Applying Behavioral Economic Theory to Problematic Internet Use: An Initial Investigation

Abstract

The widespread availability of the Internet has had profound social, educational, and economic benefits. Yet, for some, Internet use can become compulsive and problematic. The current study seeks to apply a behavioral economic framework to Internet use, testing the hypothesis that, similar to other addictive behaviors, problematic Internet use is a reinforcer pathology reflecting an overvaluation of an immediately acquirable reward relative to prosocial and delayed rewards. Data were collected through Amazon’s Mechanical Turk data collection platform. A total of 256 adults (M_age=27.87, SD = 4.79; 58.2% White, 23% Asian; 65.2% with an Associate degree or greater) completed the survey. Measures of delay discounting, consideration of future consequences, Internet demand, and alternative reinforcement all contributed unique variance in predicting both problematic Internet use and Internet craving. In aggregate models controlling for all significant predictors, alternative reinforcement and future valuation variables contributed unique variance. Individuals with elevated demand and discounting were at greatest risk for problematic Internet use. Consistent with behavioral economic research among substance abusing samples, individuals engaging in heavy Internet use report elevated motivation for the target behavior coupled with diminished motivation for other potentially rewarding activities, especially those associated with delayed reward.

Internet use has increased dramatically in recent years, nearly reaching saturation in high income countries. Only 2% of Americans report no access to the Internet, and 84% report regular Internet usage (Perrin & Duggan, 2015), with similar rates true in other developed countries (Eurostat, 2018). Further, younger populations report even greater levels of Internet usage, with 96% of those between 18 and 29 years of age reporting regular use (Perrin & Duggan, 2015). This expansion of access provides striking benefits in social connection, education, and entertainment. In fact, these benefits are so well established that many view Internet access as a basic human right (La Rue, 2011), much like access to electricity and water. Despite these benefits, for some, Internet use becomes compulsive, excessively time consuming, costly, and disruptive to other important life domains (e.g., education, work, relationships) in a manner that is similar to other addictive behaviors such as gambling or substance use disorders (Kuss, Griffiths, Karila, & Billieux, 2014). However, research has yet to establish a scientifically-based and behaviorally-informed theoretical framework for understanding problematic Internet use. Behavioral economics, a perspective that combines operant psychology with principles of microeconomics, has been used to understand substance and behavioral addictions and suggests that these behaviors can be explained as a reinforcer pathology, or an overvaluation of an immediately acquirable substance/activity relative to prosocial and delayed rewards. Considering the public health impact of problematic Internet use, the overall growing interest in behavioral addictions, and the success of behavioral economic models in informing science and practice in substance addiction (Koffarnus & Bickel, 2014; MacKillop, 2016; Murphy, Dennhardt, et al., 2012), the application of behavioral economic models to problematic Internet use is a timely and potentially fruitful endeavor. Thus, the current study seeks to apply a behavioral economic framework to problematic Internet use, testing the hypothesis that the relationships between compulsive Internet use and behavioral economic indicators will be similar to other addictive behaviors. This homology would both provide support for the notion of problematic Internet use being similar to other forms of addiction and would also extend the applications of behavioral economics to addiction more broadly, which could help guide future research and intervention.

Problematic Internet Use

Although a lack of consensus on the definition of problematic Internet use makes estimates difficult, studies suggest between 2% and 26% of Internet users report some level of compulsive use (Bakken, Wenzel, Gotestam, Johansson, & Oren, 2009; Derbyshire et al., 2013; Xin et al., 2018). At this point, it is unclear whether problematic Internet use is better understood as a metaphor or a valid psychiatric diagnosis. What is clear, however, is that the potential for harm is certainly present, as problematic internet use has been linked to academic under-achievement, failure to exercise and to engage in face-to-face social activities, sleep difficulties, negative affective states, decreased ability to concentrate, and physical problems such as carpel tunnel syndrome, back and wrist pain, and headaches (Coniglio, Muni, Giammanco, & Pignato, 2007; Lin, Wu, You, Hu, & Yen, 2018). Although general problematic Internet use is not recognized as a psychiatric disorder, the most recent Diagnostic and Statistical Manual of Mental Disorders included Internet Gaming Disorder in an appendix as a preliminary disorder needing more research (American Psychiatric Association, 2013), and the World Health Organization recently added it as an official diagnosis, highlighting an increased interest in Internet-related addictions. It has elsewhere been argued that other Internet activities, such as online shopping, social media, and “cybersex,” can also be used addictively (Brand, Young, & Laier, 2014; Young, Pistner, O’Mara, & Buchanan, 2000). A recent network analysis found that despite the fact that smartphone addiction, cybersex, and online gaming constitute distinct problem domains with unique correlates, problematic internet use also captured variance as an “umbrella construct”, suggesting that common factors may contribute to a general difficulty in regulating Internet use. Thus, while internet addiction may not be a sufficient category as a psychiatric condition, an examination of problematic Internet use may be warranted to uncover theoretically relevant mechanisms that are consistent across technologically-mediated addictions.

