Gender Bias in Sharenting: Both Men and Women Mention Sons More Often Than Daughters on Social Media
GGender Bias in Sharenting: Both Men and Women Mention Sons More Often ThanDaughters on Social Media
Elizaveta Sivak and Ivan Smirnov ∗ Institute of Education; National Research University HigherSchool of Economics, Myasnitskaya ul., 20, Moscow 101000, Russia
Gender inequality starts before birth. Parents tend to prefer boys over girls, which is manifestedin reproductive behavior, marital life, and parents’ pastimes and investments in their children.While social media and sharing information about children (so-called “sharenting”) have becomean integral part of parenthood, it is not well-known if and how gender preference shapes onlinebehavior of users. In this paper, we investigate public mentions of daughters and sons on socialmedia. We use data from a popular social networking site on public posts from 635,665 users.We find that both men and women mention sons more often than daughters in their posts. Wealso find that posts featuring sons get more “likes” on average. Our results indicate that girls areunderrepresented in parents’ digital narratives about their children. This gender imbalance maysend a message that girls are less important than boys, or that they deserve less attention, thusreinforcing gender inequality.
Keywords: gender inequality, son preference, parenthood, sharenting, social media
Gender inequality starts even before birth. Across theworld, would-be parents tend to prefer their first (or theonly) child to be a boy rather than a girl or to havemore sons than daughters [1–8]. This results in millionsof “missing girls” at birth due to sex-selective abortions[9–11]. Gender preference continues to manifest through-out childhood. In some countries couples pursue sons byhaving additional children at the cost of larger family sizeand underinvestment in daughters [12, 13]. Sons haveadvantages in nutrition [14], vaccination rates [15] andspending on health care [16, 17]. Fathers [18–20] (and insome cases both parents [21]) spend more time with sonsthan with daughters. Fathers more often marry and staymarried in families with sons [22, 23], although evidencefor this is mixed [24, 25]. At this point the reader willnot be surprised that parents also report more happinessin families with sons [26].Despite the extensive literature on gender preference,no study to date has examined whether the use of so-cial media by parents is gender biased. As social mediabecomes an integral part of parents’ life, it might be im-portant to understand if and how gender preference ismanifested in this environment. One common practicethat has recently become a widespread trend is “shar-enting” [27, 28], or parents’ habitual use of social mediato communicate a lot of detailed information about theirchildren . In this paper, we investigate gender preferencein sharenting drawing on data about 62 millions publicposts on a popular social networking site.We obtained data from VK , a Russian analogue ofFacebook and the largest social networking site in Eu-rope. VK provides an application programming interface ∗ [email protected] Sharenting, as cited in Collins Dictionary http://vk.com Maria Ermolaeva 10 Apr at 11.56Konstantin Leonidov 4 Feb at 18.14Olga Stepanova 13 July at 20.0610th of April... And only my cheerful little son was delighted with the snowMade a snowman with my daughter! The first medal! You are the best, son Michail Ivanov 29 Apr at 08.05Daughter! 3500 g, 51 cm!!!
