Organizational Nationalism from the Group-Member Exchange Perspective

The nation provides an ‘imagined community’ (Anderson, Reference Anderson1991: 7), in which individuals and organizations are perceived as members of the larger national community. Traditional nationalism research is rooted in studying exceptional moments of social transformation, such as the rise of the modern nation-state and historical movements that redefine national boundaries (Calhoun, Reference Calhoun1997). In these studies, nationalism is viewed as a political ideology that elites deploy to legitimize their rule over a territorially bounded people (Gellner, Reference Gellner1983). More recent research on nationalism, focusing on the settled times, shifts the focus from elites toward the quotidian and argues that nationalism also manifests in the micro-level dispositions demonstrated by individuals and organizations (Bonikowski, Reference Bonikowski2016). For these actors, nationalism is not only a conscious ideology but also a discursive and cognitive frame through which they understand the world and navigate social interactions (Brubaker, Reference Brubaker2004). As a quotidian ideology, nationalism provides a set of normative frameworks that shapes the perceptions and behaviors of individuals and organizations; as such, nationalism defines the ends of action, and the goal of individuals and organizations is to obtain national sovereignty and even superiority. As a quotidian practice, nationalism provides a discursive and cognitive framework through which social actors understand and interact with each other; as such, nationalism is the means rather than the ends, and individuals and organizations invoke nationalism in their sense-making and interactions to achieve other purposes. In this article, we treat firms’ rhetorical nationalism as a form of quotidian nationalism, but we do not distinguish whether firms’ adoption of rhetorical nationalism is a reflection of their intrinsic ideology or an instrumental practice through which they strategically interact with their shareholders and stakeholders.

As a nation is a community of people, nationalism can be understood from the group-member exchange perspective. Groups provide members with a feeling of attachment, and members can obtain a sense of self-identity and self-esteem through their group affiliations. Elevating the level of the group to a nation, Druckman (Reference Druckman1994: 44) argues that ‘at the level of nation, the group fulfills economic, sociocultural, and political needs, giving individuals a sense of security, a feeling of belonging, and prestige’. Once a social actor's ingroup is set at the center, outgroups are judged in relation to it (Tajfel & Turner, Reference Tajfel and Turner2004). When the love of one's nation does not imply the hate of other nations, social actors demonstrate patriotism, and a more cooperative or peaceful approach to the world (Balabanis, Diamantopoulos, et al., Reference Balabanis, Diamantopoulos, Mueller and Melewar2001; Druckman, Reference Druckman1994). By contrast, nationalism differs from patriotism in that it is distinct by ‘the blind attachment to certain national cultural values, uncritical conformity with the prevailing group ways, and rejection of other nations as outgroups’ (Adorno, Frenkel-Brunswick, Levinson, & Sanford, Reference Adorno, Frenkel-Brunswick, Levinson and Sanford1950: 107) and therefore is more associated with a competitive or militaristic approach to the world. While the group-member exchange perspective has been developed to explain individuals’ behaviors, we expect the organizational-level manifestation of nationalism to similarly include a favorable attitude toward the ingroup (one's own nation) and an unfavorable attitude toward outgroups (other nations).

Moreover, the relationship between the group and members works in both directions – while the group provides members with security, safety, and prestige, members are expected to return their loyalty and commitment to the group. Nationalism, therefore, requires country members to prioritize the interest of their nation and be willing to sacrifice for it (Druckman, Reference Druckman1994). The specific actions that country members are expected to take depend on the dominant nationalistic agenda in a specific context. For example, at the time of a foreign invasion, a country's dominant agenda is to fight the foreign military force, and the role that country members can play is to defend their nation by joining or supporting their own military force. In the de-colonial movement, the nationalist agenda in an emerging nation state is to obtain political and economic self-determination; during the process, firms cut their dependence on the former colonial power and develop alternative ties with neighboring countries and other nations (Lubinski & Wadhwani, Reference Lubinski and Wadhwani2020). Nationalism is a source of aspiration for individuals and organizations and has a powerful impact on their behaviors; it mobilizes them to forego near-term economic benefits and costs in favor of long-term national wellbeing. Therefore, when measuring firms’ rhetorical nationalism as a quotidian ideology and a quotidian practice, we expect the organizational-level manifestation of nationalism to include two other dimensions, the dominant nationalistic agenda in a particular historical context and the role that firms can play to fulfill the agenda.

Four Firm-Related Dimensions of Chinese Nationalism

Existent research on Chinese nationalism supports ingroup favoritism, outgroup unfavoritism, the dominant agenda, and firms’ role as the four key dimensions of nationalism for Chinese firms. Modern Chinese nationalism originated in the late 19th century when China experienced a series of humiliations at the hands of foreign powers. Over the years, a negative thesis of resisting Western, imperialist humiliation (i.e., anti-foreign) has persisted as a major facet of Chinese nationalism (Weiss, Reference Weiss2014; Zhao, Reference Zhao2004). Anti-foreign sentiment is particularly sensitive to the eruptions of international conflicts. Events such as the protest against the 2008 Olympic torch relay in Paris and the territory dispute with Japan in 2012 have triggered waves of anti-foreign discourse that emphasized past injustices inflicted on China by foreign countries. Recently, the US–China trade war launched by the Trump administration in 2018, and the high-profile cases of the US government's ban on Chinese telecommunication companies are likely to have led to the rise of anti-foreign sentiments in Chinese public firms.

Besides the anti-foreign thesis, a positive thesis of national pride in the country's history, culture, and collective achievements is another component of modern Chinese nationalism (Zhao, Reference Zhao2004). China has one of the oldest and longest-lasting civilizations in the world, and national pride based on the country's history and cultural heritage supplies a strong sense of national identity. In addition, firms have often leveraged the nation's cultural heritage to market their products and services, and thus, we expect Chinese public firms’ nationalistic rhetoric to demonstrate national pride. As the country's historical heritage is relatively stable, we expect that national pride derived from this element will not vary significantly across time.

In the current era, the dominant agenda of Chinese nationalism is national revival. This theme is clearly demonstrated by the state-sponsored nationalist campaigns launched by the Chinese government (Neo & Xiang, Reference Neo and Xiang2022). The most common messages in these campaigns are that the Chinese Communist Party is leading Chinese people to restore the country's former position at the center of the world (Wang, Reference Wang2014) and that the country has achieved rapid economic, technological, and infrastructural development under the Party's leadership in recent decades (Steele & Lynch, Reference Steele and Lynch2013). Although revival phrases have been used by previous Chinese leaders, slogans such as ‘the Chinese Dream’ (中国梦) or ‘the great rejuvenation of the Chinese nation’ (中华民族伟大复兴) are particularly associated with China's current president, Xi Jinping, who emphasized them in his inaugural policy pronouncement as the leader of China in 2013. Thus, we expect revival-themed words to appear in corporations’ rhetorical nationalism as a demonstration of the dominant nationalist agenda in today's China. Responding to the dominant agenda, firms are expected to play a critical role in the process. Firms can contribute to the national rejuvenation by strengthening their own economic and technological competitiveness in the international market.

