I always like to remind myself that nutritional science is a long ways from consistent peer-reviewed, widely accepted theories. What we "know" changes (dramatically!) every decade or so, and there are several groups of scientists that believe very different theories co-existing in the world today. Nutritional science in Europe is different than that in America, is different than that in Monsanto's headquarters, is different than that in the American Heart Association's headquarters, etc, etc. Basically: we are still very firmly in the "we don't know what we don't know" stage of nutritional science. That does not make our beliefs wrong, but it certainly does not make them right.
I think the question you asked is a really important one, and one I wish more people would ask about much science (and most science does have good answers for these questions!). But personally, I haven't seen an acceptable level of rigor, ethics, and consistency in nutritional science to believe any of this science as fact yet.
I want to ask: when "they" say that there is (x ammount) of (nutrient y) in 100g of (vegetable z), how do they account for the differences between vegetables grown on one farm as opposed to another?
I would very confidently say the answer to this is
no. For example, in America this data comes from the FDA. You can read about their
sampling methods here. They are not rigorous, they are subjective:
In so doing, the data base may be designed to consider and select samples based on one or more factors that may impact on nutrient variability, or, preferably, on a combination of such factors. For example, if a data base concentrates on foods selected from one state, the sampling plan might include the collection of samples according to the regions within the state where the food is grown. In addition, because some foods are known to show seasonal variation in total fat content, a relevant data base would likely want to include samples harvested at selected seasons. Furthermore, in some instances, it may be appropriate to sample according to type of processing.
This sampling methodology
assumes the distribution of data is a normal distribution, and that the average has meaning.