This is probably a sloppy way to see it, but I wonder if the idea of longitudinal datasets (to be used in ggplot2, for instance), can be exlpained as complying with the criteria of data in a 1st normal form. I know the latter comes from the relational database field, but I would see entities within a db that is 1st normal as longitudinal dataframes. Does it make sense?
Wide/fat/short and long/skinny/tall are both valid representations of multiply sampled data, each having their strengths and weaknesses. Both of these representations are valid longitudinal data sets in R as well as tables in relational databases.
Some argue that one form or the other violates database normalization concepts, but I don't see it. For example, some argue that the long/tall form repeats or duplicates keys (a violation of 1NF), but I would argue that the true row key in the long form is really a larger composite key (i.e. that includes another column), and thus not in violation of 1st normal form as composite keys don't violate normalization. In wide format, what is logically one constraint over sampled data has to be replicated across multiple columns, which is a duplication of sorts, though not a violation of 1st normal form.
In short, we can see both wide and long representation in databases without violating 1st normal form.