Within-session chasing of losses and wins in an online eCasino

Abstract

Chasing refers to the escalation of betting behaviour. It is conventionally seen when losing but can also be seen after wins. Diagnostic and screening items for gambling problems describe chasing as returning ‘another day’ to gamble. However, gamblers may also chase within sessions, and this is particularly relevant in online gambling. This study focused on two expressions of within-session chasing: (1) increasing the bet amount, or (2) a reduced probability of quitting the session, as a function of prior losses or wins. These expressions were examined across five online gambling products: slot machines, probability games, blackjack, video poker, and roulette. Our results showed that gamblers bet more and played longer sessions after immediate losses, but they bet less and played shorter sessions when losing cumulatively. The reversed pattern in the cumulative model may be due to financial constraints. For wins, gamblers bet more after both immediate and cumulative wins, but they also played shorter sessions. Chasing patterns were qualitatively similar by game type—with limited evidence for our hypothesis that chasing would be greatest for slot machines as an established high-risk category. Overall, chasing is multi-faceted, varying across the behavioural expressions, by the immediate or cumulative timeframe of prior outcomes, and by game type.

Introduction

Chasing refers to continued or escalating gambling after losing, as the gambler desperately hopes to recover their losses. As a behavioural ‘symptom’, it is directly observable (e.g., in account-based or ‘behavioural tracking’ data¹), and it is often described as a defining hallmark of the loss of control over gambling, and therefore a key boundary that separates recreational gambling from disordered behaviour². Diagnostic or screening items for chasing typically ask if the gambler “returns another day to get even”³. In gambling studies, this is often termed ‘between-session’ chasing. It is well-recognized that gamblers can also chase within sessions^4,5. For example, a gambler can increase the amount bet, or persist for a longer session^6–10. Like between-session chasing, these actions are likely to exacerbate financial losses and gambling harms. There seems little reason to think that between-session chasing is more maladaptive or ‘symptomatic’ than within-session chasing. It may actually be harder for gamblers to control behaviour within a session, due to strong emotions, or powerful states of immersion associated with fast, continuous gambling formats^11,12. The current study aimed to characterize within-session chasing tendencies in online gamblers in the eCasino of a provincially operated gambling platform in British Columbia, Canada.

Previous studies have used behavioural tracking datasets from online or land-based gambling to measure chasing patterns in various ways^6,13–19. In framing these studies, although gambling wins are relatively rare compared to losses, it is important to note that gamblers might chase after both wins and losses⁵. Between-session chasing can be expressed as the gambler returning to the gambling platform more quickly, or increasing their bet amount in the following session. We have recently examined the time interval between sessions in the eCasino dataset used in the present study. On average, gamblers returned more slowly as a function of the amount lost in the previous session, and returned more quickly as a function of the amount won. Thus, loss chasing was not observed in the aggregate, but win chasing was apparent¹⁹. Earlier work on a European online gambling dataset revealed a key boundary condition in chasing—the recent versus cumulative timeframe of outcomes¹⁶. That study analyzed weekly betting data over an eight-month period. Gamblers who lost in the most recent week reduced bet amount in the subsequent week. However, if their losses were consolidated from the first week up to the most recent week, gamblers increased their subsequent weekly bet amount as a function of the cumulative loss. After winning, gamblers also increased their subsequent weekly bet amount in response to rises in both the recent and cumulative amount won.

Characterization of within-session chasing patterns is more challenging because it requires behavioural analysis on a bet-by-bet level. In a representative sample of 2000 gamblers in land-based US casinos, Narayanan and Manchanda¹⁷ examined chasing as the probability of continuing to play and the amount bet over the course of a session, as a function of losing and winning. Approximately 90% of the casino transactions were from slot machines and around 80% of bets were placed by gamblers enrolled in a card-based loyalty program that enabled tracking. In the analysis of continuing versus quitting, gamblers did not respond reliably to an immediately prior loss, but they were less likely to continue gambling as a function of losses accumulating over the session. For wins, gamblers were less likely to continue after an immediately prior win, but they were more likely to continue as a function of cumulative wins. In the analysis of the amount bet, gamblers raised their bet immediately after a loss, but reduced their bet as a function of cumulative losses. The bet amounts were not shaped by wins of either type.

Salaghe et al.²⁰ also examined bet amount changes in land-based slot machine plays among 42,669 gamblers and found rather different chasing patterns from Narayanan and Manchanda¹⁷. They plotted gambling trajectories across consecutive wins and losses, i.e. ‘streaks’. In a winning streak, gamblers gradually increased the next bet amount after each win, but in a losing streak, gamblers did not significantly change the next bet amount after each loss. Additionally, the gamblers reacted to the cumulative outcome over the full course of the session, increased the next bet amount as a function of both cumulative wins and losses.

A recent study used a behavioural tracking dataset from a single online gambling game called ‘Mystery Arena’: a dice-based and chance-dominant game that was available through a Belgian operator⁶. Like Narayanan and Manchanda¹⁷, they examined chasing in terms of the probability of quitting the session, and the amount bet, and they also operationalized a third expression, the speed of gambling following a loss or win. Chen et al.⁶ compared the effects of an immediately prior win or loss on within-session chasing, but outcome was modelled as a binary variable, without considering outcome magnitude. Overall, the gamblers displayed heterogeneous chasing patterns across these three expressions. In the quit probability, gamblers chased a win more than loss (i.e. they tended to quit more after a loss), whereas on the amount bet and the speed of play, they chased a loss more than a win. Chen et al.⁶ did not include the effects of outcome magnitude and the immediate versus cumulative outcome timeframe.

Overall, prior studies using bet-by-bet gambling data have observed within-session loss chasing patterns on the amount bet^6,17,20, coupled with less consistent evidence for loss chasing in terms of continuing versus quitting, and with varying effects following wins. Clearly, gamblers can chase in one expression while not chasing in another. As well as modeling bet-by-bet data, some other methods have been applied to measure chasing via financial transactions including account deposits^13,14,18. Frequent deposits within a short period of time may reflect binge-like behaviour and an attempt to recover losses. Using a European dataset, Auer and Griffiths¹³ compared five potential markers of chasing in gamblers who were classified by the operator at different risk levels. Frequent money deposits within a session were the best predictor of high-risk among the putative chasing measurements.

Overall, the inconsistencies across these studies are difficult to reconcile due to a range of methodological differences. At a simple level, these studies vary by jurisdiction, by online versus land-based environment, and by game type, which may further result in distinct compositions of gamblers^21,22. The modelling approaches also vary in more complex ways; for example, Chen et al.⁶ examined the effect of binary outcomes, with a focus on the effect of the immediately prior outcome on the next bet. Narayanan and Manchanda¹⁷ and Salaghe et al.²⁰ examined the effects of losses and wins separately, with focuses on the effects of outcome magnitudes and the immediate versus cumulative outcome timeframes. Salaghe et al.²⁰ additionally considered streak effects of successive outcomes.

The present study sought to test the separate effects of wins and losses on behavioural measures of within-session chasing. We focus on two behavioural expressions: the amount bet, and the quit probability. We examine these markers as a function of both losses and wins, and separately modelling effects of the immediately prior loss or win from the cumulatively losing or winning session. Our second aim was to compare chasing tendencies across different gambling products in the eCasino, considering five game categories: slot machines, blackjack, roulette, video poker, and probability games (e.g., pachinko, reactors). Gambling games vary in their structural characteristics²³ (e.g., speed of play, ability to multiply one’s bet), in ways that might bias different forms of chasing²⁴. A typical slot machine is a fast-paced game that offers a limited number of bet options, so slot machine gamblers may be more likely to chase by continuing their session than by escalating their bet amount. In table games such as blackjack and roulette, by contrast, the gambler is able to flexibly reconfigure their bet on each event; for example, they may go ‘all in’, betting whatever funds are in their account on a single event. In these settings, a gambler may be more likely to chase by increasing their bet size than through extending their session.

