Skip to main content
NCBI home page
The Journals of Gerontology Series B: Psychological Sciences and Social Sciences logoLink to The Journals of Gerontology Series B: Psychological Sciences and Social Sciences
. 2023 Apr 25;78(9):1493–1500. doi: 10.1093/geronb/gbad061

Detecting Narcissism From Older Adults’ Daily Language Use: A Machine Learning Approach

Shiyang Zhang 1,✉, Karen L Fingerman 2, Kira S Birditt 3
Editor: Rodlescia Sneed
PMCID: PMC10461532  PMID: 37098210

Abstract

Objectives

Narcissism has been associated with poorer quality social connections in late life, yet less is known about how narcissism is associated with older adults’ daily social interactions. This study explored the associations between narcissism and older adults’ language use throughout the day.

Methods

Participants aged 65–89 (N = 281) wore electronically activated recorders which captured ambient sound for 30 s every 7 min across 5–6 days. Participants also completed the Narcissism Personality Inventory-16 scale. We used Linguistic Inquiry and Word Count to extract 81 linguistic features from sound snippets and applied a supervised machine learning algorithm (random forest) to evaluate the strength of links between narcissism and each linguistic feature.

Results

The random forest model showed that the top 5 linguistic categories that displayed the strongest associations with narcissism were first-person plural pronouns (e.g., we), words related to achievement (e.g., win, success), to work (e.g., hiring, office), to sex (e.g., erotic, condom), and that signal desired state (e.g., want, need).

Discussion

Narcissism may be demonstrated in everyday life via word use in conversation. More narcissistic individuals may have poorer quality social connections because their communication conveys an emphasis on self and achievement rather than affiliation or topics of interest to the other party.

Keywords: Electronically activated recorder (EAR), Linguistic features, personality

Narcissism is a personality trait characterized by grandiosity, entitlement, exploitativeness, and lack of empathy (Campbell & Foster, 2007). Communication is a fundamental feature of interpersonal connection (Milek et al., 2018), but it has not been widely examined with regard to narcissism. Studies suggest that narcissistic people are self-absorbed, have a sense of superiority, self-entitlement, and grandiosity, and are less sensitive to other people’s emotions (Kjærvik & Bushman, 2021) which may be evident in their communication with others. A narcissistic individual’s self-focused communications may have an adverse effect on the quality of their social connections (Seidman et al., 2020). This self-focused communication may be evident in everyday spoken language; despite an increase in the use of communication technologies among older adults (Pew Research Center, 2021), older adults still largely rely on spoken language in either face-to-face or phone communications (Marengo et al., 2021). As such, examining differences in spoken language between more narcissistic and less narcissistic older adults may provide information on how narcissism affects communication and quality of social relationship.

According to Socioemotional Selectivity Theory, compared to younger people, older adults prioritize high-quality social connections that are beneficial to emotional well-being (Carstensen, 2021; Charles & Carstensen, 2010). In contrast, narcissism may become less adaptive at this life stage where emotional goals are salient (Hill & Roberts, 2012). Although many individuals become less narcissistic with age (Chopik & Grimm, 2019), some older adults remain at relatively high levels of narcissism. The adverse effects of narcissism on social relationships may threaten individuals’ emotional well-being and life satisfaction, particularly among older adults, due to their need for high-quality social connections.

To understand the role of narcissism in late life, it is important to consider definitions of the term. Narcissism can be considered both a pathological disorder and a personality trait. Either form of narcissism may impede communication and social relationships (Weiss & Miller, 2018). The current study considers narcissism as a personality trait to capture daily communication among older adults who vary on narcissism in a nonpathological continuum.

Language as a Marker of Personality

A burgeoning literature examined the associations between linguistic markers and personality traits (Pennebaker, 2011; Stachl et al., 2020). In particular, previous studies considered two main categories of linguistic features, linguistic style, and linguistic content (Tackman et al., 2020).

Style words (e.g., personal pronouns, negations, and auxiliary verbs) represent the ways in which individuals express certain topics and they construct the basic structure of sentences (Stachl et al., 2020). Although style words do not carry information on the content of conversations, it is still informative as it comprises more than half of the words people process in daily conversations (Tausczik & Pennebaker, 2010). Additionally, the use of style words may reflect individuals’ personality because they are produced fluently and almost automatically (Ireland & Mehl, 2014). For example, extraverts used more first-person singular pronouns in text messages (Holtgraves, 2011).

In contrast, linguistic content refers to the topics that people mention in conversations (e.g., family, work, achievement) and is usually reflected in the use of nouns, verbs, adjectives, and adverbs (Ireland & Mehl, 2014). Individuals’ personality traits may also be associated with the use of content words that indicate conversation topics and emotions. For example, extraversion and agreeableness have been linked to higher usage of positive words (e.g., pretty; happy; Augustine et al., 2011; Yarkoni, 2010), whereas neuroticism has been linked to using more negative emotion words (e.g., annoying; Holtgraves, 2011). Relatedly, more conscientious young people used more work-related words and fewer leisure-related words (e.g., YouTube) on their Facebook pages (Schwartz et al., 2013). The current study examines both linguistic style and content to provide a window into older adults’ daily language associated with narcissism.

