Content uploaded by Matthew Rockloff
Author content
All content in this area was uploaded by Matthew Rockloff on Apr 18, 2017
Content may be subject to copyright.
Rollover Effect and Goal-gradient Effect
1
Running Head: Rollover Effect and Goal-gradient Effect
Jackpot Structural Features: Rollover Effect and Goal-gradient Effect in EGM
Gambling
En Li1, Matthew J Rockloff1, Matthew Browne1, Phillip Donaldson1
(1) Central Queensland University, Queensland, 4701, Australia
MANUSCRIPT SUBMITTED FOR PUBLICATION
Correspondence concerning this article should be addressed to En Li, Central Queensland
University, Queensland, 4701, Australia. E-mail: e.li@cqu.edu.au. Phone: +61-7-4923-2386.
No conflicts of interest are declared.
ACKNOWLEDGMENT:
This research was fully funded by Gambling Research Australia, a partnership between the
Commonwealth, State and Territory Governments. A previous version of this paper has been
submitted and published as part of a research report commissioned by Gambling Research
Australia.
Preferred Reference:
Li, E., Rockloff, M. J., Browne, M., & Donaldson, P. (2015). Jackpot Structural Features:
Rollover Effect and Goal-Gradient Effect in EGM Gambling. Journal of Gambling Studies,
707–20. doi:10.1007/s10899-015-9557-7
Rollover Effect and Goal-gradient Effect
2
Abstract Relatively little research has been undertaken on the influence of jackpot
structural features on Electronic Gaming Machine (EGM) gambling behavior. This study
considered two common features of EGM jackpots: progressive (i.e., the jackpot
incrementally growing in value as players make additional bets), and deterministic (i.e., a
guaranteed jackpot after a fixed number of bets, which is determined in advance and at
random). Their joint influences on player betting behavior and the moderating role of jackpot
size were investigated in a crossed-design experiment. Using real money, players gambled on
a computer simulated EGM with real jackpot prizes of either $500 (i.e., small jackpot) or
$25,000 (i.e., large jackpot). The results revealed three important findings. Firstly, players
placed the largest bets (20.3% higher than the average) on large jackpot EGMs that were
represented to be deterministic and non-progressive. This finding was supportive of a
hypothesized ‘goal-gradient effect’, whereby players might have felt subjectively close to an
inevitable payoff for a high-value prize. Secondly, large jackpots that were non-deterministic
and progressive also promoted high bet sizes (17.8% higher than the average), resembling the
‘rollover effect’ demonstrated in lottery betting, whereby players might imagine that their
large bets could be later recouped through a big win. Lastly, neither the hypothesized goal-
gradient effect nor the rollover effect was evident among players betting on small jackpot
machines. These findings suggest that certain high-value jackpot configurations may have
intensifying effects on player behavior.
Keywords Electronic Gaming Machine (EGM) · Jackpot · Progressive · Deterministic ·
Jackpot size · Gambling intensity
Rollover Effect and Goal-gradient Effect
3
Introduction
With the number of Electronic Gaming Machines (EGMs) exceeding 7.6 million worldwide
(Ziolkowski 2014), an increasing amount of research has been conducted on various aspects
of EGMs and EGM player behaviors (e.g., Blaszczynski et al. 2014; Browne et al. 2014;
Harrigan et al. 2014; Millhouse and Delfabbro 2008; Rockloff et al. 2014; Rockloff et al.
2012; Rockloff et al. 2011; Sharpe et al. 2005; Wohl et al. 2013). Despite the existing
research efforts, there is little direct evidence on the influence of the structural characteristics
of EGM jackpots on gambling behavior (for a review, see Rockloff and Hing 2013). To
redress this deficit, this present study examined two common structural features of EGM
jackpots: progressive versus non-progressive jackpots, and deterministic versus non-
deterministic jackpots. These features, as described below, were examined in an experimental
design for their potential interactive effects on intensity of gambling, and for how jackpot
size may moderate such effects.
Progressive versus Non-progressive Jackpots: Rollover Effect or Goal-gradient Effect?
Progressive jackpots incrementally grow in value as players make additional bets.
Alternatively, non-progressive jackpots generate a fixed dollar payout irrespective of the
accumulation of bets and their contributions from the players. Two conflicting views can be
postulated regarding whether progressive or non-progressive jackpots should have a more
motivating effect on players’ gambling intensity.
On the one hand, evidence from lottery betting (Beenstock and Haitovsky 2001;
Farrell et al. 1999; Rogers 1998; Rogers and Webley 2001) suggests that progressive (vs.
non-progressive) jackpots may lead to a “rollover effect” where gamblers are encouraged to
Rollover Effect and Goal-gradient Effect
4
bet more. For example, lottery sales figures in the UK tended to increase dramatically on
weeks when rollover jackpots occurred (Rogers 1998; Rogers and Webley 2001).
