Sustainability in Leadership: The Implicit Associations of the First-Person Pronouns and Leadership Effectiveness Based on Word Embedding Association Test

Abstract

The first-person pronoun is an indispensable element of the communication process. Meanwhile, leadership effectiveness, as the result of leaders’ leadership work, is the key to the sustainable development of leaders and corporations. However, due to the constraints of traditional methods and sample bias, it is challenging to accurately measure and validate the relationship between first-person pronouns and leadership effectiveness at the implicit level. Word Embedding Association Test (WEAT) measures the relative degree of association between words in natural language by calculating the difference in word similarity. This study employs the word and sentence vector indicators of WEAT to investigate the implicit relationship between first-person pronouns and leadership effectiveness. The word vector analyses of the Beijing Normal University word vector database and Google News word vector database demonstrate that the cosine similarity and semantic similarity of “we-leadership effectiveness” are considerably greater than that of “I-leadership effectiveness”. Furthermore, the sentence vector analyses of the Chinese Wikipedia BERT model corroborate this relationship. In conclusion, the results of a machine learning-based WEAT verified the relationship between first-person plural pronouns and leadership effectiveness. This suggests that when leaders prefer to use “we”, they are perceived to be more effective.

1. Introduction

The topic of leadership has consistently occupied a significant position in the investigation of the factors influencing leadership effectiveness and the improvement of leadership effectiveness in the field of organizational behavior research. Researchers have demonstrated considerable enthusiasm for these two issues, which remain enduring topics in management practice. Leadership effectiveness is defined as a leader’s behavioral ability, work status, and results in the leadership process [1]. The focus on leadership effectiveness can be observed in a leader’s leadership ability, leadership efficiency, and leadership effectiveness [2]. The ability of a leader to effectively lead their team is of paramount importance to the success of any business. Among the numerous variables that impact leadership effectiveness, the role of leader behavior and leadership style has been a prominent topic of discussion. Rather than focusing on the innate abilities of the leader, it would be more beneficial to direct attention to the leader’s observable behaviors. It can be argued that the language used by the leader is an important aspect of leadership behavior [3]. Language, as a carrier of information within human civilization, contains a multitude of individual cognitive and psychological elements [4]. The utilization of first-person pronouns constitutes an indispensable element of the linguistic process. In the past, research on the use of personal pronouns has often focused on the differences between languages or on the prediction of individual psychological states through personal pronouns [5,6]. Nevertheless, as communication between leaders and subordinates in organizations is becoming more frequent, it is currently unclear whether there is a correlation between the use of first-person pronouns and leadership effectiveness. In light of these considerations, this paper seeks to investigate the impact of first-person pronouns on leadership effectiveness. This inquiry has the potential to contribute to the expansion of research in the domains of leadership language and leadership effectiveness, offering a novel perspective for enhancing leadership effectiveness. According to relational leadership theory, it is the relationship rather than power or dominance between the leader and the employee that determines the leader’s influence [7]. This suggests that the focus on leadership should not only be on the dichotomy of the leader–employee relationship, but also on the interactive process of the dichotomy, and that leadership effectiveness should be understood in the context of leader–employee communication. Therefore, this study aims to investigate the influence of the utilization of first-person pronouns in leader–employee interactions on leadership effectiveness through a machine learning approach.

The discussion of the role of language in machine learning inevitably leads to the topic of natural language processing (NLP). The advancement of NLP technology has led to an increase in the number of studies that investigate the individual and social psycho-cognitive and behavioral patterns embedded in language and text [8]. In comparison with the time-consuming and labor-intensive methods of questionnaires and experiments, which frequently present researchers with significant challenges in directly collecting self-reported or real behavioral data from a specific group of people within a limited timeframe, the Word Embedding Association Test (WEAT) offers a solution that overcomes the sampling bias inherent in traditional episodic measurements. It provides a more direct, accurate, and rapid measurement of the implicit processes associated with the variables in question [9]. The advent of WEAT in natural language processing techniques has furnished us with instruments to investigate the collective psychological cognition present in language and text, and has enabled us to substantiate the implicit connection between the first-person singular and plural pronouns and leadership effectiveness. The aims of this research are to analyze the implicit relationship between first-person pronouns and leadership effectiveness, more specifically, to investigate the group cognitive relationship between the first-person pronouns “I” and “we” and leadership effectiveness, and to reveal how leaders influence leadership effectiveness through the use of first-person pronouns.

This study employed word vector and sentence vector analysis of the first-person pronouns “I” and “we” and leadership effectiveness using the WEAT to explore the group cognitive links between first-person pronouns and leadership effectiveness. The selection of first-person pronouns in a leader’s discourse can have significant ramifications. A multitude of factors can influence leadership effectiveness; however, traditional-scale tools are unable to detect such implicit associations. The findings of this study have significant implications for organizational contexts. Improvement in leadership effectiveness represents a crucial factor in the sustainable development of both leadership and organizations. Leader–employee interactions are a ubiquitous phenomenon. Despite the existence of numerous norms governing leader behavior in organizational contexts, there remains a necessity to regulate the language of leadership. Employees may perceive leadership effectiveness largely through the leader’s choice of first-person pronouns.

The remainder of the paper is structured as follows. Related works are presented in Section 2, which contains past research on the relationship between first-person pronouns and leadership effectiveness, as well as an introduction and application of the WEAT technique. The calculation methodology, the choice of model, and the selection of word and sentence vector database are showcased in Section 3. The detailed results and analysis of the WEAT are illustrated in Section 4. Finally, the research is discussed and concluded in Section 5 and Section 6, which contain the contributions and limitations of the study and provide ideas for subsequent research.

