Ethnicity estimates are intended to infer your ethnic origins at least 500 years ago, or even longer. So comparing them to what is known about an individual's ancestry from documentary research is unlikely to cover that period. Could that be a get-out clause for Ancestry in the event that an individual's ethnicity prediction is not what they expect? I don't know, but I often see people posting that, "My grandfather was from Ireland" or similar. Far too recent to infer anything in relation to the ethnicity estimates.
The methodology is based on reference populations. Ancestry are considered to have the best coverage, but the information they have online is illuminating. Reference populations average about 800 people per group, although obviously some will be higher and some lower. The reference population for Ireland is 794 individuals. For England and North Western Europe combined it is 1,907. I haven't bothered working that out as a percentage of the population of each of those regions, but it will be insignificant. The white paper describing their ethnicity estimation is full of references to "estimates", "inferences", "assumptions" and "probabilities". Prediction accuracy for a single-origin individual is estimated (that word again) to vary between 27 and 100 percent depending on the assigned region. Yes, you read that right - 27 percent. For people with multi-region estimates, the predicted results vary even further. Someone with four grandparents all originating from the same country is likely to have a far more accurate prediction than an individual whose grandparents originate from four different regions. For a person attributed with some English and North Western European ethnicity combined with one or more additional regions, their percentage of ethnicity for England and North Western Europe as stated in the estimate could vary by up to about 50 percent from the stated percentage, according to the white paper.
The ethnicity estimate you are given is a prediction of probabilities, not a statement of fact. It is informative and maybe educational to an extent, but with the caveat that there is no guarantee of reliability.
It likely sells DNA tests to the average punter, but for family history research the process of segment matching to identify family relationships is much more useful and reliable.