Narcissism and Language Use

Individuals who are higher in narcissism tend to apply strategies to promote their social status (Grapsas et al., 2020), and this may extend into their communication and language patterns with social partners. We apply two models of narcissism to examine language style and content. The Extended Agency Model of narcissism suggests narcissistic individuals engage in achievement-striving strategies including interpersonal attributes (e.g., confidence, charm), interpersonal strategies (e.g., self-promotion, trophy partners), and intrapsychic strategies (e.g., self-serving bias; Campbell & Foster, 2007). Likewise, the Dynamic Self-Regulatory Processing Model posits that narcissistic people perceive their social partners as sources of external validation and self-enhancement (Morf & Rhodewalt, 2001). Both models stress the interpersonal component of narcissism and it is likely that older adults who are higher in narcissism may use interpersonal communication to fuel their inflated sense of self. For example, narcissistic people may speak in a more assertive way (e.g., use less tentative words—perhaps, maybe) to show their confidence. Likewise, related to the topics of conversations, narcissistic people may be absorbed with and communicate their previous successes (e.g., use words related to achievement—win, champion) to consolidate the sense of superiority.

Previous studies have documented associations between narcissism and style words, especially personal pronouns. A particular focus has been the use of first-person singular pronouns due to narcissistic people’s excessive focus on themselves. An early study found that narcissistic young adults used more first-person singular pronouns and fewer first-person plural pronouns in impromptu monologues (Raskin & Shaw, 1988). However, using recordings of participants’ self-descriptions and group interactions, more recent studies did not observe associations between narcissism and first-person singular pronouns in young adult samples (Carey et al., 2015; Holtzman et al., 2019). In the current study, we examined associations between older adults’ style words (e.g., first-person singular pronouns, first-person plural pronouns) as a means of gauging how they view their relationships with the outside world (Pennebaker, 2011).

Studies also have examined content words that may convey features of narcissism. For instance, one study combined data from 15 studies in which college students spoke in recorded videos (e.g., self-introduction, descriptions of scenes and people) to examine correlations between narcissism and linguistic features. Narcissism was positively associated with words related to sports, second-person pronouns, swear words (Holtzman et al., 2019), and sexual language (Holtzman et al., 2010). Likewise, using a stream-of-consciousness paradigm, a study examined correlations between narcissism and word use. The study found that narcissistic young adults wrote in more open-minded ways and used more agreeable language when asked to describe their thoughts, feelings, and sensations online (Cutler et al., 2021). These paradoxical findings (e.g., swear words, agreeable language) may reflect social desirability motivations and desire to create positive impressions.

In sum, narcissism has been associated with a limited set of style and content linguistic features in young adult samples. Patterns of findings were inconsistent, however, and studies have not examined these issues in late life, when investment in social relationships may generate fewer differences in the use of language, or a different set of linguistic features associated with narcissism.

The Current Study

This study examined associations between narcissism and linguistic features of older adults’ speech in a naturalistic setting. Scholars have noted that narcissistic people may pretend to be more agreeable in public settings (e.g., impromptu speech; Back et al., 2010) compared to how they usually are. The electronically activated recorder (EAR) provides the opportunity to unobtrusively observe ambient sound and capture language as participants go about their daily lives (Mehl, 2017). As such, recording ambient sound may be a better way to understand how narcissism influences individuals’ daily social contacts and relationships.

Although previous studies have examined associations between narcissism and linguistic features (Carey et al., 2015; Holtzman et al., 2019), these correlational analyses have not been able to include narcissism and all linguistic features in one model. Instead, the current study used a machine learning approach to strategically investigate associations between narcissism and daily linguistic features. Machine learning is particularly suitable for vast data sets, as it enables the examination of a wide range of markers and the results of machine learning models are not biased by violations of model assumptions (e.g., multicollinearity, homoscedasticity; Stachl et al., 2020). A burgeoning literature has built machine learning models to make personality trait predictions associated with individual characteristics (e.g., digital footprint, facial expression; Mehta et al., 2020).

Specifically, the current study applied the random forest classifier algorithm to examine the associations. Random forest is an ensemble learning method (i.e., integrating results from multiple models to achieve better model performances) that combines the results of thousands of decision trees (i.e., tree-like models that classify data points into categories based on input features; Breiman, 2001; Sagi & Rokach, 2018). Random forest has been widely used to select and rank features that are the most predictive of a certain outcome (Rohit et al., 2020). This algorithm fits the current study because our objective is to rank the ability of each linguistic feature to predict narcissism. By using random a forest classifier algorithm, we explore the daily linguistic features that are more strongly associated with variability in narcissism and, thus, provide a deeper understanding of the manifestations of narcissism in older adults’ daily conversations. The current study draws on prior literature as a basis for examining linguistic markers of narcissism in late life. However, it is an exploratory study without specific hypotheses due to (a) the use of machine learning, a data-mining approach that applies the deduction reasoning (i.e., from observations to theories) and usually does not need preset hypotheses (Alexander III et al., 2020; Bleidorn & Hopwood, 2019), and (b) conflicting findings in prior literature (e.g., the use of first-person singular pronouns).

Method

Participants and Procedure

Data were from the Daily Experiences and Well-being Study, including 333 community-dwelling adults aged 65–92 in the greater Austin Metropolitan Statistical Area. The study oversampled older adults from high-density underrepresented and lower socioeconomic neighborhoods to represent racial and ethnic diversity as well as a wider range of socioeconomic status. Among the whole sample, over 30% of the participants identified themselves as racial or ethnic minorities.

