forum.smartpls.com

Hi,

I'm running technology acceptance models which includes moderators such as age, gender etc. I've coded them in SPSS (e.g female = 0 and male = 1) before importing them into SmartPls3. However, when I run my PLS Algorithm my construct reliability and validityy for all of my moderators shows 1. Why is this the case? Also I've noticed that it happens at any instance where I've only got 1 question and when I use the moderating effect. I've attached my my report below.

Hopefully someone can assist :)

First, single indicator variables always have perfect reliability. You cannot assess reliability based on only one item. The construct is the item.
You may specify a user-defined reliability and use PLSc if you want single items to have lower reliability. However, you need strong theoretical (or empirical; e.g., from other studies) support for this.

The reliability of interaction effects is theoretically the product of the reliabilities of the two constructs that it contains (i.e., predictor and moderator). However, technically the reliability is not directly estimable in PLS as we are usually using some workaround in form of two-stage approach (with a single indicator in the second stage; see problem above) or product-indicator approach. That is why we report the reliability as 1, but it may be lower. You may just use your reliability estimates from the predictor and moderator to approximate the reliability of the interaction effect.

Thank you for your response. I understand that I only have one item e.g. age but I'm still failing to understand why (refer to my Habit variable) my interaction e.g. Habit ->moderated by age/gender->BI is 1. Experience has 3 items (cronbach = 0.857) but even when that is used as a moderator
Habit->Exp->BI my interaction is 1 (even though all the constructs cronbachs' are not equal to 1). Your assistance once again will be appreciated :)

It is a technical issue. The reliability of an interaction effect is not directly estimable from the attached indicators. It is more conceptually the product of the two reliabilities from the predictor and the moderator. You need to calculate that yourself if you which to get that.
However, note, that it may only give you an approximation of the reliability because the product of two Cronbach’s Alpha values cannot be expected to capture the reliability accurately in this case.