Human Sexual Cycles are Driven by Culture and Match Collective Moods

To measure interest in sex, for each country, we retrieved the frequency by which people searched for the word “sex” using Google Trendstm (GT)10 (Methods 1–3); henceforth referred to as “sex-searches.” Interestingly, even in countries where English is not an official language, the English term “sex” is either more searched for than the corresponding word in the local languages or they are strongly correlated (Supplementary Table S1). Moreover, the terms most associated with searches for “sex” in GT refer to direct interest in sex and pornography (Supplementary Table S1). Therefore, GT searches for the term “sex” are a good proxy for interest in sexual behavior in the countries analyzed in this study.

Figure 1 depicts GT weekly sex-search data for 10 years from January 2004 to February 2014 for a set of Northern countries, which celebrate Christmas on December 25th. Yearly maximum peaks occur during Christmas week (red vertical lines), as previously observed for the USA11. While one may think that this increased interest in sex results simply from more free time during the holiday season, GT data is normalized by overall search volume10; even in a situation of increased general online activity, the increase in sexual interest is higher. Conversely, we could expect the holiday season to lead to a decrease in overall searches, led by school vacations for instance, originating an artificial peak for sex-related interest. However, we do not observe similar increases in weekly sex-searches for other widely observed holidays, such as Thanksgiving in the USA or Easter in France (Fig. S1A and B). Furthermore, a putative decrease in overall searches is unlikely, as a decrease in searches for school-related material can be compensated by a strong increase in searches for “presents” or “recipes”. In fact, when we control for search-volume of very common words, such as “on”, “and”, or “the”, there is some variation around the holiday period but it is in different directions for different search terms (Fig. S2A and B), probably resulting in an overall neutral change. Therefore, and although other dates lead to an increase in sex-searches (Fig. S1A and B), the Christmas holiday is uniquely associated with the highest peaks in sex-searches observed in these Northern countries. It is also known that, in Western Northern countries, conceptions peak around Christmas, in what some refer to as the “holiday effect”12. Indeed, the observed sex-search peaks match birth rate increases for this set of countries when shifted by nine months (Fig. S3A), which further confirms GT sex-searches as a good proxy for sexual activity.

Figure 1 Weekly queries for the term “sex” for a group of representative western Northern countries. The black line represents the averaged queries in a 10-year period, obtained from Google Trends, which is normalized by overall search volume. These countries are: Austria, Canada, Denmark, Finland, France, Germany, Italy, Lithuania, Malta, Netherlands, Poland, Portugal, Spain, Sweden and the United States of America. Shaded grey represents the standard deviation. The red vertical line marks Christmas week. Full size image

Compared to the observation of sex-search peaks in Northern countries that celebrate Christmas on December 25th (and corresponding increase in September birth rates where such data is available), the two hypotheses outlined above would predict quite distinct observations for other cultures and hemisphere locations. If the biological hypothesis is correct, all Northern countries should have similar sex-search peaks around the same time, and these peaks should occur in a counter-phase pattern (six months later) in all Southern countries—irrespective of culture. On the other hand, if the cultural hypothesis is true, these peaks should appear anywhere Christmas is celebrated—irrespective of hemisphere—and other similar celebrations in different cultures should lead to sex-search peaks in other times of the year.

To test these predictions, we extracted GT sex-search time-series data for all 129 countries for which GT offered consistent data. Countries were categorized according to hemisphere (North or South) and their predominant religion13,14. Countries where at least 50% of the population self-identifies as Christian were considered culturally Christian countries, and similarly for Muslim countries. Other countries, where neither of these religions is dominant, were grouped separately; Supplementary Table S2 shows the complete list of countries and categorization.

Both Northern and Southern countries show a prominent peak in sex-searches around Christmas and we observe no counter-phase pattern corresponding to the southern hemisphere winter solstice of June 21st (see Figs S4A and S5C,D). In fact, there is a strong significant correlation (R2 = 0.54, p-value < 0.001) between the mean sex-search time series of Northern and Southern countries (Supplementary Table S3). Since most Northern and Southern countries for which we have data identify as Christian (80 of 129), the observed correlation suggests that a cultural effect, rather than hemisphere location, drives the Christmas sex-search peak. Indeed, the birth data available for Christian, Southern countries peaks with Christmas sex-searches when shifted by nine months in much the same way as for Christian, Northern Countries, even though it is summer in the former and winter in the latter (Fig. S3). Furthermore, there is neither a sex-searches increase in December nor a birth peak in September for Northern countries that do not celebrate Christmas on December 25th (Fig. S7). As reliable birth data is not generally available, particularly for Southern and Muslim countries, and is only available for four Southern countries, all of them predominantly Christian, (Methods 6, Supplementary Table S9 and Figs S3 and S6), we use GT sex-search data instead to observe many more countries and address the two hypotheses.