2. Literature Review

2.1. The Relationship between First-Person Pronouns and Leadership Effectiveness

The most frequently utilized words in everyday communication are personal pronouns, which account for approximately 14% of all words used by individuals [10]. These words are arguably more socially oriented than any other type of linguistic element. The embodied cognition theory posits a close relationship between the human body, psyche, and external environment. When individuals process language, they activate their bodily experiences in the world [11]. This implies that the process of processing personal pronouns is associated with the individual’s perception of their surroundings. First-person pronouns are pervasive in human communication, being employed in a multitude of contexts, including spoken and written discourse, as well as in the creation of blogs. The use of first-person pronouns is partly a reflection of status-related psychological states. Those who are more self-focused tend to use the singular form of first-person pronouns more frequently, while those who are more collectively minded tend to use the first-person plural pronouns more frequently [12]. It can be argued that the use of first-person pronouns indicates whether we refer to ourselves as an individual or as part of a group. The use of the first-person pronoun in the singular emphasizes the self as a unique individual, whereas the use of the first-person pronoun in the plural emphasizes embeddedness in social relationships. From a psychological perspective, we have chosen to focus on first-person pronouns for the following additional reasons. Firstly, people use first-person pronouns with varying frequency, and these individual differences are relatively stable across contexts and time [13,14]. Secondly, individuals have minimal control or recollection over the manner and timing of their use of first-person pronouns [15]. Previous research has also examined the association between the use of singular and plural first-person pronouns and personality traits and other individual-level characteristics, including depressive tendencies, narcissism, and individualism [16,17,18,19].

For leaders, cognitive psychoactive verbs when paired with first-person pronouns in the plural indicate not only the cognitive, affective, and social commitment of the speaker, but also the political institutions, groups, organizations, and governments that he or she represents. Furthermore, the use of “we” as the subject of an utterance suggests that the speaker assumes full or greater responsibility for the correctness of the proposition [20,21]. In other words, the leader’s proclivity for employing first-person pronouns can be seen to reflect the leader’s personal traits and to express their emotions towards the organization and its members. Consequently, over time, organizational members will become more attuned to the leader’s implicit personality traits, which will influence the nature of the leader’s leadership style [22]. The use of first-person pronouns also reflects social status, with those who use the first-person plural pronoun “we” being more likely to be elected as leaders in a global online discussion forum than those who do not [23]. This indicates that individuals who utilize the pronoun “we” with greater frequency are more likely to garner support from others, which subsequently confers a higher social status. Furthermore, increased organizational status is associated with the emergence of more supportive and trusting leaders [24]. Also, more empathetic leaders utilize the first-person plural pronouns more frequently than the first-person singular pronouns, which also impacts the extent to which employees regard the leader in a positive light [25].

In conclusion, previous research has demonstrated that the use of first-person pronouns reflects personal traits, and that the different ways of using first-person pronouns in the singular and plural also facilitate language processing in language receivers, which in turn affects the psychological perceptions of the receivers. Nevertheless, the majority of these studies employed self-report scales, which introduces a sampling bias that compromises the validity of the measurements and the authenticity of the results.

2.2. Word Embedding Association Test

WEAT was first proposed in 2017 by Caliskan et al. [9]. The test measures the relative degree of association between target words and attribute words in natural language by calculating the difference in word similarity between two sets of target words and two sets of attribute words. In addition, Caliskan et al (2017) proposed SC-WEAT (single-category WEAT), a method that investigates the connection between a single category of target words and two categories of attribute words. The advantages of WEAT are as follows. First is the greater objectivity of its analyses. Trained by machine learning algorithms, language is analyzed without subjective human reports and with little human intervention. Second, the sample is more representative. Word vectors are usually trained from large-scale corpora, and the analyses are representative of the population as a whole. Third, the research topic is flexible. With the existing word vector database, we can analyze the implicit associations between any words or concepts, and thus flexibly select the research themes.

The proposal of WEAT and SC-WEAT has led to a series of studies on psychosocial phenomena. Regarding the generation of psychosocial phenomena, DeFranza et al. (2020) categorized 45 languages into two distinct groups: gendered language and genderless language [26]. The calculations of gender bias WEAT scores for each language indicate that gendered languages exhibit greater gender bias. This further suggests that the grammatical rules present in gendered languages may contribute to the production of social bias. Regarding the group-level psychosocial characteristics, in Toney-Wails and Caliskan’s study, the SC-WEAT method was employed to examine the relative degree of association between the target word occupation and two types of attribute word gender (male, female). The results demonstrated that occupation was more strongly associated with males than with females [27]. Furthermore, a similar approach enables the analysis of personality traits as perceived by others. Bhatia et al. (2018) calculated the cosine similarity between word vectors for the names of U.S. presidential candidates Donald Trump and Hillary Clinton, and word vectors for various personality traits (e.g., morality, competence, gentleness) were computed by analyzing a publicly available news corpus to measure the personality traits of these two candidates, as perceived by the U.S. mass media [28]. The past research indicates that the WEAT and SC-WEAT can therefore be extended beyond the use of scale tools to explore the psychological perceptions of a trait by mining and training on the extensive network data that are already available.

