How to Solve Convergent and Discriminant Validity Issues in SmartPLS

How to fix problems in Convergent and Discriminant Validity. Learn to Solve the common issues faced in establishing convergent and discriminant validity.
For Notes, please visit
researchwithfawad.com/index.p...
#DiscriminantValidity #ConvergentValidity #Validity #PLSSEM #SmartPLS #Research #Thesis #PhD #Dissertation #PaperWriting #AcademicWriting

[@DZ-fy9wi]

Thanks Fawad! I appreciate your sharing.

[@researchwithfawad]

My pleasure

[@abebenegeri1823]

Dear Dr. Fawad, thank you very much for the video. It is so really informative. Keep it up, sir!

[@researchwithfawad]

It's my pleasure

[@ehsansaeedi9967]

Thank you so much for the videos it helps a lot, please keep doing this. I am from a poor family background and cannot afford expensive research classes... thanks

[@researchwithfawad]

Pleasure. I am glad you liked it.

[@fenglin7864]

Thank you so much, Professor. Your kind teaching saved me once again.

[@researchwithfawad]

Pleasure. I am glad you liked it

[@fenglin7864]

@@researchwithfawad Professor, may I please consult you about the reference? For the solution, "if an item is cross-loading, and the difference is less than .10, REMOVE the item(s)", which paper should I refer to? I checked the paper (Farrell, 2010) you suggested in the video, and it did not mention the specific difference "0.1". Thank you.

[@shriker025]

Saved my data. Thank you.

[@researchwithfawad]

You're welcome!

[@Alhamzah_F_Abbas]

Thank you for this amazing explination

[@researchwithfawad]

Glad it was helpful!

[@AcharyaBhatt]

You are doing a wonderful job Professor. I happened to come across your channel as I was struggling with some basic concepts related to SmartPLS. Your approach and speed of delivery are both highly commendable. Please make a video on interpreting HTMT scores as well.

[@researchwithfawad]

Thanks. I am glad you liked it. For Discriminant validity please watch kzbin.info/www/bejne/l3-lfmVugN96aZY

[@halfsecond1056]

Thank you for the video Dr. Fawad. It is really informative. Please provide the link to the paper.

[@researchwithfawad]

Thanks. I am glad you liked it. Here is the paper onlinelibrary.wiley.com/doi/abs/10.1002/csr.1911

[@omerashahnawaz6933]

Thank you for the video. This video is very helpful. Can we report discriminant validity by only reporting Fornel and Larcker Criteria and not cross-loadings and HTMT?

[@researchwithfawad]

Thanks. I am glad you liked it. Actually you can, but sometimes the reviewers may ask to report HTMT as well.

[@omerashahnawaz6933]

@@researchwithfawad thank you for your reply

[@aguamarachi5832]

Thank you so much Fawad. I have a question please, My AVE is all red (less that 0.05) I have removed all the items with low factor loading but AVE is still bad. CR is okay..Can I report like that in my thesis?

[@researchwithfawad]

If it is over 0.40 and CR is over 0.60, you can report it.

[@priyankavern377]

Dear Sir, First of all, thank you so much for this video. I followed your instructions and my issues is resolved as well. But, i am unable to find any other relevant article to cite than farrell (2010). Could you please share any other relative articles as well or studies who have adopted this approach to solve the discriminant validity issue. It would be a great help. Thank you & Regards.

[@researchwithfawad]

For more you may also read the book by Hair et al, A Primer on Partial Least Square Structural Equation Modelling.

[@nishakatayan6646]

Sir all my factor loading are between the range .05 to .08 but still the AVE of some of the constructs are below .05. There are two construct who's AVE is .038 and .036 respectively. What should be done in this case?

[@researchwithfawad]

You may have to remove some items with low loadings.

[@jiefang1598]

Dear Fawad Latif, thanks so much for your selfishless help. I will definitely follow your instruction to work on my discriminant validity issuses. I have two more questions here. You mentioned that we should compare the cross loadings to see the difference. If the difference is above .1, we should consider to keep the corresponding items. Could you please tell me the relative articles for me to cite? Besides, you also mentioned that if the HTMT criterion was not satisfied, we could cautionsly continue when the cross loading and Fornell-Larcker criteria were fine. Do you have this source as well? I failed to find the sources of these statements. The article you mentioned in your video, I have read if before, which is very helpful, but no similar statements like those two I mentioned above. If you have more recommendations, please make a list. I will read them thoroughly. Once again, thanks so much and best regards.

