Register here: http://gg.gg/w6vsc
This article is available in: PDFHTMLEvaluating the Reliability and Validity of the Short Gambling Harm Screen: Are Binary Scales Worse Than Likert Scales at Capturing Gambling Harm?
Assessing gambling problems that form a quantitative index called the Problem Gambling Severity Index (PGSI). Four gambler subtypes have been identified based on the total PGSI score: non-problem, low-risk, moderate-risk, and problem gamblers. In the 10 years since the CPGI was first introduced, criticism of the scale has surfaced. The Problem Gambling Severity Index (PGSI) is a brief measure that allows for the asse. Increases in the availability of gambling heighten the need for a short screening measure of problem gambling. The Problem Gambling Severity Index (PGSI) is a brief measure that allows for the asse. Problem Gambling Severity Index 0 Never 1 Sometimes 2 Most of the time 3 Almost always. This self-assessment is based on the Canadian Problem Gambling Index. It will help you decide if you wish to seek other forms of support or information. When you gambled, did you go back another day to.Journal Information
Journal ID (publisher-id): jgi
ISSN: 1910-7595
Publisher: Centre for Addiction and Mental Health
Article Information
Article Categories: Original Research
Publication date: Spring 2020
Publisher Id: jgi.2020.44.6
DOI: 10.4309/jgi.2020.44.6
*A large, integrated survey data set provided by the Ontario Problem Gambling Centre was used to investigate psychometric properties of the Problem Gambling Severity Index (PGSI). This nine-item self-report instrument was designed to measure a single, problem gambling construct.
*Background: The Problem Gambling Severity Index (PGSI) was intended for use in epidemiological research with gamblers across the continuum of risk. (Gambling Help Online GHO) from October 2012 to December 2015 (n = 5,881) and trial data evaluating an Australian online self-directed program for gambling (GAMBLINGLESS; n = 198).James McLauchlanExperimental Gambling Research Laboratory, School of Health, Medical and Applied Sciences, CQUniversity, Bundaberg, Queensland, AustraliaMatthew BrowneExperimental Gambling Research Laboratory, School of Health, Medical and Applied Sciences, CQUniversity, Bundaberg, Queensland, AustraliaAlex M. T. RussellExperimental Gambling Research Laboratory, School of Health, Medical and Applied Sciences, CQUniversity, Sydney, New South Wales, AustraliaMatthew RockloffExperimental Gambling Research Laboratory, School of Health, Medical and Applied Sciences, CQUniversity, Bundaberg, Queensland, AustraliaAbstract
Gambling-related harm has become a key metric for measuring the adverse consequences of gambling on a population level. Yet, despite this renewed understanding in contemporary research, little exploration has been conducted to evaluate which instrument is best suited to capture the harmful consequences of gambling. This study was designed with the aim of determining whether Likert scales were better suited to capture gambling harm than binary scales. We hypothesized that the Short Gambling Harm Screen (SGHS), initially scored using a binary scale, would perform similarly to the alternate form that was Likertized for the purpose of this study. A corresponding comparison in the reverse direction was executed for the Problem Gambling Severity Index. The SGHS’s performance was assessed via a repeated-measures design in combination with three other measures of validity administered at the conclusion of the survey. In the end, we found that changing the scoring format (i.e., from binary to Likert) had negligible impact on the SGHS’s psychometric performance. We conclude that the original scoring method of the SGHS is not only appropriate but also no less suitable than Likert scales in measuring gambling harm.