For the overall chasing patterns, we hypothesize:

H1 (bet amount)

• Loss chasing: Online gamblers will increase their bet amount as a function of the size of the immediately prior loss (H1a) and the cumulative amount lost (H1b)

• Win chasing: Online gamblers will increase their bet amount as a function of the size of the immediately prior win (H1c) and the cumulative amount won (H1d)

H2 (quitting versus continuing)

• Loss chasing: Online gamblers will reduce the quit probability as a function of the size of the immediately prior loss (H2a) and the cumulative amount loss (H2b)

• Win chasing: Online gamblers will reduce the quit probability as a function of the size of the immediately prior win (H2c) and the cumulative amount won (H2d)

For Hypothesis 3, we consider differences between gambling products. We use (online) slot machines as the reference category for these analyses, given a wealth of research indicating that slot machines (and other ‘electronic gambling machines’, EGMs) are among the most high-risk forms of gambling²⁵. We expected slot machine gamblers to chase primarily through continuing (versus quitting) their session, but that other product types may be associated instead with a greater tendency to chase via increasing the bet amount. Specifically, we hypothesized that:

H3 (bet amount)

• Loss chasing: Compared to slot machines, other game categories would be associated with a greater increase in the next bet amount after an immediate loss (H3a) and as a function of the cumulative loss (H3b)

• Win chasing: Compared to slot machines, other game categories would be associated with a greater increase in the next bet amount after an immediate win (H3c) and as a function of the cumulative won (H3d)

H4 (quitting versus continuing)

• Loss chasing: Compared to slot machines, other game categories would be associated with a greater probability of quitting after an immediate loss (H4a) and after losing cumulatively (H4b)

• Win chasing: Compared to slot machines, other game categories would be associated with a greater probability of quitting after an immediate win (H4c) and after winning cumulatively (H4d)

Methods

Data overview

This study used behavioural tracking data obtained from the online gambling website PlayNow.com, which is operated by the British Columbia Lottery Corporation (BCLC) and available to BC residents. In order to access the platform, customers are required to create an account, which enables the website to track their activities over time and generate detailed timestamped records of transactional data, allowing bet-by-bet analyses. The dataset was provided to the research team by the BCLC Data Analytics team in a de-identified format where each gambler was randomly assigned a unique identification code. Ethical approval was granted by the University of British Columbia’s Behavioural Research Ethics Board to store and analyze this secondary dataset for research purposes. The dataset covers the period from 2014-10-01 to 2015-08-31 and includes 527,015,222 individual bets placed by 29,964 unique gamblers.

Analytical sample

Our analysis code is on https://osf.io/9n37d/. Analyses of within-session chasing inherently require bet-by-bet gambling data, whereas other research using behavioural tracking data can aggregate the data by session¹⁹ or week¹⁶. Due to the computational challenges that this volume of data present, we initially define a subgroup of gamblers for analysis¹⁷. We randomly selected 2000 gamblers from 9336 who fell between the 35% and 65% percentile of the distribution of total bets. This analytical sample included 1,913,251 individual bets. This approach of selecting the “typical gamblers” was taken because the higher frequency gamblers (‘whales’) would be the most computationally demanding, while the lower end of the distribution would include many cases with negligible levels of gambling (e.g., trying the platform once) who may not be expected to display chasing.

Data analysis

Variables

Gamblers typically place multiple bets within a session, and they may return to the platform multiple times within a day. The bet-by-bet data can be further organized into sessions. A ‘session’ is a series of bets placed in reasonable proximity to one another; we defined that based on 30 min or more of inactivity, at which point the gambler was logged off automatically from the PlayNow website. The first bet in a session was thus defined as any bet following 30 min or more of inactivity. See Zhang et al.¹⁹ for further discussion of a gambling ‘session’.

Our analyses aimed to measure the effects of prior outcomes on subsequent gambling behaviour within a session, using the bet amount and the probability of quitting (i.e., the session) to indicate chasing. The bet amount model used the amount of money that the gambler wagered on the next bet as the dependent variable. We log transformed this variable because of its positively skewed nature. The quit model used a binary dependent variable Quit (Stay = 0, Quit = 1) to indicate whether the gambler quits or continues their session, as a function of the prior outcomes.

The shared key predictors for both the bet amount model and the quit model were:

• Outcome: the outcome magnitude of the immediately prior bet—calculated by the paid amount minus the winning amount. We took the absolute values of these net outcomes, depending on Outcome Dummy (loss or win), we then standardized these absolute outcomes with respect to the individual gambler’s average loss and win amount in that session. Thereby a value of zero indicated that this magnitude was of an average loss or win amount in this particular session for this gambler. This standardization method fits our goal to find how individual gamblers responded to a loss (or win) when it was larger than their personal average loss (or win) amount in the session.

• Outcome Dummy: a dummy variable indicating the outcome’s valence (i.e., win or loss) on the individual bet. A negative Outcome refers to a net loss; a positive Outcome refers to a net win; a zero refers to a ‘break-even’ bet. We re-parametrized the model in two ways, one using loss as the reference category (loss = 0, win = 1) to examine the loss chasing pattern, and another using win as the reference category (win = 0, loss = 1) to examine the win chasing pattern.

• Cumulative Outcome: the cumulative outcome magnitude, from the start of the session up to the current bet. We took their absolute values and then standardized them with respect to the individual gambler’s average loss and win amount in that session, depending on Cumulative Dummy. Thereby a value of zero indicates the average session win or loss magnitude for this gambler.

• Cumulative Dummy: a corresponding dummy variable that represents this cumulative outcome’s valence—win or loss. A negative Cumulative Outcome refers to a cumulative loss, and a positive Cumulative Outcome refers to a cumulative win or a break-even. We re-parameterized the model in two ways according to the dummy, one using loss as the reference (loss = 0, win = 1), and another using win as the reference (win = 0, loss = 1)

• Game: the gambling game type, comprising slot machines (reference = 0), probability games, blackjack, video poker, roulette, and mixed sessions, in which gamblers placed bets on more than one game types within a session.

As time series data, we covaried for time nuisance variables. The covariates in the bet amount model were:

• Session order: the order of the current session in a day.

• Bet order: the order of the bet within a particular session. We tested the bet amount model with a quadratic trend effect because it yielded the best fit via BIC than a linear trend. Bet Order was standardized to avoid multicollinearity.

The covariates in the quit model were:

• Session order: same as in the bet amount model.

• Time duration: the aggregated time duration (in hours) from the first bet of the session up to the last bet within a particular session. It accounts for the possibility that fatigue may influence whether a gambler continues or not.

