SEM with AMOS: From Zero to Hero (19: Construct reliability and validity assessment)

so we already have tested the model fit and now we want to access the reliability and validity of the measurements in the previous videos i show i told you what we mean by reliability and validity but you can watch the videos but just as a recap reliability means how how consistent your measurement is right and this is the classic example for reliability and validity and validity means whether you are using the right measurement to measure something so we want measurements that are reliable and valid so what we test in sem is construct reliability and construct validity right so constructor levity there are some measures later i will show you and for construct validity there are two main validities that we assess convergent validity and discriminant validity these two are very important in sem so when we say construct validity we mean convergent and discriminability in scm there are other types of validities you can you validate you can test as well but you can read yourself more about them but these two are compulsory let's say to report in your um research results and for example if you want to test you can test normological validity means the relationship between the construct or you can access face validity content validity there are many types of validities you can test as well but these two are the most for sem so what do we mean by convergent validities it means the items that are indicators of a specific construct should converge or share a high portion proportion of variance in common um let me tell you what i mean let's go back to our model this is the model that we assess this model feed and now we want to test the reliability and validity of the measurements so convergent validity means the items of each construct should converge and let's say they should be consistent they should be consistent right so we expect high factor loadings um to have convergent validity it means the factor these items should be let's say in the same group and this is called convergent validity for this criminal discriminant validity we want to make sure that the items of each construct of course you know all these items that you can see here they have some correlations but most probably many of them not significant but not most probably that they may be non-significant or small weak but we know that our correlation between all items you see here all construct everything right but by this current discriminant validity we mean uh we want to make sure the items in each con construct have been discriminated from other items on of the from the items in other constructs so image in other words we do not want image one loaded on subjective norms as well of course it will load but not strongly so this means discriminant validity so for discriminant validity we want to make sure they are separate image 1 to image 6 are separate from subjective norm 1 to subjective norm 5. this is the meaning of discriminant validity they should be discriminated right so convergent validity we want to make sure the five items to measure subjective norms they fall into the same group and they are consistent and they converge and discriminate validity it means we want to make sure subjective norms 1 to subjective norms 5 they have been discriminated from all their items in the model all these items in the model means they are separate so they should weakly load on other constructs so subjective norms should have should subjected norms one to subjective nodes five should weakly load on others and strongly load on subjective norms and this is the meaning of construct validity right so how to test convergent and discriminant validity there are different measures you may use for construct reliability we usually report composite reliability and maximum maximal reliability and omega is something that recently i have noticed some people including myself will report for example chromebook alpha falls in this group but the assumption from cork for cornball alpha is there is no measurement error but the whole idea of scm is there is measurement error right so uh i usually report columba alpha as well but our main concern what we need to report is composite reliability and maximal reliability if you are interested there are other measures as well like omega and many more you can report them as well the threshold for all these construct reliability measures is 0.7 so as long as your construct reliability measures are greater than zero point server this means you have uh i mean the construct reliability is established what about convergent validity for convergent validity there is a something called av average variance extracted um it's very actually easy to compute you can google to know more about it uh i mean manually you we used to compute this manually but now it's a plug-in you just click so ave should be greater than 0.5 this is a threshold so as long as av average variance extracted so we compute the average variance extracted for each construct and if it is greater than 0.5 means we have convergent validity for the construct and for discriminant validity there are different methods as well one of them is htmt matrix so there is a matrix no need to do it in no need to compute any of these manually the software will give all to you so htmt should be all values should be less than 0.9 and there is another method which is the let's say the classic method it's called fornet larker the researchers who developed this 1981 um in this method i it will construct there you will have a table and the idea of the table is that average average variance extracted that we discussed here should be greater than maximum shared square variance actually not maximum share score maximum share variance anyway the maximum shared variance um between the constructs so now um i want to show you how to compute so as i said we used to do all these manually not um all provided by actually none of them were provided by aim software directly but now we have some plugins i borrowed this from james and what we do i provided them on the website as well so what you do is you just install the plugin watch the video that i have shown you how to install the plugins and then what you need to do is very easy just click on plugins and validity and reliability test master validity plugin so there is master validity plugin just click and done we got the results finished very easy right and congratulations here there is no issue how okay let's check the results cr means composite reliability it should be greater than 0.7 and you can see in this column all are greater than 0.7 so then maximal