[@researchwithfawad]

Hi. Thanks. I hope my comments have helped you. As for citation for discriminant validity, please refer to Insufficient discriminant validity: A comment on Bove, Pervan, Beatty, and Shiu (2009) Dr Andrew M. Farrell, Loughborough University As for using Fornell and Larcker instead of HTMT, these are different methods, you can put this text in your paper and i hope you will be fine.

[@jiefang1598]

@@researchwithfawad I already cited this paper and double checked it, cannot find the sentences like "if the items are cross-loaded, and the difference is less than.10, the items should be deleted" and "as long as the Fornell-Lacker and cross-loading criteria are satisfied, even if the HTMT criterion is not perfect, the discriminant validity could be cautiously accepted". If it is not too troublesome, could you please check the sources?

[@jiefang1598]

@Andre de Santis Candro unfortunately, no. I have three mediators in one model. Since one of them highly correlates to the other two mediators, I separated the original model into two models (one model has one mediator, the other one has two mediators). This solution could refer to Farrel (2010), which proposed to drop one or more constructs with insufficient discriminant validity. Farrel,(2010). Insufficient discriminant validity: a comment on Bove, Pervan, Beatty, and Shiu(2009)

[@jiefang1598]

@Andre de Santis Candro Thanks so much for your recommendation. Although I have finished the writing of my dissertation, I will read the paper you mentioned.

[@tsanin4539]

@Andre de Santis Candro Hi Sir thanks for sharing this reference, can you give detail, which paragraph that mention the procedure to remove varibles to correct discriminant validity?

[@user-bs7dt1rj9p]

Dear Professor, Is it necessary to adopt a reverse coding of questionnaire’s items in Excel for negatives ones (e.g. Strongly Disagree - 5….Strongly Agree - 1)? Thank you very much!

[@researchwithfawad]

Yes, if the item is negative in comparison to others in the scale.

[@fouziabintifarooq9341]

AssalamuAlaikum sir Sir, I am doing pilot study right now and one of my variable "personality traits" have got low loadings for some of the items and when I remove those items my Ave increases. Sir should I remove those items in this stage where my sample size is just 100 or should I take them for the actual study too because may be low loadings are the reason of inadequate sample size.

[@researchwithfawad]

No need to assess AVE for pilot study. Assess construct validity for final study. For now just assess reliability.

[@Shravan1231000]

I have got HTMT value as 1.025 which loes is between the confidence intervals (0.652, 1.059). So can I consider it to be ok..?

[@researchwithfawad]

This is still quite high. Please do check your cross loading. Also watch this video kzbin.info/www/bejne/gWGyqmCJqZxqgpY

[@joelkeats7192]

Dr.. I am facing issues with high HTMT between two constructs. There isn't much I can do because these constructs have only a few items so I cannot delete any of the items to improve HTMT. However, they are sub-dimensions of a single construct so maybe I can merge them as you suggested. Can you pls explain how I can merge them? Do you have any videos on how to do that? Thank you.

[@researchwithfawad]

Hi. Merging would mean loading them into a single latent variable. But before that, watch this video kzbin.info/www/bejne/o2fXhp99nq1sla8

[@Hellz706]

Asalam alaikum sir …. 1) please tell me how to calculate critical ratios through smartpls ? 2) Why do my all values in second order analysis are equal to 1 ….including cronbach alpha, AVE discriminant validity ??? Sir you have helped us a lot may allah bless you

[@researchwithfawad]

WAS. The t value is the critical ratio. Mean/SE. As for the Higher Order. Did you take LV Score at the first Level.

[@Hellz706]

@@researchwithfawad thank you sir for responding…yes sir i did take the lv scores for the first order

[@rongtengzhang1170]

Thanks for sharing Dr. May I ask what is the recommended value that KOL5 exceeds at 13:10? And will there be any video about how to solve the multicollinearity issues? TQVM

[@researchwithfawad]

It is loading well onto the other factor as its loading is over the generally recommended loading of .70. Although, no video yet but here are few solution i found - Check the correlations between the LVs - You can Group the independent variables in one second order LV, and further use second order LV as the predictor.