Analysis

We used R Version 4.2.1²⁶ and the lme4 package²⁷ for modelling. We use p = 0.01 as the cut-off alpha level to determine statistical significance. The bet-by-bet data are a three-level nested data structure: each gambler (level 3) bets multiple sessions, and each session (level 2) includes multiple bets, and each bet (level 1) includes variables tracking gamblers’ basic betting trajectory. Our analyses used multilevel linear modelling and focused on handling random differences of level 1 predictors between Gambler ID. We started with the a priori models (see Supplementary Information), and then selected the final models to present in this analysis section. The final selected models diverged from the a priori models because the a priori models did not converge, or could not be run (i.e., computational infeasibility), or had higher a BIC (Bayesian Information Criterion) reflecting poorer goodness-of-fit while accounting for the model complexity.

Bet amount model. To test the effects of prior outcomes on the next bet amount and their interaction with games, we used the model:

$$\begin{array}{cc}\hfill Log\left(betamoun{t}_{\mathit{ijk}}\right)& ={\beta }_{0jk}+{\beta }_{1jk}Gam{e}_{\mathit{ijk}}+{\beta }_{2jk}Outcom{e}_{\mathit{ijk}}+{\beta }_{3jk}OutcomeDumm{y}_{\mathit{ijk}}+{\beta }_{4jk}CumulativeOutcom{e}_{\mathit{ijk}}\hfill \\ \hfill & +{\beta }_{3jk}CumulativeDumm{y}_{\mathit{ijk}}+{\beta }_{5jk}Outcom{e}_{\mathit{ijk}}\ast OutcomeDumm{y}_{\mathit{ijk}}+{\beta }_{6jk}Outcom{e}_{\mathit{ijk}}\ast Gam{e}_{\mathit{ijk}}\hfill \\ \hfill & +{\beta }_{7jk}OutcomeDumm{y}_{\mathit{ijk}}\ast Gam{e}_{\mathit{ijk}}+{\beta }_{8jk}CumulativeOutcom{e}_{\mathit{ijk}}\ast CumulativeDumm{y}_{\mathit{ijk}}\hfill \\ \hfill & +{\beta }_{9jk}CumulativeOutcom{e}_{\mathit{ijk}}\ast Gam{e}_{\mathit{ijk}}+{\beta }_{10jk}CumulativeDumm{y}_{\mathit{ijk}}\ast Gam{e}_{\mathit{ijk}}\hfill \\ \hfill & +{\beta }_{11jk}Outcom{e}_{\mathit{ijk}}\ast OutcomeDumm{y}_{\mathit{ijk}}\ast Gam{e}_{\mathit{ijk}}\hfill \\ \hfill & +{\beta }_{12jk}CumulativeOutcom{e}_{\mathit{ijk}}\ast CumulativeDumm{y}_{\mathit{ijk}}\ast Gam{e}_{\mathit{ijk}}\hfill \\ \hfill & +{\beta }_{13jk}scale\left(BetOrde{r}_{\mathit{ijk}}\right)+{\beta }_{14jk}scale{\left(BetOrde{r}_{\mathit{ijk}}\right)}^2+{\beta }_{15jk}SessionOrder+e_{ijk.}\hfill \\ \hfill \end{array}$$ 1

here $i$ indicates the last bet within the session, $j$ indicates the session, and $k$ indicates Gambler ID. The model only allowed random intercepts to vary by Gambler ID. We omitted random slopes and random effects for sessions due to the computational demands of these more complex models. Similarly, although we tried to test a generalized linear mixed-effect model with a gamma distribution, this model was too complicated to converge. In the "Result" section, we transformed estimated coefficients in the bet amount model using $100\times (e^b-1)$, such that the transformed estimates can then be interpreted as the estimated percentage change in bet amount per one-unit increase in the independent variable, holding all other variables constant.

The quit model To test the effect of prior outcomes on the probability of quitting in the next bet, and to compare such effects across the different game categories, we used the model:

$$\begin{array}{cc}\hfill Qui{t}_{\mathit{ijk}}\left(1=quit\right)& ={\beta }_{0jk}+{\beta }_{1jk}Gam{e}_{\mathit{ijk}}+{\beta }_{2jk}Outcom{e}_{\mathit{ijk}}+{\beta }_{3jk}OutcomeDumm{y}_{\mathit{ijk}}+{\beta }_{4jk}CumulativeOutcom{e}_{\mathit{ijk}}\hfill \\ \hfill & +{\beta }_{3jk}CumulativeDumm{y}_{\mathit{ijk}}+{\beta }_{5jk}Outcom{e}_{\mathit{ijk}}\ast OutcomeDumm{y}_{\mathit{ijk}}+{\beta }_{6jk}Outcom{e}_{\mathit{ijk}}\ast Gam{e}_{\mathit{ijk}}\hfill \\ \hfill & +{\beta }_{7jk}OutcomeDumm{y}_{\mathit{ijk}}\ast Gam{e}_{\mathit{ijk}}+{\beta }_{8jk}CumulativeOutcom{e}_{\mathit{ijk}}\ast CumulativeDumm{y}_{\mathit{ijk}}\hfill \\ \hfill & +{\beta }_{9jk}CumulativeOutcom{e}_{\mathit{ijk}}\ast Gam{e}_{\mathit{ijk}}+{\beta }_{10jk}CumulativeDumm{y}_{\mathit{ijk}}\ast Gam{e}_{\mathit{ijk}}\hfill \\ \hfill & +{\beta }_{11jk}Outcom{e}_{\mathit{ijk}}\ast OutcomeDumm{y}_{\mathit{ijk}}\ast Gam{e}_{\mathit{ijk}}\hfill \\ \hfill & +{\beta }_{12jk}CumulativeOutcom{e}_{\mathit{ijk}}\ast CumulativeDumm{y}_{\mathit{ijk}}\ast Gam{e}_{\mathit{ijk}}\hfill \\ \hfill & +{\beta }_{13jk}SessionOrde{r}_{\mathit{ijk}}+{\beta }_{14jk}TimeDuratio{n}_{\mathit{ijk}}+e_{\mathit{ijk}}.\hfill \\ \hfill \end{array}$$ 2

here $i$ indicates the last bet within the session, $j$ indicates the session, and $k$ indicates Gambler ID. The model allowed only random intercepts and random slopes of $Outcom{e}_{i\left(\mathit{jk}\right)}\ast OutcomeDumm{y}_{i\left(\mathit{jk}\right)}$ interaction varying by Gambler ID, which allowed the degree of loss and win chasing to differ randomly across persons. We fixed all other slopes and omitted random effects for sessions because of computational infeasibility. In the "Result" section, we transformed estimated coefficients in the quit model to the odds ratio using $e^b$.

Ethical approval

The study is an analysis on the de-identified secondary gambling data. The data storage and analysis procedure received ethical approval from the University of British Columbia’s Behavioural Research Ethics Board. We performed data storage and analyses in accordance to the approved guidelines.

Results

Descriptive results

Our descriptive statistics report medians due to the heavy skewness in many variables. Over all product categories, the typical gambler placed 27 bets (SD = 119.89) per session, spent $1.00 (SD = $20.91) per bet, and had a net outcome of -$0.50 (SD = $23.68) per bet. The overall betting patterns varied by game type (see Table 1). Slot machines ranked second highest for session length, with 44 bets per session; the only category with higher persistence was the mixed sessions, which also included some slot machine play. At the same time, slot machines displayed the smallest bet amounts (Median = $0.90), which is likely due to a limited range of bet options on slot machines.

Table 1.

Descriptive summary of bet count, net outcome, bet amount.

