Analyzing Patterns and Influence of Advertising in Print Newspapers (2025)

1. Introduction

In this paper, we investigate advertising in print newspapers across India. Despite the exponential growth of digital news consumption, print newspapers continue to hold a significant position in the media landscape, both globally and within India.Internationally, print media remains a vital source of information for many. According to a study by the Pew Research Center, 53% of Americans who prefer newspapers for local news opt for the print version over digital formats (Pew Research Center, 2023a). The enduring trust in print media further emphasizes its importance. The Reuters Institute Digital News Report indicates that 62% of respondents trust print newspapers, compared to 44% for news websites (Reuters Institute, 2023). This heightened level of trust is often attributed to the perceived editorial rigor and accountability inherent in print journalism.Economically, print media continues to be a substantial revenue generator for major news outlets. As of 2023, print circulation and advertising contributed approximately 60-70% of total revenue for several leading U.S. newspapers (Johnson and Smith, 2023). The New York Times, for example, reported that about 30% of its total revenue was derived from print subscriptions and advertising (The New York Times Company, 2023). Additionally, the effectiveness of print advertising bolsters its continued relevance. A study by NewsMediaWorks found that print newspaper advertisements were 1.5 times more effective in driving purchase intent compared to digital ads (NewsMediaWorks, 2022).

In the Indian context, the significance of print media is even more pronounced. The Indian Readership Survey of 2020(Council, 2020), the primary source for newspaper readership data in India, reveals that India is the second-largest newspaper market globally, with over 110 million copies sold daily. Print media commands about 20% of India’s total advertising expenditure, a stark contrast to the global average of just 4%(Farooqui, 2024). The industry is poised for robust growth in 2025, driven by factors such as elections, with print advertising revenue anticipated to reach an all-time high.

Previous research analyzing advertising at scale has primarily focused on digital newspapers(Sridhar and Sriram, 2015) or has been conducted on a small scale(Lindstädt and Budzinski, 2011). This is largely due to the difficulty of obtaining large-scale data on print advertisements. Such data is either exclusively held by the newspapers themselves, making cross-paper comparisons impossible, or it is expensive to acquire from third-party sellers(Majid, 2024), who may not even have data from countries like India.

In this paper, we obtain digital versions of print newspapers and use advances in image processing and Optical Character Recognition (OCR) techniques to develop a pipeline to extract articles and ads from print newspapers.We obtain data from 5 popular newspapers covering multiple regions and 3 languages from India (English, Hindi, and Telugu) with an aggregated reach of over 100 million people.We obtain articles spanning multiple years from these newspapers and apply our pipeline to extract with high confidence, all ads published in the paper along with their content.

Using this rich dataset, we answer, for the first time, who advertises, what they advertise, when (e.g. what time of the week, time of year, etc), where (e.g. on what page) and how (e.g. display ads vs. text ads).We specifically focus on two important categories of advertisers – corporates and government, which together account for roughly half of the total number of advertisements and advertising expenditure. We reveal some interesting patterns in print advertising.For instance, despite the fact that print circulation is dropping significantly(Pew Research Center, 2023b), print advertising remained consistent over the past 6 years. We uncover key patterns of advertisements by various companies and governments, including the over representation of company ads on prominent pages and the disproportionate revenue that government ads contribute to print advertising.

Moreover, using this data, we try to answer an important research question at scale – does advertising in a newspaper provide favorable coverage to the advertiser?We answer this question by regressing both the coverage volume and sentiment for various entities (companies and government). We find strong evidence for this hypothesis in the case of companies, indicating a clear trend in how advertising influences coverage as well as sentiment. We show that this relationship is robust to time and popularity.

The findings help us understand and analyze print media at scale for the first time providing a key source to complement analysis of digital media.

Contributions.In this paper, we make several key contributions to the study of print media advertising and its influence on news coverage:

In an era of increasing concerns about media bias, our dataset and analysis offer critical insights into the interplay between advertisers and media outlets. By enhancing transparency and understanding of advertising practices and their potential influence on news content, we contribute to the broader discourse on media integrity, informed citizenship, and the health of democratic processes.The methodologies and tools developed in our paper are not limited to the context of (Indian) newspapers but can be adapted to processing structured image data, thus facilitating further research into the dynamics between advertising and media coverage globally.

