McCarthyism, Media, And Political Repression: Evidence From Hollywood | Using Text As Data In Policy Analysis

>> Steven Davis: Welcome to the Hoover Institution workshop on using text as data in policy Analysis.

>> Steven Davis: I'm Stephen Davis, here with my Hoover colleague and workshop co-organizer Erin Carter. Joining us today is Tianyi Wang, a faculty member in the Department of Economics at the University of Toronto. He's also a faculty research fellow at the NBER and an IZA Research Fellow.

His research is mainly at the intersection of political economy and economic history. Today he will present some recent work with Hui Ren Tan on McCarthyism, Media and Political Repression, Evidence from Hollywood. Here's our format, Tianyi will take about 30 minutes to present, then we'll turn to a discussion.

If you have a question or comment during the course of his presentation, you can put it in the Q and A box and we'll try to get to it. During our open discussion, we will run about 60 minutes in a recorded format. After that we'll turn off the recording and continue the conversation with a more informal discussion.

For anyone who wants to stick around, Tien Yi, the floor is all yours.

>> Tianyi Wang: Okay, thank you so much, Steve and Erin for having me here for the introduction and for everyone for joining here. And I'm really excited to share this work with you. So my name is Tianyi, this is still a work in progress with Hui Ren, and we are looking forward to hearing any comments or suggestions from you.

So in particular, the motivation for the talk is that demagogues have existed throughout the history of democracy. So for example, Cleon of ancient Greece has tried to convince the Athenians to kill every single man in the rebellious city of Mytilene. In the more recent centuries in American history, Alexander Hamilton also has warned Americans about those people who began their career by paying an obsequious course to the people by commencing demagogues and ending terrorists.

And even in more recent years, we have also witnessed democratic backsliding across different countries, which has only heightened such concerns. And yet, despite the importance of demagogues and their far reaching influence, there's still little empirical understanding of their impact. So that's what we are going to turn to in this study to examine a far reaching episode of demagoguery in US history.

So in the early days of the Cold War, in an episode known as the Red Scare, anti communist hysteria swept across the United States, ushered in by Senator Joseph McCarthy, who lent his name to this movement as McCarthyism, but also by other people. So the movement has led to widely publicized and unsubstantiated allegations of individuals of having communist ties or sympathies.

And this led to a climate of fear because 1 out of 5, or 13 million US workers, were subject to some form of loyalty reviews, investigations or basically a test of repression. So historians have argued this as the most widespread episode of political repression in US history. Yet there's still little empirical evidence on impact of McCarthyism, despite its historical significance.

So in this study we turn to examine a key episode during this era, that is the anti communist crusade in Hollywood. So during the Red Scare, hundreds of professionals such as writers, directors or actors in Hollywood were accused of communist ties or sympathies. So while that number may not sound alarmingly large, but during this time period the US movie theater attendance was huge.

So it's estimated roughly 40 to 80 million people would attend each week the movie theaters, making this the most popular mass media and entertainment media during those days. So we can imagine that if there were any changes in the types of people working in the movie industry or the kind of movies that were produced during this time period, if they were affected, there could be large important implications in society as well as on people's attitudes or hearts and minds because of the exposure to movies.

So in this paper we focus on both the career impacts of the Hollywood anti-communist crusade, but also its broader political implications. So in particular, to study the impact of the event of this episode of anti communism in Hollywood, we have assembled unique data measuring who were accused of having communist types of sympathies, basically who are targeted, as well as detailed individual and film characteristics during this time period spanning four decades.

So in the baseline we basically look at who are targeted, basically the determinants of the accusations, but also what's the impact of being targeted whenever look at the effects of accusations on individual career outcomes as measured by their creative output in films and TVs using a matched difference in difference approach.

And in addition, basically the idea is that we'll compare the people who are accused as well as similar co workers in the past to see how they trend over time in terms of their career outcomes. But in addition, we also examine what happens to the films themselves. In particular, we look at the changes in political sentiment in Hollywood films using some machine learning methods.

We'll talk more later. But also its broader political implications on the society such as during elections. So as a quick summary and preview of the results, in the first part of paper, when we look at the determinants of accusations. We found that people who are prominent, as well as those with progressive viewpoints, they were more likely to be targeted in Hollywood.

So in terms of the impact of accusations. Those who were accused, such as the actors or writers, they actually experienced a large and significant impact on their career outcomes and a significant decline in their creative output in both movies and tv. And it's not just quantity but also the quality of those productions.

So in terms of changes in movies, we actually document changes in the political sentiment, in particular a sharp decline in the progressiveness of the film production during the Hollywood Red Scare, which suggests the films were becoming more conservative politically during this time period. And in the last part of the paper, we studied impact of exposure to such film during this period and found that the decline in film progressiveness or as the film were becoming more conservative, exposure to such films led to counties to be more supportive of the Republican Party in presidential elections.

