Measuring Female Representation and Impact in Films over Time
MMeasuring Female Representation and Impact in Films over Time
Luoying Yang , Zhou Xu , Jiebo Luo University of Rochesterluoying [email protected] , [email protected] , [email protected] Abstract
Women have always been underrepresented in moviesand not until recently has the representation of womenin movies improved. To investigate the improvementof female representation and its relationship with amovie’s success, we propose a new measure, the fe-male cast ratio, and compare it to the commonly usedBechdel test result. We employ generalized linear re-gression with L penalty and a Random Forest modelto identify the predictors that influence female repre-sentation, and evaluate the relationship between femalerepresentation and a movie’s success in three aspects:revenue/budget ratio, rating, and popularity. Three im-portant findings in our study have highlighted the diffi-culties women in the film industry face both upstreamand downstream. First, female filmmakers, especiallyfemale screenplay writers, are instrumental for moviesto have better female representation, but the percent-age of female filmmakers has been very low. Second,movies that have the potential to tell insightful storiesabout women are often provided with lower budgets,and this usually causes the films to in turn receive morecriticism. Finally, the demand for better female rep-resentation from moviegoers has also not been strongenough to compel the film industry to change, as moviesthat have poor female representation can still be verypopular and successful in the box office. Introduction
Film is a common entertainment form that fulfills the au-dience’s desire to make emotional connections with char-acters and learn about their social world. However, it wasvery rare for women to see inspiring counterparts on the bigscreen in the past. Many key roles in film-making, such as di-rectors and cinematographers, were for many decades dom-inated almost entirely by men (Lauzen 2012), and womendid not have enough power to make demands in the film in-dustry. Consequently, women have been constantly under-represented in movies. Even when they are present, womenare often portrayed in circumscribed and subordinated waysin traditionally feminine (i.e., stereotyped) roles, such asnonprofessionals, homemakers, wives or parents, and sex-ual gatekeepers (Collins 2011). Lacking a role model on the big screen is detrimental for young girls. They are dis-couraged from pursuing ambitions and participating activelyin social affairs (Haraldsson and W¨angnerud 2019). There-fore, female under-representation is a critical issue that mustbe addressed.In more recent times, women have made inroads into var-ious fields and films have started to respond to female view-ers with strong and well-rounded female characters (Mur-phy 2015; Heldman, Frankel, and Holmes ). There are manystudies and projects devoted to studying evolving feminismin films, centering on both the upstream effects , in whichcontent is structured through the actions of major filmmak-ers in gendered organizations who presume the public’s pref-erences, as well as the downstream effects , where audiencesrespond to content and attitudes are formed and reinforced.In our work of examining gender representation in films, weask the following two questions: What factors in the film-making process have a significant impact on female repre-sentation in films, and do films that feature more female rep-resentation outperform those that do not commercially?To answer these questions, we need to define a propermeasure for female representation in films. The Bechdel testis a popular measure for examining how well-rounded andcomplete the representations of women in media are. Thetest asks only three questions:• Are there at least two women in the film who have names?• Do those women talk to each other?• Do they talk to each other about something other than aman?The test is simple and has been widely used in many filmgender studies to measure female representation(Lindnerand Schulting ; Lindner, Lindquist, and Arnold 2015). How-ever, one major criticism of the test is that it fails to re-veal the hidden gender imbalance structure(Micic 2015). Amovie can pass the Bechdel test yet still portray women asauxiliary characters with minimal screen time. For example,both Wonder Woman and The Martian passed the test eventhough the importance of female characters is very differentbetween the two movies. Therefore, we propose a new mea-sure of female representation in films: the percentage of fe-male cast members in the whole cast. This proposal is basedon the intuition that if more women are cast for a film, thenwomen will have more representation on the big screen. a r X i v : . [ c s . C Y ] O c t n addition to a new measure of female representation, wewould also like to consider different aspects of movie suc-cess. Box office return is a popular measure of movie successbecause it is directly linked to profit. However, movie ratingsand popularity are also important measures of a movie’s suc-cess. Public acclamation and popularity brings confidence tothe studios and producers to inspire similar films or sequelseven when the box office return is not substantially high.Our analyses are completed in two steps. Drawing