First, if you have 5 subscales, the internal consistency will be testes on each subscale not on the whole instrument. Then, if you use Alpha Cronbach you will need a minimum of 0.8 for diagnostic reliability. If you have only exploratory purposes, you can accept between 0.6 and 0.8, not below 0.6. There is no minimum item-scale correlation coefficient. Check how the scale will be modified if an item is deleted. The inter-items correletion coefficient should not be too high (over 0.8) because this can be a problem of multicollinearity. Remember that Cronbach is an indicator for scale reliability, not validity

Dear Cristian,

I think that both the internal consistency of each subscale and the whole questionnaire are important as we almost see both of them in psychometric reports.

You mentioned that "Check how the scale will be modified if an item is deleted". But how in terms of statistics?

Check if the internal consistency increase when the item is deleted. If the deletion of an item conducts to increasing consistency, that item should me removed. No, internal concictency shoud be checked only at scale level, not at the whole intrument because it cannot reveals a mutidimensional structure. For that I recommand to conduct an EFA or CFA study.

The squared multiple correlation column is something like the "power" of item in tour model. If the value is not significant, the item should be removed. You do not have a statistic test for the effect of item removal if that was your main issue.

- The question is that how much increase in the internal consistency is considerable?!

- There is a sleep evaluating questionnaire with 26 items categorized to 6 scales each evaluate specific type of sleep disorder. Or there is QOL questionnaire with 34 items categorized to 5 dimensions. I have seen internal consistency reported for the whole questionnaire as well as each scale in different validation reports on these measures;

- Sleep Medicine 10 (2009) 457–463 Table 2

- Qual Life Res (2007) 16:991–996 Table 2

- BioPsychoSocial Medicine 2007, 1:6 Table 2

etc.

Nunnaly (1978) has indicated 0.7 to be an acceptable reliability coefficient. . If overall alpha value >0.7, keep all the items in the construct regardless of "Cronbach's Alpha if Item Deleted" value. But check "Corrected item-total correlations" which are the correlations between scores on each item and the total scale scores. If all values >0.5, keep them all. If there is any item(s) with lower "Corrected item-total correlations", consider to remove it/them.

It is suggested values above 0.6 (Hair, Black, Babin, & Anderson, 2010).

Hair, J., Black, W., Babin, B., & Anderson, R. (2010). Multivariate Data Analysis. Upper Saddle River.

There are many different but close suggestions in the literature. One cannot say that it is not scientific approach if you consider 0.5 as acceptable.

Dear colleagues,

All the comments are right and interesting, but you must not forget that validity is the most important property of the items' scores. For this reason, be very careful removing items only according to non-adequate reliability results. It is important to analyze conjointly reliability and validity evidences. The optimal scenario is one where both reliability and validity are adequate with the minimum number of items. It is obvious that the quality of answers depends on the length of the instrument. More items, more reliability, but at the same time more probability to burden the respondents. In summary, Item analysis must be guided not only by empirical (operational) procedures, but theory-driven (validity) considerations.

I am sure that you will find the better way to improve the instrument.

All the best.

Ufuk

According to literature, the corrected item-total correlation should be > 0.3 (Maltby 2007, Brzoska 2010).

I agree with Albert Sese. Regarding the remaining discussion: There are only a few strict rules for including / excluding items from a scale, because there are so many factors to consider. One of the strict rules is: Exclude (or reverse) items with corrected item-total correlations of

Thank you Chalamala Srinivas ·

We have just published an article about scale purification, i.e. the process of eliminating items from multi-item scales. We have used the example of SCM, but our framework can be applied to any other discipline. Download: https://doi.org/10.1108/SCM-07-2016-0230 (or request via my ResearchGate page).

According to Pedhazur & Schmelkin (1991), anything above.3 should be acceptable. But it's important to remember that when you remove an item, you lose the measurability capacity that the item would offer.

The rule of thumb indicated that Cronbach‟s Alpha of coefficient 0.7 to < 0.8 is good and 0. 8 to < 0.9 is very good. And inter item and item the total validity can be in range of 0.20 for exploratory study and for others 0.30 is acceptable.

This is just a rule of thumb, indeed, not a strict cutoff value. It is important to note that Cronbach's alpha (or better: rho_T, as it is called in more recent publications) requires tau-equivalent measurement models. However, most measurement models in structural equation modeling are not tau equivalent, but congeneric. Therefore, you should not use/report rho_T (= "Cronbach's alpha"), but, instead, congeneric reliability (in the older literature called "composite" reliability). Also here, don't use these values as cutoff values, but only as a rule of thumb: See: https://en.wikipedia.org/wiki/Congeneric_reliability

This is very good information. Thank you all

Thank you for your kind and useful information.

The literature source for an inter-item correlation of 0.15 to 0.50 is:

https://doi.org/10.1111/j.1467-6494.1986.tb00391.x

and is cited here:

http://dx.doi.org/10.1037/1040-3590.7.3.309

Hi Saima Nazir . Do you have a literature source for this answer " item to total validity can be in range of 0.20 for exploratory study and for others 0.30 is acceptable. "? Especially the value of 0.2 is not verifiable. At least, I cannot find it. Thank you for your help in advance!

@Iris: Such values should never be understood as cutoff values (thresholds).

Cronbach's Alpha is frequently misunderstood. Higher does not automatically mean better. It is a measure of internal consistency, but is frequently mistaken for reliability. If you have a CA of 1 it does not mean you have perfect reliability, it simply means that you have useless items ;-)

According to Cristobal et al. (2007), the items with corrected item- total correlation lower than 0.30 are not acceptable. However, for exploratory study 0.20 is acceptable value for inter item and item-the total correlation.

@ Iris Pöschl

In the paper by Steyn et al. (2005) Article Development and validation of a questionnaire to test knowle...

"Each item ideally should correlate with the total test score. A minimum of 0.2 has been proposed as the cutoff value below which items should be discarded [33]."

33. Kline P. A handbook of test construction. London: Methuen; 1986.

Hi Mr Iris Poschl

I have mentioned the authors and the full reference for these values is;

Cristobal, E., Flavian, C., & Guinaliu, M. (2007). Perceived e-service quality (PeSQ) Measurement validation and effects on consumer satisfaction and web site loyalty. Managing service quality: An international journal, 17(3), 317-340.

thank you very much, the answers given above were very helpful.

Thank you Nazir for your useful information on corrected item- total correlation.

Dear Nazir,

" According to Cristobal et al. (2007), the items with corrected item- total correlation lower than 0.30 are not acceptable. However, for exploratory study 0.20 is acceptable value for inter item and item-the total correlation. " Can I have reference please for exploratory study where 0.20 is acceptable value for inter item and item- the total correlation. Thank you in advance.