Game	Median	Mean	SD
Bet count
Mixed	54	111.22	164.07
Slots	44	90.79	135.35
Video poker	27	62.14	92.97
Probability	21	54.09	102.64
Blackjack	14	34.32	61.03
Roulette	11	28.87	55.89
Net outcome
Slots	− 0.50	− 0.23	16.95
Video poker	− 0.50	− 0.39	26.56
Mixed	− 0.40	− 0.24	19.69
Probability	− 0.40	− 0.21	7.74
Roulette	− 0.10	− 0.32	24.69
Blackjack	0.00	− 0.41	41.63
Bet amount
Blackjack	5.00	14.45	43.22
Roulette	2.00	9.21	24.77
Video poker	1.40	4.08	10.05
Mixed	1.00	4.14	16.13
Probability	1.00	1.77	3.46
Slots	0.90	1.94	4.17

Bet amount model

Figure 1 shows the bet amount as a function of losses and wins, both on the immediately prior outcome (Fig. 1A,B) and cumulatively (Fig. 1C,D). For each model, we first describe the overall loss chasing and win chasing trends, and then compare these effects between the slot machines and the other game categories.

Fig. 1.

Chasing tendencies on the bet amount. Note On Panel (A and B), the x-axis is the standardized loss amount (A) or win amount from the immediately prior bet (B), such that zero on the x-axis is the average loss (win) amount in a session. On Panel (C and D), the x-axis is the standardized loss amount for the cumulative loss (C) or the cumulative win (D), such that zero on the x-axis is the average cumulative loss (win) in a session. Across the panels, one unit increase on the x-axis indicates that the gambler lost (won) one standard deviation more their own average loss (win) of that session. An upward slope indicates chasing, such that gamblers bet more as a function of the prior loss (win), and a downward slope would indicate an absence of chasing, that gamblers bet less as a function of the prior loss (win).

Overall, for losing outcomes, gamblers bet more as a function of the immediately prior loss (Fig. 1A), but they bet less as a function of the cumulative amount lost (Fig. 1C). Thus, gamblers chased their immediate losses but did not chase their cumulative losses (Table 2). In terms of win chasing, gamblers bet more as a function of the immediately prior amount won (Fig. 1B), and also as a function of the cumulative amount won (Fig. 1D), consistent overall with win-chasing behaviour.

Table 2.

Summary of within-session chasing patterns by game category—compared to Slots.

	Immediate outcome		Accumulative outcome
	Bet amount	Quit	Bet amount	Quit
Loss chasing
Slots	✔	✔	✖	✖
Mixed	✔*	✔	✖*	✖*
Roulette	✔*	✖*	✖*	✖*
Blackjack	✔*	✖*	✖*	✖*
Probability	✔*	✔	✖*	✖*
Video poker	✔*	✔	✖*	✖*
Win chasing
Slots	✔	✖	✔	✔
Mixed	✔*	✖	✔	✖*
Roulette	✔*	✖	✔*	✖*
Blackjack	✔*	✔*	✔*	✖*
Probability	✔	✖	✔	✔
Video poker	✔	✖	✔	✖*

‘✔’ indicates a numerical effect in the direction of chasing compared to slots; ‘✖’ indicates a directional absence of chasing under the measurement. ‘*’ indicates a significant difference (p < 0.01) in chasing relative to slot machines as the reference category.

The effects of losses on the bet amount varied somewhat by game type: the game categories had different loss chasing slopes, both as a function of the immediate losses and cumulatively. After immediate losses, all games showed an increase in the bet amount (Fig. 1A). For slot machines, every standard deviation increase in the prior loss was estimated to increase the next bet by 24.58%, p < 0.001 (see Table 3a; the value of 24.58 is derived by inputting the slope of Outcome in the formula $100\times (e^b-1)$). Compared to slot machines, video poker (p = 0.004), blackjack (p < 0.001), mixed sessions (p < 0.001), and roulette (p < 0.001) were each associated with significantly steeper slopes as a function of the immediate loss, every standard deviation increase in the prior loss was estimated to increase their bet amount by 26.81%, 28.51%, 48.23%, 60.03%, respectively (each percentage computed using the formula $100\times (e^b-1)$ with b = slope of Outcome + slope of Outcome * Game). Probability games were associated with a significantly flatter slope compared to slot machines, and every standard deviation increase in the prior loss was estimated to increase the next bet amount by 19.00%.

Table 3.

Within-session chasing regression results in the bet amount model. (a) Used loss as the reference level. (b) Use win as the reference level.