Using a first principles approach, the Internet has several properties characteristic of substances with high abuse liability. Specifically, the Internet: 1) provides immediate reinforcement; 2) is generally inexpensive and highly accessible (Perrin & Duggan, 2015); 3) shares many of the same comorbidities as substance addictions (i.e., depression, anxiety, conduct disorder; Derbyshire et al., 2013; Hyun et al., 2015; Lin et al., 2018; Ni, Yan, Chen, & Liu, 2009); and 4) activates brain circuitry associated with addiction (Balconi, Venturella, & Finocchiaro, 2017; Kim et al., 2011; Lin et al., 2012, 2015; Yuan et al., 2011; Zhou et al., 2011). Problematic Internet use shares other similarities with substance-related addictions, as well. Self-proclaimed problematic Internet users participating in a qualitative study about characteristics of problematic Internet use reported that they used the Internet for a variety of reasons, including when they felt depressed or stressed, when they were bored, as a reward, and when they were feeling happy (Li, O’Brien, Snyder, & Howard, 2015). Further, studies examining risk factors have found that greater depression, impulsivity, and virtual social support are all related to an increased risk for developing an problematic Internet use (Lin et al., 2018). Problematic Internet use also has the potential to consume large amounts of resources and time and can interfere with reward receipt from other activities.

Theories of addiction that have gained traction in substance-related fields have rarely been employed to describe the experience of problematic Internet use. Modifying currently held theories of addiction to examine problematic Internet use is a natural extension and could organize research in the area. Of course, any theory attempting to explain problematic Internet use should account for the characteristics of addictive substances listed above and should explain the research to date on etiology, motives, and maintenance in a systematic and logical fashion. In other words, a guiding theory that can account for current research, while simultaneously integrating the four characteristics of addictive behaviors mentioned above, could organize current research into a consilient framework.

Behavioral Economics

Behavioral economics integrates operant psychology and microeconomic principles to understand behavior. From this perspective, behavior allocation is the result of the perceived cost/benefit ratios of the value of each available activity, and engagement follows for activities with greater benefits and lower costs. Addictive behaviors occur when an individual demonstrates a preference for immediate, rather than delayed, rewards. Known as delay discounting, this concept is often measured by asking participants to choose whether they would rather receive a smaller amount of money now versus a larger amount of money later. Through a series of systematic choices, the point at which the individual shifts from delayed to immediate reward choice can be ascertained, creating an index of the degree of discounting due to temporal delay. Indeed, studies demonstrate a robust relation between future orientation (i.e., the degree of preference for or consideration of immediate rather than delayed outcomes) and addictive behaviors (MacKillop et al., 2011) that strengthens with increasing levels of severity (Amlung, Vedelago, Acker, Balodis, & MacKillop, 2017). Similar relationships have also been observed for obesity (Amlung et al. 2017) and attention deficit hyperactivity disorder (Jackson & MacKillop 2017), suggesting that it is a common process in conditions comprising deficits of self-regulatory capacity.

Behavioral economics also suggests that addictive behaviors occur when the substance or behavior is consistently overvalued relative to alternatives. Valuation, or demand, is typically quantified by examining level of resource allocation, either actual time or money allocation to the commodity versus alternatives (Meshesha, Dennhardt, & Murphy, 2015; Tucker et al., 2016), or purchase tasks that ask how much of a substance each participant would consume at a series of escalating prices in certain hypothetical scenarios. Using this method, consumption at each price point can be plotted to create a demand curve, which creates indices that can be used to assess risk liability. These purchase tasks have been used to assess reinforcing value for a number of substances and addictive behaviors, including, but not limited to, alcohol (Murphy & MacKillop, 2006; Teeters & Murphy, 2015), marijuana (Aston, Metrik, Amlung, Kahler, & MacKillop, 2016), cocaine (Bruner & Johnson, 2014), indoor tanning (Reed, Kaplan, Becirevic, Roma, & Hursh, 2016), gambling (Weinstock, Mulhauser, Oremus, & D’Agostino, 2016), and cigarettes (MacKillop et al., 2008). Recently, purchase tasks have been validated for unfamiliar drugs in the context of characterizing the abuse liability (MacKillop, Goldenson, Kirkpatrick, & Leventhal, 2018). One study has examined an Internet purchase task and found connections between problematic Internet use and Internet demand (Broadbent & Dakki, 2015). Overvaluation of the behavior, in combination with a preference for immediate rewards, is the hallmark of behavioral economic models of addiction and has elsewhere been labeled reinforcer pathology (Bickel, Johnson, Koffarnus, MacKillop, & Murphy, 2014).

Often, reinforcer pathology occurs in the context of diminished reinforcement from alternative activities, resulting in a greater valuation of the addictive behavior relative to available alternative activities. Thus, those with a diminished availability of alternative reward in their environment (reward deprivation) will report greater levels of addictive behaviors (Joyner et al., 2016; Vuchinich & Tucker, 1988). Research with substance-related addictions supports this, showing that greater alcohol use is related to greater levels of substance-related reinforcement relative to substance-free reinforcement (Correia, Carey, Simons, & Borsari, 2003; Joyner, Acuff, Meshesha, Patrick, & Murphy, 2018). Further, increases in substance-free activities is followed by decreases in alcohol use (Correia, Benson, & Carey, 2005), and greater substance-free reinforcement predicts better alcohol intervention outcomes (Murphy, Correia, Colby, & Vuchinich, 2005).

