# Implications of INL on the accuracy and resolution of an ADC

I was trying to understand the meaning of various specs given in a typical ADC datasheet, and I came across this article. Now, if I understand correctly, it seems to me that the accuracy of the ADC in bits can be taken as $n - \log_2{{e}}$ bits, where $n$ is the number of output bits of the device (its resolution?) and $e$ is the INL (integral non-linearity) error in terms of LSB.

My questions are:

1. Is my understanding correct? If not, what is the correct explanation?
2. If a 16-bit ADC has a INL of $\pm 4$ LSB, can we say that the accuracy of the ADC is 14-bit, and will have to ignore the lowest 2 bits for most practical purposes?
3. Here is a 16-bit ADC with a typical INL of $\pm 0.4$ LSB. Does this mean the device has a typical accuracy of more than 17-bits? How is this possible? Or does it mean the device typically has an accuracy of over 17 bits, but since it has only 16 output bits, it has to truncate the data to a lower resolution?