Dear Kara Ma

You do not have to do EFA. You can rely on CFA to validate your research model. EFA is mainly used if your research aim is to develop a scale to measure a particular latent variable, however, it is not the case here. Therefore, you can proceed with that.

Good Luck with your thesis.

Best Regards,

Belal Edries

Dear Kara Ma,

Confirmatory factor analysis will be sufficient because you are not trying to identify a new conceptual structure, you have adapted an existing scale, so I suggest you use CFA. Good luck for your study :)

Belal Edries Tuğba Karataş Thank you for helping! It really helps. after I did the CFA, and deleted several items using SmartLPS, the factor loading looks good for me now. I will adopt what you suggested!

I have another question. Some HTMT for measuring discriminant validity is higher than 0.9, some even higher than 1. Do you know what's the best solution? Is the only way to delete some items or a variable? (I am studying the path relationship from post-purchase experience to customer satisfaction to trust, then to repurchase intention, and I have found there are high correlations between trust and another two)

Dear Kara Ma

There are three measures for the discriminative validity which are the following:

• Cross-loading
• Fornell-Larcker
• HTMT

Regarding which one to rely on, there is no concrete answer whether you should use all or not, based on my observation the majority have relied on Fornell-Larcker criteria.

However, In assessing the discriminant validity, the HTMT criterion is superior to both the Fornell and Larcker criterion and the assessment of (partial) cross-loadings (Henseler et al., 2015). Although the Fornell–Larcker criterion had been long recommended to assess the discriminant validity of latent variables, it is ineffective in combination with traditional PLS-PM because it relies on consistent factor loading estimates. HTMT is a reliable tool for assessing discriminant validity, whereas the Fornell–Larcker criterion has limitations that do not justify its reputation for rigour and its widespread use in empirical research (Henseler et al., 2020).

Back to your question:

I would like to draw your attention to a set of actions that may apply and get your measurement model through the validation process.

• You probably have applied this before but I would suggest rechecking the data, for any outliers that might be still there.
• Check your sample size with relative to the number of items you have in your study. Roughly speaking, if you have between 10 to 20 respondents per item, that should be okay. However, if possible aim to collect more response.
• Drawing the factor correlation circle would be somewhat useful to present you a graphical image of the current situation, however, unfortunately, what I had is a code for drawing the correlation circle of the items using PCA using R software but I have included it anyways with a brief illustrating of each line purpose.
• In some cases, the miswording of the questions may lead to some degree of pushing one of the factors in another direction thus inverting the scale is usually applied to overcome this issue. Therefore, I would suggest rechecking the question and ensure that if the respondents read the questions carefully and they were designed to measure the constructs reflectivity, then the responses should be in one direction. If you notice that many respondents have not met this criterion then probably should consider to take an action.
• This might the least favourable action, however, if necessary to test the influence of dropping one item at once and check the result. If you notice that you will drop many times so the construct is left with less than 2 items, then I would suggest you either to try to combine the two latent variables together in a single variable if the theory enables that, while if not you are left with only an option to drop a latent variable.

Best Regards,

Belal Edries

Only CFA in case you have adopted standard scale and made minor changes. But if you have constructed t he whole scale yourself as in made new items you would have to do a efa on a separate set of samples and CFA on separate set of samples