Term	Estimate	std.error	df	Statistic	p value	conf.low	conf.high
a: The bet model with the reference of loss
(Intercept)	− 0.01	0.02	1,911,207	− 0.59	0.554	− 0.06	0.03
Outcome	0.22	0.00	1,911,207	251.43	< 0.001	0.22	0.22
Outcome dummy (Win = 1)	0.00	0.00	1,911,207	− 1.67	0.096	− 0.01	0.00
Blackjack	0.90	0.00	1,911,207	233.43	< 0.001	0.89	0.91
Mixed	0.05	0.00	1,911,207	23.10	< 0.001	0.04	0.05
Probability	0.05	0.00	1,911,207	10.50	< 0.001	0.04	0.06
Roulette	0.87	0.01	1,911,207	138.63	< 0.001	0.86	0.88
Video poker	0.12	0.01	1,911,207	12.12	< 0.001	0.10	0.14
Cumulative outcome	− 0.07	0.00	1,911,207	− 66.24	< 0.001	− 0.07	− 0.07
Cumulative outcome dummy (Win = 1)	0.18	0.00	1,911,207	99.60	< 0.001	0.17	0.18
Bet order	0.06	0.00	1,911,207	103.27	< 0.001	0.06	0.06
Bet order2	− 0.01	0.00	1,911,207	− 16.82	< 0.001	− 0.01	− 0.01
Session order	0.03	0.00	1,911,207	35.65	< 0.001	0.03	0.03
Outcome * Outcome dummy (Win = 1)	− 0.19	0.00	1,911,207	− 88.47	< 0.001	− 0.19	− 0.18
Outcome * Game (Loss chasing by game—compared to slots)
Blackjack	0.03	0.00	1,911,207	14.65	< 0.001	0.03	0.04
Mixed	0.17	0.00	1,911,207	126.21	< 0.001	0.17	0.18
Probability	− 0.05	0.00	1,911,207	− 15.71	< 0.001	− 0.05	− 0.04
Roulette	0.25	0.00	1,911,207	65.18	< 0.001	0.24	0.26
Video poker	0.02	0.01	1,911,207	2.87	0.004	0.01	0.03
Outcome dummy (Win = 1) * Game
Blackjack	− 0.05	0.00	1,911,207	− 15.01	< 0.001	− 0.06	− 0.04
Mixed	0.18	0.00	1,911,207	62.46	< 0.001	0.18	0.19
Probability	0.02	0.01	1,911,207	3.38	< 0.001	0.01	0.03
Roulette	− 0.40	0.01	1,911,207	− 68.49	< 0.001	− 0.42	− 0.39
Video poker	0.01	0.01	1,911,207	0.90	0.368	− 0.01	0.03
Cumulative outcome * Cumulative outcome Dummy (Win = 1)	0.10	0.00	1,911,207	58.85	< 0.001	0.10	0.11
Cumulative outcome * Game (Loss chasing by game—compared to slots)
Blackjack	0.040	0.00	1,911,207	18.69	< 0.001	0.04	0.04
Mixed	0.00	0.00	1,911,207	− 2.90	0.004	− 0.01	0.00
Probability	0.02	0.00	1,911,207	5.42	< 0.001	0.01	0.02
Roulette	− 0.01	0.00	1,911,207	− 3.24	0.001	− 0.02	− 0.01
Video poker	0.06	0.01	1,911,207	10.51	< 0.001	0.05	0.07
Cumulative outcome dummy (Win = 1) * Game
Blackjack	− 0.17	0.00	1,911,207	− 51.51	< 0.001	− 0.17	− 0.16
Mixed	0.04	0.00	1,911,207	13.24	< 0.001	0.03	0.04
Probability	− 0.02	0.01	1,911,207	− 3.75	< 0.001	− 0.03	− 0.01
Roulette	− 0.24	0.01	1,911,207	− 40.91	< 0.001	− 0.26	− 0.23
Video poker	− 0.01	0.01	1,911,207	− 1.17	0.241	− 0.03	0.01
Outcome * Outcome dummy (Win = 1) * Game
Blackjack	0.06	0.00	1,911,207	18.38	< 0.001	0.06	0.07
Mixed	− 0.05	0.00	1,911,207	− 18.51	< 0.001	− 0.06	− 0.05
Probability	0.05	0.01	1,911,207	8.05	< 0.001	0.04	0.07
Roulette	− 0.13	0.01	1,911,207	− 21.83	< 0.001	− 0.14	− 0.12
Video poker	0.00	0.01	1,911,207	− 0.35	0.730	− 0.02	0.02
Cumulative outcome * Cumulative outcome dummy (Win = 1) * Game
Blackjack	− 0.05	0.00	1,911,207	− 16.73	< 0.001	− 0.06	− 0.05
Mixed	0.00	0.00	1,911,207	0.76	0.447	0.00	0.01
Probability	− 0.01	0.01	1,911,207	− 2.72	0.007	− 0.03	0.00
Roulette	− 0.01	0.01	1,911,207	− 2.49	0.013	− 0.03	0.00
Video poker	− 0.06	0.01	1,911,207	− 5.83	< 0.001	− 0.08	− 0.04
b: The bet model with the reference of win
(Intercept)	0.16	0.03	1,911,207	6.37	< 0.001	0.11	0.21
Outcome	0.03	0.00	1,911,207	17.22	< 0.001	0.03	0.04
Outcome dummy (Loss = 1)	0.00	0.00	1,911,207	1.67	0.096	0.00	0.01
Blackjack	0.68	0.00	1,911,207	162.43	< 0.001	0.68	0.69
Mixed	0.27	0.00	1,911,207	79.29	< 0.001	0.26	0.28
Probability	0.05	0.01	1,911,207	6.60	< 0.001	0.03	0.06
Roulette	0.22	0.01	1,911,207	33.05	< 0.001	0.21	0.23
Video poker	0.11	0.01	1,911,207	9.33	< 0.001	0.09	0.14
Cumulative outcome	0.03	0.00	1,911,207	23.46	< 0.001	0.03	0.04
Cumulative outcome dummy (Loss = 1)	− 0.18	0.00	1,911,207	− 99.60	< 0.001	− 0.18	− 0.17
bet_order_scaled	0.06	0.00	1,911,207	103.27	< 0.001	0.06	0.06
I(bet_order_scaled^2)	− 0.01	0.00	1,911,207	− 16.82	< 0.001	− 0.01	− 0.01
Session order	0.03	0.00	1,911,207	35.65	< 0.001	0.03	0.03
Outcome * Outcome dummy (Loss = 1)	0.19	0.00	1,911,207	88.47	< 0.001	0.18	0.19
Outcome * Game (Win chasing by game—compared to slots)
Blackjack	0.09	0.00	1,911,207	35.28	< 0.001	0.09	0.10
Mixed	0.12	0.00	1,911,207	45.05	< 0.001	0.11	0.12
Probability	0.01	0.01	1,911,207	1.27	0.205	0.00	0.02
Roulette	0.12	0.00	1,911,207	27.07	< 0.001	0.11	0.13
Video poker	0.01	0.01	1,911,207	1.73	0.084	0.00	0.03
Outcome dummy (Loss = 1) * Game
Blackjack	0.05	0.00	1,911,207	15.01	< 0.001	0.04	0.06
Mixed	− 0.18	0.00	1,911,207	− 62.46	< 0.001	− 0.19	− 0.18
Probability	− 0.02	0.01	1,911,207	− 3.38	< 0.001	− 0.03	− 0.01
Roulette	0.40	0.01	1,911,207	68.49	< 0.001	0.39	0.42
Video poker	− 0.01	0.01	1,911,207	− 0.90	0.368	− 0.03	0.01
Cumulative outcome * Cumulative outcome dummy (Loss = 1)	− 0.10	0.00	1,911,207	− 58.85	< 0.001	− 0.11	− 0.10
Cumulative outcome * Game (Win chasing by game—compared to slots)
Blackjack	− 0.01	0.00	1,911,207	− 5.63	< 0.001	− 0.02	− 0.01
Mixed	0.00	0.00	1,911,207	− 1.23	0.219	− 0.01	0.00
Probability	0.00	0.00	1,911,207	0.59	0.556	− 0.01	0.01
Roulette	− 0.03	0.00	1,911,207	− 7.19	< 0.001	− 0.04	− 0.02
Video poker	0.00	0.01	1,911,207	0.21	0.832	− 0.01	0.02
Cumulative outcome dummy (Loss = 1) * Game
Blackjack	0.17	0.00	1,911,207	51.51	< 0.001	0.16	0.17
Mixed	− 0.04	0.00	1,911,207	− 13.24	< 0.001	− 0.04	− 0.03
Probability	0.02	0.01	1,911,207	3.75	< 0.001	0.01	0.03
Roulette	0.24	0.01	1,911,207	40.91	< 0.001	0.23	0.26
Video poker	0.01	0.01	1,911,207	1.17	0.241	− 0.01	0.03
Outcome * Outcome dummy (Loss = 1) * Game
Blackjack	− 0.06	0.00	1,911,207	− 18.38	< 0.001	− 0.07	− 0.06
Mixed	0.05	0.00	1,911,207	18.51	< 0.001	0.05	0.06
Probability	− 0.05	0.01	1,911,207	− 8.05	< 0.001	− 0.07	− 0.04
Roulette	0.13	0.01	1,911,207	21.83	< 0.001	0.12	0.14
Video poker	0.00	0.01	1,911,207	0.35	0.730	− 0.02	0.02
Cumulative outcome * Cumulative outcome dummy (Loss = 1) * Game
Blackjack	0.05	0.00	1,911,207	16.73	< 0.001	0.05	0.06
Mixed	0.00	0.00	1,911,207	− 0.76	0.447	− 0.01	0.00
Probability	0.01	0.01	1,911,207	2.72	0.007	0.00	0.03
Roulette	0.01	0.01	1,911,207	2.49	0.013	0.00	0.03
Video poker	0.06	0.01	1,911,207	5.83	< 0.001	0.04	0.08

Bold values are chasing estimates.

Across all product categories, gamblers reduced the next bet amount as a function of the cumulative loss (Fig. 1C). On slot machines, every standard deviation increase in the cumulative loss was estimated to reduce the next bet amount by 6.77% (p < 0.001). Compared to slot machines, probability games, blackjack, and video poker (ps < 0.001) were associated with flatter slopes as function of the cumulative loss, every standard deviation increase was estimated to lower the next bet amount by 5.13%, 2.96%, and 0.99% respectively. Mixed sessions (p = 0.004) and roulette (p = 0.001) were associated with steeper slopes as a function of cumulative loss compared to slot machines, and every standard deviation increase was estimated to lower the bet amount by 7.18% and 8.03%.

