A critical appraisal of how to employ – or not to employ – the Sexual Risk Survey in international populations

In 2009, Turchik and Garske1 constructed the Sexual Risk Survey (SRS), a self-administered material intended to assess the applicability of 23 statements related to five dimensions of sexual risk-taking behaviors: • Impulsive and unplanned sexual behaviors (e.g., item 1: “How many partners have you engaged in sexual behavior with but not had sex with?”) • Intentions to engage in risky sexual behaviors (e.g., item 4: “How many times have you gone out to bars/parties/social events with the intent of ‘hooking up’ and engaging in sexual behavior but not having sex with someone?”) • Risky sexual behaviors with uncommitted partners that one was not in a relationship with, did not know well, and did not trust (e.g., item 8: “How many partners have you had sex with?”) • Risky sexual acts such as vaginal or oral sex without a condom (e.g., item 9: “How many times have you had vaginal intercourse without a latex or polyurethane condom?”) • Risky anal sexual acts (e.g., item 13: “How many times have you had anal sex without a condom?”) Each of the five sexual risk-taking dimensions includes two to eight items that are scored on the self-reported frequency of sexual risk-taking behaviors over the past six months. Despite the inherent continuous nature of the data, Turchik and Garske1 suggested that the data be discretized into five ordered categories by binning non-null values according to population-specific percentile thresholds: “0” for null values, “1” for non-null values < the 40th percentile, “2” for non-null values ≥ the 40th percentile but < the 70th percentile, “3” for non-null values ≥ the 70th percentile but < the 90th percentile, and “4” for non-null values ≥ the 90th percentile. From the polytomous (discretized) data, they suggested that arithmetic mean scores be computed to reflect the level of endorsement of each of the five sexual risk-taking dimensions. Yet, despite Turchik et al.2 emphasizing that the validity and reliability evidence of the Sexual Risk Survey (SRS) had only been evaluated in populations of college students in the United States of America and urging that its applicability to other populations be investigated, a considerable body of research articles has employed the material in other populations without the prerequisite examination of its validity and reliability evidence. Therefore, examination of the internal structure validity and internal consistency reliability evidence of the Sexual Risk Survey (SRS) with data from different populations is warranted. Moreover, we argue that discretizing inherently continuous data into ordered categories of polytomous data by binning non-null values according to population-specific percentile thresholds, as suggested by Turchik and Garske,1 introduces unwarranted dependence among the scores of individuals within and between populations. To illustrate, let P1 and P2 be two populations. Posterior to discretization, the self-reported frequency of sexual risk-taking behaviors over the past six months of an individual populating P1 is relative and population-specific: it depends on the distribution of the scores of the other individuals populating P1 and P2. Yet, prior to discretization, such data are absolute and individual-specific: they do not depend on the distribution of the scores of the other individuals populating P1 and P2. Therefore, examination of the internal structure validity and internal consistency reliability evidence of the Sexual Risk Survey (SRS) with continuous (non-discretized) data from different populations is also warranted. Therefore, in the present research article, as urged by Turchik et al.,2 we examined the internal structure validity and internal consistency reliability evidence of the Sexual Risk Survey (SRS) with data from international populations (N = 81,060, 57% of which identified as cisgender women, 40% as cisgender men, and 3% as other gender identities) that were collected in the context of the International Sex Survey (ISS),3 a large-scale international survey conducted in 42 countries of residence (see Supplementary material for the list of countries of residence). Specifically, we investigated such evidence with (1) polytomous (discretized) data and (2) continuous (non-discretized) data from international populations. First, structural equation analyses of the Sexual Risk Survey (SRS) were performed with respect to its pre-established five-factor structure with polytomous (discretized) data. To fit the structural equation model, weighted least squares mean- and variance-adjusted robust estimation methods were employed. To assess the quality of adjustment to the data of the structural equation model, exact and approximate fit were examined. To examine exact fit, an exact fit hypothesis test was performed under the null hypothesis that the difference between the population covariance matrix and the model-implied covariance