Furthermore, a time series analysis based on the British lottery sales found that jackpot
rollovers boosted sales and potential addiction for the game (Farrell et al. 1999). The time
series lottery data from Israel also confirmed a positive relationship between rollovers and
ticket sales (Beenstock and Haitovsky 2001). If the rollover effect does apply to EGM
jackpots, EGMs with progressive (vs. non-progressive) jackpots should lead players to bet
more as each bet adds to the accumulated amount of the jackpots, and that amount may be
seen as recoverable investment in the ultimate jackpot prize.
An alternative point of view is that EGM players who consider hitting the jackpot as
their goal may experience a “goal-gradient effect” (Brown 1948; Hull 1934; Kivetz et al.
2006), where the efforts to fulfil a goal tend to be negatively related to the perceived distance
to the goal. The goal-gradient effect was originally found in animal behaviors, such that
animals ran faster and exerted stronger force when they drew closer to their food (Brown
1948; Hull 1934). Kivetz et al. (2006) proposed that the goal-gradient effect should also
apply to humans, and demonstrated the motivating effect of shorter goal distance across a
variety of human behaviors (e.g., coffee purchases; rating songs) and reward scenarios (e.g.,
buy ten coffees to earn one; rate 51 songs to earn US$25; Kivetz et al. 2006). According to
the goal-gradient effect, progressive - rather than non-progressive jackpots - may increase
EGM players’ perceived distance to the goal of “hitting the jackpot” as the jackpot value
grows after each additional bet. From this perspective, the growing jackpot amount draws
attention to the increasing number of historical bets that have been unsuccessful in winning
the jackpot, making the goal appear more difficult to attain. Additionally, higher value prizes
are associated with a lower perceived probability of winning, and a corresponding larger
Rollover Effect and Goal-gradient Effect
5
number of competing bettors. Moreover, progressive jackpots may make bettors more aware
of their contributions to the jackpot, which may be properly seen to be most likely to benefit
someone else when the jackpot goal appears distant and unachievable, adding to the power of
demotivation.
Are there situations in which either the rollover effect or the goal-gradient effect tends
to be more dominant in affecting play motivation? In order to answer this question, we
considered two other common dimensions of EGM jackpots: deterministic versus non-
deterministic format, and varying prize amount. The contribution of knowledge activation
theory in social psychology (Higgins 1996) leads us to expect that both features could
influence the relative activation of rollover and goal-gradient effects on EGM play behavior.
Knowledge Activation Theory
A classical topic in social psychology concerns what knowledge people activate and use
when encountering or responding to a stimulus event (e.g., Krech and Crutchfield 1948). An
important milestone in this line of research is the theoretical framework developed by
Higgins (1996), who formulated a series of rules on knowledge activation, among which the
‘applicability rule’ is of direct relevance to the present study. The applicability rule suggests
that the likelihood that some knowledge can be brought to mind from memory store is
determined by the degree of overlapping between the features of the knowledge and the
attended features of the stimulus being encountered, and that increased degree of overlapping
leads to increased applicability of the knowledge to the stimulus and increased probability of
the knowledge being activated and applied in decision making (Higgins 1996). For example,
consider a patron of a gambling venue who is deciding which EGM she is going to play. If
Rollover Effect and Goal-gradient Effect
6
she happens to pay attention to the jackpot (vs. free-spin) features advertised in the venue,
she may tend to rely on her knowledge about the jackpot (vs. free-spin) features to choose the
machine; as such knowledge is highly applicable to her attended aspect of the decision
environments, and consequently highly accessible in her mind. Importantly, two critical
criteria necessary to suffice the applicability rule lie in the knowledge-stimulus overlapping
and the attention drawn to the stimulus features. The subsequent section will specify how the
EGM jackpot features in question can affect whether these criteria can be met.
Deterministic versus Non-deterministic Jackpots: First Moderator
Deterministic jackpots have a guaranteed payout after a fixed number of bets, which is
determined in advance at random, but hidden from players. Non-deterministic jackpots, on
the other hand, have a potential payout assessed at random with every bet. Therefore,
deterministic jackpots incorporate a notion of goal-distance, where each bet decreases the
number of bets remaining before the jackpot triggers. For non-deterministic jackpots, the
probability of winning is constant over time (Rockloff and Hing 2013), and goal-distance
does not apply. According to the overlapping criterion of the applicability rule (Higgins
1996), players’ knowledge about their distance to a jackpot, and therefore influence of the
goal-gradient effect, would be activated only when the EGM features a deterministic rather
than non-deterministic jackpot (Fig. 1, sections a & b). The rollover effect, however, is driven
solely by knowledge regarding increasing jackpot size, and should therefore be applicable for
any progressive jackpots, including those with deterministic or non-deterministic features
(Fig. 1, sections b & d). As for EGMs with non-progressive and non-deterministic jackpots,
there is little overlapping between their jackpot features and players’ knowledge about
Rollover Effect and Goal-gradient Effect
7
jackpot distance or jackpot growth. Hence, neither goal-gradient nor rollover effect will be
activated among players betting on such EGMs (Fig. 1, section c).