Keywords: gambling harm, Short Gambling Harm Screen (SGHS), forced-choice binary, dichotomous scale, binary scale, Likert scale comparison, Problem Gambling Severity Index (PGSI)
Résumé
Les dommages liés au jeu sont devenus une mesure clé pour évaluer les conséquences néfastes du jeu à l’échelle de la population. Pourtant, malgré cette compréhension renouvelée dans la recherche contemporaine, on effectue très peu d’exploration pour évaluer quel instrument est le mieux adapté pour comprendre les conséquences néfastes du jeu. Cette étude a été conçue dans le but de déterminer si les échelles de Likert étaient mieux adaptées que les échelles binaires pour saisir les dommages liés au jeu. Nous avons émis l’hypothèse que le dépistage rapide du jeu problématique (Short Gambling Harm Screen ou SGHS), initialement évalué à l’aide d’une échelle binaire, ne fonctionnera pas différemment de la forme de Likert alternative qui a été créée aux fins de cette étude. Une comparaison correspondante dans la direction inverse a été effectuée pour l’indice de gravité du jeu excessif (PGSI). Les performances du SGHS ont été évaluées par un plan de mesures répétées, combinés à trois autres mesures de validité administrées à la fin du sondage. En fin de compte, nous avons constaté que le changement du format de pointage (c.-à-d. du binaire au Likert) avait un impact négligeable sur le rendement psychométrique du SGHS. Nous concluons que la méthode de pointage originale du SGHS est non seulement appropriée, mais également non moins appropriée que les échelles de Likert pour évaluer les dommages liés au jeu.Introduction
Contemporary research has focused on gambling-related harm as a key metric of the negative impacts of gambling at the population level (Blaszczynski, 2009; Browne, Greer, Rawat, & Rockloff, 2017; Rodgers, Caldwell, & Butterworth, 2009; Sproston, Erens, & Orford, 2000). The emphasis on harm, rather than gambling disorders, recognizes that traditional measures such as the Problem Gambling Severity Index (PGSI; Ferris & Wynne, 2001) are not well suited to measure the impact of harm on a population level. The need for a new measure of harm was met by a new 10-item screen dedicated to measuring harm—the Short Gambling Harm Screen (SGHS; Browne, Goodwin, & Rockloff, 2017). However, Delfabbro and King (2017) have raised concerns regarding the use of binary scoring of each of the harm symptomology indicators. This dispute raises the question of whether a count of the presence of symptoms, as used by the SGHS, is inferior to measures that elicit degree of frequency or intensity with respect to gambling harm. The present study aimed to evaluate this question via a repeated measures design, in which the performance of the two response formats are compared on several psychometric criteria.Harm-centred measurement approaches
A population health approach to gambling problems implies that harm, understood as a decrement to health and wellbeing, is the key outcome to be addressed. A corollary to this is that harm can occur on a continuum from mild to severe; and a practical observation is that prevalence is much lower at the severe end of the spectrum (Browne, Greer, et al., 2017). For instance, Raisamo, Mäkelä, Salonen, and Lintonen’s (2014) found that considerable harms were reported even at the lower end of gambling frequency and expenditure levels. A population study conducted in the UK has revealed similar trends, reporting individuals experiencing harms were most prevalent in the lower gambling consumption groups (Canale, Vieno, & Griffiths, 2016). In Australia, Browne and Rockloff (2018) conducted a study assessing the prevalence of harmful consequences across four problem-gambling risk categories, including no-risk, low-risk, moderate-risk, and problem gamblers. The data, again, showed that most gambling-related harms are much more common in combined categories of low-risk gamblers than the high-risk problem gamblers. Together, the evidence suggests there is merit in gauging population-level impact across the spectrum of harm, rather than relying solely on prevalence of problem gamblers as a proxy for harm.Binary scales or Likert scales?
It is perhaps intuitively appealing to suppose that Likert scales are generally more reliable and accurate than a binary response format because of their potential for capturing more information. However, the extant research suggests this is not generally the case. Grassi et al.’s (2007) study provides a useful illustration. The authors replaced the Likert scales in the 36-item short-form health survey (SF-36) with forced-choice binary scales, and found that the answering format had “no substantial effect” on the test-retest reliability or internal consistency. In another study, Geldhof et al. (2015) compared the responses collected using both binary and Likert format of the Selection Optimisation and Compensation (SOC) questionnaire and concluded that the answering formats were practically interchangeable. Further, in a study published by Litong-Palima, Albers and Glückstad’s (2018) the binary format outperformed its Likert counterparts on measures of reliability. Considering research in the marketing context, binary scales have consistently demonstrated similar reliability to Likert scales (Dolnicar & Grün, 2013a; Dolnicar & Grün, 2013b Dolnicar, Grün, & Leisch, 2011; Dolnicar & Leisch, 2012). A common thread running through these studies is the findings that binary scale do not perform significantly differently from their Likert counterparts.