2. Background and Related Work

Media Bias and Advertising Influence.The media is often regarded as the “Fourth Estate,” underscoring its vital role in upholding democratic societies by providing unbiased and factual reporting. This concept emphasizes the media’s responsibility to act as a watchdog, holding power structures accountable and fostering informed public discourse. According to McChesney(McChesney, 2004), a healthy democracy necessitates a free and independent press that delivers diverse viewpoints without undue influence from external forces. However, maintaining editorial independence has become increasingly challenging due to commercial pressures (Siles and Boczkowski, 2012) and the consolidation of media ownership.Bagdikian(Bagdikian, 2004) highlights how media conglomerates can compromise journalistic integrity by prioritizing profit over the public interest, leading to homogenized content and potential biases. The tension between upholding democratic values and meeting commercial objectives creates a dilemma for media outlets. Hamilton(Hamilton, 2004) explores how economic incentives influence news production, suggesting that market demands often shape what gets reported rather than purely journalistic considerations. In this context, auditing newspapers is crucial to ensure they adhere to ethical standards and continue to serve their role as the Fourth Estate.

The imperative for profitability significantly influences editorial decisions within media organizations. Picard (2011) examines the economics of media companies, highlighting that revenue generation—particularly through advertising—often takes precedence over journalistic ideals. This commercial focus can lead to content that appeals to advertisers or attracts larger audiences, potentially at the expense of investigative journalism or critical reporting.Croteau and Hoynes (2006) discuss how media outlets navigate the balance between maintaining editorial integrity and satisfying corporate interests. They argue that dependence on advertising revenue can result in self-censorship or the avoidance of topics that might alienate advertisers. This balancing act affects the type of news that gets published, as media companies strive to attract and retain advertisers while attempting to uphold the standards of independent journalism. Understanding these dynamics is essential for analyzing how economic factors shape media content and the implications for democratic discourse.

Media Bias Favoring Advertisers.Empirical studies have provided evidence that advertising revenue can influence editorial content, resulting in media bias favoring advertisers. Reuter and Zitzewitz (2006) examined the financial media and found that publications tend to offer more favorable coverage to firms that advertise with them. Their analysis revealed a positive correlation between the amount of advertising purchased by a company and the tone of editorial content it receives, suggesting that editors may compromise journalistic objectivity to maintain advertising relationships.Similarly, DiTella and Franceschelli (2011) investigated the impact of government advertising on media coverage of corruption scandals in Argentina. They discovered that newspapers receiving substantial government advertising were less likely to report on corruption cases involving government officials. This dependence on advertising revenue from influential entities can lead to the suppression or under reporting of negative news about those entities.Similar trends have also been observed in India(Neyazi, 2011; Dasgupta, 2017).

Ownership structures and corporate interests significantly influence media content, often leading to biased reporting. Gentzkow and Shapiro (2010) analyzed U.S. daily newspapers and found that the slant of news coverage is affected by both the ideological preferences of the audience and the ownership of the media outlet. Their study suggests that media owners may cater to audience biases to maximize circulation and profits, resulting in a consistent content slant that aligns with specific viewpoints and potentially limits the diversity of perspectives available to the public.

Gilens and Hertzman (2000) explored how corporate ownership affects news bias by examining newspaper coverage of the 1996 Telecommunications Act. They found that newspapers owned by companies with significant television interests provided more favorable coverage of the Act, which stood to benefit their corporate holdings. This indicates that corporate interests can directly influence editorial decisions, leading to reporting that serves the owners’ financial agendas.Government entities and large corporations often use advertising spending as a tool to influence media coverage, potentially leading to favorable reporting or the suppression of negative news. Prat and Strömberg (2013) discussed the political economy of mass media, emphasizing how governments can manipulate news content through financial means such as advertising budgets. Their analysis highlighted that media outlets reliant on government advertising may avoid negative reporting on governmental actions to secure continued funding, thereby compromising journalistic independence.

Challenges in print media analysis.Obtaining large-scale data from print media poses significant challenges due to limited access, high costs, and the proprietary nature of data held by newspapers or third-party vendors. Gentzkow etal. (2014) highlighted these difficulties in their historical analysis of U.S. newspapers, where data collection involved extensive archival research and manual digitization. The scarcity of digitized archives for print newspapers contrasts sharply with the abundance of data available from digital media, creating a barrier for researchers aiming to conduct comprehensive studies on print media content and its impact.

Puglisi and SnyderJr (2015) discussed the obstacles in assessing media bias and content diversity due to these accessibility issues. The proprietary nature of print media archives often requires researchers to secure expensive licenses or subscriptions, limiting the scope and scale of potential studies. Peiser (2000) noted that declining newspaper readership has led to reduced investments in archiving and data preservation, further exacerbating the problem.Advancements in technology, particularly in Optical Character Recognition (OCR) and computational methods, have begun to mitigate some of the challenges associated with analyzing print media at scale. Smith (2007a) provided an overview of the Tesseract OCR engine, an open-source tool that has significantly improved the accuracy and efficiency of converting scanned print documents into machine-readable text. This development enables researchers to process large volumes of print media content, facilitating quantitative analyses that were previously impractical due to resource constraints.

