Change in the popularity of “transgressive” content in written erotica between 2000 and 2016

Abstract

There is a widely held belief that the amount and intensity of transgressive content in pornography has been rising. Reliably assessing for such an increase, however, is complicated by methodological factors including hand-coding content using conflicting a priori definitions of what constitutes transgressive content. In response to those limitations, the present study used the results of a published empirical content analysis of ~250,000 erotic stories written over 16 years to determine if the amount or popularity of transgressive content (stories high in the themes of violence, family (incest), or BDSM) has changed in that timeframe. Results from the present study indicated no meaningful increase in either the amount of content with those themes or popularity (as measured by story views per day) of any of the three transgressive themes within the erotic narratives over the 16-year period of analysis. These results, in addition to recent research presenting similar findings within pornographic video, do not support popular perceptions that erotic material is becoming increasingly transgressive. Rather, such content within internet-based erotic material, and particularly erotic narratives, appears to be relatively consistent.

Pornography, whether text, images, audio, or video, is in wide and general use, with at least 90% of men and 60% of women reporting that they viewed pornography in the past month (see, amongst others, Regnerus, Gordon, & Price, 2016; Sabina, Wolak, & Finkelhor, 2008; Solano, Eaton, & O’Leary, 2018). While there is significant debate about the effects that pornography has on its viewers (for recent examples, see Fisher, Kohut, Di Gioacchino, & Fedoroff, 2013; Price, Patterson, Regnerus, & Walley, 2016; Wright, Steffen, & Sun, 2019; Wright, Tokunaga, & Kraus, 2016), there is little disagreement about whether or not there is more pornographic content available in 2019 than there was in 2009 or 1999 (e.g., Price et al., 2016; Regnerus et al., 2016; Sabina et al., 2008; Solano et al., 2018). It is less clear whether the content of pornography has changed in that time frame, with some scholars arguing that newer pornographic content features more and more “painful” or “brutal” sexual encounters (e.g., DeKeseredy, 2015, p. 6; and see Picker & Sun, 2011 for an activist approach). Other work (Shor & Seida, 2019) has found no meaningful increase in transgressive content.

The scholarly debate about the content of pornography is part of a line of research that seeks to understand what kind of pornography has what effect on whom. For example, Boeringer (1994) found a positive correlation between “hard-core violent and rape pornography” (p. 289) and sexual coercion, and a negative correlation between viewership of soft-core pornography and actual rape behavior. Similarly, recent work has found that pornography has substantially stronger negative effects on individuals for whom pornography itself is inconsistent with their belief structure (Perry & Whitehead, 2019). Peter and Valkenburg (2016) found a connection between viewing violent or aggressive pornographic material and increased endorsement of gender stereotypes, while others (e.g., Malamuth, 2018; Malamuth, Addison, & Koss, 2000) have suggested that violent pornography content is associated with sexual violence (amongst other outcomes). In all cases, the direction of causality remains uncertain.

In this context, the question of whether the content of pornography has become more transgressive is a pressing one. Unfortunately, several methodological issues complicate the analysis of pornographic content, including the uncertainties of manual coding schemes, inconsistent operationalization of terms, and the various forms of pornographic media (e.g., videos, images, and text) themselves (see Short, Black, Smith, Wetterneck, & Wells, 2012, for a discussion).

Most research on pornography is conducted on videos and images. Though this is consistent with the finding that the majority of pornography accessed online features images and videos (SimilarWeb, 2019), images and videos must be hand-coded for content by human observers in order to be assessed by researchers (as in Bridges, Wosnitzer, Scharrer, Sun, & Liberman, 2010; Shor & Seida, 2019). This means that the same activity (e.g., spanking) could be coded very differently, depending on the a priori definitions used. Bridges et al. (2010), for example, operationalized violence in pornography as spanning from depictions of “torturing, mutilating, or attempting murder” to “biting…spanking [and] bondage or confining” (Bridges et al., 2010, p. 1072). Their coding schema explicitly included scenarios in which the recipient of the activity appeared to be enjoying it. Therefore, the category of ‘violence’ included apparently consensual spanking, an activity that about 30% of the adult population reports engaging in (Herbenick et al., 2017) and that may be associated with greater sexual satisfaction (Frederick, Lever, Gillespie, & Garcia, 2017).