So I will skip the literature part, literature review, and also the background just because of time so we can save more time for questions and discussion later. But just want to give a quick summary of the data we used in the paper. Because one challenge we faced when we started the project was basically there were no systematic data to measure the individuals who were accused in Hollywood during the Red Scare.

So therefore we have to combine several different sources of data to measure who were targeted. So the information actually were sourced both from the publications of private organizations who named individuals as communist members or sympathizers during this time. Time period, but also some official publications from the House Congressional committees that led the investigations into Hollywood.

And this was the committee called HUAC or House Committee on Un-American Activities. So they have their own annual reports. So combining these sorts of data, we are able to compile basically a list of individuals who are accused. And it largely aligns with what historians have mentioned. About 200 or 300 individuals in Hollywood were being targeted and accused.

So in addition of the individual data they also obtained data to measure the film and also the TV show productions. So in particular from IMDb as well as the American Film Institute or AFI, we have obtained data that covers the universe of US feature films as well as TV shows over four decades.

Including the cast, the production team such as the writers or producers, as well as the characteristics measuring the content of the film, such as their genre, their subjects, which means basically their main themes that the film was exploring, but also the synopsis as well. And in addition to measure individual political activities prior to the investigations, we have also put together records of basically data measuring people who publicly opposed HUAC before at beginning of the Red Scare.

And this comes from various historical records, such as those signers of the Brits attacking HUAC or newspaper radio advertisement opposing HUAC as a measure of basically their dissenting activities against the investigation of this government body. So basically, before I end, I go into the detail basically HUAC, just briefly on historical background.

It basically was considered also quite controversial during this time period as even the President Truman actually caught this committee where iAmerica themselves because it turned out that they actually consist of several members of the KKK as well as other groups which of course were considered controversial. So basically using this type of data, we first examine the determinants of being accused.

And here comes the regression results. So there are three columns in this table and these are from three separate regressions. So in each regression we look at basically whether the type of the workers in the regression, such as actors or writers, etc, the individual were named or accused in one of the publications we put together.

Basically, they were being accused of having communist types of sympathies during this time period. And they look at what determine their accusations or who are targeted, such as their demographics or career profiles in the past, as well as a measure of the type of film they were making in the past in terms of their progressiveness.

In terms of film content. I will mention more about how we measure that, but here is basically a summary of how progressive the film they were making in the past and also whether or not they have participated in some acting activities opposing HUAC. And basically we find that we do not find much evidence in terms of demographics as a predictor of being accused, but we do consistently find that, for example, for the actors having more experience or more prominence as measured by their past Academy Awards nominations, tend to strongly predict their being targeted.

And across the board we find also past activities such as opposing HUAC increase the likelihood of being targeted. Perhaps not surprisingly, but I think what's also interesting is the film progressiveness measure, such as, for example, actors and writers. We do find that those who were involved in more progressive films in the past, such as films often sympathizing with the workers rights or minority groups during this time period, they're more likely to be targeted as well.

So while here we show some evidence of who are targeted, one thing I think before I turn to examine the impact of accusation is to mention that during this time period being accused does not have legal sanctions. So being accused of communist members or sympathizers there's no legal laws, for example against these people or requirement for them to be basically punished.

So whether or not being accused has some consequences on these people's career is largely an empirical question. And that's what we are going to turn to next. And I'll keep this part rather brief because I want to save more time later on for the discussion on the tax analysis part.

But just as a summary, when we for example look at the accused actors, we do find basically a difference in terms of the career outcomes as compared to the control group. So here I want to show basically the draw time series data comparing the people who were accused as compared to their past co-stars.

And we see that they follow largely similar trends in their raw Korea output number of titles in movie and TV in the past. But right after the start of the widespread accusations, those who were accused actually show a decline in their number of titles in movie and TV shows.

And that's also evident in the event study, basically showing quite a large and negative impact on their creative output and careers. And as a quick summary, I also find similar evidence. Here are some additional findings. Basically, for example, we also find larger effects for accusations that were made by the state, so in this case by the House committee, basically the HUAC investigators.

So the effects actually could be taken as more serious and all trustworthy. So that might explain the larger impact there. Erin, I see your hand.

>> Erin Carter: Yeah, thank you. Just a quick clarification question. What's Your explanation for the mechanism in this, your decline in productivity?

>> Tianyi Wang: Yeah, so I'm getting there actually in a few seconds actually on this slide.

Yeah, thanks. Yeah, that's something definitely we'll examine. But before getting there, we have also found some similar evidence in terms of the quality of the output as measured by the IMDb ratings. So in addition we find similar effects in terms of the on the extensive margin. So the people were not just like having less opportunities, but also when we look at whether or not they were even working in the industry or not, they also were less likely to be even employed during this period and also had less opportunities to play lead roles in films.

And we also find similar evidence from screenwriters as well, so showing a large decline in their career outcomes. So getting to Arian's questions, a natural question to explore is what might be the potential mechanisms? So it could come from both the audience side, for example, the viewers may demand and people to be taken away if they were considered too controversial.