on asample of the most widely distributed films, we first com-bine a content analysis using the Bechdel Test with film-making data such as the budget and gender of crews to ex-amine what factors are influential on encouraging more fe-male representation during the film-making process. Next,including all film characteristics such as female represen-tation, budget, genre, and many others, we further examinewhether better female representation can increase the chanceof success for films while adjusting for other possible con-founding variables. In doing so, we contribute to sociologi-cal theories about the reciprocal impact of feminism.The remainder of the paper is organized as follows. In sec-tion 2, we review the background literature that studies thegender gap in films and analyzes box office performance. Insection 3, we describe the data features and the data collec-tion process. In section 4, we present the methods and mod-els used throughout this study. We present our experimentsand the corresponding results in Section 4. In section 5, wepresent and discuss our conclusions and future directions. Related work
Most of the studies that center on gender inequality in me-dia are content analysis (Murphy 2015; Heldman, Frankel,and Holmes ; Micic 2015), which means they study howwomen are portrayed in media. While content analysis is be-yond the scope of our study, the related works have shownthat the roles of women are evolving on the big screen. Thereare also many studies about female underrepresentation, al-beit on a small scale and with simple analysis. Thomas hasshown that in the top 10 worldwide highest grossing films of2016, women speak only 27% of the lines (Thomas 2017).The finding is novel but the sample size is too small. Ateam at the Rhodes Information Initiative at Duke Univer-sity published a report that movies which pass the BechdelTest have a statistically significant higher return on invest-ment compared to films that do not (Berkman, Garland,and VanSteinberg 2017). The tech company Shift72 pub-lished a similar report which states that all films that passthe Bechdel Test surpassed the box office returns of filmsthat fail this test . However, these reports only presentedstatistics of mere comparisons between movies that pass thetest and movies that do not, which are too simple to con-clude that “more women in the film means more success(movies are more profitable).” Such analyses failed to ad-just for confounding sources such as the genres. Linder et al.have shown that after adjusting for confounding factors suchas production budgets, movies with more female representa-tion (passing the Bechdel test) tend to have smaller produc- https://shift7.com/media-research/ tion budgets and consequently earn less money. However,given the same budget, female representation does not boostthe box office return (Lindner, Lindquist, and Arnold 2015).They have conducted a similar analysis for movie critics us-ing the same set of movies and obtained a similar conclusionthat female representation has little effect on critics (Lindnerand Schulting ). The analysis in these papers is the closestto ours. However, this study only drew samples of 974 filmsfrom the 2000-2009 decade and considered only a few con-founding variables such as the budget and genres. Our anal-ysis include a larger sample from a longer time span witha larger set of confounding variables to provide deeper in-sights over time. Data
Three data sources are used in this project: the BechdelTest Movie List , the Internet Movie Database (IMDb) ,and the Movie Database (TMDb) . The Bechdel Test MovieList contains over 8000 movies crowd-sourced through theInternet with a flag showing if each of them passes theBechdel Test or not. IMDb has an up-to-date movie dataset,which is widely used among many movie related projects.However, since IMDb does not provide an official API, wehave limited access to some data, such as the full list of thecast and crew members of each movie. TMDb on the otherhand is a crowd-sourced online movie database similar toIMDb. It is used less often, but with the advantage of havingan API, we can easily access a variety of movie-related dataof interest to us. Data Acquisition
The Bechdel Test Movie List is downloaded using its API.It contains 8190 movies at the time, with 7 fields. The rele-vant features include imdbid (can be used to join with otherdatasets), title, year, and rating (0 means no two women, 1means no talking, 2 means talking about a man, 3 means itpasses the test).The IMDb dataset is downloaded from its website, whereonly the title.basic table is used. It contains 9 fields, includ-ing tconst (imdbid), titleType, primaryTitle, originalTitle,isAdult, startYear, endYear, runtimeMinutes, and genres.Getting the TMDb dataset requires more work with itsAPI. The first step is to obtain the tmdbids that are associatedwith the imdbids within the Bechdel Movie List. Then theAPIs are called iteratively to access the detail and credits ofeach movie. By rearranging the tables, we are finally ableto get our desired features including: budget, revenue, voteaverage, vote count, popularity (a comprehensive measurecalculated from release data, number of votes, number ofviews, etc. from TMDb), production companies, number ofcast members, and gender of the cast and crew members.