Even when the same definition is used, different coding teams can produce very different results. Bridges et al. (2010) found that 88.2% of videos in their sample contained physical aggression. Using the same definition of aggression, but a notably wider sampling frame (i.e., Pornhub.com, RedTube.com, YouPorn.com, and xHamster.com versus 304 randomly selected top scenes provided by AVN.com), Klaassen and Peter (2015) found that about 37.2% of videos contained aggressive acts. Further, the category of violent content is sometimes broadened to include objectifying or degrading acts (Bridges et al., 2010; Wright et al., 2016), although those terms are also inconsistently operationalized. As Lischinsky (2018, p. 158) noted, “objectification cannot be observed directly on a text’s surface,” and thus this coding method inherently involves approaching the content with an existing interpretive frame.

An alternative to the assessment of videos and images is to analyze written erotica, whether such material is presented as erotica per se (e.g., literotica.com, 2019) or as adult fiction (e.g., Fifty Shades of Grey, James, 2012). While a number of studies have used written erotica as stimulus material for in-laboratory exposures (e.g., Harris, Thai, & Barlow, 2017), others have assessed erotic narratives with various text analytic techniques, including computational linguistic approaches, to analyze the nature of the content (Koller, 2015; Lischinsky, 2018; Ménard & Cabrera, 2011; Seehuus, Stanton, & Handy, 2019). Some of these text analytic approaches have the benefit of allowing content framing to be based on the objective presence or absence of specific word patterns (see, for example, Lischinsky, 2018), and some computational linguistic techniques allow for the empirical development of themes (see, for example, Seehuus et al., 2019), eliminating any aspect of human coding.

Though the analysis of text has the benefit of data-driven theme identification, there is a fairly large difference in scope between the online consumption of visual and written erotica. Although written erotica is quite popular, with Literotica.com (the most popular erotic story site) bringing in about 45 million monthly visits over the past six months (SimilarWeb, 2019) and romance novels bringing in approximately US$1.36 billion in book sales in 2009 (Romance Writers of America, 2011, as cited in Ménard & Cabrera, 2011), it is still nowhere near as popular as video and image-based erotica. Literotica.com is ranked 54^th most frequently-visited adult website, well after video and image-based websites such as Pornhub.com, which is ranked 2^nd in this category, with 2.9 billion estimated monthly visits (SimilarWeb, 2019).

In spite of the smaller scope of text-based erotica, the advantages made possible by the use of empirical content coding (e.g., Chung & Pennebaker, 2008) mean that an analysis of written erotic narratives offers a view into possible content shifts from a related, but distinct, angle. To that end, a large-scale investigation of changes in transgressive pornographic content over time is warranted. Thus, to determine the stability of the results presented in Shor and Seida (2019), the present study examined approximately 250,000 written pornographic stories published pseudonymously online on Literotica.com (literotica.com, 2019) over the course of 16 years. Using the Meaning Extraction Method (Boyd, 2014; Chung & Pennebaker, 2008), Seehuus and colleagues (2019) identified 20 of the most prominent themes within these stories (Meaning, Sensuality, Dirty Talk, Violence, Family (incest), Body, Sleep/wake, Colors, Domesticity, Clothing, Setting, Food/drink, Automotive, Size, BDSM, Genital words, Filler words, Playfulness, Oral, and Past tense). The authors of the present study selected three of those themes that reflect content that is relatively transgressive: themes related to violence, incest (the Family theme), and BDSM. There is no gold-standard definition of transgressive pornographic content, nor are there published guidelines that establish the parameters of porn that might be considered normative; further, it is likely that there are individual differences in conceptualizations of transgressiveness. In making this judgment, the authors of the present study considered the extent to which each theme raises questions of consent, aggression, or taboo; this justification is broadly consistent with the Held and Štulhofer’s (2015) factor analysis of the types of pornography found to be arousing, although that study did not address incest as a category. With Held and Štulhofer’s approach in mind, we chose to focus on the three themes that were, in our judgement, the most transgressive relative to the other themes identified by Seehuus et al. (2019).

Therefore, the present study investigated changes in transgressive content over time and assessed the relationship between transgressive content and viewer preferences. Consistent with Shor and Seida’s (2019) recent findings, we hypothesized the following:

• H₁: The mix of content produced during the years of analysis would be found to be relatively consistent, with no notable increase in transgressive content. That is, the theme scores for the themes of Violence, Family, and BDSM will not be meaningfully correlated with date of publication.

• H₂: The preference for that content (Violence, Family, and BDSM) will be unchanged over time, which will be reflected by the absence of a meaningful interaction between theme score and date of publication in predicting story views for those themes.