We propose that behavioral economics may also provide a useful framework for understanding the choices that lead to problematic Internet use. From this perspective, the Internet provides immediate reinforcement in several life domains (e.g., social, leisure/hobbies, romantic/sexual activity) that previously required extended effort, time, and resources that might be associated with larger delayed rewards. Thus, the overall valuation of the Internet, or Internet demand, would increase due to a shift in the cost/benefit ratios of both the Internet and alternatives. Further, this would be most problematic for those with a preference for immediate, rather than delayed, rewards or in environments devoid of reinforcement from alternative activities. Indeed, this is supported by a handful of studies connecting impulsivity and delay discounting to problematic Internet use and smart phone use already (Meerkerk, Van Den Eijnden, Franken, & Garretsen, 2010; Saville, Gisbert, Kopp, & Telesco, 2010). Even some of the benefits of the Internet, such as Internet socializing, requires less effort than traditional socializing and may also come without the larger and delayed emotional and practical benefits of face-to-face socializing, reflecting reinforcer pathology. Behavioral economics may also serve to explain comorbidity between problematic Internet use and psychopathology. Several studies have found that behavioral economic mechanisms of alcohol demand and diminished alternative reinforcement mediate the relation between psychopathology (i.e., PTSD and depression) and problematic alcohol misuse (Acuff, Luciano, et al., 2018; Soltis, McDevitt-Murphy, & Murphy, 2017; Tripp et al., 2015). Although this has not been investigated with problematic Internet use, it is possible that the same mechanisms contribute to increasing problems with Internet use.

Current Study

The current study will apply the reinforcer pathology framework to problematic Internet use. We predict that future valuation variables, Internet demand, and reward deprivation will be significantly related to problematic Internet use and Internet craving. We also predict that internet demand will interact with delay discounting to predict problematic Internet use. We tested these hypotheses in a sample of adults on Amazon’s Mechanical Turk (Mturk).

Method

Participants

Participants were adults recruited through Mturk, an online crowdsourcing data collection site (www.mturk.com). Through Mturk, “requesters” can post “hits” (surveys) to be completed by “workers.” Eligibility for inclusion in the current study required that participants must: 1) be at least 18 years of age; 2) read and write English; and 3) have an acceptance rate of 85% or higher on previous hit attempts. Mturk participants who completed the survey where tagged with a “qualification” indicating that they already participated, preventing multiple participations. We included five attention check items throughout the survey to ensure valid data. Participants were required to answer at least 80% of attention check items correctly to be included in the final sample. Thus, although the initial sample included 313 individuals, the final sample only retained 256 (82%). There were no differences in gender, age, race, income, or education level between those omitted and those included in the study; there were, however, differences in problematic internet use scores between excluded (M = 48.54, SD = 16.76) and included (M = 40.10, SD = 16.84) participants, t (310) = 3.26, p = .001.

Procedures

All study procedures were approved by the University of Memphis Institutional Review Board (FWA00006815). The survey, posted on Mturk between October 2017 and January of 2018, was entitled “Study of Young Adult Life Experiences” and took between 30–45 minutes to complete. All participants were paid 3 dollars within a week of survey completion. Due to item endorsement, some participants filled out more questions and for that reason were given an extra dollar. All items that affected the length of the survey occurred following the problematic Internet use and behavioral economic items. This payment schedule is in line with ongoing rates among Mturk requesters (Horton & Chilton, 2010). We defined Internet use as “playing online games, browsing social media/other leisure sites, texting, or chatting through online forums or social groups.” Although our definition asked participants not to consider time spent doing homework or occupational work, they were asked to count time spent online for leisure/socializing while they were also doing something else (e.g., working).

Measures

Problematic Internet Use.

The Problematic Internet Use Questionnaire asks participants to rate how often (1 – Never to 5 – Always) they experience each of 18 problems related to excessive Internet use (Demetrovics, Szeredi, & Rózsa, 2008). Questions include items like “How often do you daydream about the Internet?,” “How often do you spend time online when you’d rather sleep” and “How often does it happen to you that you wish to decrease the amount of time spent online but you do not succeed?”. No clinical threshold exists for determining problematic use. Internal consistency for the PIUQ was excellent (α = .96).

We also used the Internet Craving Scale (Spada, Caselli, Slaifer, Nikčević, & Sassaroli, 2014), which contains five questions from the Penn Alcohol Craving Scale (Flannery, Volpicelli, & Pettinati, 1999) that have been modified to reflect craving for the Internet. For the current study, we added a sixth question: “How soon after waking up do you first access the Internet (e.g., playing online games, browsing social media/other leisure sites, or chatting through online forums or social groups)?” Items are summed to create a total score. Internal consistency for the Internet Craving Scale was good (α = .80).

Delay Discounting and Consideration of Future Consequences.

Delay discounting was assessed with the 8-item Delayed Reward Discounting Task (Gray, Amlung, Acker, Sweet, & MacKillop, 2014). For each item, participants choose between a smaller, sooner monetary value and a larger, delayed reward. Delay discounting was operationalized as the ratio of impulsive choices to delayed choices, with larger delay discounting values representing greater discounting of delayed rewards. We also used the Consideration of Future Consequences Scale (CFCS; Strathman, Gleicher, Boninger, & Edwards, 1994) to measure future valuation. In this measure, participants rate the degree to which each of 12 statements are characteristic of them on a 5-point scale (1 – Extremely Uncharacteristic to 5 – Extremely Characteristic). Items in the CFCS are summed to create a total score, with greater value representative of higher future valuation. Internal consistency for the CFCS was good (α = .86). Both CFC and delay discounting have been linked to misuse of a range of substances (Acuff, Soltis, Dennhardt, Berlin, & Murphy, 2018; Kirby & Petry, 2004; Murphy et al., 2015).

