What’s in a Face? Gender representation of faces in Time, 1940s-1990s

1. Abstract


Beginning with its inception in 1923, Time magazine, perhaps more than any comparable publication, has both reflected and influenced American popular attitudes to domestic and global politics. These include the changing ideas about women since the mid-twentieth century, which is the subject of this paper.

We used supervised machine learning to extract 327,322 visual images of faces from an archive of Time magazine, which contains 3389 issues ranging from 1923 to 2014, and computationally classified the faces as male or female. We then closely read selected Time articles to make sense of this quantitative data against the background of postwar feminism writ-large, and the history of the magazine itself. Our focus is on the period between the 1940s and the 1990s, which witnessed significant changes in attitudes toward women, and where our data of the proportion of female faces exhibits significant fluctuation.

We found four clear phases in the visual representation of women in Time from the 1940s to the 1990s: a peak in the mid-to-late 1940s, a dip from the mid-1950s to early 1960s, another peak in the 1970s, and another dip in the 1980s. The number of female faces depicted in Time then rises steadily since the early-1990s. We interpret these variations through an interdisciplinary framework. Through our combined quantitative and qualitative approach, we found that the percentage of female faces found in Time between 1940 and 1990 correlates with attitudes towards women in both the larger historical context as well as within the textual content of the magazine.

Methods and Results

We first collected data through human labor using Amazon Mechanical Turk (AMT) to identify and tag faces from the archive. This data was used to train a RetinaNet detector[1] to automatically identify and extract faces from the remainder of the archive. Using an accuracy threshold of 90 percent yielded 327,322 faces. A pre-trained face descriptor convolutional neural network VGGFace[2] was then fine-tuned and used to classify each face as either male or female.

Similar patterns emerge from both the AMT data and the automated data, featuring an increase in the proportion of female faces from the 1920s to 1945, a post-Second World War dip, a rebound beginning in the mid-1960s, followed by a decrease in the 1980’s, and a final rebound beginning in the early-1990s. Since similar trend lines were found using the AMT and the automatic extraction data, our analysis focuses on the latter, more comprehensive data set.

The proportion of women in each issue is shown in Figure 1. While there is a significant amount of variance per issue (compared to when the data is aggregated per year), a clear trendline emerges when the data is Lowess smoothed. Charting the proportion of women in each issue was useful for identifying outliers for our close reading analysis.

Figure 1: The percentage of women’s faces in each issue. The solid line is a Lowess smoothed version of the data.

To interpret the image data, we analyzed Time within the broader historical context of the 20th century. Our analysis consisted of a close reading of selected issues and articles, chosen based on the following criteria: 1) outlier issues from Fig. 2 defined as those between 1940 and 1990 with > 40% women, 2) issues and articles that were referenced in our secondary sources, and 3) results from EBSCO’s Academic Search Complete database, which we used to retrieve all articles in which the word ‘woman’ or ‘women’ or ‘housewife’ was mentioned within our dates of interest.

Our quantitative data coincides closely with the findings of our qualitative analysis. The number of images of women in Time magazine increases during the second world war as women’s role in the workforce expanded beyond traditional ‘feminine’ occupations to fill the gap and to satisfy increased production in the defence industry[3]. In the post-war years, the number images of women decreases as women across North America were instructed by “social engineers, such as psychologists, that they needed to be good wives and mothers in order to fit normally into post-war life”[4]. Our close reading analysis revealed that during these years, notable women who made the news bore the label of ‘housewife’, regardless of whether or not they were actually a housewife. The number of images of women increases once again as the women’s liberation movement began to take shape in the mid-sixties, peaking at the height of the movement in the 1970s. During this time, the women’s movement pushed the institutional structure of Time magazine to change from a gender-based caste system to a more equitable work place[5]. In the 1980s, there is a decrease in images, and we believe that this drop in the representation of women is consistent with the analysis in Susan Faludi’s well-known book Backlash[6], in which she describes the 1980s as a decade that rejected feminism. The representation of women then starts increasing once again in the 1990s and onward.


We found that a distant reading of the images of faces in Time magazine is consistent with a historical analysis of American socio-political trends and with a close reading of the magazine’s content. Specifically, we found that the percentage of female faces peaks during eras when women have been more active in public life, and wanes in eras of backlash against women’s rights. This finding is particularly relevant in our contemporary post-literate world in which people absorb culture through images, and spend more time scanning images than reading print content.

[1] Tsung-Yi Lin, Priya Goyal, Ross Girshick, Kaiming He, and Piotr Dollar, “Focal Loss for Dense Object Detection,” IEEE Transactions on Pattern Analysis and Machine Intelligence (2018): 1–1, https://doi.org/10.1109/TPAMI.2018.2858826;

Tsung-Yi Lin, Priya Goyal, Ross Girshick, Kaiming He, and Piotr Dollár, “Focal Loss for Dense Object Detection,” ArXiv:1708.02002 [Cs] (August 2017), http://arxiv.org/abs/1708.02002.

[2] P Omkar M. Parkhi, Andrea Vedaldi, and Andrew Zisserman, “Deep Face Recognition,” in Proceedings of the British Machine Vision Conference 2015, 41.1–41.12 (Swansea: British Machine Vision Association, 2015), https://doi.org/10.5244/C.29.41; Refik Can Malli, “VGGFace Implementation with Keras Framework. Contribute to Rcmalli/Keras-Vggface Development by Creating an Account on GitHub,” GitHub, [2016] 2019, accessed September 28, 2019, https://github.com/rcmalli/keras-vggface.

[3] Ellen Carol DuBois and Lynn Dumenil, Through Women’s Eyes: An American History with Documents, 2nd ed. (Boston: Bedford/St. Martin’s, 2009), 548–53.

[4] Mona Gleason, Psychology, Schooling, and the Family in Postwar Canada (Toronto: University of Toronto Press, 1999), 52.

[5] Curtis Prendergast and Geoffrey Colvin, The World of Time Inc: The Intimate History of a Changing Enterprise 1960 – 1980 (New York: Atheneum, 1986)

[6] Susan Faludi, Backlash: The Undeclared War Against American Women (Anniversary edition) (New York: Broadway Books, 2006).

Ana Jofre (jofrea@sunypoly.edu), SUNY Polytechnic, United States of America, Josh Cole (acole3@gmail.com), Queen's University, Canada, Michael Reale , SUNY Polytechnic, United States of America and Vincent Berardi , Chapman University, United States of America

Theme: Lux by Bootswatch.