Participants initially completed a 90- to 120-min face-to-face baseline interview in which they reported on their social ties, well-being, and background characteristics. Participants then carried a study-provided Android device for 5–6 days. An EAR was installed on the Android device to capture participants’ daily language and ambient sound unobtrusively (Mehl, 2017). The EAR was activated for 30 s every 7 min during waking hours for 5–6 days. Narcissism was assessed at the end of the study via a self-report questionnaire. Participants received $50 for completing the baseline interview and another $100 for the daily component (e.g., EAR). The study was approved by the University of Texas at Austin Institutional Review Board.

The current study included 281 English speakers (Mage = 74.07, SDage = 6.47) who completed the measure of narcissism and had valid sound snippets. The excluded participants (n = 52) scored lower on self-rated health (t(331) = −3.32, p = .001), reported worse education (χ2(1,N = 333) = 13.58, p = .001), and were more likely to identify as racial or ethnic minorities (χ2(1,N = 333) = 54.65, p < .001). Sample information is reported in Table 1.

Table 1.

Sample Descriptive Information

Participants (N = 281)
M SD Range
Age 74.07 6.47 65–89
Self-rated healtha 3.62 1.00 1–5
Narcissismb 0.20 0.17 0–0.75
Number of language snippetsc 100.51 52.36 9–301
Proportions
High narcissismd 0.39
Female 0.53
Married 0.59
Racial/ethnic minority 0.25
Education
 High school or less 0.12
 Some college 0.29
 College or more 0.59
Open in a new tab

Notes: SD = standard deviation.

a1 (poor), 2 (fair), 3 (good), 4 (very good), and 5 (excellent).

bContinuous score of narcissism calculated as the proportion of narcissism-consistent responses out of 16 items.

cNumber of recorded sound snippets in which participants talked; 30 s out of 7 min.

dNarcissism dichotomized into high narcissism and low narcissism groups based on the mean value.

Measures

Narcissism

Participants completed the 16-item version of Narcissism Personality Inventory (NPI-16; Ames et al., 2006). For each item, participants chose between a narcissism-consistent response (e.g., I know that I am good because everybody keeps telling me so; scored one) and a narcissism-inconsistent response (e.g., When people compliment me I sometimes get embarrassed; scored zero). We calculated the average score across items to represent narcissism (Cronbach’s α = 0.75). The possible range of the narcissism measure was 0–1. Mahalanobis distance was calculated to detect insufficient effort responding; no participant was identified as an insufficient effort responder based on established cut-off scores in the literature (Meade & Craig, 2012).

Linguistic features

Throughout the 5- to 6-day intensive data collection, the EAR generated 147,990 sound files among which 67,572 had audible sound (e.g., conversations, music, environmental sound) as opposed to silence. Of those files, 28,323 included meaningful language that was produced by the participants.

Trained research assistants transcribed participants’ speech verbatim; each file was transcribed by two different assistants to assure accuracy in transcription. Transcribers coded whether another social partner was speaking (other than the participant), but we did not transcribe speech by any other person.

We used the Linguistic Inquiry and Word Count 2015 (LIWC) to code 93 linguistic features in these transcripts. LIWC is a text analysis software that has been widely used in social science to analyze word frequency and linguistic features, including content and style words (Mehl & Gill, 2010; Pennebaker et al., 2001). LIWC applies a comprehensive dictionary of words classified into distinct categories by comparing the text against its internal word lists. It provides a calculated score of the percentage for each linguistic category (out of the total number of words in the transcripts; Pennebaker et al., 2003). By comparing the input documents with its embedded dictionary, LIWC identifies style words (e.g., pronouns, articles) as well as content words including psychological processes (e.g., positive and negative emotion categories), relativity words (e.g., time), and traditional content dimensions (e.g., achievement; Pennebaker et al., 2001). We included 81 nonpunctuation linguistic features out of the total 93 features in our analysis, because we transcribed participants’ speech from sound files and the punctuation might be arbitrary. Supplementary Table 1 lists the linguistic features used in the current study. The development and psychometric properties of the linguistic features are illustrated in the LIWC documentation (https://mcrc.journalism.wisc.edu/files/2018/04/Manual_LIWC.pdf).

Analytic Strategy

We first examined the bivariate associations between linguistic features and narcissism. Then, we examined associations between narcissism and the proportion of time in which (a) the participant spoke or (b) a conversation partner spoke as a function of the total number of speech sound files. Next, a random forest classifier was used to evaluate associations between narcissism and linguistic features. Random forest is a supervised machine learning algorithm that is widely used to classify observations into different categories (e.g., high vs low narcissism) based on a wide range of inputs (e.g., linguistic features). Although random forest can be used for continuous and categorical outcomes, the algorithm achieves optimal performance with categorical outcomes (Chauhan & Yafi, 2022). Thus, we dichotomized narcissism into a categorical variable based on the mean value; 109 (39%) participants were classified as more narcissistic; and 172 (61%) as low narcissistic.

To minimize bias created by feature selection, the results of random forest were calculated by assembling 2,000 decision trees together. Decision trees were set to have a maximum of five layers (which represent the depth of decision trees) to avoid overfitting (Sagi & Rokach, 2018). Each decision tree selected features that provided the most information gain (i.e., reduction in entropy). Entropy was calculated using the formula below, in which pi denotes the probability for an observation to fall into a certain category i.