Together, our review of the research on quotidian nationalism and our analysis of Chinese nationalism pertinent to firms reveal anti-foreign, national pride, national revival, and corporate role as four distinct dimensions relevant to Chinese public firms. However, whether Chinese public firms have increased their rhetorical nationalism during the past 20 years, and if so, on which dimensions, are still open empirical questions. In addition, some firms may benefit more from demonstrating rhetorical nationalism than others, and thus there are likely to be firm-level heterogeneities in their deployment of rhetorical nationalism.

Firm Characteristics and Rhetorical Nationalism

As we study nationalism as both a quotidian practice and a quotidian ideology, we argue that firms’ deployment of rhetorical nationalism is related to their strategic motivations and socialization. Strategic motivations look at firms’ adoption of rhetorical nationalism as a practice to manage their relationships with important stakeholders. Past research has shown that firms strategically use language to signal their beliefs, values, and legitimacy to attract employees, consumers, and investors (Certo, Reference Certo2003; Connelly, Certo, Ireland, & Reutzel, Reference Connelly, Certo, Ireland and Reutzel2011). Firms that signal a commitment to social responsibility and environmental protection can improve employee attraction and distinguish themselves from competitors (Burbano, Reference Burbano2016; Jones, Willness, & Madey, Reference Jones, Willness and Madey2014). Firms also adopt patriotic rhetoric to absorb external resource constraints, the effectiveness of which is affected by the societal-level sentiments of nationalism and firms’ own involvement in overseas markets (Mohr & Schumacher, Reference Mohr and Schumacher2019).

Data Source

We collected the data from the annual reports of all Chinese publicly listed firms. In particular, we analyzed the MD&A section of the annual report where managers discuss and analyze financial information and major events to facilitate investors’ understanding of the company's operating results and financial conditions (China Securities Regulatory Commission, 1999). Compared with other sections in financial reports that mainly consist of boilerplate and tables with quantitative data, managers usually disclose relevant information in the MD&A via narrative communication. Narratives in MD&As are not subject to third-party assurance and thus provide discretion for managers to strategically release information in ways that help create a better impression or mitigate a negative impression of the company to investors. Thus, the MD&A serves as an accessible channel through which firms can exhibit rhetorical nationalism. A few existing studies have shown that managers strategically manipulate the wording in the MD&A to influence audiences’ impression of the firms (Caserio, Panaro, & Trucco, Reference Caserio, Panaro and Trucco2019; Muslu, Radhakrishnan, Subramanyam, & Lim, Reference Muslu, Radhakrishnan, Subramanyam and Lim2015). Scholars have also evidenced that the qualitative narrative in MD&As can be mined to obtain valuable information about firms’ future performance, operational uncertainty, managerial mentality, and corporate culture (e.g., Li, Reference Li2008, Reference Li2010; Loughran & McDonald, Reference Loughran and McDonald2016; Muslu et al., Reference Muslu, Radhakrishnan, Subramanyam and Lim2015).

Appendix I provides three examples of Chinese public firms using nationalistic rhetoric in the MD&A section. Example 1 is an excerpt from the MD&A section of the Tonghua Dongbao Pharma Co., a biopharmaceutical company. When discussing the impact of their innovation regarding human insulin in the 2005 annual report, it used the words ‘Chinese pharmaceutical industry as a leader of the world … and winning honor for our motherland’. Example 2 is an excerpt from the MD&A section of the Shanghai Laiyifen Co. Ltd., a snack food company. When elaborating its corporate strategy in the annual report of 2019, it said ‘its mission is to become a practitioner and disseminator of Chinese culture, enhance public welfare, and serve the motherland with the responsibility of conscience’. Example 3 is an excerpt from the MD&A section of the Dandong Xintai Electric Co. Ltd., which is an electric equipment manufacturer. In its 2015 annual report, the company stated that it would ‘strive for the realization of the Chinese Dream of the great rejuvenation of the Chinese nation, and the prosperity of the enterprise’. These nationalistic expressions, which are not directly related to firms’ finances, motivate us to investigate the phenomenon of rhetorical nationalism among Chinese public firms.

We obtained the annual reports of all listed companies during the period 2000–2020 from CNINFO, an official disclosure website of all listed companies in China. We started from the year 2000 because there are substantial missing data for public firms’ annual reports before 2000, and we ended in 2020, for which the most recent data are available. We obtained 45,713 unique annual reports in PDF format and applied the Tika package in Python to extract the text content of those PDF files for further textual analysis. Our sample comprised 43,434 annual reports after dropping 2,279 files that failed on text extraction because of file damage or incompatibility with Tika. Because the word embedding model relies on the relationship between the word and its neighboring words within the sentence to learn the representation of semantic meaning, coherent text data are desired for training. However, many sections in annual reports mainly consist of tables with quantitative data and boilerplate. To efficiently train the word embedding model, we included only the MD&A section of the annual reports because this section is mainly composed of narrative text on the company's business and financial conditions. We applied the regular expression matching operations in Python to extract the MD&A section from 41,285 annual reports and dropped 2,149 annual reports in which the MD&A section could not be detected or contained fewer than 100 valid words. The sample size was further reduced when matching with data about firms’ operational performance because of missing values in the dataset.

Word Segments

Unlike alphabet-based languages such as English, Chinese is a character-based language. A word might be represented in one or multiple characters and form sentences without clear punctuation. For example, the sentence ‘今天天气很好’ (‘The weather is good today’) consists of the words [‘今天,’ ‘天气,’ ‘很好’] with no blank spaces between them. To prepare the training data for a word embedding model, we applied the Jieba Footnote ¹ package in Python, a widely used Chinese word segmentation tool, to cut sentences into sequences of words.

We then further applied the named entity recognition (NER) procedure to recognize the named entity of words and replace named entities (locations, times, and number) with a predefined tag. For example, ‘2019 年 我司 营业 收入 100 万 人民币’ (‘In 2019, the earnings of our company is 1 million CNY’) would be transformed to ‘【时间】 我司 营业 收入 【数量】 万 人民币’ (‘[TIME], the earnings of our company is [NUMBER] CNY’). Applying the NER to preprocess the text can help reduce the number of tokens that the word embedding model needs to learn. Prior studies have used this technique to train word embedding models and found it helpful to enhance model performance (Li et al., Reference Li, Mai, Shen and Yan2021; Sugathadasa et al., Reference Sugathadasa, Ayesha, de Silva, Perera, Jayawardana, Lakmal and Perera2017).

Intuition of Word Embedding

Word embedding is used to represent the semantic meaning of words using a fixed-dimension numeric vector, where words with similar semantic meanings will have similar numeric values in the vectors. The construction of word embedding is based on the distribution hypothesis (Harris, Reference Harris1954): words that occur in similar contexts (with similar neighboring words) tend to have similar meanings. One naïve implementation to represent the semantic meaning of a word is to construct a count vector that records the frequency of all words in the corpus that appear near the focal word in the training corpus data. To implement the count vectors, we can first assign a unique ID (from 1 to N, where N is the size of vocabulary) to each word in the vocabulary and then use the position value (P, V) in a focal word's count vector to represent the frequency of its neighboring words: the value V in position P refers to the number of times that the word whose ID is P appears near the focal word. After the construction of word count vectors, the semantic similarity between words can be represented by the cosine similarityFootnote ² between their count vectors.

