Which method is more appropriate to be done first: is it EFA or CFA in order to validating an existing instrument? or it is necessary to do both methods regardless of the order?
In order to validate an instrument there are many methods to confirm construct validity, both CFA & EFA suitable in general, which one? Depends on your purpose, if you want to confirm the structure of the instrument depends on a certain theory you use CFA, if you want to explore the construct of the instrument and define its dimension without depending on the theory you use EFA which after do it you judge the instrument construction. For more accurate results you might do them both that every method could provide you special advantages but be aware here to do them by use two separate samples for validation. Usually researcher used to do EFA before CFA. You can use SPSS for EFA and LISREL for CFA. Or FACTOR which is free and easy to use.
In some cases, you will see certain articles have done both EFA and CFA. CFA is a confirmatory technique, whereby it confirms the factor structure based on a theory whereby EFA is more towards exploring the number of factors for a particular instrument. In some cross-cultural studies, both EFA and CFA are done since the instruments are generally translated and hence the number of factors should be explored followed by a confirmatory technique
If you are developing an instrument with still unknown factors even if it is founded on a particular theory, EFA may be more suitable. Because it allows you to explore some latent factors of your instrument. However, if you have an instrument with established factors, CFA is used because you need to confirm the factor structures of that instrument.
you can use any one of them if you want to validate it in the context of the confirmation of the quality of instrument, most popular is to make EFA which provide you information about dimentions (Factors) . EFA provide you Eigen values for factors that is one of determining number of factors depending on them when exceeds 1, and so parallel analysis ...etc.
EFA then show that for example the instrument dimentions are four, which is mentioned by the original instrument developer
If you are using a valid instrument for your study, there is no need to use EFA. Just use CFA to see whether it can also apply to your sample (i.e., to validate an instrument). However, if you add some new items or modify the items to meet the needs of your study, then both EFA and CFA (on a new sample) are required.
First do EFA in SPSS. Then, assess the extracted structure via CFA in AMOS or other appropriate software. You can approve the CFA is fit by the fit indices model and by showing all of the factor loading are greater than 0.50.
Very often instruments are introduced without sufficient and rigorous validity studies or flawed studies. A case in point is the measurement of self construals. Then there is a bandwagon effect and numerous studies use an unproven measure. It is always better to use LISREL or R to check the validity of the measure in your study and to report the validity indicators for your sample (RMSEA, GFI, CFI, chi square). This adds to the confidence with which we can use the measure. This is especially important when a scale developed with college students in North America is used with minority participants or in other countries. My experience shows "valid" scales often show substantial issues of validity.
I agree whit you Mary... that is why we need to do lots of CFA even instruments that reports high validity indices. These testing the internal structure through factor analysis improves the validity of any instrument, at least internally. It is however important to ensure that instruments have external validities.
EFA can be conducted to make an empirical assessment of the dimensionality of the study scale items. Normally, most factors are correlated with many variables, making interpretation difficult.
Dimensionality can be investigated by looking at the loading of each scale item on the factors. The test of unidimensionality is that each summated scale should consist of items loading highly on single factor . If a summed scale is proposed to have multiple dimensions, each dimension should be reflected by a separate factor.
EFA is also a data reduction technique as items with low and ambiguous loading can be dropped from further analysis. Generally, factor loading below 0.4 are considered low, and the associated items should be dropped, whereas factor loading of 0.5 or greater are considered practically significant. Also, any item loading on more than one factor, that is, with a loading score equal to or greater than 0.5 on more than one factor must be dropped from the analysis since cross loading variables makes interpreting the factor difficult.
Then, factor loading and communalities need to be investigated. The communality of a variable represents how much it explains the variance in the factor solution communalities less than 0.5 are too low so such variables can be deleted as they have insufficient contribution to the variance.
Thus items must be eliminated if (a) loadings were less than 0.5 (b) loadings were greater than 0.5 for more than one factor (c) communality was less than 0.4.
i am planning to do a validation study of SGRQ. this tool had validated more than 20 languages. but most of the studies are done convergent and discriminant validly including original development study. There are 4 studies conducted the Factor analysis. out of the four, two studies done only EFA. other ONE , they did EFA and CFA. Other one did bifactor analysis. All factor analysis did not showed the number of factor structures mentioned in the original developmental research. In this case do i need to do both EFA and CFA in my studies? Can I use SPSS AMOUS software to do this analysis?