Furthermore, the corpus trained in different languages or in different eras also yields disparate results [29]. Li et al. (2020) monitored potential semantic patterns over 150 years and discovered that the frequency of the word “risk” has fluctuated over time, accompanied by an increase in its negative connotations [30]. Hamamura et al. (2021) conducted corpus training in the Chinese context and discovered that the words “individualism” and “collectivism” were associated with attribute words in China from 1950 to 2000. Collectivism was linked to the attribute words “positive”, “negative”, and “achievement” from 1950 to 2000 in China [31]. In conclusion, it can be posited that the WEAT based on a large-scale corpus trained in a specific context can reflect the cognitive, psychological, and behavioral information of individuals in a multitude of specific contexts.

3. Method

3.1. Data Collection

3.1.1. Selection of Word/Sentence Vector Database

In order to reflect the cognitive attitude of the group more accurately, a large number of corpora are needed in the training process of word embedding [32]. Therefore, we choose the word vector databases based on a large amount of corpus training. Given that different corpora may yield disparate results pertaining to group cognition, we made distinct cultural considerations in the selection of word vector database, opting for one Chinese word vector database and one English word vector database for word vector analysis [33]. Based on these two criteria, we selected two word vector databases, Beijing Normal University (BNU) word vector database and Google News word vector database. In order to reflect the current state of Chinese people’s social cognition, the BNU word vector database was selected as the Chinese word vector database. The BNU word vector database is a Chinese word vector database that was trained in 2018 and it is the most complete Chinese word vector database available. It was constructed using corpora from various sources, including Baidu Encyclopedia, Chinese Wikipedia, and People’s Daily. The BNU word vector database comprises 2668 million data points [34]. The English word vector database was selected as the Google News word vector database, which contains 3 million pieces of data collected through the Google News platform; this means that it contains a great deal of information about the psychology and behavior of the English-speaking community and is able to accurately describe the psychology and behavior of the English-speaking community [35].

In order to more fully exploit the contextual relationship of word vectors, we use the BERT model to obtain sentence vectors to characterize leadership effectiveness [36]. Because the BERT model is very expensive to train, this study directly adopts the Chinese Wikipedia BERT model, which is publicly available from Google, to complete the subsequent analysis task [37]. Through it, it can help us better verify the relationship between first-person pronouns and leadership effectiveness.

3.1.2. Selection of Words and Sentences

The purpose of this study is to investigate the relationship between first-person singular and plural pronouns and leadership effectiveness. Therefore, we chose the first-person pronouns “I” and “we” as attribute words and “leadership effectiveness” as the target word. Following the process of Dhingra et al. [38], for the selection of words related to the target word “leadership effectiveness”, we firstly selected 110 articles on Google Scholar with the keyword of leadership effectiveness from 2000 to 2023, and collected the definitions and dimensions of leadership effectiveness in each article as keywords for the word frequency analysis, and obtained 111 keywords. Figure 1 shows the most frequent words of “leadership effectiveness”. Finally, we selected the ten words with the highest word frequency as the words describing “leadership effectiveness”, including “performance”, “satisfaction”, “commitment”, “competence”, “objective”, “organization”, “behaviour”, “effectiveness”, “achievement”, “attitude”.

In the BNU word vector database, the attribute words were the Chinese first-person singular pronoun “I” and the first-person plural pronoun “we”. In the Google News word vector database, we used the English words “I” and “We” as attribute words, in which the English first-person singular pronoun “I” included “I\me\my\mine” and the first-person plural pronoun “We” included “We\us\our\ours”. The ten words that were used as target words for “leadership effectiveness” in both corpora included “performance”, “satisfaction”, “commitment”, “competence”, “objective”, “organization”, “behaviour”, “effectiveness”, “achievement”, “attitude”. For the selection of sentences in the Chinese Wikipedia BERT sentence vector database, following the process of Guo and Caliskan [39], we referred to the leadership effectiveness scale developed by previous researchers, in which 37 sentences related to “leadership effectiveness” were selected, including sentences such as “As a manager, I manage effectively” [40,41].

3.2. Calculation Methodology

The word embedding association test is a method in natural language processing that uses high-dimensional vectors to represent words. It turns a word into a vector by bringing the word embedding into the vector space, called the word vector [42]. Word embedding training is the process of predicting a target word in a fixed-length segment of text using the rest of the text except for the meaning of the target word, and this training can help to discover relationships between words. The CBOW model is a neural network model for generating word vectors, and given a word’s context, the model can perform word prediction [43]. Compared with traditional count-based or matrix decomposition-based methods to generate word vectors, the CBOW model can be trained using large-scale corpora and learn high-quality dense vectors with low dimensionality (usually between 50 and 300), saving storage space and computational resources. At the same time, the CBOW model is also able to capture complex relationships between words, such as synonyms, antonyms, and analogies. Therefore, in this paper, we adopt the CBOW model for word embedding of selected words. Both BNU word vector database and Google News word vector database used the CBOW model to generate word vectors.