However, using count vectors for word embedding is inefficient. Given a corpus with N unique words, we need N count vectors with length N to represent the words in the corpus. Given that N is usually a large number (about 350,000 words in the third edition of The American Heritage Dictionary of the English Language), it would be extremely challenging to compute and maintain the resultant N × N matrix. Furthermore, the number of combinations of co-occurrence of words is enormous when N is large, and it is impossible for a corpus with a limited size to cover all the combinations, leading to sparse vector representations of words (i.e., many values in the count vector are zero).

Word2vec Model for Generating a Compact Word Embedding Vector

To generate a compact vector representation of words, many word embedding models have been proposed to map each word to a continuous vector space with a much lower dimension compared to the vocabulary size (Deerwester, Dumais, Furnas, Landauer, & Harshman, Reference Deerwester, Dumais, Furnas, Landauer and Harshman1990; Mikolov, Sutskever, Chen, Corrado, & Dean, Reference Mikolov, Sutskever, Chen, Corrado and Dean2013; Mikolov, Yih, et al., Reference Mikolov, Yih and Zweig2013; Rohde, Gonnerman, & Plaut, Reference Rohde, Gonnerman and Plaut2006). Among these methods, the recently proposed word2vec (Mikolov, Sutskever, et al., Reference Mikolov, Sutskever, Chen, Corrado and Dean2013; Mikolov, Yih, et al., Reference Mikolov, Yih and Zweig2013) achieves a breakthrough performance by training a neural network on the semantic meaning of words. The word2vec model learns to extract a word's semantic meaning from a large training corpus using the skip-gram (SG) schemeFootnote ³: given a specific word in a sentence, the word2vec model first maps the word to a fixed-length numeric vector (used as word embedding after the training) and then uses the numeric vector to predict all the neighboring words around the focal word. The word2vec model is trained via the SG scheme to capture the co-occurrence relationship between the focal word and its neighboring words. The training objective is to maximize

$${\cal L}_{SG} = \displaystyle{1 \over {\vert W \vert }}\sum\nolimits_{w_t\in W} {\log P( w_{c_1}, \;\ldots , \;w_{c_k}\vert w_t) } , \;$$

where W is the set of all word tokens in the corpus, w _t is the target (focal) word for a specific prediction, $w_{c_1}, \;\ldots , \;w_{c_k}$ are context words (neighboring words) of the target word w _t in a certain window, and k is the window size. The conditional probability P(w _c|w _t)is approximated by a multinomial logistic regression of |C| classes, which take the word embedding vector of the focal word $V_{w_t}$ as input:

$$P( {w_c{\rm \vert }w_t} ) = \displaystyle{{\exp {{\beta }^{\prime}}_{w_c}V_{w_t}} \over {\mathop \sum \nolimits_{w\in V} \exp {{\beta }^{\prime}}_wV_{w_t}}}, \;$$

where β _w are the coefficients associated with the word w, and C is the vocabulary of the corpora. However, calculating the conditional probability in this way is computationally inefficient because we need to go through all unique words in the vocabulary to obtain the denominator: $\sum\nolimits_{w\in V} {\exp {{\beta }^{\prime}}_wV_{w_t}}$. To alleviate the burden of computation, Mikolov, Yih, et al. (Reference Mikolov, Yih and Zweig2013) and Mikolov, Sutskever, et al. (Reference Mikolov, Sutskever, Chen, Corrado and Dean2013) apply a negative sampling strategy to reduce the multi-class problem to a binary classification. They first sample some focal word-neighboring word pairs (w _t, w _c) from the training corpus as positive samples and construct some pairs that are unlikely to exist in the training corpus as negative samples. Then, given a focal word-neighboring word pair (w _t, w _c), the word2vec model is asked to determine whether the pair is from positive samples or not. This binary classification problem can be formalized as P(Y|w _t, w _c), where Y = 1 if the pair (w _t, w _c) comes from the training corpus, and Y = 0 otherwise. Thus, the training objective when applying the negative sampling strategy is to maximize

$${\cal L}_{w2v} = \mathop \prod \limits_{( {w_t, w_c} ) \in D} P( {Y = 1{\rm \vert }w_t, \;w_c} ) \mathop \prod \limits_{( {w_t, w_c} ) \in {D}^{\prime}} P( {Y = 0{\rm \vert }w_t, \;w_c} ) , \;$$

where D is the set of positive samples and D ^′ is the set of negative samples. The conditional probability is calculated as $P( {Y = 1{\rm \vert }w_t, \;w_c} ) = {{\exp ( {{{\beta }^{\prime}}_{w_c}V_{w_t}} ) } \over {1 + \exp ( {{{\beta }^{\prime}}_{w_c}V_{w_t}} ) }}$.

The gist of this training framework is to capture the co-occurrence relationship between the focal word and its neighboring words. By training the word2vec model on a large corpus and updating the parameters of the model via backpropagation (a standard algorithm for training neural networks), the model gradually learns to represent the semantic meaning of words as fixed-length numeric vectors and associate the vector representations of focal words with their neighboring words. After the training is finished, the mapped numeric vectors are used as the word embedding for corresponding words.

Defining Sub-Dimensions and Corresponding Seed Words

We manually defined the seed words for the four dimensions of nationalism related to firms: national pride, national revival, corporate role, and anti-foreign. We first performed a media search of news articles with key words on corporations and nationalism (i.e., ‘企业’, ‘爱国主义’, ‘民族主义’) in the Baidu search engine. From the top 100 search results, we generated 10–15 words for each of the four dimensions of nationalism. We then asked six experts (two corporate executives, two business school professors, and two college students) to individually evaluate the relevance of each word and kept only the words with an approval rate higher than 67%. Nonetheless, defining these word lists manually was subject to noise and error, and thus further cleaning was needed.

We then inspected whether the semantic meanings of the proposed seed words aligned with the key concept of nationalism by applying the trained word2vec model to investigate the properties of their word embedding vectors. For each word of interest, we first determined its word embedding vector and then calculated the vector's cosine similarities to all other words’ word embedding vectors. We then ranked the cosine similarities from high to low and obtained the five synonyms of a word based on the cosine similarities in the context of MD&As to the focal word. Table 1 provides the list of manually generated seed words and their top synonyms. The synonyms provided a tool to understand the underlying semantic meaning of the seed words. The synonyms were conceptually similar to the seed words and well related to each dimension of nationalism, thus confirming the validity of the seed words.

Table 1.

Seed words and their top synonyms

Using Seed Words and the Trained Word2vec Model to Generate an Extended Word List

After obtaining the seed words for a specific dimension of nationalism, one straightforward method to measure companies’ strength in the focal dimension was to count the frequency of seed words in their MD&A reports. However, because of the limitation of human knowledge and preference, many words with semantic meanings aligned to the key concepts of the focal dimension were not included in the seed word list. Measuring the strength of the focal dimension based merely on its seed words may have led to bias (caused by human preference when selecting seed words) and imprecision (caused by an incomplete word list for the focal dimension of nationalism).