Our hypotheses are consistent with some recent research (e.g., Shor & Seida, 2019), but inconsistent with popular media narratives about pornography, which would alternatively suggest that transgressive theme scores would (a) increase over time and (b) be increasingly related to story views. Relying on the findings of recent empirical work, we predicted no such increase or preference.

Method

The data for this analysis were the same data that were used in Seehuus et al. (2019); data collection, theme extraction, and theme scoring for the present study were identical to that work. In brief, the data were derived from a corpus of text-based erotic stories (N = 296,884) that were downloaded between April and May of 2016 from Literotica.com (literotica.com, 2019), the most-visited website that pseudonymously publishes user-submitted erotic text (SimilarWeb, 2019). These stories were submitted to the website between 2000 and 2016 and were programmatically captured from the website over the course of several months. Stories that were not prose (e.g., poetry, or those stories that contained embedded images, n = 24,663), were not written in English (n = 13,636), or were fewer than 100 words long (n = 13,962) were excluded, leaving a corpus of 244,623 stories.

The stories were analyzed using the Meaning Extraction Method (MEM; Chung & Pennebaker, 2008), an approach that empirically identifies content themes in text, without requiring or allowing any pre-determined theme construction or definition. That is, the themes generated by the MEM derive solely from the text itself, and solely through an empirical process. This increases both the ecological validity of the analysis and the possibility of unanticipated findings. The Meaning Extraction Helper (Boyd, 2014), a purpose-built software package, was used to automate the first steps of the Meaning Extraction Method (for a more thorough description of the process, see Seehuus et al., 2019).

A principal components analysis generated a set of themes, or words commonly co-occurring in the individual stories. The themes were first independently named by each of the three authors; the names individually generated were very similar, and all authors agreed on the final theme names used. Theme scores were calculated by counting the number of times that each word in a theme occurred in a particular story, then summing that value for all words from a particular theme and dividing by the number of words in the story. That value was then multiplied by the sign of the factor loading for that word. Thus, theme score is the percent of the story that consists of words from a particular theme. Theme scores were multiplied by 100,000 to avoid leading zeros. As discussed earlier, the Violence, Family, and BDSM themes were selected as they were the themes that were thought to reflect transgressive content. See Table 1 for the words that comprise each of these three themes and their respective factor loadings (from Seehuus et al., 2019).

Table 1.

Themes of interest

Theme and Word (loading)
Violence
pain (0.43), rip (0.40), scream (0.40), tear (0.35), blood (0.34), bitch (0.30), slam (0.30), growl (0.28), cry (0.27), burn (0.24), shove (0.24), yell (0.24), bite (0.23), slap (0.22)
Family
mother (0.62), father (0.61), mom (0.61), dad (0.60), daughter (0.56), son (0.53), parent (0.39), family (0.38), sister (0.31), brother (0.29)
BDSM
command (0.46), master (0.42), serve (0.35), train (0.31), leather (0.24)

Analytic Plan

H₁: Theme scores will not change over time.

To test H₁, we estimated correlations between the theme scores of each of the three themes of interest (Violence, Family, and BDSM) and publication date; if the theme scores increased or decreased over time, this correlation would be meaningfully different than zero. The size of this sample (N = 244,623) rendered traditional null hypothesis significance testing (NHST) unhelpful, since correlations as small as r = .0040 are significant at p < .05, and we do not generally find correlations of that size to be meaningful. To test for the absence of change over time, we used equivalence testing (see Lakens, 2017, for a modern tutorial; see Schuirmann, 1987, for a more theoretical introduction), which uses, in this instance, two one-sided tests (TOST). These results answer two different questions: (a) is the parameter of interest statistically different than zero, which is to say, is there sufficient evidence to say the parameter is greater or less than zero; and (b) is the parameter of interest statistically equivalent to zero, which tests whether the parameter is within pre-specified bounds of meaningfulness.

Given the sample size, we anticipated that the answer to the question of whether the parameter was statistically significantly different than zero would almost certainly be yes, and that this result would not inform our interpretation of these results. Thus, we considered question (b) to be more relevant to H₁. If the TOST procedure found that the observed correlations were within (and thus closer to zero than) our bounds of meaningfulness, that would be evidence in support of H₁. That finding would suggest that there was no meaningful relationship between date of publication and each of the three themes of interest.

