed analysis for. The keyword IND stands
! for indirect. The distal outcome is given to the left of
! IND and the distal predictor is given to the right of IND. 
! All mediational chains linking these two variables will be 
! analyzed by Mplus
 
   WLOSS IND TREAT ;

! The below command tells Mplus you are going to specify what
! you want on the output

OUTPUT:

! SAMP requests descriptive statistics
! STAND(STDYX) requests a fully standardized solution
! MOD(ALL 4) requests to show all modification indices > 4
! RESIDUAL requests to show residual analyses
! CINTERVAL requests to show confidence intervals
! PATTERNS requests to show patterns of missing data
! TECH4 requests technical output that is useful
! Details of the below are provided in my book

SAMP STAND(STDYX) MOD(ALL 4) RESIDUAL
   CINTERVAL PATTERNS TECH4 ;

! I sometimes use STAND(ALL) instead of STAND(STDYX)
! when I request standardized output. See Chapter 11 of 
! my book for the rationale.

! To see the output, go to the page 'Output' on this website

# This syntax assumes you have read your data into R into a
# data frame called "XXX". If you do not know how to do so
# see the "Programs" or "Syntax II" tab on this webpage

# line below loads the R library lavaan
 
library(lavaan)

# line below makes the variables in "data" available for
# reference

attach(XXX)

#Formula type                 Operator    Mnemonic

#latent variable definition      =~       is measured by
#regression                      ~        is regressed on
#(residual) (co)variance         ~~       is correlated with
#intercept                       ~1       intercept

# line below indicates you are going to specify the model.
# You can use any name you want. I used the word "model". 
# Follow the word by <- which is read as "equals" and
# the single apostrophe signals the start of the model 
# specification

model<- '

# lines below specify the measurement model. The =~ notation
# is read as "is indicated by". The name of the latent
# variable is on the left of =~ and the indicators are on the
# right of it. The first listed indicator is assumed to be 
# the reference indicator. SR1 and SR2 are in the data
# frame "data". The names are case sensitive. You make up 
# the names of the latent variables. No length restrictions
# but best to avoid special characters and long names
 
LSREG =~ SR1 + SR2
LEFF =~ EFF1 + EFF2

# lines below specify the causal model. The ~ indicates a
# regression equation is being specified. The outcome is
# on the left of ~ and the predictors are on the right of
# ~. You must include the + signs. I assign labels for
# the regression coefficients using p notation to stand
# for "path coefficient" but such labels are optional. I
# use them because I need to make use of the labels for 
# mediation analysis.  
  
LSREG ~ p1*TREAT
LEFF ~ p2*TREAT
WEIGHT ~ p3*LSREG + p4*LEFF

# to conduct mediation analysis, I must use labels to
# indicate the mediational chains I want to analyze. I
# must use path tracing rules to specify them, which 
# are described in my book. The := notation indicates
# that path coefficients are going to algebraically
# manipulated. The name of the variable that will contain
# the result of the manipulations is on the left of :=.
# You can use any term you want (I used "p1p3" and "p2p4").
# The first line below is the chain from the treatment 
# to the latent self regulation variable to weight loss.
# The second line below is the chain from the treatment 
# to the latent efficacy variable to weight loss.
 
p1p3 := p1*p3 
p2p4 := p2*p4

# For this model, the total effect of the treatment on
# weight loss is the sum of the two mediational chains.
# The line below calculates it. An advantage of Mplus
# is that it automatically identifies the relevant chains
# and does the calculations without you specifying the
# algebra

tot := p1p3 + p2p4
'
# The above apostrophe indicates you are finished 
# describing the model

# The line below tells lavaan to fit the model to the data
# using the sem function, identifying the model by its 
# assigned name and the data frame where the data are.
# The MLR estimator is the robust maximum likelihood with
# Huber-White adjustments. By including the mimic option
# the analysis will mimic the modeling choices made in 
# Mplus. Otherwise, lavaan makes its own choices, some of
# which differ from Mplus
  
fit <- sem(model, data=XXX,estimator='MLR',mimic='mplus')

# The line below prints out the model results including a
# standardized solution
 
summary(fit,standardized=TRUE)

# The line below prints out modification indices

modindices(fit)

# The line below prints out the predicted covariances by
# the model

fitted(fit)

# The line below prints out an analysis of residuals, 
# including standardized residuals

resid(fit, type="standardized")

# The line below prints out the log likelihood for the model
 
logLik(fit)

# The line below prints out the BIC for the model

BIC(fit)

# The line below line prints out a large number of global
# fit indices, albeit in a clunky format

fitMeasures(fit)

Mplus Syntax (! are comments ignored by Mplus) Lavaan Syntax (# are comments ignored by lavaan)

To see a diagram of your model after running it from Mplus, click on the menu item within the Mplus interface called ‘Diagram’ and then click on ‘View Diagram.’ You can edit the figure that appears using the 'View' menu item within the graphic file. To see a diagram in lavaan, you can use the R package SemPlot and input your executed lavaan model into it (see the command below).

Mplus Generated Diagram

Mplus gives you the option to add to the diagram a variety of results in the form of unstandardized coefficients, standardized coefficients, etc. You can export the diagram to a pdf file and from that file, to an image file. You can edit the figure in multiple ways

semPlot Generated Diagram (semPaths(fit,title=FALSE,curvePivot=TRUE,intercepts=FALSE)

semPlot gives you the option to add to the diagram a variety of results in the form of unstandardized coefficients, standardized coefficients, etc. You can export the diagram to a pdf file, a jpg file, and many other formats.