To mitigate the problems of bias and imprecision, a comprehensive word list for the focal dimension was needed. We utilized the trained word2vec model and the inspected seed words for each dimension to develop an expanded word list for the focal dimension. Given the trained word2vec model can capture the semantic similarities between words in the context of MD&A for each dimension, we could use the trained model to obtain the synonyms of the seed words (with high cosine similarity) and take the synonyms as relevant words that could be used to measure the strength of the focal dimension. The procedure of combining word embedding models and some predefined keywords to obtain a comprehensive word list has been adopted in several previous studies and shown to be capable of generating quality word lists (Li et al., Reference Li, Mai, Shen and Yan2021; Theil, Štajner, & Stuckenschmidt, Reference Theil, Štajner and Stuckenschmidt2020; Tsai & Wang, Reference Tsai and Wang2014).

Specifically, we first obtained the representative vector of the focal dimension by taking the average of word embedding vectors of all related seed words. We computed the cosine similarities between the word embedding vector of each word with the representative vector of the focal dimension and then selected words with high cosine similarities as the expanded word list for the focal dimension of nationalism. We further manually inspected the selected words and removed words that were not related to nationalism. For each dimension, we retained 100 wordsFootnote ⁴ and thus had 400 words related to nationalism in total. We followed the approach of Li et al. (Reference Li, Mai, Shen and Yan2021) to deal with words that overlapped sub-dimensions of nationalism: if a word appeared in multiple word lists, we included it only in the word list that had the highest cosine similarity between the representative vector of the dimension and the embedding vector of the word. Table 2 shows the word list for each dimension of nationalism, in order of descending similarity to the representative vector for each dimension.

Table 2.

Extended word list for each dimension

Computational Measurement of Rhetorical Nationalism

After obtaining the extended word list for each dimension of nationalism, we measured the strength of a focal dimension in an MD&A report by the weighted count of words belonging to the extended word list divided by the total number of words in the MD&A. We applied the term frequency–inverse document frequency (TFIDF) weighting scheme to adjust the word importance when calculating the word ratio, such that words with a higher frequency in the document and less coverage in other documents in the corpus were assigned larger weights. The use of the aggregation of the TFIDF weights of selected words to measure the qualitative attributes of text data has been widely applied in text analysis for its simplicity and effectiveness (Li et al., Reference Li, Mai, Shen and Yan2021; Loughran & McDonald, Reference Loughran and McDonald2011). Formally, given an MD&A and a list of selected words reflecting the targeted sub-dimension of nationalism [word _i, …, word _n], we have:

$$Score = \displaystyle{{\mathop \sum \nolimits_i^n TF_i \, \ast \, IDF_i} \over D} \, \ast \, 100, \;$$

where TF _i is the count of word _i in the MD&A, IDF _i is the natural logarithm of one plus the ratio of the total number of MD&As in the corpus divided by the number of MD&As containing $word_i\log \left({1 + {{{\rm Number\;of\;all\;MD}\& {\rm As}} \over {{\rm Number\;of\;MD}\& {\rm As\;containing}\;word_i}}} \right)$, D is the total word count in the MD&A, and the multiplier 100 is used to represent the frequency score as a percentage.

Following the procedure of generating the value of Score for each firm's MD&A report in a year, we applied the word list for each dimension of nationalism to generate the score for ‘national pride’, ‘corporate role’, ‘national revival’, and ‘anti-foreign’. We then obtained the aggregated score of nationalism by summing the scores of ‘national pride’, ‘corporate role’, ‘national revival’, and ‘anti-foreign’.

Validating the Constructs of Nationalism

It was important to validate our measures of rhetorical nationalism, as applying a word embedding model to extract a nationalism-related word list is a new methodology. We validated the constructed measures via socialization markers: (1) state ownership (Lavelle, Reference Lavelle2008), (2) central SOEs, and (3) firm's founding year according to imprinting theory (Deng, Reference Deng2009; Marquis & Qiao, Reference Marquis and Qiao2020). These socialization markers are backed by well-established relationships with corporate nationalism and are unlikely to be driven by corporate nationalism (i.e., reverse causality). Following Li et al. (Reference Li, Mai, Shen and Yan2021), we validated the association between our constructed measures and the aforementioned markers via regression, with industry-, year-, and province-fixed effects. We also included firm size and ROA to control for the basic characteristics of firms.

Following Lin et al. (Reference Lin, Fu and Fu2021), we define SOEs using an indicator variable that takes the value of one if the biggest shareholder controlling the firm of a year is a state-owned organization and if the biggest shareholder has more than 20% direct or indirect voting rights.Footnote ⁵ We expect that SOEs would be more nationalistic than non-SOEs. Besides general state ownership, we also investigate nationalism under different varieties of state capitalism (Musacchio, Lazzarini, & Aguilera, Reference Musacchio, Lazzarini and Aguilera2015). We follow Lin et al. (Reference Lin, Fu and Fu2021) to classify a firm's state ownership as local government ownership or central government ownership. Specifically, central SOEs are those SOEs managed by the SASAC-SC or supervised by the MF, while other SOEs are local SOEs. In our sample, about 30% of the SOEs are central SOEs. As central SOEs usually operate in industries that are concerned with national livelihood, defense security, or strategic priority, we expect that these firms would demonstrate more rhetorical nationalism than local SOEs. Table 3 reports the results and has five panels that include both the aggregation (Panel A) and each of the four sub-dimensions (Panels B–E) of nationalism. Column 1 of Table 3 presents the results using state ownership as the independent variable. Expectedly, we find that SOEs have stronger rhetorical nationalism across all sub-dimensions and their aggregation.

Table 3.

Comparison of the nationalism value on markers

Notes: p-values in parentheses. ***p < 0.01, **p < 0.05, *p < 0.1.

Notes: p-values in parentheses. ***p < 0.01, **p < 0.05, *p < 0.1.

Notes: p-values in parentheses. ***p < 0.01, **p < 0.05, *p < 0.1.

Notes: p-values in parentheses. ***p < 0.01, **p < 0.05, *p < 0.1.

Notes: p-values in parentheses. ***p < 0.01, **p < 0.05, *p < 0.1.

Column 2 of Table 3 presents the results when separating state ownership into that affiliated with the central government and that affiliated with the local government. We find that the association between the central SOE and rhetorical nationalism is larger (p < 0.01 in a Wald test) than that between the local government and rhetorical nationalism. Specifically, Wald tests show that the central SOE's rhetorical nationalism is higher than that of local SOEs in corporate role (p < 0.01) and anti-foreign (p < 0.01) but lower in national pride (p < 0.05). In addition, the coefficients are not statistically different when it comes to national revival. Compared with local SOEs, central SOEs are expected to serve national strategic interests and fulfill important social obligations, and thus their rhetorical nationalism should demonstrate their important corporate role. Similarly, they are also likely to compete more in the international market and hence have a higher anti-foreign score than local SOEs. The exception of national pride is understandable because local SOEs are usually profit-driven and pursue commercial purposes to serve local economic development objectives (Lin et al., Reference Lin, Fu and Fu2021). In our sample, local SOEs are more likely to develop products from regional culture or traditions (e.g., Chinese liquor and traditional Chinese medicine), and thus, these firms show a higher level of national pride. The indifference in national revival between central SOEs and local SOEs may be attributed to the fact that both types of firms are subject to the same external environment and thus adopt similar rhetoric related to the dominant societal agenda. Together, the overall higher level of rhetorical nationalism demonstrated by central SOEs than local SOEs suggests the validity of our measurement.