We set the equivalence bounds against which we tested the correlations at |r| ≥ .05, which is half of Cohen’s famous small correlation boundary (1988) and one quarter of Hemphill’s (2003) empirically-derived “lower third” boundary. These boundaries are inherently arbitrary. It is our judgement that a real and reliably measured effect of smaller than |r| ≥ .05 would not describe a relationship of empirical or theoretical interest in this content domain, and thus not meaningfully inform our understanding of the way that people produce and engage with erotic narratives. Although arbitrary, we believe that, as Cohen suggested, these boundaries are likely to “be found reasonable,” and that any effect smaller than our boundaries would be “so small that seeking them…is a bootless task” (1988, p. 13).

We used the TOSTER package (Lakens, 2017) in R (R Core Team, 2019) to test whether the absolute values of the each of the three correlations exceeded the equivalence boundaries (|r| ≥ .05) described above. If the correlation had exceeded that guideline, we would have further tested whether it fit in any of the classic rule of thumb categories (Cohen, 1988; Hemphill, 2003).

H₂: Date of publication will not interact with theme score to predict story views.

This hypothesis was tested by estimating the effect of the interaction between date of publication and theme content for each of the three themes of interest in a regression model predicting story views. The primary measure of interest was the change in R² associated with the addition of the interaction term to the regression model, as this describes the amount of variability in story views accounted for by the interaction between theme content and publication date. A meaningfully large ΔR² would suggest that, contrary to H₂, interest in a particular theme had increased (or decreased) with time. We selected a very conservative standard of ΔR² > .02 as a guideline for an interaction of sufficient size to be meaningful, recognizing again that any such standard is inherently arbitrary.

Results

As shown in Table 2, the correlations between theme scores and date of publication were generally small, although, per the discussion above about sample size, almost all were statistically significant. Positive correlations indicated that theme scores were higher for newer stories (i.e., that more recent stories contain more content associated with that theme).

Table 2.

Correlations amongst variables of interest

		1	2	3	4	5
1	Violence
2	Family	−.0230
3	BDSM	.1210	−.0630
4	Publication date	−.0170	.0230	.0110
5	Story views	−.0210	.2540	−.0570	−.2590
	M	239.70	213.72	61.85	3254.65	56682.69
	SD	243.73	442.00	176.39	1604.18	91343.63

Note that given the sample size (N = 244,623), significance is not a good indicator of meaningfulness, and is thus not marked. Any correlation with an absolute value of more than approximately .0040 is significant at p < .05, and any with an absolute value of more than approximately .0052 is significant at p < .01. Also note that more than the usual number of significant digits are reported, reflecting the sample size and the number of correlations very close to zero.

The Violence (r = −.017), Family (r = .023), and BDSM (r = .011) themes were found to correlate weakly with publication date, which suggests a very small decrease (i.e., Violence) or very small increase (i.e., Family and BDSM) in the amount of theme content over time. Equivalence tests were run for each of the three correlations with TOSTER (Lakens, 2017) for R (R Core Team, 2019), using the equivalence bounds of [−.05, .05] and an α of .05, as explained above. For all three correlations, both the null hypothesis test (p < .0001) and the equivalence test (p < .0001) were significant. This suggests that, in the language of Lakens, Scheel, and Isager (2018), although each correlation is statistically different from zero, it is statistically equivalent to zero. See Figure 1 for a graphical representation of the theme scores for Violence, Family, and BDSM by year. Given the negligible influence of time on the amount of Violence, Family, and BDSM content produced, these findings support H₁.

Figure 1.

The amount of content from the Violence, Family, and BDSM themes has not changed meaningfully over the period of analysis.

Regressions predicted story views by modeling the interaction of theme score and date of publication. See Table 3 for complete results of those models, noting again that significance is not a good measure of meaningfulness in a sample of this size. Whereas publication date predicted approximately 6.8% of story views, with newer stories having fewer views (which is likely at least in part because older stories have had more time to be viewed), the interaction between theme score and number of story views was generally not meaningful, with the largest ΔR² of .0149, which was associated with adding the interaction term for the Family theme. The interaction of Family and publication date predicted an additional 1.5% of the variability in story views. That means that over the 16-year period of analysis, Family content has slowly become less predictive of story views, although that effect was of minimal strength. See Figure 2 for an illustration of this interaction, and Figures 3 and 4 for illustrations of the two other, weaker interactions.

Table 3.

Regression models predicting story views with theme content, publication date, and the interaction of theme content and publication date.