Data Preprocessing
As mentioned from previous sections, we have three datasources: Bechdel Test Movie List, IMDB, and TMDB. The https://https://bechdeltest.com https://developers.themoviedb.org revious two are well established as tabular forms and re-quire minimal processing.For the TMDb data, new features are created to better rep-resent the movie measurables of interest. First, the revenue-budget ratio is calculated through simple division to repre-sent the profitability of each movie. There are some excep-tions: a few non-commercial movies have budgets of 0, sowe removed these rare cases from our dataset. Some movieshave a long list of production companies, but only those atthe top of the list provide major contributions, so we excisethe list and set an upper limit of 5 production companies foreach movie. In addition, the movie credits features (number,gender, and title of cast and crew members) are aggregatedto extract new features such as female cast ratio, number ofcore crew members (directors, producers and screenplays),and the female core crew ratios.Finally, the Bechdel Movie List, IMDb, and TMDbdatasets are joined according to the imdbid. The categoricalvariables are then one-hot encoded and the numerical vari-ables are min-max normalized in preparation for modeling.The final table contains 4232 observations and 67 variables. Methodology
Our primary goal is to evaluate what factors have an ef-fect on female representation in films and the impact of fe-male representation on films’ success. Given this purpose,the primary method of our analysis is an explanatory modelwhich focuses on modeling the true generation of sampledata rather than making the best prediction. Many believethat models with high explanatory power are inherently ofhigh predictive power as well. Conflation between expla-nation and prediction is common, however the type of un-certainty associated with explanation is of a different na-ture than that associated with prediction ( ? ). Therefore, weadd a prediction model as the secondary task to comparethe predictive performance with our primary model. For ourprimary model, we consider Generalized Linear Regressionwith a penalty term for model selection. Generalized linearregression is one of the most commonly used statistical mod-els for explanatory modeling, and the addition of the penaltyterm puts constraints on regression coefficient estimation tokeep only the relevant variables from the model. It is idealfor our goal of explanatory modeling with variable selection.For our secondary model, we consider Random Forest. Ran-dom Forest is known for being robust in predictive model-ing with feature ranking by importance. We expect RandomForest to have very satisfactory performance as a predictivemodel for our data and to use its performance to comparewith the predictive performance of our primary model. Inorder to compare the coefficient estimates of different vari-ables from the regression model, variables are normalizedby min-max normalization to [0 , so that they are on thesame scale, and the same dataset is fitted to both the regres-sion model and Random Forest model. After obtaining thevariable selection and predictors that rank results from thetwo methods, we compare the set of selected variables andthe top ranked predictors to see if they are consistent. Forboth methods, we split the samples by a 7:3 ratio into the training and test sets and compare the prediction accuracy tosee if they have similar performance. Generalized Linear Regression with L Penalty
Generalized linear regression is a broad class of modelsthat includes regular regression for continuous response aswell as models for discrete responses. The relationship be-tween a dependent variable and one or more independentvariables can be explained by the magnitude and sign ofthe regression coefficient estimates. Depending on the typesof response variables, we include the regular linear regres-sion for continuous responses and logistic regression for bi-nary responses in our analysis. In addition, we also includea L regularization term to shrink the coefficient estimatestoward 0 so that only highly correlated variables remain inthe model (Tibshirani 1996). A tuning parameter controlsthe degree of shrinkage such that varying the tuning pa-rameter value results in different estimates. We employ theBayesian Information Criterion (BIC) to choose the optimalmodel, which focuses on finding the true model that gener-ates the data. If using an error-based or accuracy-based cri-terion such as cross-validation for variable selection, whencorrelated variables enter the model, any one of the corre-lated variables being selected can give a good predictiveperformance; however BIC selects the variables that yieldthe best likelihood for the model and removes the rest, thusserving our purpose of explanatory modeling perfectly. Random Forest
The Random Forest is one of the most effective methodsfor predictive analysis as either a classification algorithm orregression model (Ho 1995). By selecting the nodes ran-domly and aggregating the trees for pooled results, RandomForest is known to be robust to noise and outliers, as wellas the value changes of parameters on a fine scale, includ-ing the number of trees aggregated and the number of nodesbeing selected at each node split. Therefore, we do not man-ually vary the parameter values on a fine scale in order toachieve the optimal results. The model selection, which isthe ranking of predictors by their changes in purity, is doneinternally through computing the Out-of-bag (OOB) errorfrom aggregating the trees.