Figure 1: Selected examples of posts with mentions of chil-dren. All names and dates have been changed. (API) that allows the systematic download of publiclyavailable information. We used the VK API to collectpublic posts that were made in 2016 by 635,665 usersfrom Saint Petersburg (fourth largest European city),aged 18-50 years (see SI Text for details on our sampleand data collection). We then identified posts with men-tions of children by looking at posts that contained thewords “daughter” and “son” along with their differentforms, e.g. “dochenka” (daughterling) or “soooooooon”(see Methods for details). Common topics for such postsincluded celebrations of different achievements and im-portant events (e.g. births and birthdays or starting andfinishing school), expression of love, affection, and pride,and reports on spending time with the children (see Fig.1 for illustrative examples and SI Text for more informa-tion about common topics).We computed the proportion of female and male usersfrom each cohort who mentioned sons or daughters in a r X i v : . [ c s . S I] M a r their posts at least once along with the average numberof mentions of children for these users. In our analysis,we used various definitions for “mentions of children” toensure that the results were not influenced by a specificchoice of words (see Methods). We also collected infor-mation about the number of “likes” that posts featuringchildren obtained, on average. We used these data to in-vestigate whether the social network environment mightreinforce gender bias by rewarding posts featuring chil-dren of one gender more than those of another. Results
Fig. 2 shows the proportion of users who mentionedchildren in their public posts at least once in 2016. Theproportion of women increases sharply until 31-32 yearsold and then gradually falls. The peak matches the aver-age age of women at first childbirth, which is 30 years inSaint Petersburg [29]. The proportion of men who men-tion children is significantly lower and steadily increaseswith age.In almost all cohorts of users, sons are mentioned bya larger proportion of both men and women. This dif-ference cannot be explained by the sex ratio at birthalone (1.06 in Russia) and thus indicates gender prefer-ence in sharing information about children. Those userswho mention children at least once also write slightlymore posts about sons. There are 2.3 posts about sonsper woman and 2.1 posts about daughters per woman( p = 0 . p = 0 . Discussion
Studies of gender preference in parental practices usu-ally have to rely on self-reports, e.g. reports about timespent with children [18–21]. Self-report studies havesome benefits, but their results are affected by various bi-ases including social desirability bias or recall bias. Men-
20 25 30 35 40 45 50
Age P r o p o r t i o n o f u s e r s Women: sonWomen: daughterMen: sonMen: daughter
Figure 2: The proportion of users who mentioned children intheir public posts at least once in 2016. Sons are mentionedby a larger proportion of both men and women. Vertical barsare standard errors.Table I: The average number of likes per post. All daugh-ters/sons differences are statistically significant with p < − . Written by women Average number of likesby women by menfeaturing daughters 6.7 1.1featuring sons 10.7 1.8Written by men Average number of likesby women by menfeaturing daughters 5.3 2.6featuring sons 6.7 3.7 tions in posts are directly observable and present a clearand simple metric, which can be used on easily accessi-ble data to measure parents’ gender bias. We used thismetric on a large dataset of public posts of more than sixhundred thousand users and found that both men andwomen exhibited son preference on the social networkingsite: sons were mentioned significantly more often thandaughters. This result is remarkably stable, and holdstrue across age cohorts, different measures, and sets ofwords. We also found that writing posts in which sonsare mentioned is more rewarded: these posts get around1.5 times more likes than stories featuring daughters.Son preference in traditional societies and developingcountries is a well-known phenomenon. Our results con-firm that son preference is also prevalent in countries notimmediately associated with gender disparity .Gender preference in “sharenting” may seem quiteharmless in comparison with such layers of gender dis- Russia is above average in the ranking of countries by genderparity [30]. parity as sex-selective abortions or underinvestment ingirls. However, son bias online may affect girls as theymay feel underappreciated and less visible. It may alsohave broader effects on gender parity. Even moderatebias might accumulate given the widespread popularityof social media. Son preference in “likes” can additionallyamplify the bias, acting as social media’s built-in positivefeedback loop. Millions of users are exposed to a genderbiased news feed on an everyday basis and, without evennoticing, get the reaffirmation that it is normal to paymore attention to sons.Previous studies have shown that children’s books aredominated by male central characters [31, 32]. In text-books, females get fewer lines of text, fewer named char-acters, and fewer mentions than men [33]. Additionally,in movies there are on average twice as many male char-acters as female ones in front of the camera [34]. Whilefemale coverage on Wikipedia compares favorably withsome other lists of notable people [35], there are still 4times more articles about men than women [36]. Genderimbalance in public posts may send yet another messagethat girls are less important and interesting than boysand deserve less attention, thus presenting an invisibleobstacle to gender equality.