But it could also reflect the employer's own concern, for example, their concern for perceived backlash or risks of hiring these individuals or losing potential sponsors for the show. So basically in the paper we show that we have defined no evidence for audience backlash. So we actually examined box office revenue for films that actually involved some accused individuals and we actually found no evidence that such films suffered in terms of their box office revenue.

Even when we look at more plausible. Exogenous variation just around, in which the exposure or the appearance on the film was more exogenous, such as being named right before and after the publication of some of these basically accusations. And we still find no evidence for a popular boycott, basically suggesting it's likely not the audience who are really driving the results.

And in the paper we provide more qualitative evidence, using some historical narratives as well, consistent from what the people were actually mentioning during this period, that the effects were more likely to be driven by employers who are taking actions in order to avoid potential controversies or perceived risks, as they were worried potential government backlash or losing the important sponsors for the show, et cetera.

And therefore they take actions to not hire the controversial individuals. Okay, so of course, well, so far we are still looking at these individual careers. A more important question is what happens to the types of films being made. And in particular, I'm going to show a quote here basically from the Investigator or HUAC, which suggests what kind of films they were particularly targeting or investigating.

So here it mentions that they are less interested in a film that has explicit communist context where only few people would watch it, but instead they are more interested in some ordinary Mary or John Mary pictures, which more people would have watched, but even with a single drop of progressive thought in it.

So clearly they were trying to target films with more progressive content in them. And similarly, during this time period, there were screenwriters, such as Dan Ren, who published the Screen Guide for Americans, basically advising filmmakers what to make and what to avoid. So she advised filmmakers to, for example, do not smear the free enterprise system or industrialist or any profit motive.

So basically do not attack anything that's similar to capitalism and also to not glorify depravity, failure or anything about a common man as well. So it suggests that there could be some changes in the film content as filmmakers may become more cautious of pursuing certain subjects as they may be worried of getting the attention from the government investigators or being targeted.

So is that the case? So how to measure changes in such political content? A challenge we faced is that there's actually no systematic measure that basically quantify what's the sentiment of films over a long span of time. So the existing ones were more qualitative in nature. So that's why to overcome this challenge and make progress, we turn to machine learning methods and in particular word embedding to study changes in film content.

The goal is basically to measure the progressiveness but also conservativeness of each film over the study period, over four decades or so. And to our knowledge, this paper actually provides the first empirical evidence systematically that use machine learning to measure the political sentiment of films. So and a little bit more on how we implement the word embedding in this context.

So in particular, as some of our audience may have already known, so word embedding basically converts, what it does is to convert text to some high dimensional vectors which can then be compared with one another to measure the similarities between a text. So basically it's a measure of similarity between different text.

And in particular, we perform this embedding method on using text data that measures each film's major subjects which identify the main themes of each films. So basically we can identify how similar is each film relative to another film based on the text data of their mean subjects using word embedding.

And just as an example to show what was meant here, basically we can look at one example which shows the data, actually the text data, the raw data from the AFI catalog or American Film Institute. And this catalog basically lists all the subjects for each film, measuring its key theme.

And this is an example from Charlie Chaplin's Modern Times, a classic film. And we can see that the main subject or major subjects mentions things such as class distinction or the depression, factory workers or anti unemployment, which aligns well with the main themes of the film. Well, when we look at minor subjects, it mentions things like arrests or cafes or even for example roller skating, et cetera.

So this is of course mentioned in the film, but not close to the core of the core theme of the movie. So that's why when we look at, when we measure film content, we focus on the major subjects of the film. But the results are also similar when you look at the minor subjects altogether, even with the synopsis as well.

So basically to perform the word embedding, we use a pre trained embedding model from Cohere. It's a leading AI and large language model platform similar to ChatGPT's OpenAI. But one thing we realized when we work on this is that the embedding model from ChatGPT sometimes can still produce some stochastic changes each time we ran.

Well, the cohere one is more consistent every time we run it. So that's why we chose cohere over OpenAI. But the results are qualitatively similar when we use the other as well. Okay, so just to show a proof of concept of performance of embedding in this context, to measure movies using their mean subjects, here is basically just a proof of concept looking at embedding of some classic films in history on a 2D embedding space.

So the movies are trivial. Sourced from the American Film Institute's 10 top 10 list, which lists some of the most classic films in American history from 10 different genres. And here we focus on five genres with sufficient difference one way or another for the purpose of presentation, such as Court Rule Drop, courthouse drama, gangster movies, romantic comedy, etc.

And here what we find is that films from each genre indeed tend to be clustered closer to one another. So all the gangster movies were closer to one another, all the western movies were close to one another, similar to sci-fi, etc. Which suggests that when we look at the mean subject of the film as data, the embedding actually can identify films that are similar to one another.

>> Steven Davis: What are the dimensions? What's on the vertical and horizontal scales?

>> Tianyi Wang: So basically this is actually generated from web. So it's actually basically a measure of the relative position of each film in this 2D embedding space. So this is like, a vector space, but there's actually a downsized to just 2D for presentation.