[@rongtengzhang1170]

@@researchwithfawad Understood Dr, thanks again

[@madinanarysheva1455]

Assalamu alaikum, Dr Fawad! I want to say Thank you so much for your tutorial. In order to solve Convergent and Discriminant Validity I've deleted 8 items. Initially, in Heterotrait-monotrait (HTMT) - Matrix there were 7 values greater than 0.9, and deleting these 8 items improved the Heterotrait-monotrait (HTMT) - Matrix and Fornell-Larcker criterion, in the first table only one value is more than 0.9, in the second table only one value is not greater than all the values underneath. Do you think discriminant validity in this case can be established? Thank you in advance!

[@researchwithfawad]

WAS. Thanks for watching. You can say to a certain extent it is. If the two constructs are sub-dimensions to a higher order, why dont you merge them.

[@madinanarysheva1455]

Thank you so much for your reply @@researchwithfawad

[@artimalik1126]

Good morning Sir, any video of you for calculating sample size

[@researchwithfawad]

Not yet but hopefully soon. If your population size is known, please refer to Krejcie and Morgan (1970) table If your study is based on perception, try to have sample size over 200 (Roscoe, 1975). Using SEM, have sample size n * 10, where n is the no. of items

[@tsanin4539]

Hi, I have a second order model, now I have a problem that the HTMT value are more than 0.90. Is there any related paper that explain how to solved htmt problem? In my case I've htmt score more than 0.90.

[@researchwithfawad]

Please follow the steps in the video they will help you solve the issue. You may also watch this video. kzbin.info/www/bejne/o2fXhp99nq1sla8

[@tsanin4539]

@@researchwithfawad Yes, I've follow the step from this video. But, when I open the paper from this video I am not found the statement when htmt are more than 0.90 then remove the indicator. Can I get the paper that I can cite for my research? Thank you in advance

[@researchwithfawad]

Please read the whole paper, if there is high discriminant validity you do not have any other option but to either merge it or remove it. also, not everything has a reference.

[@rashikakumari2621]

Sir, if AVE value is 0.449 but the cronbach alpha and composite reliability is 0.886 and above 0.7 then can we say that the model is fit?

[@researchwithfawad]

AVE and CR are not for model fit. But yes, you can say construct is reliable and valid

[@rashikakumari2621]

@@researchwithfawad Thank you so much sir. What are the parameters in model fit in smart pls4?

[@user-bs7dt1rj9p]

Dear Professor, May I ask you how to combine 2 constructs into one? I have 2 variables with 3 indicators each of them. Their convergent validity is less than 0.7. I want to combine them. Thank you!

[@researchwithfawad]

You can just put all items into a single construct.

[@user-bs7dt1rj9p]

@@researchwithfawad ok, but that 2 variables were also measured through the questionnaire. To be clearer, I have 2 formative constructs (2 variables with 3 indicators each). I also measured each one of that 2 global variables through the questionnaire and I have different responses. I understand that i have to put together all the 6 indicators, but how should I do with those 2 variables?

[@tumchooyoong7042]

Htmt red in value what can be conclude on this? How to interpret?

[@researchwithfawad]

Please watch this video kzbin.info/www/bejne/l3-lfmVugN96aZY

[@stouts0899]

EssalamAlaykom Dr, Pls I have one construct with 6 indicators and AVE< 0.5 but when I drop with 2 items with Factors loading

[@researchwithfawad]

WAS. Overall from the model, are you deleting more than 20%, further removing the items shall not deviate the scale from the conceptualization for the construct.

[@stouts0899]

​@@researchwithfawad Big thanks Dr, you are doing a great job helping us with our studies. Could you share with me references to support this idea.

