In literature, the average inter-item correlations in the range of 0.15 to 0.50 is considered as an indicator for an acceptable level of consistency. Why do the high average inter-item correlations affect the consistency? And give poor internal reliability?
1 Answer
Looking at inter-item correlations for reliability is just one part of many for looking at the validity of a scale. Typically, items below 0.15 have poor inter-item correlations, suggesting they're really not that well related to each other and might not be suitable for measuring a single construct. However, items that are above .50 tend to be very similar to each other, almost to the point that they're redundant. There's no point in having two items on a scale that measure the construct in exactly the same way. Rather, you want to measure the depth and different aspects of a specific construct. For this reason, researchers aim to keep the inter-item correlation between .15 and .50. Once the correlation reaches a certain point, it no longer improves the validity of the measure. In fact, it can actually hurt it.
-
2$\begingroup$ May I ask which literature you refer to? (about the inter item correlation level between 0.15 to 0.50) $\endgroup$ Jun 19, 2017 at 6:44
-
$\begingroup$ i akes a similiar question that may be related - can you help?stats.stackexchange.com/questions/405643/… $\endgroup$– user1607May 13, 2019 at 18:44
And give poor internal reliability?
That's not obvious why. Please expand it with reasons/example. 2) Note also thataverage inter-item correlations
are directly related to the standardized Cronbach's alpha which is considered mostly as a "reliability" index. 3) In my own opinion "internal consistency" is a bad term. I would rather use concepts "item-total (or item-construct) homogeneity" (a facet of validity) and "item-item homogeneity" (a facet of reliability). $\endgroup$