Page 1 of 1

Removing indicator leaving single indicator

Posted: Tue Jun 27, 2017 4:26 am
by iputu.ariyasa
I have an exogenous formative higher order construct with four reflective lower order constructs. One of the LOC has two indicators, and after data collection stage, one the indicators has low outer loading and resulting in low reliability & validity of the LOC. Regarding the relationship between the LOC and HOC, whether i keep the indicator or not, the weight of the LOC showed a low significance level

My question is:
  • 1. should i remove this indicator and leaving the LOC with only single indicator (and possibly reducing the content validity)?
    2. can anybody point me to a reference(s) that could justify the removal or retention the indicators?
thanks

Best Regards,
I Putu Ariyasa

Re: Removing indicator leaving single indicator

Posted: Wed Jun 28, 2017 8:11 pm
by jmbecker
I don't have a specific reference in mind, but generally a two indicator reflective construct is often problematic in terms of achieving the desired reliability criteria as it is not really a common factor type of construct, which would need at least three indicators.
I, personally, would keep the indicator and opt for higher content validity. In addition, higher reliability and validity after removal are an illusion as a single item construct is not able to assess reliability and validity and thus defines the reliability as 1 though it is probably lower. Thereby you have no chance to evaluate, whether your removal really improves the reliability and validity. Thus, keeping the item seems more plausible to me.

Re: Removing indicator leaving single indicator

Posted: Thu Jun 29, 2017 3:49 pm
by iputu.ariyasa
jmbecker wrote:I don't have a specific reference in mind, but generally a two indicator reflective construct is often problematic in terms of achieving the desired reliability criteria as it is not really a common factor type of construct, which would need at least three indicators.
I, personally, would keep the indicator and opt for higher content validity. In addition, higher reliability and validity after removal are an illusion as a single item construct is not able to assess reliability and validity and thus defines the reliability as 1 though it is probably lower. Thereby you have no chance to evaluate, whether your removal really improves the reliability and validity. Thus, keeping the item seems more plausible to me.
Thanks for your reply, i think i would keep the item too

Re: Removing indicator leaving single indicator

Posted: Fri Jan 26, 2018 7:24 pm
by manjap
Hello. I am new to statistics and SmartPLS and have read a few posts on single indicators.
In my model, I had to remove several indicators that were below 0.7 with LPM and PEU leaving just 1 single measurement item left and a resulting loadings of 1.0.
Is there anyone who can advise if I have to take any action on the loadings of 1.0 in SmartPLS, specifically how to handle reliability and validity for LPM and PEU with the loading of 1.0?

Attached is 2 files of my model: stage 1: red circles with all indicators below 0.7; stage 2 (revised): all items below 0.7 removed

Many thanks, Manja

Re: Removing indicator leaving single indicator

Posted: Tue Jan 30, 2018 9:07 am
by jmbecker
First, I would not blindly delete indicators. I would try to understand why they have performed so poorly and if there are any data reasons (missing values, (reverse) coding, etc.) that could contribute to the problem.
Second, I would delete the indicators sequentially (lowest loading first) and re-estimate the model each time. This may lead to increased loadings of the other remaining indicators and therefore deleting fewer indicators.
Third, I would not use a single indicator instead of two or more somewhat unreliable indicators (loading 0.5 - 0.7). The one indicator (although looking good in terms of loadings, etc.) might represent your conceptual variable very badly. You probably had a good theoretical foundation for collecting multiple indicators. Thus, try to stick with as many indicators as possible even if they are slightly unreliable. This is still better than removing too many. The single indicator is usually worse than the composite of unreliable indicators. How can a single indicator suddenly become very good, just because you delete other indicators? --> Because it hides his unreliability behind the lack of evaluation criteria.
Only if you think that in a given situation a single-indicator is sufficient (usually you choose the indicator to be a single indicator a priori because it represents the conceptual domain very well, e.g. "Are you satisfied" for satisfaction) you should use it, but not if it is a result of indicator removal.