What do you mean by SEM?

welcome everyone in the last session we discussed about structural equation modelling we're just briefed about structural equation modelling and what is the role of structural equation modelling right so in this case when you talk about structural equation modelling which is nowadays highly used in all kind of researches structural equation modelling is build up built upon basically three kind of statistical techniques you can say right

basically the three are regression or if I say it is multiple regression right so multiple regression factor analysis right factor analysis and the third I say is the chi-square right the three well the three statistical measures that are highly used in structural equation modelling are these three right I squared so the chi-square talks about the goodness of fit right so goodness of fit index so now it tells about the whether to the obs

erved and the expected model the object model and the expected model are they more or less same or different and if they are same then it is a good thing right if their clothes are same that is good but if there is a lot of difference between the observed and expected then it is not good okay so what in the in the first part we talked about the measurement model right so we said that the entire structural equation modelling can be said as

a combination of the measurement model and then the structural model structural model okay so so the measurement model is what we say is a confirmatory factor analysis right so apart if you if you can go understand the word so there are exploratory factor analysis and other was confirmatory factor analysis okay so this part confirmatory factor analysis part is a measurement model okay so confirmatory factor analysis is used as a measure

to reduce the measurement error of a study right that means what when you have measured certain things and that those things could have been measured and you know the researcher with the the the respondent might not have taken seriousness serious interest or something and something has happened right so in a cfa case what happens is here it tries to take build a latent construct basically a latent construct which is made up of certain va

riables right so this variables together will represent the the construct let's say okay so this construct in the measurement models we say what we do is we basically create a covariance matrix or a covariance structure right in a covariance structure a covariance structure if you remember we have also discussed that it is like a correlation structure only but a correlation is a special case of covariance right so here we try to see whet

her the model there is a covariant relationship covalence relationship between the constructs right and finally what we do is we try to see in the CFA in the CFA case we first check the construct and for its reliability and validity okay so you checked for validity through different the construct you know we have the construct validity so so to do the construct validity we use certain ways for example we convergent validity discriminant

validity right and nomological validity right so convergent validity I am sure you are very clear because in convergent validity only thing you want to see is that whether the items are converging towards the main construct okay so if the items are sufficiently having a larger you write the correlation between the item and the construct is high or interitum correlation you can say is high then we say this is a high reliability so that he

lps us also to understand the convergent validity there is a convergent convergence but discriminability occurs when we try to measure key whether two constructs are sufficiently separate from each other okay so construct a and B should be sufficiently different from each other so to do that what we use is the average variance extraction right value and compare it with the let us say the correlation right correlation and we always say th

at the average variance extraction extraction should be higher than the squared correlation value right the squared correlation value whatever comes it should be the a ve should be higher than that then there shows a discriminant validity okay but let us say now comes a case once we have done the measurement we have it's a measurement model so we have measured and we found that the construct is justified and it is explaining what it shou

ld explain okay it is doing its job with the help of the model fit indices we have said now the model fitting business are basically we use the Chi square by Chi square by degree of freedom which we always say should be as less as possible so in between let us say anything less than 3 is good we say right we have taken a magic value of 3 right but some people some researchers also say anything up to 5 is still good right so what does it m

ean the Chi square by degree of freedom basically talks about that it compares against the baseline model that means there is a model there is a model and it compares give what happened what is the saying is the observed and the expected model the absurd model in the expected model the difference should not be very high right so when we are when we are measuring the Chi square by degree of freedom and if it is sufficiently threes around l

ess than 3 then we say that the model is a fit model okay there are other fit indices also for example the goodness effect with GFI IFI incremental fit in this right and you have CF I write confirmatory fit indices so these are some of the absolute fit indices which are used and you have a magical number we say of 0.9 so 0.9 so if it is if these values are above 0.9 and you can you might not take all the values you might take only one of

from the absolute fit indices and one from the incremental fit indices right so if you take these values and see that this is above 0.9 then along with the chi-square by degree of freedom and these we say that the model is a fit model right and we also find something called the RMS CA this is also an important thing which takes the root mean square error basically and RMR right so these this is the residual the root mean of the residual

s right so residuals are the unexplained so unexplained part the residual the square of the unexplained right basically it takes right and the RMS yield now if these two values are sufficiently good below 0.8 we say you know for our MSA is 0.08 8 percent below 8 percent less than 8 percent then it is a fit model so after doing all this we have understood that there is a the measurement model the measurement model has been used to check f

or the constructs validity and reliability but now conscious situation where we have to check whether there is a relationship between the constructs or not or what kind of a relationship exists between the constructs rates so to do that we use the path analysis with the path analysis right so the path analysis basically is the structural part only right so what is happening here we are saying model a construct a is effective let's say co