Fig. 1 Knowledge activation under different jackpot features
As previously stated, the rollover effect and the goal-gradient effect tend to exert their
motivating force in opposite directions. Under progressive jackpots, players’ gambling
intensity should increase when the rollover effect is activated. The goal-gradient effect, on
the other hand, can only strengthen betting intensity on EGMs featuring non-progressive
jackpots. Furthermore, the activation of goal-gradient effect tends to demotivate players
betting on progressive EGMs.
Based on this theorization, it is expected that the rollover effect activated by
progressive features should positively affect play behavior on non-deterministic EGMs (i.e.,
Non-progressive Progressive
Deterministic
Non-deterministic
a) Goal-gradient
effect
d) Rollover
effect
b) Goal-gradient
& rollover effect
c) No effect
Rollover Effect and Goal-gradient Effect
8
causing greater betting intensity in section d rather than c in Fig. 1). Moreover, the goal-
gradient effect activated by deterministic features should lead to greater betting intensity on
non-progressive (Fig. 1, section a) rather than progressive EGMs (Fig. 1, section b), since the
goal-gradient and rollover effects tend to cancel each other out in the latter situation.
Jackpot Size: Second Moderator
Jackpot size may also moderate the influence of rollover and goal-gradient motivations on
EGM play. As the size of a reward is positively related to its salience and its ability to attract
attention (Eisenberger and Selbst 1994), a small (vs. large) reward can lead to less (vs. more)
attention being focused on its relevant aspects (Easterbrook 1959). Hence, EGMs with small
(vs. large) jackpots should reduce players’ attention to their jackpot features. Following the
attention criterion of the applicability rule (Higgins 1996), both the solo and the combined
jackpot effects proposed above would be less likely to occur among players betting on EGMs
featuring small rather than large jackpots.
Hypotheses
Based on the theory outlined in the previous sections, the following hypotheses are proposed:
H1 (rollover effect > no effect): When players bet on EGMs featuring non-deterministic
jackpots, their betting intensity will be greater if the jackpots are progressive rather than non-
progressive.
Rollover Effect and Goal-gradient Effect
9
H2 (goal-gradient effect > combined effect): When players bet on EGMs featuring
deterministic jackpots, their betting intensity will be greater if the jackpots are non-
progressive rather than progressive.
H3 (size effect): Both effects in H1 and H2 will diminish if the jackpots are small rather than
large.
Purpose of the Experiment
The present experiment was devised to investigate the hypothesized interactive effects of
(non-) progressive jackpots, (non-) deterministic jackpots, and jackpot size on EGM
gambling behavior. In particular, we wished to test which EGM jackpot configurations could
lead to the hypothesized effects on aspects of play in the forms of bet size, betting speed and
betting persistence, as well as on players’ subjective and physiological reactions.
Methods
Participants
One hundred and twenty-three (51 male, 72 female) participants, aged 18 - 82 (M = 50.4, SD
= 16.4) completed the experiment following recruitment from newspaper-flyer
advertisements in Bundaberg, Queensland Australia. The majority of the participants had a
personal income level between $2001 and $799 per week (70.8%) and an Australian cultural
background (92.7%). As calculated from the 9-item Problem Gambling Severity Index
1 Any money amount reported in this article is measured in Australian currency, unless otherwise stated.
Rollover Effect and Goal-gradient Effect
10
(PGSI, Ferris and Wynne 2001), 34.1%, 21.1%, and 10.6% of the participants were low-risk,
moderate-risk, and problem gamblers. Furthermore, among the total participants, 93.5% had
gambling experiences before, 64.2% had gambled on EGMs within the last 12 months, and
51.2% regarded EGM as their preferred form of gambling. Please see Table 1 for more
details on the experiment participants.