Entropy= −∑ci=1pilog2pi

In addition to classifying observations, random forest is able to evaluate the associations between the input and output variables by calculating the feature importance that represents each feature’s contribution in distinguishing high and low narcissism. Specifically, the current study used the permutation feature importance. This method, using importance, measured the predictive value of a certain feature by calculating the decrease in model performance when this feature was removed (Altmann et al., 2010). The permutation feature importance suits the current study as it is not biased toward features that are more unique (i.e., high-cardinality features). In the equation, s represents the accuracy of the random forest model; K is the number of repetitions; and j is the column (i.e., feature) that is being shuffled.

Importancej=s−1K∑Kk=1sk,j

We used a k-fold cross-validation technique to evaluate the performance of the random forest model (Wong & Yeh, 2020). Compared to the conventional approach, which utilizes all the observations in the testing set all at once, the cross-validation method guarantees that each data point has an opportunity to be validated against all other data points. In this case, data were split into five subsets (i.e., k = 5) where models were estimated using k − 1 subsets for each fold, with the kth subset serving as the validation sample.

Results

Preliminary Analysis

We first examined the sample characteristics (Table 1) and the descriptive statistics for each linguistic feature generated by LIWC (Supplementary Table 1). Participants were most likely to use common verbs (e.g., eat, come), informal language (e.g., ok, fillers), personal pronouns, as well as words related to present focus (e.g., today, now) and social processes (e.g., mate, talk) regardless of narcissism level.

Next, the bivariate correlations between narcissism and the linguistic features were calculated (Supplementary Figure 1). Narcissism was significantly correlated with personal pronouns (r = −0.12, p = .04), first-person plural pronouns (r = 0.19, p = .001), second-person pronouns (r = −0.14, p = .02), third-person plural pronouns (r = −0.13, p = .03), auxiliary verbs (r = −0.14, p = .02), as well as words expressing causation (e.g., because; r = −0.12, p = .049), achievement (r = 0.15, p = .01), and assent (e.g., ok; r = 0.13, p = .03).

We also compared the frequency of participant speaking and other party speaking (which might be taken as a proxy for listening) between more narcissistic and less narcissistic people using independent t tests. Narcissistic people did not speak more (t(279) = −0.43, p = .67) or listen less (t(279) = −1.15, p = .25) compared to less narcissistic people.

Results From the Random Forest Classifier

Results showed that the random forest classifier model achieved an accuracy of 0.64, indicating an acceptable level of classifying performance. The random forest classifier calculates the feature importance of each linguistic feature. Table 2 summarizes the feature importance of 20 linguistic features that are the most powerful in differentiating high narcissism and low narcissism. The top five features that obtained the highest feature importance were first-person plural pronouns (e.g., we), words that are related to achievement (e.g., win), work (e.g., office), sex (e.g., erotic), and desired state (e.g., want; this category was named “discrepancy” in LIWC documentation; representing discrepancy between wishes and reality). Figure 1 depicts the differences in using these five categories of words between individuals who scored higher and lower on narcissism.

Table 2.

Feature Importance of the Top 20 Linguistic Features that Distinguished Narcissism

Feature importance
First-person plural pronouns 0.041
Achievement 0.030
Work 0.024
Sexual 0.023
Discrepancy 0.021
Second-person pronouns 0.019
Leisure 0.017
Anger 0.017
Hear 0.017
Drives 0.017
Netspeak 0.016
Impersonal pronouns 0.016
Affiliation 0.015
Past focus 0.015
Words >6 letters 0.015
Authentic 0.015
Negative emotions 0.014
Personal pronouns 0.014
Third-person plural pronouns 0.014
Informal language 0.014
Open in a new tab

Notes: The feature importance was calculated by the decrease in model performance after a feature is randomly shuffled. Higher feature importance represents a stronger association between the feature and the outcomes (i.e., narcissism).

Figure 1.

Figure 1.

Open in a new tab

Frequencies of linguistic features among more narcissistic and less narcissistic individuals. The standardized scores of frequencies were calculated to provide comparisons across different linguistic features. The features are displayed in order of the magnitude of the differences in standardized frequencies between people with higher and lower narcissism scores.

Sensitivity Analysis

We conducted a sensitivity analysis using narcissism as a continuous variable. Due to the change from a categorical outcome to a continuous outcome, the machine learning model used in the sensitivity analysis was the random forest regression model rather than the random forest classifier model. The random forest regression model resulted in similar performance and results as the random forest classifier model (accuracy = 0.71). The top five linguistic features selected by the random forest regression model were first-person plural pronouns, work, swear words, and language output: number of words per sentence, total word count. Although random forest regression and classifier are based on different types of decision trees and may yield different findings, the linguistic features detected by the regression and classifier model overlapped and highlighted the importance of first-person plural pronouns and work-related words as linguistic characteristics of narcissism.