The lack of evidence for the superiority of Likert over binary response formats is counterintuitive considering the greater potential for informational content in an interval scale. Likert scales provide participants with the opportunity to choose from a range of responses to denote a degree of agreement, frequency, or severity. These ordered responses, typically between four to seven points (Adelson & McCoach, 2010), provide the potential for participants to indicate a more precise response to the probe. Nevertheless, there is an absence of guidelines on the way in which Likert scales are to be designed. For instance, there are several options for answer stems (e.g., likely, agree, most of the time, etc.). There is also no definitive way by which the resulting scores should be aggregated. For example, certain researchers advocate for the use of neutral mid-points (Raaijmakers, Van Hoof, ’t Hart, Verbogt, & Vollebergh, 2000; Velez & Ashworth, 2007), while others warn against them (Guy & Norvell, 1977; Wakita, Ueshima, & Noguchi, 2012). The optimal number of rating categories vary from two (McCallum, Keith, & Wiebe, 1988) to eleven (Cummins & Gullone, 2000; Leung, 2011). Certain researchers argue that reliability increases with the number of scale points (Lozano, García-Cueto, & Muñiz, 2008; Weng, 2004), while others have found evidence suggesting that reliability is largely independent of the number of scale points (Bendig, 1954; Komorita, 1963; Matell & Jacoby, 1971).
Theoretical considerations may also explain why Likert response formats do not, in practice, tend to perform better than their binary counterparts for many applications. Given that Likert items typically yield scores (e.g., 0, 1, 2, 3) that are then summed across items to create a scale score, this format requires the strong (item-response theoretic) assumption that each step in the ordered response represents an identical difference of degree on the hypothesized latent construct (Michell, 2012). Binary scales involve only the weaker assumption that the various items are similarly related to, or load onto, the underlying construct. It is also worth considering the higher degree of cognitive effort employed by respondents in answering with a Likert scale, and the degree to which differences in ordered responses might therefore reflect either noise, or a systematic bias in terms of minimising or maximising responses. Binary responses, such as reporting whether an event happened or alternatively whether a symptom is present, are arguably inherently more concrete and less ambiguous, and may therefore be less vulnerable to these forms of error.
Despite the heavy reliance on surveys as the main method for data collection on gambling harm, the question of response format has not yet been explored within gambling research. Even though the SGHS and the FocaL Adult Gambling Screen (FLAGS; Schellinck, Schrans, Schellinck, & Bliemel, 2015) are both scored using a binary response format, neither has been subject to a similar analysis in response to the concerns raised by Delfabbro and King (2017). The aim of the present study is, therefore, to examine the influence of different response formats have on the psychometric properties of the SGHS. More specifically, the research objective is to compare the reliability of the SGHS, initially scored using a binary scale, against a Likert version of SGHS to determine which scale format is more suited for capturing gambling harm. The present study hypotheses that psychometric performance of the binary SGHS will not differ substantially (i.e., the difference will be below the p < .05 threshold), in both reliability and validity, from the alternate Likert form.MethodsParticipants
Adult gamblers (n = 618) who gamble at least two to four times a month were recruited for this study through TurkPrime, a North American online research panel recruitment service. Participants who had missing answers (n = 42), showed pattern responding (n = 17), or scored greater than 2 standard deviations apart in their responses between the repeated measures were excluded (n = 4). Additional multivariate outliers (n = 23) were identified using Mahalanobis distance with a p < .05 threshold, and subsequently removed from the sample. A total of 532 (female = 204) participants aged from 18 to 87 (M = 42.07, SD = 13.13) were included for analysis. See Table 1 for the participant demographic characteristic summary.Design
