forum.smartpls.com

Hi every one,
I hope i find an answer in your forum because I really have a big problem :
I have a second order construct which is multidimentionnal in the theory. up to here all is fine. When it comes to empirical issues I found low positive correlation between the 1st order dimensions which normally not allows me to build a 2nd order construct and a lack of convergent validity is up but the surprise when i test the second order construct and follow theory i found not bad value of regression weights between 1st and 2nd order (.4 to .7). so it does it mean that i am faced with formative LV ? can we increase correlation between 1st order dimensions or it's possible to consider the 2nd order and violate the rule of high correlation as recommanded by Jarvis & al (2003)? any advice on this knowing that in litterature it's always reflective contruct not the inverse. by the way I'm using smartPLS 3.
Thank you

Because the second-order construct will be based on the first-order components it will not be surprising that you will find some strong correlations / regression weights between first and second-order constructs. You need to consider whether it is possible from a theoretical point that your first-order constructs form the second-order construct (formative) measurement or not.
Low or high correlation by themselves of not evidence of anything without substantive theoretical support.

jmbecker wrote:Because the second-order construct will be based on the first-order components it will not be surprising that you will find some strong correlations / regression weights between first and second-order constructs. You need to consider whether it is possible from a theoretical point that your first-order constructs form the second-order construct (formative) measurement or not.
Low or high correlation by themselves of not evidence of anything without substantive theoretical support.

Thank you Dr for your fast reply.
I'm little confused did you mean : for reflective construct if one condition or criterion of Jarvis & al is not respected it should be formative? Well in the theory it considered reflective even these dimensions are independant and moreover litterature doesn't mention any thing about whether they use it as reflective or formative. few paper discuss little about debate on that but don't give any clear way. I use the jarvis criteria and i find it more logic to be formative but i read somewhere that choosing formative need to be proven by a very big data to say that only these dimensions form the construct. Please any references talking about how much i need as sample size for formative contruct? I'm really upset.