The development of the Problematic Series WatchingScale (PSWS).

Background and aims The goal of the present study was to create a short ProblematicSeries Watching Scale (PSWS). Methods On the basis of the six components model of Griffiths ( 2005 ), six items were identifiedcovering all components of problematic series watching. Confirmatoryfactor analyses were carried out on two independent samples (N1  = 366, N2  = 752). Results The PSWS has appropriate factor structure and reliability. Theamount of free time was not, but the series watching time was associatedwith PSWS scores. Women had higher scores than men. Discussion Before PSWS, no prior scale has been created to measure problematicseries watching. Further research is needed to properly assess itsvalidity and reliability; and for examining whether extensive serieswatching can lead to health-related and psychosocial problems. Conclusions In the increasingly digitalized world there are many motivationalforces which encourage people watching online series. In the lightof these changes, research on problematic series watching will beprogressively relevant.

Introduction

Since the end of the 1990s problematic Internet use has been acontroversial topic: some researchers and clinicians (e.g. Ko, Yen, Chen, Chen, & Yen, 2005 or Young, 1998) have claimedthat problematic Internet use deserves a classification as a psychiatricdisorder in its own right, while others (e.g. Griffiths, 1999, 2000 or Yellowlees & Marks,2007) have claimed that people who spend excessive time onthe Internet are not really addicted to it, but rather to the specificactivities they can pursue through this medium. Recently, severalscreen-related online activities were identified as problematic onlinebehaviors such as Facebook use (Andreassen,Torsheim, Brunborg, & Pallesen, 2012; Ryan, Chester, Reece, &Xenos, 2014), online gambling (Chóliz, 2015; Griffiths &Barnes, 2008), online gaming (Demetrovics et al., 2012; Grüsser, Thaleman, & Griffiths, 2006), or onlinepornography use (Grubbs, Sessoms,Wheeler, & Volk, 2010; Kor et al., 2014).

According to the research of Pontes, Szabó, and Griffiths(2015), when participants hadto indicate their three most preferred online activities, watchingvideos and movies was mentioned by one-third of the respondents besidesaccessing general information and news, social networking, e-mailingand online chatting, and gaming or gambling. Although more than one-thirdof the examined population mentioned watching online content as theirpreferred non-work online activity, to our best knowledge no priorscientific study examined problematic series watching.

It might be a relevant issue for many peoplebecause accessing series by downloading or streaming is (a) very cheap(or free), (b) it is available for almost everyone who has broadbandInternet access, (c) it does not depend on a certain place and time (i.e. playing squash depends on a certainplace and time), (d) series have a high variety – everyone canfind one which fits his/her interest, (e) they are not age- and socio-economicstatus-dependent, (f) it does not take effort to watch them, (g) andthey are constructed to be highly enjoyable and often contain cliffhangerswhich motivate the viewer to continue. These characteristics are highlysimilar to the ones mentioned by Cooper (1998) regarding Internet and pornography. Based on these reasonsand Pontes et al.’s (2015) findings, we assume that problematic series watching deserves scientificinvestigation.

Sussman and Moran’s (2013) review on television addiction did not differentiate between thetypes of content that can be seen on television, they define televisionaddiction as a subjective craving for watching anything on televisionincluding both classical and online content. In our research, we aimedto differentiate problematic series watching from the concept of televisionaddiction as we focused on the content of the problematic use (serieswatching) rather than on the medium through which the problematicuse happens (television). In our research, we observed problematicseries watching which could be done either through a television (i.e. classical TV series) or a screen attached toa computer (i.e. Netflix). The device itself is just a medium throughwhich the user can reach the content with the latter being in ourmain focus.

According to a recent study from a leadingInternet television company with over 40 million members, 76% of series viewers mentioned that watching several episodesof a TV show is a favored escape from their busy lives and 73% oftheir respondents have positive feelings towards binge streaming TV(Netflix Media Center, 2013). In the literature of problematic behaviors, escapism and moodmodification are good candidates in predicting a problematic activity.We assumed that if three quarters of the series streamers watch severalepisodes in succession for escaping from the problems of the everydaylife, then it is possible to assume that problematic series watchingcan appear among them.

On the basis of Griffiths’ components model (2005), we can distinguish six coreelements of problematic series watching: (a) salience (series watchingdominates thinking and behavior); (b) mood modification (series watchingmodifies/improves mood); (c) tolerance (increasing amounts of serieswatching are required to achieve initial effects); (d) withdrawal(occurrence of unpleasant feelings when series watching is discontinued);(e) conflict (series watching compromises social relationships andother activities); and (f) relapse (tendency for reversion to earlierpatterns of series watching after abstinence or control).

