Abstract
This study adopts Agenda Setting’s Framing Theory to examine how Singapore newspapers framed one of the nation’s most controversial laws—Section 377A (s377A)—which criminalised consensual sex between male adults. Through a qualitative content analysis of 808 news articles published over a 15-year period, we found that coverage of s377A peaked in 2022, with notable surges in 2007, 2009, 2013, and 2018. The most salient news frame was Legal (5,637 sentences), followed by Societal Morality (5,523) and Religion (4,660), with Medical (1,299) a distant last. These findings revealed that Singapore’s news framing of s377A was influenced by local activities and regional legal developments concerning similar laws. In addition, the two news sources analysed largely reflected the conservative stance of the Singapore government. Beyond these findings, this study makes an important methodological contribution to thematic analysis by employing a mix of inductive and deductive methods. This study also leveraged NVivo’s auto-code function to effectively analyse large textual datasets whilst ensuring reliability through multiple iterative rounds of open coding by human coders. This hybrid approach mitigates the limitations of both fully automated and fully manual coding in content analysis, offering a novel coding method that balances efficiency with interpretive depth.
Keywords
Introduction
Section 377A (s377A) was a controversial Singapore law that criminalised consensual sex between male adults. The law stated that “any male person who, in public or private, commits, or abets the commission of, or procures or attempts to procure the commission by any male person of, any act of gross indecency with another male person, shall be punished with imprisonment for a term which may extend to 2 years” (Penal Code 1871, s. 377A). Crucially, Singaporean courts have emphasised that the law criminalised the homosexual act and not a person’s sexual orientation or identity (Kurohi, 2020).
This law was first introduced to Singapore’s Penal Code in 1938 when the country was under British colonial rule. During the Penal Code Review in 2007, s377A garnered much public attention and debate as both supporters and opponents of the law held strong opinions on why it should be repealed or retained. The Singapore Parliament eventually voted s377A to be retained in the legislation and this decision was justified by then-Prime Minister Lee Hsien Loong, stating that Singapore was generally a conservative society and abolishing the law would not resolve debates on whether homosexuality is acceptable or morally right. Keeping the status quo on s377A, yet not proactively enforcing the law, would therefore help preserve the country’s social cohesion and uphold the traditional, heterosexual family values that most Singaporeans supported (Parliament of Singapore, 2007).
Although s377A was not actively enforced, LGBT (an acronym for lesbian, gay, bisexual, and transgender) groups continued to face strong opposition from local religious organisations in their efforts to advocate for the law’s repeal (Lee, 2016; Ng and Andanari, 2015). Between 2010 and 2019, s377A was also legally challenged for its unconstitutionality in five separate cases. However, all these cases were dismissed and the law remained (Lim and Seet, 2025).
The announcement to repeal s377A was finally made during the National Day Rally on 21st August 2022. Then-Prime Minister Lee Hsien Loong stated that the country’s attitudes towards homosexuality had shifted considerably over the years, with most people now accepting that a person’s sexual behaviour is a private matter. He also conceded that sex between men should not be a criminal offence (Goh, 2024). Subsequently, on 29th November 2022, the Parliament passed the bill to repeal s377A, and the law was officially repealed on 3rd January 2023.
Literature review
Media portrayal of S377A
S377A is a highly contentious issue in Singapore and the discourse surrounding it has garnered significant media attention. Yet, little is known about how the issue has been represented in Singapore’s newspapers. With the media often serving as an expression of a nation’s values or a representation of its collective consciousness (Edelstein, 1982), examining media portrayals of s377A is particularly relevant in the Singaporean context. As a generally conservative city-state with a long-standing ban on LGBT topics, s377A was likely carefully framed by Singapore’s traditional media as such issues were historically deemed too controversial and sensitive for prominent coverage. These framing choices would have shaped how narratives surrounding s377A were constructed and presented to the public.
While audiences do not always passively accept media messages, research shows that the media nonetheless plays a crucial role in shaping public attitudes (Philo, 2008). Studies have demonstrated the influence of the media on liberalising attitudes toward homosexuality by mainstreaming certain narratives and making them more socially acceptable (Andersen and Fetner, 2008; Loftus, 2001). Hence, studying how s377A has been framed (Framing Theory) in Singapore’s newspapers over the years can provide valuable insights into how the media has represented this contentious issue and how it may have shaped public opinion towards LGBT issues in Singapore.