Cosine similarity is the cosine of the angle between two word vectors existing in vector space, which measures the semantic relatedness that exists between two words [44]. This means that measuring and focusing on cosine similarity helps us to understand the implicit relationship between two words. The value of cosine similarity lies in the range of [−1, 1], which means that if the cosine similarity is 1, the two words measured are completely related, and if the cosine similarity is 0, the two words measured are completely unrelated, and the cosine similarity is seldom negative in practical scenarios. The specific formula for cosine similarity is shown in Equation (1), where A and B represent two different word vectors:

(1)$similarity(A,B)=cos(\overrightarrow{A},\overrightarrow{B})={\displaystyle \frac{{\sum }_{i=1}^n(A_i\times B_i)}{\sqrt{{\sum }_{i=1}^n{A}_i^2}\times \sqrt{{\sum }_{i=1}^n{B}_i^2}}}$

SC-WEAT is used to calculate the similarity difference between the target word of a single category and the attribute words of both categories, and the SC-WEAT score indicates the relative semantic similarity. Therefore, the higher the SC-WEAT score, the stronger the relative connection between the target word and the attribute word. When the SC-WEAT score is positive, it means that the target word w is more closely related to the attribute word A, and when it is negative, it means that the target word w is more closely related to the attribute word B. In this study, w is “leadership effectiveness”, A is “I”, and B is “we”. The specific calculation formula is shown in Equation (2):

(2)$es(\overrightarrow{w},A,B)={\displaystyle \frac{mea{n}_{a\in A}cos(\overrightarrow{w},\overrightarrow{a})-mea{n}_{b\in B}cos(\overrightarrow{w},\overrightarrow{b})}{std\_de{v}_{\chi \in A\cup B}cos(\overrightarrow{w},\overrightarrow{\chi })}}$

Using the above calculations, it is possible to explore the implicit relationship between first-person pronouns and leadership effectiveness in the Chinese context through a natural language approach. In this study, the CBOW model was used to train a word vector model using the BNU word vector database, which was used to calculate the relationship between attribute words and target words. The attribute words included the first-person singular and plural pronouns “I” and “we”, and the target words were ten words related to “leadership effectiveness”. The cosine similarity and SC-WEAT of “I-leadership effectiveness” and “We-leadership effectiveness” were calculated, respectively, and finally, we repeated the validation using the Google News word vector database.

The BERT model is a bidirectional self-encoding semantic model proposed by Google in 2018, which solves the defect of unidirectionality of semantics in the traditional model, and achieves the best results in 11 tasks, such as question answering, natural language inference, and so on, and it will be the most important bit of progress in the field of natural language approaches in the near future [45]. The BERT model introduces two tasks in the training process; the first task is to train the language model by using the MaskLM approach. When inputting a sentence, some words are randomly selected for prediction, and then they are replaced with a special symbol. After that, we let the model learn the words that should be filled in at these places according to the given labels. The second task adds an extra sentence-level continuity prediction task to the bidirectional language model, i.e., predicting whether the two text segments input to BERT are consecutive texts, and the introduction of this task can better enable the model to learn the relationship between consecutive text segments. BERT can do concurrent execution, extracting relational features of the words in the sentence at the same time, and extracting relational features at many different levels, thus reflecting more comprehensively the relationship between the words and the segments. Compared to word embedding through the CBOW model, the BERT model can obtain word meanings based on the context of the sentence, thus avoiding ambiguity. The BERT model can be used to encode sentences for semantic similarity calculation.

In this paper, we directly adopted the Wikipedia open-source BERT model for analysis and generated vectors of first-person singular and plural pronouns and leadership effectiveness target sentences, and analyzed the implicit relationships embedded in large-scale texts by calculating the cosine similarity relationship between the vectors. We inputted the attribute words and target sentences into the BERT model, stored the output sentence vectors in groups, calculated the cosine similarity between the target words and sentence group vectors, computed the SC-WEAT scores, and carried out subsequent analyses.

The overall calculation process of this paper is shown in Figure 2. Firstly, we select the target words and sentences that can reflect leadership effectiveness, as well as the related first-person pronouns. Then, we word-embedded these words and sentences. Considering the differences between different cultures, for the word embedding model, we used two word vector databases, which contain the CBOW model trained based on the BNU corpus and the CBOW model trained based on the Google News corpus. Moreover, in order to measure the effect of first-person pronouns on leadership effectiveness more comprehensively, we also used the BERT model trained by the Wikipedia corpus to word-embed relevant utterances to obtain sentence vectors. Finally, we calculated the cosine similarity and SC-WEAT scores of the target vectors and attribute vectors obtained from the three models to explore the implicit association of first-person pronouns on leadership effectiveness.

4. Results

First of all, we establish a word vector computational model through the data of the BNU word vector database to verify the connection between the first-person singular and plural pronouns and leadership effectiveness. In the study, there are two first-person pronouns, “I” and “we”, and ten words representing “leadership effectiveness”, and we calculate the cosine of each attribute word and the word representing “leadership effectiveness”, respectively.

After calculating the cosine similarity between each attribute word and the word representing “leadership effectiveness”, we can obtain a total of 2 × 10 cosine similarities (see Figure 3). The numbers on the x-axis in Figure 3 are each of the ten words representing “leadership effectiveness”, and the y-axis represents the cosine similarity values between each word and the first-person pronouns. Figure 3 shows that the cosine similarity between the first-person plural pronouns and each of the ten representative words of “leadership effectiveness” is higher. The means of the cosine similarity between the first-person singular and plural pronouns and leadership effectiveness are shown in Figure 4. The results of the independent samples t-test indicated that the cosine similarity of “I-leadership effectiveness” ($M=0.07$, $SD=0.21$) was significantly smaller than the cosine similarity of “We-leadership effectiveness” ($M=0.10$, $SD=0.14$), $t\left(18\right)=-3.495$, $p<0.05$, $Cohen’sd=-0.168$, $95\%CI[-0.01,-0.05]$.