Table 1 Demographic and Gambling Characteristics of Experiment Participants (n = 123)
Characteristic n %
Gender
Male 51 41.5
Female 72 58.5
Age
30 or below 21 17.1
31-40 14 11.4
41-50 22 17.9
51-60 23 18.7
Above 60 43 35.0
Income
Less than $80 per week 6 4.9
$80-$199 per week 15 12.2
$200-$499 per week 60 48.8
$500-$799 per week 27 22.0
$800-$1,099 per week 6 4.9
$1,100 and above per week 9 7.3
Cultural background
Australian 114 92.7
English, New Zealand, South African, or German 8 6.5
Other 1 0.8
PGSI
Rollover Effect and Goal-gradient Effect
11
0 (no identifiable problems) 41 33.3
1-2 (low-risk) 42 34.1
3-7 (moderate-risk) 26 21.1
8 or more (problem gamblers) 13 10.6
Missing data 1 0.8
Participants who had gambling experiences before 115 93.5
Participants who had gambled on EGMs within the last 12 months 79 64.2
Participants who reported EGM as their preferred form of gambling 63 51.2
Computer Simulated EGM Programs
Nine versions of a computer simulated EGM were created in Visual Basic for this study,
including eight EGMs with different jackpot features and one EGM without a jackpot. When
participants started those jackpot EGMs, they first encountered an instruction screen where
the corresponding EGM’s jackpot features were explained. Then the program proceeded to
the game interface, it displayed the game information on three areas; including a top area
showing the present amount of the jackpot, a central area showing 3 reels of symbols, and a
bottom area showing betting options and results (see Fig. 2). When participants were playing,
they chose the bet size (i.e., 25, 50, or 100 cents) and clicked on the “Spin!” button. At the
conclusion of a spin, the reels rested on a 3 symbol-combination outcome. The EGMs had the
typical sounds associated with spinning reals, winning bets, and losing bets. As participants
kept playing, the changes of the betting results and the jackpot amounts (for EGMs with
progressive jackpots) were reflected on the screen after every single bet. The no-jackpot
EGM was identical to the jackpot EGMs in most aspects, except the lack of jackpot
instruction screen and jackpot information in the game. Furthermore, unbeknownst to the
study participants, all EGMs were programmed to generate the same winning outcomes (i.e.,
10 times the bet size) on five specific bets (i.e., 2nd, 6th, 8th, 13th, and 20th bet) and the same
Rollover Effect and Goal-gradient Effect
12
losing outcomes (i.e., 0 times the bet size) on all other bets. EGMs were programmed without
any Jackpot wins. However, all participants who completed the experiment were put into an
end-of-study lottery draw for a $500 jackpot prize.
Fig. 2 The game interface of a computer simulated EGM program in the experiment
Procedures
Ethical clearance for this research project was gained through the Human Research Ethics
Committee in the researchers’ university. Participants were given $20 upon arrival at their
session as compensation for their time. After completing a brief questionnaire including
demographic questions, gambling experience questions, and the Lie-Bet Scale (Johnson et al.
Rollover Effect and Goal-gradient Effect
13
1988), participants were asked whether they would like to wager their $20 compensation on
the EGM. Two people walked away with the $20 (an older male and female who arrived
together). All other participants agreed to gamble with their $20. Once these participants
agreed, the $20 was retrieved from them and loaded to the EGM for their subsequent play.
Given the modest sample size, stratified random assignment based on participants’ gender,
age, and Lie-Bet score was utilized to allocate participants to play the EGM in the different
conditions (as described below). Prior to and during the EGM play, each participant had a
finger sensor attached to his or her non-dominant hand, which measured a few physiological
indices including skin conductance. After the EGM play was concluded (i.e., when
participants indicated that they wanted to stop playing or when all credits had been lost),
participants also completed a second questionnaire including questions on their attitudes
toward the EGM utilized in the experiment, the Problem Gambling Severity Index (PGSI,
Ferris and Wynne 2001), and questions on their personality and experiences related to
gambling. Participants were also asked how truthful they had been in the survey. Except one
participant who missed that question, all reported “completely truthful” or “mostly truthful”2.
Finally, participants were debriefed and the remaining credits on the EGM (or $20 when the
credits were less than $20) were paid back to them before they left.
Design
The experiment was based on a 2 (progressive vs. non-progressive) × 2 (deterministic vs.
non-deterministic) × 2 (small jackpot vs. large jackpot) factorial design with an additional no-
jackpot control condition. Participants in the (non-) progressive and the (non-) deterministic
2 The participant who missed that question was included in the analysis, since excluding the participant didn’t
change the substance of the results.
Rollover Effect and Goal-gradient Effect
14
conditions were informed of the mechanisms of the specific jackpot features on their EGM
both verbally and with an instruction screen prior to play.
In order to create authenticity feelings among our participants, we showed them the
real jackpots they had the opportunity to win in the experiment. In the small jackpot
conditions, they were shown a jar with $500 cash and told about the opportunity to win $500
as a cash jackpot. In the large jackpot conditions, we showed them a jar with 500 instant
scratch-it tickets and told them about the opportunity to win instant scratch-it tickets for a
$25,000 jackpot. Note that utilizing instant scratch-it tickets is the only credible way to
appear to offer a high value jackpot. Since participants would doubt our willingness to offer a
real cash prize of $25,000 from the university (as they should), offering instant scratch-it
tickets, which we showed to them, has much greater credibility. A further advantage of this
jackpot manipulation lies in the different winning likelihoods perceived by participants in the
large and small jackpot conditions. Participants might feel that they needed to win twice (i.e.,
win the EGM jackpot first, then win the instant scratch-it ticket game) in the large jackpot
conditions and only needed to win once (i.e., win the EGM jackpot) in the small jackpot
conditions. Consequently, participants should perceive winning the large ticket jackpot as
substantially less likely than the small cash one. Since we were testing the relative attraction
of small but somewhat likely jackpots versus large but very improbable jackpots, the exact
likelihoods that participants estimated were not important, as long as they perceived these
differences.
