Problem with significant t and small beta

[estherxing]

Hi, I encountered problem with significant t and small beta value, like t=2.576683 and beta=-0.139495; t=1.648455, beta=-0.058015; t=2.090919, beta=-0.086341.

As t=beta/VAR, such phenomenon infers that VAR is too small. It might because of high alpha value for the measurement items of a construct, or respondents all give the same answer. So I wonder how the Standard Deviation (STDEV) is calculated in PLS? It is based on the different measurement items of a construct, or different responses given to one measurement item (i.e. one question)?

Also, I have some constructs measured by only one item, so will that also contribute to this phenomenon, i.e. significant t and small beta?

Thank you very much for your kind help!

[Diogenes]

Hi,

1) With bootstrap procedure, for example, n=100 cases with 1000 re-sampled models (samples) we´ll have 1000 betas (one for each re-sampled model), then the mean, standard deviation and t-values are computed from these 1000 betas.

2) Even been significant, how about practical significance?

Best regards.

Bido

[dhenderson17]

Have you checked the distribution of your independent/dependent variables? I'm wondering if your responses may be highly skewed and non-normal. So, perhaps (and this just came to mind), you could take transform the data--take the square root for example---to make it more normally distributed...just a thought.

[viswadatta]

PLS is a non parametric method and is independant of the distribution of the variables.

Small beta denotes a lower influence of the construct on the endogenous construct which is nonetheless significant

[estherxing]

Thank you for your insightful explainations! Yes, as Prof. Bido noted, the practical significance is too low, so I would treat it as non-existence (i.e. for those beta value = 0.01, 0.02)..