Exploring Seniors’ Casino Gaming Intention

Senior Gaming Motivations

In the past few decades, casino gaming has gained exposure as a new form of recreation and entertainment for the senior population; as a result, many researchers and marketers have initiated studies to learn more about this market. While profiling the older casino gaming market, researchers have debated the underlying motives of senior casino gamblers. Motivation has been recognized in the literature as an important driving force encouraging participants to engage in different types of behaviors (Jang, Bai, Hu, & Wu, 2009). Motivation is the key element influencing seniors’ decision-making processes in patronizing a casino and, thus, a better understanding will help casino operators meet senior casino visitors’ diverse needs.

Several senior gaming motivations can be identified from the existing literature. Escaping from their ordinary routine is one motivation for seniors to participate in gambling. Seniors have reported that they want to get away from their normal living environment, such as retirement communities or senior care facilities. Lifestyle changes that often accompany aging (widowhood, declining health, and learning to live on fixed incomes) can act as catalysts for negative feelings. Seniors can also be lonely and bored with uneventful living environments or a lack of social involvement (Sullivan, 2001). They might be attracted to casino gaming as a release for some of their negative feelings and boredom. In fact, casinos can be a perfect platform for temporarily forgetting negative feelings (Smith & Abt, 1984). Relieving feelings of isolation, loneliness, and boredom is cited as the number one reason middle-aged and older women gamble (Chrostowski, 1997).

The biggest attraction for older adults visiting casinos is the fun and excitement of the casino experience. The Las Vegas Convention and Visitors Authority (1996) claimed that casinos could provide a special form of excitement and entertainment. Hope and Havir (2002) and McNeilly and Burke (2001) found that casino gaming has become a popular form of entertainment for most senior participants. Other studies have found similar results (American Gaming Association, 2002; National Opinion Research Center, 1999; O’Brien Cousins, Withcer, & Moodie, 2002). Just being in a casino can be fun. Something is always happening in gaming halls. Watching people win or just passing time can be very entertaining for seniors. The bright lights in the gaming halls and sounds from game machines can provide a multisensory experience for seniors (Loroz, 2004).

Seniors might also participate in casino gambling for the opportunity to interact socially with others. Most elderly female participants from Michigan (Tarras, Singh, & Moufakkir, 2000) and elderly Detroit residents (Zaranek & Chapleski, 2005) considered going to casinos an opportunity for social activity. In relation to the first motivation, trips to a casino provide seniors with a chance to meet and talk to other people outside of their daily routines. Interestingly, McNeilly and Burke (2001) found that many seniors take more day-trips to casinos than any other social activity, such as trips to shopping malls, local libraries, zoos, theaters, or restaurants. Hagen et al. (2005) also found that among the participants in their study, gambling was viewed as a “pleasant social activity.” The social aspects of gambling also helped participants combat loneliness.

Life experiences may further explain seniors’ attraction to casino gambling. Many seniors 65 years and older were raised during the Great Depression, and casino gambling was something they could not afford to enjoy. They had nothing to waste, and they worked hard for their money. Spending money on any type of entertainment or recreation was unimaginable when they barely made enough to make ends meet. Not only did they have little money to spend on gambling, it was also not legally available or acceptable when they grew up. Gambling was also considered sinful under the cultural norms of the time. Socially and personally, the situation is now very different. Casino gaming is viewed as a pleasant and legal leisure activity. After providing for all the needs of their families, and with more time and income at their disposal, seniors can now afford an interest in casino gaming. With the availability of casino gambling in their later life (Hope & Havir, 2002), seniors may be curious about this leisure activity, making curiosity another major motivation for casino activity.

According to Hope and Havir (2002), the prospect of winning money is not a major motive for seniors playing casino games. However, they also found that men tended to be more curious and more likely to go to win money than women, who were more likely to go for fun. The money seniors spend on their gaming (money lost) at the casino is viewed as entertainment expenses, just as they would spend money on any other type of leisure or recreational activity. Most of these seniors are fully aware that the odds are always against them and set their limits for responsible gambling. However, some older adults go to casinos to win money, and it is difficult to entirely disregard winning money as a reason seniors play casino games.

In sum, the five most common senior motivations for visiting casinos extracted from past studies include escaping, enjoyment (fun, excitement, and entertainment), socializing, curiosity, and winning. These five motivation dimensions were thus used for further analysis in this study.