Parsing all countries by religion (Fig. 2A and B, Fig. S4 and Supplementary Table S3), it is clear that the mean sex-search time-series are periodic but uncorrelated between Christian and Muslim countries (R2 = 0.19, p-value < 0.001). The difference in sex-search behavior between these two sets of countries is further revealed in Fig. 2C and D, where we averaged the sex-search yearly time-series across all ten years centered on Christmas week (for Christian countries) or centered on Eid-al-Fitr, the major family holiday, that ends Ramadan (for Muslim countries). In Christian countries, the only clear peak occurs during the Christmas week. In contrast, in Muslim countries there is a peak during the week of Eid-al-Fitr and a second peak during the week of Eid-al-Adha, the other major religious and family celebration in Muslim culture; also noteworthy is a steep decrease during Ramadan, consistent with that period of general abstinence (as further discussed below). Both of these groups of countries clearly show sex-search peaks associated with distinct cultural celebrations, rather than with hemisphere. Indeed, it is worth noting that the Muslim calendar does not follow the solar calendar: every year Ramadan shifts by 10 days relative to its date during the previous Gregorian calendar year. Nevertheless, sex-searches peak during the moving week of Eid-al-Fitr (and Eid-al-Adha) in Muslim countries. The moving sex-search peaks associated with major religious events in Muslim countries further emphasizes the cultural driver behind such collective behavior.

Figure 2 Weekly queries for the term “sex” in culturally different countries. (A) Normalized and averaged queries for all available countries identified as Christian (dark red line). (B) Normalized and averaged queries for all available countries identified as Muslim (dark green line). (C) Searches in all Christian countries centered around Christmas week21. (D) Searches in all Muslim countries centered around Eid-al-Fitr week24. See Supplementary Table 2 for country identification and availability on GT. The vertical red lines mark Christmas week, the shaded light green area represents Ramadan, with the darker green lines marking Eid-al-Fitr (solid) and Eid-al-Adha (dashed). Shaded areas around the lines in C and D show the standard deviation. Full size image

To resolve the incompatible predictions of the biological and cultural hypotheses we made country-specific comparisons between hemisphere and culture, beyond the group-average behavior described above. We averaged the yearly sex-search time-series for each of the 129 individual countries across all years in four different ways: centered on Christmas week (fixed relative to the solar calendar), centered on Eid-al-Fitr week (moving relative to the solar calendar), and centered on each of the solstices, fixed on June 21st and December 21st (Methods 4, Supplementary Tables S4–6 and Fig. S5). We then measured the response of countries to a holiday as the sex-search z-score deviation above the mean at Christmas, Eid-al-Fitr and the two solstice weeks (Methods 5 and Supplementary Table S7). Figure 3 shows a world map with color-coded countries: shades of red indicate countries whose highest sex-search deviation from mean occurs during the Christmas week, and shades of green indicate countries whose highest sex-search deviation from mean occurs during Eid-al-Fitr week (Methods 7). It is clear that this response yields a map organized according to culture rather than hemisphere.

Figure 3 World-wide sex-search profiles. The world map is color-coded according to the z-score of each individual country’s sex-search time-series. Shades of red represent a higher z-score (larger increase in searches) during Christmas week (on Christmas-centered data). Shades of green represent a higher z-score (larger increase in searches) during Eid-al-Fitr week (on Eid-al-Fitr centered data). Light grey denotes countries with no significant variation above mean in either of these weeks. Dark grey countries are those for which there is no GT data available. Black line represents the equator separating the hemispheres. Built using: https://mapchart.net/. Full size image

We then compared this new country classification (according to the individual countries’ sex-search profile, Supplementary Table S7 and Supplementary Methods S1) with our previous identification based only on the proportion of the population that self-identified as Christian, Muslim or Other (Supplementary Table S2)13,14. Out of the 30 countries originally identified as Muslim14, 77% show a significant increase (z > 1) in sex-searches during the week of Eid-al-Fitr, and out of the 80 countries originally identified as Christian13, 80% show a significant increase (z > 1) during the Christmas week, regardless of the hemisphere. It is important to note that this correspondence is even higher (91%) when we identify as “Other” the ten Christian countries that do not celebrate Christmas on December 25th. In fact, we do not see an increase in sex searches around December 25th in any of these Northern Russian and Serbian Orthodox Christian countries, which celebrate Christmas in early January, and this further supports the cultural hypothesis (Methods 2, Supplementary Methods S1, Supplementary Figure S7). Moreover, only 14% of Southern countries showed a significant increase in sex-searches during the June solstice (Supplementary Tables S7 and S8B), demonstrating that there is no significant counter-phase sex-search peak in the southern hemisphere, contradicting the biological hypothesis.