The instruction described each feature in functional terms without emotive words. As
an example, the deterministic, progressive, large jackpot condition participants were told:
“The $25,000 prize amount will be shown on the top of the screen once you begin. You’ll
notice that the jackpot prize grows with every bet you make. The ticket jackpot will payout
Rollover Effect and Goal-gradient Effect
15
after a certain number of bets have been placed. The number of bets that must be made before
the jackpot is triggered has been determined in advance and at random.”
A design feature worth mentioning was the starting jackpot amounts displayed in the
game interface of the simulated EGM programs. Specifically, $476, $500, $25000, and
$25000 were displayed as the starting amounts for the small progressive, small non-
progressive, large progressive, and large non-progressive jackpot conditions. The intention of
setting $476 for small progressive jackpot EGMs was to ensure the average jackpot size
displayed on those EGMs to be around $500 during the play (i.e., similar to the jackpot size
displayed on the small non-progressive jackpot EGMs). Since all EGMs were programmed to
generate winning outcomes on five out of the first 20 bets and losing outcomes on all other
bets, participants experienced on average one win every four bets in their first 20 bets. Then
when they placed another four bets (i.e., had placed their 24th bet) and did not win, they
should start feeling a decreasing chance of winning. Hence, the 24th bet represented a critical
turning point in the middle of participants’ play. They would keep playing for a while until
they finally realized the inevitability of losing and stopped playing. As for the EGMs with the
small progressive jackpots, their starting jackpot of $476 would grow to $500 after 24 bets
and keep growing for a while until the betting ended, leading to an average jackpot size
around $500 during the entire play. We kept $25000 as the starting amount for large
progressive jackpots since that was a much larger figure and the growth of the jackpot would
not change its average amount to dramatically different from $25000.
Results
Data Analysis
Rollover Effect and Goal-gradient Effect
16
The primary response variables of interest included the behavioral outcomes of average bet
size, betting speed (bets per minute), persistence (number of bets), the self-reported attitudes
toward the EGM utilized in the experiment (a three-item, six-point Likert scale), and the
physiological arousal measured by skin conductance changes due to the EGM play. Each
outcome was analyzed with an ANCOVA model using (non-) progressive feature, (non-)
deterministic feature, and jackpot size as the primary predictor variables in a crossed design.
Gender, age, personal income, and PGSI (Ferris and Wynne 2001) were entered as
covariates, as each of these variables could play certain roles in influencing participants’
gambling behavior (e.g., Browne et al. 2014; Johansson et al. 2009).
Treatment of Missing Data
The missing data were treated through two ways. Firstly, minor missing data were found for
less than 5% of the total participants on their EGM attitudes or PGSI items. These missing
data were replaced through the Expectation Maximization (EM) technique. Secondly, due to
equipment/recording failure, data on 22 participants’ (i.e., about 18% of the total participants)
skin conductance changes were also missing. These participants were excluded from any
analysis involving the physiological arousal (i.e., listwise deletion).
Average Bet Size
The first ANCOVA model used average bet size as the outcome, and showed a significant
main effect of PGSI ( 24.14)98,1( F,001.p,13.
2
p
). A follow-up partial correlation
analysis was conducted between average bet size and PGSI, controlling for gender, age, and
Rollover Effect and Goal-gradient Effect
17
personal income. The result demonstrated that players more at risk of gambling problems
tended to bet more on EGMs with jackpots ( 35.
p
r,001.p).
Consistent with our hypotheses, the ANCOVA model also demonstrated a significant
three-way interaction effect of (non-) progressive feature, (non-) deterministic feature, and
jackpot size on participants’ average bet size (
33.4)98,1( F
,05.p,04.
2
p
). The results
of this ANCOVA model are listed in Table 2, and the three-way interaction is illustrated in
Figure 3.
To further explore the three-way interaction, we decomposed the interaction for
participants in large and small jackpot conditions. When the jackpot was large (Fig. 3, Panel
A), there was a significant two-way interaction between (non-) progressive and (non-)
deterministic features (
16.8)48,1( F
,
01.p
,15.
2
p
). Simple effect revealed that
participants in large deterministic jackpot conditions placed higher bets on the EGM with
non-progressive ( 93.54M cents, 33.23SD ) rather than progressive jackpot ( 04.38M
cents, 83.12SD ,
08.5)48,1( F
,05.p,10.
2
p
). In contrast, participants in large non-
deterministic jackpot conditions, bet on average more on the EGM with progressive (
79.53M cents, 82.26SD ) rather than non-progressive jackpot ( 59.41M cents,
10.14SD ), although the simple effect fell short of significance (
32.3)48,1( F
,10.p,
07.
2
p
).