We also examined the Spearman’s correlations between narcissism and linguistic features to reduce the possible bias introduced by the skewed distributions of linguistic features (de Winter et al., 2016). The results of Spearman’s rho were similar to the Pearson correlation reported previously. Narcissism was correlated with the use of total function words (rs = −0.12, p = .04), personal pronouns (rs = −0.15, p = .01), first-person plural pronouns (rs = 0.21, p = .001), second-person pronouns (rs = −0.15, p = .01), third-person plural pronouns (rs = −0.13, p = .03), auxiliary verbs (rs = −0.09, p = .00), words related to sex (rs = 0.12, p = .04), affiliation (rs = 0.14, p = .02), and assent (rs = 0.12, p = .04).

Finally, to address the potential issue of the excessive zeros in the outcome variable (i.e., narcissism), we also estimated a series of zero-inflated Poisson models. Results showed the same pattern as our random forest model, such that first-person plural pronouns (B = 0.43, p = .001) and achievement (B = 0.43, p = .005) were significantly associated with higher narcissism, and work was associated with lower probability of excess zeros (B = −10.07, p < .001; Supplementary Table 3).

Discussion

Narcissism may have a profound impact on individuals’ social lives as it influences the nature of their interactions with social partners (Grapsas et al., 2020; Morf & Rhodewalt, 2001). Although language is an important part of interpersonal communication, we know little about how narcissism is manifested in daily life through language use (Holtzman et al., 2019). This study contributes to the literature by examining a wide range of linguistic features collected in participants’ natural conversations and demonstrating the way that narcissistic older adults speak and the topics they choose in daily conversations. Compared to less narcissistic older adults, more narcissistic older adults tend to speak in the first-person plural and to use words associated with self-achievement (rather than affiliation).

The discussion focuses on understanding how the linguistic features reflect the characteristics of narcissism from the perspectives of (a) linguistic style (i.e., we) and (b) linguistic content (i.e., words related to achievement, work, sex, discrepancy).

Linguistic Style: A Sense of Personal Superiority

Linguistic style refers to how people organize their conversation topics (rather than the topics themselves), and it is reflected by the use of pronouns, auxiliary verbs, and other words that do not convey concrete meanings (Tausczik & Pennebaker, 2010). Findings showed that narcissistic people used more first-person plural pronouns in their daily talk. Prior studies have shown that more narcissistic young adults use the first-person singular pronouns more than less narcissistic young adults (Raskin & Shaw, 1988). Although the use of “we” may indicate affiliation and interdependence in some cases (e.g., we have experienced a lot together), it may also be used to express exclusiveness and superiority (Tausczik & Pennebaker, 2010). In fact, speakers can use “we” to indicate commands, as in the Royal We (Pennebaker & Lay, 2002; Schimpff, 2019). For instance, the sentence “we need to get the work done by tomorrow” may indicate the speaker assigns a task to the listeners instead of working on the task collectively. By using the Royal We, the speaker automatically takes the superior position in a conversation (Kacewicz et al., 2014; Schimpff, 2019). It is likely that narcissistic people are more dominant in conversations compared to less narcissistic people, and use “we” in an authoritative and superior way.

Linguistic Content: A Constant Pursuit of Status

Content words indicate the topics of conversations (Tausczik & Pennebaker, 2010), and the content individuals choose to communicate with social partners may provide information about their personalities. Narcissistic people are preoccupied with a sense of self-importance with an expectation of others’ adulation (Krizan & Herlache, 2018). Reflecting on past achievements may be an effective approach to attract social partners’ attention and garner respect. Indeed, narcissistic college students were more likely to be rated by their partners as bragging about their academic and other accomplishments (e.g., athletics; Buss & Chiodo, 1991). Findings on achievement-related words in this study suggest more narcissistic older adults continue to pursue higher status, even in late life when continuous achievement may be less important.

More narcissistic older adults also talked more about work. As noted earlier, narcissistic people crave social status (Grapsas et al., 2020), and they perform better in the workplace when they perceive more opportunities for glory (Roberts et al., 2018). Additionally, it is notable that the current sample consisted of older adults who did not work full time for pay. Compared to younger people, older adults typically prioritize their emotional needs rather than work-related goals (Carstensen, 2021). In this regard, the linguistic feature reflects how narcissistic older adults may select conversation topics associated with accomplishments in outside domains, at the life stage that their counterparts are more likely to talk about their families and share emotions with social partners (Carstensen, 2021).

Narcissistic older adults used sexual words more frequently than less narcissistic older adults, though references to sexual content were rare in this study overall. Previous studies have found narcissistic young adults require less relationship commitment for engaging in sexual intercourse (Foster et al., 2006). Similarly, more narcissistic young adults used more sexual words in their daily life (Holtzman et al., 2010, 2019) and Twitter profiles (Sumner et al., 2012). Among older adults, topics related to sex and intimacy may drive other people’s attention even more effectively because being sexually active in late life has been linked to elevated self-worth and a feeling of being respected (von Humboldt et al., 2021). Narcissistic older adults may talk about sex-related topics to indicate their continued attractiveness and capabilities, though the relatively low frequency of these words warrants additional research to draw conclusions.

Finally, this study showed that compared to less narcissistic older adults, more narcissistic older adults mentioned their desired states more frequently by using words that indicated the discrepancy between hopes and reality. Narcissistic people may believe they are entitled to special treatment (Piff, 2014), yet the privileges that narcissistic people wish for may be hard to achieve in reality. As such, narcissistic people may complain about their current lives, reflecting the contrast between narcissistic people’s self-concept and their images in other people’s eyes (Back et al., 2010).

