Hi,
At present I have two constructs (one independent and one dependent) each with 3 reflective indicators. However, one indicator (in both constructs) falls below the 0.707 threshold. When I remove this indicator (e.g. loading = 0.640) from one construct I am left with 2 indicators (but I believe that having two reflective indicators is a problem when analysing data and thus, one indicator is more appropriorate).
I was wondering if you had an option to use:
(a) the 0.707 threshold and reduce three reflective items to one item or (b) reduce the threshold to 0.6 and keep all three items
which would you apply and why?
Furthermore, my research is exploratory!
Which of these approaches would be more accepted for publication purposes? Any papers for guidance would be much appreciated.
Thanks.
Factor Loading - 1 reflective item or reduce threshold value
Factor Loading - 1 reflective item or reduce threshold value
Yvonne O' Connor
UCC, Ireland.
UCC, Ireland.
-
Diogenes
- PLS Super-Expert
- Posts: 899
- Joined: Sat Oct 15, 2005 5:13 pm
- Real name and title:
- Location: São Paulo - BRAZIL
- Contact:
Hi,
In a theoretical example, if we have all indicators with 0.707, the AVE will be 0.5.
In a more realistic example, some indicators will be greater than 0.707 e others lower, but we could assess the value of AVE if it is greater than 0.5.
In your case I’ll keep the three indicators, because “content validity”.
1) These threshold are more “fuzzy than crisp” in the sense that they are proposed as an acceptable value in the general cases. Even if AVE is a little lower than 0.5, I do not see a problem.
See:
LITTLE, T. D.; LINDENGERGER, U.; NESSELROADE, J. R. On selecting indicators for multivariate measurement and modeling with latent variables: When “good” indicators are bad and “bad” indicators are good. Psychological Methods, v. 4, n. 2, p. 192–211, 1999.
2) In a general sense, single indicator is not a good option, see:
DIAMANTOPOULOS, A.; SARSTEDT, M.; FUCHS, C.; WILCZYNSKI, P.; KAISER, S. Guidelines for choosing between multi-item and single-item scales for construct measurement: a predictive validity perspective. Journal of the Academy of Marketing Science, v. 40, n. 3, p. 434–449, 2012.
LISREL do not have problems with attenuation (when the correlations between LV are underestimated), but PLS-PM has (consistency at large), for this reason many authors have recommended a “generous” number of indicator by LV.
Best regards,
Bido
In a theoretical example, if we have all indicators with 0.707, the AVE will be 0.5.
In a more realistic example, some indicators will be greater than 0.707 e others lower, but we could assess the value of AVE if it is greater than 0.5.
In your case I’ll keep the three indicators, because “content validity”.
1) These threshold are more “fuzzy than crisp” in the sense that they are proposed as an acceptable value in the general cases. Even if AVE is a little lower than 0.5, I do not see a problem.
See:
LITTLE, T. D.; LINDENGERGER, U.; NESSELROADE, J. R. On selecting indicators for multivariate measurement and modeling with latent variables: When “good” indicators are bad and “bad” indicators are good. Psychological Methods, v. 4, n. 2, p. 192–211, 1999.
2) In a general sense, single indicator is not a good option, see:
DIAMANTOPOULOS, A.; SARSTEDT, M.; FUCHS, C.; WILCZYNSKI, P.; KAISER, S. Guidelines for choosing between multi-item and single-item scales for construct measurement: a predictive validity perspective. Journal of the Academy of Marketing Science, v. 40, n. 3, p. 434–449, 2012.
LISREL do not have problems with attenuation (when the correlations between LV are underestimated), but PLS-PM has (consistency at large), for this reason many authors have recommended a “generous” number of indicator by LV.
Best regards,
Bido