Theory of Planned Behavior

The main concept of TPB is that most human behavior is under volitional control. People engage in actions because they want to act in a certain behavioral way, and their conscious motives trigger them to engage in that action (Ajzen & Fishbein, 1980). Volitional behaviors are influenced by behavioral intention, which is the likelihood to act (Fishbein & Ajzen, 1975) and the immediate determinant of a behavior (Ajzen, 1985). Senior casino gaming behavior and behavioral intentions can be expressed in the following algebraic equation:

B ≈ BI = ( AT ) W 1 + ( SN ) W 2 + ( PBC ) W 3 ,

where B is the expected casino gaming behavior and BI is senior casino gaming (behavioral) intention. The approximate sign (≈) between B and BI indicates that measuring gaming intention can predict seniors’ gaming behaviors only when the intention does not change from the predictors of intention. Fishbein and Ajzen (1975) suggested that the proximal cause of a behavior is the intention to enact the behavior. This means that researchers need an accurate measurement of casino gaming intention to fully understand senior casino gaming. The antecedents of intentions are better understood than the antecedents of behavior, so this study aims only to accurately measure senior casino gaming intention, not the behavior itself. AT in the equation is thus seniors’ attitudes toward casinos, SN is the subjective norm, and PBC is perceived behavioral control. The signs w₁ through w₃ are empirical weights indicating the relative importance of the three terms in the model.

TPB claims that attitude toward a behavior is, at the most basic level, a function of behavioral beliefs and outcome evaluations. Attitude thus derives from behavioral beliefs, referring to the subjective probability that a person’s behavior will lead to a certain consequence. This also includes outcome evaluations, which refer to a person’s evaluation of each consequence (Ajzen & Fishbein, 1980). The value of the outcome evaluation contributes to the attitude toward the behavior. Applied to senior casino patronage, a senior who perceives the consequences of going to casinos as positive and considers these consequences important would have a favorable attitude toward visiting casinos. He or she might evaluate playing casino games as an activity that provides fun, excitement, and the opportunity to meet people and may thus view these as important outcomes of going to a casino. Thus, this individual holds a positive attitude toward casinos and is more likely to intend to visit a casino (S. Moore & Ohtsuka, 1999; Oh & Hsu, 2001). On the other hand, a senior who believes that casinos are sinful places and a way to take people’s money, outcomes they consider negative, will hold a negative attitude toward casinos. Thus, he or she is less likely to visit a casino to play games.

Subjective norm refers to the perceived social pressure from important others to perform or not to perform a behavior (Ajzen & Fishbein, 1980). People are more likely to perform a behavior when they get support from their referents than when they do not get support from their referents (Ajzen & Fishbein, 1980). Subjective norm is also determined by an individual’s normative beliefs and motivation to comply with the beliefs of others who are important to them. Normative beliefs are based on perceptions of other people’s preferences about whether an individual should or should not perform a behavior (Ajzen & Fishbein, 1980). In terms of casino gaming behavior, normative beliefs can be measured by rating the statement “my spouse thinks that I should/should not participate in casino gaming.” The motivation to comply is the individual’s tendency to conform to the expectations of people who are important to them. It is the level of a person’s willingness to act in accordance with the ways in which their referents want them to act. For example, one can measure the level of motivation to comply by rating the statement “I want to do what my spouse thinks that I should do.” Here, the individual’s decision to play casino games relies heavily on the referent’s (spouse) approval or disapproval of casino gaming. When a senior believes that his or her most important referents (e.g., spouse, children, and friends) approve of casino gaming, then he or she is more likely to go to a casino to game.

Perceived behavioral control is the degree to which an individual feels that the performance or nonperformance of the behavior in question is under his or her volitional control (Ajzen, 1985, 1988). It is “the person’s belief as to how easy and difficult performance of the behavior is likely to be” (Ajzen & Madden, 1986, p. 457). Ajzen (2002) asserted that, all else being equal, having a high level of perceived control reinforces an individual’s intention to perform a behavior and increases his or her effort and determination to act on the behavior. Perceived behavioral control is determined by control beliefs, referring to a person’s beliefs about the presence of factors that may facilitate or impede performance of the behavior (Ajzen, 2001). Perceived behavioral control is measured by multiplying the strength of their beliefs and the power of the control factor to facilitate or inhibit the performance of a behavior; the resulting products can then be summed up across all control beliefs. Seniors’ casino patronage decisions could be influenced by various perceived constraints and barriers. Inhibiting factors could be either internal (e.g., skills, abilities, poor health) or external (e.g., time, transportation, insufficient financial resources). For instance, even if someone has a positive attitude toward casino gaming and approval from important referents, if the individual does not live close to a casino or does not have transportation to a casino, casino gaming will be perceived as a difficult activity. Therefore, the intention to participate in casino gaming will decrease. Oh and Hsu (2001) studied the role of perceived behavioral control by measuring four control factors (budgetary affordability, time availability, self-control, and gambling skills) and found that all control factors except budgetary affordability significantly affected gambling behavioral intention.