Framing theory
This study adopts Framing Theory as its theoretical framework. According to Entman (1993: 52), “to frame is to select some aspects of a perceived reality and make them more salient in a communicating text.” Salience is central to this process and is defined as “making a piece of information more noticeable, meaningful, or memorable to audiences.” (Entman, 1993: p. 53). Salience can be empirically measured in two ways: (1) by assessing the volume of media coverage over time or (2) through content analysis of news narratives, analysing attributes, keywords, and themes that are emphasised in news stories (Chyi and McCombs, 2004). When specific attributes, keywords, or themes become more salient, they can shape the public’s interpretations of issues, influencing both what people think about and how they think about it (Entman, 2007).
Beyond shaping individual perceptions, media framing also influences the collective understanding and consensus in a community. Nelson and Kinder (1996: 1058) suggest that frames act as a “mental recipe” to help the general population grasp complicated issues, by which communities can form similar opinions. However, media frames are not neutral. The framing process is often influenced by the norms and values of newsmakers, which can shape how certain groups or issues are represented through deliberate “selection, emphasis, and exclusion” (Gitlin, 1980: 7). In other words, frames serve as interpretive tools that organise news messages, shaping how issues are reported and discussed in the news. This makes framing a potent tool for influencing public opinion, particularly on divisive topics such as LGBT, where the media can either challenge or reinforce existing biases. Thus, Framing Theory provides a valuable lens to analyse how Singaporean newspapers have thematically framed s377A, shedding light on the salient narratives surrounding the law.
Common news frames associated with homosexuality
Prior research on the media framing of LGBT topics and homosexuality has been examined across diverse sociopolitical environments, including Muslim-majority nations such as Malaysia, Lebanon, Indonesia, and Turkey (Adamczyk et al., 2017; Ting et al., 2023), Protestant-majority nations such as the United States, South Africa, Uganda, and Ghana (Adamczyk et al., 2017; Baisley, 2015; Nartey, 2022), more liberal European contexts such as Belgium (Jacobs and Meeusen, 2020), as well as atheist state such as China (Chang and Ren, 2017). These studies consistently identify several recurring frames: morality, religion, crime, and rights frames. Firstly, the morality frame tends to portray homosexuality as a violation of ethical and social norms. News stories using this frame associate homosexuality with sexual morality issues and non-normative behaviours, such as drug abuse and sex parties (Adamczyk et al., 2017), drawing parallels between homosexuality and detrimental social conduct. LGBT individuals are also depicted as immoral or engaging in deviant sexual behaviours, posing threats to societal heterosexual values (Baisley, 2015; Chang and Ren, 2017; Nartey, 2022). Secondly, the religion frame is frequently used to argue against homosexuality by referencing religious beliefs. Such news articles cite teachings from leading religions, including Christianity, Catholicism, and Islam, to condemn homosexual behaviours as sinful or contrary to traditional religious values (Adamczyk et al., 2017). Crime frames commonly appear in coverage of police actions, such as the detention of individuals suspected of engaging in gay sex or participating in LGBT parades and marches (Ting et al., 2023). This frame may also depict gay people as victims of crimes, often attributing their victimhood to inherent weaknesses or flaws of homosexuality, such as promiscuity, cowardice, or shame of one’s sexual orientation (Chang and Ren, 2017). Finally, contrary to the negative portrayal of LGBT groups in the previous frames, the rights frame emphasises LGBT individuals as legitimate members of society fighting for recognition and equal opportunities (Jacobs and Meeusen, 2020). This frame tends to parallel LGBT individuals’ struggles for equal rights with those of other marginalised groups, such as racial minorities or persons with disabilities (Baisley, 2015).
