Dear Researchers, I have 6 independent variables and 1 mediate variable and 1 dependent variable, out of these 6 independent, 1 variable is non-significant using Co variance base SEM (FCA) via smartPLS. I used different measures such as Cronbach’s alpha , CR and AVE along with model fit measures and T-Test...Please guide me, should I delete the non significant one or should I explain it...
Kind Regards
Muhammad Khan
I have seen it done both ways. Sometimes people will trim the model to delete nonsignificant paths, but most of the time people will leave it in to show the nonsignificant parameter estimate. Leaving it in shows results for the model as you originally proposed it, which is the model you should test and evaluate.
Respected and dear Paul E. Spector & Michael R. Frone thanks millions for your valuable comments...
I had faced the same problem during my research and I have discussed with one of the senior professor who is very good in statistics and Sir advised me to drop that variable. I did that then the model fit was suddenly improved and other indicators are also improved and the paths were significant.
So my suggestion to you please remove that variable.
Hi,
you shouldn't drop the variables. In addition what Paul and Michael said, variables in a model most often have a control or adjustment function. This function serves causally identifying the parameters of other variables (if relevant controls are in the model). As any causal identification this is a conceptual issue located on the population level, not the sample. Hence, even if the sample estimate may be non-significant, the controlling function works, as long the variable is in the model (in most of the cases, the estimate won't be exactly zero). Removing the variable, hence, biases the effect of the other variables.
Beyond this parameter-focused issue, by removing the variable, you change the model structure and its inherent restrictions, thus influencing the model tests.
You created the model based on theory, now stick to it :)
Best,
Holger
Respected and dear Holger Steinmetz Thanks a lot for your valuable comments, I will not drop the variable as you mentioned it effect also the other variables as I tried...
Non-significance is not a crime. Infact it should be discussed in detail and also give your recommendations for next researchers.
Your non significant path may help the next researcher to
1. estimate the sample size for desired effect
2. some variation in tool construct, etc.
And of course there will be citation for you :)
So, Dear Keep it.
If one of the multiple variables is not significant should it be deleted from the model or kept in?
Suggesting not to delete but instead - to discuss / explain your rationale why & how this observation is possible after further literature review & provide your future research recommendation - this is also part of knowledge contribution. Reason being you are conducting a covariance-based SEM using FCA to confirm the model in which the particular insignificant independent variable should had carried certain weight / importance in previous research models.
Brother Khan...I think in the first place your model was a proposed model which can only be confirmed when tested for validity. However, if any variable among the proposed variables appear to be insignificant either way the model can be reported as its is and in the discussions you make effort in explaining why you have such model or you can remove it and run the model again before having a final validated model. Remember in discussions you need to review different similar researches
the nonsignificant variable must be discussed. As the reason may be change in the cultural context, the mindset of sample or any other. So, in my opinion it must be discussed along with the reason.
Dear Muhammad Khan, have you tried with a multivariate analysis? I mean that when you use a multivariate analysis some variables that could not be significant alone, could be relevant in combination with some others, so it depends on what are you looking for.
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques
- Structural Equation Modeling