Factor	B	SE B	p
Violence (Model R2 = .0676, ΔR2due to interaction < .0001)
Constant	104705.88	403.47
Violence	−2865.46	404.01	< .0001
Publication date	−14.75	0.11	< .0001
Violence x Publication date	0.17	0.11	0.14
Family (Model R2 = .1496, ΔR2due to interaction = .0149)
Constant	105916.36	385.32
Family	45198.73	368.75	< .0001
Publication date	−15.05	0.11	< .0001
Family x Publication date	−6.77	0.10	< .0001
BDSM (Model R2 = .0705, ΔR2due to interaction = .0006)
Constant	104468.03	402.78
BDSM	−9589.22	399.16	< .0001
Publication date	−14.69	0.11	< .0001
BDSM x Publication date	1.44	0.11	< .0001

Note that given the sample size (N = 244,623), the presence of significance is not a good indicator of meaningfulness, although a p value of more than .05 can be safely interpreted as lack of evidence of a relationship. The value of ΔR² associated with the interaction is the best indicator of the meaningfulness of the interaction. Theme scores were standardized, and publication date calculated such that a negative coefficient indicates a reduction in views over time

Figure 2.

The interaction between Family theme content and year of publication predicts a small amount of variability in story views, with Family content being less predictive of story views for more recent stories. The slope of each line represents the relationship between the amount of content in the Family theme and story views. R² = .1496; ΔR² due to interaction = .0149. Due to the sample size (N = 244,623), significance is not a good indicator of meaningfulness and is thus not shown.

Figure 3.

The interaction between Violence theme content and year of publication does not meaningfully predict story views. Each line represents stories from the year indicated, and the slope of the line represents the strength of the relationship between amount of theme content and story views for that year. R² = .0676; ΔR² due to interaction < .0001. Due to the sample size (N = 244,623), significance is not a good indicator of meaningfulness and is thus not shown.

Figure 4.

The interaction between BDSM theme content and year of publication does not meaningfully predict story views. Each line represents stories from the year indicated, and the slope of the line represents the strength of the relationship between amount of theme content and story views for that year. R² = .0705; ΔR² due to interaction = .0006. Due to the sample size (N = 244,623), significance is not a good indicator of meaningfulness and is thus not show

The very small ΔR² for the Violence (ΔR² < .0001) and BDSM (ΔR² = .0006) themes indicates that there is effectively no meaningful change in the relationship between the Violence and BDSM themes and story views across the period of analysis; see Figures 3 and 4. Since none of the ΔR² values exceed the .02 threshold, these findings support H₂; the preference for Family, Violent, or BDSM content has not meaningfully changed over time.

Discussion

This analysis used empirically-derived themes to assess changes in the amount and influence of transgressive content in 244,623 erotic stories posted online between 2000 and 2016. The results of this study suggest that the transgressive themes of Family (incest), Violence, and BDSM are neither increasing in the amount of content being produced within erotic literature, nor meaningfully changing the extent to which that content drives story popularity. The popularity of the Violence and BDSM themes is almost entirely flat over the period of analysis, regardless of whether popularity is measured by the sheer amount of content or by the extent to which the content drives story views. The Family theme has increased in popularity by a negligible (but measurable) amount over the period of analysis, but the extent to which Family content drives readership has, if anything, decreased by a very small amount.

Though the internet has increased the accessibility and availability of pornographic material, these results are consistent with Shor and Seida’s (2019) analysis, which similarly did not find any changes. Although neither that study nor the present study explored the entirety of pornography available online, the amount of transgressive content available on the most popular video- and text-based erotic websites does not appear to be increasing. These findings may help allay concerns that changes in the content of pornography may be a catalyst for increased sexual aggression among pornography viewers (e.g., DeKeseredy, 2015; Wright et al., 2016). Although researchers, community stakeholders, and other relevant groups may attribute decrements in sexual or mental health to a perceived increase in transgressive or violent pornographic content (e.g., Lim, Carrotte, & Hellard, 2016; Bridges, Wosnitzer, Scharrer, Sun, & Liberman, 2010) the results of this analysis do not support these perceptions for this sampling frame. Rather, consistent with recent work in this area and with the historical prevalence of incest content in literature, whether classical (out of many examples, see Ford, 1690/2017; Irving, 1981; Sophocles, 429BC/1977) or popular (Minahan, 2011; Pornhub, 2018), our findings suggest no such increase.

