In advocating an approach based on structural characteristics rather than whole games, Griffiths and Auer [11••] followed methodological advances in identifying taxonomies of the structural characteristics associated with risk . They found five relatively specific game characteristics as potentially relevant, based on previous research: event frequency, bet frequency, event duration, in-play gambling, and payout interval. These dimensions are motivated by a general impression that “the speed and frequency of the game” (p. 1) are the main force behind the potential for gambling-related harms. The authors point out that the frequency of an “event” (the entity that forms the identity of winning and losing bets, such as the result of a soccer match or the drawing of lottery numbers) may not match the frequency of betting, as one may buy more than one lottery ticket or make multiple distinct bets on a single soccer match. Further, events with relatively long duration may foster “in-play gambling” , such as betting on the result of a particular corner kick. Finally, payout interval is used to refer both to the time that elapses from the completion of the event until the payment is made and also the schedule of reinforcement according to which bets are won. As has been noted many times, behaviors reinforced on variable ratio schedules , which are characteristic of EGM play, are particularly robust and resistant to extinction.
These characteristics, as a set, leave a number of questions open. Event frequency and bet frequency appear to be usefully distinguished in lotteries, in which one may buy multiple tickets for the same drawing, but not for other games such as poker or sporting event outcomes. For in-game betting, the choice to designate a corner kick as something other than an event may be arbitrary. While it may be reasonable in the case of soccer to designate the higher-order event (the whole soccer game) as the true “event” in distinction to the smaller unit, there are other settings where the designation is less clear. Is the “event” in poker the single hand or the outcome of a sequence of hands across an evening, or a tournament? Also, delay of reinforcement is generally viewed as distinct from the interval associated with a schedule [15, 16], such that the “payout interval” characteristic may need further elucidation and, possibly, division into separate concepts.
Interestingly, the earlier research [10••], despite a primary focus on games as a whole, did analyze games in terms of two dimensions previously discussed in the literature, namely pure chance versus combined skill and chance  and bank games versus social game . The bank versus social dimension reflects that social games have expected value = 0 because no commercial interest takes a portion of the bets, whereas bank games have expected value <0. This pair of potential dimensions is not especially compelling, as the chance-social cell contains only coin tossing and rock-paper-scissors, which are not strongly associated with actual gambling, and the combined-social cell contains poker, which is undeniably serious gambling, plus the games of backgammon, bridge, and rummy, which are not. In short, the category of social games serves primarily to separate poker from all other forms of gambling (this may be related to the fact that  was addressed solely to poker). Among the bank games, blackjack, sports betting, and horserace betting are considered to contain an element of skill, and a long list of other games (including some poker variants) are considered pure chance games. For these reasons, it is not surprising that the paper’s conclusions rested on whole-game considerations rather than specific characteristics.
It is telling that both of these recent papers [10••, 11••] explore potential systems of conceptualizing relevant game characteristics that, in addition to their significant respective limitations, have little overlap when viewed simultaneously (as the sole point of congruence, the concept of event duration appears similar to the designation of deferred gambling assigned to one empirical cluster comprising several diverse games by [10••]). It is appropriate to conclude that such systems are in the early stages of development.
Rachlin and colleagues [[19••]; updating ] provide an example of theoretically grounded models that take account of structural characteristics of games and the literature of temporal discounting. This model assumes that gambling events are experienced and encoded as a string of losses concluding with a single win. Given this model structure, wins are always coded as recent and therefore memorable, and long strings of losses include many which are remote and therefore dim memories. Because the reinforcing value of wins is always recent and strong, and the punishing value of losses is often substantially diminished by the delay from their occurrence to their encoding, gambling in this model is always maintained at greater rates than its normative value would suggest. Assuming further that cumulative gains and losses are calculated following each win, long losing streaks lose their punishing value by the time from the first bet until the eventual win, whereas short losing streaks with favorable resolutions by a quick win are correspondingly advantaged in their reinforcement value. Furthermore, individuals who discount past events more strongly will show this effect more strongly than those with lesser delay discounting. In fact, problem gamblers demonstrate greater delay discounting than nonproblem (but still frequent) gamblers .