3

I would like to use a 5-item subscale to measure the effects of a behavioural intervention program on the use of behavioural strategies. I obtain significant results when I analyse it in models however the reliability of this measure is very low at .5 and the mean inter-item correlation is below .2. Can I use this measure or should I exclude it from all subsequent analyses?

Thank you.

Melissa Duncombe
  • 79
  • 1
    I assume that you measured reliability using something like Cronbach's $\alpha$ - which actually does not measure reliability, but internal consistency. A low internal consistency does not necessarily mean that there is low reliability. For example, in formative measures, you can even get negative internal consistencies, while the measure itself can be highly reliable. Could you post your items? – Felix S Jun 27 '12 at 07:27
  • The items are based on a 5-point likert scale ranging from never to always and are from a parent-rated questionnaire and include "you ignore your child when he/she is misbehaving", "you take away privileges or money from your child as a punishment", "you send your child to his/her room as punishment", "you calmly explain to your child why he/she has done something wrong", "you use time out as a punishment" and " you give your child extra chores as a punishment". Thank you. – Melissa Duncombe Jun 27 '12 at 09:22
  • 1
    @Felix, alpha is still considered among measures of reliability. Internal consistency is a term better to avoid, because some take it as inter-item consistency (a facet of reliability) and some as item-total consistency (a facet of validity). Melissa, you are free to choose to use your items as one scale or not. "Apples" and "Oranges" are quite dissimilar, but both represent construct "fruits". – ttnphns Jun 27 '12 at 17:07
  • @ttnphns: You're right; my first comment was somewhat overstated - alpha is a proxy for one aspect of reliability, but it misses others aspects. And there are cases, where perfectly reliable measures have zero or negative alpha. Concerning the items, I could well imagine that these are alternative approaches of parenting styles (which are not interchangeable), and therefore I would take a look into the "formative measure" approach. (One key problem is that it is hard to determine the reliability empirically - you would have to build a MIMIC model, and things get complicated ...) – Felix S Jun 28 '12 at 07:27

0 Answers0