In this work, we explore these dynamics in a fully automated manner by leveraging physical newspaper data to conduct a systematic analysis of the ad ecosystem and quantify the impact of governmental and corporate advertisements on editorial bias. By employing computational techniques, we aim to analyze large-scale datasets, capturing the nuances of how advertisements affect the tone and volume of media coverage in Indian newspapers. To the best of our knowledge, we are the first to conduct such an analysis using an automated mechanism, and all data and code used in this study are available ²²2https://github.com/harsha20032020/Code_and_Data_for_Analyzing_Patterns_and_Influence_of_Advertising_in_Print_Newspaper to facilitate reproducibility and further exploration by the research community.Our approach extends the existing literature by offering a data-driven, scalable method to assess the role of advertiser influence across various factors such as source and entity type.

3. Data

Our data collection involves scraping the archives of the print newspapers (also known as ‘epapers’) in India.We focus on five major news sources, which are among the country’s most popular: The Times of India,Hindustan Times,and The Telegraph(English sources), as well as Dainik Bhaskar(in Hindi), and Sakshi(in Telugu).Times of India is India’s most circulated English newspaper; Hindustan Times is a close second, and The Telegraph is also widely read and most circulated in the state of West Bengal.Dainik Bhaskar and Sakshi are among the most circulated Hindi and Telugu newspapers. Some of these sources are chosen because of their high circulation and consistent availability of e-papers for the time period of interest.For each source, we crawled e-papers from the respective sites from the timelines mentioned in Table 1. We also collected papers from the same sources from multiple metropolitan cities to understand any regional effect.

The data collected primarily consists of page-level images. However, additional metadata was available for The Times of India and Hindustan Times. This metadata includes bounding box information for different page segments, division of articles and advertisements, and textual data corresponding to articles, though advertisements lacked associated text. This availability of metadata aided us in making a custom image segmentation model for processing other sources. Since the article text was already provided, we only needed to apply Optical Character Recognition (OCR) to the advertisements for these sources.

4. Analysis of the Print Ad Ecosystem

Our dataset offers a unique opportunity to analyze the nature of advertisements in print newspapers at an unprecedented scale. The methodology we developed is highly scalable and applicable to any newspaper, regardless of language or nation, potentially providing much-needed transparency in the print advertising landscape.

We focus on differentiating between government and corporate advertising expenditures.³³3Unless explicitly specified, all the plots and analyses refer to the combined dataset; separate plots are available in the Appendix These two categories of advertisers roughly make up 50% of both the volume and spending on advertising. Advertising on the first, third, and last pages of a physical newspaper holds significant importance due to higher visibility and reader engagement(MediaScience, 2024; Reynolds Journalism Institute, 2024). These pages are considered premium spots and command higher advertising prices.

Keyword Identification.For identifying articles and ads related to governments, we first identify articles related to government corruption using a manually curated set of keywords such as “bribe,” “scam,” “corrupt,” and other relevant terms commonly found in corruption-related articles. Government advertisements are classified by detecting keywords like “government,” “state,” and “tender” in the ad text. In this study, “government” is used specifically to refer to state and national institutions rather than local political entities or parties. The complete set of keywords used to identify both government articles and ads is provided in AppendixG. Our keyword-based approach demonstrates strong performance, achieving an F1 score of 0.94 and an accuracy of 0.94 when validated against 100 advertisement samples, confirming the completeness of our keyword selection.For the corporate analysis, we select a subset of the most prominent advertisers reported in TAM Media Research India’s quarterly reports.⁴⁴4https://tamindia.com/ These companies represent key sectors such as Fast-Moving Consumer Goods (FMCG), Automobile, Technology, Education, and other major industries. Keywords associated with each company are derived from popular products and common terms used to refer to that company. A comprehensive list of the companies and their corresponding keywords can be found in AppendixH. Our methodology assumes that any potential misclassifications at the keyword identification stage are randomly distributed across page numbers and page areas, which are the only two factors that could potentially influence our regression analyses and graphical representations, thus minimizing systematic bias in our findings.Using these keywords, we extract relevant articles and advertisements for each entity, enabling a focused analysis of their advertising patterns.