Past Experience

Trafimow (2000) claimed that when a person engages in novel behaviors, the variables (attitude, subjective norms, and perceived behavioral controls) used in the TPB will predict the intention well. However, when a person engages in a behavior out of habit, the predictive power of the variables will be reduced. Accordingly, many empirical studies include past behavior as a predictor of behavioral intentions and future behavior (Bentler & Speckart, 1981; Quellette & Wood, 1998) in the TPB. Triandis (1980) first suggested including past behavior in the TPB, noting that learned behavior from repeated performance is another cause of behavior. This line of theoretical development posits that if an individual habitually engages in a particular behavior, that person has no need to perform the evaluations and reasoning assumed by the TPB. For example, seniors who have already participated in casino gambling might not go through the cognitive evaluations (attitude, subjective norms, and perceived behavioral controls), making their decision based on past experience instead. They will be more familiar with various facilities, services, types of machines and games, and even people in the casinos than seniors who have not visited casinos at all or do not go to casinos regularly. Behaviors that are more habitual than planned can be measured directly from the repeated past performance of the behavior. The direct effect of past casino experience on intentions means that seniors who have been to casinos in the past and who are regular casino goers are more likely to visit casinos without hesitation.

This research investigates whether attitude, subjective norms, and perceived behavioral controls, along with motivation, influence seniors’ casino patronage intentions differently based on whether seniors perceive themselves as regular or irregular casino patrons. As Ajzen (1991) explained, the relative importance of attitude, subjective norms, and perceived behavioral controls in predicting behavioral intention should vary across behaviors and situations; thus, investigating the moderating effects of seniors’ past casino experiences in the model should enhance our understanding of senior casino patronage behaviors. In summary, Figure 1 exhibits the extended TPB model to explain senior casino gaming intention.

OPEN IN VIEWER

Figure 1

A Proposed Model

Measurement of Variables

We set up a questionnaire with three parts. The first part assessed the TPB model based on three levels, behavioral intentions based on the three predictor variables (attitude, subjective norm, and perceived behavioral control), each rooted in a relevant belief (behavioral beliefs for attitude, normative beliefs for subjective norm, and control beliefs for perceived behavioral control; Ajzen, 2002). For attitude, respondents assessed their attitude toward casino gaming based on the statement “All things considered, for me, going to a casino would be _______ . . .” A set of six bipolar adjectives were provided for respondents to complete the sentence: not enjoyable/enjoyable, unpleasant/pleasant, bad/good, boring/fun, harmful/beneficial, and foolish/wise. The attitude construct is based on a behavioral belief, the sum of the belief strength multiplied by the outcome evaluation (∑bb_i · be_i). Thus, two questions were asked for each of the four items based on a 7-point Likert-type scale. For behavioral beliefs (bb_i), respondents evaluated their belief strengths based on four benefits of visiting casinos. For the outcome evaluations (be_i), respondents evaluated four salient beliefs about visiting casinos.

Subjective norm was tested by asking respondents to rate a relevant referent’s level of approval of the respondent patronizing a casino. Three relevant referents were extracted from leisure literature on older adults: spouses, children, and friends. Both 7-point Likert-type and semantic differential scales were used. Next, subjective norm was represented by a normative belief construct, which is the sum of normative beliefs (nb_i) multiplied by the motivation to comply (mc_i) with each referent group (∑nb_i · mc_i). The normative beliefs (nb_i) were measured by asking respondents to rate on a 7-point scale the level of influence each referent group has on the respondent’s decision to visit a casino. Motivation to comply (mc_i) was measured by a respondent’s general motivation to comply with his or her referent’s opinion, using a 7-point scale.