MethodsCounting mentions of children
We used the API of VK to download all public posts ofusers from Saint Petersburg that were made in 2016. Wethen computed vector representations of Russian words by training a fastText [37] model on the collected corpus.We used this model to identify words similar to “son” and“daughter”, namely the closest words in the vector spacemeasured by cosine distance. We manually excluded un-related words. For instance, both the words “son” and“granddaughter” are unsurprisingly semantically close tothe word “daughter” according to the model. However,these are not synonyms to the word “daughter” and wedo not treat them as mentions of daughters. After exclu-sion of unrelated words we obtained a list of the 30 clos-est synonyms to the word “daughter” and the 30 closestsynonyms to the word “son”. The posts that included atleast one of these words were considered as posts men-tioning children. The use of word embeddings trainedon the VK corpus allowed us to take into account wordsor their forms that cannot be found in dictionaries butwhich are used by the users of the social network, e.g.“sooon” instead of “son”. We performed an additionalanalysis to make sure that our results were not drivenby a particular choice of words (see SI Text). We alsoremoved potentially fake accounts and filtered posts thatwere not made by users themselves (see SI Text for de-tails on data preprocessing) and then computed the pro-portion of users who mentioned children at least once intheir posts, and the average number of such mentions peruser.
Acknowledgements
Support from the Basic Research Program of the Na-tional Research University Higher School of Economicsis gratefully acknowledged. [1] Xu Tian, Xiaohua Yu, and Stephan Klasen. Gender dis-crimination in China revisited: a perspective from fam-ily welfare.
Journal of Chinese Economic and BusinessStudies , 16(1):95–115, 2018.[2] Karsten Hank and Hans-Peter Kohler. Gender prefer-ences for children in Europe: Empirical results from 17FFS countries.
Demographic research , 2(1), 2000.[3] Pauline Rossi and L´ea Rouanet. Gender preferences inAfrica: A comparative analysis of fertility choices.
WorldDevelopment , 72:326–345, 2015.[4] Anthony Abeykoon. Sex preference in South Asia:Sri Lanka an outlier.
Asia-Pacific Population Journal ,10(3):5–16, 1995.[5] John Bongaarts. The implementation of preferences formale offspring.
Population and Development Review ,39(2):185–208, 2013.[6] Edgar Dahl, Manfred Beutel, Burkhart Brosig, andKlaus-Dieter Hinsch. Preconception sex selection for non-medical reasons: a representative survey from Germany.
Human Reproduction , 18(10):2231–2234, 2003.[7] Edgar Dahl, Ruchi S Gupta, Manfred Beutel, YveStoebel-Richter, Burkhard Brosig, Hans-Rudolf Tin-neberg, and Tarun Jain. Preconception sex selection de- mand and preferences in the United States.
Fertility andsterility , 85(2):468–473, 2006.[8] Sara Raley and Suzanne Bianchi. Sons, daughters, andfamily processes: Does gender of children matter?
An-nual Review of Sociology , 32:401–421, 2006.[9] G´eraldine Duth´e, France Mesl´e, Jacques Vallin, IrinaBadurashvili, and Karine Kuyumjyan. High sex ratios atbirth in the Caucasus: Modern technology to satisfy olddesires.
Population and Development Review , 38(3):487–501, 2012.[10] Barbara D Miller. Female-selective abortion in Asia: Pat-terns, policies, and debates.
American anthropologist ,103(4):1083–1095, 2001.[11]
World Development Report 2012: Gender equality anddevelopment . The World Bank, 2012.[12] Batool Zaidi and S Philip Morgan. In the pursuit of sons:Additional births or sex-selective abortion in Pakistan?
Population and development review , 42(4):693–710, 2016.[13] Onur Altindag. Son preference, fertility decline, and thenonmissing girls of Turkey.
Demography , 53(2):541–566,2016.[14] Silvia Helena Barcellos, Leandro S Carvalho, and Adri-ana Lleras-Muney. Child gender and parental invest- ments in India: Are boys and girls treated differ-ently?
American Economic Journal: Applied Economics ,6(1):157–89, 2014.[15] Vani K Borooah. Gender bias among children in Indiain their diet and immunisation against disease.
Socialscience & medicine , 58(9):1719–1731, 2004.[16] Lina Song. In search of gender bias in household resourceallocation in rural China.
IZA Discussion papers , (3464),2008.[17] Bela Ganatra and Siddhivinayak Hirve. Male bias inhealth care utilization for under-fives in a rural com-munity in western India.