The Christmas and Eid-al-Fitr holidays carry significant cultural and religious meaning, but they are not directly associated with sex. It is, in fact, very counter-intuitive to think of Christmas and Eid as the times of the year with the most online searches for sex. However, these events may trigger specific and collective moods, leading to a striking correspondence between these holidays and sexual interest. To investigate the emotional factors involved we measured changes in public sentiment on Twitter15,16,17. The analysis was performed before, during, and after Christmas and Eid-al-Fitr in a set of seven countries with sufficient Twitter traffic in our data: Australia, Argentina, Brazil, Chile, Indonesia, Turkey, and the USA (Methods 9 and Fig. S8). Although it is not possible to know whether the Google and Twitter populations are the same per country, given the large volume of Google searches and tweets, it is very likely that they provide a significant sample of the same populations.

Twitter sentiment was quantified by rating a random 10% sample of all tweets posted between September 2010 to February 2014 using the Affective Norms for English Words (ANEW) lexicon18 (Methods 8 and 9). The ANEW lexicon consists of 1,034 English words that carry a sentiment score along three dimensions: Arousal (a), Dominance (d), and Valence (v), corresponding respectively to whether the word makes human raters feel calm vs. excited, controlled vs. in-control, and sad vs. happy. The sentiment value of a single tweet is defined as the mean ANEW score of its words. We translated the lexicon to Spanish and Portuguese to capture public sentiment in those languages as well, but did not have the ability to translate into additional languages. To avoid bias from holiday-related language, we ignored all words used in traditional greetings for all known holidays in the World (Supplementary Table S13); we also removed the word “Christmas” and “valentine” from the lexicon, which does not include other holiday names.

We first observed that the weekly volume of sex-searches significantly correlates with the mean weekly sentiment derived from the three ANEW dimensions in a multiple linear regression (Supplementary Methods S2, Supplementary Table S10). In every country, valence yields a positive coefficient, while dominance a negative coefficient; thus the happier but less in-control the population mood is, the more sex-searches tend to increase in every country (Methods 10 and Supplementary Methods S2). Interestingly, while public sentiment displays a strong linear relationship with sex-search volume when all mood dimensions are considered, there is little correlation with each ANEW dimension on its own (Supplementary Table S11).However, the observed linear correlation does not allow us to characterize the population mood in the target cultural celebrations. To investigate if days that are similar in mood to Christmas in Christian Countries or to Eid-al-Fitr in Muslim Countries also tend to observe increased volume of sex-searches, we need a more nuanced characterization of the mood profile each week.

Because collective mood sentiment, as measured here, is derived from many tweets of large and diverse populations, it can contain distinct and informative components. Thus, we employed an eigenvector-based analysis19 to characterize the distribution of sentiment values, rather than just average sentiment. We thus obtain the components of public sentiment that explain most of the variance in the data not attributable to regular language use, hereafter referred to as “eigenmoods.” Specifically, an eigenmood is a small set of components (eigenvectors) of a matrix. In this matrix, the rows denote sentiment scores in a given range or bin, and the columns denote the weeks (Methods 11 and Supplementary Methods S3), and elements are the number of tweets during a week that fall in that bin. Thus, an eigenmood is not an average sentiment value (per week in our analysis), but rather a change in the distribution of sentiment that explains a significant proportion of the variation in the time-series data20.

We found that two components were sufficient to describe public sentiment associated with each holiday and country – a characterization that is independent of sex-search volume, and relies only on measurement of sentiment on Twitter (Methods 10–12, Supplementary Methods S3–5, and Supplementary Figs S10 and S11). Figure 4 (Column A), Figs S9 and S14 show the sentiment distribution of the selected eigenmoods that best characterize the holidays of interest, per every week of the year; redder (greener) colors represent increased (decreased) numbers of tweets falling in the respective mood dimension bins – e.g., for valence, upper bins on vertical axis denote increased happiness and lower bins denote increased sadness. The sentiment distributions of rows 1, 2, and 3 in Fig. 4 column A are centered on Christmas for USA (Northern, Christian) and Brazil (Southern, Christian), and Eid-al-Fitr for Indonesia (Southern, Muslim). While the eigenmood that describes Christmas in the USA uses only the valence dimension of ANEW, the best eigenmood for Christmas in Brazil requires valence and arousal, and for Eid-al-Fitr in Indonesia requires valence and dominance. The sentiment distribution of these eigenmoods per week clearly shows that significant and unique changes in sentiment occur during the target holidays. In all these cases, the public mood of the holiday in question generally shifts to “happy” bins (more red in higher valence) and away from “sad” bins (more green in lower valence). In Brazil, the mood also shifts to more “calm” bins during Christmas week (more red in lower arousal), and in Indonesia it also shifts to neither “in-control” nor “controlled” bins during the Eid al-Fitr week (more red in mid dominance). More detailed characterization of eigenmoods and their selection for each country is provided in Supplementary Material (Supplementary Methods 3–5 Fig. S12 and 13).