4.1. Who is advertising?

Figure2 illustrates the absolute spending by government and companies. The data indicates that government entities are significant contributors to print advertising, surpassing corporate spending in total expenditure.Companies account for $890 million in advertising expenditure in our dataset, while government spending reaches $1.02 billion—with the majority of newspaper ads being government-placed. These cost estimates are derived by multiplying the ad’s area with the cost per square centimeter that accounts for multiple factors, including each advertisement’s page area, placement location, page number, and the specific newspaper source, along with the corresponding cost per square centimeter. A detailed breakdown of these variations and their associated rates is provided in Table 6.

4.3. How do they advertise?

To understand the methods and strategies employed by different advertisers, we analyze the sizes and placements of advertisements, which reveal insights into the types of ads provided, the approximate amounts spent, and the fraction of the page area they cover. By computing the size of the ads, we can discern patterns in advertising behavior between government entities and companies.

Figure4 presents the cumulative distribution function (CDF) of advertisement sizes for both government and corporate advertisers. The data reveals a stark contrast between the two groups. Approximately 85% of the ads placed by government entities are small, occupying less than 10% of a page. Full-page ads by the government are exceedingly rare, constituting only about 1% of their total advertisements.In contrast, corporate advertisers display a different pattern. The CDF for companies shows distinct jumps at the 25%, 50%, and 100% marks, corresponding to quarter-page, half-page, and full-page ads, respectively.

Further analysis uncovers additional observations regarding advertising strategies:We document an odd-page bias in advertisement placement. Advertisers prefer to place ads on odd-numbered pages, which are typically the right-hand pages in a newspaper layout. This preference may be an artifact of the way newspapers are read, as right-hand pages are more immediately visible when flipping through the pages. Our data supports this bias, with corporate ads appearing more frequently on odd pages (42,724 instances) compared to even pages (29,393 instances). This finding aligns with industry practices that consider odd-numbered pages as premium spots due to higher reader engagement.

Our dataset also allows us to compute the targeting priorities of various companies. For example, companies like Amazon and Patanjali have been primarily advertising in Hindi-language newspapers, as illustrated in Appendix Figure12. We also observe interesting differences in advertising approaches across national English-language newspapers and regional publications. Companies tend to place more full-page ads in English-language newspapers, as shown in Appendix Figure10.In general, most advertisements are heavily concentrated at smaller sizes, covering between 5% to 10% of a page, as depicted in Figure9 (Appendix). Additionally, clear trends emerge when analyzing advertisement sizes across different industries and sectors. Educational institutions predominantly use full-page ads, as shown in Appendix Figure13.In contrast, insurance companies typically opt for half-page ads (Appendix Figure14), whereas technology companies exhibit clustering around quarter-page, half-page, and full-page ads (Appendix Figure15).

4.4. When are the ads being placed?

To understand the temporal dynamics of advertising in print newspapers, we conducted an analysis of time-based trends, as depicted in Figure5. This figure illustrates the monthly average percentage of newspaper space occupied by advertisements from May 2018 to May 2024.Overall, the data reveals that advertisements consistently occupy approximately 35% of the total newspaper area throughout the years.

However, significant fluctuations are observed during certain periods:In early 2020, coinciding with the onset of the COVID-19 pandemic, there is a marked decline in the percentage of newspaper space devoted to advertisements. This drop reaches its lowest point in April and May 2020, where advertising space plummets to around 10%. The decline can be attributed to the economic uncertainty and reduced consumer activity during nationwide lockdowns, prompting advertisers to cut back on spending.A similar but smaller drop can also be found in June and July 2021, which aligns with the second wave of COVID-19 in India. This period was characterized by severe health impacts and renewed economic disruptions, leading to a temporary reduction in advertising activity.

As the largest English daily, with a readership of over 15 million, Times of India (TOI) consistently has a higher percentage of its space occupied by advertisements, averaging around 40%. This indicates a strong preference among advertisers for TOI, likely due to its wide reach among English-speaking and urban audiences.In contrast, Dainik Bhaskar, a leading Hindi-language daily, exhibits an average advertising space of approximately 20%, nearly half that of TOI.

Despite short-term fluctuations, the overall stability in the proportion of advertising space over the 6 year period suggests that print media remains a vital platform for advertisers. While digital media has grown substantially, print advertising continues to play a significant role, especially in reaching certain demographics and regions where newspapers are a primary source of information.

4.5. What are the ads about?

To understand the content focus of the advertisements, we conducted topic classification on the ad texts. We used an off-the-shelf topic classification model from Hugging Face,⁵⁵5https://huggingface.co/dstefa/roberta-base_topic_classification_nyt_news specifically the RoBERTa-base model fine-tuned on a New York Times dataset.