Internet Demand.

We used a modified version of an existing Internet Purchase Task (IPT) to measure Internet demand (Broadbent & Dakki, 2015). Participants first read to the following text:

Imagine you are on a 5-hour flight. The airline offers access to the Internet that you can purchase in 30-minute increments. For each of the prices below, please indicate how many 30-minute increments of Internet access you would buy at each price for social/recreational purposes (i.e., to check your e-mail, social networking sites, movies, music etc. rather than using the Internet for school, work, etc.).

Participants then reported how many 30-minute increments of Internet access they would purchase at a series of escalating prices ($0, $0.50, $1.00, $1.50, $2.00, $2.50, $3.00, $4.00, $5.00, $7.50, $10.00, $15.00, $25.00). Consumption (number of 30-minute increments) at each price was then plotted to create a demand curve, from which quantifiable demand indices are extracted. For this study, we examined five demand indices. Observed indices included intensity (consumption when the price is zero), breakpoint (price when consumption reaches zero), and O_max (point of maximum expenditure). We derived indices of elasticity and Q₀ with the modified, exponentiated version (Koffarnus, Franck, Stein, & Bickel, 2015) of the exponential demand equation (Hursh & Silberberg, 2008):

$$Q=Q_0*{10}^{k\left(e^{-\alpha Q_{0}C}-1\right)}$$

Where, Q = quantity consumed, k = the range of the dependent variable (i.e., 30-minute increments of internet access) in logarithmic units, P = price, C = predicted consumption at a unit price of P, and α = elasticity of demand. In the current study, k was held constant and was calculated by subtracting the log10-transformed average consumption at the highest price from the log10-transformed average consumption at the lowest price. In our sample, k was held constant at 1.60. The Koffarnus equation allows for the inclusions of zero; thus, zeros were not removed before elasticity was calculated. Data were cleaned using the Stein macro (Stein, Koffarnus, Snider, Quisenberry, & Bickel, 2015) before indices were extracted. The Stein macro detects nonsystematic data based on three criteria: 1) Trend (detection limit for ΔQ = .025); 2) Bounce (detection limit for B = .10); 3) Reversal from zero (detection limit number for reversals = 0). These criteria removed 25 participants (9.67%) from data analysis for nonsystematic responding. Test-retest reliability, concurrent validity, and discriminant validity for purchase tasks are robust for multiple commodities (Acuff & Murphy, 2017; Chase, MacKillop, & Hogarth, 2013; Few, Acker, Murphy, & MacKillop, 2012; Murphy et al., 2013; Murphy, MacKillop, Skidmore, & Pederson, 2009), although this has not been established with the Internet purchase task.

Reward Deprivation.

We used the Reward Probability Index (RPI; Carvalho et al., 2011) to assess reward deprivation. Participants report how much they agree with 20 statements related to reward functioning on a 4-point rating scale. The RPI is further divided into two factors. The first factor, experience of reward, is made up of 11 items and measures one’s ability to experience reward (e.g., “I have the abilities to obtain pleasure in life”). The second factor, access to reward, uses the remaining nine items and measures the degree of access to reward in the environment. The items for each factor are summed to create total scores. RPI factors demonstrate convergent validity with relevant reward variables (Carvalho et al., 2011). In the current sample, internal consistency was good for both the reward probability (α = .87) and the access to reward factors (α = .85).

Data Analysis

Outliers were identified using a criterion of Z = 3.29, and were changed to one unit above the highest nonoutlying value (Tabachnick & Fidell, 2013). Next, we checked variables for nonnormality according to previously established limits (−2 and 2; Trochim & Donnelly, 2006). Q₀ and elasticity were nonnormal and were corrected using the Base 10 log method. First, we examined bivariate relations among study variables. Next, we used hierarchical regressions to examine the incremental utility of each behavioral economic variable in predicting Internet use, PIUQ score, and Internet craving score after controlling for demographic covariates (i.e., gender, age, and annual household income). Finally, we examined a final model that includes all predictors that were significant in the hierarchical regressions. To test the reinforcer pathology model, we examined interactions between demand indices and delay discounting in predicting problematic Internet use. Independent variables were mean centered before use in the interaction terms. In tests of the interaction, we included demographic covariates in the first step; delay discounting was included in the second step; demand indices (separately) were included in the third step; and the interaction term was included in the final step. We analyzed all data using SPSS v25 (IBM Corp. in Armonk, NY).

Results

Descriptive statistics and correlations among study variables

Demographic information and means of study variables are presented in Table 1. Participants spent an average of 36.22 (SD = 24.22) hours per week on the Internet engaged in nonwork-related activities. Participants reported average scores of 40.10 (SD = 16.84) on the PIUQ and 20.11 (SD = 6.52) on the Internet Craving Scale (Scale: 0 to 36). The exponentiated demand curve equation provided a good fit for individual data (R² = .88) and an excellent fit for aggregate data (R² = .99). Correlations among study variables are reported in Table 2. Delay discounting, CFC, Demand Q₀ and breakpoint, and experience of and access to reward were all correlated with Internet craving and PIUQ scores in the expected directions.