Participants completed two tests with alternative forms of SGHS and PGSI over a one-week test-retest interval. They were randomly allocated to either complete the same form (i.e., Likert-Likert or Binary-Binary) or the alternative forms (i.e., Likert-Binary or Binary-Likert) at the one-week follow-up. Though the participants might complete different forms of the SGHS and PGSI across time one and time two, the forms did not differ in the same testing (i.e., if a participant received the Binary SGHS at time-one, they will also receive the binary PGSI at time-one). See Table 2 for a summary of the different permutations. Approximately 44% (n = 234) completed repeat assessment of the same form, while the remaining 56% (n = 298) completed the alternate form at follow-up. Participants also completed several other validation measures (described below) at the end of the one-week follow-up survey.Procedure
Participants were recruited to participate in two online surveys. They were compensated in the form of either reward points, cash or gift cards of their choice. This study was approved by the Central Queensland University Ethics Committee (approval number 0000021464), and informed consent was obtained at the outset of the first survey.Analysis
Each measure’s internal consistency was calculated using Cronbach’s alpha calculated on either the tetrachoric (for binary) or polychoric (for Likert) item correlation matrix. The SGHS’s test-retest reliability, alternate-form reliability, convergent validity and discriminant validity was computed using Spearman correlations. Comparisons between two forms of SGHS was done using Fisher’s Z test (Myers & Sirois, 2006), which provides a test of significance between nonparametric correlation coefficients by converting them into standardized (z) scores (Zar, 2005).Measures
In addition to the SGHS and PGSI, several measures were included to assess external validity for each version of the scale. The Kessler Psychological Distress Scale (K6; Kessler et al., 2002) and the Personal Wellbeing Index (PWI; Cummins, 1997) served as outcome-oriented validity; that is, harms and problems should be associated with increased distress and lower wellbeing. Additionally, impulsivity is a known risk factor for both gambling problems and harm (Browne et al., 2019; Russell, Hing, Li, & Vitartas, 2019), and therefore trait-impulsivity was measured using the Barratt Impulsiveness Scale-Brief (BIS-Brief; Steinberg, Sharp, Stanford, & Tharp, 2013).Gambling Harm
The SGHS (Browne, Goodwin et al., 2017) is designed to assess a respondent’s degree of gambling-related harm. The original SGHS is scored on a forced-choice binary scale and contains 10 items derived from a comprehensive 72-item checklist (Browne, Goodwin et al., 2017, Langham et al., 2015). The SGHS includes harmful consequences that are more prevalent (e.g., “decreased savings” or “sold personal items”) and uses a binary response scale. It has been shown to be a good proxy of the full checklist (r = .94) and enjoys a negative linear relationship to wellbeing (Browne, Goodwin et al., 2017). This scale was compared with a four-point Likert version of the same measure with 0 being “never,” 1 denoting “sometimes,” 2 representing “most of the time,” and 3 indicating “almost always.” Alpha reliability in the present study for the original binary SGHS was .95 at both times one and two. The Likert-scored SGHS alpha reliability was also identical across surveys: .97 at time one and two. Refer to Table 6 for the combined response distribution of both forms of SGHS.Problem Gambling
Participants completed the PGSI (Ferris and Wynne, 2001), a standard tool for assessing degree of gambling problems in surveys. The nine-item PGSI contains questions such as “has gambling caused you any health problems, including stress or anxiety?” It is scored on a four-point Likert scale, with 0 representing “never” and 3 indicating “almost always.” An alternate binary form of the PGSI was also calculated for this study. Alpha reliability in this study for the standard (Likert) PGSI was .96 at time one and .98 at time two; the alternative binary version had a reliability of .89 at both time one and two. See Table 5 for the response distribution of the PGSI.Psychological Distress
The Kessler Screening Scale for Psychological Distress (K6; Kessler et al., 2002) was chosen to measure the presence of distress among the participants. The K6 consists of six items scored on a five-point Likert scale from 0 (“none of the time”) to 4 (“all the time”). Coefficient alpha in the current study was .94. As noted above, the SGHS should predict greater psychological distress (Brown, Oldenhoff, Allen, & Dowling, 2016).Personal Wellbeing
The PWI, adapted from the Comprehensive Quality of Life Scale (Cummins, 1997) was used to measure general life satisfaction. It is an eight-item questionnaire designed to measure multiple domains associated with quality of life, including living standards, health, achievements, safety, sense of belonging and future prospects. These items are sco
https://diarynote.indered.space
This article is available in: PDFHTMLEvaluating the Reliability and Validity of the Short Gambling Harm Screen: Are Binary Scales Worse Than Likert Scales at Capturing Gambling Harm?