Fig. 3 Panel B shows the average bet size on small jackpot EGMs in different
conditions. Only the main effect of PGSI was significant ( 11.18)46,1( F,001.p,28.
2
p
), and the PGSI score was positively correlated with the average amount of bet placed on
Rollover Effect and Goal-gradient Effect
18
small jackpot EGMs after controlling for gender, age, and personal income ( 53.
p
r,
001.p). None of the other main or interaction effects were significant ( 05.' sp ).
Furthermore, both simple effects in small deterministic and non-deterministic jackpot
conditions proved non-significant ( 05.' sp ).
Table 2 ANCOVA predicting average bet size from (non-) progressive feature, (non-
) deterministic feature, and jackpot size (large or small)
Variable df MS F ƞp2
Progressive 1 663.23 1.90 .02
Deterministic 1 318.23 .91 .01
Size 1 328.59 .94 .01
Progressive × Deterministic 1 1326.98 3.80 .04
Progressive × Size 1 282.78 .81 .01
Deterministic × Size 1 484.57 1.39 .01
Progressive × Deterministic × Size 1 1512.00 4.33* .04
Gender 1 2.38 .01 .00
Age 1 368.88 1.06 .01
Personal income 1 121.65 .35 .00
PGSI 1 4972.64 14.24*** .13
Error 98 349.15
Total 110
Rollover Effect and Goal-gradient Effect
19
* p < .05; ** p < .01; *** p < .001.
Fig. 3 Average bet size by (non-) progressive feature, (non-) deterministic feature, and
jackpot size (Panel A and B)
Panel A – large jackpot condition Panel B – small jackpot condition
Speed of Betting (Bets per Minute)
The second ANCOVA model, using the speed of betting as the outcome, found no significant
effects of the jackpot features, the jackpot size, or their interactions ( 05.' sp ). Among the
covariates, only the effect of PGSI was significant (
10.4)98,1( F
,05.p,04.
2
p
).
However, the partial correlation between bets per minute and PGSI did not reach
significance, after controlling for gender, age, and personal income ( 18.
p
r,05.p).
Persistence (Number of Bets)
47.15
34.70
52.71
43.29
30
35
40
45
50
55
60
Deterministic Non-deterministic
Average bet size (cents)
Progressive Non-progressive
38.04
53.79
54.93
41.59
30
35
40
45
50
55
60
Deterministic Non-deterministic
Average bet size (cents)
Progressive Non-progressive
Rollover Effect and Goal-gradient Effect
20
The third ANCOVA model examined the total number of bets placed by participants as the
response variable. Since all bets were programmed as losses past the 20th bet, and participants
across all jackpot conditions placed on average 79.32 bets ( 86.32SD , only six participants
placed 20 or fewer bets on jackpot EGMs), the number of bets placed by participants
essentially reflected their persistence while losing. According to the results of this ANCOVA
model, the jackpot features or size did not reliably predict continued play, nor did any
interactions or covariates ( 05.' sp ). Removing those six participants who placed 20 or
fewer bets did not change these analysis results.
EGM Attitudes
Participants’ attitudes toward the EGM they played were measured by a three-item, six-point
Likert scale (the Poker Machine was enjoyable, the Poker Machine was exciting, I would like
to play the Poker Machine again; 1 = strongly disagree, 6 = strongly agree). The average of
the three items was calculated to form an EGM attitudes composite ( 79.
), which was
entered into the fourth ANCOVA model as the response variable. The results showed no
significant effects of the jackpot features, the jackpot size, their interactions, or any covariates
on the EGM attitudes ( 05.' sp ).
Physiological Arousal
A fifth ANCOVA model used as the response variable participants’ physiological arousal due
to the EGM play, which was measured by the changes in skin conductance (i.e., the
differences between the average experimental skin conductance during play and the average
Rollover Effect and Goal-gradient Effect
21
baseline skin conductance prior to play ). None of the jackpot features, the jackpot size, or
their interactions had any significant effects on skin conductance changes ( 05.' sp ).
Among the covariates, only the effect of gender was significant ( 93.5)78,1( F,05.p,
07.
2
p
), with male participants showing larger skin conductance increases than females due
to playing the jackpot EGMs, even after controlling for age, personal income, and PGSI (
23.
p
r,05.p).
No-jackpot Condition
Each of the eight jackpot conditions in the factorial design was compared with the no-jackpot
condition through ANCOVA models using the conditions as the primary predictor variables
and gender, age, personal income, and PGSI as the covariates. None of the eight jackpot
conditions were significantly different from the no-jackpot control condition for average bet
size (all 05.' sp ), bets per minute (all 05.' sp ), number of bets (all 05.' sp ), EGM
attitudes (all 05.' sp ), or physiological arousal (all 05.' sp ).