Experiments
In this section, we present the basic features of femalerepresentation (i.e. the Bechdel test result and female castratio), movie success (i.e. the revenue/budget ratio, movierating and movie popularity), and our modeling results oftheir associations.
Preliminary
We examine the two most important measures of femalerepresentation in movies, the Bechdel test result and the fe-male cast ratio. About 55.01% of the movies in our datasetpass the Bechdel test. As Fig.1 shows, in the 21st century,the movies that pass the test have outnumbered the onesthat do not. The average female cast ratio is 0.26, and Fig.1shows that the movies in latter time have higher femaleast ratios compared to the former time. Using a simpleunivariate linear regression, we conclude that as time pro-gresses, the number of movies that pass the Bechdel test in-creases (coefficient estimate = 1.61, p-value < . ) andmore females are cast in the movies (coefficient estimate =0.11, p-value < < < < < < < < < < The Improvement in Women Representation
We first evaluate the factors that have strong associationswith the Bechdel test result. Setting the Bechdel test resultas the response variable of a logistic regression with penaltyand a Random Forest, we include the variables of year, adultcontent, run minutes, budget, number of cast members, fe-male director ratio, female producer ratio, female screen-play writer ratio, 26 production company indicators, and 25genre indicators as the predictors. The same set of predic-tors is also fitted to the models with the female cast ratio be-ing the response variable. The variable selection results areshown in Table 1. For both response variables, the regres-sion model and the Random Forest model have very similarpredictive performance. The female screenplay writer ratiois selected in both regression models to have positive as-sociation with the Bechdel test result (coefficient estimate =0.79) and female cast ratio (coefficient estimate = 0.042) hasvery high rankings in both Random Forest models (Bechdeltest=6, female cast ratio=5), implying that women participa-tion in story-writing is key to determining how women arerepresented on screen, outperforming the female director ra-tio and female producer ratio. These two variables are alsoselected with relatively smaller magnitude and decent rank-ing in the Random Forest model, due to the fact that thesethree variables are correlated and their effects on the re-sponse variables are adjusted for the presence of the femalescreenplay writer ratio. Year is also a very strong predictorof female representation, which is consistent with Fig.1 inthat as time progresses women become better represented.For the female cast ratio, the number of cast members isthe most influential factor (coefficient estimate=-0.19, rank-ing=1) as a higher number of cast members leads to a lowerfemale cast ratio. Such an association implies that male ac-tors are more likely to be hired than females. Many genreswere identified to have better female representation, such asRomance, selected by both regression models (Bechdel test:0.18, female cast ratio: 0.042) with decent rankings in bothRandom Forest models (Bechdel test: 14, female cast ratio:6), and Horror, selected by the Bechdel test regression (co-igure 1: Women representation in movies over time. (a) The distribution of the Bechdel test results over the last 12 decades;(b) The distribution of the female cast ratio over the last 12 decades.efficient estimate = 0.20) with decent rankings in both Ran-dom Forest models (Bechdel test: 11, female cast ratio: 14).Here, Horror was paid special attention due to its associ-ation with movie success in future analysis. Some genres,like Action and Crime were identified to have less femalerepresentation. Although significant correlations of budgetwith the female cast ratio and Bechdel test results are foundin the preliminary statistical analysis, budget is not selectedin either of the regression models. However, this fact can beexplained by the gender preference of genres. From Fig.3,we can see that Action and Crime movies, which are iden-tified in our models to be negatively associated with femalerepresentation, are likely to receive a much higher budgetthan movies that are not associated with these genres. Mean-while, they also tend to feature more male cast members thanfemale cast members. Romance and Horror movies (surpris-ingly) feature more females than males on the average. How-ever, they also tend to receive a lower budget. The imbalancebetween the budgets of genres and gender preference of gen-res leads to the imbalance of budgets on movies with differ-ent levels of female representation.