Assessing gambling problems that form a quantitative index called the Problem Gambling Severity Index (PGSI). Four gambler subtypes have been identified based on the total PGSI score: non-problem, low-risk, moderate-risk, and problem gamblers. In the 10 years since the CPGI was first introduced, criticism of the scale has surfaced. The Problem Gambling Severity Index (PGSI) is a brief measure that allows for the asse. Increases in the availability of gambling heighten the need for a short screening measure of problem gambling. The Problem Gambling Severity Index (PGSI) is a brief measure that allows for the asse. Problem Gambling Severity Index 0 Never 1 Sometimes 2 Most of the time 3 Almost always. This self-assessment is based on the Canadian Problem Gambling Index. It will help you decide if you wish to seek other forms of support or information. When you gambled, did you go back another day to.Journal Information
Journal ID (publisher-id): jgi
ISSN: 1910-7595
Publisher: Centre for Addiction and Mental Health
Article Information
Article Categories: Original Research
Publication date: Spring 2020
Publisher Id: jgi.2020.44.6
DOI: 10.4309/jgi.2020.44.6
*A large, integrated survey data set provided by the Ontario Problem Gambling Centre was used to investigate psychometric properties of the Problem Gambling Severity Index (PGSI). This nine-item self-report instrument was designed to measure a single, problem gambling construct.
*Background: The Problem Gambling Severity Index (PGSI) was intended for use in epidemiological research with gamblers across the continuum of risk. (Gambling Help Online GHO) from October 2012 to December 2015 (n = 5,881) and trial data evaluating an Australian online self-directed program for gambling (GAMBLINGLESS; n = 198).James McLauchlanExperimental Gambling Research Laboratory, School of Health, Medical and Applied Sciences, CQUniversity, Bundaberg, Queensland, AustraliaMatthew BrowneExperimental Gambling Research Laboratory, School of Health, Medical and Applied Sciences, CQUniversity, Bundaberg, Queensland, AustraliaAlex M. T. RussellExperimental Gambling Research Laboratory, School of Health, Medical and Applied Sciences, CQUniversity, Sydney, New South Wales, AustraliaMatthew RockloffExperimental Gambling Research Laboratory, School of Health, Medical and Applied Sciences, CQUniversity, Bundaberg, Queensland, AustraliaAbstract
Gambling-related harm has become a key metric for measuring the adverse consequences of gambling on a population level. Yet, despite this renewed understanding in contemporary research, little exploration has been conducted to evaluate which instrument is best suited to capture the harmful consequences of gambling. This study was designed with the aim of determining whether Likert scales were better suited to capture gambling harm than binary scales. We hypothesized that the Short Gambling Harm Screen (SGHS), initially scored using a binary scale, would perform similarly to the alternate form that was Likertized for the purpose of this study. A corresponding comparison in the reverse direction was executed for the Problem Gambling Severity Index. The SGHS’s performance was assessed via a repeated-measures design in combination with three other measures of validity administered at the conclusion of the survey. In the end, we found that changing the scoring format (i.e., from binary to Likert) had negligible impact on the SGHS’s psychometric performance. We conclude that the original scoring method of the SGHS is not only appropriate but also no less suitable than Likert scales in measuring gambling harm.
