SEM Series Part 1: Developing a good model and hypotheses

[@Gaskination]

Here's a fun pet project I've been working on: udreamed.com/. It is a dream analytics app. Here is the KZbin channel where we post a new video almost three times per week: kzbin.info/door/iujxblFduQz8V4xHjMzyzQ Also available on iOS: apps.apple.com/us/app/udreamed/id1054428074 And Android: play.google.com/store/apps/details?id=com.unconsciouscognitioninc.unconsciouscognition&hl=en Check it out! Thanks!

[@Truth789]

Sir James Gaskin, I have no words to praise you but can say that you are more than my physical so-called guide/supervisor.

[@PastelBlueiscol]

Dear James, you are the smartest guy on the youtube... Your videos are very informative and comprehensive...

[@Gaskination]

Thanks! You're nice :) I'm glad these videos are helping people out.

[@Gaskination]

+Borith Long For some reason KZbin won't let me reply directly to your comment on my video, so I'll reply here. I recommend looking at the kurtosis and also the outliers for both variables. You may discover that there are a few that are throwing it off. If that is not the case, then you may have simply uncovered an interesting phenomenon. Or it may be because those who have received such violations are simply more aware (reverse causality). The issue may be that your data was collected at the same time for both variables, instead of increasing awareness (or just assessing awareness) and then observing behavior over the subsequent months.

[@borithlong4769]

Thank you very much. :)

[@JoSeoyeon]

James, I would like to thank you. Your videos have really helped our research group on our paper on SEM. Our proposal got accepted last night all thanks to the info from your videos mostly. Cheers from the Philippines!

[@Gaskination]

Wahoo! Congrats!

[@Gaskination]

Yes. The difference in scale doesn't actually make a big difference since we are examining relative movements rather than absolute movements.

[@Gaskination]

Perhaps in the future, but not soon. I will put it on my Gaskination To Do list. I just have too many research ideas I want to pursue at the moment.

[@chrisross9355]

Your video just helped a desperate grad student over here. Good Karma your way. Thanks a ton!!!

[@mssr1965]

As I see your 1st video of SEM series will be amazing.

[@aminamin34]

Dear Professor, Thanks So much. All videos are useful. I just ask you, if possible continue your series with PLS.. Thanks

[@sajeeb2005]

In Nepal there is no institutions teaching SEM yet, your video will surely help me for my doctoral dissertation.

[@MohitYadav-wq7jr]

Dear james, plz guide me on the correlated and uncorrelated models in SEM .How to interpret the models?

[@borithlong4769]

Hello James, thank you very much for ur informative video. I would like to ask you a question regarding the hypothesis and sign of estimated coefficient. Suppose we hypothesis that Awareness of drink driving will have negative (negative sign) effect on traffic violation...and we know that it must be like that but the result showed that estimate coefficient is positive. and I think there is no mistake regarding the questionnaire (answer from respondent are mostly disagree but i still get positive result)...Do you know how to deal with this issue? problem with covariance?? Thank you very much in advance.

[@sonaligarg9633]

First of all, I would like to thank you for your clear and concise video !! I would like to ask a question regarding the technique that should be used. Suppose we have 3 independent variables that are interrelated, 1 mediating variable, 1 dependent variable and various control variables. Will you please suggest me about the technique? Thank you in advance.

[@Gaskination]

The techniques shown in this playlist (kzbin.info/aero/PLnMJlbz3sefJaVv8rBL2_G85HoUko5I--) will walk you through the appropriate steps. SEM is definitely the preferred approach for mediation.

[@dipankarbiswas3834]

Thank you for your smart presentation. It helps me a lot..

[@ALI_B]

This is brilliant Dr. I need to develop a model on my subject (marketing) : measure the image of a city in order to test if there is a coherent brand image among city users. The constructs are : familiarity with the city, knowledge about the city, appreciation, evaluation. How do you see these test the image of th city?

[@Gaskination]

Ali Bezzaa I suppose if you are also collecting data on their impression about the city brand, then you could just do a simple regression from these variables to the city brand.

[@bayankhalifa1543]

Dear Prof Gaskin, many thanks for videos and for answering questions. Is it accepted to have non-parametric variables as endogenous variables in a model? Thanks.

[@Gaskination]

bayan khalifa Amos assumes normally distributed variables when running the Maximum Likelihood algorithm. I'm pretty sure the Bayesian algorithm available in AMOS is intended for non-parametric variables.

