@sumelic While I can't offer the statistical data you're looking for, I can point out that the data you present doesn't point to the opposite conclusion. In Rovai 2015, the difference between 9.7% and 12.5% is statistically insignificant. Both Kantor's data are touching the noize floor, and the seemingly two-fold difference must be disregarded.
Most importantly, the two papers you cite are open to be weaponised to argue for two diametically opposite conclusions regarding the short /ŭ/. In Rovai's Delos data, 39 out of 39 instances of /ŭ/ are transcribed as Gr. <Ο>; in Kantor's data, 169/188 or 89.89% are transcribed as <ΟΥ>, and only 18/188 or 9.57% as <Ο>. What is an uninformed observer to make of this? Why, whatever the unscrupulous author wants them to, of course. Or, as someone who probably never said it said,
“There are three kinds of lies: lies, damned lies, and statistics”. (Among other things, a careful observer will notice that the nominative
-us has been excluded from consideration by Kantor).
The importance of the 39/39 number appears when juxtaposed to homogeneous data, i.e. to the transcription of /ĭ/ in the same corpus, and when both are looked at in light of historical-comparative evidence. This evidence tells us that no Romance variety merges /ŭ/ with /ō/ without merging /ĭ/ with /ē/; but Daco-Rumanian together with some Southern Italian varieties do the opposite (
the Lausberg link). Thus, if /ŭ/ was [ʊ] on Delos as the data indicates, then /ĭ/ must have been [ɪ]. Bohemian Czech offers a cross-linguistic parallel, since in it /ĭ/ and /ī/ are clearly differentiated by laxness but /ŭ/ and /ū/ are not. No satsifying explanation of the Delos transcriptions of front vowels can be offered without first establishing the nature of the vowel system of Delian Greek, which like all Koiné Greek must have been characterised by the tendency to raise long front vowels.
(Personally, I would suggest that <Ε> is a graphical shortening for the digraph <EI> to mean “the short vowel that has the quality of the long vowel spelled as <EI>”. Latin itself sometimes uses this spelling for both the long and the short /i/, especially after another <I>, so that the nom. pl. of personal names in <IVS> is spelled as <IE>. This probably indicates dissimilatory lack of raising of the original diphthong /ei/, which regularly merged with /ī/.)
As for Calabrese's paper, it's a theoretical proposal intended to be read by specialists in phonological theroy. It has been dredged up and used in the most irresponsible ways by people who falsely pretend to understand what it says. Not only do they not, the paper itself is filled with self-contradictions. On pp. 79-81, Calabrese goes to great length to state that the elimination of [-ATR] (= lax) vowels [ɪ, ʊ] can result not only in [e, o] but in [i, u] and even [ɛ, ɔ]. He postulates the operation of Negation to explain the former change, and Delinking to explain the latter.
All three reflexes presuppose [ɪ, ʊ] as the starting point. Nowhere in the paper is the existence of the stage [ɪ, ʊ] denied. It's only the chronology that's at issue.
(The “two separate identical shifts” suggestion is plainly untenable, as there are well-established linguistic links between Southern Italy, Africa and Sardinia. Southern Italy is where African innovations penetrate into Italy. These resulted in whole areas of Sardinian-type vocalism, as well as various intermediary systems. Arguably, the actual explanation is that Africa was settled by speakers of Southern Italian varieties of Latin; these and other innovations later came back to roost from Africa).
And there's the rub. While Allen only confidently describes this state of the vowel system for the 4-5th centuries AD, and regards earlier evidence is not conclusive enough, Calabrese instead says the following on p.76 (emphasis mine), basing this exclusively (it seems) on Pompeiian evidence:
Given that the inscriptional records shows that that the replacement of ě with ae begins to occur around the first century AD, we must conclude that the process differentiating the [ATR] values of short and long vowels must have already occurred around this time.
...which places it smack-dab in the middle of Classical Latin. This is not the place to discuss all the ways in which Calabrese's is self-contradictory (in fact it's three different earlier papers glued together). What I'm trying to demonstrate is that people who cherry-pick Calabrese's paper to claim lack of differentiation in laxness (what some modern frameworks describe in terms of [
Advanced
Tongue
Root], not uncontroversially) in Classical Latin can be cherry-picked right back at using that same paper, especially if one actually has enough grounding in theoretical phonology to understand what it says.
Please forgive this reply's argumentative tone, but the situation with that paper is a specimen of ideologically weaponised science used to promote an esthetically-driven language ideology, which is
reactionary to what the OP calls “the Germanisation of the Latin language” and hence can be called
the Italianisation of the Latin language, and I wish this misguided campaign could be stopped for good.
[ˈmɪŋkja, -k͜ça] from Latin mentula(m). The absence of [ɪ] in major Romance languages has caused to appear unlikely its presence in Classical Latin; minor Romance languages and dialects must be studied to have a thorough overview of the fact.