Previous evidence has found that when educating about the effects of sexism, presenting information in a traditional lecture-based format may result in rejection of information or a lack of motivation to change behaviors. Incorporating alternative learning methods may be more effective in reducing sexism or sexist beliefs among group members in a community or organization. Here, the authors investigate how experiential learning techniques which involve hands-on methods such as group games and discussions about sexism affect understanding of the impacts of everyday sexism.
Male and female participants were randomly assigned to participate in the WAGES intervention, receive information about sexism without an experiential component, or participate in a group activity with no educational component. The Perceived Harm of Everyday Sexism PHES evaluation was used to measure knowledge of the harms of gender inequity before the intervention, immediately after, and one week after the intervention. The WAGES intervention increased awareness of the harms of subtle everyday sexism as well as limiting the reactance and lack of self-efficacy seen in an information-only intervention that did not incorporate experiential learning.
Overall, the experiential learning-based WAGES intervention was successful in increasing awareness of the impacts of sexism when compared to traditional lecture-based methods.
The authors argue that education on sexism must be framed appropriately, involving experiential learning techniques to minimize reactance and maximize dialogue and feelings of empowerment. Specifically, WAGES may prove valuable in driving effective social change within communities and organizations. Analysis was limited to the individuals completing all phases of the study. During the initial baseline phase of the study, participants completed the Perceived Harm of Everyday Sexism PHES scale as part of a mass screening survey.
Items assessed the perception of how harmful it is for a woman to be encouraged to perform stereotypical activities and discouraged from non-stereotypical activities, and attitudes towards gender conformity and nonconformity. Participants were randomly assigned to participate in the WAGES intervention, receive information without an experiential component, or participate in a group activity with no educational component.
Participants in the WAGES intervention played a modified board game intended to illustrate the impacts of small gender-based advantages and disadvantages over time, followed by a group discussion. For group activity participants, a board game was played followed by a discussion about reducing intergroup conflict, with no information about gender. Table 2. One should note that, because of the way in which topic modeling was implemented in this work, topics would emerge with comparable sizes in terms of the number of documents assigned to them.
Hence these histograms might be biased and considering the fact that the original dataset has its own natural biases of self-reported sample, the frequency analysis cannot be used to draw any conclusions. Figure 1. In the next step, we consider the similarity between topics. This can be done in two ways: 1 by comparing how words are assigned to each pairs of topics and 2 by comparing how documents are assigned to each topic. The first approach is more suitable when we have smaller number of topics and hence larger overlap between the words assigned to each topic whereas the second approach can be used when there are more topics and each document is assigned to multiple topics at the same time.
We quantified these similarities by calculating the cosine similarity between the vectors of word weights and topic weights in the word-topic and topic-document matrixes. In these diagrams, each node circle represents a topic and the edges lines represent the strength of the similarity between each pairs of topics.
Figure 2. The weight of the connections between pairs of the topics is based on the similarity of how the words are assigned to them in the left panel and how topics are assigned in the right panel. We removed connections with cosine similarities smaller than 0. The color-code of the right panel is based on the communities that are detected using the Gephi implementation of the Leuven algorithm of community detection in networks.
In the case of 20 topics, we can also try to cluster topics into groups based on simple clustering algorithms in network science that group nodes of a network based on the strengths of their connections, i. The right panel of Figure 2 shows such grouping color-coded based on the Leuvain algorithm Blondel et al.
In order to shed further light on the topics and automated annotation of the posts, we also performed qualitative coding based on human judgment on a sample of posts.
First, we coded a sample of randomly selected posts into the 7 categories that are presented in Table 1 , by two independent coders. The intercoder agreement has been calculated using a set of measures and are reported in Table 3. Table 3. Intercoder agreement scores for a randomly selected sample of posts coded by two human coders.