[@bayankhalifa1543]

James Gaskin Thanks, Prof Gaskin. I want to ask whether I can run a model with both nominal and scale variables. is it possible to incorporate the two types of variables in the same model and test their regressions in the same way displayed in your videos. Many thanks

[@Gaskination]

bayan khalifa Yes, but only if the nominal variables are binary, or are split up into dummy variables.

[@Gaskination]

Not sure I understand the question. We covary the exogenous variables, we do not covary endogenous variables.

[@marwanabil5475]

Your videos is very simple and helps people to learn easily.you are very talent,smart and professional to educate people well.... I would like to learn from your experience on SEM and frankly, I need your help

[@Gaskination]

I am holding an SEM Boot Camp at BYU in Provo, UT from May 8th to May 10th. You are welcome to attend.

[@samrande]

Dear James Is it possible to find the relationships of variables with different scales and connect to each other in a model. For example, I have a construct with five likert type scale and one construct with a 9 type scale ( 1-9).

[@fazlihaleem6603]

Hi James , I have one categorical variables with four categories and I want it to treat it in the regression as an independent variable. I have made four dummies D1, D2, D3, D4. when included simultaneously they are all insignificant, but if i put one by one they are significant. I am a little confused about it. could you please help me here.?

[@Gaskination]

SEM is a network of interdependent effects. Each predictor explains some of the variance in the DV. However, there is some overlap among the variance they explain. Therefore, when you include multiple predictors, they "steal" explained variance away from each other. Also, I've rarely found it convenient to use categorical variables as predictors. They work much better in t-tests and ANOVAs.

[@smrdgn35]

Dear James, thank you very much for such valuable sharing of you. I've got a question, what is the problem with the model when an estimate is over one? How should we comment on it?

[@Gaskination]

If it is unstandardized, then there is no problem. If it is standardized, that means there is an estimation problem. This sometimes happens when you only have two items for a factor, or when you have wild error variance (like in the negatives). These are all fixable issues. You can email me if you need some assistance: "[e-mail]"

[@smrdgn35]

James Gaskin Dear James, I've got the issue, thanks a lot for your kind assistance. I will contact you on e-mail if any other problem occurrs.

[@julieemminger]

Have you ever heard of incorporating SEM with Failure Mode and Effects Analysis?

[@Gaskination]

No, I have never heard of that. I've never heard of failure mode.

[@julieemminger]

@@Gaskination From what I gather, which is only based on my manic coffee fueled quick review, is that it's engineering math. Ive been trying to figure how to integrate a risk or failure score as an indicator with SEM. My quest continues. Thank you for the reply and your videos. I've got your SEM course on my summer to do list.

[@christholst13]

Hallo professor Gaskin. We are currently doing an exam project for our Philosophy of Science and Methodology course, where one of the options it to perform a SEM. However we have not yet been introduced to the subject by our lecturer and we are a bit pressured on time, since we also have three exam projects. The data that we have, is from a EPSI survey made amongst the customers of the Danish real estate agent EDC. Now, we would like to hear, if the EPSI is in anyway similar to the model that you have used? And if it could potentially be applied in a similar way, since your guides to how to perform the SEM is excellent and makes it understandable, how practally to do it, not just the theory :) Kind regards, four Danish University students :)

[@Gaskination]

I'm not familiar with EPSI. If it is a survey that collects multiple items around a few constructs, then yes, it is very similar to the data I use here, and can be analyzed in roughly the same way.

[@enlitechaos]

Dear James I built a model with 7 variables.. Would you please tell me where can I find null model in AMOS? I need to know model fit values (eg. RMSEA ) of null model. I can find independ model and sature model, but no null model sincerely.

[@Gaskination]

Do you mean the default model? This is the model we usually report values for. It should be listed with the independent and saturated model. If not, then it may be because the model did not run properly. Perhaps you have a negative error variance, or the model is not positive definite, or the iteration limit was reached. Or you have zero degrees of freedom. Try resolving these first, then the default model should appear.

[@amritakodye9525]

Hi James, The values of my model are below the required level for example,GFI, AGFI, NFI, RMSEA. I have tried to improve model fit by adding covariances to the error terms but still no such improvement was made. When I created cross loadings between constructs, the values became even worse. I will request you to advise a solution Thanking you in advance

[@Gaskination]

Do an EFA first. This should reveal where the real problems are.

[@ahmedelmelegy8833]

Thanks James. Perfect, simple; yet vey informative videos. I have a question here. What is the criteria to accept or reject the mediation hypothesis. If one path is significant and the other is insignificant, do we accept or reject the hypothesis?