And so being close to one another in this 2D space means basically close to another to closer to one another. Okay, but it's hard to interpret without some notion of what the content along each scale. Yeah, basically it's basically embedding a vector space. Basically, to me, I think the way to interpret this is more about the relative distance between these dots, which measures how similar they are, whether to the left or right.

A single dot doesn't really can convey that much message, which I agree with you, Steve. Yeah, and while this is only for a subset of movies, we find similar evidence when we look at all movies. When we do the embedding on the full set of American feature films looking at their main subjects and across different genres, we find that.

For example, basically all the films within each genre tend to be clustered around each other. For example, here the brighter the color is, that means a higher density of films being clustered to one another. So you can see all the westerns tend to be clustered around this position, and the mysteries ones tend to be here, etc.

Which, again, provide some support to the embedding model, that it is doing what it's supposed to do, basically identifying film similar to one another. So you might be wondering basically why we do this, why we show this, right? How do we use the films that are similar to each other, right?

The similarities between films. So basically here it comes to the key part to identify the film political sentiment. So in particular, in order to measure each film's relative political sentiments, such as progressiveness, we following several steps. So basically the first step is to identify the similarities between each film to a known set of progressive films.

Basically, we want to know for the given film how similar is this film to a set of films that we know are progressive. And we know that they are progressive because we check that using a few scholars publications, books. In fact, three different books to identify themes that were considered progressive on social issues and from a progressive point of view during this time period.

>> Steven Davis: Are you going to define what progressive means? Cuz word has different connotations in different eras.

>> Tianyi Wang: Yes, yes, yes. I'm going to show you some example in a second. Of course, yeah, just some illustrations. And here basically let me show you all the films that were considered, I think consistently identified by these three books, film scholars book and the type of issues they were exploring.

For example, this shows, across three books, the movies that were identified with progressive topics. For example, the first one is the Grapes of Wrath, and topic was on unemployment, poverty, the Great Depression, etc. And as well as other films such as those on anti-Nazism or anti-Semitism, basically attacking anti-Semitism or racism in history, etc.

Or unemployment, etc. The class difference between rich and poor or inequality, homelessness. And basically this forms kind of the benchmark set of movies that we know film scholars has identified as consistently as progressive films. Basically films with progressive views on social issues. And as another illustration in terms of the subjects, right?

What these films are about. We can see basically mentioning on depression, which means the Great Depression. But also workers issues, a little bit on African-American, or unemployment, or for some racism, etc. So again, what we do is for each film we have in the data, we compare their main subjects relative to basically, this set intuitively compared to this set of words.

Basically to see how similar is each film relative to topics like them, right? If they're very close or very similar, then it suggests they are more progressive potentially. And we also do something on the conservative side. So basically to measure how conservative each film was over time, we basically we focused on anti-communist film.

Because they were clearly conservative in terms of the political outlook. So we obtained this from the University of Washington's red scale filmography. And we again can look at the subjects. Here on the right-hand side you can see words clearly like communism, etc. Or also mentioning on patriotism, war or some Russians, etc.

And using basically these two benchmarks we can then compare on net how progressive is each film, how conservative it was. And by basically taking the difference between the Psim, or progressive similarity, and the Csim, or the conservative similarity between the films. And to measure the net difference between them.

And we take the net take this difference because of two reasons mainly. The first one is that the text data measuring this film characteristic, like the main subjects, were also changing over time, like their length were changing over time. So this would affect both measures individually, potentially. Just because the length of the text was becoming more and there are more words to compare with over time.

And so we take the difference to net up this overall secular change in the length of the text data. But also because a film can be both similar to progressive films and also conservative films. And so it'll take difference to measure the net difference, or the net how progressive it is, relatively speaking, basically.

Okay, so basically after these following steps we can basically assign a score as the net progressive score for each film. And basically we then look at how it evolves over time, on average across years. And here this just shows you the descriptive figure that shows you the average film progressiveness, where we measure the progressiveness of each film, then averaged across all films in each year and plotting the average over time.

And we can see that there was some reduction in progressiveness, such as during World War II periods. And that's because there were more war films during this period with patriotism themes on wars, etc. So that looks more conservative. But what's interesting also is we find this sharp decline in film progressiveness right after the start of the Hollywood investigations by HUAC.

And if anything, also a larger decline later, years after the more intensive period of the Cold War, potentially because more films were exploring these themes on anti Communism. Yes, Erin.

>> Erin Carter: Thanks, yes. So do you know what the time is from writing a script to actually having a movie in theaters?

I'm just curious for how long. Long it could take in theory, for this sort of rapid change. Cuz this is a really rapid change after.

>> Tianyi Wang: Yeah, during this time period filmmaking was actually quite efficient, quite fast. Like even just a few months. So the average time you look at the data for production time is actually just two months.

Yeah, but plus a few, like, you know, sometime on average, like it's a few months only in terms of production.