This result shows a considerable agreement between independent coders that indicates the robustness of the extracted coding scheme using topic modeling. In the next step we coded a sample of posts and compared the results with the categories assigned to each post by the topic model algorithm. Here we see less agreement between the computational model and the human coding. Table 4 shows the intercoder agreement scores. Table 4. Intercoder agreement scores for a randomly selected sample of posts coded by a human coder and the topic model. In the process of coding, we observed that some posts contain multiple stories and experiences and that potentially is a problem for the computational coding, particularly when we force the algorithm to select only one category for each post.
Moreover, the context, and layered nature of the posts that are interpretable by human readers can be out of reach to the computational model. Another observation here is that the topic model works less accurately for shorter posts and where there are complicated references to concepts and use of abbreviations and specific jargons.
In order to understand the mismatch between the topic model assignments and the coding by humans, we considered the posts that are assigned to topics by the algorithm and human coder differently. Figure 3. A representation of the mismatch between topic model assignments and human coding. The edges between topics are weighted proportional to the number of posts that are co-assigned to the corresponding topics by the human coder and the algorithm.
The similarity between the networks shown in Figure 3 and the left panel of Figure 2 clarifies the relationship between human coding and computational coding. Where, the loadings of a post to topics are less localized on one topic and the topic model detects more than one significant topic in the post represented in the left panel of Figure 2 , there is a higher chance of a mismatch between the human coding and topic model assignment Figure 3.
This often happens when we have a post that accounts for multiple experiences or reports on multifaceted stories. Analysis of the Everyday Sexism data has hitherto largely been qualitative in nature, with themes and sites associated with experiences of sexism drawn out in Bates' book, Everyday Sexism Bates, and journalism.
- Aptitude Test Problems in Physics (Science for Everyone)!
- Gone (Alex Delaware, No. 20)?
- New York City NEW SHSAT Test Prep 2017, Specialized High School Admissions Test (Argo Brothers).
- Poetic Astronomy in the Ancient Near East: The Reflexes of Celestial Science in the Ancient Mesoptamian,Ugaritic and Israel Narrative.
- Clathrate Hydrates of Natural Gases, Third Edition (Chemical Industries Series).
This finding bears out the qualitative categorizations of the data set by Bates, and offers an important understanding of how topic modeling could be useful in processing and beginning to understand similar data sets that have not yet been analyzed. Although the Everyday Sexism accounts are submitted through the website, or via Twitter, the purpose of the site is to log everyday instances of sexism, both on and offline, and the majority of topics relate to offline experiences of sexism.
One of the main findings from this study is that experiences of sexism, even loosely grouped in the ways that we have described, are located everywhere. They are connected and they are pervasive. When we increase the number of topics to 20 Table 2 , this allows us to break down these experiences into separate but connected sites of sexism. For example, young women are clearly experiencing sexism in their learning environments, as evident in Topic S4 of our initial analysis but in the larger sample, we see sexism being experienced in both the school and University Topics L11 and L13 , areas connected by being associated with learning and with formative experiences of gender relations and expectations.
The patterns of sexism experienced in the classroom at school may well pave the way for similar behavior in the lecture hall or university classroom, with the majority of words in both topics overlapping. Subtle differences in the ways in which these educational, professional and leisure spaces operate can be exposed by this more finely tuned analysis. The home is a hugely influential space in which children begin to witness and absorb expectations around gendered roles and behavior. Topic L12 draws together a picture of sexism in the family, and points to the power of the home and familial relationships in encoding attitudes to sexism.
Analysis of the larger number of topics draws out numerous topics associated with what we may cluster together as street harassment, or Women in Public Spaces. Separating these topics out allows us to arrive at a more complex view of the reports that generate these clusters. Topics 10 and 17 suggest the frequency of accounts of women being verbally harassed, followed and threatened in the street while simply going about their daily lives.
It is also possible to extract themes in the data through relationships between topics exposed through our analysis, shown in Figure 2. Other connections are superficially less clear. While this may seem baffling at first glance, it ties in with Bates' observations of how sexism is reinforced in the home when victims of sexual harassment are subject to judgment and blame when reporting incidents to those close to them Bates, , pp. The groups are identified based on the strength of the connections between topics assessed through the overlap of documents co-assigned to them. This picture shows how various sub-topics are interconnected and the experience of sexism is not isolated in one shape or form.