This model provides classification into eight classes. Evaluation of the classification model can be found in AppendixB. Figure6 presents the distribution of topics in the advertisements compared to the articles. Our analysis reveals several interesting observations. Business, health, and technology ads are significantly higher in proportion than articles in these topics.Conversely, topics such as crime and death, which are more prevalent in articles (including obituaries in the case of ads), are less represented in advertisements.

5. Do advertisers get positive coverage?

Building upon the data and insights presented in the previous sections, we address a pivotal research question: Does advertising in a newspaper influence the coverage that advertisers receive in the news content? Specifically, we aim to explore how advertising revenue from both corporate and government entities may affect the portrayal of news, regarding content bias and the extent of coverage. Leveraging our extensive dataset, we conduct a thorough analysis to examine this potential relationship.

Our investigation focuses on two primary dimensions: the coverage of entities, measured by the number of articles mentioning them, and the sentiment of that coverage, determined through sentiment of the articles. By examining these factors in relation to the advertising expenditure of the entities, we seek to understand whether there is a correlation between advertising spending and favorable media representation.

Inspired by previous literature that explores similar questions in specific contexts –such as the influence of car industry advertising on news coverage(DiTella and Franceschelli, 2011) or the reporting of government corruption(Smet and Vanormelingen, 2012)– our approach builds upon established models from economics. We introduce measurable proxies for both advertising spending and media coverage to facilitate a quantitative analysis.

For advertising spending, we employ the weighted ad ratio as a proxy. This ratio quantifies the prominence and investment in advertisements by considering both the size of the ads and their placement within the newspaper. The ratio is calculated using the following formula:

The scaling factor is computed to indicate the relative importance and cost associated with the placement of an ad. Advertising rates vary depending on factors such as the page number, edition, and city. We obtained ad rates for various editions and page numbers for all the newspapers in our dataset to compute these scaling factors. For instance, if the base rate per square centimeter for an ad in The Times of India is $\$x$, and an ad on the front page costs $\$y$, the scaling factor for front-page ads is calculated as $y/x$. The scaling factors used in our analysis are based on actual rates from the newspapers, as detailed in Table6.A comprehensive description of the costs and scaling factors can be found in AppendixF.⁶⁶6While we could directly use the estimated costs from SectionB, variations in costs across advertisers and the approximate nature of these estimates led us to use the area of the ads, adjusted by scaling factors, as a more robust measure.

By incorporating the scaling factor and normalizing by the page area, the weighted ad ratio allows for meaningful comparisons across different pages and publications, despite differences in page sizes and advertising costs. This normalization ensures that our analysis accounts for the relative prominence and investment in advertisements, rather than absolute measures that may be skewed by publication-specific characteristics.

Our regression analysis examines the relationship between the sentiment and volume of media coverage over a given time period and the weighted ad ratio of advertisements placed by the entities. By focusing on the difference in sentiment and coverage volume, we aim to explore whether increased advertising expenditure correlates with more favorable or extensive media portrayal, without relying on direct financial data that may be proprietary or unavailable.

Sentiment analysis of the articles was conducted using the model from (Camacho-Collados etal., 2022), one of the most widely adopted and downloaded models on Hugging Face providing sentiment classification of the content: -1, 0, or 1 corresponding to negative, neutral, or positive sentiment.Since we saw in Section4 that government spending and types of ads are very different from other corporates, we will do the analysis separately.

6. Conclusion

In this paper, we introduced a unique pipeline for collecting and analyzing advertisements in print newspapers at scale. Our methodology is highly scalable and applicable to any newspaper, irrespective of language or region, opening new avenues for large-scale research on the effects of advertisements in print media. By open-sourcing our code and dataset, we aim to enable researchers and practitioners to further explore the intricacies of print advertising and its impact on media content.

Our comprehensive analysis across multiple languages and newspapers revealed significant insights into the scale and nature of both government and corporate advertising. By examining who advertises, where, when, how, and what they advertise, we uncovered distinct patterns in advertising strategies. Government entities are major advertisers, with a wide distribution of smaller ads across various pages, while companies tend to invest in premium placements and larger ad formats to maximize visibility.

Through our regression analysis, we found a clear, statistically significant correlation between corporate advertising expenditure and both the volume and tone of the media coverage they receive. This suggests that higher advertising spending by companies may correlate with more favorable and extensive coverage in the news content. In contrast, the correlation for government entities was less pronounced, possibly due to the obligatory nature of many government ads, which are legally mandated and broadly disseminated, making it challenging to influence coverage at scale.These findings raise important questions about media bias in favor of advertisers and its potential effects on public discourse, informed citizenship, and democratic processes. The influence of advertising revenue on editorial content underscores the need for transparency and ethical considerations within the media industry.