Bulletin of the World HealthOrganization , 72(1):101, 1994.[18] Michael Baker and Kevin Milligan. Boy-girl differencesin parental time investments: Evidence from three coun-tries.
Journal of Human Capital , 10(4):399–441, 2016.[19] Kathleen Mullan Harris and S Philip Morgan. Fa-thers, sons, and daughters: Differential paternal involve-ment in parenting.
Journal of Marriage and the Family ,53(3):531–544, 1991.[20] Joan Aldous, Gail M Mulligan, and Thoroddur Bjar-nason. Fathering over time: What makes the differ-ence?
Journal of Marriage and the Family , 60(4):809–820, 1998.[21] Shelly J Lundberg. The division of labor by new parents:does child gender matter?
IZA Discussion papers , 2005.[22] Gordon B Dahl and Enrico Moretti. The demand forsons.
The Review of Economic Studies , 75(4):1085–1120,2008.[23] Francine D Blau, Lawrence M Kahn, Peter Brummund,Jason Cook, and Miriam Larson-Koester. Is there stillson preference in the United States? Technical report,National Bureau of Economic Research, 2017.[24] Shelly Lundberg, Sara McLanahan, and Elaina Rose.Child gender and father involvement in fragile families.
Demography , 44(1):79–92, 2007.[25] Andreas Diekmann and Kurt Schmidheiny. Do parents ofgirls have a higher risk of divorce? An eighteen-countrystudy.
Journal of Marriage and Family , 66(3):651–660,2004.[26] Hans-Peter Kohler, Jere R Behrman, and Axel Skytthe.Partner + children = happiness? The effects of partner-ships and fertility on well-being.
Population and devel-opment review , 31(3):407–445, 2005.[27] Alicia Blum-Ross and Sonia Livingstone. “Sharenting,”parent blogging, and the boundaries of the digital self.
Popular Communication , 15(2):110–125, 2017.[28] Anna Brosch. When the child is born into the Internet:Sharenting as a growing trend among parents on Face-book. 43(1):225–235, 2016.[29] Interfax. The highest average age of women at firstchildbirth in Russia was recorded in Saint Petersburg. , 2017. [Ac-cessed 21.03.2018].[30]
Global Gender Gap Report 2016 . The World EconomicForum, 2016.[31] Janice McCabe, Emily Fairchild, Liz Grauerholz, Ber-nice A Pescosolido, and Daniel Tope. Gender intwentieth-century children’s books: Patterns of dispar-ity in titles and central characters.
Gender & society ,25(2):197–226, 2011.[32] Mykol C Hamilton, David Anderson, Michelle Broad-dus, and Kate Young. Gender stereotyping and under-representation of female characters in 200 popular chil- dren’s picture books: A twenty-first century update.
SexRoles , 55(11-12):757–765, 2006.[33] Rae Lesser Blumberg. The invisible obstacle to educa-tional equality: Gender bias in textbooks.
Prospects ,38(3):345–361, 2008.[34] Stacy L Smith, Marc Choueiti, Katherine Pieper, TraciGillig, Carmen Lee, and Dylan DeLuca. Inequality in700 popular films: Examining portrayals of gender, race& LGBT status from 2007 to 2014.
USC Annenberg ,2015.[35] Claudia Wagner, David Garcia, Mohsen Jadidi, andMarkus Strohmaier. It’s a man’s Wikipedia? Assessinggender inequality in an online encyclopedia. In
ICWSM ,pages 454–463, 2015.[36] Swedish Ministry for Foreign Affairs. Wikigap. https://meta.wikimedia.org/wiki/WikiGap , 2017. [Accessed21.03.2018].[37] Piotr Bojanowski, Edouard Grave, Armand Joulin, andTomas Mikolov. Enriching word vectors with subwordinformation. arXiv preprint arXiv:1607.04606 , 2016.
