前期出版
前期出版
頁數：55﹣112
臺灣報紙中空污風險的新聞框架： 跨時的演變及其內涵
News Framing of Air Pollution Risks in Taiwanese Newspapers: A Longitudinal Study of Changes
研究論文
作者(中)
譚躍、蘇靖雅
作者(英)
Yue Tan, Ching Ya Su
關鍵詞(中)
不確定性、主題、框架、時間序列分析、情感、電腦內容分析
關鍵詞(英)
air pollution, computer-assisted content analysis, frames, time series analysis, tone, topics, uncertainty
中文摘要
臺灣近十年來，空污相關的健康及環境問題不斷惡化，但人們對它的理解高度仰賴媒體的建構。為理解臺灣媒體近年來對空污議題建構的內容和可能的影響，本研究使用電腦內容分析法，考察2017至2021年臺灣三大報《聯合報》、《蘋果日報》、《自由時報》空污報導的數量、主題、情感方向、不確定性及其跨時性的變化。本研究使用時間序列分析，考察空污報導的數量和內容趨勢，並探討實際空污程度對空污報導的影響。發現媒體報導量並未隨著空污總體的改善而減少，反而升高；報導語氣有稍微惡化的趨勢；報導內容中的不確定性總體維持不變。本研究使用Entman（1993）對於框架功能的四個分類：定義、原因、解決方案和道德判斷，來歸納、對比和討論這五年媒體報導中主題內容的變化趨勢。研究發現三大報較少報導空污的健康危害，且逐年減少。對於污染源的討論逐漸縮小到了對農／工業污染、火災和汽機車，忽略了營建業這個最大的污染源。三大報雖大量討論空污的解決方案，但都與政府政策相關，反而使得與政治選舉相關的主題逐年增加。最後，本研究結合文獻，針對每一個主題的報導數量、情感分數、不確定性的變化和對社會可能的影響進行了討論。
英文摘要
The 2020s have witnessed a steady increase in the rates of haze-related diseases in Taiwan. Indeed, German sociologist Ulrich Beck (1992) criticized current development in societies that has produced unintended and unforeseen side-effects to modern life. He described that the new risks are historically unprecedented, invisible, technologically sophisticated, and highly uncertain.
Social knowledge of contemporary environmental risks, independent from reality, is mainly acquired from mass media. Social constructionists argue that the media influence what and how things come to be defined as risks and facility developments of risk consciousness (Kitzinger, 1999; Beck, 1992). Therefore, to understand how Taiwanese society responds to haze-related risks, it is important to first know how news media report these risks and their consequences. The answers to these questions are not only needed to understand how media influence public understanding of the air pollution issue, but also to help researchers make more specific suggestions on how to promote public awareness of air-related risks and protection behaviors through news media in Taiwan.
The concept of framing offers a powerful framework for understanding how a news report provides the “schemata of interpretation” that enable individuals to make sense of an otherwise meaningless succession of public events (Goffman, 1974). To frame is to make a persistent selection, emphasis, and exclusion (Gitlin, 1980). Consistent with this emphasis-based definition, Entman (1993) proposed a classical definition of framing with a focus on problem definition, causal interpretation, moral evaluation, and/or treatment recommendation for the issue described. As suggested by van der Meer (2018), we use these four functions to organize our analysis and discuss the topic themes that we obtain from unsupervised machine learning. Because uncertainty is the core component of risks (黃俊儒, 2014) and affective heuristics are the key determinant of risk perception and risk prevention behaviors (Nabi et al., 2020), both uncertainty and tone are analyzed in terms of their trends and their relationships with topic prevalence.
The study examines air-pollution news from three major newspapers in Taiwan, including United Daily, Apple Daily, and Liberty Times from 2017 to 2021. Their news articles have been downloaded from their websites with web scrapers. The keywords for searching include “air quality” and “air pollution”. After manual cleaning of irrelevant articles, the total number of articles is 8,509.
