Please read following article it will help you

http://journals.sagepub.com/doi/abs/10.4256/mio.2010.0030

See also here: https://www.youtube.com/watch?v=K34sF_AmWio

If SEMs are giving you a bit of confusion, you may want to consider its close cousing the directed acyclic graph (DAG) as a method. 

e.g.: https://www.researchgate.net/publication/278717528_Graphical_Causal_Models

Chapter Graphical Causal Models

Hi Murumba,

Regarding your inquiry, I had dificulty to understand what you mean with "control variable" term. As far I understand, a control variable is one that the researcher purposely neutralizes or cancels in a research in order to prevent it from interfering in the analysis of the relationship between the independent and dependent variables. In biological sciences it could be temperature, humidity, and so on. So you do'n't want them to interfere in your analysis, so you try to keep it constant and not to add it to the model.

On the other hand, if you can't control this variable, such as gender or , considering your research field, let say national and multinational companies, you use this variable as a moderator in a multigroup analysis.

Basically you build a baseline model and split into the groups provided by the moderator and compare results between these groups. Usualy you have to dichotomize the moderator if it is not nominal variable, such as gender. If it is a continuous variable you can use median to split in two groups.

So, as far I understand, a moderator can be interpreted as a control variable, if I really capture the core of your question.

If you want to continue this discussion, feel free to call me back.

Regards,

Alcides Barrichello

Thank you all for your answers and recommendations.

Hi Alcides Barrichello, I was a bit confused when I saw in some articles, the authors already has moderator and mediator incorporated in their model, and in addition include control variables like corporate size, Leverage on the ground that previous studies do that. Then in interpretation of results you see statement like Corporate size or leverage is not significant or it is significant. This made me think that if you begin to talk about control variable being significant or not then they are not different from any other independent or predictor variable used in the model. 

Hi Murumba,

Now I understood your point and agree with you: this is really weird !

Looking to some literature I have (unfortunatelly it is only Portuguese), I found two possibilities but both do not address your issue.

First one is that you can have a mediator which is, at the same time, a moderator (you call it a moderator-mediator), a nominal moderator, like corporate size for instance (big company / small company), in such way the variable is, at the same time, mediator and moderator, depending upon the way you treat it. But, when seen as a moderator, you don't have to include it in the model and there will be no significance to report for it, only for the other variables when the company is big or small.

Second situation is when you have a multigroup moderator and again company size could be one to act like this in such way you could compare structural models between two different groups, big and small companies, but, again, you don't have to include this is the model and there will be no significance either for this sort of variable.

You know, this made me curious. Would you mind to send me a copy of one of those articles in such way I can better understand how this is reported by the authors ?

Seeing what they say maybe I can give you some help and/or learn something new about the way I have to present the results of my future papers. 

Regards,

Alcides Barrichello

Okay Alcides Barrichello. I will try and get one of those papers across to you.

Regards.

Murumba Inekwe

Hello Alcides Barrichello please see the attached article and the part I highlighted. Am confused on the impact it makes to the study. If the results were the other way round would have made any difference considering they are control variables.

Hi,

I,ll read as soon as I can and return to you.

Alcides Barrichello

Hi Murumba,

 I think I got the issue, at least for my understading.

 Instead of using "control variable" expression, it would be better if Bukit and Iskandar (2009) would have used "confounding variable" or "confounder" ("A confounding variable is characterized as one that correlates with both a treatment (or predictor of interest) and outcome (Anderson et al.1980, chap. 5). It is also considered to occur causally prior to the treatment (see Cook and Campbell 1979). In other contexts, the confounding variable might be referred to as a “disturber” (Steyer and Schmitt 1994), a “covariate” (Holland 1986, 1988), or, drawing from the language of R. A. Fischer, a “concomitant” variable (Pratt and Schlaifer 1988)." - Frank, K. A. (2000). Impact of a confounding variable on a regression coefficient. Sociological Methods & Research, 29(2), 147-194).

 Considering this sort of concept (sorry if I'm so detailed, but this was new for me), you must consider the confouders otherwise the conclusion you get about IVs and DVs may be incomplete, even if, at the end, the relationship involving them are not significant (you only know this after running the regression).

 Not using the confounder can incur in what is called OVB (omitted variable bias).

 On the other hand, Frank (2000) "extended existing techniques for sensitivity analyses by indexing the impact of a potentially confounding variable on the

statistical inference with regard to a regression coefficient." which can be useful to accept or not hte use of a specific confounder.

 The key point is the theory behind the confunder to be used, I mean, theoretical base to define if a variable is really a confounder or not.

 Becker, DeFond, Jiambalvo & Subramanyam (1998) are more clear regarding to the use of confounders, but still didn't go deeper in the theme because it is much more a tool to be used instead of part of their article (see attached).

 I found some videos at YouTube which are quite helpful to clarify this (see below).

 1) https://www.youtube.com/watch?v=KMqEQI7QO0E

 2) https://www.youtube.com/watch?v=5tMMESxjDBg#t=970.701

 3) https://www.youtube.com/watch?v=1Cn1smM3kbQ

 4) https://www.youtube.com/watch?v=mFIvsqVir48

 I hope that this can help you.

 See you.