Country-level factors that influence news framing of homosexuality
It is also important to note that the media framing of homosexuality varies considerably across national contexts, influenced by factors such as a country’s dominant religion, level of economic development, and degree of democracy. In Adamczyk et al.’s (2017) analysis of seven nations with high levels of religiosity (Muslim vs Protestant), as well as varying levels of economic development and democracy, they found that while Muslim- and Protestant-majority nations were similarly likely to frame homosexuality in relation to religion, Muslim-majority nations were four times more likely than Protestant-majority nations, to frame it in the context of morality. The study also found that nations with lower gross domestic product (GDP) are more likely to frame homosexuality as a moral issue, whereas more democratic nations typically frame it in the context of human rights (Adamczyk et al., 2017).
Singapore, being one of the most religiously diverse country in the world with no single dominant faith (Liu, 2014), together with its rapid economic growth over the past decades (Teutem, 2025), and its highly regulated media environment that primarily promotes state-approved perspectives (Reporters Without Borders, 2025), presents a unique context for studying the media framing of LGBT issues. The distinctive combination of religious diversity, economic prosperity, and the state’s tight media regulation raises critical questions about how LGBT topics, particularly the contentious s377A, have been framed in its media. Additionally, analysing the evolution of media frames over a prolonged period is significant given the notable legislative challenges surrounding LGBT rights in Singapore over the 15 years leading up to the official repeal of s377A. This study also investigates whether the framing of s377A varied across different Singaporean newspapers, which can provide valuable insights into the diversity—or lack thereof—of perspectives on LGBT issues presented to the public. Hence, the objectives of this study are (1) to analyse the thematic framing of s377A, tracking its evolution across the years; and (2) to compare the coverage of s377A across local newspapers. It seeks to answer the following research questions:
Research method
Sample selection
Two leading local newspapers, The Straits Times (ST) and TODAY (TY), were selected for this study due to their high readership and extensive circulation in Singapore, with both ranking among the top five news sites in terms of weekly online reach in a 2024 study by the Reuters Institute (Tandoc and Chew, 2024). ST has the highest offline reach and the third-highest online reach in the country, accessed by over 35% of Singapore residents. With a long history dating back to 1845, ST is regarded as the most trusted news source in Singapore (Tham, 2025), renowned for its comprehensive coverage of national issues (Ho, 2016). On the other hand, TY was ranked 5th amongst the most widely read online newspapers in Singapore with about 2.9 million monthly unique visitors to its website (TODAY, nd). With a digital-first orientation, TY reaches a broad and diverse audience, including younger readers. Together, these two highly influential newspapers with substantial readership provide a robust basis for a comprehensive examination of how local mainstream media have framed s377A.
Despite Singapore’s linguistic diversity, this study focused on English-language newspapers as English remained the most frequently spoken language at home in Singapore, used by 48.3% of the population (Department of Statistics Singapore, 2021). Moreover, prior research has established precedent for focusing on English-language newspapers in analyses of media framing in Singapore (e.g., Phillips, 2021; Tandoc et al., 2025).
Articles from both ST and TY were sampled from the period of 22nd October 2007 (which corresponds to the first day where the issue was extensively raised in the Parliament) to 1st December 2022 (which extends slightly beyond the repeal decision made on 29th November 2022 to capture immediate post-decision coverage). This study opines that this period would be appropriate for studying the framing effect of newspaper reporting. The search keyword “377A” was used in Factiva, and “identical duplicates” were removed to ensure the integrity of the sample and to cut away waste. This resulted in 808 relevant articles from both ST and TY.
Development of themes
Adopting the thematic analysis steps as outlined by Braun and Clarke (2006), this study started with the authors heuristically reading news articles about s377A. The authors also shortlisted two strategic Parliamentary Reports (transcripts of the Penal Code Review on 22nd and 23rd October 2007) for reading to understand the topic and data—this is consistent with Braun and Clarke’s (2006) first step of familiarising with the data. The two Parliamentary Reports were chosen as they dealt extensively with the issue of s377A.
Careful line-by-line reading of the Parliamentary Reports by the authors resulted in the emergence (inductive approach) of themes, which were later meticulously reviewed. This process, as prescribed by Braun and Clarke (2006), enabled the authors to come up with four major themes, as well as their associated keywords, in the discussion of s377A: Medical, Legal, Societal Morality, and Religion. The next section describes the themes in detail and their associated keywords (some of which emerged during the coding process—see the section on “Coding” below).