Perceived behavioral control was tested, using both 7-point Likert-type and semantic differential scales, by asking questions about respondents’ confidence in visiting casinos. This was determined by testing respondents’ self-efficacy for and control over going to casinos. The point of determining self-efficacy was to measure how easy or hard it was to get to a casino. The point of determining controllability was to measure how much personal control respondents have over going to casinos. The control belief is the basis for perceived behavioral control. Three control belief items (transportation, proximity to a casino, and health conditions) were identified from the senior leisure literature. Control beliefs reflect perceived behavioral control and consist of two parts, control beliefs (cb_i) and perceived control power (pp_i). Control beliefs (cb_i) were measured by asking respondents to rate how much each of the control belief items influenced their decision to go to a casino using a 7-point Likert-type scale. Perceived control power (pp_i) was measured using a 7-point Likert-type scale by asking the respondents the level of self-control they believed they had over each of the control beliefs. The control belief construct is also the sum (∑cb_i · pp_i) of control beliefs (cb_i) multiplied by perceived control power (pp_i).

The questionnaire also asked respondents to evaluate the level of their motivation to go to a casino. A statement, “If I were to go to a casino, I would go there to _______. . .,” was presented. Thirty-four items drawn from existing gaming motivation studies (Chantal, Vallerand, & Vallières, 1994; Lee, Lee, Bernhard, & Yoon, 2006; Neighbors, Lostutter, Cronce, & Larimer, 2002) were provided to complete the sentence.

Respondents’ intentions to visit casinos in the future were measured using three questions that probed their likelihood of visiting casinos in the near future. These three questions were all modified from Ajzen’s (2002) study: “I would like to visit a casino in the near future”; “I intend to visit a casino in the near future”; “I plan to visit a casino in the near future.” Respondents’ past casino visits were measured by asking whether they regularly visited casinos. The respondents’ demographic information, gender, ethnic background, education, and annual income, was also collected.

Data Collection

In this study, seniors are defined as those who are 65 years and older (Jang & Ham, 2009; Jang, Ham, & Hong, 2007).The target population for this study was a consumer database of U.S. citizens who are 65 years or older. An online survey was administered using an external market research firm that maintains a by-invitation-only panelist database. Each person invited to be a panelist had to answer a list of questions sent by the research firm. The online questionnaires were distributed to the panelists via e-mail invitations. A soft launch was conducted to ensure the proper functioning of the survey and e-mailing system. Sixty one completed responses were collected in the soft launch and were used in a pilot test of the study. All Cronbach’s alphas ranged from .95 (intention) to .72 (behavioral beliefs), which is sufficient for internal consistency (Nunnally, 1978). The main launch went out to 5,000 e-mail addresses, and a total of 681 participants filled in the survey questionnaire. The response rate was 13.62%. Of the 681 respondents, 61.1% were female (n = 418) and 38.5% were male (n = 263). Most respondents (92%, n = 629) were Caucasian/White, highly educated (53.2%, 4-year college or graduate degree), and married (72.1%, n = 493). A total of 66% of the respondents reported a household income of more than $40,000 in the previous year (64.4%, n = 440), owned their homes (90.4%, n = 618), and were retired (54.1%, n = 370).

Data Analysis

The data were screened for any violations of underlying assumptions by conducting descriptive statistics using the Statistical Package for the Social Sciences (SPSS v. 17). The multivariate and univariate outliers were checked. All cases were retained for further analysis. The data were then analyzed according to Anderson and Gerbing’s (1988) two-step approach, which involves a preliminary step of confirmatory factory analysis (CFA) to test whether the measured variables reliably reflect the hypothesized latent variables using AMOS (analysis of moment structures), a statistical analytic software used in implementing structural equation modeling (SEM; Arbuckle, 2007). First, construct reliability and validity of the construct measurements were tested, as well as the overall fit of the measurement model. Then, the structural model was tested to investigate the relationships among the constructs. Finally, metric invariance (i.e., measurement and structural invariance) tests were conducted to assess the moderating effect of seniors’ past casino experiences.