 Alcides Barrichello

Hi  Alcides Barrichello. Thank you very much. It was confusing to me as control variable because the way i understood control variable is that once you have taken care of it base on your theoretical reasoning or literature support you dont need to interprete the results of the individual  control variables again. I mean if ,for example you have corporate size as a control variable, saying that firm size is significant or not after your analysis,,I see no need for that. Because does that affect the results which you were trying to control by introducing the control variable? Your explanation is much clearer.

Regard.

Murumba

Hi

Please refer to the references:

1. Donna L. Schminkey Timo von Oertzen Linda Bullock(2016). Handling Missing Data With Multilevel Structural Equation Modeling and Full Information Maximum Likelihood Techniques. https://doi.org/10.1002/nur.21724.

Volume39, Issue4, 286-297. https://onlinelibrary.wiley.com/doi/full/10.1002/nur.21724

2. Data Analysis: Structure Equation Modeling (SEM)

January 2016, DOI: 10.1007/978-981-10-0053-9_4. In book: Expatriate Manager’s Adaption and Knowledge Acquisition.

https://www.researchgate.net/publication/301256051_Data_Analysis_Structure_Equation_Modeling_SEM

3. Rufus Lynn Carter (2006). Solutions for Missing Data in Structural Equation Modeling. Research & Practice in Assessment, Volume One: Winter 2006, 1-7,

https://files.eric.ed.gov/fulltext/EJ1062693.pdf

4. Paul D. Allison (2003). Missing Data Techniques for Structural Equation Modeling. Journal of Abnormal Psychology, Vol. 112, No. 4, 545–557.

https://statisticalhorizons.com/wp-content/uploads/Allison-2003-JAP-Special-Issue.pdf

Regards,

Zuhair

Thanks al-Hemyari. I appreciate

thank you Professor Alcides Barrichello and Murumba for great discussions, I was there following. I also have a problem of control variable. when add control variable to my model in amos are not significant to other constructs in my model and fit indices become worse , when removed model fit is ok. what can I do with such situation ? please advise me.

Hi Bruce,

Nice to receive your message.

Regarding your question, let`s assume that what you mention as a control variable is a variable that the researcher purposely neutralizes or cancels in a research in order to prevent it from interfering in the analysis of the relationship between the independent and dependent variables.

For instance, if company size is a control variable this means that you must use just one type of company size in your analysis in such way there will be no variation regarding company size when deal with the statistics. The parameters to define the company size you will use is yours, but important is that you will deal with company sizes quite similar.

In this sense, you don`t have to use the control variable in your model since it was controlled previously to data collection.

On the other hand, if you didn't do such control but you suspect that a certain variable will influence the responses of your model, you may check if this variable is a moderator, influencing the relation IV --> DV depending upon its intensity.

Using the same example of company size, you have large companies and small companies in your sample, you must build a general model, w/o using company size in it and after that check if there is any difference in the relation IV --> DV when you consider a subsample of small size and another one of large size.

This can be done by Testing for Multigroup Invariance and Byrne's book (

Byrne, B. M. (2010). Structural Equation Modeling with AMOS: basic concepts, applications, and programming (2nd ed.). New York: Routledge.), chapter 7 (Testing for the factorial equivalence of scores from a measuring instrument - First-order CFA model) should be quite useful for that.

You can also try some James Gaskin videos in YouTube; this guy is really good in AMOS use.

Hope this helps.

See you.

thank you Professor Alcides Barrichello . Answer is very clear

Professor Alcides Barrichello , I see that you are very active and responsive , I would like to take this opportunity and ask you another Question regarding CFA. I followed Dr. Gaskin video and I run CFA in Amos. by modification indices ( suggestions from Gaskin), my model fit indices are now within acceptable Range.( ie. CFI=.945/ RMSEA=.59/TL=.93/ PCLOSE=009/ CIM/DF=2.452 & SRMR=045).

My question is; I saw many papers measuring second order factor but honestly I don't know if it is a must. should i go for second order factor? or just report my results above. what is your suggestion

thank you in advance

Hi Bruce.

In fact all depends upon the theory you use (actually statistics is always a tool and can be totally meaningless if you don't have a good theory behind the model).

If the theory says a construct A is a 2nd order one, compound by constructs B and C (let's call them "subconstructs" from now on; this is more a poetic license than a real used word) and your results confirm this, that's OK. But, if the model doesn't fit in this approach you should try directly the subconstructs in place of the 2nd order construct.

Attached is a paper I published where the model has some 2nd order constructs but I decided to check if this was the best approach and try some 1st order alternative (actually 3 alternative models). At the end I confirmed that 2nd order model was the best.

In the other way around, if your theory treats the constructs as 1st order (it seems this is your case), there is no reason to try a 2nd order approach, even because what would this 2nd order construct be ?.

If your variables were presented in some papers as a 2nd construct, them you should try; if not, forget about and keep 1st order approach.

See my paper, remembering that, in my case, theory says there are some 2nd order constructs and I decided to check if this was the best approach much more because in Social Sciences sometimes things are not straitghforward.

Hope this can help you.

Any doubt, feel free to call me back.

Regards.

Thank you so much , Professor Alcides Barrichello