nstruct B this is a simple case like a regression only but suppose now we are adding a third variable let's see the third construct now we are saying that if if in this case it is becoming more than a simple multiple regression in a multiple regression case you had this was the illicit dependent variable this is the independent variable but now in this case this has change its characteristic and it has become a new independent variable a

nd this has become the final dependent variable so in such a condition which is actually what it is happening in real life right so one becomes the father of a house is depend the family's depend on the father but the father is may be dependent on his on the office or the you know is office ambience of his the climate or is let us say the boss right or whatever ever he is working so if he is tomorrow let's say suspended from his job then

his family will get affected or if he is given a promotion then his family will be productive positively affected so let's say if I take it although it is a very crude way of understanding this is the boss or the employer let us say the boss is a bad word the employer the the father and let's say the children so the children are dependent on the father the father is dependent on the employer right so this is the case where you want to es

tablish such kind of relationships to do that we use structural equation modelling so it's a process for stress testing a structural theory right a structural theory is a conceptual representation of the hypothesis the hypothesis relationships between the constructs now this relationship that we have built there hypothesized relationships we are yet to prove it right so how do we prove it now we'll see we'll see so how do we prove it so

after you know the mechanism becomes like exactly like a multiple equation only okay so it can be expressed in terms of a construct more a structural model that represents the theory because set of structural equations so everywhere there is a new equation it is a case like a simultaneous equation right so you are creating different equations right from here to here here to here so this equation that you are building now you are testing t

hem with a represent the structural model repels the theory with a set of such equations and is usually depicted with a visual diagram the visual diagram is this okay so what are we focusing here we are focusing on the overall and the relative model fit right and the size direction and significance of the structural parameter estimates now the estimates the word estimates if you these are this is some simple the regression estimates ther

e are the regression coefficients right now with the help of the estimates we can say whether a hypothesis is to be accepted or rejected the estimates will help us to say whether a hypothesis is to be accepted or rejected now this is a structural model but one thing before I get into the structural model please remember that the structural model the relationship that you have built in this case was the C right it should always emerge fro

m an existing theory it is not that we liked something like you know the is affecting C we will say ah our C is getting affected by a now that is something like out of bases out of reason right so just pulling an arrow to make some you know to just prove your point or prove somebody's point is not the right thing to do your model should be built upon very clear structure theoretical premise okay so there are two constructs here customer s

hare and customer commitment customer share has got four variables customer commitment has again got four variables right this lambda if you see the lambda these are the factor loadings as you saw a factor loadings in factor analysis expletive factor analysis similarly they're also the factor loadings okay so my special theory hypothesis is that customer commitment results from customer share the customer share right so the arrow moves f

rom that means the arrow is moving from customers share to customer commitments because thermo commitment is a dependent variable right so structural equation modeling there are five six stages the one two three four here if you see defining the construct of developing the overall measurement model designing the study to produce empirical results assessing the measurement model validity is a part of the measurement model of the CFA right

the five and six specifying the structural model and then the assessing the structural model validity is the five and six are the part of the structural equation right so the structural model so remember when you have let's say after you have done the measurement model right so whatever you're the Fitness in descended right the structural model should be better than the measurement model the structural model should always be better or h

ave a richer value or higher value than the measurement model if it is not happening then there is not much of a significant improvement right so we need to understand that the measurement model was only measuring the constructs but structural model you are putting in some relationships right so the structural model has to be more powerful and you know robust then the measurement model okay okay so how do you convert a measurement model i

nto a structural model so now you see the a this is a measurement model so ABCD are these are all there's a covariance relationships right so there is a double-headed arrow to form from one construct to the other right so this double-headed arrow is basically nothing but still talks about the covariance relationship right but here if you look at the structural model it is very clear in saying that a and B FX c c then FX d okay now these t

hings that you see that for example this this symbols right these are called the error terms so if you these are the sometimes we use it simple we use e 1 e 2 right kind of error terms now what is the error terms time and again I've been explaining the error terms are nothing but the unexplained part right so in this case there are 4 items or variables related with each Latin constant and 4 4 4 4 and there are 2 error terms here right be

cause every dependent variable will have a error term right the depend obviously the independent variables do not have an inner term right so why then this is also an independent variable then why it is having additive now because it is also independent variable for these okay so that is why now I'm not getting into the recursive and non recursive model but just to if you want to understand because a recursive model is something where th

ere is a if you look back this is a non recursive model and the recursive model is only the only difference is that when there is a one-sided when the arrow moves from one direction towards one direction only we say it is a recursive model on the other hand when there is a dyadic relationship or a double the path is moving both sides right so in that case this is called a non recursive SEO model right so now let us look at this case this

is a it's back from the book of the thumb and black had an understand so from there we have brought so these are some of the constructs and their relationships so these are the endogenous variable this is the endogenous variable these two are the exogenous variables exogenous means independent and endogenous means at the dependent okay now hypothesis is there are several hypothesis if you can see the h1 is saying EP now what is EB if we