Together, the findings revealed that more narcissistic people tend to be in dominant positions during daily conversations, as well as to engage in conversation around topics (e.g., past achievements) that might garner admiration or convey their own self-importance.

Limitations and Future Directions

Several limitations are evident in the current study. Although we detected differences in language use as a function of narcissism, the most commonly used words did not differ as a function of narcissism. As such, the linguistic differences may be subtle and difficult to detect for social partners. Further, we do not know who the participant was speaking to. Words may be understood differently depending on the conversation partner. Also, LIWC applies the word frequency approach, and it does not take the language context into consideration. Thus, LIWC does not detect the meaning of sentences that include negations (Schwartz et al., 2013; Tausczik & Pennebaker, 2010). For example, in the sentence “I don’t like you,” LIWC recognizes the word “like” and classifies it as positive emotions. This may lead to inaccurate scores for some linguistic features (e.g., positive emotions). As such, more fine-tuned sentiment analysis tools (e.g., Latent Dirichlet Allocation, Bidirectional Encoder Representations from Transformers) may generate a more nuanced understanding of each text corpora.

Furthermore, the accuracy of the reported random forest classifier was 0.64. Although an accuracy of 0.70 is generally considered to represent a well-validated model, the accuracy and standard of random forest model evaluation vary across different fields (Izquierdo-Verdiguier & Zurita-Milla, 2020; Pavey et al., 2017). Considering we conducted supervised machine learning and relied purely on information extracted from daily conversations to classify individuals’ narcissism, an accuracy of 0.64 is acceptable. However, the results need to be interpreted cautiously because the reliability of feature importance (i.e., the strength of the association between narcissism and each linguistic feature) is also related to model performance (Gregorutti et al., 2017). Future studies may seek ways to improve model performance (e.g., using larger data sets and increasing the number of trees in the model) and examine daily language use.

Additionally, the current study used NPI-16 to measure narcissism. The measure has been validated in previous studies (Ames et al., 2006). However, NPI-16 largely measures grandiose narcissism, yet other facets of narcissism (e.g., vulnerable components; Krizan & Herlache, 2018) need to be considered as well to cover a full spectrum of narcissism.

In conclusion, utilizing sound files collected throughout the day, the current study revealed that core features of narcissism (e.g., superiority, entitlement) may be manifested in older adults’ conversational style via style of words and conversation topics. This speaking style may weaken more narcissistic older adults’ social ties. Future research is necessary to establish whether social partners detect these subtleties in language, as well as the behavioral features of conversation that may weaken these social ties.

Supplementary Material

gbad061_suppl_Supplementary_Material
Click here for additional data file. (329.3KB, pdf)

Contributor Information

Shiyang Zhang, Department of Human Development and Family Sciences, The University of Texas at Austin, Austin, Texas, USA.

Karen L Fingerman, Department of Human Development and Family Sciences, The University of Texas at Austin, Austin, Texas, USA.

Kira S Birditt, Institute for Social Research, University of Michigan, Ann Arbor, Michigan, USA.

Funding

This research was supported by grants from the National Institute on Aging: R01AG046460 (Fingerman, PI) and P30AG066614 (awarded to the Center on Aging and Population Sciences at The University of Texas at Austin), as well as a grant from the Eunice Kennedy Shriver National Institute of Child Health and Human Development P2CHD042849 (awarded to the Population Research Center at The University of Texas at Austin). The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

Conflict of Interest

None declared.