Our dataset remains highly underutilized, offering rich opportunities for further research. Notably, we possess the actual text of the articles that appeared adjacent to the advertisements, enabling studies on contextual influence and placement effects. Additionally, our dataset serves as a valuable resource for comparative analyses with digital media, allowing for explorations of cross-media advertising strategies and their implications.

References

• (1)
• Bagdikian (2004)Ben Bagdikian. 2004.H., 2004, The New Media Monopoly.
• Beattie etal. (2021)Graham Beattie, Ruben Durante, Brian Knight, and Ananya Sen. 2021.Advertising Spending and Media Bias: Evidence from News Coverage of Car Safety Recalls.Manage. Sci. 67, 2 (feb 2021), 698–719.https://doi.org/10.1287/mnsc.2019.3567
• Camacho-Collados etal. (2022)Jose Camacho-Collados, Kiamehr Rezaee, Talayeh Riahi, Asahi Ushio, Daniel Loureiro, Dimosthenis Antypas, Joanne Boisson, Luis Espinosa-Anke, Fangyu Liu, Eugenio Martínez-Cámara, Gonzalo Medina, Thomas Buhrmann, Leonardo Neves, and Francesco Barbieri. 2022.TweetNLP: Cutting-Edge Natural Language Processing for Social Media.arXiv:2206.14774[cs.CL]https://arxiv.org/abs/2206.14774
• Council (2020)Media ResearchUsers Council. 2020.Indian Readership Survey 2019 Q4.https://mruc.net/uploads/posts/cd072cdc13d2fe48ac660374d0c22a5d.pdf.
• Croteau and Hoynes (2006)David Croteau and William Hoynes. 2006.The business of media: Corporate media and the public interest.Pine forge press.
• Dasgupta (2017)Buroshiva Dasgupta. 2017.Tackling ‘bias’ and fake coverage in the Indian media.(2017).
• DiTella and Franceschelli (2011)Rafael DiTella and Ignacio Franceschelli. 2011.Government Advertising and Media Coverage of Corruption Scandals.American Economic Journal: Applied Economics 3, 4 (October 2011), 119–51.https://doi.org/10.1257/app.3.4.119
• Farooqui (2024)Javed Farooqui. 2024.Strong outlook: Indian print media industry is poised to record robust growth in 2024.https://economictimes.indiatimes.com/industry/media/entertainment/media/strong-outlook-indian-print-media-industry-is-poised-to-record-robust-growth-in-2024/articleshow/107801911.cms.[Accessed 13-10-2024].
• Gala etal. (2023)Jay Gala, PranjalA Chitale, AK Raghavan, Varun Gumma, Sumanth Doddapaneni, AswanthKumar M, JankiAtul Nawale, Anupama Sujatha, Ratish Puduppully, Vivek Raghavan, Pratyush Kumar, MiteshM Khapra, Raj Dabre, and Anoop Kunchukuttan. 2023.IndicTrans2: Towards High-Quality and Accessible Machine Translation Models for all 22 Scheduled Indian Languages.Transactions on Machine Learning Research (2023).https://openreview.net/forum?id=vfT4YuzAYA
• Gentzkow and Shapiro (2010)Matthew Gentzkow and JesseM. Shapiro. 2010.What Drives Media Slant? Evidence From U.S. Daily Newspapers.Econometrica 78, 1 (2010), 35–71.https://doi.org/10.3982/ECTA7195arXiv:https://onlinelibrary.wiley.com/doi/pdf/10.3982/ECTA7195
• Gentzkow etal. (2014)Matthew Gentzkow, JesseM Shapiro, and Michael Sinkinson. 2014.Competition and ideological diversity: Historical evidence from us newspapers.American Economic Review 104, 10 (2014), 3073–3114.
• Gilens and Hertzman (2000)Martin Gilens and Craig Hertzman. 2000.Corporate ownership and news bias: Newspaper coverage of the 1996 Telecommunications Act.Journal of Politics 62, 2 (2000), 369–386.
• Hamilton (2004)JamesT Hamilton. 2004.All the news that’s fit to sell: How the market transforms information into news.Princeton University Press.
• Jocher etal. (2023)Glenn Jocher, Ayush Chaurasia, and Jing Qiu. 2023.Ultralytics YOLOv8.https://github.com/ultralytics/ultralytics
• Johnson and Smith (2023)A. Johnson and B. Smith. 2023.The State of News Media Finances.Columbia Journalism Review (June 2023).
• Lindstädt and Budzinski (2011)Nadine Lindstädt and Oliver Budzinski. 2011.Newspaper vs. Online Advertising–Is There a Niche for Newspapers in Modern Advertising Markets?Online Advertising–Is There a Niche for Newspapers in Modern Advertising Markets (2011).
• Majid (2024)Aisha Majid. 2024.Top 25 US newspaper circulations: Print circulations of largest titles fall 14% in year to September 2023.https://pressgazette.co.uk/media-audience-and-business-data/media_metrics/us-newspaper-circulation-2023/.Accessed: 13-10-2024.