Keywords: gambling harm, Short Gambling Harm Screen (SGHS), forced-choice binary, dichotomous scale, binary scale, Likert scale comparison, Problem Gambling Severity Index (PGSI)
Résumé
Les dommages liés au jeu sont devenus une mesure clé pour évaluer les conséquences néfastes du jeu à l’échelle de la population. Pourtant, malgré cette compréhension renouvelée dans la recherche contemporaine, on effectue très peu d’exploration pour évaluer quel instrument est le mieux adapté pour comprendre les conséquences néfastes du jeu. Cette étude a été conçue dans le but de déterminer si les échelles de Likert étaient mieux adaptées que les échelles binaires pour saisir les dommages liés au jeu. Nous avons émis l’hypothèse que le dépistage rapide du jeu problématique (Short Gambling Harm Screen ou SGHS), initialement évalué à l’aide d’une échelle binaire, ne fonctionnera pas différemment de la forme de Likert alternative qui a été créée aux fins de cette étude. Une comparaison correspondante dans la direction inverse a été effectuée pour l’indice de gravité du jeu excessif (PGSI). Les performances du SGHS ont été évaluées par un plan de mesures répétées, combinés à trois autres mesures de validité administrées à la fin du sondage. En fin de compte, nous avons constaté que le changement du format de pointage (c.-à-d. du binaire au Likert) avait un impact négligeable sur le rendement psychométrique du SGHS. Nous concluons que la méthode de pointage originale du SGHS est non seulement appropriée, mais également non moins appropriée que les échelles de Likert pour évaluer les dommages liés au jeu.Introduction
Contemporary research has focused on gambling-related harm as a key metric of the negative impacts of gambling at the population level (Blaszczynski, 2009; Browne, Greer, Rawat, & Rockloff, 2017; Rodgers, Caldwell, & Butterworth, 2009; Sproston, Erens, & Orford, 2000). The emphasis on harm, rather than gambling disorders, recognizes that traditional measures such as the Problem Gambling Severity Index (PGSI; Ferris & Wynne, 2001) are not well suited to measure the impact of harm on a population level. The need for a new measure of harm was met by a new 10-item screen dedicated to measuring harm—the Short Gambling Harm Screen (SGHS; Browne, Goodwin, & Rockloff, 2017). However, Delfabbro and King (2017) have raised concerns regarding the use of binary scoring of each of the harm symptomology indicators. This dispute raises the question of whether a count of the presence of symptoms, as used by the SGHS, is inferior to measures that elicit degree of frequency or intensity with respect to gambling harm. The present study aimed to evaluate this question via a repeated measures design, in which the performance of the two response formats are compared on several psychometric criteria.Harm-centred measurement approaches
A population health approach to gambling problems implies that harm, understood as a decrement to health and wellbeing, is the key outcome to be addressed. A corollary to this is that harm can occur on a continuum from mild to severe; and a practical observation is that prevalence is much lower at the severe end of the spectrum (Browne, Greer, et al., 2017). For instance, Raisamo, Mäkelä, Salonen, and Lintonen’s (2014) found that considerable harms were reported even at the lower end of gambling frequency and expenditure levels. A population study conducted in the UK has revealed similar trends, reporting individuals experiencing harms were most prevalent in the lower gambling consumption groups (Canale, Vieno, & Griffiths, 2016). In Australia, Browne and Rockloff (2018) conducted a study assessing the prevalence of harmful consequences across four problem-gambling risk categories, including no-risk, low-risk, moderate-risk, and problem gamblers. The data, again, showed that most gambling-related harms are much more common in combined categories of low-risk gamblers than the high-risk problem gamblers. Together, the evidence suggests there is merit in gauging population-level impact across the spectrum of harm, rather than relying solely on prevalence of problem gamblers as a proxy for harm.Binary scales or Likert scales?
It is perhaps intuitively appealing to suppose that Likert scales are generally more reliable and accurate than a binary response format because of their potential for capturing more information. However, the extant research suggests this is not generally the case. Grassi et al.’s (2007) study provides a useful illustration. The authors replaced the Likert scales in the 36-item short-form health survey (SF-36) with forced-choice binary scales, and found that the answering format had “no substantial effect” on the test-retest reliability or internal consistency. In another study, Geldhof et al. (2015) compared the responses collected using both binary and Likert format of the Selection Optimisation and Compensation (SOC) questionnaire and concluded that the answering formats were practically interchangeable. Further, in a study published by Litong-Palima, Albers and Glückstad’s (2018) the binary format outperformed its Likert counterparts on measures of reliability. Considering research in the marketing context, binary scales have consistently demonstrated similar reliability to Likert scales (Dolnicar & Grün, 2013a; Dolnicar & Grün, 2013b Dolnicar, Grün, & Leisch, 2011; Dolnicar & Leisch, 2012). A common thread running through these studies is the findings that binary scale do not perform significantly differently from their Likert counterparts.