Finally, we leverage two imprinting effect-related markers to validate our constructed measures: (1) firms founded before 1992 and (2) firms founded after 2001. We code dummy variables to indicate whether a firm was founded before 1992 (after 2001). Columns 3 and 4 of Table 3 show that firms founded before 1992 have significantly higher nationalism, while those founded after 2001 have significantly lower nationalism, both in aggregation and in sub-dimensions. In Column 5 of Table 3, we simultaneously test the three markers (state ownership, firms founded before 1992, and firms founded after 2001) and find the patterns of results to be largely consistent. These tests further suggest that the markers are not completely overlapped and that our rhetorical nationalism measurement can be verified by multiple markers.

Convergent Validity and Discriminant Validity

We assess the convergent validity and discriminant validity of our rhetorical nationalism measures. Convergent validity refers to the degree to which a measure that is designed to assess a particular construct correlates with other tests that assess the same construct, while discriminant validity refers to the degree to which a variable that is designed to measure a particular construct does not correlate with variables that measure different constructs. We use data from the Chinese private enterprise survey (CPES)Footnote ⁶ and test correlations between our measurements and private enterprises’ self-reported nationalism. The CPES is the most representative survey of private enterprises in China (Chen, Lu, Lin, & Song, Reference Chen, Lu, Lin and Song2019), and its survey questions vary by year. In particular, the 2014 and 2018 surveys asked leaders of private firms questions related to the four dimensions of private firms’ nationalistic values. For example, the 2018 survey asked whether a firm leader would agree that the literature of traditional Chinese culture should be required reading for children's elementary education, and this question can be used to assess a firm leader's value on national pride. Similarly, three 2018 survey items that can be used to measure the anti-foreign value are (1) trade protectionism, (2) conspiracy belief that foreign forces attempt to curb China's development, and (3) the influence of Hollywood on Chinese culture. Two survey items related to the ‘China Dream’ in 2014 can be used to measure the national revival value, and one survey item related to whether a firm leader would put the interest of the firm before that of the national rejuvenation can be used to measure the corporate role value. Together, these CPES items provide the survey-based measurements of private firms’ nationalism. Appendix II shows the selected questions for each sub-dimension of nationalism and the corresponding rationale.

However, one challenge of using private firms’ survey responses is that we do not know the identities of these firmsFootnote ⁷ and thus cannot match them to the list of publicly traded firms in our sample. Therefore, we choose to verify the correlation between our constructed measures of rhetorical nationalism and private enterprises’ self-reported nationalism at the city level. We aggregate firms’ rhetorical nationalism and self-reported nationalism by taking their average at the city level according to their headquarter cities. We remove those SOE firms in our samples when aggregating their rhetorical nationalism, as the CPES surveys only private enterprises. To align with the survey year of the self-reported nationalism, the city-level rhetorical nationalism is calculated using observations in the last three years to the survey year in order to mitigate the problem of meager observations of public firms in some cities. Cities with less than ten valid observations of rhetorical nationalism or self-reported nationalism are omitted. For sub-dimensions with multiple questions, we apply principal component analysis to combine the responses in multiple questions into a one-dimension value.

Table 4 shows the correlations between our computed rhetorical nationalism and the survey-based nationalism at the city level. The sub-dimensions of computational nationalism and the corresponding sub-dimensions of self-reported nationalism are all positively correlated, ranging from 0.1609 to 0.2484. That is, cities whose firms’ computational nationalism is high are inclined to have high self-reported nationalism in the corresponding dimension. Besides, a sub-dimension in computational nationalism is most correlated to the corresponding sub-dimension in the survey-measured nationalism but less correlated to other sub-dimensions, suggesting that our computational measurement of rhetorical nationalism has convergent as well as discriminant validity. These correlations are all nonsignificant, which may be attributed to the indirect test at the city rather than at the firm level and the small samples after aggregating at the city level. We also report the correlations among the sub-dimensions of computational nationalism in Table 7 and find that the correlations among the sub-dimensions are not too high. The maximum correlation is 0.26 between corporate role and national revival, which is understandable given that firms’ role corresponds to the dominant agenda of nationalism. These results suggest that our computational measurement of rhetorical nationalism has convergent as well as discriminant validity.

Table 4.

Convergent and discriminant validity of computational measure

Notes: p-values in parentheses; We did not report the correlation between the aggregated nationalism because the self-reported data is from two surveys and cannot be directly added up.

The Growth of Rhetorical Nationalism

Table 5 presents the average scores of the nationalism measures for all available firm observations across years. The corresponding graphic representation in Figure 1 demonstrates the growing trend of nationalism among Chinese public firms from 2000 to 2020. Overall, the aggregated nationalism score of Chinese public firms based on the MD&A section of their annual reports increases from 0.368 to 0.725 (an absolute increase of 0.357 and a percentage increase of 97%) from 2000 to 2020. Out of the four dimensions of nationalism, national revival and corporate role are the two most significant components of nationalism, accounting for 74% of the nationalistic rhetoric used by Chinese public firms in 2020. National revival has the largest growth (an increase of 0.151 from 2000 to 2020), contributing 42.2% to the growth of the aggregated nationalism score, while corporate role has the second largest growth (0.115 from 2000 to 2020), contributing 32.1% to the growth of the aggregated nationalism score. The change in these sub-dimensions indicates that national revival has become the most important theme of rhetorical nationalism for Chinese public firms, and that public firms have emphasized their role in this process.

Table 5.

Nationalism score for each dimension and the aggregation across years

Figure 1.

The trend of rhetorical nationalism across years

The trend of rhetorical nationalism among Chinese public firms from 2000 to 2020. For each year, we calculate the average percentage scores of sub-dimension and aggregation of all public firms’ rhetorical nationalism. The data of Nationalism (aggregation) are shown in the Y-axis on the right.

Although national pride and anti-foreign rhetoric account for smaller proportions of the aggregate nationalism score (10.3% for national pride and 16.0% for anti-foreign in 2020), they have both experienced substantial growth during the past 20 years (90.1% for national pride and 97.3% for anti-foreign from 2000 to 2020). Therefore, Chinese public firms’ rhetorical nationalism has grown in all four dimensions. As national pride is derived from historical heritage, its value, as shown in Figure 1, is the most stable. By contrast, the anti-foreign value shows substantial variation over time and has various peaks corresponding to foreign policy events such as the Beijing Olympics protest, China–Japan territory disputes, and the trade war in 2018.

Despite the overall growth trend, Chinese public firms’ rhetorical nationalism experienced a substantial decrease between 2002 and 2004. This may have been caused by China's entry into the WTO in late 2001, and these three years mark the lowest points of Chinese public firms’ rhetorical nationalism during the past two decades. However, the low level did not last. We note that the strident increase in Chinese public firms’ rhetorical nationalism in 2005 (0.288 increase in the aggregated nationalism score) coincided with massive anti-Japanese demonstrations against Japan's bid to join the United Nations Security Council, Japan's approval of history textbooks that whitewashed Japanese wartime atrocities, and its pledge to help the US defend Taiwan in the event of an attack by Beijing. In that year, all of the sub-dimensions of nationalism experienced a significant increase, suggesting soaring rhetorical nationalism among public firms at that time.