Before doing the two-step analysis, the motivation measurement items were purified, and the scale’s psychometric properties were examined using various tests, such as item analyses, exploratory factor analyses (EFAs), CFA, assessments of scale reliability, and unidimensionality, as well as convergent and discriminant validity (Anderson & Gerbing, 1988). Seven items exhibited low correlations (.5 or less) and were deleted after careful consideration. An EFA with varimax rotation was carried out using the 27 remaining motivation items. Six more items with low communalities, high cross-loadings, and low loadings were deleted. At this point, 21 motivation items remained, representing five motivation factors, which are Winning, Socializing, Enjoyment, Curiosity, and Escaping. The cumulative total variance explained approximately 67.93% of the factors. The Kaiser–Meyer–Olkin measuring of sample accuracy was .90, well above the recommended threshold of .60 (Tabachnick & Fidell, 2001). All five factors had Cronbach’s alphas higher than .70, indicating good reliability (Hair, Anderson, Tatham, & Black, 1998). A confirmatory factor model using the maximum likelihood method was estimated to improve measurement properties in the proposed scale (Anderson & Gerbing, 1988; Bagozzi, 1980). The model fit indices from the initial CFA did not show an acceptable threshold, χ²(121) = 487.58, p = .000; normative fit index (NFI) = .88; comparative fit index (CFI) = .91; root mean square error of approximation (RMSEA) = .082. Three more items were deleted after inspection of item squared multiple correlations and modification indices. The results of the second CFA using 18 items showed improved model fit, χ²(117) = 383.01, p = .000; NFI = .91; CFI = .93; RMSEA = .07. The result of the final CFA indicated that each observed item on the latent constructs showed high loadings from .64 to .83. The Cronbach’s alpha estimates ranged from .70 to .87, and all average variances extracted (AVE) ranged from .51 to .60, indicating acceptable reliability. We also compared the AVE with the squared correlations (Fornell & Larcker, 1981) and found that the squared correlations (ranging from .12 to .28) between each pair of constructs were all less than the AVE (ranging from .51 to .60), providing evidence of discriminant validity. More detailed information about this process can be found in a study done by Phillips, Jang, and Canter (2010).

To make the model more parsimonious, senior casino visiting intention was regressed against all five motivation factors (Winning, Socializing, Enjoyment, Curiosity, and Escaping). The test showed that the dimensions Socializing (β = 1.17, p < .05) and Escaping (β = .99, p < .05) did not significantly influence casino visiting intention. Thus, two motivation constructs, Socializing and Escaping, were deleted from the model and only Winning, Enjoyment, and Curiosity were included for further analysis in testing the model.

Measurement Model

A CFA for the proposed model using 10 constructs (attitude, behavioral beliefs, subjective norm, normative beliefs, perceived behavioral control, control beliefs, Winning, Enjoyment, Curiosity, and Intention) was then conducted, and the model showed an acceptable level of fit indices, χ²(355) = 1230.9, p = .00; NFI = .93; CFI = .95; RMSEA = .05. Overall, the measurement model showed a good fit for the data. Convergent validity was assessed by the significant loadings between the observed variables and each latent variable (Anderson & Gerbing, 1988). All observed variables were loaded at least .50 on their delegated latent variables and were statistically significant (p < .01). As shown in Table 1, all AVE were more than the recommended threshold value of .50 (Fornell & Larcker, 1981), ranging from .55 to.88, which supported adequate internal consistency. Next, the composite reliabilities of all constructs exceeded the cutoff value of .70 (Hair et al., 1998), ranging from .71 to .96. Thus, the multiple item scales were acceptable for measuring each of the constructs. Comparing the AVE with the squared correlations between constructs tested for discriminant validity (Fornell & Larcker, 1981). The squared correlations between each pair of constructs were all less than the AVE. Thus, discriminant validity was satisfied. Overall, the measurement model showed good fit to the data.

OPEN IN VIEWER

Table 1

Measure Correlations, the Squared Correlations, and Measurement Properties (N = 681)