Win chasing slopes varied by game type too (Table 3b). On slot machines, for every standard deviation increase in the immediate win, gamblers increased their next bet amount by 3.37% (p < 0.001). Compared to slot machines, blackjack, mixed sessions, and roulette gamblers raised their next bet amount significantly more, as every standard deviation increase in the prior win was estimated to increase their next bet amount by 13.54%, 16.42%, and 16.71%, respectively (ps < 0.001). Probability games (p = 0.205) and video poker (p = 0.084) did not differ significantly from slot machines.

Similarly, the cumulative win (Fig. 1 Panel D) slopes varied by game type. For slot machines sessions, every standard deviation increase in the cumulative amount won was estimated to increase the next bet amount by 3.30% (p < 0.001). Roulette and blackjack sessions raised their bet amount significantly less than slot machine sessions (ps < 0.001), and every standard deviation increase in the cumulative win was estimated to increase the next bet amount by 0.47% and 1.97%. Mixed sessions (p = 0.219), probability games (p = 0.011), and video poker (p = 0.018) did not differ significantly compared to slot machines.

Quit model

Figure 2 shows the quit probability as a function of losses and wins, for the immediately prior outcomes (Fig. 2A,B) and cumulatively (Fig. 2C,D). Although the chasing trajectories in the bet amount model were qualitatively similar across product categories, the effects on the quit probability showed more heterogeneity by game type (see Table 2). After an immediate loss, gamblers in four of six game categories reduced their quit probability (Fig. 2A), showing chasing. After losing cumulatively, gamblers increased the quit probability across all game categories (Fig. 2C). After an immediate win, gamblers in five of six games increased their quit probability (Fig. 2B), and after winning cumulatively, gamblers in four of six games increased the quit probability (Fig. 2D).

Fig. 2.

Chasing tendencies on the quit probability. Note A downward slope indicates that the quit probability decreased as a function of the prior loss (win) amount, reflecting a pattern of chasing, whereas an upward slope indicates that the quit probability increased as a function of the prior loss (win), reflecting the absence of chasing.

For the model testing immediate losses, in slot machine sessions every standard deviation increase in the loss amount was estimated to significantly lower the odds of quitting the session by factor of 0.87 (i.e., 13% decrease; the OR is computed using formula $e^b$ with b = log odds) (Table 4a), indicating increased chasing intensities. On mixed sessions, probability games, and video poker, gamblers did not significantly differ from this rate. For blackjack and roulette sessions, gamblers displayed significantly lower chasing intensities than for slot machines, such that every standard deviation increase in the loss amount was estimated to increase their odds of quitting the session by factor of 1.10 and 1.20 (the is computed using formula $e^b$ with b = slope of log odds of Outcome + slope of Outcome * Game), respectively.

Table 4.

Within-session chasing regression results in the quit model. (a) Used loss as the reference level. (b) Use win as the reference level.

Term	OR	0.50%	99.50%	Log odds	p value
a: The bet model with the reference of loss
(Intercept)	0.00	0.00	0.00	− 5.63	< 0.001
Outcome	0.87	0.84	0.90	− 0.14	< 0.001
Outcome dummy (Win = 1)	0.74	0.67	0.82	− 0.30	< 0.001
Blackjack	6.21	5.51	7.01	1.83	< 0.001
Mixed	1.09	0.96	1.24	0.09	0.074
Probability	1.96	1.60	2.40	0.67	< 0.001
Roulette	9.64	8.21	11.33	2.27	< 0.001
Video poker	2.62	1.85	3.70	0.96	< 0.001
Cumulative outcome	8.90	8.45	9.36	2.19	< 0.001
Cumulative outcome dummy (Win = 1)	2.83	2.55	3.13	1.04	< 0.001
Session order	0.92	0.90	0.95	− 0.08	< 0.001
Time duration	0.92	0.92	0.93	− 0.08	< 0.001
Outcome * Outcome dummy (Win = 1)	1.25	1.12	1.39	0.22	< 0.001
Outcome * Game (Loss chasing by game—compared to slots)
Blackjack	1.26	1.19	1.34	0.23	< 0.001
Mixed	1.04	0.99	1.10	0.04	0.051
Probability	0.99	0.91	1.09	− 0.01	0.886
Roulette	1.38	1.26	1.51	0.32	< 0.001
Video poker	1.05	0.85	1.30	0.05	0.544
Outcome dummy (Win = 1) * Game
Blackjack	1.16	1.04	1.31	0.15	< 0.001
Mixed	1.22	1.07	1.40	0.20	< 0.001
Probability	1.34	1.07	1.68	0.30	< 0.001
Roulette	1.00	0.85	1.17	0.00	0.972
Video poker	1.27	0.91	1.77	0.24	0.068
Cumulative outcome * Cumulative outcome dummy (Win = 1)	0.09	0.08	0.10	− 2.39	< 0.001
Cumulative outcome * Game (Loss chasing by game—compared to slots)
Blackjack	0.35	0.33	0.38	− 1.04	< 0.001
Mixed	0.67	0.62	0.72	− 0.40	< 0.001
Probability	0.75	0.66	0.85	− 0.29	< 0.001
Roulette	0.30	0.28	0.34	− 1.19	< 0.001
Video poker	0.54	0.44	0.66	− 0.61	< 0.001
Cumulative outcome dummy (Win = 1) * Game
Blackjack	0.36	0.31	0.41	− 1.03	< 0.001
Mixed	0.60	0.51	0.70	− 0.52	< 0.001
Probability	0.53	0.41	0.68	− 0.63	< 0.001
Roulette	0.32	0.27	0.38	− 1.14	< 0.001
Video poker	0.49	0.32	0.74	− 0.72	< 0.001
Outcome * Outcome dummy (Win = 1) * Game
Blackjack	0.70	0.62	0.80	− 0.35	< 0.001
Mixed	0.94	0.82	1.09	− 0.06	0.297
Probability	1.10	0.86	1.40	0.10	0.308
Roulette	0.71	0.60	0.84	− 0.35	< 0.001
Video poker	1.00	0.70	1.42	0.00	0.994
Cumulative outcome * Cumulative outcome dummy (Win = 1) * Game
Blackjack	7.23	6.51	8.03	1.98	< 0.001
Mixed	2.34	2.07	2.65	0.85	< 0.001
Probability	1.28	1.03	1.59	0.25	0.003
Roulette	8.52	7.36	9.86	2.14	< 0.001
Video poker	2.58	1.77	3.74	0.95	< 0.001
b: The bet model with the reference of win
(Intercept)	0.01	0.01	0.01	− 4.89	< 0.001
Outcome	1.08	0.98	1.20	0.08	0.043
Outcome dummy (Loss = 1)	1.35	1.23	1.49	0.30	< 0.001
Blackjack	2.58	2.26	2.96	0.95	< 0.001
Mixed	0.80	0.69	0.92	− 0.23	< 0.001
Probability	1.40	1.09	1.79	0.34	< 0.001
Roulette	3.08	2.58	3.69	1.13	< 0.001
Video poker	1.61	1.08	2.41	0.48	0.002
Cumulative outcome	0.82	0.77	0.87	− 0.20	< 0.001
Cumulative outcome dummy (Loss = 1)	0.35	0.32	0.39	− 1.04	< 0.001
Session order	0.92	0.90	0.95	− 0.08	< 0.001
Time duration	0.92	0.92	0.93	− 0.08	< 0.001
Outcome * Outcome dummy (Loss = 1)	0.80	0.72	0.89	− 0.22	< 0.001
Outcome * Game (Win chasing by game—compared to slots)
Blackjack	0.88	0.79	0.99	− 0.12	0.006
Mixed	0.98	0.86	1.12	− 0.02	0.746
Probability	1.09	0.88	1.37	0.09	0.296
Roulette	0.97	0.84	1.13	− 0.03	0.653
Video poker	1.05	0.79	1.39	0.05	0.656
Outcome dummy (Loss = 1) * Game
Blackjack	0.86	0.76	0.97	− 0.15	< 0.001
Mixed	0.82	0.71	0.94	− 0.20	< 0.001
Probability	0.74	0.59	0.93	− 0.30	< 0.001
Roulette	1.00	0.85	1.18	0.00	0.972
Video poker	0.79	0.56	1.10	− 0.24	0.068
Cumulative outcome * Cumulative outcome dummy (Loss = 1)	10.89	10.02	11.83	2.39	< 0.001
Cumulative outcome * Game (Win chasing by game—compared to slots)
Blackjack	2.55	2.36	2.76	0.94	< 0.001
Mixed	1.57	1.42	1.73	0.45	< 0.001
Probability	0.96	0.80	1.14	− 0.04	0.523
Roulette	2.59	2.33	2.88	0.95	< 0.001
Video poker	1.39	1.02	1.91	0.33	0.007
Cumulative outcome dummy (Loss = 1) * Game
Blackjack	2.80	2.46	3.18	1.03	< 0.001
Mixed	1.68	1.44	1.96	0.52	< 0.001
Probability	1.88	1.47	2.41	0.63	< 0.001
Roulette	3.12	2.62	3.72	1.14	< 0.001
Video poker	2.06	1.34	3.16	0.72	< 0.001
Outcome * Outcome dummy (Loss = 1) * Game
Blackjack	1.42	1.25	1.62	0.35	< 0.001
Mixed	1.06	0.92	1.22	0.06	0.297
Probability	0.91	0.71	1.16	− 0.10	0.308
Roulette	1.41	1.19	1.68	0.35	< 0.001
Video poker	1.00	0.70	1.42	0.00	0.994
Cumulative outcome * Cumulative outcome dummy (Loss = 1) * Game
Blackjack	0.14	0.12	0.15	− 1.98	< 0.001
Mixed	0.43	0.38	0.48	− 0.85	< 0.001
Probability	0.78	0.63	0.97	− 0.25	0.003
Roulette	0.12	0.10	0.14	− 2.14	< 0.001
Video poker	0.39	0.27	0.56	− 0.95	< 0.001

