Undergrad Understanding Yates' correction

[Tyto alba]

I was studying Biostatistics by Zar, the Yates' correction part and stumbled upon this:

... Our need to determine the probability of a calculated X² can be met only approximately by consulting Appendex Table B.1, and our conclusions are not taking place exactly at the level of alpha which we set. This situation would be unfortunate were it not for the fact that the approximation is a very good one, except when df=1. In the case of df=1, it is usually recommended to use Yates correction for continuity.

My doubts:

1. The conculsions are not taking place exactly at the level of alpha- what does this mean?

Do the values of Chi square mentioned in table not correspond to the alpha(probability) indicated? Is the corresponding value actually a range?

1. I also don't understand what it means by 'This situation would be unfortunate were it not for the fact that the approximation is a very good one, except when df=1. In the case of df=1, it is usually recommended to use Yates correction for continuity.'

Why is Yates' correction specifically done when df=1?

 

[MarneMath]

The statistics for the table simply do not have a chi squared distribution. We have a fixed count and hence a discrete distribution whereas the chi-squared is continuous. Nevertheless the chi-squared distribution is a reasonable approximation for the contingency table. I don't agree that you should use Yates correction when the degrees of freedom is one. I argue that you should never use Yates correction.