The lack of evidence for the superiority of Likert over binary response formats is counterintuitive considering the greater potential for informational content in an interval scale. Likert scales provide participants with the opportunity to choose from a range of responses to denote a degree of agreement, frequency, or severity. These ordered responses, typically between four to seven points (Adelson & McCoach, 2010), provide the potential for participants to indicate a more precise response to the probe. Nevertheless, there is an absence of guidelines on the way in which Likert scales are to be designed. For instance, there are several options for answer stems (e.g., likely, agree, most of the time, etc.). There is also no definitive way by which the resulting scores should be aggregated. For example, certain researchers advocate for the use of neutral mid-points (Raaijmakers, Van Hoof, ’t Hart, Verbogt, & Vollebergh, 2000; Velez & Ashworth, 2007), while others warn against them (Guy & Norvell, 1977; Wakita, Ueshima, & Noguchi, 2012). The optimal number of rating categories vary from two (McCallum, Keith, & Wiebe, 1988) to eleven (Cummins & Gullone, 2000; Leung, 2011). Certain researchers argue that reliability increases with the number of scale points (Lozano, García-Cueto, & Muñiz, 2008; Weng, 2004), while others have found evidence suggesting that reliability is largely independent of the number of scale points (Bendig, 1954; Komorita, 1963; Matell & Jacoby, 1971).
Theoretical considerations may also explain why Likert response formats do not, in practice, tend to perform better than their binary counterparts for many applications. Given that Likert items typically yield scores (e.g., 0, 1, 2, 3) that are then summed across items to create a scale score, this format requires the strong (item-response theoretic) assumption that each step in the ordered response represents an identical difference of degree on the hypothesized latent construct (Michell, 2012). Binary scales involve only the weaker assumption that the various items are similarly related to, or load onto, the underlying construct. It is also worth considering the higher degree of cognitive effort employed by respondents in answering with a Likert scale, and the degree to which differences in ordered responses might therefore reflect either noise, or a systematic bias in terms of minimising or maximising responses. Binary responses, such as reporting whether an event happened or alternatively whether a symptom is present, are arguably inherently more concrete and less ambiguous, and may therefore be less vulnerable to these forms of error.
Despite the heavy reliance on surveys as the main method for data collection on gambling harm, the question of response format has not yet been explored within gambling research. Even though the SGHS and the FocaL Adult Gambling Screen (FLAGS; Schellinck, Schrans, Schellinck, & Bliemel, 2015) are both scored using a binary response format, neither has been subject to a similar analysis in response to the concerns raised by Delfabbro and King (2017). The aim of the present study is, therefore, to examine the influence of different response formats have on the psychometric properties of the SGHS. More specifically, the research objective is to compare the reliability of the SGHS, initially scored using a binary scale, against a Likert version of SGHS to determine which scale format is more suited for capturing gambling harm. The present study hypotheses that psychometric performance of the binary SGHS will not differ substantially (i.e., the difference will be below the p < .05 threshold), in both reliability and validity, from the alternate Likert form.MethodsParticipants
Adult gamblers (n = 618) who gamble at least two to four times a month were recruited for this study through TurkPrime, a North American online research panel recruitment service. Participants who had missing answers (n = 42), showed pattern responding (n = 17), or scored greater than 2 standard deviations apart in their responses between the repeated measures were excluded (n = 4). Additional multivariate outliers (n = 23) were identified using Mahalanobis distance with a p < .05 threshold, and subsequently removed from the sample. A total of 532 (female = 204) participants aged from 18 to 87 (M = 42.07, SD = 13.13) were included for analysis. See Table 1 for the participant demographic characteristic summary.Design