Bold values are chasing estimates.

For the cumulative loss model, every standard deviation increase in the loss amount increased the odds of quitting by factor of 8.90 in slot machine sessions. On all other game categories, gamblers increased the odds of quitting significantly less than in slot machine sessions, reflecting greater chasing intensities in the other product categories as a function of the cumulative loss.

For the immediate win model (Table 4b), in slot machine the amount won did not significantly change the odds of quitting. Blackjack gamblers had a significantly smaller win chasing slope compared to the slot machine gamblers: every standard deviation increase in the win amount reduced the odds of quitting by factor 0.96, reflecting a greater chasing intensity. Mixed sessions, probability games, roulette, video poker gamblers did not differ significantly from slot machines.

For the cumulative win model, every standard deviation increase in the amount won reduced the odds of quitting by factor of 0.82 in slot machine sessions. Blackjack, mixed sessions, roulette, and video poker gamblers reduced chasing intensities, such that every standard deviation increase in the win amount was estimated to increase their odds of quitting by factor of 2.09, 1.28, 2.12, and 1.14, respectively. These rates were each significantly greater than slot machine sessions, whereas Probability games did not differ significantly from slot machines.

Discussion

The present study characterized bet-by-bet within-session chasing patterns across multiple game categories online, comprising slot machines, probability games, blackjack, video poker, roulette, and mixed sessions. Sessions of online gambling were delineated based on 30 min or more of inactivity, to approximate logging on and logging off from the website. Within-session chasing was operationalized as an increase in the bet amount, or a reduction in the quit probability (i.e., session persistence), as a function of the amount lost or won. Chasing tendencies were seen to vary to some extent on several dimensions: the outcome valence (losses or wins), the chasing expression (bet amount or quit probability), the timeframe of the prior outcomes (immediately prior or cumulative), and the game category (see Table 2).

Across most games, win chasing differed markedly by chasing expression, while the immediate versus cumulative outcome timeframe exerted a more consistent effect. After winning, both immediately and cumulatively, gamblers on most games were more likely to quit their session, but simultaneously increased their next bet amount (i.e. on those sessions when they did not yet quit). One consideration for this expression of chasing is the increase in available funds for further betting (i.e., a ‘wealth effect’²⁸). The house money effect²⁹ is also relevant here, which explains how windfall payments can encourage further risk-seeking³⁰. People betting after a gambling win may not yet have internalized those funds as their own, and therefore feel as if they are playing with ‘house money’. The wealth effect and house money effect speak most clearly to the increase in the bet amount, but do not readily explain the increased the likelihood of quitting seen in the present data. Alternatively, the winning scenario might encourage gamblers to play more strategically with available funds; they can afford to bet more but they did not wish to lose those profits, hence shortening the session. By expressing chasing in one form but not another, this suggests the gamblers are not simply impulsive risk-seekers³³.

By contrast, in the losing scenario, immediate versus cumulative losses led to categorically distinct chasing tendencies, while there was more consistency between the two chasing expressions. After losing more in the immediately prior bet, gamblers tended to increase their bet amount and were more likely to continue (less likely to quit) their session. These illustrations of within-session loss chasing cannot be readily explained by reduced funds or financial constraints. However, after losing more cumulatively, gamblers generally bet less and were more likely to quit. This absence of loss chasing in the cumulative analysis may be explained simply by diminishing financial resources to continue, i.e., the inverse of the wealth effect described for winning. Such an argument was supported by chasing results from another online gambling dataset. Compared to our bet-by-bet observations, Ma et al., (2014) analyzed loss chasing in the longer-term with week-by-week data and found that gamblers bet less in the following week as a function of the losses in the most recent week, but bet more in the following week as a function of the losses over aggregating weekly net losses. Although these effects seem quite different to our observations, if one considers the time frame in the order of weeks, gamblers will likely have some opportunities to ease financial constraints (e.g., after paydays³¹) and can then afford to chase accumulating losses, compared to our single session timeframe. Future studies may test this hypothesis by linking gambling data with bank information³² and investigating the relationship between within-session chasing and one’s paydays and billing cycles.

Another possible explanation for the juxtaposition of loss chasing to immediate losses and the absence of loss chasing to cumulative losses is the break-even effect. This posits that people's risk attitude is driven by the goal of breaking even, and thus they will be more risk-seeking so long as the prior loss is small enough to be recoverable by the win associated with the risky choice²⁹. A single bet is characterized by a likely but small loss, and gamblers may be hopeful of recovery after small immediate losses, driving our observed chasing for both bet amount and quit probability. However, with cumulative losses over a session, the possibility of breaking even via chasing diminishes, so loss chasing would be expected to dissipate. A US study of land-based slot machine gambling found largely consistent chasing patterns with our results and also suggested that loss chasing patterns might be explained by the break-even effect¹⁷. Those slot machine gamblers increased their bet amount, while the likelihood of quitting stayed stable as a function of the immediately prior loss. But after losing cumulatively, they reduced the bet amount and only continued playing as long as the loss was small. By stopping when the loss became too large, they expressed an intolerance to large losses, perhaps because the possibility of breaking even had diminished.