>> Erin Carter: What about like sourcing a script, getting all the financing?

>> Tianyi Wang: Yeah, so like that could take some more time, but. Yeah, so, yeah, that's a good question.

I think I'll also look more into that. But on average. So it can maybe potentially it should be faster, I think quite fast. Once you source the scripts, etc., shouldn't be much longer than a few months, yeah. So can I just check how much time I have left?

I know you should.

>> Steven Davis: Trying to wrap up the next five minutes or so.

>> Tianyi Wang: Okay, yeah. So while this figure is highly suggestive because we see a decline in film progressiveness during the start of the Red Scare. And also it kind of came back after the decline of Red Scare.

So suggesting that the Red Scare in Hollywood likely have decreased the progressiveness of films. But again, this is only other things that could have contributed to the decline in the film content. So basically a question we're still trying to answer, trying to investigate is was there a chilling effect on filmmakers?

Because there were some anecdotes suggesting filmmakers were indeed become more careful scared of pursuing certain subjects. So could some people be just staying away from more progressive topics? But this turned out to be a bit puzzling to us because we try to investigate this, for example, looking at the accused filmmakers as well as their past coworkers in terms of their progressiveness over time.

But we didn't find much evidence in terms of any changes in the differential change in their film progressiveness relative to other filmmakers. So this perhaps just implies that on average, everyone was actually becoming more careful and trying to shy away from these topics. So there was not much difference between the accused or their past co workers, etc.

But of course, we welcome any suggestions if you have ideas on how we may investigate further. For example, especially in a more rigorous way causal way to show some people might be becoming more basically chilled in terms of what they want to make in films. And in the last part of the paper, we basically look at the more conservative sentiment in Hollywood films during this era actually may have affected society as we saw the decline in progressiveness in the films.

So to do that, we actually need a measure of local exposure to movies. And that's why we have collected data to measure across US Counties the exposure to movie theaters as the number of movie theaters per thousand people in 1940 before the start of the Red Scare. So here we can see that within each region, although there are some of course, regional differences, but within each region there's a lot of also variation across different counties in how much exposure to movies to movie theaters people had.

And we then use a difference in difference regression to look at basically in place with more movie theaters across different counties. What happens when films were becoming or changing in terms of their progressiveness like the one is basically the figure which we saw a few slides ago measuring their average progressiveness over time and controlling for both county fixed effect.

And also state by year fixed effects, as well as some robustness check looking at county controls interacting with year Dummies. And the outcome variable is basically the Republican vote share in presidential elections. So what we find is basically in this table so far, not just focus on the first two columns as the last two are more about some falsification test.

So in the first column, for example, very conditional on only county fixed effects of state by year fixed effects. Counties with more exposure to movie theaters actually became more Republican supporting when there's a reduction in the net progressiveness of the film average. So here is a negative coefficient.

But remember, we saw a decline in film progressiveness later in the year. So such a change could have increased the support for Republican Party during this time period. And the results are robust to controlling for county controls. And while also we can look at some falsification tests looking at other types of movie theaters in columns 3 and 4, such as art house movie theaters or black movie theaters, because these movie theaters produced much different films during this era and the effects likely would be much more limited.

And indeed we find like when we look at exported to these alternative film theaters, we do not find to see the same effect on Republican voting. And the last part of the paper also explores a little bit about the different types of films, a little bit closely focusing on anti communist films such as those on internal and external communist films.

And I think I have one more minute, so I'll be really brief here. But essentially we want to compare the effects of anti communist films focusing on whether those who are targeting internal communists such as like internal subversiveness, from internal spies within America, all those targeting external threats from communism like from Soviet Union.

And we compare a difference effect because those on internal communism is more consistent with the message from McCarthy basically. So if we find some effect from this type of movie on internal communists, it's more likely to suggest the effect was induced by McCarthyism. And that's what we find.

Basically we find although we look at overall effects, for example, the share of anti communist films overall each year, we do find exposure to more anti communist films, actually increased Republican vote shares. But the effects are mainly coming from anti communist films from targeting internal communities, which is consistent with McCarthyism and as compared to those targeting external communist threats.

So the results suggest that at least a large part of the effects from these anti communist films were actually induced by the message from McCarthyism. Okay, so let's come to the conclusion slides basically. So in this paper we examine an important chapter in American history that has received little investigations.

That's the impact of McCarthyism. And using several novel data sources and methods from machine learning, we can identify basically the impacts of accusations, but also changes in the film content, which could have important political implications. So which highlights again the importance of text as data and to provide new opportunities to study films or entertainment media and their broader political impacts.

So, yeah, that's all. So thanks everyone and please feel free to let me know the questions or comments.

>> Steven Davis: Thank you Tianyi. Let me start with some high level comments. I really enjoy the presentation, but I guess I have a few things to say which might seem critical.

First, I think it's not actually so helpful to lump so many things into the progressivism category. So if you look at what you've got there and your net difference between progressivism and anti communism, you have at least two very distinct themes going on. One is we fought a war against the Nazis and our allies were the Soviets during the war.