References

  1. Alexander III, L., Mulfinger, E., & Oswald, F. L. (2020). Using big data and machine learning in personality measurement: Opportunities and challenges. European Journal of Personality, 34(5), 632–648. doi: 10.1002/per.2305 [DOI] [Google Scholar]
  2. Altmann, A., Toloşi, L., Sander, O., & Lengauer, T. (2010). Permutation importance: A corrected feature importance measure. Bioinformatics, 26(10), 1340–1347. doi: 10.1093/bioinformatics/btq134 [DOI] [PubMed] [Google Scholar]
  3. Ames, D. R., Rose, P., & Anderson, C. P. (2006). The NPI-16 as a short measure of narcissism. Journal of Research in Personality, 40(4), 440–450. doi: 10.1016/j.jrp.2005.03.002 [DOI] [Google Scholar]
  4. Augustine, A. A., Mehl, M. R., & Larsen, R. J. (2011). A positivity bias in written and spoken English and its moderation by personality and gender. Social Psychological and Personality Science, 2(5), 508–515. doi: 10.1177/1948550611399154 [DOI] [PMC free article] [PubMed] [Google Scholar]
  5. Back, M. D., Schmukle, S. C., & Egloff, B. (2010). Why are narcissists so charming at first sight? Decoding the narcissism–popularity link at zero acquaintance. Journal of Personality and Social Psychology, 98(1), 132–145. doi: 10.1037/a0016338 [DOI] [PubMed] [Google Scholar]
  6. Bleidorn, W., & Hopwood, C. J. (2019). Using machine learning to advance personality assessment and theory. Personality and Social Psychology Review, 23(2), 190–203. doi: 10.1177/1088868318772990 [DOI] [PubMed] [Google Scholar]
  7. Breiman, L. (2001). Random forests. Machine Learning, 45(1), 5–32. doi: 10.1023/A:1010933404324 [DOI] [Google Scholar]
  8. Buss, D. M., & Chiodo, L. M. (1991). Narcissistic acts in everyday life. Journal of Personality, 59(2), 179–215. doi: 10.1111/j.1467-6494.1991.tb00773.x [DOI] [PubMed] [Google Scholar]
  9. Campbell, W. K., & Foster, J. D. (2007). The narcissistic self: Background, an extended agency model, and ongoing controversies. In Sedikides C. & Spencer S. J. (Eds.), The Self (pp. 115–138). Psychology Press. [Google Scholar]
  10. Carey, A. L., Brucks, M. S., Küfner, A. C. P., Holtzman, N. S., große Deters, F., Back, M. D., Donnellan, M. B., Pennebaker, J. W., & Mehl, M. R. (2015). Narcissism and the use of personal pronouns revisited. Journal of Personality and Social Psychology, 109(3), e1–e15. doi: 10.1037/pspp0000029 [DOI] [PubMed] [Google Scholar]
  11. Carstensen, L. L. (2021). Socioemotional selectivity theory: The role of perceived endings in human motivation. Gerontologist, 61(8), 1188–1196. doi: 10.1093/geront/gnab116 [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. Charles, S. T., & Carstensen, L. L. (2010). Social and emotional aging. Annual Review of Psychology, 61, 383–409. doi: 10.1146/annurev.psych.093008.100448 [DOI] [PMC free article] [PubMed] [Google Scholar]
  13. Chauhan, R., & Yafi, E. (2022). Applicability of classifier to discovery knowledge for future prediction modelling. Journal of Ambient Intelligence and Humanized Computing, 1–16. doi: 10.1007/s12652-022-03694-3 [DOI] [Google Scholar]
  14. Chopik, W. J., & Grimm, K. J. (2019). Longitudinal changes and historic differences in narcissism from adolescence to older adulthood. Psychology and Aging, 34(8), 1109–1123. doi: 10.1037/pag0000379 [DOI] [PubMed] [Google Scholar]
  15. Cutler, A. D., Carden, S. W., Dorough, H. L., & Holtzman, N. S. (2021). Inferring grandiose narcissism from text: LIWC versus machine learning. Journal of Language and Social Psychology, 40(2), 260–276. doi: 10.1177/0261927X20936309 [DOI] [Google Scholar]
  16. de Winter, J. C. F., Gosling, S. D., & Potter, J. (2016). Comparing the Pearson and Spearman correlation coefficients across distributions and sample sizes: A tutorial using simulations and empirical data. Psychological Methods, 21(3), 273–290. doi: 10.1037/met0000079 [DOI] [PubMed] [Google Scholar]
  17. Foster, J. D., Shrira, I., & Campbell, W. K. (2006). Theoretical models of narcissism, sexuality, and relationship commitment. Journal of Social and Personál Relationships, 23(3), 367–386. doi: 10.1177/0265407506064204 [DOI] [Google Scholar]
  18. Grapsas, S., Brummelman, E., Back, M. D., & Denissen, J. J. (2020). The “why” and “how” of narcissism: A process model of narcissistic status pursuit. Perspectives on Psychological Science, 15(1), 150–172. doi: 10.1177/1745691619873350 [DOI] [PMC free article] [PubMed] [Google Scholar]
  19. Gregorutti, B., Michel, B., & Saint-Pierre, P. (2017). Correlation and variable importance in random forests. Statistics and Computing, 27(3), 659–678. doi: 10.1007/s11222-016-9646-1 [DOI] [Google Scholar]
  20. Hill, P. L., & Roberts, B. W. (2012). Narcissism, well-being, and observer-rated personality across the lifespan. Social Psychological and Personality Science, 3(2), 216–223. doi: 10.1177/1948550611415867 [DOI] [Google Scholar]
  21. Holtgraves, T. (2011). Text messaging, personality, and the social context. Journal of Research in Personality, 45(1), 92–99. doi: 10.1016/j.jrp.2010.11.015 [DOI] [Google Scholar]
  22. Holtzman, N. S., Tackman, A. M., Carey, A. L., Brucks, M. S., Küfner, A. C., Deters, F. G., Back, M. D., Donnellan, M. B., Pennebaker, J. W., Sherman, R. A., & Mehl, M. R. (2019). Linguistic markers of grandiose narcissism: A LIWC analysis of 15 samples. Journal of Language and Social Psychology, 38(5–6), 773–786. doi: 10.1177/0261927X19871084 [DOI] [Google Scholar]