A primary concern some hold about violent or transgressive content within erotic material is that consuming such content may increase the perpetration of sexual violence (for a comprehensive discussion, see Shor & Seida, 2019). This is a valid concern. If this were the case, our data and the work of others would suggest that the rate of sexual violence should be relatively stable over time. However, a review of US governmental statistics on violence against women (both reported and unreported to law enforcement) shows a decline from a peak rate of 5.0 acts of sexual violence per 1,000 women in 1994 to 1.8 acts of sexual violence per 1,000 women in 2010 (Planty, Langton, Krebs, Berzofsky, & Smiley-McDonald, 2013). This decline, combined with data indicating an increase in overall viewership of pornography over approximately the same time period (Gorman et al., 2010; Klaassen & Peter, 2015; Price et al., 2016; Wright et al., 2016), suggests that, on the macro level, pornography viewership may have no effect (or even a protective effect, as per Fisher et al., 2013) on violence. Alternatively, any effect of violent pornography on sexual violence may have been countered by other societal forces. This kind of ecological data, however, must be interpreted with great caution, as there are a large number of factors that influence broad national trends (see Lim, Carrotte, & Hellard, 2016, for a discussion of these factors). For example, it is possible that individuals who choose to consume more violent sexual material may be drawn to sexual violence, prone to engage in sexual violence, or have beliefs that align more with violence against women. These possibilities are supported by the Confluence Model of sexual aggression (Malamuth, Linz, Heavey, Barnes, & Acker, 1995; Malamuth, Sockloskie, Koss, & Tanaka, 1991), which suggests that pornography consumption and an individual’s likelihood of engaging in sexual aggression are interdependent when predicting attitudes that support violence against women (Malamuth, Hald, & Koss, 2012). Furthermore, this model suggests that the consumption of violent or transgressive pornography alone is unlikely to lead to such outcomes. Less is known about the relationship between viewing videos or reading stories that are characterized by incest-related content and engaging in either consensual or nonconsensual sexual behavior with a family member. However, the Confluence Model similarly proposes that it is unlikely that the consumption of incest content or an increase in the prevalence of incest content (which was not demonstrated in this study) alone would lead to sexual activity with family members.

This study has important limitations. Although these findings are consistent with those of other studies that analyzed pornographic videos (Shor & Seida, 2019), there may be meaningful differences between individuals who view erotic videos or images and those who read erotic stories. Because we did not have access to any reliable demographic data on the writers and readers of the erotic narratives, we could not assess potential differences between these individuals and those who view erotic videos. Web analytics suggest that men are overrepresented (when compared to the general internet-using population) for both Pornhub.com and common erotic video websites, whereas women are much less underrepresented for story sites (Alexa Web Analytics, 2019). That is, there is some evidence that women are more likely than men to read erotica, although men are more likely to engage with pornography in general, regardless of the medium. In addition, the information collected and included in our models was based on total story views, not unique story views. Therefore, we were unable to distinguish between ten people reading a story one time and one person reading a story ten times (Seehuus et al., 2019). However, given that numerous stories were viewed over 50,000 times, we can reasonably conclude that many different people read these stories over time. Finally, it is important to acknowledge that transgressiveness is a nuanced topic that is worthy of its own empirical review. The three themes that we chose to analyze were transgressive relative to the other themes isolated by Seehuus and colleagues (2019); this does not mean that BSDM or consensual aggression during sexual activity is inherently transgressive. Similarly, there are a broad range of behaviors that may fall under the categories of BDSM or consensual sexual aggression, and some may be considered more transgressive than others.

Further, although Literotica.com is by far the most-visited website dedicated to erotic text content (literotica.com, 2019; SimilarWeb, 2019), it is not the only website dedicated to that content, and our findings do not address the existence or the potential emergence of websites either dedicated to more transgressive content or websites which have seen an increase in such content. Further studies should seek to evaluate the changes in erotic content available as the technology to perform those analyses develops.

The methodological approach implemented in this study also has important strengths. The use of a computational method to categorize and quantify the influence of story content facilitated the use of a much larger sample than would otherwise have been feasible, which both enabled the identification of relatively small effects and allowed for increased confidence in those estimates, particularly considering that the stories span 16 years. Further, by using empirically-determined themes, Seehuus et al. (2019) – and by extension, this analysis – could not have been swayed by preconceptions about what kinds of content was in the story, or what kinds of content might be connected to story views. This allowed for an unbiased generation of results.

Overall, these results add to the growing body of evidence that, while pornography may be expanding in reach, “there is no new thing under the sun” (Ecclesiastes 1:9, King James Version). Both the prevalence and popularity of transgressive content in written erotic narratives on the largest website dedicated to such content are largely unchanged in the past 16 years.