And then later, after the war was over, there was a global contest between the Soviet bloc and the US Led west, so to speak. And those were enormous global. You could even argue existential struggles. So there's a lot of reasons why you might expect that to influence the content of films.

Okay, apart from the activities of the House on American committees, then there's a separate stream running through your progressivism construct, which includes things like concern for. For workers, concerns about racism, concerns about anti Semitism. I understand that they have some connection to the anti communism because the workers rights were sometimes wrapped up in socialist or leftist movements that in some cases, but not all had sympathy for communist ideologies and regimes.

But they're quite distinct. So I don't find it so useful to lump them up altogether, I'd like to see you separate them out. And just one smaller point along these same lines. You know, you had antisemitism as a feature of progressivism in your list that at some point along the way, since, since the 1940s and 50s, antisemitism was no longer a.

Viewed as a, how to put it, it would no longer have the same relationship to what we think of as progressivism as it did then. Okay, so I just don't find the term progressivism very helpful. I would reduce it to more primitive concepts that have a more durable and clearer meaning.

So that's kind of point one. Second, I find it'd be interesting to have more content, more context for your study. So we know that there are pressures on film writers. I'm sorry about that.

>> Tianyi Wang: All right.

>> Steven Davis: And film writers and, and producers and actors all the time.

And you had a phrase there, employers, meaning the film producers, I take it, taking actions to avoid controversies and perceived risks. Well, what happened after the George Floyd murder in 2020? All of a sudden, artists, film writers, film producers, writers and so on, were moving away from pro police or type films and so on.

So these things go on all the time. And it's true that the House UN American Committee may have had some, its own marginal effects, but there are other things happening in the broader society, some of which I've already mentioned. So just, it would be nice to just characterize descriptively, to start with how the content of film shifted over a longer period of time in the wake of salient developments.

So I'm pretty sure you'd see a big shift after the pandemic struck because it was tied up with these other things. There was a time when it was, I think, viewed as risky, career wise to be openly in favor of Reagan Republicanism that eventually faded. The stigma that attaches to being pro Trump, which was enormous in 2016, has certainly shifted.

And I suspect all these things affect what types of movies get made, what kinds of actors and actresses. So I just wanted some broader context that would help me then think about the marginal effect of the House UN American Committee's actions had in this particular period.

>> Tianyi Wang: Yeah, so Steve, I agree with you.

I think yeah, on the first point on for example, how do we measure progress? That's indeed a challenging aspect because you know, there are different concepts of progressivism and also progressiveness and also it can evolve over time. So and that's why we actually also was trying to think of like, you know, ways because this hasn't been attempted in the past and would be some ideal ways to measure that.

So that's why when we discovered this actually from 1950s there was this study by contemporary film censorship board in America. They actually basically argued that during this time period there was a decline in social problem films. It's called social problem films because oftentimes they explored kind of important social issues during the day and they actually gave a few examples.

And one thing they mentioned was anti-semitism during this time period. So that's why we didn't want to ignore that at least.

>> Steven Davis: Well, I wouldn't wanna ignore it, but separate out the national security, international global conflict aspects of what's in your progressivism definition from the social problems aspect.

I would just dispense with the word progressivism entirely and try to at least make that distinction. And then one question I have is how many of the people that the house, American House, the hwac, how many of the people they went after were really motivated by things that had little to do with communism and really had more to do with these social.

Yeah, the social problem. I don't know. I think it's a pretty interesting question. Maybe the anti communism aspect of it was maybe just a veneer to go after something else. I agree with you.

>> Tianyi Wang: Yeah, that's the context part. I agree with you. That's a very well put actually a way of putting it.

It's a veneer of targeting people. But actually beneath that anti communism veneer they actually work going after people who are not communist. But they were actually sympathizing with progressive or you know, basically left leaning viewpoints. For example, there were films being named such as It's a Wonderful Life.

It's has been named by FBI as a controversial film because of it's considered a class distinction between rich and poor and the portrayal of the- You should get better in the paper. Yeah. And also Etsy, one of the most famous movies of all time. Nobody would see it today.

Some kind of anti-American film. So FBI went after It's a Wonderful Life and also like similar movies which were clearly sympathizing with the lower class in society inequality. So that's why we actually went after this social problem film because this context they were not really just targeting explicit communities of films and they were not actually much theme on communism in first place.

It was mostly, yeah. Sympathetic is the left, for example. Yeah, and I would welcome any suggestion. For example, if we don't use progressive maybe social issues social problem films for example. Maybe more accurate reflection.

>> Steven Davis: There's the global anti-communist struggle. Well first they were allies, then they were mortal enemies

That's one element of what's happening. But then there's the social problem aspect which they're not completely divorced, but I would separate those out. I wouldn't lump them in my. I would avoid the use of the word progressivism. Anyway, I've gone on enough on this. Let me see what Erin wants to ask.

