Randal Collins' "Violence" is a theory of what happens when humans (in individuals or groups small enough to be individually interacted with) find themselves in potentially violent situations, what physical actions they take there, and what emotions take them from one place to another. Why they find themselves in potentially violent situations in the first place (why wars start, say) is more or less entirely placed to the side, but that is fine - there are already plenty of books about why wars start, or whatever, and this is book is already an ambitious one. He is also only interested in direct, physical violence in which one human being seeks to physically injury or kill another with their own body or tools directly held by them. (Sorry, SJWs and libertarians; neither hate speech nor theft count as violence.)
Collins' philosophical anthropology, more fully elaborated in "Interaction Ritual Chains" (which I highly recommend) is a highly, almost hypertrophically, socially oriented one. What people want (at a basic, fundamental level) is emotional energy, which they get from holding common attention with others in ritualized ways, and which is invested in the objects of this ritual attention. For an example of a very straightforward and indeed intentionally designed interaction ritual, consider how reciting the pledge of allegiance creates emotional attachments to the flag, that can then be met with violence when the sacred object is disrespected. The same thing applies to sports teams, popular bands, and actual religious symbols, and in a comparatively attentuated (but more ubiquitous) form in common objects of attention that are not worshipped at mass gatherings, but in smaller, less emotionally intense interactions. Even introverted and iconoclastic individuals are constantly obsessed with the objects of attention in their subculture, and in receiving attention from their fellows. And all of this is possible, moreover, because humans are intensely aware of the feelings of people around them, and subconsciously want to be in sync with them.
This is Collins' philosophical anthropology; his evidence base is mixed, but overwhelmingly consists of individual reports, photographs, and videos. (Not all of this evidence is ideal, but I will address that later.) This book is a product of the WorldStar era; we can directly observe to the second people's reactions and their actual behavior, far more so than before photography and especially video. And the philosophical anthropology is marshaled to explain the data: people overwhelmingly avoid conflict. Most situations are not violent, most situations in which violence is threatened come to nothing, most situations in which violence happens, happens in a very underwhelming way, and most people present do not participate. The human tendency to seek emotional rapport with each other is too strong and typically wins out. Collins looks at three broad classes of violence: the "forward panic" of mobs and cowboy cops, "heroic combat" such as between boxers or duelists, and the "cold" violence of specialists such as flying aces, snipers, and contract killers.
"Forward panic" describes a specific pathway: a sustained period of adrenal stress and tunnel vision which is presented with the sudden opportunity for resolution with a weak opponent. This is the most typical kind of violence - unexpected, incompetent, and uneven. Soldiers fire wildly and inaccurately, cops or protestors kick men down on the ground. This is matched by a certain stance by victims - crumpled, accepting of violence, fleeing wildly or lying fetal, even when a moment before they were engaged in the same bluster as their opponents; the violence of forward panic usually only begins after one side has folded, and among the victors, only a few engage in it, the others cheering them on or looking about as in a stupor.
Of all the forms of violence Collins looks at, this is the most common, and also the form that most fits his main thesis, which is that it is the logic of the immediate situation - rather than that of the persons or broader social structure - that is the determinant of how things play out. (In this moment of (rightly) highly politicized discussion of police killlings it's worth emphasizing that Collins really does see police violence as unexceptional and driven by the same factors as every other kind of sordid violence.) The other forms of violence fit less easily into this broad characterization, because they draw on particular kinds of participants and particular kinds of social institutions or subcultures.
The staged, fair violence of "heroes" involves confrontation between evenly matched fighters between an audience. This is sometimes explicitly entertainment - medieval jousts, MMA fighting rings - but also includes real attempts to hurt each other by people who are really angry at each other, such as in aristocratic duels or in high school fights of the "fight! fight! fight!" sort; either way, this kind of violence follows certain rules. This is the only form of violence in which participants are "evenly" matched beyond differences of size or skill (and even too great a difference in that would be regarded as disqualifying); and there are limits to the kinds of violence that can be employed. (Collins gives the examples of a Mike Tyson's biting his opponent's ear off, which inspired rage from his opponent and from the audience even though it was far less objectively harmful than many boxing injuries; and of a high school fight between two girls in which a larger boy entered to wail on one of the combatants and was chased by the whole audience and eventually turned over to the police.) The winner gains prestige especially from these fights, but even the losers have been the focus of sustained crowd attention, and their status as members of the fighting elite is sustained; this, in particular, allows for this form of violence to settle for the settling of grudges. When participating in this kind of violence serves as marker of more generalized (e.g., class) elite status, boundary policing may occur - that only a samurai may wield a sword, or that only a gentleman may challenge another to a duel.
The "cold" violence of solitary specialists has the least obvious audience, as well as the least obvious emotional involvement. The sniper or hitman, unlike the hero or man in forward panic, does not experience tunnel vision, but patient analytical involvement in his techniques. But the audience and the emotional involvement are there after all - they are interested in performing for the elite of others who know and appreciate the techniques, in cultivating their reputation of other specialists, and in their own sense of mastery over the craft of killing. Moreover, they seek to kill without a real combat situation arising; it is this that prevents tunnel vision or forward panic from arising, or from getting too emotionally involved with their victims.
There are a few sections where Collins relies on a few dubious sources, or one source alone; Elijah Anderson's "Code of the Street" (aside from some accounts of violent episodes) forms his basis for one long section, Anderson is a sociologist in good standing but I don't know how theoretically laden or controversial his account is; SLA Marshal (who a friend suggests is unreliable) and David Grossman (in the news now as a peddler of seminars designed to make cops more violent) form the basis of much of the forward panic literature. This, combined with my ignorance of the literature on this subject and (luckily) lack of direct experience, means I'm a poor evaluator of it. My official if tentative recommendation is to read "Interaction Ritual Chains" first, because it's mind-blowing and serves as the theoretical background to this.