Know-How Exchange : Research/Metrics

Hi,

I assume you mean the reliability of a multi-item construct. For instance, you measure customer satisfaction with 3 questions and want to measure the reliability index; so that you can average the 3 answers into one 'customer satisfaction' number.

The most commonly used reliability index is 'Cronbach's alpha', which basically measures how much the answers to these 3 questions correlate with eachother. People differ slightly on how much is acceptable, using 0.6 to 0.8 as the minimum acceptable. For me, 0.7 is a minimum for acceptable reliability. Note too that Cronbach alpha can be 'too high': if it is above e.g. 0.95, it implies that answers to the three questions are virtually identical. In that case, I would drop some of the questions in later surveys, which reduces respondent fatigue

Dear Omarfdawood

You may also want to consider what it is you're setting out to measure your goals and expectations against:

a competitor,
a national average,
an industry standard?

Any "acceptable reliability index" is also going to hinge on who you are asking the questions of, where, when, in what context, and weighed against what.

But beware. Statistics can be melded, welded, and shunted around to match our expectations, desires, and wishful thinking.
Once you've come up with your index, you must stick to it no matter what, and regardless of your liking or not liking the answers.

I hope this helps.

Gary Bloomer
Wilmington, DE, USA