Medical
Good physical health refers to the proper functioning of all biological components (Amzat and Razum, 2014), and good mental health refers to having a “sense of coherence,” which is the ability to understand, manage, and find meaning in situations (Antonovsky, 1979: 123). This thematic scope encompasses any references to medical and psychiatric conditions that cause harm to an individual’s physical and mental well-being. This includes any mention of diseases, illnesses, disorders, and mental illnesses. A sample of keywords associated with the Medical theme is AIDS, Biology, Disease, Gene, Health, HIV, Medical, and STD.
Legal
A scheme of social control, backed and sanctioned by the power of the state, for the protection of social interests, by means of legal capacities and legal redress (Willis, 1926). This thematic scope encompasses any references to the sources of law in Singapore and the legality of 377A. This includes any allusion to the constitution, legislation, subsidiary legislation, judge-made law, and the various organs of the state. A sample of keywords associated with the Legal theme is Constitution, Criminalise, Enforce, Imprison, Legal, Legislation, Policy, Political, Prosecute, Punish, Regulation, and Rule.
Societal Morality
A set of shared and accepted social conventions about right and wrong human conduct (Horner, 2003). This thematic scope encompasses any references to societal norms and the prevailing morality of the population. This includes any mention of societal constructs, including marriage and family, traditions, and prevalent values and beliefs. A sample of keywords associated with the Societal Morality theme is Community, Conservative, Conventional, Culture, Family, Marriage, Moral, Norm, Social, Society, Tradition, and Value.
Religion
A shared set of beliefs and actions related to powerful supernatural entities (Frankenberry, 2014). This thematic scope encompasses any references to religious matters and concepts. This includes any mention of religious faiths, organisations, institutions, beliefs, practices, and activities. A sample of keywords associated with the Religion theme is Belief, Buddhist, Catholic, Christian, Church, Faith, Hindu, Islam, Muslim, Pastor, Priest, Religion, Secular, Sikh, and Taoist.
Coding
Due to the large volume of sampled articles, this study relied on the algorithm (auto-coding) afforded by the coding software NVivo. Reliance on auto-coding is not new, especially for large sample sizes. For example, Kaefer et al.’s (2015) study relied on auto-coding to handle their sample of 230 articles. Hence, this study did the same.
Generation of keywords
Keywords under each of the four main themes.
Algorithm training
With the 47 keywords, auto-coding function was activated to code sentences that contain the root keywords into the various themes. For NVivo, there was a choice on how strictly the algorithm was to operate. Less coding option would mean the algorithm adopts a “higher threshold to determine similarity when comparing each text passage to each code.” More coding option would mean the algorithm adopts a “lower threshold to determine similarity when comparing each text passage to each code”—this would typically result in more sentences being coded. This study chose the option between Less and More—the average option—in order to have a balance between having more sentences (to train the algorithm) and ensuring the relevancy of the sentences being coded. With this option, the data went through the first round of auto-coding (under the pattern-coding function). Thereafter, the authors reviewed 10 auto-coded sentences per keyword. These 10 auto-coded sentences (per keyword) were randomly chosen, and after a review by the authors, were appropriately coded further or removed. As both authors were conducting the analyses and reviewing the coding together, the calculation of inter-rater reliability was deemed unnecessary.
Therefore, in the first round of auto-coding, the authors manually coded 470 sentences (47 keywords x 10 sentences per keyword). The selection of sentences was based on their contextual accuracy to the keywords. During this process, some keywords were also reassigned to more appropriate themes. For example, reviewing sentences with the keyword ‘belief’ showed that it was more commonly used in the context of religious beliefs rather than traditional societal beliefs. As a result, ‘belief’ was reassigned from Societal Morality theme to Religion theme. The algorithm was instructed to learn from the refreshed schema before running the auto-coding process again in the same format. In the second round, the authors reviewed 10 randomly selected auto-coded sentences per keyword, removing any sentences that were irrelevant. For instance, when examining sentences for the keyword “AIDS”, it was found that while some sentences referred to the medical condition (AIDS), others used “aids” to mean assistance. Thus, sentences that did not match the intended keyword definition were removed. One final round of auto-coding was conducted, in which the algorithm was set to a high threshold (More coding option) to allow for more sentences being coded now that the algorithm is more mature through training. This resulted in 17,119 references (sentences) being coded into the four themes.