Interestingly, another year with significant growth of rhetorical nationalism is 2008 (0.146 increase) when Beijing hosted the Olympics and the global financial crisis occurred. While the Olympics was supposed to promote the connection between China and the rest of the word, anti-China protests in Paris, London, San Francisco, and New Delhi during that year's Olympic torch relay catalyzed an outpouring of nationalism in China (Blanchard, Reference Blanchard2008). The People's Daily, the mouthpiece of the government, issued a signed commentary titled, ‘How can patriotism be more powerful?’, calling on people to focus on building up the country's overall strength (NBC News, 2008). Meanwhile, the global financial crisis, although rooted in Western countries, shocked the Chinese economy, and this might have contributed to a feeling of being victimized among Chinese firms. Finally, another important trend is that, during the more recent period from 2013 to 2020, Chinese public firms’ rhetorical nationalism has experienced steady growth.

The Yearly Distribution of Rhetorical Nationalism

To rule out the possibility that the overall growth in rhetorical nationalism was driven by some outlier firms, we further unpacked the yearly distribution of the nationalism score at the beginning, middle, and end of the two decades (2000, 2010, and 2020). Figure 2 shows that, consistent with the findings in Figure 1, the average of the nationalism value increased over time, suggesting that nationalistic rhetoric has become more and more common in Chinese public firms’ annual reports. Moreover, the distributions show a steady rightward movement, which suggests that the increasing trend is not driven by outlier firms.

Figure 2.

The histogram of rhetorical nationalism

The histogram (density) of scores of the aggregation and sub-dimension of public firms’ rhetorical nationalism in years 2000, 2010, and 2020. (A) Nationalism, (B) national pride, (C) national revival, (D) corporate role, and (E) anti-foreign.

The Geographical Distribution of Rhetorical Nationalism

To investigate whether the increasing trend of rhetorical nationalism was caused by public firms located in certain geographical areas, we calculated the average rhetorical nationalism of all public firms headquartered in a province. Figure 3 shows the province-level distribution in the years 2000, 2010, and 2020. We observe that the use of nationalistic rhetoric increased in public firms located all over the country and that the trend of increasing nationalism is not driven by firms in specific regions. This finding further confirms that rising nationalism is a nation-wide trend.

Figure 3.

Geographical distribution

Geographical distribution of public firms’ rhetorical nationalism in years 2000, 2010, and 2020.Footnote ¹⁰ For each year, we calculate the geographical average score of rhetorical nationalism based on all public firms established in the district. Regions without valid data points are set to be gray. (A) 2000, (B) 2010, and (C) 2020.

The Industrial Distribution of Rhetorical Nationalism

To shed some light on the industrial distribution of rhetorical nationalism, we calculated the average rhetorical nationalism of all public firms in each industry. We used the criterion provided by the China Securities Regulatory CommissionFootnote ⁸ to classify public firms into 19 industries. Panel A of Table 6 shows the industry-level distribution of public firms’ rhetorical nationalism in the full sample. To further validate the heterogeneity of public firms’ rhetorical nationalism among Business-to-consumer (B2C) and Business-to-business (B2B) industries, we manually identified the industries operating under the B2C model. Specifically, we constructed a dummy variable ToC that takes the value of one if the industry is among ‘wholesale and retail’Footnote ⁹, ‘lodging and catering’, ‘resident services, repair and other services’, ‘Education’ and ‘Culture, sports and entertainment’ (with the industry codes of F, H, O, P, R), and zero otherwise. We then validated the association between rhetorical nationalism and ToC via OLS regression. Panel B of Table 6 presents the regression results, with year- and province-fixed effects to control for unobserved heterogeneities related to time and region. We also included the markers and control variables identified in the section ‘Validating the constructs of nationalism’ to control for firms’ basic characteristics. We found that firms operating in B2C industries have an overall higher nationalism score. These firms were especially active in using phrases related to ‘national pride’ and ‘national revival’ but less likely to express sentiments related to ‘anti-foreign’ and ‘corporate role’. These consumer-facing industries are not usually in a position to lead China in international economic and technological competition, and they may also serve overseas markets and therefore demonstrate lower scores for ‘anti-foreign’ and ‘corporate role’.

Table 6.

The industrial distribution of rhetorical nationalism

Notes: p-values in parentheses. ***p < 0.01, **p < 0.05, *p < 0.1.

Strategic Motivation and Rhetorical Nationalism

In this section, we investigate the kinds of public firms that are more likely to exhibit nationalism in their disclosure. Based on our analysis in the section ‘Four firm-related dimensions of Chinese nationalism’ regarding the types of firms that are either motivated or socialized to demonstrate nationalistic sentiment, we have already tested the relationship between a firm's nationalism score and three socialization variables, being an SOE, being founded before 1992, and being founded after 2001. We further investigate the relationship between a firm's rhetorical nationalism and the three variables related to their strategic motivation for adopting nationalistic rhetoric: (1) government subsidies, measured as the natural logarithm of the number of subsidies that a firm received from the government in the last fiscal year, (2) overseas sales, measured as the natural logarithm of sales from the overseas market in the last fiscal year, and (3) number of individual investors, measured as the natural logarithm of the firm's number of individual investors at the end of the last fiscal year.

We used the OLS model to test the association between the firm attributes of interest and the expressed nationalism in firms’ MD&A. For each regression, industry-, year-, and province-fixed effects were included to control for unobserved heterogeneities related to industry, time, and region. In addition, we controlled for the three markers that we used to validate the computational rhetorical nationalism. We also controlled for (1) firm size, measured as the natural logarithm of the firm's total assets at the end of the fiscal year, (2) operating performance, measured as the ROA in the fiscal year, and (3) market-to-book ratio, measured as the natural logarithm of the market value divided by the book value at the end of the fiscal year. We collected these firm-level characteristics from the CSMAR Database. Table 7 presents the summary statistics and pairwise correlations for firm characteristics and nationalism scores. The correlations among explaining variables are not very high, indicating that multi-collinearity is not a significant concern.

Table 7.

Summary statistics and correlation table

Note: Correlation coefficients with absolute values greater than 0.01 are statistically significant at the p ≤ 0.05 level of confidence.

Table 8 presents the results of using the rhetorical nationalism scores as the dependent variable. The sample size was smaller than that in the analysis in the section ‘Measuring the nationalism score of Chinese public firms’, as some variables had missing values. For example, in the CSMAR Database, data about overseas sales were available only from 2007, leading to the absence of around 20% of the values for the variable; similarly, about 32% of values for the variable of government subsidies were missing. Column 1 of Table 8 uses the aggregated nationalism score as the dependent variable and shows that firms with higher total assets (firm size), better operating performance (ROA), more investors holding stocks, and lower sales from overseas are more active in adopting rhetorical nationalism. Overall, our results paint a picture that larger and more profitable firms with more investors and lower sales from overseas are those that demonstrate the highest level of rhetorical nationalism. However, government subsidies have no significant impact on firms’ adoption of rhetorical nationalism, and the results indicate that the rhetorical nationalism manifested in Chinese public firms’ annual reports may be geared toward consumers and investors more than the government. We need to note that the lack of correlation between government subsidies and firms’ rhetorical nationalism may also be caused by the substantial missing data in the measurement of government subsidies. Thus, future research should further test the relationship between the two. In addition, all three marker variables are significantly associated with the overall level of rhetorical nationalism expressed by Chinese public firms after controlling for other firm-level characteristics.

