Cronbach's alpha is essentially the mean of all possible split-half reliability coefficients. So if you calculate alpha, you shouldn't bother with split-half. 

My professor would want me to do several reliability tests. However, my rrl just used chronbach's on their reliabiliy so I also dont find the need to use the split half. But if ever he would still insist, would it not affect my CFA since the scale would be split into two?

Sure, if you want to divide the scale in half & treat each half as a separate measure, then you should calculate the reliability for each half. Just run alpha for each of the two subsets of items - that's my guess.

Well, technically ryff's scale cannot be treated as separate measures since its well-being. So if every i'll do a split-half and my measures will be separated, will it not affect the CFA? Should the CFA be separated as well?

Agree with other colleagues that chronbach's alpha  is the best for your scale testing. It include all the possible split halfs.

Cronbach's alpha is a problematic measure. It presupposes tau-equivalency, or a Rasch model, meaning that all loadings of the items on the factor to be measured have to be equal. If that is the case, alpha underestimates reliability. If loadings are not equal - as they typically are not - it is not possible to know if reliabitly is under- or overestimated. This is explained in papers by Tenko Raykov and others. A good treatment of reliability and its different measures is found in Raykov and Marcoulides "Introduction to Psychometric theory" (2011).

he instructed me to do chronbach's, split half, and alternate forms. Thoughts on this?

That may depend on your relation. Is he capable of learning something new, and can he accept that it comes from you?

no. he is very authoritative im afraid. so i really dont have any choice. By the way, how long can i administer the other form in alternate form reliability?

If I understand you correctly, you do not have two alternate forms of the Ryff scale. What you are doing is to take the Ryff scale and split it in two. If you do that, then of course you can calculate the correlation between the two halves, treating each as a separate measure. And that's all alternate-forms reliability is, really.

The split-half method simply takes the correlation between the two halves and applies the Spearman-Brown prophecy formula. This compensates for the fact that the halves are shorter than the whole - a longer measure will be more reliable. So you (a) double the correlation, (b) add 1 to the correlation, and then (c) divide the result from step (a) by the result from step (b). This will NOT be the reliability of either half taken alone, though.

If your advisor is being quite dogmatic, then just run the analyses s/he demands. But don't be surprised to find that Cronbach's alpha comes out very close to the split-half value. That happens most of the time.

BTW: It's Cronbach, not Chronbach!

He instructed me to do an alternate form reliability using the 42-item test with that of the 18-item test. What I understood in alternate forms is that the 2 instruments should be administered days apart. I was recommended by other researchers here to just extract the answers for the 18-item instrument from the 42-items since they have the same items. Is it advisable? 

Oh, thank you for the correction btw!

If you administer them days apart, then you are not just looking at alternate-form reliability; there's also a test-retest element. I would say to do both at once.

Since the items for one scale are embedded in the other, it does seem silly to have people respond to both - though I suppose it might make a difference. (Sometimes my inner reaction to having answered one question may affect how I respond to the next one.) What the other researchers advise isn't quite the same thing. I'd be careful here; I don't think your adviser sounds like he'll be happy with the solution.

If he is at all open to suggestion, what you could propose is this: Administer just the 42-item version. Calculate (a) the correlation between the 18 items also found on the short scale and the total 42-item scale, and also (b) the correlation between the 18 short form items and the remaining 24 items.

However, neither of those findings establishes the reliability of either scale. For that, I'd still recommend Cronbach's alpha or possibly split-half.

It is not clear to me why you need to test the validity of the Ryff scales. They are used in international research for years and their validity has been evidenced in many ways. 

Having said this, I agree that you should do what your supervisor suggests for the needs of the study. Years ago, I was instructed to use split half AND Cronback in a Greek journal, I refused to do so, and they refused to publish the article (for this and other reasons...). 

Finally, as I am familiar with the Ryff scales, I would advise you to be cautious about what you test if you use the 18- and the 42-item scales. In the former scales, the reliability has been proven to be rather low, and thus, researchers prefer to use this as one scale measuring well-being. In the latter scales, the reliability of the six scales is also acceptable and thus you can use them to measure the six separate constructs. 

If you decide to use a correlation between the 42- and the 18-item scales, try to define beforehand what association this correlation will represent. 

I suggest you this study (own authorship) dealing with psychological wellbeing for organizing and presenting results and/of internal consistency issues.

You can report both coefficients (Cronbach's Alpha and two-half reliability statistics) in the same paragraph. It will give a better support for your results.

Best,

Sergio U.

Article The Satisfaction With Life Scale (SWLS): Psychometric Proper...

Stephen Joy

so the correlation I'll do between the 42 and 18 items are not alternative forms reliability?

Georgios Vleioras,

I initially was not going to do psychometric properties on my study however, the class has a kind of disagreement with the adviser. Nevertheless, thank you for the info!

Sergio Useche

thank you for the recommendations! Will gladly note it