Findings & discussion
Number of coded sentences under each of the four main themes by year.
Framing influenced by local and regional activities
The authors found that the issue of s377A was featured most frequently in 2022, with peaks also found in 2007, 2009, 2013, and 2018. The peaks observed in 2007 and 2022 were unsurprising as they coincide with the parliamentary debate over the repeal of s377A—this shows that Singapore's broadsheets are typically aligned with the agenda of the parliamentary debates. The peaks in 2018, 2013, and 2009 were closely examined by returning to the data to identify the causes of these spikes. The authors found that India repealed a similar law sometime in September 2018 (Safi, 2018), which accounted for many articles in Singapore having the same agenda in 2018. Also in September 2018, high-profile politicians in Malaysia (the closest neighbouring country to Singapore) discussed the legality of their sodomy laws, recognition of LGBT culture, and same-sex marriages in the country (The Straits Times, 2018). In late 2012, there was a legal contest to s377A locally, which accounted for widespread discussions in the newspapers and consequently led to a peak in 2013. And in 2009, there was the AWARE Saga in Singapore. AWARE (Association of Women for Action and Research), which is a prominent non-profit organisation that advocates for gender equality and LGBT rights, underwent a sudden leadership change from March to May 2009 (Nair, 2019). It was revealed then that many members of the new executive committee were from the same church and espoused anti-homosexual views. Their takeover of AWARE was aimed at steering the organisation towards a more conservative direction. While the takeover only lasted 2 months, this episode drew extensive local media coverage in 2009 and renewed discussion around homosexuality in Singapore and s377A.
The authors also found that Legal (5,637) was the most prominent theme, followed closely by Societal Morality (5,523), then Religion (4,660). Medical (1,299) came in a distant last. The findings suggest a few things. Firstly, considering Singapore’s conservative cultural climate and its penchant for religious harmony, it is no surprise that Legal and Societal Morality frames are more prominent. The prominence of Legal frames may be due to news articles introducing the legality of s377A before delving into other aspects. In addition, as discussed above, the framing of s377A in Singapore was influenced by regional legal developments concerning similar laws. Meanwhile, the prominence of Societal Morality frames may be reflective of the state’s largely conservative leaning, given that the repeal of s377A was accompanied by a constitutional amendment to protect the definition of marriage (Leo, 2022). This suggests that the state has consistently prioritised “family values and social norms” over the years (Ang, 2022).
It is also important to note that Singapore’s news industry operates largely as a duopoly between Mediacorp and SPH Media Trust (Tandoc and Chew, 2024). ST is operated by the government-funded SPH Media Trust, while TY is operated by the state-owned Mediacorp. Thus, the editorial positions from these two state-run news outlets may have been postured to avoid or prioritise certain news frames in relation to discussions of s377A. This may help explain why discussions of s377A from these two newspapers tend to focus on Legal and Societal Morality frames, which are largely reflective of Singapore’s conservative stance.
However, it is surprising that Religion theme follows closely behind and Medical theme came in a distant last. The authors would have expected that a robust discussion in Singapore’s public sphere would largely centre around Legal, Societal Morality, and Medical frames, leaving the sensitive Religion frame out of the discourse. That said, it is to be expected (consistent with studies in our literature review) that religious groups will be vocal about homosexuality and s377A. This could indicate an evolving reporting landscape where religious groups or persons—for example in Lee (2016), where reporting is not accompanied by any scriptural text—being involved in social issues is no longer considered taboo. Although this could also suggest that Singapore’s government-leaning newspapers prioritise robust debates, do not steer away from sensitive issues, and allow religious perspectives to be expressed. And by extension, the result of the nation having greater political processes and a pluralistic posture, which is a key dimension of democracy (Adamczyk et al., 2017).