Women Representation and Movie Success
Setting the movie revenue/budget ratio (R/B ratio), rating,and popularity each as the response variable of an individualregression model with penalty and a Random Forest model,we include the variables of year, adult content, run time min-utes, number of cast members, female cast ratio, female corecrew ratio, the Bechdel test, 26 production company indica-tors, and 25 genre indicators as the predictors.The female cast ratio is selected to have strong positive as-sociation with R/B ratio (coefficient estimate= . × − )and ranks as the most important predictor in the RandomForest model (ranking=1). Meanwhile, three similar genreshave been identified to have high positive association withthe R/B ratio: Horror (coefficient estimate= . × − , ranking=7), Mystery (coefficient estimate= . × − ,ranking=5), and Thriller (coefficient estimate= . × − ,ranking=6). As we have acquired the knowledge from pre-vious analysis that the female cast ratio is positively as-sociated with the Horror genre, we are confident to con-clude that with the presence of Horror and the other twocorrelated genres in the model, the female cast ratio has apositive effect on a movie’s profitability after consideringthe effects of genres. This same set of variables is also se-lected for movie rating, however their associations with rat-ing are in the opposite direction in contrast to the R/B ratio.The female cast ratio has a negative association with rat-ing (coefficient estimate= − . × − ) and a very highranking in the Random Forest model (rank=4). Horror (co-efficient estimate= − . , ranking=6) and Mystery (coeffi-cient estimate= − . , ranking=16) are identified to be neg-atively associated with rating. It seems that the movies fea-turing more women, horror, and mystery elements are likelyto make more money, but receive more criticism. Neitherof the female representation measures, the female cast ratioand Bechdel test result, are selected for movie popularity.However, the female cast ratio still receives a high rankingin the Random Forest model (ranking=4), possibly due to itscorrelation with other variables.Due to the fact that the revenue/budget ratio is calculateddirectly from budget, and that in the previous analysis wehave learned that budget does not have a direct impact onfemale representation, we exclude budget from our modelsince its strong correlation with the response variable wouldoverwhelm the effects of other variables. However, our mod-els still demonstrate the relationship between budget andmovie success. The run-time in minutes and the number ofcast members are identified to have very strong associationwith all three aspects of movie success and receive very highranking in the Random Forest models, and they are a proxyof a movie’s budget as a higher budget leads to a larger num-echdel test Female cast ratioVariable Regression Random Forest Regression Random ForestTest accuracy 64.9% 63.1% Test MSE . × − . × − Female screenplay ratio 0.79 6 0.042 5Female director ratio 0.15 21 1.25 × − < < . × − , ranking=6) and Adventure (coeffi-cient estimate= . × − , ranking=5), as well as one pro-duction company Disney (coefficient estimate= . × − ,ranking=8). Meanwhile, it also identifies one genre beingnegatively associated with popularity, Drama (coefficientestimate= − . × − , ranking=10). Fig.4 shows that Ac-tion, Adventure, and Sci-Fi movies, which tend to be verypopular, also tend to feature fewer female cast members;Drama movies, on the other hand, feature slightly more fe-male cast members but also tend to be less popular than othergenres. Disney, known for creating strong and well-roundedfemale characters, in fact still features more male cast mem-bers (or male voices in animation) than female cast mem-bers in their productions. Therefore, the high popularity of Disney movies does not help associate better female repre-sentation with high movie popularity. Conclusions