SC-WEAT ($score=-1.271$) also shows that “leadership effectiveness” is more closely related to “we”. The above results show that the word vector index of the BNU word vector database verifies the implicit relationship between leadership effectiveness and the first-person plural pronouns. The results indicated that in the Chinese context, people are more likely to think that those who often use “we” as the subject have more leadership effectiveness, while those who use “I” as the subject are more likely to have lower leadership effectiveness.

Similarly, we modeled word vectors with data from the Google News word vector database to re-examine the implicit relationship between the first-person singular and plural pronouns and leadership effectiveness. After calculating the cosine similarity between each attribute word and the word representing “leadership effectiveness”, we can obtain a total of 2 × 10 cosine similarities (see Figure 5). The numbers on the x-axis in Figure 5 are each of the ten words representing “leadership effectiveness”, and the y-axis represents the cosine similarity values between each word and the first-person pronouns. Figure 5 shows that the cosine similarity between the first-person plural pronouns and each of the ten representative words of “leadership effectiveness” is higher. The means of the cosine similarity between the first-person singular and plural pronouns and leadership effectiveness are shown in Figure 4. The results of the independent samples t-test show that the cosine similarity of “I-leadership effectiveness” ($M=0.011$, $SD=0.083$) is significantly smaller than “We-leadership effectiveness” ($M=0.022$, $SD=0.091$), $t\left(78\right)=-5.647$, $p<0.001$, $Cohen’sd=-0.126$, $95\%CI[-0.014,-0.007]$.

SC-WEAT ($score=-1.077$) also validated the direct implicit relationship between “leadership effectiveness” and “we”. The results show that the word vector index of the Google News word vector database verifies the implicit relationship between leadership effectiveness and the first-person plural pronouns. The results indicated that in the English context, people also think that those who often use “we” as the subject have more leadership effectiveness, while those who use “I” as the subject are more likely to have lower leadership effectiveness.

Then, we calculate the sentence vectors of leadership effectiveness through the BERT model to verify the implicit relationship between the first-person pronoun and leadership effectiveness again. By generating vectors representing the first-person pronouns “we” and “I” and the target sentences of “leadership effectiveness”, and calculating the cosine similarity between the vectors separately, we finally obtain 2 × 37 cosine similarity values (see Figure 6). The numbers on the x-axis in Figure 6 are each of the 37 sentences representing “leadership effectiveness”, and the y-axis represents the cosine similarity values between each sentence and the first-person pronouns. Figure 6 shows that the cosine similarity between the first-person plural pronouns and each of the 37 sentences representing “leadership effectiveness” is higher. The means of the cosine similarity between the first-person singular and plural pronouns and leadership effectiveness are shown in Figure 4. The results of the independent samples t-test showed that the cosine similarity of “I-leadership effectiveness” ($M=0.424$, $SD=0.06$) is significantly smaller than “We-leadership effectiveness” ($M=0.465$, $SD=0.055$), $t\left(72\right)=-3.027$, $p<0.01$, $Cohen’sd=-0.712$, $95\%CI[-0.067,-0.013]$. In other words, the cosine similarity of the sentence vectors, which are semantically closer to the reality of the organizational context, suggests that the relationship of “we-leadership effectiveness” is greater than the relationship of “I-leadership effectiveness”.

5. Discussion

How to promote leadership effectiveness has been an important theoretical and measurement issue, but previous researchers have mostly measured its exogenous processing through scales, ignoring the implicit effects on language receivers that exist during the use of first-person pronouns in language. Is there a measurable implicit processing of leadership effectiveness, and is the relationship between first-person plural pronoun and leadership effectiveness significantly higher than that between first-person singular pronoun and leadership effectiveness? This study attempts to give an answer through the WEAT, which verifies the implicit relationship between the first-person plural pronoun and leadership effectiveness through word vector and sentence vector model analysis. Specifically, word vector analysis based on the BNU word vector database and the Google News word vector database revealed that the cosine similarity of the word vectors of “I-leadership effectiveness” is significantly smaller than that of “We-leadership effectiveness”. The sentence vector analysis based on the BERT model also shows that the cosine similarity of the sentence vectors of “I-leadership effectiveness” is significantly smaller than the cosine similarity of “We-leadership effectiveness”. The semantic relatedness of “we-leadership effectiveness” is also closer than “I-leadership effectiveness”.

The contributions of this study are as follows. First, through machine learning and WEAT, this study found that online users showed semantic associations between “we” and leadership effectiveness in the implicit expression of text, which strongly proved that the association between first-person pronouns and leadership effectiveness was implicitly present and could be accurately measured. This study validates to some extent that leaders’ use of the plural form of the first-person pronoun can lead to employees perceiving their leaders as more effective. Cross-cultural influences were also taken into account in the study, as both Chinese and English corpora were used and analyzed to obtain the same results. This suggests that people who often use “we” in both Chinese and English regional cultures are perceived to have leadership effectiveness. This implicit association may be due to the following reasons. First, the use of “we” is perceived as a positive and friendly behavior, and a leader who possesses this type of behavior will increase employees’ liking for him or her, leading to positive perceptions and associating the leader with leadership effectiveness. Based on social identity theory, individual identification with a group is the basis for group behavior [46]. By using the first-person plural pronoun “we”, the leader stimulates the group attributes of the employees, which makes the social ties between individuals stronger, strengthens the collective identity, and makes the employees willing to support the leader and cognitively trust the leader’s abilities.

