Bootstrapping

[estherxing]

Sorry recently I have flushed the forum with my enquiries, I am now doing data analysis for my thesis and a bit tight on my schedule. Thanks for your kind understanding n support :)

I have a problem with Bootstrapping. When I run it at the case = 255 (my sample size) and run time = 500, the software run EXTREMELY slow and when it has run for 30 sub-samples or so, the bootstrapping window would close automatically. Sometimes, it even gives me error msg 'A singular matrix occured during the estimation of the outer weights of a block. Using more indicators could solve the problem.' However, all the constructs are measured with sufficient indicators using a 7-point Likert type scale.

I wonder whether it is because of large amount of data. When I cut the model into half, it works. However, for my thesis, I need to get the full model. So may I know anyone has any suggestions on this? :)

Thank you and a lovely day!
Esther :)

[cringle]

Dear Esther,

the bootstrapping performance depends on you system (processor speed and memory). By default, Java uses 256 MB memory. you can use the settings in the SmartPLS preferences to add more memory. Alternatively, you can start SmartPLS from the windows command box with

Code: Select all

java -Xms 512M -Xmx 1024M -jar C:\Programme\SmartPLS2\startup.jar

or similar(see sun java webpages for details) - the command simply determines the minimum and maximum memory to use (don't state more than you have).

This should speed up bootstrapping.

About the singular matrix: this usually happens when all data points of a certain variable are identical. You may also check, on using missing values and having appropriate settings in SmartPLS (case wise deletion may results in a too small number of cases for PLS computations).

Best
Christian

[estherxing]

Thank you Christian for your insightful suggestions! :D