In the present study our goal was to measure problematic serieswatching. Therefore, we created a scale with appropriate factor structureand reliability which is based on the six-component model of Griffiths(2005).

Methods

Participants and procedure

The research was conducted with an online questionnaire system,the filling out lasted approximately six minutes. The data collectionoccurred in two waves: the first one in April 2014 (Sample 1), thesecond one in April–May 2015 (Sample 2). In both cases, participantswere informed about the goal and the content of the study, also, theanonymity and the confidentiality of their answers were ensured. Theywere asked to check a box if they agreed to continue and participate.The first part of the questionnaire contained questions regardingdemographic data, such as gender, age and level of education. Participantswere also asked to estimate the amount of free time they have on anaverage weekday and weekend. In the subsequent part, the items ofthe problematic series watching were presented.

Sample 1 consisted of 366 Hungarian respondents (Female = 258; 70.5%) who wereaged between 18 and 82 (Mage = 22.83, SDage = 6.23). 250 of them(68.3%) lived in the capital, 35 (9.6%) in county towns, 58 (15.8%)in towns, and 23 (6.3%) in villages. Concerning their level of education,14 (3.8%) had a primary school degree, 270 (73.8%) had a high schooldegree, 82 (22.4%) of them had a degree in highereducation (bachelor, masters, or doctoral). On an average weekdaythey had 4.12 hours of free time (SD =3.11 hours) and watched series for 62.39 minutes (SD = 60.77). On an average weekend they had 8.38 hoursof free time (SD = 5.24) andwatched series for 88.81 minutes (SD = 88.13).No outliers were detected using theMahalanobisdistance test, thus all 366 cases were retained in this sample.

Sample 2 consisted of 754 Hungarian persons (Female = 550; 72.9%). They wereaged between 18 and 67 years (Mage = 27.25, SDage = 8.70). 315 (41.8%) lived in the capital,151 (20.0%) in county towns, 199 (26.4%) in towns and 89 (11.8%) ofthem lived in villages. 48 (6.4%) people from this sample had an elementarydegree, 420 (55.7%) had a high school degree, 286 (37.9%) had a degreein higher education. Regarding their free time, on an average weekdaythey had 4.82 hours of free time (SD = 3).On an average weekend they had 9.10 hours of free time (SD =4.44). In this case we did not measure the time theyspent on series watching. After examining the data with the Mahalanobisdistance test, two outliers were found in this sample. These caseswere removed from further analyses, resulting in a final number of752 valid cases.

Measures

To our best knowledge, currently there is no measurement that canassess problematic series watching. For this purpose, on the basis of the Bergen Work AddictionScale (BWAS – Andreassen, Griffiths,Hetland, & Pallesen, 2012), we created a new scale –Problematic Series Watching Scale (PSWS) –to measure the six core elements of problematic series watching interms of (a) salience, (b) tolerance,(c) mood modification, (d) relapse, (e) withdrawal, and (f) conflict. We chose this measure as a basisbecause it grasps each possible element of problematic behaviors andthe basis of this questionnaire is widely used to measure other onlineand offline problematic behaviors. Besides Work Addiction, Andreassen,Torsheim et al. (2012)used this set of items to measure Facebook addiction and also shoppingaddiction (Andreassen et al.,2015). We chose to use the label of “Problematic SeriesWatching” instead of “Series Watching Addiction”because we have no solid evidence regardinghealth-related negative consequencesof this behavior and this might be a too specific and peripheral behaviorto call it addiction (Billieux, Schimmenti,Khazaal, Maurage, & Heeren, 2015).

The BWAS contains seven items; however, we wantedto have a larger initial item pool to ensure a better content validityregarding the framework of problematic behavior. Moreover, the originalscale contains an additional dimension (health problems), but ouraim was to create a measure using the above-mentioned six dimensions.Therefore, those 12 items have been chosen that were the initial itemsof Andreassen, Griffiths et al. (2012) and belonged to one of the six core components. First,all 12 items were translated following the protocol of Beaton, Bombardier,Guillemin, and Ferraz (2000).Then, the items were modified to reflect the individual’s serieswatching habits by replacing the subject “work” in eachitem with the word “series watching”. Grammatical errorswere also corrected. Respondents had to answer using a 5-point scale(1 = Never; 2 = Rarely; 3 = Sometimes;4 = Often; 5 = Always).