Correlations Between Latent Constructs (Squared)
Measures	BB	NB	CB	AT	SN	PBC	Winning	Enjoyment	Curiosity	BI	AVE
BB	1										0.61
NB	.32 (.10)	1									0.61
CB	.23 (.05)	.17 (.03)	1								0.65
AT	.49 (.24)	.28 (.08)	.20 (.04)	1							0.68
SN	.40 (.16)	.55 (.31)	.14 (.02)	.36 (.13)	1						0.79
PBC	.35 (.12)	.29 (.08)	.03 (.00)	.26 (.07)	.36 (.13)	1					0.54
Winning	.41 (.17)	.10 (0.1)	.14 (.02)	.32 (.10)	.18 (.03)	.10 (.01)	1				0.55
Enjoyment	.56 (.31)	.21 (.04)	.19 (.04)	.44 (.19)	.37 (.14)	.26 (.07)	.43 (.18)	1			0.61
Curiosity	.21 (.04)	.03 (.00)	.12 (.01)	.15 (.02)	.13 (.02)	.00 (.00)	.39 (.15)	.34 (.12)	1		0.55
BI	.55 (.30)	.31 (.10)	.19 (.04)	.51 (.26)	.37 (.14)	.36 (.13)	.39 (.15)	.40 (.16)	.12 (.01)	1	0.88
Mean	5.6	5.13	4.7	5.41	5.17	6.10	3.9	5.19	3.65	5.64
SD	0.92	0.95	1.6	1.21	1.38	1.01	1.47	1.33	1.48	1.5
Composite reliability	0.75	0.82	0.79	0.91	0.92	0.78	0.86	0.76	0.71	0.96

Note: BB = behavioral beliefs; NB = normative beliefs; CB = control beliefs; AT = attitude; SN = subjective norm; PBC = perceived behavioral control; BI = behavioral intention; AVE = average variance extracted. Model measurement fit: χ² = 1230.93 (df = 355, p < .001), RMSEA = 0.052, CFI = 0.95, NFI = 0.93.

Structural Model

A structural model was estimated to examine the hypothesized relationships in the extended TPB model with external constructs of senior gaming motivations. The results showed that the goodness-of-fit indices (goodness-of-fit statistics: χ²(375) = 1519.43, p < .00, CFI = .93, NFI = .91, RMSEA = .06) exceeded their acceptance level, suggesting that the model is adequate. Behavior beliefs explained about 41% of the variance in seniors’ attitudes toward casinos, and normative beliefs explained approximately 31% of the variance in subjective norm. The predictor variables explained 37% of the variance in overall intention.

Table 2 shows that all t values, except the path from perceived behavioral control to behavioral intention, were significant at the .01 level. The t values of coefficients between behavioral beliefs and attitude (β _{BB AT} = .64, t = 15.42) and normative beliefs and subjective norm (β _{NB SN} = .56, t = 15.24) indicated a significant positive relationship. However, the control beliefs did not influence perceived behavioral control (β _{CB PBC} = −.04, t = −0.98). All three predictor variables for BI in the TPB model were found to have positive significant influences on casino visiting BI (β _{AT BI} = .31, t = 8.87; β _{SN BI} = .12, t = 4.22; β _{PBC INT} = .22, t = 7.85). Not all the senior casino visiting motivation constructs had positive significant relationships with casino visiting intention. Only the Winning and Enjoyment motivation constructs had positive significant effects on intention (β _{WIN BI} = .23, t = 5.49; β _{ENJOY BI} = .17, t = 3.37). However, the last motivation factor, Curiosity, was negatively associated with visiting intention (β _{CURIO BI} = −.10, t = −2.55).

OPEN IN VIEWER

Table 2

Standardized Maximum Likelihood Parameter Estimates (N = 681)

Paths			Standardized Coefficients	t Value	Results
BB	→	AT	0.64	15.42**	Supported
NB	→	SN	0.56	15.24**	Supported
CB	→	PBC	−0.04	−0.98	Not supported
AT	→	BI	0.31	8.87**	Supported
SN	→	BI	0.12	4.22**	Supported
PBC	→	BI	0.22	7.85**	Supported
Winning	→	BI	0.23	5.49**	Supported
Enjoyment	→	BI	0.17	3.37**	Supported
Curiosity	→	BI	−0.10	−2.55	Not supported
R2(AT)			0.41
R2(SN)			0.31
R2(PBC)			0.01
R2(Intention)			0.37

Note: BB = behavioral beliefs; NB = normative beliefs; CB = control beliefs; AT = attitude; SN = subjective norm; PBC = perceived behavioral control; BI = behavioral intention; CFI = comparative fit index; NFI = normative fit index; RMSEA = root mean square error of approximation. Goodness-of-fit statistics: χ² = 1519.43 (df = 375, p < .001), χ²/df = 4.05; CFI = .93, NFI = .91, RMSEA = .06.

*

p < .05. **p < .01.