Table 8.

Associations between firm characteristics and nationalism

Notes: p-values in parentheses. ***p < 0.01, **p < 0.05, *p < 0.1.

Columns 2–5 of Table 8 present the results using the sub-dimensions of nationalism as dependent variables. We see that firms that demonstrate a high level of national pride tend to be those that are more profitable, have a larger number of investors, and have lower sales from overseas. Firms that demonstrate more national revival-related nationalism are those that are larger and more profitable, and have a lower market-to-book ratio, a larger number of investors, lower sales from overseas, and fewer subsidies from the government. Firms that are more likely to demonstrate corporate role-related nationalism are those that are larger, more profitable, and with a higher market-to-book ratio, lower sales from overseas, a larger number of investors, and more subsidies from the government. The positive effect of government subsidies may be explained by the fact that firms that have obtained the government's support are motivated to highlight their contributions to the society. Finally, firms that demonstrate high anti-foreign nationalism are those that are larger, and with a lower market-to-book ratio, a larger number of investors, and fewer subsidies from the government. These patterns of results are largely consistent with the analysis of the aggregate nationalism score in Column 1.

Robustness Checks with Missing Values and the Lagged Dependent Variable

Furthermore, to address the concern of the missing data about government subsidies and overseas sales, we reran the models without these two variables. Table 9 reports the results and shows that firms with a larger size, a high level of profitability, and a large number of individual investors demonstrate more nationalistic rhetoric in their annual reports. These results are consistent with those reported in Table 8.

Table 9.

Associations between firm characteristics and nationalism (excludes overseas sales and received subsidies)

Notes: p-values in parentheses. ***p < 0.01, **p < 0.05, *p < 0.1.

To mitigate the concerns of reverse causation between firm characteristics and rhetorical nationalism, we introduced the lag term of the dependent variable as a control variable and reran the regression models. The results are represented in Table 10. We found that the lag term of the dependent variable (last year's rhetorical nationalism) is positively associated with the current year's rhetorical nationalism, suggesting that public firms discourse nationalism in a relatively constant style. Meanwhile, though the statistical power of firm characteristics decays to some extent, the sign coefficients that are statistically significant in Table 9 remain the same in Table 10. These patterns are expected, as many firm characteristics will also not vary significantly across time, and thus adding the lag term of the dependent variable will also absorb some of the explanatory power of other predictors. The slightly reduced sample size is mainly attributed to the imperfect extraction of firms’ MD&As, and some firm-year observations do not have extracted text data, leading to missing values in the previous year's rhetorical nationalism.

Table 10.

Associations between firm characteristics and nationalism (excludes overseas sales and received subsidies and with lag terms of dependent variables)

Notes: p-values in parentheses. ***p < 0.01, **p < 0.05, *p < 0.1.

Rhetorical Nationalism and Future Economic Performance

Finally, we investigated the associations between a firm's rhetorical nationalism in the current year (t) and the following year's (t + 1) economic performance and presented the results in Table 11. We found that firms that demonstrate a high level of rhetorical nationalism tend to have better future economic performance. All sub-dimensions of nationalism are positively associated with future ROA, but only national revival has a statistically significant effect. This may be attributed to the fact that firms with a strong sense of national revival are more inclined to invest in activities that directly benefit the country's progress. By doing so, these firms also contribute to the nation's socioeconomic growth and position themselves favorably for long-term success. However, we would like to caution that establishing causal relationships between sub-dimensions of rhetorical nationalism and firm performance is beyond the scope of our article. We merely presented the factual findings of correlations between sub-dimensions of nationalism and firm ROA, and called for future research that can adopt causal identification strategies to further rigorously test the relationships.

Table 11.

Associations between nationalism and firm's future performance

Notes: p-values in parentheses. ***p < 0.01, **p < 0.05, *p < 0.1.

We then further tested whether rhetorical nationalism could help firms obtain profit from the domestic market, which in turn contributes to their better performance. We introduced the operating profit from the domestic market in the next year scaled by the total asset (DProfitability _t+1) as a mediator. We also include the current-year domestic market profit as a control variable (DProfitability _t). We validated the mediating effect of DProfitability _t+1 using structural equation modeling (Ullman & Bentler, Reference Ullman, Bentler and Weiner2012). The results are shown in Table 12. The sample size in Table 12 is reduced compared to Table 11 because the data on domestic profit are available only from 2007. Table 12 shows that, with all control variables, rhetorical nationalism is positive in predicting a firm's domestic market profit (β = 0.002; p = 0.087). Furthermore, we found that a firm's domestic market profit is positive and significant (β = 0.946; p = 0.000) in predicting future ROA. And rhetorical nationalism is also significant in predicting future ROA (β = 0.001; p = 0.071). A firm's domestic market profit thus partially mediates the relationship between rhetorical nationalism and firm future ROA. The proportion of the total effect that is mediated is 0.69. In unreported analysis, we also explored the mediating effect of firms’ received subsidy from the government in the next year and the firm's amount of long-term loans, but we do not find any significant mediating effect (untabulated). These results reveal that rhetorical nationalism could help firms improve their future ROA, mainly by earning more profit from the domestic market as opposed to obtaining support from the government or better accessing financial resources.

Table 12.

Structural Equation Modeling (SEM) examining the mediating role of domestic profitability

Notes: SE, standard error; LL, Lower level; UL, Upper level.

Theoretical Contributions

While our primary contribution in this article is to develop a computational measurement of Chinese public firms’ rhetorical nationalism, this article also makes important theoretical contributions to the organizational research on nationalism. While the global surge of nationalism has been depicted as the ‘most powerful force in the world’ (Mounk, Reference Mounk2018; Walt, Reference Walt2019), organizational scholars have paid relatively scant attention to nationalism. The existent organization and strategy literature has mostly studied nationalism at the country level (e.g., Click & Weiner, Reference Click and Weiner2010; Ertug et al., Reference Ertug, Cuypers, Dow and Edman2023; Fisman et al., Reference Fisman, Hamao and Wang2014; Lubinski & Wadhwani, Reference Lubinski and Wadhwani2020) and has not studied how nationalism is manifested in the quotidian domain, especially in organizational activities (Bonikowski, Reference Bonikowski2016; Bonikowski & DiMaggio, Reference Bonikowski and DiMaggio2016; Takeda, Reference Takeda2021). One related work is Mohr and Schumacher's (Reference Mohr and Schumacher2019) study of firms’ patriotic rhetoric. However, patriotism is conceptually different from nationalism (Ertug et al., Reference Ertug, Cuypers, Dow and Edman2023), and the focus of their study is on the contingent impact on firm performance rather than developing the theoretical construct of the firm-level nationalism and its measurement. Our paper fills this literature gap by contributing to the organizational research on nationalism on two fronts. First, we argue that, as a quotidian form of nationalism, firms’ manifestation of nationalism can be a result of the organizational goal of pursuing national superiority or a set of strategies that organizations adopt to interact with their external environment. So, organization-level nationalism can be both the ends and means, and we find support for both motivations in our study of rhetorical nationalism of Chinese public firms. Firms that are founded to accomplish the strategic goals of the government or in the era with a strong nationalistic imprinting effect demonstrate a higher level of rhetorical nationalism. In addition, as nationalism fosters a united culture and offers a collective identity for citizens of a nation, domestic investors and consumers cast a more positive view on firms that demonstrate a high level of nationalism. In turn, firms can strategically deploy nationalism in their routine interactions with shareholders and stakeholders and benefit from doing so. We provide supportive evidence regarding firms’ strategic motivation and performance consequences in deploying nationalism.