Table 1.

Demographic Characteristics of Adults (N = 256) Recruited from Amazon’s Mechanical Turk Database for a Survey of Behavioral Economic Predictors of Problematic Internet Use.

Variable	N	Percentage or M (SD)
Gender (% female)	93	36.3%
Age		27.87 (4.79)
Race	255
White	149	58.2%
American Indian	5	2%
Asian	59	23%
Black	31	12.1%
Other	12	4.7%
Current Student (% current)	63	24.3%
Education Level	255
Less than high school	14	2.5%
Some college	75	29.4%
Associate’s degree	31	12.2%
Bachelor’s degree	119	46.7%
Master’s degree	15	5.9%
Professional degree	1	0.4%
Current Household Income	256
$0 - $15,000	32	12.5%
$15,001 - $25,000	42	16.4%
$25,001 - $35,000	41	16%
$35,001 - $50,000	57	22.3%
$50,001 - $75,000	51	19.9%
$75,001 - $100,000	21	8.2%
More than $100,000	12	4.7%
Delay discounting	255	.53 (.27)
CFC	256	41.55 (8.26)
Experience of reward	243	32.30 (5.92)
Access to reward	250	24.07 (5.60)
Intensity-derived	232	12.30 (12.88)
Omax	234	12.09 (12.36)
Elasticity	211	.030 (.037)
Breakpoint	255	9.50 (9.53)
Problematic Internet Use	234	40.10 (16.84)
Internet Craving	254	20.11 (6.52)

Note. CFC = Consideration of Future Consequences; Delay discounting = ratio of choice for immediate v. delayed rewards on the 8-item delay discounting measure; Experience of reward = reward probability factor of the Reward Probability Index (RPI); Access to reward – environmental suppressors subscale of the RPI; Intensity-derived = internet consumption when cost was free on the IPT; O_max = maximum expenditure on the IPT; Elasticity = rate of change as a function of price on the IPT; Breakpoint = price at which internet consumption reaches zero; SD = Standard deviation.

Table 2.

Correlations among Study Variables from a Survey of Behavioral Economic Predictors of Problematic Internet Use.

	Domain	1	2	3	4	5	6	7	8	9
1. Internet Craving	Internet Involvement	-
2. Problematic Internet Use	Internet Involvement	.55***	-
3. Q0	Valuation	.13*	.16*	-
4. Elasticity	Valuation	−.13	−.04	−.38***	-
5. Breakpoint	Valuation	.21**	.15*	.23***	−.49***	-
6. Omax	Valuation	.13	.06	.30***	−.69***	.76***	-
7. Delay Discounting	Reward Immediacy	.16**	.15*	.06	.02	.10	.09	-
8. CFC	Reward Immediacy	−.46***	−.31***	−.12	−.01	−.10	−.07	−.23***	-
9. Experience of Reward	Alternative Reinforcers	−.23***	−.25***	−.05	−.04	.07	.07	−.02	.41***	-
10. Access to Reward	Alternative Reinforcers	−.49***	−.34***	−.11	.06	−.12	−.09	−.12	.43***	.47***

Note. CFC = Consideration of Future Consequences; Delay discounting = ratio of choice for immediate v. delayed rewards on the 8-item delay discounting measure; Experience of reward = reward probability factor of the Reward Probability Index (RPI); Access to reward – environmental suppressors subscale of the RPI; Intensity-derived = internet consumption when cost was free on the IPT; O_max = maximum expenditure on the IPT; Elasticity = rate of change as a function of price on the IPT; Breakpoint = price at which internet consumption reaches zero; SD = Standard deviation.

^***p< .001

^**p< .01

^*p< .05.

Hierarchical regressions

Hierarchical regressions controlling for gender, age, and income indicated that greater demand breakpoint and demand intensity and lower CFC, experience of reward, and access to reward were associated with both greater Internet craving and problematic Internet use in the expected directions (Table 3). Demand elasticity and O_max were also associated with problematic Internet use in the expected direction. A final model that included all significant predictors suggested that consideration of future consequences and access to reward were unique predictors of problematic Internet use, and that consideration of future consequences was a unique predictor of Internet craving. Our models accounted for 29% and 21% of the variance in problematic Internet use and Internet craving, respectively.

Table 3.

Hierarchical Regressions of Behavioral Economic Variables Predicting Problematic Internet Use and Internet Craving in a Sample of Adults from Amazon’s Mechanical Turk (N = 256)