Statistical analysis

Preliminary statistical analysis comprised of the descriptive analysisin SPSS 22 such as means, standard deviations, frequencies, and skewness-kurtosisvalues. Later, estimation of Cronbach’s alpha values, correlations, t-tests and ANOVAs were performedwith this software as well. All cases had complete data in both samples. 1

Participants were asked to estimate the amount of free time they have on an average weekday and weekend.According to the survey of the Hungarian Central Statistical Office(2011), on an average day,a Hungarian individual (aged between 15 and 74) spends 712 minuteswith satisfying his/her physiological needs (i.e. sleeping, eating,hygiene) and the remaining 728 minutes is for other activities. Basedon these results, we set a threshold for the amount of possible maximumfree time one individual can have in our sample in order to minimizebias. Reported amount of free time more than 728 minutes were recodedas missing data, however, these cases were not completely deleted.To have a single indicator of free time for an average day of theweek, the reported values were weighted by the following formula:(weekday time*5 + weekend time*2)/7. After recodingthe high values, 344 cases were retained where a realistic amountof time was indicated, whereas 22 cases were consideredas missing for this variable.

Prior to the confirmatory factor analyses (CFA), the data was investigatedfor normality. Regarding univariate normality, the simulation study of Curran, West, and Finch (1996) concluded that problems canarise from having a skewness value above 2.0 and a kurtosis valueabove 7.0. Also, multivariate normality was examined in Mplus 7.3(Muthén & Muthén,1998–2012) using two-sided test of fit for skewness andkurtosis (Wang & Wang, 2012). As these tests were statistically significant in both Sample 1and Sample 2, the assumption of multivariate normality was violated.

In order to investigate the factor structure of this new measurement,a series of CFAs were conducted using Mplus 7.3. Since the data didnot have multivariate normal distribution, the robust maximum-likelihoodestimator (MLR) was used instead of maximum-likelihood (ML). Multiplegoodness of fit indices were taken into consideration based on therecommendations of Brown (2015): the comparative fit index (CFI), the Tucker–Lewis index(TLI), the root mean square error of approximation (RMSEA) and its 90% confidence interval, and test of close fit (CFit),the standardized root mean square residuals (SRMR). Guided by suggestionsof several methodologists (Bentler,1990; Brown,2015; Browne & Cudeck, 1993; Hu & Bentler, 1999; Schermelleh-Engel, Moosbrugger, & Müller,2003), good or acceptable model fit was defined by the followingcriteria: CFI (≥.95 for good, ≥.90 for acceptable), TLI (≥.95for good, ≥.90 for acceptable), RMSEA (≤.06 for good, ≤.08for acceptable), CFit (≥.10 for good, ≥.05 for acceptable),and SRMR (≤.05 for good, ≤.10 for acceptable).

Moreover, to test whether the amount of free time could have aneffect on PSWS scores, a multiple indicators multiplecauses (MIMIC) analysis (Brown, 2015) was carried out. The MIMIC model consists of a measurement model(previously established in the CFA) and a structural model which makesit possible to estimate the effect of indicators (spare time) on thelatent variable (PSWS) while controlling for other variables.

Finally, regarding reliability, internal consistencywas measured by Cronbach’s alpha taking Nunnally’s (1978) suggestions into considerationregarding the acceptability of thevalue (.70 is acceptable, .80 is good). However, as Cronbach’salpha value could have biases which could inflate or deflate the reliability(see Osburn, 2000 or Sijtsma, 2009), two additional valueswere calculated: factor determinacy and composite reliability. Factordeterminacy refers to the correlation between the estimated and thetrue factor scores and it describes how well the factor is describedby the indicators. It ranges from zero to one with higher scores indicatinghigher levels of reliability (Muthén &Muthén, 1998–2012). Composite reliability can also be considered when assessing a model.Values above .70 should be considered acceptable (Nunnally & Bernstein, 1994). Additionally, inter-itemcorrelations were also observed with values between .15 and .50 consideredacceptable (Clark & Watson, 1995).

Ethics

In case of both Sample 1 and Sample 2, the studies were conductedin accordance with the Declaration of Helsinki and were approved bythe Institutional Review Board of Eötvös Loránd University,Budapest, Hungary. All subjects were informed about the studies theyparticipated in and all provided informed consent.