Our second contribution to the organizational studies of nationalism is that we develop a four-dimensional theoretical framework on rhetorical nationalism for organizations (firms). We extended the group-member exchange perspective to study organization-level nationalism and found that ingroup favoritism manifests as national pride in a nation's cultural heritage and recent achievements, whereas outgroup unfavoritism manifests as competition with foreign firms and concerns over uncertainties in the international market. In addition, we added the dominant nationalistic agenda and organizations’ role in the process as two additional dimensions of organizations’ rhetorical nationalism. Although our context is Chinese public firms, the four-dimensional theoretical framework can be applied generally to study organizational-level nationalism in other contexts.

Our third contribution to the organizational research of nationalism is to develop a computational method of analyzing organizations’ routine communications. Although researchers have already analyzed firms’ patriotic rhetoric using earnings calls and press releases (i.e., Mohr & Schumacher, Reference Mohr and Schumacher2019), the method of measuring a firm's values based merely on its use of seed words may lead to bias due to human preferences on seed word selection and imprecision due to the lack of a comprehensive word list. The word embedding method can effectively overcome these limitations by evaluating a word's relevance through its position in its neighboring context and therefore capturing the cultural meanings embedded in words’ relationships (Kozlowski et al., Reference Kozlowski, Taddy and Evans2019; Li et al., Reference Li, Mai, Shen and Yan2021). Word embedding techniques have become increasingly popular in studying firms’ values and culture (e.g., Li et al., Reference Li, Mai, Shen and Yan2021; Lawson, Martin, Huda, & Matz, Reference Lawson, Martin, Huda and Matz2022), and our study is the first to apply this method to measure firms’ rhetorical nationalism. The method that we have developed can be applied to study organizational-level nationalism in other countries, and the extended word list that we developed can be adopted by future researchers to study the relationship between Chinese public firms’ rhetorical nationalism and many other variables. Together, we hope the novel measurement that we have developed in this article can lay down the methodological foundation for a vibrant school of research about firms’ rhetorical nationalism.

Besides the general contribution to the organizational study of nationalism, our article contributes to the specific research area of Chinese nationalism by advancing the research on firms. Firms are important players in the rising nationalism in China, and they are important forces that influence geopolitical conflicts, the global supply chain, and technology competition. However, the research on Chinese nationalism has traditionally focused on investigating the sentiment of the public (Gries, Reference Gries2004; Han, Reference Han2018; Johnston, Reference Johnston2016; Schneider, Reference Schneider2018) or the policy of the government (Weiss, Reference Weiss2014; Zhao, Reference Zhao2004). As a result, how nationalism can manifest in the corporate domains has been rarely studied (see Fisman et al., Reference Fisman, Hamao and Wang2014 and Tian et al., Reference Tian, Tse, Xiang, Li and Pan2021 for exceptions). Our article provides the first set of rigorous, quantitative evidence regarding rising rhetorical nationalism in the population of Chinese public firms in the last two decades. We show that the rhetorical nationalism of Chinese public firms has nearly doubled in the past two decades, and the growth is cross-board rather than driven by specific sectors or regions. In addition, while the popular media have often painted the Chinese government as the primary driver of the rising nationalism in China (e.g., Doshi, Reference Doshi2021), we show that firms’ attempts to appeal to consumers and investors may have played an even larger role in affecting their adoption of rhetorical nationalism. Chinese firms would have strong incentives to appear nationalistic because they are financially rewarded for doing so. Therefore, the rising nationalism in China is not just a result of a top-down process but also a bottom-up one driven by firms’ attempts to manage their shareholders and stakeholders and obtain returns.

Limitations and Future Research Directions

It is also important to point out that our article is not without limitations, and recognition of these limitations can lead to future research opportunities. One limitation is that we study Chinese public firms’ rhetorical nationalism demonstrated only in the MD&A section of annual reports. Firms may also demonstrate nationalism in other forms of communication, such as CEO speeches and social media releases to the public. However, it is important to point out that the MD&A section provides the most reliable source to measure Chinese public firms’ rhetorical nationalism. Unlike other forms of firm communication, public firms’ annual reports are mandated, and the China Securities Regulatory Commission has built the CNINFO system for the public to freely and easily access firms’ annual reports. Chinese public firms, unlike their US counterparts, do not usually have earnings call records. In this regard, the annual report is arguably the most important channel for external stakeholders to obtain information about public firms. The non-extemporaneous nature of the writing and managers’ flexibility in choosing what to cover in the MD&A makes it a place where firms could strategically release information to external stakeholders, as managers can carefully craft the language and content in MD&As before releasing them to the public (Caserio et al., Reference Caserio, Panaro and Trucco2019; Li, Reference Li2008). Meanwhile, communications in other forms, such as CEO speeches and social media releases, can be sporadic, subject to communicative interaction with others and external events, and therefore may not be used to systematically measure firms’ rhetorical nationalism. Nevertheless, future research should try to gather a more comprehensive set of communications by firms and investigate whether firms’ nationalistic rhetoric varies by communication channels. In addition, future research should investigate the relationship between firms’ rhetorical nationalism and material actions. From an intrinsic perspective, firms that demonstrate higher rhetorical nationalism should have more material actions. However, from an instrumental perspective, firms that demonstrate higher rhetorical nationalism may actually have fewer material actions if the latter is more costly. Investigating the relationships between different measurements of organizational nationalism is likely to yield fruitful outcomes.

The second limitation of our study is that we are not able to test the convergent and discriminant validity of our measures through firm-level analysis. Firms that were surveyed by the CPES may not necessarily be publicly listed firms. Although the patterns of correlations based on samples that may not even match provide evidence regarding our measurements, future research should look for firm-level survey data to provide more rigorous validation tests. The third limitation of our research is that we establish only correlations between firms’ rhetorical nationalism and other characteristics. As we acknowledged in the article, the relationships between some firm-level variables, such as performance, and rhetorical nationalism adoption can be inverse. As our primary goal in this article is to develop a computational method to measure Chinese public firms’ rhetorical nationalism and to describe the general trend of the past two decades, we did not intend to establish causal relationships between firms’ rhetorical nationalism and other firm-level variables. However, future researchers should exploit experiments, policy shocks, and other causal identification schemes to more clearly delineate the factors that contribute to the growing manifestation of nationalism in firms and investigate the consequences of adopting nationalistic rhetoric for firms.

Future research should also investigate how firms’ nationalism discourse can interact with state-led nationalistic campaigns and active grassroots nationalistic forces. Our article shows that national revival and firms’ role in this process are the two most important components of nationalistic rhetoric manifested by Chinese public firms. These results indicate that firms are a positive force for achieving nationalism goals and national superiority in the international sphere. Our article also shows that firms adopt nationalistic rhetoric to cater to the public's rising appetite for nationalism. Therefore, Chinese nationalism in the political, social, and economic spheres may have reinforced each other, and investigating the relationships of nationalism in these three spheres can be another fruitful question for future research.