Medical theme being a distant last in Singapore’s news media framing of s377A is unexpected given the long history of equating homosexual sex with AIDs (Santa and Bell, 2006). A few reasons may explain this low prominence. Firstly, this could be the result of sampling bias that led to some broad medical discussions about homosexuality being excluded. For example, a 2019 Straits Times article that mentioned HIV and homosexuality (Lau and Yong, 2019) was not captured in our search because it did not include the keyword ‘377A’. Another possible reason could be due to the strong backlash faced by a Nominated Member of Parliament following the 2007 parliamentary debate, during which the individual raised multiple medical discussions supporting the retention of s377A (Ng, 2007). This could have muted subsequent medical discussions on s377A, and therefore, lowered the presence of Medical framing in local news coverage.
Comparison of framing between ST and TY
The authors also did a comparison between how ST and TY framed the issue. In such comparison studies, it would be appropriate to work with percentages by year as a measure of salience, instead of the number of sentences by year as above. For example, in 2007, there were 83 sentences under the Legal theme by TY, and a total of 238 sentences in the four themes. Therefore, the percentage of Legal frames is 83 divided by 238, giving 34.87%. This means that calculations were done to ascertain the salience of the theme within TY in a specific year. Figures 1 and 2 below show the percentage of each frame across the years. Salience of themes across the years. Relative salience of themes for TY and st across the years.
The authors found that both newspapers showed similar trends over the years. For instance, Legal theme taking the most salient position in many years. The findings show a large gap in ST’s feature of the Medical theme, compared to other themes within ST, and compared to TY’s too. However, the authors opine that the gap is not significant enough to draw any further conclusions other than those already explained above.
Methodological contributions
Importantly, beyond the thematic differences, the main strength and academic contribution of this paper lie in its methodological approach. Content analysis can be conducted in various ways. Some studies rely entirely on manual coding by human coders (e.g. Baysha and Hallahan, 2004; Jacobs and Meeusen, 2020; Rahmadan and Setiawati, 2021; Ting et al., 2023), while others employ full auto-coding using natural language processing and machine learning algorithms integrated in qualitative data analysis software (QDAS) (e.g. Burscher et al., 2016). This study adopts a combination of manual and auto-coding to thematically analyse a large sample size of 808 news articles. QDAS-assisted content analysis is not uncommon and has been adopted in varying forms in previous research (e.g. Kaefer et al., 2015; Schotel, 2023). However, what sets this study apart is first, the use of both deductive and inductive coding methods, which has been coined by Kaefer et al. (2015: 6) as “multi-level coding process,” and second, the training of QDAS machine learning algorithm through multiple iterative refinements, which is a form of supervised machine learning method (Grimmer and Stewart, 2013). With this unique approach, this study has found ways to further establish the integrity, robustness, and validity of using QDAS to assist content analysis.
As detailed in the methodology section, the authors have identified the four main themes and their associated keywords for content analysis on NVivo using a combination of inductive coding (heuristic reading of relevant news articles and Parliamentary Reports and running word cloud analysis across the entire dataset) and deductive coding (fitting keywords into the four main themes). This process parallels Kaefer et al.’s (2015) study of 230 newspaper articles on international media perceptions of New Zealand’s environmental performance. In their approach, keywords were first generated inductively through word frequency queries, with references for each respective keyword extracted and saved through text search queries. Open coding was then conducted with the authors opening the references saved for each keyword, reading the content line-by-line, and coding salient themes deductively. This multi-level coding process is akin to what this study did, but the approach was extended by conducting three rounds of open coding, as compared to their single round. Crucially, this study applied the multi-level coding approach to a much larger dataset of 808 news articles than theirs of 230.
This coding approach also builds on Schotel’s (2023) study, which examined how German and Dutch newspapers framed lesbian, gay, bisexual, transgender, and intersex (LGBTI) people across a 10-year period. Schotel (2023) analysed an extensive dataset of 15,877 news articles using a combination of auto-coding and manual coding. Auto-coding was conducted through Latent Dirichlet Allocation (LDA) topic modelling, a type of unsupervised machine learning algorithm aimed at discovering thematic structure inductively in large datasets without predefined keywords (DiMaggio et al., 2013). This exploratory process was complemented with critical frame analysis, where Schotel (2023) randomly selected samples of news articles and coded them deductively into a set of predefined frames, allowing for a more nuanced interpretation of the topic modelling results. Schotel’s (2023) approach is principally aligned with what was done in this study. However, beyond the methods of Schotel (2023) and Kaefer et al. (2015), this study conducted multiple rounds of open coding during the auto-coding process in NVivo. During these rounds of iterations, keywords were regrouped or excluded accordingly before the analyses were re-run. By incorporating this form of supervised machine learning, this study was able to improve the quality of the training data, enhance the integrity and robustness of the algorithm and, by extension, the validity of the analysis outputs.