Participants completed two tests with alternative forms of SGHS and PGSI over a one-week test-retest interval. They were randomly allocated to either complete the same form (i.e., Likert-Likert or Binary-Binary) or the alternative forms (i.e., Likert-Binary or Binary-Likert) at the one-week follow-up. Though the participants might complete different forms of the SGHS and PGSI across time one and time two, the forms did not differ in the same testing (i.e., if a participant received the Binary SGHS at time-one, they will also receive the binary PGSI at time-one). See Table 2 for a summary of the different permutations. Approximately 44% (n = 234) completed repeat assessment of the same form, while the remaining 56% (n = 298) completed the alternate form at follow-up. Participants also completed several other validation measures (described below) at the end of the one-week follow-up survey.Procedure
Participants were recruited to participate in two online surveys. They were compensated in the form of either reward points, cash or gift cards of their choice. This study was approved by the Central Queensland University Ethics Committee (approval number 0000021464), and informed consent was obtained at the outset of the first survey.Analysis
Each measure’s internal consistency was calculated using Cronbach’s alpha calculated on either the tetrachoric (for binary) or polychoric (for Likert) item correlation matrix. The SGHS’s test-retest reliability, alternate-form reliability, convergent validity and discriminant validity was computed using Spearman correlations. Comparisons between two forms of SGHS was done using Fisher’s Z test (Myers & Sirois, 2006), which provides a test of significance between nonparametric correlation coefficients by converting them into standardized (z) scores (Zar, 2005).Measures
In addition to the SGHS and PGSI, several measures were included to assess external validity for each version of the scale. The Kessler Psychological Distress Scale (K6; Kessler et al., 2002) and the Personal Wellbeing Index (PWI; Cummins, 1997) served as outcome-oriented validity; that is, harms and problems should be associated with increased distress and lower wellbeing. Additionally, impulsivity is a known risk factor for both gambling problems and harm (Browne et al., 2019; Russell, Hing, Li, & Vitartas, 2019), and therefore trait-impulsivity was measured using the Barratt Impulsiveness Scale-Brief (BIS-Brief; Steinberg, Sharp, Stanford, & Tharp, 2013).Gambling Harm
The SGHS (Browne, Goodwin et al., 2017) is designed to assess a respondent’s degree of gambling-related harm. The original SGHS is scored on a forced-choice binary scale and contains 10 items derived from a comprehensive 72-item checklist (Browne, Goodwin et al., 2017, Langham et al., 2015). The SGHS includes harmful consequences that are more prevalent (e.g., “decreased savings” or “sold personal items”) and uses a binary response scale. It has been shown to be a good proxy of the full checklist (r = .94) and enjoys a negative linear relationship to wellbeing (Browne, Goodwin et al., 2017). This scale was compared with a four-point Likert version of the same measure with 0 being “never,” 1 denoting “sometimes,” 2 representing “most of the time,” and 3 indicating “almost always.” Alpha reliability in the present study for the original binary SGHS was .95 at both times one and two. The Likert-scored SGHS alpha reliability was also identical across surveys: .97 at time one and two. Refer to Table 6 for the combined response distribution of both forms of SGHS.Problem Gambling
Participants completed the PGSI (Ferris and Wynne, 2001), a standard tool for assessing degree of gambling problems in surveys. The nine-item PGSI contains questions such as “has gambling caused you any health problems, including stress or anxiety?” It is scored on a four-point Likert scale, with 0 representing “never” and 3 indicating “almost always.” An alternate binary form of the PGSI was also calculated for this study. Alpha reliability in this study for the standard (Likert) PGSI was .96 at time one and .98 at time two; the alternative binary version had a reliability of .89 at both time one and two. See Table 5 for the response distribution of the PGSI.Psychological Distress
The Kessler Screening Scale for Psychological Distress (K6; Kessler et al., 2002) was chosen to measure the presence of distress among the participants. The K6 consists of six items scored on a five-point Likert scale from 0 (“none of the time”) to 4 (“all the time”). Coefficient alpha in the current study was .94. As noted above, the SGHS should predict greater psychological distress (Brown, Oldenhoff, Allen, & Dowling, 2016).Personal Wellbeing
The PWI, adapted from the Comprehensive Quality of Life Scale (Cummins, 1997) was used to measure general life satisfaction. It is an eight-item questionnaire designed to measure multiple domains associated with quality of life, including living standards, health, achievements, safety, sense of belonging and future prospects. These items are sco
https://diarynote.indered.space
コメント