jmaslow posted on Thursday, December 08, 2011 - 12:24 pm
I am estimating a multiple group model using wlsm estimation because some indicators are categorical. It seems from other responses that I should use STD, not STDYX coefficients for standardized results. However, STD produces factor loadings greater than 1. Does this mean STDYX is the appropriate choice? Thank you!
I have estimated a latent growth curve model. I estimate two latent intercepts and two latent slopes. I have two latent factors that predict the latent intercepts and slopes. i also include a set of binary (dummy codes for race and gender) and a continous SES variable as predictors of all the latent factors in the model. Based upon Linda's previously post that says,
"... depends on the scale of the dependent and independent variable involved. for example,
factor ON factor STD factor ON binary x STDY or STD factor on continuous x StdYX
I am assuming the dependent variable is a factor. If it is an observed variable,
observed ON binary x Stdy observed ON continuous x StdYX"
I would use a combination of STD and StdYX.
If this is the case, can i compare STD and StdYX estimates to each other?
I have categorical factor loadings on my latent variables but the covariates that I'm controlling for are continuous. Once I add them into the model, the output for the standardized results does NOT include the standard errors or p-values - just the betas. The unstandardized results still include the SE's and p-values for each of the betas. Can you tell me how I can fix the problem of the standardized output not including the full information? Here is the syntax for my model:
use variables are bri1_r bri12_r bri19_r bri28_r bri33_r bri42_r bri51_r bri57_r bri69_r bri72_r d8 d9 d10 d13 d15 age; categorical are bri1_r bri12_r bri19_r bri28_r bri33_r bri42_r bri51_r bri57_r bri69_r bri72_r d8 d9 d10 d13 d15; Missing are all (-9999); Model: EMOCTRL by bri1_r bri12_r bri19_r bri28_r bri33_r bri42_r bri51_r bri57_r bri69_r bri72_r; HARDIS by d8 d9 d10 d13 d15; HARDIS on EMOCTRL; EMOCTRL on age; d13 with d15 (p); output: standardized; modindices;
We don't provide these values for models with covariates and WLSMV.
janemcready posted on Thursday, April 24, 2014 - 12:00 am
As you stated above using STDY or STDYX depends on the scale of the dependent and independent variable involved. How can I calculate and find which one can be used. What is the common things we should note in such cases.
It depends on the independent variable. StdYX with continuous. StdY with binary.
Leah Lipsky posted on Tuesday, September 08, 2015 - 11:56 am
I'm confused as to which standardized estimate to report for a path in a structural equation model with several latent and observed variables including 1 binary observed variable (b1regimen). The input statement is pasted below, and the path I am unsure about is indicated by the statement: l1a1c on ... b1regimen. Thanks very much.
VARIABLE: NAMES are MASTERID AGE C1DEMAND C1DFR C1PCC C1PI_P1 C1PI_P2 C1PSYCTL C1RESPON G1DFR G1PCC L1A1C P1DEM21 P1DFR P1DSMP P1PCC B1REGIMEN;
MISSING are all .;
USEV are age c1respon C1DEMAND C1PCC C1PI_P1 C1PI_P2 C1PSYCTL L1A1C P1DEM21 P1DSMP P1PCC B1REGIMEN;
l1a1c on p1dem21 p1dsmp b1regimen; b1regimen on p1dem21; p1dsmp on p1dem21 pcc cpi; pcc on p1dem21 auth; cpi on auth age; auth on p1dem21; c1pi_p1 on age; c1pi_p2 on age; p1pcc on age; c1respon on age;
You use WLSMV so your b1regimen predictor is its underlying continuous latent response variable. Therefore you should use STDY, standardizing relationships among DVs, which in fact is the same as STDYX in this case.
For my PhD research, I run a model in which my outcome variable is dichotomous and my independent variables are continuous. I would like to assess the mediating effect of two continuous variables in the relationship between three continuous variables and the outcome variable (which is binary). Therefore, in the output of the command, I wrote the following syntax:
Output: stdyx CINT(BCBOOTSTRAP);
I would like to ask which results I should report? For the direct effects on the outcome variable, should I report the standardized (STDYX) or no standardized estimates? I would like to say that for running this model, I inserted the latent variables, and not the observed variables. In addition, for the bootstrap, should I report the standardized (again STDYX)or no standardized estimates? Many thanks in advance for your help,
Different fields have different traditions. Some require standardized but unstandardized is fine as long as you understand how big the effects are. If you have an observed continuous X variable and you want to standardize you should use STDYX. All this is explained in our book.
I have several latent variables in my model. I treated the the indicators as categorical because they are measured with likert-type scales. I have used WLSMV as the estimator.
I am having factor loadings greater than 1 for one of the indicators. That specific item is on a 7-point Likert-type scale, different from rest of my indicators which are usually on a 5-point scale for other latent variables.
For all estimates (STD STDY STDYX) the value is greater than 1. I was wondering what your recommendation would be. Can you also help me understand why this might be happening? I have always used the WLSMV as the estimator with categorical indicators, never had this before.
Just to clarify: when I check "Residual Variances" for the latent variables at the end of STDYX, none of the values are negative.
I cannot see anything printed as "remainders" but, when I check the R-square section for the observed variables, I can see that only the item with an estimated value greater than 1 has a negative residual variance. The same row says "Undefined" for the Estimate column and the S.E. = 0.10280E+01.
What does having the negative residual variance indicate for this item? Does this mean that there is a misfit in this model? I reckon this might be because the item is highly stigmatising (asking about illegal behaviours) so may be due to a floor effect. What is your advice in these cases? to try fitting an alternative model?
Right - you get the negative residual variance info in the R-square section. This means that the model should not be used as is but needs modification. If you have more than one factor, you can try an EFA to see if that gives you ideas.