Results

Since we had two independent datasets, the analyses were carriedout as it follows: Sample 1 was used for the investigation of thefactor structure and for the MIMIC analysis, whereas Sample 2 wasused for cross-validation and for examining differences across gender,age, educational level, and place of residence. Several solutionswere observed: (1) a 12-item unidimensional solution, (2) a 6-itemunidimensional solution based on the BWAS, (3) a 6-item unidimensionalsolution which retained one item per factor based on modificationindices as suggested by Brown (2015). In a second step, using the more diverse Sample 2 for cross-validatingthis factor structure, we carried out the analysis with the 6-itemversion which was based on modification indices.

Structural analysis

For the initial 12 items of the PSWS, skewnessvalues ranged from −.24 to 1.94, and kurtosis values rangedfrom −1.11 to 2.60 which were within the acceptability rangeproposed by Curran et al. (1996). Next, a series of confirmatory factor analyses were conductedon Sample 1 in order to test alternative models. The comparison ofthe examined models can be seen in Table 1. The results showed that the unidimensional 6-item model showed acceptable model fit in the caseof both Sample 1 (CFI = .98, TLI = .97,RMSEA = .04 [90% CI .00–.08], CFit = .64,SRMR = .03) and Sample 2 (CFI = .96, TLI = .93,RMSEA = .07 [90% CI .05–.09], CFit = .06,SRMR = .03).

Table 1.

Confirmatory factor analyses results of the Problematic SeriesWatching Scale (PSWS)

		CFI	TLI	RMSEA [90% CI]	CFit	SRMR
Sample 1	12-item unidimensional model (all initial items are included)	.65	.58	.15 [.14–.16]	.00	.09
	6-item unidimensional model of the BWAS (items2, 3, 6, 7, 10, 11)	.94	.89	.07 [.04–.10]	.13	.04
	6-item unidimensional model of the PSWS (items2, 3, 6, 7, 9, 12)	.98	.97	.04 [.00–.08]	.64	.03
Sample 2	6-item unidimensional model of the PSWS (items 2, 3, 6, 7, 9,12)	.96	.93	.07 [.05–.09]	.06	.03

Notes: CFI = comparativefit index; TLI = Tucker–Lewis index; RMSEA = rootmean square error of approximation; CFit = RMSEA’stest of close fit; SRMR = standardized root meansquare residuals.

Standardized factor loadings, reliability indices (Cronbach’s alpha, factor determinacy, compositereliability, and inter-item correlations) and descriptive statisticsregarding both Sample 1 and 2 can be seen in Table 2. Factor loadings were acceptable(ranging from .43 to .62 in Sample 1 and from .52 to .68 in Sample2). Although the Cronbach’s alpha value was borderline in thecase of Sample 1 (α = .69), other reliabilityindices had adequate values and inter-item correlations were alsowithin the acceptable range for both samples. These results indicatethat the PSWS has good factor structure and acceptable reliability.For the final Hungarian and English versions, see Appendix.

Table 2.

Standardized factor loadings, reliability indices and descriptivestatistics of the Problematic Series Watching Scale (PSWS) with two independent samples

		Sample 1 (N = 366)	Sample 2 (N = 752)
Standardized factor loadings	Salience 1	.54	.54
	Tolerance 2	.42	.60
	Mood modification 3	.41	.52
	Relapse 4	.62	.59
	Withdrawal 5	.62	.67
	Conflict 6	.69	.68
Reliability indices	Cronbach’s alpha	.69	.76
	Factor determinacy	.85	.88
	Composite reliability	.73	.77
	Mean inter-item correlations	.29	.36
Descriptive statistics	Mean (SD)	13.85 (4.67)	12.62 (4.41)
	Skewness (SD)	.62 (.13)	.85 (.09)
	Kurtosis (SD)	–.24 (.25)	.67 (.18)

MIMIC model analysis

In order to investigate the effect of the amount of free time onproblematic series watching scores, a MIMIC analysis was applied.The model fit indices showed that the model remainedacceptable (CFI = .99, TLI = .99,RMSEA = .02 [90% CI .00–.06], CFit = .91,SRMR = .03). Daily average spare time was notsignificantly (β = .01, p = .93) associated with the PSWS latentvariable.