Discussion
The present experiment sought to examine the effects of (non-) progressive jackpots on EGM
playing behavior, and how (non-) deterministic jackpots and jackpot size may moderate such
effects. In a crossed-design, the results revealed a significant interaction between the (non-)
progressive feature, (non-) deterministic feature, and jackpot size on participants’ average bet
size on the EGM. In particular, the largest bets were made on high jackpot machines
($25,000) that were represented as deterministic (i.e., a jackpot payoff after a fixed number of
Rollover Effect and Goal-gradient Effect
22
bets, which is determined in advance at random) and non-progressive (i.e., payoff at a fixed
jackpot amount). These machines may have appeared more valuable because of the
dominance of goal-gradient effect. The incremental bets may have encouraged larger bet
sizes as the players felt they were drawing nearer to an inevitable payoff event. That is, each
bet places the gambler closer to the goal of winning the jackpot prize. The experiment
demonstrates that intensity of betting, at least by measure of bet size, is largest in the
presence of this type of jackpot. In fact, the average bet size for the large, deterministic and
non-progressive jackpot ( 93.54M cents, 33.23SD ) was 20.3% higher than the bet-size
average for all studied jackpot feature combinations ( 67.45M cents, 33.20SD ).
Importantly, large jackpots that were non-deterministic (potential payoff assessed at
random with each bet) and progressive (each bet adds to the jackpot prize) also promoted
high average bet sizes. This is a common configuration for jackpots in gaming venues, and in
this condition demonstrated bet-sizes ( 79.53M cents, 82.26SD ) were 17.8% higher than
the average for all studied jackpot configurations ( 67.45M cents, 33.20SD ). This
jackpot configuration may be attractive due to the rollover effect that was demonstrated to
make rollover lottery prizes more attractive (Beenstock and Haitovsky 2001; Farrell et al.
1999; Rogers 1998; Rogers and Webley 2001). Players may justify the placing of large bets
that add to the large jackpot prize, as they can potentially be later recouped through attaining
the jackpot prize.
Furthermore, jackpot size affected EGM players’ susceptibility to both rollover and
goal-gradient effects. These effects were observed on high-jackpot machines, but they
disappeared in low-jackpot conditions. Hence, all three findings mentioned above were
consistent with the hypotheses formulated through the knowledge activation theory.
Rollover Effect and Goal-gradient Effect
23
In recent years, the knowledge activation theory has been increasingly utilized in
explaining various addiction phenomena such as smoking or alcohol use (e.g., Cervone et al.
2007; Friedman et al. 2005; Friedman et al. 2009). The present study, to the best of our
knowledge, is among the first gambling research that builds upon the knowledge activation
theory. This critical theory may provide a useful framework for examining other important
gambling phenomena in the future.
In addition to the results that supported the key hypotheses, the present study also
demonstrated that players who reported higher PGSI scores tended to bet more on EGMs
with jackpots, and that males experienced larger physiological arousal than females due to
playing the jackpot EGMs. These findings also add to the growing gambling literature on
PGSI or gender effects (e.g., Browne et al. 2014; Johansson et al. 2009).
Limitations
As is true for all lab-based experimental studies, there are concerns about the external validity
of the results. The artificial environment of the lab, as well as the way in which the jackpots
were described, may not be entirely true to the information players receive in a real venue.
Nevertheless, the jackpots were described in simple and functional language, and these
descriptions produced reliable effects on average bet size. We take a view on the importance
of experimental realism over mundane realism, where it is important to understand the
psychological constraints and contingencies that operate in jackpot EGMs (offering real
money prizes), rather than attempting to faithfully recreate every detail of the gambling
environment.
Rollover Effect and Goal-gradient Effect
24
Conclusion
This experiment provided some evidence that common jackpot configurations are associated
with higher average bet sizes, which are one component of gambling intensity. That such
jackpot configurations are relatively common in venues is likely a consequence of the natural
evolution of EGMs, where only the most popular and profitable machines survive
competition on the gaming floor.
References
Beenstock, M., & Haitovsky, Y. (2001). Lottomania and other anomalies in the market for
lotto. Journal of Economic Psychology, 22, 721-722.
Blaszczynski, A., Gainsbury, S., & Karlov, L. (2014). Blue Gum gaming machine: An
evaluation of responsible gambling features. Journal of Gambling Studies, 30(3), 697-
712.
Brown, J. S. (1948). Gradients of approach and avoidance responses and their relation to
level of motivation. Journal of Comparative and Physiological Psychology, 41(6),
450-465.
Browne, M., Langham, E., Rockloff, M. J., Li, E., Donaldson, P., & Goodwin, B. (2014).
EGM jackpots and player behavior: An in-venue shadowing study. Journal of
Gambling Studies. Advance online publication. doi: 10.1007/s10899-014-9485-y
Cervone, D., Orom, H., Artistico, D., Shadel, W. G., & Kassel, J. D. (2007). Using a
Rollover Effect and Goal-gradient Effect
25
knowledge-and-appraisal model of personality architecture to understand consistency
and variability in smokers' self-efficacy appraisals in high-risk situations. Psychology
of Addictive Behaviors, 21(1), 44-54.