• McChesney (2004)RobertD McChesney. 2004.The problem of the media: US communication politics in the twenty-first century.NYU Press.
• MediaScience (2024)MediaScience. 2024.The Benchmark Series: The Powerful Impact of Ad Placement.MediaScience.https://www.mediascience.com/case-studies/results-the-largest-cross-media-advertising-effectiveness-study-in-australia/
• NewsMediaWorks (2022)NewsMediaWorks. 2022.The Power of Print: Advertising Effectiveness in Newspapers.Research Report. NewsMediaWorks.
• Neyazi (2011)TaberezAhmed Neyazi. 2011.India on television: how satellite news channels have changed the way we think and act.
• PaddlePaddle Contributors (2024)PaddlePaddle Contributors. 2024.PaddleOCR: Awesome multilingual OCR toolkits based on PaddlePaddle (practical ultra lightweight OCR system, support 80+ languages recognition, provide data annotation and synthesis tools, support training and deployment among server, mobile, embedded, and IoT devices).https://github.com/PaddlePaddle/PaddleOCR.Accessed: 14-10-2024.
• Peiser (2000)Wolfram Peiser. 2000.Cohort replacement and the downward trend in newspaper readership.Newspaper Research Journal 21, 2 (2000), 11–22.
• Pew Research Center (2023a)Pew Research Center. 2023a.Local News Preferences in the Digital Age.
• Pew Research Center (2023b)Pew Research Center. 2023b.Newspaper Readership Demographics.
• Picard (2011)RobertG Picard. 2011.The economics and financing of media companies.Fordham Univ Press.
• Prat and Strömberg (2013)Andrea Prat and David Strömberg. 2013.The political economy of mass media.Advances in economics and econometrics 2 (2013), 135.
• Puglisi and SnyderJr (2015)Riccardo Puglisi and JamesM SnyderJr. 2015.Empirical studies of media bias.In Handbook of media economics. Vol.1. Elsevier, 647–667.
• Reuter and Zitzewitz (2006)Jonathan Reuter and Eric Zitzewitz. 2006.Do Ads Influence Editors? Advertising and Bias in the Financial Media.The Quarterly Journal of Economics 121, 1 (2006), 197–227.https://EconPapers.repec.org/RePEc:oup:qjecon:v:121:y:2006:i:1:p:197-227.
• Reuters Institute (2023)Reuters Institute. 2023.Digital News Report 2023.
• Reynolds Journalism Institute (2024)Reynolds Journalism Institute. 2024.The importance of ads on article pages: Balancing revenue and reader experience.Reynolds Journalism Institute.https://rjionline.org/news/the-importance-of-ads-on-article-pages-balancing-revenue-and-reader-experience/
• Siles and Boczkowski (2012)Ignacio Siles and PabloJ. Boczkowski. 2012.Making sense of the newspaper crisis: A critical assessment of existing research and an agenda for future work.New Media & Society 14, 8 (2012), 1375–1394.https://doi.org/10.1177/1461444812455148arXiv:https://doi.org/10.1177/1461444812455148
• Smet and Vanormelingen (2012)Dries Smet and Stijn Vanormelingen. 2012.The Advertiser is Mentioned Twice. Media Bias in Belgian Newspapers.(01 2012).
• Smith (2007a)Ray Smith. 2007a.An overview of the Tesseract OCR engine. In Ninth international conference on document analysis and recognition (ICDAR 2007), Vol.2. IEEE, 629–633.
• Smith (2007b)Ray Smith. 2007b.An Overview of the Tesseract OCR Engine. In ICDAR ’07: Proceedings of the Ninth International Conference on Document Analysis and Recognition. IEEE Computer Society, Washington, DC, USA, 629–633.https://storage.googleapis.com/pub-tools-public-publication-data/pdf/33418.pdf
• Sridhar and Sriram (2015)Shrihari Sridhar and Srinivasaraghavan Sriram. 2015.Is online newspaper advertising cannibalizing print advertising?Quantitative Marketing and Economics 13 (2015), 283–318.
• The New York Times Company (2023)The New York Times Company. 2023.2023 Annual Report.
• Vik Paruchuri (2024a)Vik Paruchuri. 2024a.GitHub - VikParuchuri/surya: OCR, layout analysis, reading order, table recognition in 90+ languages.https://github.com/VikParuchuri/surya.[Accessed 14-10-2024].
• Vik Paruchuri (2024b)Vik Paruchuri. 2024b.GitHub - VikParuchuri/surya: OCR, layout analysis, reading order, table recognition in 90+ languages.https://github.com/VikParuchuri/surya?tab=readme-ov-file#limitations.[Accessed 14-10-2024].

