Post Your Answer

 Hi Ruchika, first of all I would like to ask what your research objective is. If it is fulfilled with the simple correlation approach, then why do you want to look for other ways? Because as you yourself said, it is a tried and tested method in other instances and provided appropriate results too.

Replied 3 years ago

By Ruchika Tuli

Hi Amardeep, it’s not that I want to make it complex. I was just thinking if we can use other methods as well. For example, using regression. Won’t it help in testing convergent validity too?

Hi Ruchika, as far as I understand answering your question is not really straightforward. Also, terms like convergent and discriminant validity can differ in meanings from the perspectives of different statisticians or researchers (methodologists). So, please clear me whether you are working on latent variables (or constructs) and relationship between them through the nomological networking perspective or are you working on a single scale perspective and trying to find the relationship between the different items and factors included in it. This will help me to help you better. In any case, I am attaching these papers’ links below so that you can understand both these points.   This is the link to the Psychological Bulletin by L. Cronbach and P. Meehl- https://doi.apa.org/doiLanding?doi=10.1037%2Fh0040957   And this is the link to he article Construct Validity: Advances in Theory and Methodology by M. E. Strauss and G. T. Smith- https://www.annualreviews.org/doi/10.1146/annurev.clinpsy.032408.153639   Both these links belong to the first option as I really think you are asking in that context. If not, then please let me know. I will try to find relevant papers for the other option as well. Good luck.  

Hello dear, I see you are doubting because you are trying to get variance and yes, that matters too. But the way I understood your question, your main focus is testing “convergent validity”. This is actually new to me because this means that you are not focusing on testing the prediction and finding correlation among the pairs is enough.  I am not saying it’s wrong, in fact it is good. The collinearity between predictors can cause a problem if you will test the prediction, so this is interesting. I know some researchers want to try new ways, different ways and it is okay to have that researcher’s diligence in yourself but I’d suggest not to choose a multivariate multiple regression approach for this. It might lead to further problems. Also, it is completely okay to believe in what “everybody” believes sometimes. Still if you have doubts, consult some experts. They may also help you to choose the right sample size for regression in case you are fine with starting anew.  https://www.bostondissertationhelp.com/ is one such site. I guess they have experts who help in other countries too. Best wishes, N.  

Hello, aren’t you talking about discriminant validity rather than convergent validity? I mean the explanation and correlation that you are talking about seem to me like that. This can be evaluated by Cross loading option in the HTMT ratio or you can use the Fornell Larcker Criteria.   Anyways, for convergent validity, I am mentioning a few points below, please consider them. The convergent validity requires VIF for the outer model or the factor loadings. It also requires the AVE (Average Variance Extracted), Composite Reliability, and Cronbach Alpha.   The conditions which need to be met to test establish the convergent validity are as follows. VIF < 5 Factor Loading > 0.70 AVE > 0.50 Composite Reliability > 0.70 Cronbach Alpha > 0.70   I hope it is understandable.  

Hi, I agree with Sohail. The convergent validity is tested to find the relation between the observed and their latent variables. I won’t go into details as Sohail has already mentioned the requirements. I would just like to suggest checking and clarifying what exactly your research needs from your supervisor or guide.