Problematic Internet Use							Internet Craving
Variable	Beta (S.E.)	C.I. Lower, Upper	β	t	p	ΔR2	Beta (S.E.)	C.I. Lower, Upper	β	t	p	ΔR2
Step 1
Gender	.02 (2.21)	−4.33, 4.37	.00	.01	.99	.08	1.10 (.85)	−.58, 2.77	.08	1.29	.20	.08
Age	.15 (.22)	−.28, .59	.04	.69	.49		.04 (.09)	−.13, .22	.03	.48	.63
Income	−2.83 (.61)	−4.03, −1.63	−.28	−4.64	<.001		−1.00 (.24)	−1.46, −.53	−.26	−4.22	<.001
Step 2 (separate)
DD	6.87 (3.93)	−.87, 14.61	.11	1.75	.08	.01	2.41 (1.53)	−.61, 5.43	.10	1.57	.12	.01
CFC	−.85 (.12)	−1.08, −.63	−.42	−7.38	<.001	.17	−.21 (.05)	−.30, −.11	−.26	−4.35	<.001	.07
Experience of Reward	−.53 (.17)	−.87, −.19	−.19	−3.05	.003	.04	−.22 (.07)	−.35, −.09	−.21	−3.36	.001	.04
Access to Reward	−1.34 (.18)	−1.68, −.99	−.45	−7.62	<.001	.18	−.33 (.07)	−.47, −.18	−.28	−4.41	<.001	.07
Q0	8.38 (3.81)	.88, 15.89	.14	2.20	.03	.02	4.04 (1.57)	.95, 7.14	.16	2.56	.01	.03
Omax	1.24 (.58)	.10, 2.37	.14	2.14	.03	.02	.28 (.24)	−.20, .76	.08	1.17	.25	.01
Elasticity	−155.96 (70.94)	−295.84, −16.09	−.15	−2.20	.03	.02	−32.10 (29.79)	−90.84, 26.64	−.07	−1.08	.28	.01
Breakpoint	.35 (.11)	.14, .56	.21	3.33	.001	.04	.10 (.04)	.01, .19	.15	2.28	.02	.02
Omnibus Model Controlling for Significant Predictors
CFC	−.50 (.13)	−.76, −.24	−.27	−3.74	<.001	.21	−.13 (.06)	−.24, −.02	−.17	−2.35	.02	.13
Experience of Reward	−.04 (.19)	−.41, .34	−.02	−.20	.84		−.07 (.08)	−.24, .09	−.07	−.85	.40
Access to Reward	−.74 (.21)	−1.16, −.32	−.27	−3.47	.001		−.18 (.09)	−.36, .007	−.15	−1.90	.06
Q0	1.54 (4.07)	−6.49, 9.58	.03	.38	.70		1.67 (1.64)	−1.55, 4.90	.07	1.02	.31
Omax	−.27 (1.12)	−2.49, 1.96	−.03	−.24	.81		-	-	-	-	-
Elasticity	−30.31 (92.43)	−212.70, 152.07	−.03	−.33	.74		-	-	-	-	-
Breakpoint	.23 (.15)	−.07, .53	.14	1.52	.13		.09 (.05)	−.006, .18	.12	1.85	.07

Note. CFC = Consideration of Future Consequences; O_max = Maximum Expenditure; S.E. = Standard Error; C.I. = Confidence Interval; DD = Delay discounting; Q₀ = Intensity-derived.

The Reinforcer Pathology Model: Tests of interaction effects between delay discounting and demand

In the linear regression analysis that evaluated the interactions between delay discounting and demand breakpoint in predicting problematic Internet use, the overall model including the delay discounting by demand breakpoint interaction term was significant F(6, 225) = 6.51, p < .001, R² = .13, and the interaction term significantly predicted problematic Internet use (β =.13, p = .04; ΔR² = .02). We conducted follow up regressions of demand breakpoint predicting PIUQ scores for those with low and high delay discounting (median split). For those with low delay discounting, demand breakpoint was not a significant predictor (β =.03, p = .76, ΔR² = .00). However, for those with high delay discounting, demand breakpoint was a significant predictor (β =.35, p < .001, ΔR² = .12; Figure 1a).

Figure 1.

Demand curves for average a) consumption and b) expenditure, separated into groups of those who scored in the bottom and top thirds on the Problematic Internet Use Questionnaire (PIUQ). X- and y-axes are log-transformed. a) Those who score in the bottom third on the PIUQ report less internet consumption at lower prices. As price (x-axis) increases, consumption decreases at a slower rate for those who score in the top third of the PIUQ compared to those who score in the bottom third. b) Those who score in the top third on the PIUQ are willing to spend more money on internet than those in the bottom third. As price increases, those who score in the bottom third on the PIUQ reach a maximum expenditure (O_max), while those who score in the top third demonstrate a willingness to continue to spend money. The escalating expenditure slope for those who score in the top third of the PIUQ demonstrates a ceiling effect that is present with the current iteration of the internet purchase task, which only reaches $25.

The overall model including the delay discounting by demand O_max interaction term was significant F(6, 225) = 5.18, p < .001, ΔR² = .10, and there was a non-significant trend-level effect for the interaction in predicting problematic Internet use (β =.11, p = .077; ΔR² = .01). We also conducted follow up regressions of demand O_max predicting PIUQ scores split by delay discounting scores. For those with low delay discounting, demand O_max was not a significant predictor, (β = −.06, p = .54; ΔR² = .00). However, for those with high delay discounting, demand O_max was a significant predictor, (β =.28, p = .002; ΔR² = .08; Figure 1b). All other interactions were nonsignificant.