Easterbrook, J. A. (1959). The effect of emotion on cue utilization and the organization of
behavior. Psychological Review, 66(3), 183-201.
Eisenberger, R., & Selbst, M. (1994). Does reward increase or decrease creativity?. Journal
of Personality and Social Psychology, 66(6), 1116-1127.
Farrell, L., Morgenroth, E., & Walker, I. (1999). A time series analysis of U.K. lottery sales:
Long and short run price elasticities. Oxford Bulletin of Economics and Statistics,
61(4), 513-526.
Ferris, J., & Wynne, H. (2001). The Canadian Problem Gambling Index: Final report.
Prepared for the Canadian Centre on Substance Abuse.
Friedman, R. S., McCarthy, D. M., Förster, J., & Denzler, M. (2005). Automatic effects of
alcohol cues on sexual attraction. Addiction, 100(5), 672-681.
Friedman, R. S., McCarthy, D. M., Pedersen, S. L., & Hicks, J. A. (2009). Alcohol
expectancy priming and drinking behavior: The role of compatibility between prime
and expectancy content. Psychology of addictive behaviors, 23(2), 329-333.
Harrigan, K., Dixon, M., & Brown, D. (2014). Modern multi-line slot machine games: The
effect of lines wagered on winners, losers, bonuses, and losses disguised as wins.
Rollover Effect and Goal-gradient Effect
26
Journal of Gambling Studies. Advance online publication. doi: 10.1007/s10899-013-
9436-z
Higgins, E. T. (1996). Knowledge activation: Accessibility, applicability, and salience. In E.
T. Higgins & A. Kruglanski (Eds.), Social Psychology: Handbook of basic principles
(pp. 133-168). New York: Guilford.
Hull, C. L. (1934). The rats’ speed of locomotion gradient in the approach to food. Journal of
Comparative Psychology, 17(3), 393-422.
Johansson, A., Grant, J. E., Kim, S. W., Odlaug, B. L., & Götestam, K. G. (2009). Risk
factors for problematic gambling: A critical literature review. Journal of Gambling
Studies, 25(1), 67-92.
Johnson, E. E., Hamer, R. M., & Nora, R. M. (1988). The Lie/Bet questionnaire for screening
pathological gamblers: A follow-up study. Psychological Reports, 83(3 Part 2), 1219-
1224.
Kivetz, R., Urminsky, O., & Zheng, Y. (2006). The goal-gradient hypothesis resurrected:
Purchase acceleration, illusionary goal progress, and customer retention. Journal of
Marketing Research (JMR), 43(1), 39-58.
Krech, D., & Crutchfield, R. S. (1948). Theory and problems of social psychology. New
York: McGraw-Hill.
Millhouse, H., & Delfabbro, P. (2008). Investigating preferences for gaming machine
Rollover Effect and Goal-gradient Effect
27
features in problem and non-problem gamblers using a consumer choice methodology.
Gambling Research: Journal of the National Association for Gambling Studies
(Australia), 20(1), 37-49.
Rockloff, M. J., Donaldson, P., & Browne, M. (2014). Jackpot expiry: An experimental
investigation of a new EGM player-protection feature. Journal of Gambling Studies.
Advance online publication. doi: 10.1007/s10899-014-9472-3
Rockloff, M. J., Greer, N., & Evans, L. G. (2012). The effect of mere presence on electronic
gaming machine gambling. Journal of Gambling Issues, (27), 1-14.
Rockloff, M. J., Greer, N., Fay, C., & Evans, L. G. (2011). Gambling on electronic gaming
machines is an escape from negative self reflection. Journal of Gambling Studies,
27(1), 63-72.
Rockloff, M., & Hing, N. (2013). The impact of jackpots on EGM gambling behavior: A
review. Journal of Gambling Studies, 29(4), 775-790.
Rogers, P. (1998). The cognitive psychology of lottery gambling: A theoretical review.
Journal of Gambling Studies, 14(2), 111-134.
Rogers, P., & Webley, P. (2001). “It could be us!”: Cognitive and social psychological
factors in UK National Lottery play. Applied Psychology: An International Review,
50(1), 181-199.
Sharpe, L., Walker, M., Coughlan, M. J., Enersen, K., & Blaszczynski, A. (2005). Structural
changes to electronic gaming machines as effective harm minimization strategies for
non-problem and problem gamblers. Journal of Gambling Studies, 21(4), 503-520.
Wohl, M. J., Gainsbury, S., Stewart, M. J., & Sztainert, T. (2013). Facilitating responsible
Rollover Effect and Goal-gradient Effect
28
gambling: The relative effectiveness of education-based animation and monetary limit
setting pop-up messages among electronic gaming machine players. Journal of
Gambling Studies, 29(4), 703-717.
Ziolkowski, S. (2014). The World Count of Gaming Machines 2013. Prepared for the Gaming
Technologies Association.
