Hi David,

You can indeed use a single large sample to conduct both EFA and CFA given the same survey. Given your sample size you may want to randomly split the data into two sets of 500+ (50% each). The EFA data set can then be used for finding a best fitting model. This model can be decided upon by fit indices and/or theory. You can really do anything with this data set (e.g., remove items, pick any factor structures, condense response options) because this is an exploratory analysis where you are building a probable explanation of your data. After you have decided on a model from the EFA data, you can then use the remainder of the data to run a CFA. This CFA should be the final model you decided upon with your EFA. The entire idea behind this is that you are trying to find a best model with the EFA data and you are trying to test that model with the CFA data.

You could also get a sample of individuals and run an EFA, make changes to the questionnaire based on the EFA, then collect new data and run a CFA on this new data based on the revised scale (or another EFA and repeat the process). Both methods are valid.

Dear David,

I didn't get what you mean by new questionnaire. But i guess you want to develop new measures for some constructs. If its the case, you need to follow steps as suggusted in prior research such as:

A paradigm for developing better measures of marketing constructs by Churchill J 1979.

For your question, you need to collect the data to purify the scales using efa then collect a new data based on the revised scale and use the cfa.

Dear David,

Indeed, you can split the sample and perform EFA and CFA. A rule of thumb is to get nearly 10 participants by item.

My guess is that if you make minor changes to the structure (for instance remove few items) according to EFA results, you can perform a CFA with the second portion of the same data set. However in subsequent work you should consider testing the new version of the scale with a new data collection.