In addition, this study overcame the sampling bias of traditional episodic measures, as well as the time-consuming and labor-intensive problems of traditional implicit association tests (IATs), to achieve a more direct, accurate, and rapid measurement of the implicit processes of leadership effectiveness. No previous study has used machine learning or WEAT to measure the correlates of leadership effectiveness, so the research methodology used in this study is methodologically innovative in the measurement of the relationship between leadership efficacy and other variables. It is also informative for research on the measurement of cognitive attitudes in other groups. By using a machine learning model trained on the word vector database of Internet text corpora, we can determine the implicit psychological cognition of Internet users. In the past, it was often difficult for researchers to collect self-reported or real behavioral data of a specific group in a short time, so the word-embedded association test helps researchers obtain the target research data conveniently and effectively.

Finally, using a machine learning approach, we validate the influence of the leader–employee interaction process on leadership effectiveness in relational leadership theory, and provide ideas for how relational leadership theory can be better applied to real organizational situations by focusing on a more specific part of the communication, the way in which first-person pronouns are used. It opens up a new way of thinking for enterprises to improve leadership effectiveness and promote their sustainable development. The sustainable development of an organization is based on a macro view of the trend of the times and a micro view of each employee striving towards the same vision. Ensuring that employees have the same goal depends on the effectiveness of leadership. Through the use of “we”, employees will trust the leader’s leadership effectiveness in terms of implicit cognition. Combined with the leader’s own efforts to promote the fusion of implicit cognition and external ability, it is the inexhaustible driving force for the development of the enterprise.

Limitations and Future Directions

There are also some limitations in this study. First, only groups in Chinese and English regions were considered in the study, and the results may not be applicable to other language regions. Moreover, it was difficult to collect demographic information and personality information of users in this study, and we do not know whether this implicit cognitive relationship exists in a specific group, so we need to conduct targeted research to obtain cognitive information of a specific group. Therefore, subsequent research can collect data for different languages and groups in specific scenarios to build a more complete analytical model.

In addition, because leadership effectiveness has a strong contextual and group nature, the measurements of online big data on which this study is based do not capture measurements of web users’ or non-users’ real behavior offline. Therefore, the correlation between leadership effectiveness and first-person pronouns measured in this study may have contextual bias. And it is worth noting that our word vector database contains data trained on news, Google, Baidu, and other news media texts, and it is likely that the texts in these platforms have been screened and filtered and have been circulated in a way that is acceptable to the general public, and thus, the trained data may already have certain tendencies that do not reflect the completely true psychology of the population. In order to better understand whether the relationship between first-person pronouns and leadership effectiveness can be validated in spontaneous messages, in the future, we could collect more data on users’ personal statements, including their personal social media data, which might be closer to the real situation.

Furthermore, this study focused on group-level perceptions and did not explore the individual leader level. The mismatch between leaders’ implicit perceptions and explicit behaviors may have other implications for leadership effectiveness, as leaders may know inwardly that it is better to use “we”, but still use “I” in communication. At the same time, there are different tasks in the leadership process, and the use of different first-person pronouns in different task scenarios may also show differences in leadership effectiveness. Therefore, further consideration of leaders’ personal-level factors and situational factors is needed in future research.

6. Conclusions

Leadership effectiveness is a fundamental need for successful organizations, and ways to increase leadership effectiveness exist in every aspect of daily work and life, including leaders’ preferences for the singular and plural forms of first-person pronouns in their verbal expressions. Using machine learning and WEAT, this study found that online users’ implicit textual expressions showed a strong semantic association between the first-person plural pronoun and leadership effectiveness. Leaders were perceived to be more effective when they used the word “we”. In addition, the study demonstrated the feasibility of using the WEAT to measure leadership effectiveness or other organizational behavioral variables, which is an extension of the traditional-scale instrument. The results also remind us that the seemingly vast and unorganized online world contains users’ psychological perceptions, and that word-embedded association tests provide a new tool for exploring social perceptions.

Author Contributions

Conceptualization, Q.Y. and Y.Z.; methodology, Q.Y.; software, J.C.; validation, Q.Y., Y.Z. and J.C.; formal analysis, Q.Y.; writing—original draft preparation, Q.Y.; writing—review and editing, Q.Y. and Y.Z.; visualization, J.C.; supervision, Y.Z. All authors have read and agreed to the published version of the manuscript.

Institutional Review Board Statement

Not applicable.

Informed Consent Statement

Not applicable.

Data Availability Statement

The data can be made available on request.

Acknowledgments

The authors extend their appreciation to the organizations that provided open-source word vector and sentence vector data.

Conflicts of Interest

The authors declare no conflicts of interest.

Footnotes

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Figures

Figure 1. A visual representation of the most frequent words as a word cloud map for leadership effectiveness.

Figure 2. The flowchart of the calculation methodology.

Figure 3. The cosine similarity results based on BNU word vector database.

Figure 4. The cosine similarity of first-person pronoun and leadership effectiveness.

Figure 5. The cosine similarity results based on Google News word vector database.

Figure 6. The cosine similarity results based on the BERT model.