[@Gaskination]

If the IV--Med or Med--DV are not significant, then no mediation. If they are both significant, but there is no change in the IV--DV after adding the mediator, then probably no mediation. If the indirect (bootstrapped) effect between IV and DV is significant, then you have mediation.

[@mohammedbargoti3681]

Thank you very much, this is really good. just a question, I am trying to perform a path analysis using AMOS, however, every time i run the software , a message pops up saying that one of the observed variable is represented by an ellipse in the path diagram. I do not know what should i do fix this problem.Many Thanks

[@Gaskination]

This means you have a variable in your dataset with the exact same name as one of the ellipses in your AMOS model. This happens quite often when someone names their variables and their residuals e1, e2, etc.

[@chompm]

I would be interested in seeing how you use the control variables (Gender, Age, education). Is that exercise covered in another video session?

[@Gaskination]

Yes it is. Probably toward the end when we're dealing with mediation and moderation. Essentially, you just keep them in and treat them like independent variables. You regress them upon anything you think they may affect (usually the DVs). You covary them with any exogenous variables you think they may be correlated with. Then you report their effects in the tables with everything else, but just shade them out a bit. Only mention interesting effects from controls (if at all).

[@chompm]

James Gaskin many thanks James. Your very prompt and thoughtful reply(s) make subscribing to your channel a real pleasure. Peter

[@zohorasuliman5796]

Dear Prof, can you please give any guideline regarding the hypothesis rejection ( If 10 out of 13 hypotheses are rejected) can that candidate pass the PhD viva by considering the performance on that day? Or is there any threshold for the number of rejected hypotheses? Please give your valuable comments along with ref. if possible. Thanks

[@Gaskination]

Usually the dissertation is judged on the merit of the work, not on the outcomes. Even if all hypotheses are rejected, as long as you had a worthwhile research question that you pursued with rigor, you should be able to advance science in your field, and therefore, you should be worthy of graduating.

[@ismas.9409]

Dear Prof. Gaskin. Do we always have to include control and moderator variable on SEM? In this case, I’m using CSR and service quality as dependents, with corporate image and loyalty as the independents. How do I implement these variables using SEM on SPSS, Sir? I'd like to hear some advice from you.

[@Gaskination]

You only use control variables when you want to rule out alternative explanations for the DV. Moderators are for when you want to explain under what contexts the relationships hold up. In SPSS, you can control for something by including it in the regression as an IV. To test for moderation in SPSS, you can either do an interaction (multiply moderator by IV) or run the regressions with different subsets of the data (e.g., male/female)

[@ismas.9409]

James Gaskin Thank you sir, i was clueless to operate it on SPSS. But it sure helps me out

[@ammaraali8136]

Hello JAMES GASKIN i have four dimensions of moderator and my model is not running its giving some errors like sample size . can you guide how i can solve this problem.

[@Gaskination]

I would have to know more about the model and sample. Feel free to email me. You can find my email by googling my name.

[@AmgadBadewi]

Really, you are brilliant. Thank you very much

[@khalilrahman3559]

Hi James I have dependent and independent variables in my study plan.I have 3 dependent variables and each Independent variable has further 4 dimensions which mean a total of 12 dimensions. similarly I have selected 2 independent variables having each two dimensions means 4 dimensions. Please explain me how I manage this in the model. I read the first order and second order measurement but I am still confused that how the second order measurement of my dependent variables having 12 dimensions will be further correlated with my independent variables and its 4 dimensions.

[@Gaskination]

Here is a video which may help: kzbin.info/www/bejne/fnO0gaSga5iMbdU

[@khalilrahman3559]

Hi James.... reference to our previous contact and your help on my data, I once again need your help. I also treated my data on Chi-square test to calculate association and strength of direction between variables. I got a very high level of chi square/Pearson value like 90.435 or 70.456 or 87.342.... Please tell me did this higher chi square value will effect my results. Is this come under the recommended range? I am quite confused with this...

[@Gaskination]

I'm not sure I understand your question. Chi-square inflates with sample size and with model complexity.

[@syedmirfan5060]

Dear James its excellent

[@ArtistHridya_2023]

dear sir Morning ppts of your presentations must be available for reviewing the things

[@Gaskination]

Many of these presentations are available from the StatWiki: statwiki.kolobkreations.com/

[@wsiriw4444]

Thank you for the knowledge you share about SEM which is very useful for the research especially in social sciences. I still have some doubt that can I run SEM with second order of three constructs, AMOS program? Thank you very much.