Discussion

Considering ubiquitous Internet availability, excessive Internet use has the potential to be a major societal concern. Much of the work to date, however, has been atheoretical, and thus the current study examined problematic Internet use as a form of reinforcer pathology. Consistent with our hypotheses, low future reward valuation and access to reward, and high Internet demand were significantly associated with Internet craving and problematic Internet use in a sample of adult Internet users. Participants’ reported Internet consumption on a hypothetical Internet purchase task fit well with a quantitative demand curve model that has been used as a measure of reinforcing efficacy for other addictive behaviors and substances (Murphy & MacKillop, 2006; Reed et al., 2016; Weinstock et al., 2016). The results suggest that, although Internet use is sensitive to price generally, it may be less sensitive for individuals experiencing problematic Internet use (Figure 1). One previous study has linked Internet demand to problematic use (Broadbent & Dakki, 2015), and many previous studies demonstrate a relation between various elevated demand and addictive behaviors (Aston, Metrik, & MacKillop, 2015; Murphy & MacKillop, 2006). Our results are consistent with these studies, suggesting that hypothetical purchase tasks are useful measures capturing the reinforcing efficacy of the Internet. Our understanding of problematic Internet use is still nascent, and consequently there are few theoretically grounded measures that captures severity of use. Demand could provide a severity index that is easy to administer, sensitive to laboratory manipulations and treatment, and theoretically grounded in animal and human studies of drug use liability (Bickel, Marsch, & Carroll, 2000; Bickel, Snider, Quisenberry, & Stein, 2017).

Our study suggests that delay discounting and CFC, two conceptually and empirically distinct measures of future valuation, are associated with problematic Internet use and Internet craving. This is consistent with previous research connecting delay discounting to excessive smart phone use (Wilmer & Chein, 2016), in addition to a study that found differences in delay discounting between those with self-reported problematic Internet use and those without (Saville et al., 2010). Consistent with addictive behaviors (Bickel & Marsch, 2001; MacKillop et al., 2011), the Internet provides immediate reinforcement for activities from domains that require greater effort/time expenditure in the real-world (e.g., dating, games, gambling, shopping). Thus, those with a tendency to devalue future rewards can access these rewards promptly via the Internet. This could lead to increased levels of Internet use and, potentially, disengagement in activities with more temporally distal rewards. Interestingly, delay discounting was no longer significantly associated with problematic Internet use and Internet craving after accounting for annual household income. Thus, the effect of delay discounting on problematic internet use is attributable to covariance with income. In our sample, lower income was indeed associated with greater delayed discounting values (r = −.24), which may, in turn, result in engagement in addictive behaviors that favor immediate reward at the expense of greater future reward.

Behavioral economics also suggests that reinforcer pathology is most likely to occur in an environment with diminished access to rewarding activities. Both animal and human studies support this in substance-related addictions (Alexander, Coambs, & Hadaway, 1978; Correia, Simons, Carey, & Borsari, 1998; Herrnstein, 1979; Murphy et al., 2005), and interventions that increase substance-free reward are associated with reduced substance use (Kristjansson et al., 2016; Kristjansson, James, Allegrante, Sigfusdottir, & Helgason, 2010; Murphy, Skidmore, et al., 2012). Our study is among the first to examine reward deprivation as a risk factors for problematic Internet use, and, suggests that experience of reward and access to reward were protective against problematic Internet use. Individuals with less access to non-digital social and recreational opportunities may be at greater risk of the development of problematic Internet use. Although conjecture, Internet-related social engagement may not provide the same ensuring emotional and instrumental benefits as in-person interactions, operating as an “inferior good” in economic terms. Further, diminished reward experience (i.e., anhedonia) is a known risk factor for and outcome of prolonged substance use (Acuff, Soltis, et al., 2018; Leventhal et al., 2011; Meshesha, Pickover, Teeters, & Murphy, 2017), and interventions have been designed specifically to target engagement in rewarding experiences (Magidson et al., 2011; Murphy, Dennhardt, et al., 2012). The Internet may operate in a similar way as these substances, providing reward that is easily obtained, but that might undermine engagement in non-digital social/recreational activities associated with delayed but perhaps ultimately larger rewards (e.g., satisfying interpersonal relationship or hobbies). It is important to note, however, that these associations may not be causal, and that internet use can often facilitate non-electronic rewarding experiences (e.g., meeting someone online can lead to face-to-face contact). Future longitudinal research is needed to determine if there are bidirectional and causal relations between reward deprivation and internet use.

Previous theoretical work has highlighted delay discounting and demand as potentially important and interacting phenotypes defining reinforcer pathology, such that the presence of both (high discounting and high demand) would result in problematic engagement (Bickel, Johnson, et al., 2014; Bickel, Moody, Quisenberry, Ramey, & Sheffer, 2014); however, previous empirical studies have not demonstrated this interaction in predicting alcohol (Acuff, Soltis, et al., 2018; Lemley, Kaplan, Reed, Darden, & Jarmolowicz, 2016) or marijuana use (Aston et al., 2016) among humans. We found that demand breakpoint and O_max (trending) interacted with delay discounting to predict problematic internet use. Those with high demand for the Internet, but low delay discounting, may have greater control over compulsive use, which results in lower problems. Likewise, those with high delay discounting, but low internet demand, are not likely to engage with the Internet problematically. The interaction found in our study suggests that those with high delay discounting may experience compulsive Internet use only when their demand for Internet is also high. In this scenario, the Internet (an immediate reinforcer) is highly valued, and immediate rewards are preferred.

