I do not know how to solve this problem. Please, can somebody help me?
I applied the FIMIX method and I got two segments.
When I am running the bootstrapping procedures on the first segment I get the message:
Calculation error
An error occurred during the estimation outside. Maybe there are too few observations.
When bootstrapping procedures is applying on the total sample and on the second segment I have not this problem.
FIMIX and bootstrapping
-
christian.nitzl
- PLS Expert User
- Posts: 248
- Joined: Sat Jul 25, 2009 1:34 pm
- Real name and title:
Dear Christian, thank you for your replay,
The sample size in the subsample is 66 cases. The model has 4 constructs, and the large construct (block) has 4 variables.
I really need bootstrap procedure results as standard errors because I need to calculate t - values for path coefficients.
I tried all bootstrap method (as sign change), but it didn’t help.
What do I need to do in this case?
Semina
The sample size in the subsample is 66 cases. The model has 4 constructs, and the large construct (block) has 4 variables.
I really need bootstrap procedure results as standard errors because I need to calculate t - values for path coefficients.
I tried all bootstrap method (as sign change), but it didn’t help.
What do I need to do in this case?
Semina
-
christian.nitzl
- PLS Expert User
- Posts: 248
- Joined: Sat Jul 25, 2009 1:34 pm
- Real name and title:
That’s sounds a little bit strange to me because 66 cases should be enough for such a model. Perhaps there is a variable with to low variance. That means for one item there are almost the same numbers for all cases which causes the failure. Do you use binary coded variables? Use also another algorithm for the path coefficient estimation (e.g. path weighting sheme). How many samples do use for the bootstrapping since 100-200 are could be enough.
Greetings,
Christian
Greetings,
Christian
I have the same problem:
FIMIX gave me 3 segments with one very relevant, as some indicators are exactly the same number for all cases.
The problem now is, that the related construct is the most important one, which should be explained.
I have tried to vary the results a little bit as an a priori segmentation, but this does not return good results at all.
Do you have any thoughts on this?
and if there is no solution: how could this be reported in my dissertation?
segment 1 and 2 work fine and have good results, but segment 3 has problems because the segment is so sharp`?
FIMIX gave me 3 segments with one very relevant, as some indicators are exactly the same number for all cases.
The problem now is, that the related construct is the most important one, which should be explained.
I have tried to vary the results a little bit as an a priori segmentation, but this does not return good results at all.
Do you have any thoughts on this?
and if there is no solution: how could this be reported in my dissertation?
segment 1 and 2 work fine and have good results, but segment 3 has problems because the segment is so sharp`?
Dear Christian,
bootstrapping doesn't work for me either and the msg that occurres says "Using more indicators could solve the problem."
I have some measurement models with only one indicator: e.g. age, driving experience in years...
Could it be the reason? How can I solve the problem if I still need those models in my construct?
Thank you for your support!
bootstrapping doesn't work for me either and the msg that occurres says "Using more indicators could solve the problem."
I have some measurement models with only one indicator: e.g. age, driving experience in years...
Could it be the reason? How can I solve the problem if I still need those models in my construct?
Thank you for your support!
-
christian.nitzl
- PLS Expert User
- Posts: 248
- Joined: Sat Jul 25, 2009 1:34 pm
- Real name and title:
Hi Tatjana,
Did this message really appear when you run the bootstrap procedure or did the message also appear when you run the normal PLS algorithms? Did you try different algorithms’?
Normally there is no technical problem to use single items in PLS. I am already calculating a model in PLS with only single items. This is like a normal regression analysis. Perhaps there is one item/construct that caused the problem (to less variance…) or your sample size is too small for the complexity of your model.
Try to find out if there is a single indicator which causes the problem. Delete one after another of the items and calculate the model once again.
Try also to double your sample in the original data table by copy and paste. That can give you a hint if there is a problem with your sample size.
Good luck!
Christian
Did this message really appear when you run the bootstrap procedure or did the message also appear when you run the normal PLS algorithms? Did you try different algorithms’?
Normally there is no technical problem to use single items in PLS. I am already calculating a model in PLS with only single items. This is like a normal regression analysis. Perhaps there is one item/construct that caused the problem (to less variance…) or your sample size is too small for the complexity of your model.
Try to find out if there is a single indicator which causes the problem. Delete one after another of the items and calculate the model once again.
Try also to double your sample in the original data table by copy and paste. That can give you a hint if there is a problem with your sample size.
Good luck!
Christian