>> Erin Carter: Fantastic, so I really enjoy this paper. There's a lot going on here, so lots of discusses. It's really fascinating. I'll just focus on two comments, last question. So first is that I entirely agree with Steve about progressivism. I think that label obscures more than it clarifies. And what I would love to see instead is making more use of the topic labels.

Right. So and I think given the huge shifts in culture taste salience of particular movie topics over time, I think what would be most compelling would be doing topic salience in a DID framework. Then when you're looking at the topics that act Actors, writers, et cetera, are taking on for those who are treated with the targeting versus those who aren't.

And I think that would be really compelling and a bit more inductive telling us about what they're working on changed as a result of being targeted rather than, you know, broader cultural shifts and domestic international issues, et cetera. So, you know, and I think that, you know, as a subset of that, you could get even more inductive by doing a distinctiveness algorithm, right?

Just like looking at the people who were targeted, getting the text of all of their movie scripts and comparing that to the corpus of people who were not targeted in their movie scripts and seeing what terms are most distinctive in the work that they're doing after being targeted, I think that would be a really great illustration.

That'd be really revelatory. So second, I'm still really interested in the mechanism here and I think. So you have tons going, three very different parts of this paper that you break down to different papers. The part that to me is sort of like the most paper style, most interesting way of what's going on is the shifts in behavior after being targeted.

And there, I think I would love to see more about the mechanism. So in particular, I'm still not totally convinced about if this is self-censorship or not being able to find work because of employer decisions. And I think that there's something that doesn't quite make sense to me, which is that if this is not driven, if you've shown this is not driven by popular shifts in demand, there aren't these popular boycotts, then why do the employers care, right?

If they know that people aren't going to boycott a film, why do they react so strongly to this sort of targeting? So I think I'd have one suggestion about how you could look at that a little bit. So in particular you have, I think this IMDb data. So you can do a lot with professional networks.

So if you look at the professional network of actors who are targeted, if you could see, so basically how wide does the stigma or how wide does the social effect, the fear go, really? So if someone is targeted, you see all of their coworkers or past professional network start to self censor as well, to change what they're working on, to sort of avoid certain projects, I think if you find an effect there in that broader professional network, that would be evidence that there is this sort of self censorship, fear dynamic going on at the actor level.

And then it's not really about employers, but I think like teasing apart Those two mechanisms would help, you know, build out sort of a richer paper. But overall I really enjoyed this. I think that it's very interesting.

>> Tianyi Wang: Thank you very much Erin. I agree with you and I think, it's a very nice situation also to look at difference in difference and using the type, the text, like labels, like topics, keywords, et cetera.

At one point, we did that at the beginning before we did try the progressive machine learning method. But we were thinking maybe to make it more kind of systematic, maybe using some other more advanced methods. But I think it can shed light also when you look at topics as well.

Yeah, definitely. And also past connection actually we did look at that past connection to the accused to see whether the past connected co star or co writers even they were self censoring themselves including having less progressive films. But we actually don't find much evidence that the past connected co-stars, co-workers, worked less on the progressive films based on the current measure.

But we did find some evidence using initially the another measure based on some dictionary based keywords. Some suggestive evidence like some topics worth shifting a little bit. But yeah, I think that's worth going back to also explore more. Yeah. And any suggestions to show self-censorship or tuning effect would be very welcome as well.

All right, can we let Elizabeth Elder, she's got some questions and comments. We'll let her speak.

>> Elizabeth Elder: Hi, thank you so much for this, this really interesting talk. So like Erin, one of my reactions with this, this seems like a lot, a couple of papers in one, maybe.

And in particular to me, it felt like the evidence on particular creative individuals and their kind of persecution and sanctions and reactions feels pretty separate from the kind of market level film production kind of, you know, tone and sentiment and political and content and effects. And in particular it seemed to me like you have, I, I was compelled by your case that you know, these people were being targeted individually because of probably their individual politics and political activities and their employer's risk aversion.

And this wasn't driven by kind of public, you know, boycotts or demands or things like that. But I wasn't quite as convinced that that logic really applied to the kind of changes in the kinds of films that were being produced. McCarthyism didn't, I think by the end he was pretty unpopular, but it didn't come out of nowhere.

It seemed like there, there probably was a broader public thermostatic kind of demand for more, more conservative content. And by conservative I mean like using your measure more kind of pro US like war film. Like you're talking about changes from social problem films to war films. There was probably a lot of public demand for that.

I don't really see how anything that you've shown eliminates the possibility that this is what the public wanted and not that this was a reaction to elite or government demands for sanction. So I wonder, you know, how much we can, you know, think of that as. I can't personally think of any research designs that could tease that out, if you can.

Wonderful. But I wonder if maybe there's kind of a paper initially on, you know, sanctioned to individuals and then a separate paper on changes in the composition of films being produced. Who knows why that was. But that change as kind of a treatment on the public and then looking at the vote changes after that.