The unexpected significant negative relationship between Curiosity and intention can be explained by a suppressor variable (Horst, 1941). To determine if the negative relationship between Curiosity and intention was affected by a suppressor, a series of regression analyses was conducted. The analysis showed that two other motivation factors, Winning and Enjoyment, had something to do with the suppression effect. When these two variables are combined with Curiosity, Winning makes the relationship between Curiosity and intention insignificant (β = −.04) and Enjoyment makes the relationship even more negatively significant (β = −.25). This is a negative suppression (Darlington, 1968), which occurs when a variable (Curiosity, in this case) has a negative weight after being included in a regression equation where all the variables have positive intercorrelations. This negative suppression could be caused by a variable in the regression equation that has a negative correlation with the dependant variable or a variable that has a high correlation with another predictive variable (McNemar, 1945). However, neither of these cases seems relevant to this study. None of the variables showed a negative correlation with the criterion variable, and the primary independent variables within the TPB model have acceptable correlations (Table 1). The multicollinearity among independent variables does not seem to have caused the negative suppression either.

Measurement Invariance and Structural Invariance Test

The moderating effects of past casino experience were examined by testing the measurement invariance. First, the sample was split between regular casino visitors and irregular casino visitors. Of the 681 total samples, 533 considered themselves regular casino visitors, and 148 considered themselves irregular visitors. Measurement invariance was accessed by conducting chi-square difference tests. A model is invariant, and full invariance is supported, when no significant chi-square difference exists between the nonrestricted model and the full metric invariance model (restricting factor loadings across two groups; Yoo, 2002). The moderating effects of past casino visits between predictor variables (attitude, subjective norm, perceived behavioral control, and motivations) and intention were the focus of this study. Therefore, attitude, subjective norm, perceived behavioral control; three motivation factors (Winning, Enjoyment, and Curiosity); and intention were included in the chi-square testing model.

Table 3 shows the results of the measurement invariance test. The chi-square difference was significant between the freely estimated base model and the fully restricted model, Δχ²(16) = 46.5, p < .01; full metric invariance was not supported. This means that the factor loadings between casino visitors and nonvisitors were not the same. When full metric invariance is not supported, Yoo (2002) recommends a partial invariance test. After a careful examination of the modification indices and changes in parameter, four of the invariance constraints across two groups were relaxed, and a partial metric invariance model was supported, Δχ²(12) = 17, p > .01.

OPEN IN VIEWER

Table 3

Measurement Invariance Test

Models	χ2	df	Δχ2	Δdf	NFI	CFI	RMSEA
Base model	955.5	408	—	—	.93	.96	.04
Full metric invariance of CFA	1002.0	424	46.5 a	16	.93	.96	.04
Partial metric invariance of CFA	972.45	420	17.0 b	12	.93	.96	.04

Note: CFA = confirmatory factor analysis; NFI = normative fit index; CFI = comparative fit index; RMSEA = root mean square error of approximation.

a.

Chi-square difference test Δχ²(16) = 46.5, p < .01 (significant), full metric invariance is not supported.

b.

Chi-square difference test Δχ²(12) = 17.0, p > .01 (insignificant), thus partial metric invariance is supported.

A structural invariance test assessed whether the parameter estimates were equivalent across the two groups. The test was performed much like the measurement invariance test. First, to create the base model, a structural equation model (partial metric invariance) was simultaneously estimated across regular visitor and irregular visitor groups. A chi-square difference test was conducted between the base model (the partial metric invariance of CFA) and a full metric invariance structural model in which all proposed causal paths were fixed to be invariant across the groups. As shown in Table 4, both models fit the data well. The results showed that the chi-square difference between the partial metric invariance SEM and the full path invariance model was insignificant (p > .05), indicating that the paths across regular casino visitors and irregular visitors were not different.

OPEN IN VIEWER

Table 4

Structural Invariance Test

Models	χ2	df	Δχ2	Δdf	NFI	CFI	RMSEA
Partial metric invariance of structural model	972.5	420	—	—	.93	.96	.04
Full path invariance model	991.9	426	19.4 a	6	.93	.96	.04

Note: CFA = confirmatory factor analysis; NFI = normative fit index; CFI = comparative fit index; RMSEA = root mean square error of approximation.

a. Chi-square difference test Δχ²(6) = 19.4, p > .01 (insignificant), thus paths across two groups are not different.