Figure 4 Mood distributions and their correlations with sex-searches. Rows: 1 - USA centered on Christmas, 2 - Brazil centered on Christmas, 3 - Indonesia centered on Eid al-Fitr. Columns: (A) Heatmaps of sentiment distribution reconstructed from selected eigenmoods. Vertical axis specifies the bins of the ANEW distribution for a given mood dimension, from low (bottom) to high (top) values. Eigenmood components were selected to best characterize the respective holiday and country (after removing the first component). In the case of the USA (Row 1), the two selected components both fall in the “valence” dimension and are labelled valence1 and valence2; for Brazil (Row 2) and Indonesia (Row 3) the first component also falls in the “valence” mood dimension, but the second falls in the “arousal” and “dominance” dimensions, respectively. Horizontal axis specifies the week of the centered, averaged year (52 weeks for the Gregorian calendar, 50 for the Muslim Calendar). The dotted line in the center marks the holiday of interest, on week 26 for Christmas, or week 25 for Eid al-Fitr. Color represents the weight of the eigenmood per bin per week. (B) Projections of weeks into the space formed by the selected eigenmood components. Each axis specifies the projection of week onto each component that defines the eigenmood. See text for details and supplemental materials for more information on component selection. (C) Linear regressions between GT sex search volume (vertical-axis) and similarity to holiday center in the Twitter eigenmood space depicted in column B (horizontal-axis) for averaged weeks. The weeks of Ramadan are shown with increasing color intensity from more yellow to more green as they approach Eid-al-Fitr. The R2 values for the regressions are 0.380 for Christmas in the USA, 0.504 for Christmas in Brazil, and 0.407 (0.637 without the Ramadan weeks) for Eid-al-Fitr in Indonesia. Full size image

Figure 4, column B, shows all weeks in the data projected onto the selected eigenmood space of two components for each country. It is clear that in this space Christmas weeks (red diamonds) cluster together for the USA and Brazil, and Eid-al-Fitr weeks (green circles) cluster together for Indonesia, demonstrating that the eigenmoods are consistent in different years for each holiday in each country. Figure 4 column C depicts the linear regression between sex-search volume as calculated before (vertical axis), and mood similarity to the target holiday in the respective eigenmood space (horizontal axis) for all weeks in the data set denoted by black circles in the plot (Methods 14 and Supplementary Methods S6). We observe a significant correlation for all countries studied, with R2 ≥ 0.38 for Christmas in all Christian Countries and R2 ≥ 0.34 for Eid-al-Fitr in all Muslim Countries, irrespective of hemisphere (Supplementary Table S12). Thus, in Christian countries we can say that the more the public mood of any week resembles the Christmas eigenmood, the larger the volume of observed sex-searches tends to be. Similarly, in Muslim Countries the more public mood is similar to the Eid-Al-Fitr eigenmood, the larger is the volume of sex-searches. In the case of both Muslim Countries studied (Indonesia and Turkey), there is a striking result pertaining to Ramadan: those 4 weeks (4 lowest green triangles in Fig. 4C, bottom right, for Indonesia), have the lowest sex-search volume by far in the data, consistent with the period of abstinence that marks Ramadan (see also Fig. 2B,D). The public mood during these weeks of Ramadan is also quite distinct from the Eid-al-Fitr mood (horizontal axis in Fig. 4C, bottom right), but, becomes more similar the closer the week is to Eid-al-Fitr in time; and as the mood becomes closer to the Eid-al-Fitr mood as Ramadan approaches its end, the sex-search volume also increases. Naturally, due to the low, outlier sex-search volume during Ramadan weeks, the linear regression is much stronger if those weeks are removed, with R2 ≥ 0.64 (Supplementary Table S12).

Thus, not only there are specific moods associated with Christmas and Eid-al-Fitr, the eigenmoods that best characterize these holidays significantly correlate with increased interest in sex throughout the calendar. This is true in all countries studied, in both hemispheres and cultures. Moreover, and although these moods occur at different times in different cultures, they seem to be similar in essence and, in general, the “happier” the mood, the more it associates with sex interest.