want to see then I think it is B it will be here EP is environmental perception AC is attitude towards coworkers yes I is staying in tension OC is organizational commitment and Jas is job satisfaction okay so environmental and what this co-workers I think attitude towards the coworkers are affecting the job satisfaction and the organization climate finally affecting the staying in tension so the hypothesis is that environmental effects

jjs job satisfaction elemental also positively affects OC AC effects AC effects Jas AC effects OC right and then Jessie J s sorry effects si and OCFS si so one two two here four five six seven right and Jase you know see there must be one so if you see job satisfaction leads to organizational commitment this is one more right kind of a relationship so this is a very complex relationship right so so I'm not getting into back into CFA now

so yes so so let's look at this right so if you are having such a condition now you have established relationship now what should you do after you have got the structural model intact then what is what you need to do is you need to find out the first the model fit right so as you did in the measurement model similarly you need to find the model fit for this case also and I said that the model fit should be better than the measurement mod

el right so once you have the measurement model then what you do is after the you check for the model fit suppose it is the model is fit then you go to the next thing that is to check for the fitness industry and assess also you have suppose your check and they are all coming point 9 and above so that means the model is now clear it is a fit fit model right there is no problem with it but in case in case there is a problem of let's say t

here's a poor model fitness is showing then a most has a facility to improve the you know to improve the mo see the software which you use for little and a most a most being a very popular software they are to improve the model you can do one thing you can use the modification indices so modification indices indices are one such value which is used to improve the model fit now if you have the modification indices then this indices what i

t does is basically try to find out which are the error terms which have got lot of you know there's a which there's a lot of unexplained part into it so that those two errors are combined so that the unnecessary are as unexplained part will get reduced and a model will improve better right so so uh okay so once you've done this then what happens is let's go back to the hypothesis now what do you do is the you need to go for the estimates

right the estimates so the estimates we'll assume in a most at least helps you in providing this estimates this estimates as I told you as nothing but the coefficient regression coefficients for each value now it will also give you with it the estimates as well as its significance right so when it gives you the significance right from the significance if suppose something is significant right as the significance is told shown in triple

star or double star look depending on point zero one or point zero five or point zero zero one right kind of a significance level so when when you are going through that estimates you need to check that relationship suppose J s ep2 Jas now if that relationship is it will give you an you know significant or non significant relationship it will explain that so if there if your significant relationships significant then we say that yes Jas

is really getting affected by EP and it is not due to chance that it has happened only this time right similarly suppose out of it suppose you found only one of them that's a Jas and OC j s job satisfaction leading to organizational climate leading to ordinal climate this is not coming significant that means what in this model we have theoretically assumed that job satisfaction also needs to organize the climate and thus it affects stayin

g in tension right so this is not coming true that means this hypothesis is a false hypothesis we cannot we cannot say that your J is actually effect so C and then that effects si right so this is one thing that is that you are given right so after the end after the end you will get all the estimates and with their significant no value significance values and through that you can accept or reject your hypothesis right so in this case the

re are seven hypotheses and we said tentatively let's say only one of the hypotheses is not accepted or it is rejected let's say so it is rejected so we say in this study there are six seven hypotheses out of which the first six hypothesis we're significant and they were they it showed a positive effect as the relationship you have to explain through the relationship but only one of them did not show a significant relationship that means

it shows that there is no effect okay okay so when you do this when you do this you need to understand you need to understand that there is something also for example let's say there are two constructs let's say two constructs this is a third construct let's say this is a fourth construct let's say right now there is something we need to understand ABCD for example okay okay so now if you see in this case a is affecting D also through C

okay B is also affecting B through C and as well as directly in this case if you shall look at it right so B has a direct effect also and B has an indirect effect also so let us say these two are point the factor loadings are point five and point five right and this is let us say point four right and this is let us say point three these are the factor loadings I'm giving now what will you understand from here there are two types of effec

ts one called the direct effect and the other called the indirect effect now in this case you have to understand that the indirect effect from A to D through C an indirect effect means when the the variable passes through intervening variable that means a mediating variable a intervening you can understand a mediating variable right which is behaving like a mediator in between right so 0.5 0.5 in to 0.4 is equal to 0.2 so the indirect eff

ect is 0.24 a to do a 2 D and similarly from b to it is also it's a this could have been point food also so 0.4 into let's say point four so 0.16 let's say so this is from and what is the direct effort from B to D now point three so here we say that the direct effect from B to D is higher than the indirect effect and in this case but the total indirect effect the total indirect effect is 0.2 plus 0.16 okay so many a times it happens it

is of interest to a researcher to know give whether if we if mediating variable comes into the picture whether it is advantageous or it is it is not required sometimes it's like a catalyst we have to understand like in a catalyst the presence of a catalyst will change the reaction right so direct effect might be let's say the direct effect might be significant might not be significant right so if it is whatever and the value could be les

