Online Sex Addiction After 50: an Exploratory Study of Age-Related Vulnerability

Abstract

The population of Internet users is ageing, yet online sex addiction research remains limited to younger age groups. Our study aimed to explore the association between online sex addiction and vulnerabilities related to older age, such as the absence of a partner, changes in work career, and boredom. Out of 2518 respondents who participated in an online survey, 158 (6.3%) were aged 50–77 and constituted the primary focus of the study. Linear regression analyses showed that occupational status, boredom (reasons for Internet use), and involvement in cybersex predicted online sex addiction, and that relationship status and offline sex had no effect on addictive behaviour. There was no evidence for a moderating effect from the occupational status on the relationship between boredom and online sex addiction. The results suggest that older age does not protect against the development of online sex addiction, and age-related vulnerability may amplify the risks.

Annex 1

Procedure

The psychometric properties of scores on the s-IAT-Sex were evaluated to exclude possible biases in studying excessive internet use for sexual purposes at an older age with a confirmatory factor analysis and measurement-invariance testing. All the testing procedures were conducted on imputed data. The imputation was carried out on the whole dataset (n = 2471) (for further details, see “Statistical Analyses” section). We used a confirmatory factor analysis and measurement invariance testing. While conducting a confirmatory factor analysis, the DWLS estimator, with the robust correction using a polychoric correlation matrix, was used due to the ordinal and skewed nature of the data (Li 2016), which was an issue for this measurement (with the majority of items being positively skewed with a skewness of > 1). Therefore, all reported fit indexes are based on the robust model fit statistics. Then, we carried out the measurement invariance testing across the age groups of people aged 18–49 (n = 2360) and persons aged 50 and more (n = 158). McDonald’s omega coefficient (McDonald 1999) was used to evaluate the reliability in terms of the internal consistency of the s-IAT-Sex. Compared to Cronbach’s alpha, omega does not assume tau-equivalence of the items’ factor loadings (Trizano-Hermosilla and Alvarado 2016), which has been neither met nor assumed in the current factor model. The omega coefficient should therefore produce more accurate estimates of the scale’s reliability.

Results

Analyses of the psychometric properties of the scores on the s-IAT-Sex showed that the two-factor model, proposed by Wéry et al. (2015) and Pawlikovski, Altstötter-Gleich, and Brand (2013), provided only a marginally better fit (CFI = 0.960, TLI 0.951, RMSEA = 0.90, 90% CFI_RMSEA (0.085;0.094), SRMR = 0.048, χ2 (53) = 1105.819) than the one-factor model (CFI = 0.955, TLI = 0.945, RMSEA = 0.95, 90% CFI_RMSEA (0.090;0.099), SRMR = 0.051, χ2 (54) = 1245.499). For both models, TLI and CFI indicate a good model fit (Hooper et al. 2008); RMSEA is an acceptable but mediocre model fit (MacCallum et al. 1996); and SRMR indicates a good model fit (Hooper et al. 2008). Moreover, the intercorrelation between the two factors was high (r > 0.9), indicating the redundancy of the two-factor solution in the current sample. The one-factor model was therefore used for the measurement-invariance testing across age groups. Internal consistency measured by McDonnal’s omega (1999) indicated the adequate consistency (ω = 0.91) of the one-factor model.

The scalar level of measurement invariance was achieved, indicating that the age groups of 50 and higher and below 50 do not differ significantly in terms of factor structure, factor loadings and item thresholds (Table 4). The Satorra-Bentler (2010) tests indicate a non-significant decrease in the model fit, as indicated by chi-square statistics, and CFI and RMSEA increased in the more constrained models. However, the results should be interpreted with caution because the fit indices might be inflated due to the groups’ size differences (Yoon and Lai 2018). The SRMR, the index that is not influenced by the chi-square statistic, decreased only slightly. To address the possibility of the homogenisation of the imputed data due to the disproportional sample size of older and younger subsamples (n_older = 149; n_younger = 2322), measurement-invariance testing was repeated with datasets where imputations were first carried out separately for both age groups, and imputed datasets were then merged for the purposes of measurement-invariance testing. The differences in the results were negligible (fit index differences were around 0.001). Only results from the whole-sample imputed datasets are thus reported.