Taken together, our study suggests that behavioral economics provides a theoretical model that can account for a significant portion of the experience of problematic Internet use. The Internet provides immediate reinforcement (e.g., pornography, social networking sites, video games, online shopping) in domains that may require more effort and time to obtain in real-life parallels. Further, individuals with problematic Internet use may have limited access to rewarding alternatives to the Internet. For some people excessive Internet use may conflict with attainment of delayed, prosocial alternatives (e.g., meaningful relationships, work, physical fitness) thereby creating a vicious cycle whereby the relative value of Internet activity increases. Our study also supports the examination of behavioral addictions in general. In conjunction with other behavioral economic analyses of behavioral addictions (Becirevic et al., 2017; Reed et al., 2016; Weinstock et al., 2016), our study suggests that behavioral addictions operate in a similar fashion as substance-related addictions. A major distinction between the behavioral and substance addictions is related to the presence of neurological effects. Research does support the presence of neurological changes and deficits among those with behavioral addictions (Balconi et al., 2017; Beutel, Wo, Duven, & Mu, 2014), albeit at a lower intensity than substance-related addictions. This, however, suggests a “quantitative rather than qualitative difference” between these addictions (Wise, 2002), and highlights the possibility of a spectrum of addictive behaviors that could partially be dictated by strength of neurological effect while still acknowledging the behaviorally similar aspects of the phenomena.

Strengths, Limitations, and Future Directions

Our study used a moderately sized sample of Internet users collected through Mturk and it is the first study to apply a comprehensive behavioral economic measurement approach to describing problematic Internet use. Our results provide support for behavioral economic models of problematic Internet use and set the stage for future research in this domain, given that behavioral economic research has guided a variety of laboratory, clinical, and policy-relevant research with other addictive behavior (Hursh & Roma, 2013; Kristjansson et al., 2016; MacKillop, 2016; Murphy, Dennhardt, et al., 2012; Reed et al., 2016). However, our study had several limitations. First, given the use of self-report measures, the potential for bias or measurement error is certainly possible. However, these measures (and similar ones) have been used in numerous studies on substance-related addictions, and validation studies using actual substances have been highly supportive (Amlung, Acker, Stojek, Murphy, & MacKillop, 2012; Amlung & MacKillop, 2015; Madden, Begotka, Raiff, & Kastern, 2003). A further form of measurement error is evident in Figure 1 in the possibility of ceiling effect for the Internet purchase task, as problematic Internet users reported persistent nonzero consumption across prices. Future studies should include a larger range of prices. The iteration of the IPT used in the current study also presents a scenario that may be unfamiliar to most people (i.e., purchasing internet on a flight) and may not represent typical internet habits. Future research should explore other possible IPT scenarios including purchasing internet at home or in a cafe. Second, we used a cross-sectional design with a sample collected through an online data collection website from individuals who are seeking employment online, which may not reflect the typical Internet user’s experience or generalize to the national population. The population was also non-representative of the US population in other ways as it included a higher number of males, Caucasians, Asian Americans, and individuals with college degrees. Our sample, however, did include participants with a wide-range of Internet use, suggesting that Mturk may provide a useful, albeit non-representative sample to examine a range of problematic Internet use. There were also significant differences between excluded and included individuals based on problematic internet use, which may have influenced our findings. Future studies should use alternative methodological approaches to increase the confidence of our findings.

Finally, another limitation is that our assessments focused on problematic Internet use in general, not specific types of use. One enduring question for the field is whether problematic Internet use is the behavior of concern or whether the concern only pertains to certain specific types of use (e.g., online gaming, cybersex, and smartphone addiction; Baggio et al., 2018). From this perspective, the Internet might serve as an accelerant for other forms of addiction, rather than the object of addiction per se, and the current study cannot address specific types of Internet engagement. Baggio et al. found that problematic Internet use may be useful as an “umbrella construct” for technologically-mediated addictions, but that these phenomena are also distinct, such that a) they are associated with unique outcomes (e.g., gambling can result in large financial loses whereas internet pornography may have significant relationship consequences) and b) individuals often have difficulty regulating one of these domains but not the others. Our study suggests that the umbrella construct of problematic internet use can be partially explained through behavioral economics and sets the precedent for further examination within these individual domains (e.g., cybersex, online gaming, etc.). In some cases, however, the internet may serve as a healthier substitute for other addictive behavior (e.g., for an individual who socializes or plays games online instead of gambling or using drugs).

Conclusion

Our study supports the application of behavioral economics, a theoretical perspective that guided research on a variety of health risk and addictive behaviors (Bickel, Johnson, et al., 2014), to the study of problematic Internet use. Individuals who report elevated delay discounting, low consideration of the future, elevated internet demand and general deficits in access to and experience of reward are most likely to engage in problematic Internet use. These results support future research that applies behavioral economic models to understand the etiology and developmental course of problematic internet use, and to guide prevention and treatment approaches.

Figure 2.

Interactions reflecting the reinforcer pathology model. All independent variables were mean centered before use in the interaction term. a) The interaction between demand breakpoint and delay discounting was significant (p = .04). Those with both high delay discounting and high breakpoint report higher levels of problematic internet use compared to other groups. b) The interaction between demand O_max and delay discounting demonstrated a nonsignificant trendlevel relationship with problematic internet use (p = .077). Those with both high delay discounting and high O_max report higher levels of problematic internet use compared to other groups.