I think you'll need to show the complete output for things to make sense. For example, are you calculating item-total correlations for the 18 items all toghether? Which is your case?

1. Three scales which are alternative measures for the same latent trait?

2. Three scales that represent three primary order factors under a second order factor?

3. Three scales that represent completely different latent traits that are correlated?

4. Three scales that represent any other kind of different orthogonal structure or other exotic types?

The increase in your alpha and item total-correlation due to the higher number of subjects is not that obvious - not sure if that's what you mean. The change might be due to distribution, variability, etc that changed.

So... another thing that has to be clear is your objective by testing reliability. Have you checked dimensionality before? Funny thing, I tend to check on dimensionality and factorial structure first, before trying to play around with reliability (internal consistency). Remember that Cronbach's alpha and other reliability measures depend on the scale representing a single factor - except for the Omega.

Thank you Pedro for your insightful comments. In my case, the three scales measure three distinct concepts and the EFA suggests three different oblique factors. However, I do not find any possibility for a higher order factor. I am calculating corrected item-total correlations for each scale separately, not for all 18 items together.

If possible, can you please clarify and elaborate a little more on your statement "The change might be due to distribution, variability, etc that change?

My point, were 200- odd participants less for a EFA of this sort? So when the no of participants increased to 433, the reliability analyses gave the real picture.

I have checked the data distribution is approximately normal and no errant outlier and influential case.