Comparing the effect of wins and losses on chasing patterns, we concluded that losses were more likely to induce overall chasing than wins in most games (see Table 2), assuming gamblers can afford to chase. After losing the immediately prior bet, when the loss was small and affordable, gamblers consistently chased by increasing their bet amounts over a longer session. After winning, immediately and cumulatively, gamblers bet strategically by betting larger only over a shorter session. This effect of losses being greater than wins aligns with loss aversion in the general population and corroborates the central role of losses in developing gambling problems among screening protocols for Gambling Disorder (e.g., Problem Gambling Severity Index or PGSI; the Diagnostic and Statistical Manual of Mental Disorders or DSM-5). This behavioural potency of losses may be explained by conceptualizing chasing as a form of urgency—the affect-related component of impulsivity³³. In supporting this conceptualization, empirical studies on impulsivity showed that losses prompted a faster reaction time to the next gambling trial than wins^6,34, and laboratory-induced negative affect from film clips extended gambling persistence³⁵.

As hypothesized, the chasing patterns varied by game category. As a key diagnostic feature of Gambling Disorder, high chasing intensities may represent the warning sign for potential gambling harm. Gambling products associated with greater chasing behaviours may also warrant regulatory scrutiny. Our analyses by game type used the online slot machine games in the eCasino as the reference level, as a gambling format with a relatively well-established risk profile^36,37. Overall, the slot machine category was associated with the longest sessions (in terms of number of bets) but the smallest bet amounts; this contrast is consistent with their structural design that facilitates continuous game-play but with a limited range of bet options, and this may further shape the two expressions of chasing.

Our behavioural data showed that slot machines were not the game with the highest chasing intensity. For immediate losses, slot machine gamblers chased on both bet amount and quitting odds, but slot machines were not the game category associated with the steepest slopes on either measure. Instead, gamblers on video poker, blackjack, mixed sessions, and roulette chased more intensively on bet amount, while gamblers on video poker, probability, and mixed sessions had similar quitting odds as slot machines. We did not see a cumulative loss chasing effect across any game category, including slot machines, on either measure. For wins, slot machine gamblers chased their immediate wins on the bet amount but not by session persistence, and again slot machines were not the category associated with the greatest chasing tendencies. Compared to slot machines, gamblers on blackjack, mixed sessions, and roulette chased more intensively on bet amount, while gamblers on blackjack chased more intensively on quitting odds. For cumulative wins, slot machine gamblers chased on both the bet amount and by continuing, but slot machines were not the only game chased among the highest intensity by these two accounts. Gamblers on mixed sessions, probability, and video poker had similar chasing intensity on bet amount as slot machines, while gamblers on probability games had similar chasing intensity on quitting odds as slot machines. Overall, our findings provide little support for the hypothesis that slot machines are the most chasing intensive game category.

Chasing has two major forms, between-session and within-session, and this study focused on within-session chasing expressions using bet-by-bet tracked behaviour. Clinical and epidemiological screening tools for gambling problems, such as DSM-5³ and PGSI³⁸, currently emphasize the between-session construct (e.g. ‘returning another day to get even’). The within-session sample in the current study was a subgroup of the sample used in a separate study of between-session effects¹⁹; see sample comparison in Supplementary Information). We characterized between-session chasing as the time interval to return to the eCasino as a function of the amount won or lost in the last session. Comparing the results across the between-session and within-session analyses, the chasing patterns did not consistently act in tandem (Supplementary Information Table 1). For instance, the average gambler in most games did not chase losses between sessions, but they would chase the immediately prior loss within a session. The average gambler would chase wins both between and within sessions, but they chased only on bet amount and not on quitting odds, reflecting a strategic chasing style.

In characterizing chasing systematically across gambling forms, the heterogeneity of the behavioural responses stands out, and we see this as being at odds with the emphasis given to between-session chasing in the clinical assessment of disordered gambling. Further research is required to resolve which of these two forms of chasing is the better behavioural marker for gambling problems. On the one hand, gambling problems inherently manifest in long-term behaviour across days³⁹. On the other hand, during fast and continuous gambling, gamblers’ self-control may be further compromised by emotional influences and states of in-game immersion¹², as well as memory requirements to track recent outcomes. Although online gambling datasets typically lack detailed information on clinical status⁴⁰, they can contain some proxy markers, to investigate chasing forms in gamblers with likely problems. such as voluntary self-exclusion or account closure¹.

The use of behavioural tracking data is gaining popularity in academic and industry sectors for examining online gambling behaviour. In this detailed examination of betting behaviour across multiple gambling formats on a single Canadian platform, our data highlight both the potential benefits but also the significant challenges with this methodology. Computational feasibility of modelling the bet-by-bet data limited our analysis to a subgroup of gamblers, who were sampled from the mid-range (35 to 65 percentile) in terms of the bet volume. As such, our behavioural patterns reflect the typical gambler but offer limited insight into the chasing behaviours of more highly engaged (and, potentially, disordered) gamblers at the upper end of the distribution. Future studies may benefit from dedicated servers and advanced data processing techniques (e.g., parallel programming techniques⁴¹).

Author contributions

K.Z. conceptualized ideas, analysed data, and drafted the manuscript. J.R. advised on data analysis and manuscript review and editing. X.D. contributed to data management and server maintenance. T.L. contributed to conceptualization and manuscript review. L.C. supervised the project and contributed to manuscript review and editing. This study used a secondary dataset provided by BCLC to the Centre for Gambling Research. All authors have approved the final manuscript.

Data availability

The dataset for this study was provided to the researchers by the BCLC, under a Non-Disclosure Agreement that prohibits further data sharing.

Competing interests

The Centre for Gambling Research at UBC is supported by the Province of British Columbia government and the BCLC. LC holds a Discovery Award from the Natural Sciences and Engineering Research Council (Canada). KZ held the Graduate Fellowship in Gambling Research, a fellowship supported by the BCLC and adjudicated by the UBC Faculty of Arts. KZ held the Graduate Fellowship in Gambling Research (2021–2022), a fellowship supported by the BCLC and adjudicated by the UBC Faculty of Arts. LC is the Director of the Centre for Gambling Research at UBC, which is supported by funding from the Province of British Columbia and the BCLC. The Province of BC government and the BCLC had no role in the preparation of this manuscript, and imposed no constraints on publishing. LC has received remuneration from the International Center for Responsible Gaming (travel; speaker honoraria; academic services), the Institut fur Glucksspiel und Gesellschaft (Germany; travel; speaker honoraria), GambleAware (UK; academic services), Gambling Research Australia (academic services), Alberta Gambling Research Institute (Canada; travel; academic services), German Foundation for Gambling Research (advisory board; travel). He has been remunerated for legal consultancy by the BCLC. LC receives an honorarium for his role as Co-Editor-in-Chief for International Gambling Studies from Taylor & Francis, and he has received royalties from Cambridge Cognition Ltd. relating to neurocognitive testing. KZ and LC have not received any further direct or indirect payments from the gambling industry or groups substantially funded by gambling. The remaining authors declare no conflict of interest.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-024-70738-3.