References

1. Van Knippenberg, B.; Van Knippenberg, D. Leader self-sacrifice and leadership effectiveness: The moderating role of leader prototypicality. J. Appl. Psychol.; 2005; 90, 25. [DOI: https://dx.doi.org/10.1037/0021-9010.90.1.25] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/15641888]

2. Chen, J.C.; Silverthorne, C. Leadership effectiveness, leadership style and employee readiness. Leadersh. Organ. Dev. J.; 2005; 26, pp. 280-288. [DOI: https://dx.doi.org/10.1108/01437730510600652]

3. Awamleh, R.; Gardner, W.L. Perceptions of leader charisma and effectiveness: The effects of vision content, delivery, and organizational performance. Leadersh. Q.; 1999; 10, pp. 345-373. [DOI: https://dx.doi.org/10.1016/S1048-9843(99)00022-3]

4. Reitz, M. Leading questions: Dialogue in organizations: Developing relational leadership. Leadership; 2017; 13, pp. 516-522. [DOI: https://dx.doi.org/10.1177/1742715015617864]

5. Na, J.; Choi, I. Culture and first-person pronouns. Ersonality Soc. Psychol. Bull.; 2009; 35, pp. 1492-1499. [DOI: https://dx.doi.org/10.1177/0146167209343810] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/19713570]

6. Holtzman, N.S. A meta-analysis of correlations between depression and first person singular pronoun use. J. Res. Personal.; 2017; 68, pp. 63-68.

7. Uhl-Bien, M. Relational leadership theory: Exploring the social processes of leadership and organizing. Leadersh. Q.; 2006; 17, pp. 654-676. [DOI: https://dx.doi.org/10.1016/j.leaqua.2006.10.007]

8. Hirschberg, J.; Manning, C.D. Advances in natural language processing. Science; 2015; 349, pp. 261-266. [DOI: https://dx.doi.org/10.1126/science.aaa8685]

9. Caliskan, A.; Bryson, J.J.; Narayanan, A. Semantics derived automatically from language corpora contain human-like biases. Science; 2017; 356, pp. 183-186. [DOI: https://dx.doi.org/10.1126/science.aal4230]

10. Kacewicz, E.; Pennebaker, J.W.; Davis, M.; Jeon, M.; Graesser, A.C. Pronoun use reflects standings in social hierarchies. J. Lang. Soc. Psychol.; 2014; 33, pp. 125-143. [DOI: https://dx.doi.org/10.1177/0261927X13502654]

11. Niedenthal, P.M.; Barsalou, L.W.; Winkielman, P.; Krauth-Gruber, S.; Ric, F. Embodiment in attitudes, social perception, and emotion. Personal. Soc. Psychol. Rev.; 2005; 9, pp. 184-211. [DOI: https://dx.doi.org/10.1207/s15327957pspr0903_1]

12. Davis, D.; Brock, T.C. Use of first person pronouns as a function of increased objective self-awareness and performance feedback. J. Exp. Soc. Psychol.; 1975; 11, pp. 381-388. [DOI: https://dx.doi.org/10.1016/0022-1031(75)90017-7]

13. Robbins, M.L.; Jonnalagadda, P.; Spahr, C.M. Rebalancing social & personality psychology methods: The case for naturalistic observation. Soc. Personal. Psychol. Compass; 2024; 18, e12917.

14. Pennebaker, J.W.; King, L.A. Linguistic styles: Language use as an individual difference. J. Personal. Soc. Psychol.; 1999; 77, 1296. [DOI: https://dx.doi.org/10.1037/0022-3514.77.6.1296]

15. Pennebaker, J.W.; Chung, C.K. Expressive writing, emotional upheavals, and health. Foundations of Health Psychology; Oxford University Press: Oxford, UK, 2007; pp. 263-284.

16. Yin, Y.; Wakslak, C.J.; Joshi, P.D. “I” am more concrete than “we”: Linguistic abstraction and first-person pronoun usage. J. Personal. Soc. Psychol.; 2022; 122, 1004. [DOI: https://dx.doi.org/10.1037/pspa0000285]

17. Twenge, J.M.; Campbell, W.K.; Gentile, B. Changes in pronoun use in American books and the rise of individualism, 1960–2008. J. Cross-Cult. Psychol.; 2013; 44, pp. 406-415. [DOI: https://dx.doi.org/10.1177/0022022112455100]

18. Carey, A.L.; Brucks, M.S.; Küfner, A.C.; Holtzman, N.S.; Back, M.D.; Donnellan, M.B.; Pennebaker, J.W.; Mehl, M.R. Narcissism and the use of personal pronouns revisited. J. Personal. Soc. Psychol.; 2015; 109, e1. [DOI: https://dx.doi.org/10.1037/pspp0000029]

19. Zimmermann, J.; Brockmeyer, T.; Hunn, M.; Schauenburg, H.; Wolf, M. First-person pronoun use in spoken language as a predictor of future depressive symptoms: Preliminary evidence from a clinical sample of depressed patients. Clin. Psychol. Psychother.; 2017; 24, pp. 384-391. [DOI: https://dx.doi.org/10.1002/cpp.2006]

20. Fetzer, A. “And I Think That Is a Very Straightforward Way of Dealing with It” The Communicative Function of Cognitive Verbs in Political Discourse. J. Lang. Soc. Psychol.; 2008; 27, pp. 384-396. [DOI: https://dx.doi.org/10.1177/0261927X08322481]

21. Dunmire, P.L. Political discourse analysis: Exploring the language of politics and the politics of language. Lang. Linguist. Compass; 2012; 6, pp. 735-751. [DOI: https://dx.doi.org/10.1002/lnc3.365]

22. Kalish, Y.; Luria, G. Traits and time in leadership emergence: A longitudinal study. Leadersh. Q.; 2021; 32, 101443. [DOI: https://dx.doi.org/10.1016/j.leaqua.2020.101443]

23. Cassell, J.; Huffaker, D.; Tversky, D.; Ferriman, K. The language of online leadership: Gender and youth engagement on the Internet. Dev. Psychol.; 2006; 42, 436. [DOI: https://dx.doi.org/10.1037/0012-1649.42.3.436]