matrix is null. Adequate exact fit was determined by a fixed threshold value: a p ≥ 0.050. To examine approximate fit, four model-implied fit indices were employed: the comparative fit index (CFI), the Tucker-Lewis index (TLI), the root mean square error of approximation (RMSEA), and the standardized root mean square residual (SRMR). Adequate approximate fit was determined by fixed threshold values: a CFI ≥ 0.950, a TLI ≥ 0.950, an RMSEA ≤ 0.060, and an SRMR ≤ 0.080. Subsequently, internal structure validity and internal consistency reliability evidence were examined through (1) model-implied χ2 test statistics along with their corresponding degrees of freedom and probability values, (2) model-implied approximate fit indices (i.e., CFI, TLI, RMSEA, SRMR), (3) model-implied λ factor loading values, and (4) model-implied McDonald ω internal consistency values, which are as follows: χ2 (220) = 75,621.740 (p < 0.001), CFI = 0.796, TLI = 0.765, RMSEA = 0.142, SRMR = 0.080, λ ∈ [0.678, 0.953], ω ∈ [0.744, 0.907]. Considering exact and approximate fit (see Supplementary material for model-implied graph drawings), we cannot recommend employing the Sexual Risk Survey (SRS) with polytomous (discretized) data from international populations. Second, structural equation analyses of the Sexual Risk Survey (SRS) were performed with respect to its pre-established five-factor structure with continuous (non-discretized) data, strictly following the aforementioned structural equation analysis protocol, yet by employing maximum likelihood robust estimation methods to fit the structural equation model. Subsequently, internal structure validity and internal consistency reliability evidence are as follows: χ2 (220) = 1,466.281 (p < 0.001), CFI = 0.901, TLI = 0.886, RMSEA = 0.056, SRMR = 0.050, λ ∈ [0.368, 0.876], ω ∈ [0.548, 0.822]. Considering exact and approximate fit, alongside low model-implied λ factor loading and McDonald ω internal consistency values (see Supplementary material for model-implied graph drawings), we cannot recommend employing the Sexual Risk Survey (SRS) with continuous (non-discretized) data from international populations. In conclusion, the present data did not suggest that the validity and reliability evidence of the Sexual Risk Survey (SRS), which had only been evaluated in populations of college students in the United States of America, is applicable to international populations. Accordingly, inasmuch as sexual (risk-taking) behaviors are intrinsically associated with sociodemographic determinants, no such material can be presumed applicable to international populations.4 Nevertheless, we recommend that researchers interested in populations other than college students in the United States of America (1) examine the validity and reliability evidence of the material in their population(s) – notably by investigating alternatives to the pre-established internal structure of the material – or (2) employ continuous (non-discretized) data from the self-reported frequency of sexual risk-taking behaviors over the past six months (i.e., item scores) rather than arithmetic mean scores computed to reflect the level of endorsement of each of the five sexual risk-taking dimensions (i.e., factor scores). Full funding information is available in Supplementary material. Full disclosures are available in Supplementary material. Zsolt Demetrovics, PhD, Mónika Koós, PhD, Shane W. Kraus, PhD, Léna Nagy, PhD, Marc N. Potenza, MD, PhD, Rafael Ballester-Arnal, PhD, Dominik Batthyány, PhD, Sophie Bergeron, PhD, Peer Briken, MD, Julius Burkauskas, PhD, Georgina Cárdenas-López, PhD, Joana Carvalho, PhD, Jesús Castro-Calvo, PhD, Lijun Chen, PhD, Giacomo Ciocca, PhD, Ornella Corazza, PhD, Rita I. Csako, PhD, Marco de Tubino Scanavino, MD, David P. Fernandez, PhD, Elaine F. Fernandez, PhD, Hironobu Fujiwara, MD, PhD, Johannes Fuss, MD, Roman Gabrhelík, PhD, Ateret Gewirtz-Meydan, PhD, Biljana Gjoneska, MD, PhD, Mateusz Gola, PhD, Joshua B. Grubbs, PhD, Hashim T. Hashim, MD, Romain Icick, MD, PhD, Mohammad S. Islam, PhD, Martha C. Jiménez-Martínez, PhD, Tanja Jurin, PhD, Ondrej Kalina, PhD, Verena Klein, PhD, András Költő, PhD, Chih-Ting Lee, MD, Sang-Kyu Lee, MD, PhD, Karol Lewczuk, PhD, Chung-Ying Lin, PhD, Christine Lochner, PhD, Silvia López-Alvarado, PhD, Kateřina Lukavská, PhD, Percy Mayta-Tristán, PhD, Dan J. Miller, PhD, Olga Orosová, PhD, Gábor Orosz, PhD, Kyeongwoo Park, PhD, Fernando P. Ponce, PhD, Gonzalo R. Quintana, PhD, Gabriel C. Quintero-Garzola, PhD, Jano Ramos-Diaz, PhD, Kévin Rigaud, PhD, Ann Rousseau, PhD, PhD, Marion K. Schulmeyer, PhD, Pratap Sharan, MD, PhD, Mami Shibata, MD, Sheikh Shoib, MD, Vera L. Sigre-Leirós, PhD, Luke Sniewski, PhD, Ognen Spasovski, PhD, Vesta Steibliene, PhD, Dan J. Stein, PhD, Julian Strizek, PhD, Aleksandar Štulhofer, PhD, Norman Therribout, PhD, Banu C. Ünsal, PhD, Marie-Pier Vaillancourt-Morel, PhD, Marie C. van Hout, PhD, Cora von Hammerstein, PhD.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.