Overall, our findings regarding female representation in-dicate that female representation is critically influenced bythe work of female crew members, especially female screen-play writers, in the film-making process. In addition, it isalso evolving throughout time as a result of other factorsoutside movies and thus not included in the dataset (i.e. thevariable year). Our findings also reflect some difficulties ac-tresses face, for example, male actors are more likely to behired than females, genres that tend to receive higher bud-gets also prefer male actors to portray the stories, and so on.From our investigation into the relationship between femalerepresentation and movies’ success, we discover that com-pared to the Bechdel test result, the proposed female castratio is more directly linked to a movie’s success as it is of-ten selected multiple times. In comparison, the Bechdel testresult is not selected once. Moreover, movies featuring morewomen tend to have a lower budget which leads to a higherrevenue/budget return. However, they also suffer from morecriticism, possibly due to the low budget invested. Consider-ing that the number of cast members and run-time minutesare both selected to have a very high positive impact on amovie’s rating and that they are both a proxy of the budget,this is a very plausible explanation. The female cast ratiois not directly linked to a movie’s popularity. However, wehave also discovered that genres likely to be popular, suchas adventure and action, also tend to feature fewer femalecast members. Disney has made contributions to better fe-male representation on the big screen and its productions areoften very popular. However, their productions still featuremore males than females. /B ratio Rating PopularityVariable Regression Random Forest Regression Random Forest Regression Random ForestTest MSE . × − . × − . × − . × − . × − . × − Female cast ratio . × − − . × − − . × − . × − . × − . × − . × −
10 0.00 9Drama − . × − − . × − . × − . × − − . × −
11 -0.018 15 . × − . × − . × − Table 2: Regression coefficient estimates and Random Forest predictors importance ranking for variables selected by the regu-larized regression models of movie revenue/budget ratio, rating and popularity.Our findings have demonstrated that the difficultieswomen in the film industry face are both upstream anddownstream: female filmmakers, especially female screen-play writers, are instrumental for movies to have better fe-male representation, but the percentages of female filmmak-ers are very low; only 6.0% of directors, 9.7% of produc-ers, and 12.2% of screenplay writers are female. Meanwhile,lower budgets are provided to support the movies that couldtell good stories about women, thus causing the films to re-ceive more criticism. The demand of better female represen-tation from viewers is also not strong enough to press thefilm industry for change, as movies that have poor femalerepresentation can still be very profitable.
Discussions
Our study provides a larger picture of female represen-tation in movies and how it is perceived by the audienceover time. Unfortunately, underrepresentation of women inmovies is not the only difficulty women are facing. Apartfrom the low ratio of female cast members, portrayal ofwomen in stereotypical ways that reflect and sustain sociallyendorsed views of genders, and depictions of relationshipsbetween men and women that emphasize traditional rolesand normalize violence against women, are another two im-portant themes of how media portray gender (Wood 1994).Although our newly-defined measure of female representa-tion, the female cast ratio, has shown to be more successfulin evaluating movie success than the Bechdel test, it alsohas limitations that fail to further explain how women areportrayed in movies. For example, our findings indicate thathorror movies like to feature more women than men, and thereason behind this phenomenon is that the audience enjoysthe victimization of women more than the victimization ofmen (Tamborini, Stiff, and Zillman 1987). In other words,the root of ”favoritism” toward women in horror movies isstill that people want to see women being helpless and pas-sive rather than men. Another genre identified by our anal-ysis that favors women, Romance, is also likely to portraywomen in a stereotypical way. A linguistics study of themovie ”The Best of Me” shows some differences in the ex- pressions between men and women, in that men prefer touse commanding directives while women use directives forsuggesting or requesting, and that men tend to use swearwords to express anger and women tend to use swear wordsto express bad feelings (Sulastri, Laila, and Hum 2019).Even in a movie that targeted the female audience, womenwere portrayed as docile and submissive compared to men.Meanwhile, some studies have used dialogue speaking timeto evaluate how much women talk in movies and discov-ered that even in some female-led movies (such as Disneyprincess movies), the lead female’s speaking time could beoutnumbered by male cast members (Anderson and Daniels2016). Whether they speak to express their ideas, feelings,or make