Automated content analysis of newspaper articles is conducted to measure media coverage in terms of their volumes, topics, tone, and uncertainty. Tone and uncertainty are measured with established dictionaries (CLIWC, Linguistic Inquiry, and Word Count). The topical themes are measured with unsupervised machine learning (topic modeling), whose algorithms learn hidden clusters (topics) in text data. A typical topic model observes word frequencies in each document in terms of a suitable weighted mixture of topical word frequencies where the weights indicate the different proportions of topics that appear in the document (Guo et al., 2016). Guo et al. (2016) found that LDA-based analysis performs better than a dictionary-based approach in many aspects. In this study we employ the most widely used topic model algorithm, Latent Dirichlet Allocation (LDA), with the R package of STM. Ten topics are chosen according to the four criteria from the R package of ldatuning.
Current conditions for air pollution are operationally defined by 24-hour-average PM2.5 concentrations for each day. Such data are publicly available from the official websites of the Environmental Protection Administration, Executive Yuan of Taiwan (www.epa.gov.tw) and the U.S. Environmental Protection Agency (www.epa.gov). Among all kinds of harmful airs (e.g., PM2.5, PM10, O3, CO, SO2, NO2), we choose PM2.5 as the main indicator, because scholars use it the most often (Apte et al., 2018; Hayes et al., 2020; Colmer et al., 2020), and governments as the main index for overall air quality.
Trend analyses are first performed to examine the linear and non-linear trends in term of news coverage volume, tone, and uncertainty with the R package of Forecast. The R package of STM, with the prevalence of each topic as a dependent variable, allows us to model the main effects of time, tone, uncertainty, and their interaction. Statistically, time series analysis (Vector Autoregressions Model, VAR) and Granger causality tests are conducted to examine the relationship between news media coverage and real-word condition of PM2.5 concentrations. The R package Var is then used to automatically determine the best time lag for each agenda-setting relationship. The time unit is set at one day. The analysis is conducted for each shared topic yielded by LDA.
Findings show that the amount of media coverage does not decrease with the overall improvement of air pollution, but rather increases. The tone of a report deteriorates slightly over time. However, uncertainty in the content of the report remains generally unchanged.
Employing Entman’s (1993) definition of frame functions (problem definition, causal interpretation, treatment recommendation, and moral evaluation), this study summarizes, compares, and discusses trends in the topical themes over the past five years. The results present that newspapers only devote a small proportion of coverage to the health hazards of air pollution, and this proportion has been decreasing year by year. The discussion of pollution sources gradually narrows down to industrial and agricultural pollution, fires, and transportation, ignoring pollution from the construction industry.
Although the three major newspapers have discussed numerous solutions to air pollution, the solutions all relate to government policies, which has led to a rise in topics related to political elections year by year. The politicization of air-pollution news may lead the public to expect solutions from governmental regulations and even party elections, ignoring the seriousness of health risks and reducing their willingness to take individual protective actions (Adger, Quinn, Lorenzoni, Murphy & Sweeney, 2012). Finally, based on the literature, this study discusses the theoretical and practical implications of the results.
Social knowledge of contemporary environmental risks, independent from reality, is mainly acquired from mass media. Social constructionists argue that the media influence what and how things come to be defined as risks and facility developments of risk consciousness (Kitzinger, 1999; Beck, 1992). Therefore, to understand how Taiwanese society responds to haze-related risks, it is important to first know how news media report these risks and their consequences. The answers to these questions are not only needed to understand how media influence public understanding of the air pollution issue, but also to help researchers make more specific suggestions on how to promote public awareness of air-related risks and protection behaviors through news media in Taiwan.