[@Gaskination]

You can run it, but only if it is a 2nd order reflective construct. For formative ones, you need to use PLS software.

[@johnparker5882]

Dear James I have a hypothesis to run a data how much will this cost to purchase the software or rent for a year ?

[@Gaskination]

Amos and SPSS cost about $65 each if you have a student email address. You can get them at the academic super store (online). If you don't have a student email address, then I would recommend using SmartPLS for free instead.

[@sethulakshmig2999]

Sir, your classes are really helpful. May I ask one doubt? while modifying a model can we add covariance between a latent variable and error term?

[@Gaskination]

Generally speaking, it is bad practice to covary errors with other variables, even with other errors. The exception is that you can covary the errors of parallel mediators or dependent variables.

[@sethulakshmig2999]

@@Gaskination Thank you sir

[@mckennacisler01]

First of all, thank you for your clear and concise videos! I was wondering, would it make sense to use a scale item as both a moderator (split at the median into two groups) and a control (just as the scale)?

[@Gaskination]

+Mckenna Cisler If you do, do not do it at the same time, or else our model will not run. You can use it as a moderator or as a control, but not both together.

[@bayankhalifa1543]

Dear Prof, I have two data sets collected from two different samples? Could you please let me know how should I run a model containing variables from both sets? Best

[@Gaskination]

bayan khalifa You can combine the datasets if they have the same variables and are from different samples. You will want to do a Levene's homogeneity of variance test (in the ANOVA menu in SPSS) to demonstrate that they can be analyzed together as a single sample.

[@bayankhalifa1543]

James Gaskin thanks, Prof. But the datasets contain different variables and I want to investigate the relationships between variables in dataset1 and variables in dataset2. thanks.

[@Gaskination]

bayan khalifa This is only possible if the respondents are the same across the two datasets, or if they have some binding factor (like different members of the same team).

[@bayankhalifa1543]

James Gaskin Dear Prof, the respondents are different across the two datasets, but they have a binding factor, university type. I aggregated the datasets using university type as a break variable, ended up with a dataset containig two cases, public university and private university, with all the variables of the two datasets. Then, I tried to run my model using Amos, but it reported that the model cannot be run with a too small sample, only two cases. How can I handle the problem. Thank you very much.

[@Gaskination]

bayan khalifa This cannot be overcome unless you have a lower level of binding so that you can have more sample size. Or you will have to use the datasets separately.

[@lilybaeee]

Hello! Thanks for your help. I would like to ask, can true or false survey question be analysed by SEM? Or all questions must be asked in likert scale type?

[@Gaskination]

True and false are less useful because they provide very little variance. If the questions really need to be true/false, then you can create scores (like a sum or average) to represent the extent of a particular construct. That new score can be used in a path analysis or a regression.

[@adelhesham8744]

Hi Dr James Gaskin what is ranges for path coefficient values? Can path coefficient values more than 1?

[@Gaskination]

Standardized path estimates should be between -1 and +1. Unstandardized estimates can be unbound. They might be in the millions.

[@adelhesham8744]

Thank you dr James Gaskin

[@ilmamufidah6272]

Could you tell me what is the function of covariance there in the model?

[@Gaskination]

AMOS is a covariance-based SEM program, so it assumes all exogenous variables are covaried.

[@ilmamufidah6272]

James Gaskin Thank you very much of your answers. I have more further questions. What is the meaning if all latent variables are covaried with each other in amos? what is the hypothesis? do we always need hypothesis in every CFA we conduct? then, if I draw my model in which all latent variables are covaried, what should be my hypothesis in CFA? Do we always make something like null hypothesis and alternative hypothesis too in CFA? if we do, then what is the null hypothesis and alternative hypothesis, if I have a model in which all latent variables are covaried each other, and the covariance is the one and only arrows that connect all latent variable. I have no correlation arrow in my model. All latent variables are connected with covariance, just like in your model, but I got confuse with the hypothesis.

[@bibinful]

how to do sem in amos when there is 2 dependent variable and 10 independent variables ?

[@Gaskination]

same way as the approach shown in this playlist.

[@asifkhaliq9086]

Thank you very much for your explanation . . .

[@BMM351]

Thank you!!!!

[@waleednadeem377]

whats your email i want to show u my theoretical framework

[@Gaskination]

+waleed nadeem [e-mail], but I mostly offer statistical advice, not theoretical advice.

[@mohammedbargoti3681]

Thanks a lot .Problem solved