reliability all greater than 0.7 good so based on these two measures your construct reliability is established then ave is used for convergent validity av is greater than 0.5 in all of them that is very good so convergent validity is established and then um for discriminant validity this is the former marker table this part i mean from from subjective norms to the end from this column to the end so what does it show you see there are some bold values here these are the squared root of av i can i can prove this to you square root of [Music] 0.610781 781 is the same or this one zero point five six one four seven four eight so almost if you round it up it will be seven four so these values these bold values these are uh square root of ave and what else you see the other values in this table are just a correlation between the constructs for example 394 is the correlation between subjective norms and image and then image with job relevance is 544 and so on right and this asterisk shows that the correlation is significant statistically significant at 0.01 if there are three has three okay now the criteria for this current validity is discriminant validity based on this method is these values should be greater than all values in the same column and same row guys we don't look at these four the first four columns there are different things so all starts from this to the end right so this table and here 781 is greater than all of them in this column image 749 is greater than all here and this is and 749 should be greater than this one as well the same row and same column so 774 is greater than all here and here the same for all of them so we do not have any discriminant validity issue right what is the meaning of this table this means that ave should be ave which shows the how uh how uh the the proportion of each av is an indicator of the convergent validity means how com how items have been converged together are converged together uh how items are converged uh in each construct so the average of their values yeah should be so this basically means that have a the average variance extracted for each construct means of these values should be greater than the correlation between a construct and other constructs right so this means they should be discriminated we do not want these items to have correlation with other items or other constructs in the model so this is the idea of this this table there is another table that we discussed is hdmt this is something new compared with the previous method the formular care method and here you just it's a table you just check to make sure there is no value greater than 0.9 if there is no this means discriminant validity is established so if there is any issue it will be highlighted here and the color will change to red hopefully actually luckily we don't have any issue but what if we had any issue if construct reliability had uh was not i mean the value was less than 0.7 what you can do is you need to go back to your model and for example suppose image has construct reliability issue then you drop you exclude the weakest item and here the weakest item is for images 0.59 so you remove these items 0.59 is related to image four so you just remove image four right you remove it and then run model again the model again and assess the reliability and validity if not good again you have to remove one more item so you need to remove the weakest items in the construct if constructor levity for a construct had a problem what about av if av was not reached to 0.5 you do the same drop the weakest one for this respective construct what about this criminal validity if there was this criminal validity if it's about correlation between the constructs so if for example there was an issue of this criminal validity between let's say image and job relevance image and job relevance you just run efa exploratory factor analysis in spss here you cannot run exploratory factorances you have to go back to spss run efa only on the items of image and job relevance and then find the one that makes problem which one the one that has cross loading the one that has cross loading the one that for example you run efa and then you find out that oh image 4 is loaded on image at the same time it's um somehow strongly has loaded on job relevance too then you drop you exclude image four so you just take the items for these two constructs if they have construct discriminant validity issue you take them to efa run efa only on these items the two constructs items and then find which item is loaded somehow on both of them then drop it right then your discriminability issue should improve discriminability should improve and yeah [Music] so this is the way to assess reliability and validity but just to share with you sometimes av does not reach to 0.5 and you do your best but it's still not good but we don't want to exclude we don't want to remove many items right or some cases you have only two or three items in your model and we don't want to have any construct with less than two items never ever no construct with less than two items or so in this case you cannot remove more items so i tell you something sometimes ave 0.4 or something you cannot reach to 0.5 that is fine you can refer to i have a paper i have included in the references here you can cite that paper or you can cite my book so av is too a script measure for convergent validity in those cases you act knowledge that your av did not reach to 0.5 and this happens sometimes especially when the measurement is new or maybe in some um constructs that uh you know if you even check the literature you see the av is not that high for those constructs this may happen so if this happened then you can cite my book or my paper and justify it so you may mention that yeah we understand this is i'm not didn't reach to 0.5 however av is too strict measure for conversion validity so how to access convergent validity as long as construct reliability is greater than ave and construct reliability is greater than 0.7 you would accept it and another thing just a secret we do sometimes construct reliability does not reach to 0.7 there are some references that you they say even 0.6 is accepted right so if you had no chance okay you made you can cite my book or there are other books like i think andy fields book they say 0.6 is ok as well for construct reliability measures anyway hope you don't face any issue in real reliability and validity however if you fail these are some of the methods you can use to address the issue

Channel: saeed sharif

Views: 1,270

Rating: 5 out of 5

Keywords:

Id: bACJxpFwdNQ

Length: 17min 11sec (1031 seconds)

Published: Wed Mar 17 2021