Supplementary InformationSample and data preprocessing
VK is the largest European social networking site, withmore than 100 million active users. It was launchedin September 2006 in Russia and provides functional-ity similar to Facebook. According to VK’s Terms ofService: “Publishing any content on his / her own per-sonal page, including personal information, the User un-derstands and accepts that this information may be avail-able to other Internet users taking into account the ar-chitecture and functionality of the Site”. VK providesan application programming interface (API) that enablesdownloading of information systematically from the site.In particular, it is possible to download user profiles fromparticular educational institutions and within selectedage ranges. For each user, it is possible to obtain a listof their public posts. VK’s support team confirmed to usthat the data downloaded via their API may be used forresearch purposes.One limitation of VK is that its API returns no morethan 1000 users for any request. To collect data on usersfrom Saint Petersburg on a larger scale we created a listof all high schools in Saint Petersburg and then accessedIDs of users from each age cohort (from 18 to 50 years)who graduated from each of these schools. As each ofthese requests returns less than 1000 users, we were ableto collect information about all users who indicated theirhigh school on VK. Note that not all users in the sam-ple currently live in Saint Petersburg, and not everyoneon VK indicated their (former) high school in their pro-file. However, this approach allowed us to collect a largesample of VK users in a systematic way. Another advan-tage of our approach is that it provides an opportunityto effectively remove fake profiles. To achieve this wedid not include in the final sample the users who hadno friends on the social networking site from their highschool. The data were collected as part of the DigitalTrace project and the data collection procedure was ap-proved by the Institutional Review Board of the NationalResearch University Higher School of Economics.We made sure that only posts with authentic contentwere included in the analysis. We excluded reposts andposts with exactly the same content made by multipleusers. We also did not include posts containing URLs toaccounts for potential automatic posting and advertise-ments by websites or VK applications (e.g. invitationsto visit a web site or to beat the score in a game). Notall the posts in the resulting sample are necessarily aboutusers’ own children. Some posts include mentions of chil-dren of friends or relatives, or jokes, etc. By our estimate,the proportion of such posts is around 9%, and their re-moval does not affect the observed son bias (see SI Topicanalysis). P r o p o r t i o n o f u s e r s Women: sonWomen: daughterMen: sonMen: daughter a) Dictionary size N u m b e r o f p o s t s Women: sonWomen: daughterMen: sonMen: daughter b) Figure S1: Proportion of users who mention children at leastonce (a) and the total number of posts with mentions of chil-dren (b), as a function of the number of words included in theanalysis.
Dictionary analysis
The exact estimate of son bias might depend on theselection of words that are counted as mentions of chil-dren. However, we found that the observed preferencefor sons holds true irrespective of the choice of words.To show this, we selected the N most frequent synonymsand forms of the words “daughter” and “son” from ourcorpus. We then used these sets of words to computethe proportion of users who mentioned children at leastonce (Fig. S1a), as well as the total number of posts withmentions of children (Fig. S1b). The son bias holds truefor all N = 1 , ...,
40. Any changes for larger N are negli-gible. The list of the most frequent words along with thenumber of occurrences of each word is presented in TableS1.
Topic analysis
We identified the main topics of posts with mentionsof children by analyzing a sample of posts from one agecohort (30 years old). We coded all the men’s posts
Figure S2: Examples of posts featuring children. Authors ofthe posts provided consent to the use of their posts in thispublication. All posts were originally written in Russian andwere translated verbatim. (879 posts) from this cohort and randomly selected 20%(1521) of the women’s posts. At the first stage of coding, for each post we wrote down the category which mostfully grasped the post’s content. At the second stage wecollapsed similar categories into broader ones.Only 9% of the posts are not related to the users’own children (8% among women’s posts and 18% amongmen’s). These posts include mentions of other people’skids as well as jokes, news, and stories about pets. Af-ter filtering out the irrelevant posts, the son bias forwomen was unchanged, and for men it remained signif-icant: women wrote 15% more posts about sons thanabout daughters and men wrote 34% more posts aboutsons in this age group.Among relevant posts, the most common topics werereports of spending time with children (27% of posts),expressions of positive feelings, mostly love, affection,or pride (26%), and celebrations of births and birthdays(19%; see Fig. S2 for examples). These three categoriesaccounted for 72% percent of all posts about user’s ownchildren. Note that the distribution of topics most likelydepends on the age of a child, and might be different forother cohorts.