commands, men speak more than women and theirinfluence on the audience is ultimately stronger.We will also consider many other factors that can influ-ence the image of a character in movies in our future studiesto help us better understand how gender works in shaping acharacter’s image and how they influence the audience’s per-ception of movies. For example, race and ethnicity are bothimportant factors that impact the audience’s expectation fora character. For example, studies have shown that AfricanAmerican music videos were significantly more likely toportray sexual content and sexualized female characters thanWhite videos (Turner 2011). In addition, the first AfricanAmerican performer to win the Academy Award was a fe-male, and she won much earlier than her male counterparts.The reason behind the phenomenon that minority womenare more popular than minority men is possibly that womenand minorities were traditionally both considered to be sub-missive to white males, thus a minority woman portrayedin such a way is more acceptable to the mainstream than aminority man. Besides race and ethnicity, age is another im-portant factor. While in movies both women and men in their60s and older are dramatically underrepresented comparedto their representation in the U.S. population (Lauzen andDozier 2005), older actresses experience greater difficultyin finding a job than older actors even for celebrities (Tremeand Craig 2013). The majority of male characters are in their30s and 40s, and the majority of female characters are inigure 2: Movie budget, revenue, rating and popularity overthe last 12 decades. their 20s and 30s. For male characters, leadership and occu-pational power increase with age; however, as female char-acters age, they are less likely to have goals (Lauzen andDozier 2005). This observation coincides with the main-stream expectation of genders in that men hold authority andleadership which usually increases with age, while womenare in more subordinate roles and often sexualized, which iswhy younger women are preferred.In conclusion, we hope our findings send a message tothe film-making industry that movies with better femalerepresentation are more likely to succeed. Improving fe-male representation in movies requires both upstream effortand downstream effort: female crew and cast members con-tribute to better female representation in movies, and posi-tive feedback from the audience encourages investment intomovies to hire more female crew and cast members. There-fore, we encourage the film-making industry to employ morewomen as a start. Men have been the leading voices of story-telling in movies for decades, so we believe that stories fromwomen’s perspectives are worth exploring and would attractlarger audiences.
Acknowledgement
The authors would like to thank Joyce Luo and WilliamArtman for their help with writing.
References [Anderson and Daniels 2016] Anderson, H., and Daniels, M.2016. Film dialogue from 2,000 screenplays, broken downby gender and age. https://pudding.cool/2017/03/film-dialogue/index.html .[Berkman, Garland, and VanSteinberg 2017] Berkman, S.;Garland, S.; and VanSteinberg, A. 2017. Quantifiedfeminism and the Bechdel test. Technical report, DukeUniversity.[Collins 2011] Collins, R. L. 2011. Content analysis of gen-der roles in media: Where are we now and where should wego?
Sex Roles
Feminist Media Studies
Sexualization, Media, & Society
Proceedings of 3rd international conference on documentanalysis and recognition , volume 1, 278–282. IEEE.[Lauzen and Dozier 2005] Lauzen, M. M., and Dozier, D. M.2005. Maintaining the double standard: Portrayals of ageand gender in popular films.
Sex roles
Women’s Media Center
Socius
Sociological Inquiry .[Murphy 2015] Murphy, J. N. 2015.
The role of women infilm: Supporting the men–An analysis of how culture influ-ences the changing discourse on gender representations infilm . Undergraduate Honor Thesis, Department of Journal-ism, University of Arkansas.[Sulastri, Laila, and Hum 2019] Sulastri, S.; Laila, M.; andHum, M. 2019.
Characterizing Men And Women LanguageIn The Best Of Me Movie . Ph.D. Dissertation, UniversitasMuhammadiyah Surakarta.[Tamborini, Stiff, and Zillman 1987] Tamborini, R.; Stiff, J.;and Zillman, D. 1987. Preference for graphic horror featur-ing male versus female victimization.
Human Communica-tion Research .[Tibshirani 1996] Tibshirani, R. 1996. Regression shrinkageand selection via the lasso.
Journal of the Royal StatisticalSociety: Series B (Methodological)
Applied Economics Letters
Sex Roles