However, the two topics, L5 and L7, appear unconnected to the other topics, with no ties either strong or weak. In fact, these topics appear to be quite general, remaining both distanced from and yet relevant to other topics. In Topic L7, the use or misuse of appropriate titles and forms of address are experienced as everyday sexism, which may emerge across a range of backdrops. What can topic modeling of the Everyday Sexism data set tell us about experiences of sexism?
The topic modeling approach delivers word bags containing highly distilled elements of commonly experienced sexist encounters, creating stark pictures of interrelated sites, languages and relationships in which sexism is enacted. This analysis suggests that sexism is fluid; it's not limited to a certain space, class, culture, or time. It takes different forms and shapes but these are connected. Sexism penetrates all aspects of our lives, it can be subtle and small, and it can be violent and traumatizing, but it is rarely an isolated experience. What does this method add to a qualitative analysis, and how can this sort of study be useful?
In summary, topic modeling provides an effective means of analyzing a large data set to produce high level as well as subtler and more finely drawn themes and commonalities. Using a data set like the Everyday Sexism reports, which have already been subject to extensive qualitative analysis, allows us to test this method against qualitative findings, producing consistent results.
One concern at the beginning of this project was that this method may seem reductive, producing the most common and therefore, it could be argued, most affecting or important experiences or sites of sexism. The topic modeling approach in fact offers a largely inclusive set of findings, highlighting distinct topics but visualizing connections between these topics, providing the opportunity to tease out connected but subtly different topics, which can then be contextualized by qualitative readings of the reports.
The results presented here are based on preliminary analysis, but in the future a more sophisticated approach both in the sense of methods of topic modeling and using larger and more representative datasets could potentially improve the results significantly. This could allow researchers to use computational methods to extract concepts and patterns that could then inform policy agendas. KE and TY designed the research. SM and TY performed the computational analysis. KE performed the qualitative analysis. The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
We would like to thank Laura Bates for useful comments and suggestions throughout the project. We have taken care to avoid a quantitative approach that might count and rank experiences of sexism from most common and therefore important to least. Barnett, R. Ageism and Sexism in the workplace.
Generations 29, 25— Google Scholar. Bastian, M. Gephi: an open source software for exploring and manipulating networks. Bates, L. Everyday Sexism [online]. Becker, J. Yet another dark side of chivalry: Benevolent sexism undermines and hostile sexism motivates collective action for social change. Blei, D. Latent dirichlet allocation.
Laura Bates: How Can We End Everyday Sexism?
Fast unfolding of communities in large networks. Bonilla, T. Elevated threat levels and decreased expectations: how democracy handles terrorist threats. Poetics 41, — Brandt, M. Sexism and gender inequality across 57 societies. Buchanan, N. Russo and H. Carstensen, T. Gender Trouble in Web 2. The Second Sex. Harmondsworth: Penguin Books. Eccles, J. Gender role stereotypes, expectancy effects, and parents' socialization of gender differences.
Everyday Sexism by Laura Bates
Femfuture Firestone, S. London: Cape. Foster, M. Tweeting about sexism: the well-being benefits of a social media collective action. Friedan, B. The Feminine Mystique. PubMed Abstract. Ghosh, D. Mapping tweets with topic modeling and Geographic Information System.
- France’s everyday sexism.
- Way Too Cool: Selling Out Race and Ethics?
- Xenotransplantation: Basic Research and Clinical Applications!
- The Complete TurtleTrader: How 23 Novice Investors Became Overnight Millionaires.
- Antineoplastic and Immunosuppressive Agents Part I!
- Spatial Reasoning and Planning: Geometry, Mechanism, and Motion;
Glick, P. The Ambivalent Sexism Inventory: differentiating hostile and benevolent sexism. Hall, D. Harper, A. Auburn University. Hartsock, N.