second order construct: repeated indicators approach

zfayena · Post by **zfayena** » Wed Jan 01, 2014 10:10 am

Dear Experts

My name is Izak , and I am a doctorate student at the University of Manchester, UK. My topic is about the ways international high-tech entrepreneurs learn about the business opportunities. The research is conducted as a mixed method study. Firstly, Please allow me to express my appreciation for your prompt feedback and replies, I find them very helpful. I have been developing my PLS model, step by step, using mainly the book written by Hair et al. (2013) and their guidelines and 'rules of thumbs'.

Secondly, and if I may so, I'd like to get your professional opinion on an issue, which is a little bit confusing and unclear to me.

I'm modelling a PLS model (SmartPLS), with a second order construct (HOC) and three first order constructs (LOCs). The measurement model is considered as reflective-reflective. In the first step I followed Hairs' (2013) instructions how to assess the reliability and validity of the measurement model. I conducted this assessment for each one of the LOCs. The LOCs produced a satisfactory internal consistency, convergent validity and discriminant validity. In the second step, I evaluated the HOC reliability and validity by assessing the secondary indicators' (the repeated indicators) outer loadings (i.e. the loadings of each of the repeated indicators on the HOC directly), than I have assessed the CR and the AVE (the AVE was computed by averaging the squared repeated indicators' loadings) and finally I have evaluated the discriminant validity by comparing the HOC to the other constructs in the model, as described in your book. My questions are:
1. 1. Have I conducted the right steps and evaluated the HOC in the second step as it should be? If no, may I ask what are your detailed recommendations?
2. 2. I am asking this question because in an article, which was published by Wetzels et al. (2009) they described the repeated indicators approach differently. If I understand them correctly, according to their approach, the repeated indicators process is used to compute the latent scores for each of the LOCs, and then to use these scores to establish the reliability and validity of the model and not as was instructed in Hairs' book (which in my humble opinion makes more sense)
I would really appreciate your reply, as I am a Doctorate student who plans to submit his thesis in the near future,

I wish you a happy and healthy new year,

Izak

Diogenes · Post by **Diogenes** » Wed Jan 01, 2014 9:58 pm

Hair et al. and Wetzels (see p. 181) give the same recommendation.
Your rmodel is ok.

We use the LV score approach when we have some LOC with many indicators and others with few indicators (If we repeate these indicators, the load from HOC to LOC will be bigger in the LOC with more indicators).

Best regards,
Bido

zfayena · Post by **zfayena** » Thu Jan 02, 2014 12:49 pm

However, Wetzles uses the loadings of the first order construct to assess the validity of the second order. If I understand correctly they compute the AVE of the second order based on the first order constructs' loading and not by using the loadings of the repeated indicators. What do you recommend?

Diogenes · Post by **Diogenes** » Thu Jan 02, 2014 1:05 pm

See: viewtopic.php?t=640
Best regards,
Bido

zfayena · Post by **zfayena** » Thu Jan 02, 2014 1:50 pm

Thanks for your prompt reply!,

Hair (PLS premier book) used the mv of the second order to compute the AVE of the second order construct (Hair et al. 2013, p. 237). but, in your professional opinion, and in the same vein as Wetzles et al. (2009), I shall use the path coefficients between the first order and the second order as the loadings of the second order construct. These values will be used for further calculation of the AVE and etc. Can you please instruct me how to justify this? on the basis of what references?

Best wishes,

Izak