Gender, age, educational level and place of residencedifferences

PSWS scores moderately correlated with series watching time [r(364) = .27, p < .001)].For further measuring demographic differences, we used the largerand more diverse sample (Sample 2) of 752 cases. Women (Mfemale = 12.91, SDfemale = 4.50) had higher scores [t(750) = −2.90, p < .01] than men(Mmale = 11.86, SDmale = 4.08). Relatively weak negative correlations were found betweenage and PSWS [r(750) = −.21, p = .001]. Using one-way ANOVA (withBonferroni-corrected post-hoc test), educational level-related differenceswere measured between the three groups [F(2, 749) = 6.44, p < .01]. Those who have higher educationdegree (M = 11.97, SD = 4.19) scored significantly lower on PSWSthan those who have elementary school (M = 13.96, SD = 4.96, p = .011)or high school degree (M = 12.92, SD = 4.44, p = .013).Using the same method no place of residence-related differences werefound.

Discussion

Our results show that the PSWS has an appropriate factor structureand reliability. Respondents watch series more than one hour per daywhich is more than one-fifth of their free time which indicated thatseries watching might be an important free time activity. However,the amount of free time one has is not associated with PSWS scores.Women had higher scores on PSWS and respondents with higher educationhad lower scores on it.

We have to consider the concept of overpathologization (Billieux et al., 2015) whichargues that everyday activities are being turned into behavioral addictions.Indeed, it is not obvious that problematic series watching affectsa large part of the population. Moreover, clinically validated diagnostictests with adequate levels of sensitivity and specificity are requiredfor establishing an accurate diagnosis (Maráz, Király, & Demetrovics, 2015). Inlight of these suggestions, it can be said that the problematic serieswatching belongs to a group of problematicbehaviors – along with for instance dance (Maráz, Urbán, Griffiths, & Demetrovics,2015) and smartphone use (Wang, Wang, Gaskin, & Wang, 2015) – that are not necessarily as addictive and prevalent as classical substance addictions. Still, it mightbe important to consider these problematicbehaviors in today’s changing era.

Given the lack of empirical research on series watching, we supposedthat it might be similar to other problematic screen-related behaviors(e.g. online gaming, Internet or Facebook use). Although health-relatedvariables were not included in this research, intense screen-timecan be related to various health problems, such as reduced physicaland psychosocial health (Tremblayet al., 2011), increased cardiovascular risk (Grøntved et al. 2014),sleeping problems (Do, Shin, Bautista, &Foo, 2013), lower levels of life satisfaction (Mentzoni et al., 2011), or interpersonalproblems (Lo, Wang, & Fang, 2005).

Some limitations of this research need to be addressed. This wasa questionnaire-based study which is prone to bias. However, thislimitation could be overcome by implementing more objective measuresthat would respect the individual’s privacy as well. Althoughthe two samples were diverse, neither was representative which limitsthe generalization of the results. Regarding the PSWS, the resultswere based on a correlational design which does not make it possibleto infer causality. More research is needed to examine its temporalstability, convergent, divergent, and predictive validity in differentcultures.

In this new research area where cross-sectional studies are rare,a longitudinal design could be fruitful in examining how series watchingis affected by different life events and also how it might impactone’s health. In terms of clinical practice, prevalence andincidence should be investigated. Further research is needed to explorewhether problematic series watching and other problematic online behaviorshave the same roots. It is possible that they have the same negativeconsequences. Other possible covariates could be examined in the futuresuch as loneliness or urgency. Also, further investigation is neededwhether extensive series watching can lead to health and psychosocialproblems.

Conclusions

To our best knowledge no prior study examined problematic serieswatching. In the present study a 6-item Problematic Series WatchingScale was created on the basis of the six-component model of Griffiths(2005). The scale has goodfactor structure and reliability. PSWS scores are positively relatedwith time spent on series watching, whereas the amount of free timedoes not have an effect on PSWS scores. In the more and more digitalizedworld there are many forces which encourage people watching onlineseries. In the light of these changes, research on problematic serieswatching will be increasingly relevant.

Authors’contribution

All authors (GO, BB, ITK) contributed to the manuscript equally.

Conflictof interest

The authors declare no conflict of interest.

ENGLISH VERSION

Below you find 6 questions related to series watching. Answer eachof the 6 questions by selecting one response alternative (rangingfrom “never” to “always”) that best describesyou.

1 – Never.

2 – Rarely.

3 – Sometimes.

4 – Often.

5 – Always.

KEY: Add the scores of the items then divide by the numberof the items.

HUNGARIAN VERSION

A következőkben 6 sorozatnézéssel kapcsolatostételt olvashatsz. Válaszolj a kérdésekreaz alábbi skála segítségével:

1 – Soha.

2 – Ritkán.

3 – Néha.

4 – Gyakran.

5 – Mindig.

KIÉRTÉKELÉS: A tételek pontszámát összekell adni, majd elosztani a tételek számával.