The concept of framing offers a powerful framework for understanding how a news report provides the “schemata of interpretation” that enable individuals to make sense of an otherwise meaningless succession of public events (Goffman, 1974). To frame is to make a persistent selection, emphasis, and exclusion (Gitlin, 1980). Consistent with this emphasis-based definition, Entman (1993) proposed a classical definition of framing with a focus on problem definition, causal interpretation, moral evaluation, and/or treatment recommendation for the issue described. As suggested by van der Meer (2018), we use these four functions to organize our analysis and discuss the topic themes that we obtain from unsupervised machine learning. Because uncertainty is the core component of risks (黃俊儒, 2014) and affective heuristics are the key determinant of risk perception and risk prevention behaviors (Nabi et al., 2020), both uncertainty and tone are analyzed in terms of their trends and their relationships with topic prevalence.
The study examines air-pollution news from three major newspapers in Taiwan, including United Daily, Apple Daily, and Liberty Times from 2017 to 2021. Their news articles have been downloaded from their websites with web scrapers. The keywords for searching include “air quality” and “air pollution”. After manual cleaning of irrelevant articles, the total number of articles is 8,509.
Automated content analysis of newspaper articles is conducted to measure media coverage in terms of their volumes, topics, tone, and uncertainty. Tone and uncertainty are measured with established dictionaries (CLIWC, Linguistic Inquiry, and Word Count). The topical themes are measured with unsupervised machine learning (topic modeling), whose algorithms learn hidden clusters (topics) in text data. A typical topic model observes word frequencies in each document in terms of a suitable weighted mixture of topical word frequencies where the weights indicate the different proportions of topics that appear in the document (Guo et al., 2016). Guo et al. (2016) found that LDA-based analysis performs better than a dictionary-based approach in many aspects. In this study we employ the most widely used topic model algorithm, Latent Dirichlet Allocation (LDA), with the R package of STM. Ten topics are chosen according to the four criteria from the R package of ldatuning.
Current conditions for air pollution are operationally defined by 24-hour-average PM2.5 concentrations for each day. Such data are publicly available from the official websites of the Environmental Protection Administration, Executive Yuan of Taiwan (www.epa.gov.tw) and the U.S. Environmental Protection Agency (www.epa.gov). Among all kinds of harmful airs (e.g., PM2.5, PM10, O3, CO, SO2, NO2), we choose PM2.5 as the main indicator, because scholars use it the most often (Apte et al., 2018; Hayes et al., 2020; Colmer et al., 2020), and governments as the main index for overall air quality.
Trend analyses are first performed to examine the linear and non-linear trends in term of news coverage volume, tone, and uncertainty with the R package of Forecast. The R package of STM, with the prevalence of each topic as a dependent variable, allows us to model the main effects of time, tone, uncertainty, and their interaction. Statistically, time series analysis (Vector Autoregressions Model, VAR) and Granger causality tests are conducted to examine the relationship between news media coverage and real-word condition of PM2.5 concentrations. The R package Var is then used to automatically determine the best time lag for each agenda-setting relationship. The time unit is set at one day. The analysis is conducted for each shared topic yielded by LDA.
Findings show that the amount of media coverage does not decrease with the overall improvement of air pollution, but rather increases. The tone of a report deteriorates slightly over time. However, uncertainty in the content of the report remains generally unchanged.
Employing Entman’s (1993) definition of frame functions (problem definition, causal interpretation, treatment recommendation, and moral evaluation), this study summarizes, compares, and discusses trends in the topical themes over the past five years. The results present that newspapers only devote a small proportion of coverage to the health hazards of air pollution, and this proportion has been decreasing year by year. The discussion of pollution sources gradually narrows down to industrial and agricultural pollution, fires, and transportation, ignoring pollution from the construction industry.
Although the three major newspapers have discussed numerous solutions to air pollution, the solutions all relate to government policies, which has led to a rise in topics related to political elections year by year. The politicization of air-pollution news may lead the public to expect solutions from governmental regulations and even party elections, ignoring the seriousness of health risks and reducing their willingness to take individual protective actions (Adger, Quinn, Lorenzoni, Murphy & Sweeney, 2012). Finally, based on the literature, this study discusses the theoretical and practical implications of the results.
出版資訊
706次下載
2023/ 冬
No.154
