Error message for SEM

juliakatharinapfarr

I'm trying to do SEM in Jasp. Unfortunately, the program wouldn't take my model and giving me this error: 

"Error in nlminb(start=start.x, objective=objective_function, gradient=GRADIENT, : NA/NaN gradient evaluation"

Does anybody know what this means? 

This is my model (I'm trying to model the association of Neuroticism, Anxiety, Anhedonia, Cognitive Control and brain structure (=cluster). ):

#measurement model 

Anxiety =~ STAIS_Sum

Anhedonia =~ SHAPS_Sum

Neuroticism =~ NEOFFI_Neurotizismus

CognitiveControl =~ PfadfinderB

Cluster =~ Neuro_Eigenvalue_FA_anteriorthalamicradiation_R_neg_CLUSTER49

Kontrolle =~ PfadfinderA

Age =~ Alter

Sex =~ Geschlecht

#regressions

Anxiety ~ Neuroticism

Anhedonia ~ Neuroticism

Cluster ~ Neuroticism

Anhedonia ~ Cluster

CognitiveControl ~ Cluster

#correlations

Anhedonia ~~ Anxiety

CognitiveControl ~~ Anhedonia

Thank you for any help!

Julia

Ihnwhi

Hi @juliakatharinapfarr,

Thank you for this error report! We will try to figure it out what it means by.

Meanwhile, I am wondering about the two points:

(1) Do all latent variables have only one indicator? I am wondering about this because, usually, one factor has at least two indicators.

(2) It seems like Kontrolle, Age, and Sex are not used in either regressions or correlations. In this case, these three latent variables are treated as exogenous latent variables (i.e., explanatory latent variables). Then, the covariances among Kontrolle, Age, Sex, and Neuroticism are automatically estimated, and I am wondering whether this is what you want.

Cheers,

Ihnwhi

juliakatharinapfarr

Hi @Ihnwhi ,

Thank you for your reply!

about

(1) Yes, all latent variables have only one indicator, because we're using the total scores of the questionnaire measurements.

(2) I'm not sure if this is what I want, honestly. I want to control for Kontrolle, Age and Sex but don't want any of these variables to be used in either regressions or correlations. Do you have an advice how this is done correctly?

Thanks,

Julia

Ihnwhi

Hi @juliakatharinapfarr,

Thanks for your follow-up questions!

1. I see. If you a solid and concrete theory that a single indicator is sufficient for the latent variables (i.e., a single indicator captures the latent construct), that is not problematic! Otherwise, I would consider using the path analysis (a special case of SEM where all the latent variables are replaced with the observed variables). With regard to a discussion about the usage of single indicator in latent variable models, please refer to Hayduk and Littvay (2012) or Petrescu (2013).
2. If you want to control for Kontrolle, Age, and Sex, regressions should be specified from those three variables to the endogenous variables (endogenous variables refer to variables that are predicted by other variables in SEM). For example, if you want to control for the effects of Kontrolle, Age, and Sex on Anhedonia, you should additionally specify the regression below and see the parameter estimates:

Anhedonia ~ Kontrolle + Age + Sex

• One additional thought: for the control variables, I would directly use three observed variables rather than using latent variables!

Cheers,

Ihnwhi