In comparison to studies that rely on full auto-coding (e.g. Burscher et al., 2016), this study's multi-level coding approach may be less efficient for handling very large datasets. Burscher et al.’s (2016) study on news framing in the nuclear power debate is an example of content analysis that relies entirely on auto-coding. Using k-means cluster analysis and automatic sentiment analysis, Burscher et al. (2016) conducted thematic analysis on 4,286 news articles. This inductive approach allowed the algorithm to identify frames and define frame elements with minimal human intervention. While this method is highly effective for large datasets and may mitigate issues of potential human bias, yet leaving frame identification entirely to the algorithm risks overlooking subjective meanings and interpretive nuances present in the content. This is because machines cannot fully grasp the contextual subtleties that human coders can perceive, which can also result in data points being repeatedly coded into multiple themes. Furthermore, the results of Burscher et al.’s (2016) study depended on extensive pre-processing and carefully established coding rules, showing that a fully auto-coded process still requires significant human input and oversight.
Accordingly, the methodological approach employed in this study—a multi-level coding process with multiple iterative refinements of the algorithm during the auto-coding process—demonstrates how QDAS can be effectively integrated into content analysis with large datasets whilst ensuring the validity and reliability of auto-coding. This study's approach seeks to balance efficiency with interpretive depth, mitigating the limitations of both fully automated and fully manual methods.
Conclusion
This study relied, to some measure, on the algorithm to code, which presents a few limitations that have already been explored extensively in the discussion section. Firstly, in large qualitative content analysis studies, auto-coding might miss coding some content. This discrepancy can affect the comprehensiveness of the analysis. Next, auto-coding does not consider context. It cannot recognise sarcasm, slang, idioms, or ambiguity. Context matters, and auto-coding does not take it into account. This may have impacted the Legal theme, as an article could introduce the legality of s377A, but wholly frames it from other aspects. Additionally, only “377A” was used to search for news articles. News discussions that broadly cover homosexuality and/or legal debates that do not mention “377A” could have added value to the study if there were deeper contextual coding done to confirm the articles’ relationship to s377A. However, since this study relied heavily on algorithmic coding, attribution of news articles to s377A would be strengthened only if “377A” was used. Future studies could look at expanding the search words to include, for example, “homosexual*”, if more contextual analysis were undertaken. Moreover, themes were originally generated inductively using Parliamentary Reports. Future studies could look into building themes based on newspaper reports too. The study acknowledges that the exclusion of news sources in other languages, such as Mandarin, Malay, and Tamil, may have biased the findings towards English-speaking communities. The selection of two state-run news outlets may have also influenced the range of frames present in the dataset, and as a result, led the findings to largely reflect the perspectives commonly represented within mainstream media. A more comprehensive analysis would therefore benefit from the inclusion of non-English news sources, as well as a broader range of news outlets including both state-run and independent news sources. Finally, a common concern with auto-coding is the risk of “over-coding” (Kaefer et al., 2015: 16), which is the tendency to repeatedly run analyses in search of new patterns simply because the software makes it easy to do so. However, in this study, the multiple rounds of analyses were not instances of over-coding but a deliberate process of fine-tuning the algorithm to enhance the accuracy and reliability of the findings.
This study represents the first comprehensive examination of news coverage related to this controversial topic in Singapore. The findings indicate that the media has, to some extent, prioritised robust public debate over the maintenance of religious sensitivities, though additional sentiment analysis would be required for further validation. Furthermore, this research introduces an effective approach for coding a large volume of articles through algorithm training, offering a significant methodological contribution to the framing literature.
Footnotes
Declaration of conflicting interests
The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The authors received no financial support for the research, authorship, and/or publication of this article.
Ethical considerations
Ethics Approval was not required as data was collected from publicly available newspaper sources and did not deal with any humans.
Data Availability Statement
Data is not available