s or more whatever but suppose if it goes through a mediating effect the mediator then suppose we see that it is significantly improving the the value is improving then we say that there is a presence of a mediation effect right so let us see what is this mediation as you can see here right so k goes through to e through m m being the mediator right so M mediates the relationship between K and E if the diet effect of K on E is diminished

when M is also a predictor of E so we will if suppose you have a direct effect nobody wants a mediator until unless the direct effect is not so good right so if your mediator only adds to the explanation or adds to the strength of the relationship then only the mediated is a justification otherwise there is no justification for mid-age mediation okay now adding direct effects so how do you check for mediation although this will not expl

ain much I will do it directly here let's say there are two types of mediation so one is a partial mediation and we said the other is a full mediation that means we will Hey the mediator has a partial effect or the mediator has a full effect now what is this full and partial let us say a M let us say B okay so this is something like this so when a has a significant effect on the mediator and M has a significant effect on the dependent va

riable right then and suppose a and B also this is a barren and Kanis method is a very although some people say is over the old method but still the it is a very basic method and the very clear method suppose the presence of a mediator and the absence of a mediator if there is a presence of the mediator and the absence of a redditor if the significance level changes let's say the direct effect the direct effect let's say when you bring in

a mediator let said the relationship becomes strongly significant right and the diet effect shows not so or trust strong then we say it is a case of a full mediation right that means when suppose a to be was easily now was earlier is significant now also is not significant let us say and through this this is becoming significant right then we say there's a powerful mediation suppose both are significant but this has got a lesser value t

hen this then we'll say it's a partial mediation right so these are the ways of just justifying or you know identifying okay this is ridden so mediation involves the comparison of a direct effect with the two constructs while also including an indirect way through the third conserve full mediation is found when the direct effect becomes non significant as I said so when it is non significant and in the presence of the indirect effect wher

eas full meditation is when it is non significant whereas partial mediation occurs when the direct effect is be reduced but still significant that means what if I go directly also I will get some money for example but if I go through somebody I will get more money that is a partial mediation but suppose if I go directly I will get no money but if I go through somebody I will get money then it is a full mediation as good as that right you

have to understand so so well it says some of the things okay there is one more important thing this is called a after a mediation weeks is there is something called a moderation also now what is a moderator a moderator this is always a you know something like there are some things which are used as a moderator now some things automatic moderate the relationship okay now what is the moderating relationship now for example many a cases g

enerally what is used is gender gender we say is a moderating variable now what do you mean by moderating but that means whatever the relationship you studied in this case for example or the seven hypothesis that might behave differently for male and differently for female okay so if you have a few exercises the same thing for male and you exercise the same relationships and test the relationships for female that means it is different so

if you take only gender is one that means there is you are not able to get into deep and you are not explaining it well so if you do a moderation test a moderation test then we can we can easily explain that whether gender also moderate is gender is moderating the relationship between let us say let's say I am saying here I am in putting in a moderator okay moderator now I am saying the relationship between a and B is moderated this ther

e is a mediator also but this is a mediating and moderating effect both are there right so this if the moderator is affecting the relationship right is affecting the relationship so how do you check for moderation it is very simple there's nothing very complicated in case you have continuous variables also you can use a moderator because generally there is a confusion between researchers that moderator should be a categorical variable or

not right if you have a continuous variable also what you can do is you can use some logic and create some categories in it for example let's say income from 0 to or let's say any 0 to 10 lakhs is let's say 1:10 to let's say 20 lakhs is to 22 let's say above is 3 now why doing this what I have done is I have categorized the variables right so moderation can be tested with multi group SDM right so what you do is simple in a moderation te

st you first define the groups and then you run it by taking each part of the each group individually so by taking it will have two values will generate one suppose in this case the all the seven hypothesis that seven you know Jas and all these things so if you have for male you have will have a particulars the seven hypothesis the relationships right seven hyper similarly for female of so you will get it right now then how do you know ki

that moderator what should you do to look at the effect of the moderation the best is to see is there any variable or any relationship that is significant in male and baby C not significant in female or vice versa only pick up those those relationships which are significant in both the cases the ones which are not significant in though in either one of them they are items for deletion okay so by doing this basically what you are doing i

s you are you are using a moderation technique hi this is another way also which I was saying the bod resting variable can be collapsed into groups which I explained or cluster analysis also can be used to identify the groups right that is also by the hierarchical clustering you can identify the groups also that is one more method so whatever it is so you understood that structural model helps in identifying the relationships how they go

and there is if there is an immediate ringing effect we can check and finally if there is a moderator how does the moderator affecting the relationship that also can be checked right so this is all we have for SVM now thank you

## 0 Comments