And then, this is kind of a tack on to that second point. So sorry for going on a little bit, but I also was a little bit curious, you know, how much of the change in the films that are being produced is because of changes between genres shifting towards war films, spy films, whatever, versus changes within genres and how progressive the content is.

Because it seems like the former to me is probably more a story of aud audience demand. Like these are the kinds of films people want to watch right now. And the latter seems more consistent with the story of people are changing what's in the films that they're producing to kind of be in line with government priorities.

And regardless, public is demanding. So sorry. Sorry that, but this is really interesting, and I'm excited. Excited to be able to engage on it. So thank you.

>> Tianyi Wang: Thanks so much, Elizabeth. I think. Yeah, I agree with you. The film's progressiveness measure, like the time series one of the show.

Yeah, we didn't take that as a causal or as only coming from McCarthyism. We agree with you. There are other things that could go. They're going that measure, and it would be interesting to separate them out. But I think it's a suggestive piece of evidence, not descriptive one.

But we will hope to also do more than that and basically look at some ways to identify. For example, what Erin Steven mentioned, like self-censorship. Truly in fact, beyond just changes across genders because of popularity, but also there, within gender as well. Yeah, so any suggestion on that.

Yeah, will be helpful as well. Thank you.

>> Steven Davis: All right. Thank you. Can we? How about Brett Carter? He'd like to speak. Can we give him speaking privileges?

>> Brett Carter: Yeah, fantastic. This was a really fascinating presentation. I enjoyed it very much. I was curious if you could go back to the figure maybe about, you know, a third of the way into the presentation, maybe halfway through the presentation, that showed a dec.

Kind of the apparent progressivism of, yes, exactly, poems. And so the HVAC would then be 47. So what's kind of striking to me. You know, I mean, sure, you know, one can sort of see kind of a secular decline there. But, but the other thing that's kind of striking is, you know, the number of the spikes, you know, their, their volume.

I'm kind of curious to what extent you think there's kind of clustering, perhaps in kind of the political quality of films. I'm also curious if you were to extend the. I mean, it's kind of hard to know I guess, how to think about the magnitude of that decline, right?

I mean, especially, you know, kind of given that, you know, this is, you know, an original measure along the Y axis. So I'm curious if you were to extend the X axis out, you know, to what extent, you know, that would continue and then maybe that would give us some sort of way to kind of anchor your measure along the Y.

Along the Y axis as well. But anyway, yeah. So I'm kind of curious if you were to extend the X axis, kind of what that would look like, whether this is really part of a secular very long term decline. And then the bit about clustering as well, to what extent you think there's kind of clustering in the film industry that may or may not be independent of the HUAC and what might drive that clustering.

>> Tianyi Wang: Yeah. So thank you very much, Brett, for the idea to look at maybe even more recent years. Extending the X axis. Yeah, Yeah. I think that's a very interesting idea. Also we could do, we could explore, definitely. But also like yeah In terms of clustering, I wonder if I understood correctly.

But if you mean like, for example, like how many films, the volume of films, like, you know, taking into that, taking that into account as well, I think certainly we can also weight the. Yeah.

>> Brett Carter: I'm sorry, so, it's not, I guess I'm curious what exactly. I mean, the question about the mechanism, I think a number of people have brought up.

And so in some sense the question is about that. I mean, to what extent? So there are these sort of spikes around 1958 through 1960. I mean, you can sort of see this kind of upward trend that kind of begins maybe in 1954. And sort of rises and then again, you know, in kind of the early 1960s.

Right. And then obviously there's the trough in 1960, 1964. So if I were to look at that, you know, I guess, sure, one could fit some sort of regression line and show this kind of secular decline. But to me, the story, maybe the more interesting story is what exactly is driving these sorts of sort of fits and starts.

>> Tianyi Wang: Right. I guess I'm kind of curious whether there's sort of another, more interesting story here other than this kind of long term secular decline. And I think, you know, that's sort of another sense in which, you know, extending the X axis sort of might.

>> Brett Carter: We fought a war against the communist regime in the early 1950s.

Yeah, Korea. And we started another one in the early 1960s and got involved in another one in Vietnam. So this is back to my point about there's the global conflict here between the west and the US and communist regimes is probably playing a significant role in driving some of these movements.

And that's a very distinct concept from the social problems, concern for the workers and so on. Because I wanted to separate them out. I look at that and see less kind of secular decline, which, sure, you can make the argument, but more kind of sort of discrete trends that are driving these sort of spikes and declines anyway.

And so I think that to me is if I were sort of thinking about this, that would be kind of the most interesting area on which to focus.

>> Steven Davis: Okay, we're going to wrap up now. We've kind of had time, but I wanna thank you, Timi, for a very interesting presentation and paper.

There's more to do here, but you can see that people are quite interested and excited about what you're doing. So thanks so much. We're gonna turn off the recording now, but anyone who wants to is welcome to stick around for an informal discussion. So thanks so much.

>> Tianyi Wang: Thank you everyone as well for your helpful suggestions and comments.

Feel free to email me if you have more. Thank you.