24. Hogue, M.B.; Yoder, J.D.; Ludwig, J. Increasing initial leadership effectiveness: Assisting both women and men. Sex Roles; 2002; 46, pp. 377-384. [DOI: https://dx.doi.org/10.1023/A:1020457312806]

25. Meinecke, A.L.; Kauffeld, S. Engaging the hearts and minds of followers: Leader empathy and language style matching during appraisal interviews. J. Bus. Psychol.; 2019; 34, pp. 485-501. [DOI: https://dx.doi.org/10.1007/s10869-018-9554-9]

26. DeFranza, D.; Mishra, H.; Mishra, A. How language shapes prejudice against women: An examination across 45 world languages. J. Personal. Soc. Psychol.; 2020; 119, 7. [DOI: https://dx.doi.org/10.1037/pspa0000188]

27. Toney-Wails, A.; Caliskan, A. Valnorm quantifies semantics to reveal consistent valence biases across languages and over centuries. arXiv; 2020; arXiv: 200603950

28. Bhatia, S.; Goodwin, G.P.; Walasek, L. Trait associations for Hillary Clinton and Donald Trump in news media: A computational analysis. Soc. Psychol. Personal. Sci.; 2018; 9, pp. 123-130. [DOI: https://dx.doi.org/10.1177/1948550617751584]

29. Garg, N.; Schiebinger, L.; Jurafsky, D.; Zou, J. Word embeddings quantify 100 years of gender and ethnic stereotypes. Proc. Natl. Acad. Sci. USA; 2018; 115, pp. E3635-E3644. [DOI: https://dx.doi.org/10.1073/pnas.1720347115]

30. Li, Y.; Hills, T.; Hertwig, R. A brief history of risk. Cognition; 2020; 203, 104344. [DOI: https://dx.doi.org/10.1016/j.cognition.2020.104344]

31. Hamamura, T.; Chen, Z.; Chan, C.S.; Chen, S.X.; Kobayashi, T. Individualism with Chinese characteristics? Discerning cultural shifts in China using 50 years of printed texts. Am. Psychol.; 2021; 76, 888. [DOI: https://dx.doi.org/10.1037/amp0000840]

32. Torregrossa, F.; Allesiardo, R.; Claveau, V. A survey on training and evaluation of word embeddings. Int. J. Data Sci. Anal.; 2021; 11, pp. 85-103. [DOI: https://dx.doi.org/10.1007/s41060-021-00242-8]

33. Bao, H.W.S. PsychWordVec: Word Embedding Research Framework for Psychological Science. Computer Software. 2022; Available online: https://psychbruce.github.io/PsychWordVec/ (accessed on 1 September 2023).

34. Beijing Normal University Word Vector Database. Available online: https://github.com/Embedding/Chinese-Word-Vectors (accessed on 26 April 2018).

35. Google News Word Vector Database. Available online: https://code.google.com/archive/p/word2vec (accessed on 30 July 2013).

36. Devlin, J.; Chang, M.W.; Lee, K. Bert: Pre-training of deep bidirectional transformers for language understanding. arXiv; 2018; arXiv: 1810.04805

37. Chinese Wikipedia BERT Sentence Vector Database. Available online: https://huggingface.co/google-bert/bert-base-chinese (accessed on 15 November 2018).

38. Dhingra, B.; Zhou, Z.; Fitzpatrick, D.; Muehl, M.; Cohen, W.W. Tweet2vec: Character-based distributed representations for social media. arXiv; 2016; arXiv: 160503481

39. Guo, W.; Caliskan, A. Detecting emergent intersectional biases: Contextualized word embeddings contain a distribution of human-like biases. arXiv; 2019; arXiv: 2006.03955

40. Denison, D.R.; Hooijberg, R.; Quinn, R.E. Paradox and performance: Toward a theory of behavioral complexity in managerial leadership. Organ. Sci.; 1995; 6, pp. 524-540. [DOI: https://dx.doi.org/10.1287/orsc.6.5.524]

41. House, R.J.; Podsakoff, P.M. Leadership effectiveness: Past perspectives and future directions for research. Organizational Behavior; Routledge: London, UK, 2013; pp. 55-92.

42. Mikolov, T.; Chen, K.; Corrado, G.; Dean, J. Efficient estimation of word representations in vector space. arXiv; 2013; arXiv: 1301.3781

43. Liu, B. Text sentiment analysis based on CBOW model and deep learning in big data environment. J. Ambient. Intell. Humaniz. Comput.; 2020; 11, pp. 451-458. [DOI: https://dx.doi.org/10.1007/s12652-018-1095-6]

44. Lenci, A. Distributional models of word meaning. Annu. Rev. Linguist.; 2018; 4, pp. 151-171. [DOI: https://dx.doi.org/10.1146/annurev-linguistics-030514-125254]

45. Singh, M.; Jakhar, A.K.; Pandey, S. Sentiment analysis on the impact of coronavirus in social life using the BERT model. Soc. Netw. Anal. Min.; 2021; 11, 33. [DOI: https://dx.doi.org/10.1007/s13278-021-00737-z]

46. Turner, J.C.; Oakes, P.J. The significance of the social identity concept for social psychology with reference to individualism, interactionism and social influence. Br. J. Soc. Psychol.; 1986; 25, pp. 237-252. [DOI: https://dx.doi.org/10.1111/j.2044-8309.1986.tb00732.x]