You might have several reasons for that. That is one of the reason that it is recommended that every instrument should be validated for specific sample.

Here I will assume that you use the same methods as in the previous validation process.

There are some constructs that  are very dependable on culture. However, I don't think this is the case, since previous studies was based with different culture.

Other reason might be  the translation process. Usually, you have two different people translating the original language to the Spanish (in this case). Then, you should ask two different people to make a backwards translation, which means translating back from Spanish to the original language.  

A third reason might be the sample. You might have a very different sample from previous studies and that might affect the factorial solution. 

Further, it is not because they have the same factor structure that they respect the non invariance for the parameters, which means that not necessary they are measuring the same construct. 

I suggest that you have a look into the theory and see if it is acceptable to have 11 factors. Further, you might contact the authors from previous studies and ask for the script and more information regarding the type and what kind of estimator they analyzed the factorial structure of the test. 

If all this issue was already considered, you might find different factorial solution for the same instrument. There a few of this examples in the literature. At the same time, I don't remember by heart such a big difference. 

Best,

Hi,

I would add:

a) the factor solution in the original studies were invalid (check how they did it)

b) the samples have to be comparable with regard to other influences of the structure. For instance, if the English sample consisted of individuals with low education, whereas your sample consisted of individuals with a high education, this difference could lead to a more differentiated structure. Sometimes more direct causes are responsible and cultural differences are only spurious. 

Best,

--Holger

Culture and language can have effects on the psychometric properties of instruments. We found different factor structures comparing Americans and Singaporeans on a job satisfaction scale in the following paper:

Spector, P. E., & Wimalasiri, J. (1986). A cross-cultural comparison of job satisfaction dimensions in the United States and Singapore. International Review of Applied Psychology, 35, 147-158.

The following paper discusses more general issues in translating and transporting scales across countries/languages:

Spector, P. E., Liu, C., and Sanchez, J. I. (2015). Methodological and substantive issues in conducting multinational and cross-cultural research. Annual Review of Organizational Psychology and Organizational Behavior, 2:9, 1.9-31. [Early view]

Hi Cristobal,

did you find the 11 factors in a confirmatory or exploratory analysis?

Anyway it could be because sample features, translation or the theoretical construct measured.

Regards

Thank for your support, we will discuss your ideas with mi team! It was the exploratory analysis Enrique.

Best Regards!

so, you can try the two models (11 and 4 factors) in a confirmatory analysis and see which one get the best fit. 

Best Regards!

I agree with Enrique: a CFA with the already validated 5-factor model, in your case, would be best. If and only if the CFA doesn't fit and there is no obvious solution to the problem, run the EFA again... but being aware of a couple of things: 

1) If you determined the number of factors to extract in the EFA using eigenvalues, you very, very probably overestimated their numerosity (I suggest Parallel analysis instead, but even simple screeplot inspection is much better). 

2) If you have highly-correlated factors, it may be useful to "force" the number of factors on a lower number and see if those factors merge into one. 

3) Factors with less than 3 high saturations are usually not useful, and may be discarded (even at the cost of removing the items from your scale). 

Of course, if your CFA fails and you have to modify the model with EFA / item removal, you will need a second sample to run a CFA again and thus confirm your modifications.

I agree with Marcelo, never you trust in eigenvalues. You "must to" use Scree test and better a Parallel Analysis. At least four items by factor (in EFA) is recomended. Also, you should consider the meaningfull, explained variance percentage by factor. And dont hesitate to delete items: bidimensional, low loading

I'm so courious about the difference of language & culturer..Better to validate and do some pilot test