Back in the day, I was running simulations for my degree. And I obtained a series of perfectly acceptable results, indeed slightly better than the accepted literature, but sill well within reason. There was, nonetheless, a distinctive pattern in my results – a periodic, much better than average, performance, depending on parameter tuning. Therefore I came up with a somewhat plausible explanation and brought it to my advisor. After hearing me out for half a minute they basically told me: the data are fine for what you need. If you want to prove your idea, you need a whole lot more than that.

And they were very right.

Data come with a purpose. They are collected for a reason. And their acceptable and reasonable degree of accuracy varies with that purpose and depends on that reason. The fact that they happen to be included within a peer reviewed piece, by a serious author, and published in a much respected publication does not mean that one can just take them and use them as “facts”.What may be perfectly fair data (there is, to my knowledge, a non indifferent friction on this point – but this is definitely not a field I am confident enough with to argue the merits of this particular piece) in so far as showing a general trend of decline in violent crime across the centuries, may be completely unacceptable in so far as arguing that such specific oscillation occurred, let alone speculate on its causes.

This is not meant as criticism of you, or your question. But the practice of lifting data from published studies and re-purposing them in order to support some very different argument is unfortunately quite common, and it's probably worth being mindful of this sort of process. Aside from the fact that – in my modest opinion at least – it's quite shoddy; it's a practice which has been used for less than dignified purposes.

Such criticism anyways does not apply to the present study. The author – Manuel Eisner, Univ. Cambridge, PhD, Professor of Criminology – does a pretty good job at explaining what data he is using and what's their purpose.

It is my opinion that, if the author's contention was that Italian murder rates increased substantially – for some reason – from mid XVIII Century to early XIX Century, they would have either sought to provide more and better estimates, or they would not have argued that. And indeed they don't. The author is attempting to draw a more general conclusion here. As Eisner explains, describing the purpose of his collation of many different studies, of different accuracy, from various periods and European regions:

Taken together, these data first confirm the Europe-wide massive drop – roughly by a factor of 10:1 to 50:1 over the period from the fifteenth to the nineteenth century – first observed by Gurr on the basis of English data (1981).

The author then proceeds to illustrate the origin of the data he used in his computation, going as far as to point out that

I included all studies [...] that at least present information on the respective territory covered, the type of source, and the time period covered. If it was sufficiently clear how the information had been gathered, I ignored occasional warnings by the author against using homicide counts for computing rates.

And further points out that

Althought somewhat less thoroughly covered than the other areas, Italy is the fifth region with a series of studies that permit empirically based extrapolations. Studies of medieval and renaissance cities include Bologna (Blanshei 1981, 1982), Florence (Becker 1976), and Venice (Ruggiero 1978, 1980). Romani (1980) has examined court records in late sixteenth century Mantova, and a fascinating study by Blastenbrei (1995) analyzes wounding reports by medical professionals and judicial records in late sixteenth-century Rome. Finally, a series of studies on Padova (Zorzi 1989), Citra (Panico 1991), Sardinia (Donneddu 1991), Tuscany (Sardi 1991), and Rome yields another cluster of estimates for the late eighteenth and early nineteenth century before the onset of national statistics.

Meaning that, besides data being sparse, they are also not geographically homogeneous. Which means that the contribution of regional and particular differences to any fluctuation internal to the Italian series can't certainly be discounted (if any such difference were to emerge within statistical error). The author itself later suggests a higher degree of violent behavior in the Southern regions – or, better, a more belated decline from late-medieval levels of interpersonal violence. Especially he maintains that, during mid to late XIX Century

In Italy, homicide rates were higher in the rural south with its low literacy rates than in the more industrialized north.

But that's really not something which we need to indulge ourselves too much with, being a whole different contention from the specific one we are seeking confirmation for. As the author again explains that

my interest is not to compare differences in the magnitude of 50 or 60 percent over time [my highlighting], but a ten- to possibly 50-fold decline. Therefore quite considerable inaccuracy […] can be accepted.

If you check the article cited as source for the graph you posted, you can see the author's original presentation of the data (which I obviously can't reproduce here) for pre-statistical records, which helps better conveying the relative sparsity of the data and the meaning of the series thus produced. While the data reproduced in the graph you linked, are only compiled to allow a direct comparison with the original study by Gurr (1981) for England.

Specifically, accounting for upper an lower estimates of population – by the well established practice of eye-interpolation – I estimate for the interval you are interested in, an uncertainty of about a factor 10 with regards to Italy.

The overall picture is consistent with what the author claims:

a continent-wide decline of serious interpersonal violence.

The author also agrees with my estimate, observing that

for each century, the estimates show a large degree of dispersion, with a ration between the highest and the lowest bundle of estimates typically being around 1:10. […] Based on our knowledge of the large local variability of homicide rates in present society, we should not expect anything else when working with historical data.

Which, in some way, answers your question. Not really whether such a fluctuation has a plausible cause, or whether it actually occurred, but simply whether the available data call for us to make a speculative assumption. And, it seems to me that they don't.

 

Eisner, Manuel. “Long-Term Historical Trends in Violent Crime.” Crime and Justice, vol. 30, 2003, pp. 83–142. JSTOR, www.jstor.org/stable/1147697. Accessed 3 May 2020.

One minor edit for clarification: where the author simply refers to Citra, he means the Principato Citra, a region of the Kingdom of Naples, of present day Salerno.