I am doing a research using Theory of Planned Behavior to study tourist intention. In my study, Perceived Behavior Control (PBC), Subjective Norms(SN), Attitude (ATT) are Independent variables. Intention is dependent variable (IN). And Moderator. In PBC, SN, ATT there are 2-3 subsets umbrella under each of this independent variable. There are minimum 2 questions in each subset.
I got 40 responds for the pilot test. When I run the preliminary test to check reliabilty there is an issue arise. There is one variable that its Cronbach's (0.228)
Clearly, there is a difference between normal and reliable data.
Normality and reliability are two separate concepts with two different measurement methods.
The low alpha tells you the items making up that composite scale are not highly intercorrelated. Check to make sure they all correlate positively with one another. If one of them is correlated negatively with most of the others, either it doesn't belong in the scale (because it measures something else), or it needs to be reverse-coded (because it was asked in the negative compared to how the others where phrased).
If one (or more) of the variables in the scale correlates near zero with most of the others, and/or has a low item to total correlation, it (or they) should be dropped from the scale as not really measuring the same thing as the others. If most of the items correlate near zero with one another, the items are not measuring the same thing and cannot be combined into a meaningful scale. That's the whole point of using alpha to examine the internal consistency of the items in the scale.
A high mean for the scale is not definitively diagnostic about the scale. Sometimes it just says most of the cases in your sample have high levels of whatever the scale is measuring. In this case, however, it could be that the questions are worded poorly, and induce the respondents to pick a high value for their response. The items would then have low variance, and much of their variance would be just random noise, or error, not true-score variance. That would tend to produce low inter-item correlations, and a low alpha. Again, alpha would be doing what you want it to do, by identifying a low-reliability scale (in this case, due to the low reliabilities of the items in that scale).
- Badges
- Science topic
- Similar topics
- Chemistry
- Pharmacology