Appendix A Segmentation Model and Outputs

The image segmentation model processes newspapers’ pages and identifies segments that are either articles or advertisements. One such example is given below

The segments are then cropped and processed through the rest of the pipeline. The segmentation model processes the images with confidence and IOU thresholds of 0.1, and then the identified regions are cropped with a margin of 10 pixels. The model and code used for inference will be made public.

We also used 1024 pages for training the model despite having access to larger training datasets, primarily since the performance degraded with larger datasets. Given the importance of a balanced approach in identifying both articles and advertisements accurately, we selected the first model for further analysis due to its superior mAP, which provides a better overall measure of detection performance across all classes.

Appendix B Exploratory analysis plots

Approximately 2.4 million articles and advertisements were identified across the analyzed news sources. A detailed breakdown of the number of articles and advertisements per source is provided in Table 1.To understand the distribution of topics covered in the articles and advertisements, we employed a topic classification model that was fine-tuned using a dataset from The New York Times⁷⁷7https://huggingface.co/dstefa/roberta-base_topic_classification_nyt_news with an accuracy and F1 score of 0.91. This model allowed us to categorize the content into predefined topics, providing a clearer understanding of the thematic focus within articles.

Figure9 shows the distribution of ad ratios across all papers. Most ads are small, occupying less than 10% of the space.

Figure10 shows the distribution of company ad ratios across the 5 papers we study. Companies typically give larger ads. We can see that there is significant distribution around quarter, half, and full-page advertisements.

Figure11 shows the ads by governments. We can see that across sources, the area page ratio is small, indicating that most ads are small in comparison to the page.

Figure 12 presents the distribution of the number of ads among the top 15 companies, broken down by contributions from various newspapers. Companies like Tata and Maruti Suzuki have a broad presence across multiple sources, while others focus their spending more selectively. This chart offers insight into the target audience for their industry.

Certain sectors specifically give certain types of ads. e.g., education sector typically gives larger full-page ads (figure13. Insurance typically gives half-page ads (figure14)

In Figure 17, we can see how both entities spend over time for a source, Times of India. Company spending exhibits clear seasonality, with fluctuations that may correspond to specific periods. Government spending appears more stable over time, with fewer pronounced peaks and stays close to company spending.

We can also observe the raw counts of articles and advertisements for various sources below.

Appendix C Surya OCR in Reading Order

For English Articles, We leverage Tesseract OCR and its Page Segmentation Modes to extract text in reading order, but due to no similar functionality in Surya OCR, We first generate a reading order from the image segment using Surya OCR, then process the image into a manner which is sequentially in reading order and then run Surya OCR on the processed image. A detailed run of the process can be seen in Fig 23

Appendix D Impact of Advertising and Scandals on Sentiment

Figure 24 clearly demonstrates the influence of advertising on sentiment, particularly in how scandals impact conglomerates. This conglomerate was significantly affected by negative media coverage following a scandal in Jan 2023, prompting an increase in advertising efforts to mitigate the fallout. The data highlights the response, which is buying more ads, and gradually, the sentiment shifts following a scandal and gradually returns to pre-scandal levels.

Appendix E Regression Tables

The supplementary panel regressions on interactions with popularity with Corporate ads are provided in Tables 4 and 5.

Appendix F Price Variations and Scaling Factor

Appendix G Keywords To Identify Government-based articles and advertisements

The keywords used to identify Government-based articles and ads can be found in Table 7.Articles are identified if there is a match with any keyword from the corruption-related list and the presence of a government-related term (”government,” ”govt,” ”state,” or ”central”) to ensure government corruption-related articles. In contrast, advertisements are classified if they contain any keyword from the advertisement-related list.

Appendix H Keywords To Identify Companies

The keywords are mostly curated by adding subsidiaries and brands that are uniquely identifiable to a brand.Table8.