[@researchwithfawad]

Primarily you can use the following references Collier, J. E. (2020). Applied structural equation modeling using AMOS: Basic to advanced techniques. Routledge. Hair JF, Hult GTM, Ringle CM, Sarstedt M. Thousand Oaks, USA: Sage Publications; 2014. A Primer on Partial Least Squares Structural Equation Modeling (PLS-SEM) Hair Jr, J. F., Hult, G. T. M., Ringle, C. M., Sarstedt, M., Danks, N. P., & Ray, S. (2021). Partial least squares structural equation modeling (PLS-SEM) using R: A workbook

[@humairaahasan]

Assalamualaikum dr, I have a question. Is there a required gap for a good discriminant validity between variables? My variable discriminant validity has higher loading on its own variable but the gap between the other variable is quite close (less than 0.2). For example: the discriminant validity for my variable satisfaction when loading on its own is 0.782 but the loading with another variable is 0.620, Is the close gap acceptable?

[@researchwithfawad]

WAS. If the gap is over 0.10 you are good to proceed with further analysis

[@humairaahasan]

@@researchwithfawad Alhamdulillah. Thank you for the fast reply dr. Does it have a reference I can cite? for justification. If you don't mind sharing it with me

[@researchwithfawad]

The cut-off value for loadings was a significant level of 0.7 (Hair et al., 2013). Thus, items with a loading higher than 0.7 on two or more factors were deemed to have significant cross-loadings. Hair Jr, J. F., Hult, G. T. M., Ringle, C., and Sarstedt, M. (2013), A Primer On Partial Least Squares Structural Equation Modeling (PLS-SEM). SAGE Publications, Incorporated

[@humairaahasan]

@@researchwithfawad Thank you for the reply dr

[@hanahoah7325]

Assalamu'alaikum Sir. I have questions. If my hypothesis: -H1= one tailed, 5% -H2= two tailed, 5% What options should i choose in significance level when i want to run bootstrapping, should i do: @two step -step 1= choose one tailed, 0,05 -step 2= choose two tailed, 0,025 Or; @one step I just do like step (choose two tailed, 0,025 and for p values:2 to generate h1 hypothesis p values) Or how sir? With my respect for ur respon sir

[@researchwithfawad]

Do two tailed. But for those hypotheses that are one tailed, divide p value by 2 and then report that p value.

[@hanahoah7325]

@@researchwithfawad two tailed with 0,025, right sir?

[@researchwithfawad]

No, the p value you obtain will be divided by two to make it one tailed.

[@hanahoah7325]

@@researchwithfawad sir sorry im still not understandur answear U mean, u reference me choose two tailed, and what sig level that i should choose ? 0,05 Or 0,025 ?

[@researchwithfawad]

The p value that you get from the results, divide that by 2 to get one-tailed results.

[@kokshiongpong1690]

Hi Dr, after deleted, HTMT looks fine but the relationship become not significant. How to explain this? Thanks.

[@researchwithfawad]

How many items were removed. Delete one by one and see hownit affects the significance. You may use tue more liberal measire of 0.90 for HTMT.

[@kokshiongpong1690]

@@researchwithfawad 1 - 2 items for each dimension since a few HTMT were above .90. I did it one by one, the HTMT value reduced to < 0.9 for a particular variable. However, after two of the HTMT values scored below 0.9, the relationship become not significant and there are still 1-2 values of HTMT above 0.90. Thanks.

[@researchwithfawad]

Try removing items from difference constructs to what you have removed from now, if using cross loading and check again.

[@abdullahalmahroqi8166]

Sir, I have a variable with perfect reliability and discriminant validity, but its AVE is 0.4. So, can I keep it or I must delete it ?

[@researchwithfawad]

If CR is over 0.70 and AVE is 0.40 or over, this is acceptable.

[@abdullahalmahroqi8166]

@@researchwithfawad Thank you sir. May Allah reward you

[@syedandrabi489]

Sir ..how can we deal with lower AVE if cronbachs alpha and composite reliability is excellent

[@researchwithfawad]

Please check for low loadings

[@syedandrabi489]

@@researchwithfawad sir what if the deletion of certain items does not impact the value of AVE much..it still.remains below 0.5

[@researchwithfawad]

If its over 0.40 and CR is over 0.70, you are good to go.

[@syedandrabi489]

@@researchwithfawad ok sir thankyou very much... almighty bless u.. Sir can i get a citation of the author so that I can get justification for O.40 ave

[@researchwithfawad]

Here is the title of the paper The impact of customer knowledge management capability on project performance-mediating role of strategic agility Syed Arslan Haider and Umar Nawaz Kayani See text under table 2