Page 1 of 1

Urgent: t-test for observed variables

Posted: Sun Apr 12, 2020 7:56 am
by hurdemirbilek
Hi to everyone,

In my model one of the 2nd order construct that has three LVs. The structure is reflective-reflective. (as attached, numbers in paranthesis is t value, numbers in outside paranthesis is path coefficinets)
When I do bootstrapping, surprisingly t statistics of all indicators above are below 1.96 with minimum 1.613 maximum 1.919 as below:
LV1 has 4 indicators
t of Ind1: 1.672
t of Ind2: 1.686
t of Ind3: 1.669
t of Ind4: 1.650
LV2 has 4 indicators
t of Ind1: 1.741
t of Ind2: 1.725
t of Ind3: 1.714
t of Ind4: 1.613
LV3 has 2 indicators
t of Ind1: 1.919
t of Ind2: 1.912

By the way t statistics of all other LV and OV are OK.
What am I suppose to do now?

Re: Urgent: t-test for observed variables

Posted: Sun Apr 12, 2020 8:07 pm
by jmbecker
What are the loadings? Are they sufficiently high (i.e., larger than 0.7) or not? Usually, loadings are only insignificant if they are small. If they are large, sign changes in the bootstrap could be a problem (i.e., that they flip from +0.8 to -0.8, creating large variation in the estimates).

Or maybe the indicators are just not reflective indicators, but formative indicators?

Re: Urgent: t-test for observed variables

Posted: Mon Apr 13, 2020 9:59 am
by hurdemirbilek
Thanks for your attention.
In fact, factor loadings are not too small.
They were reflective in previous studies.

I revised the model and the problem still exists. I have high loadings but they are not significant. What am I supposed to do?

Lastly, the cause of the problem seems about "no sign changes" option but it is not configurable, I am using 3.2.9. How can I change it?

Re: Urgent: t-test for observed variables

Posted: Tue Jun 23, 2020 7:30 am
by Ng11
In my case, I have experienced the similar problem where my t-test value is too slow which does not exceed 1.65 and also my p value is too high almost approaching to 1. May I know any solution for this issue? Or how should i justify this based on the obtained results?

Re: Urgent: t-test for observed variables

Posted: Tue Jul 07, 2020 2:19 pm
by jmbecker
One way to decrease the variability and to increase t-values is to increase sample size. If you sample size is too small and your model too complex (you seem to have complex higher order constructs with repeated-indicators) you will get instable results.
Another problem could be insufficient identification of the constructs. If they do not have paths to other constructs that are sufficient strong, it is hard to empirically identify the model parameters. Hence, you need to rethink your model.