  23. Holtzman, N. S., Vazire, S., & Mehl, M. R. (2010). Sounds like a narcissist: Behavioral manifestations of narcissism in everyday life. Journal of Research in Personality, 44(4), 478–484. doi: 10.1016/j.jrp.2010.06.001 [DOI] [PMC free article] [PubMed] [Google Scholar]
  24. Ireland, M. E., & Mehl, M. R. (2014). Natural language use as a marker of personality. In Holtgraves T. M. (Ed.), The Oxford handbook of language and social psychology (pp. 201–218). Oxford University Press. [Google Scholar]
  25. Izquierdo-Verdiguier, E., & Zurita-Milla, R. (2020. ). An evaluation of guided regularized random forest for classification and regression tasks in remote sensing. International Journal of Applied Earth Observation and Geoinformation, 88, 102051. doi: 10.1016/j.jag.2020.102051 [DOI] [Google Scholar]
  26. Kacewicz, E., Pennebaker, J. W., Davis, M., Jeon, M., & Graesser, A. C. (2014). Pronoun use reflects standings in social hierarchies. Journal of Language and Social Psychology, 33(2), 125–143. doi: 10.1177/0261927X13502654 [DOI] [Google Scholar]
  27. Kjærvik, S. L., & Bushman, B. J. (2021). The link between narcissism and aggression: A meta-analytic review. Psychological Bulletin, 147(5), 477–503. doi: 10.1037/bul0000323 [DOI] [PubMed] [Google Scholar]
  28. Krizan, Z., & Herlache, A. D. (2018). The narcissism spectrum model: A synthetic view of narcissistic personality. Personality and Social Psychology Review, 22(1), 3–31. doi: 10.1177/1088868316685018 [DOI] [PubMed] [Google Scholar]
  29. Marengo, D., Montag, C., Sindermann, C., Elhai, J. D., & Settanni, M. (2021). Examining the links between active Facebook use, received likes, self-esteem and happiness: A study using objective social media data. Telematics and Informatics, 58, 101523. doi: 10.1016/j.tele.2020.101523 [DOI] [Google Scholar]
  30. Meade, A. W., & Craig, S. B. (2012). Identifying careless responses in survey data. Psychological Methods, 17(3), 437–455. doi: 10.1037/a0028085 [DOI] [PubMed] [Google Scholar]
  31. Mehl, M. R. (2017). The electronically activated recorder (EAR) a method for the naturalistic observation of daily social behavior. Current Directions in Psychological Science, 26(2), 184–190. doi: 10.1177/0963721416680611 [DOI] [PMC free article] [PubMed] [Google Scholar]
  32. Mehl, M. R., & Gill, A. J. (2010). Automatic text analysis. In Gosling S. D. & Johnson J. A. (Eds.), Advanced methods for conducting online behavioral research (pp. 109–127). American Psychological Association. doi: 10.1037/12076-008 [DOI] [Google Scholar]
  33. Mehta, Y., Majumder, N., Gelbukh, A., & Cambria, E. (2020). Recent trends in deep learning based personality detection. Artificial Intelligence Review, 53(4), 2313–2339. doi: 10.1007/s10462-019-09770-z [DOI] [Google Scholar]
  34. Milek, A., Butler, E. A., Tackman, A. M., Kaplan, D. M., Raison, C. L., Sbarra, D. A., Vazire, S., & Mehl, M. R. (2018). “Eavesdropping on happiness” revisited: A pooled, multisample replication of the association between life satisfaction and observed daily conversation quantity and quality. Psychological Science, 29(9), 1451–1462. doi: 10.1177/0956797618774252 [DOI] [PMC free article] [PubMed] [Google Scholar]
  35. Morf, C. C., & Rhodewalt, F. (2001). Unraveling the paradoxes of narcissism: A dynamic self-regulatory processing model. Psychological Inquiry, 12(4), 177–196. doi: 10.1207/s15327965pli1204_1 [DOI] [Google Scholar]
  36. Pavey, T. G., Gilson, N. D., Gomersall, S. R., Clark, B., & Trost, S. G. (2017). Field evaluation of a random forest activity classifier for wrist-worn accelerometer data. Journal of Science and Medicine in Sport, 20(1), 75–80. doi: 10.1016/j.jsams.2016.06.003 [DOI] [PubMed] [Google Scholar]
  37. Pennebaker, J. W. (2011). The secret life of pronouns: What our words say about us. Bloomsbury Press. [Google Scholar]
  38. Pennebaker, J. W., Francis, M. E., & Booth, R. J. (2001). Linguistic inquiry and word count: LIWC 2001, Vol. 71. Lawrence Erlbaum Associates. https://www.researchgate.net/profile/Ryan-Boyd-8/publication/337731895_Linguistic_Inquiry_and_Word_Count_LIWC2015/links/6088fef0907dcf667bcadc17/Linguistic-Inquiry-and-Word-Count-LIWC2015.pdf [Google Scholar]
  39. Pennebaker, J. W., & Lay, T. C. (2002). Language use and personality during crises: Analyses of Mayor Rudolph Giuliani’s press conferences. Journal of Research in Personality, 36(3), 271–282. doi: 10.1006/jrpe.2002.2349 [DOI] [Google Scholar]
  40. Pennebaker, J. W., Mehl, M. R., & Niederhoffer, K. G. (2003). Psychological aspects of natural language use: Our words, our selves. Annual Review of Psychology, 54(1), 547–577. doi: 10.1146/annurev.psych.54.101601.145041 [DOI] [PubMed] [Google Scholar]
  41. Pew Research Center. (2021). Social media fact sheet. https://www.pewresearch.org/internet/factsheet/social-media/
  42. Piff, P. K. (2014). Wealth and the inflated self: Class, entitlement, and narcissism. Personality and Social Psychology Bulletin, 40(1), 34–43. doi: 10.1177/0146167213501699 [DOI] [PubMed] [Google Scholar]
  43. Raskin, R., & Shaw, R. (1988). Narcissism and the use of personal pronouns. Journal of Personality, 56(2), 393–404. doi: 10.1111/j.1467-6494.1988.tb00892.x [DOI] [PubMed] [Google Scholar]
  44. Roberts, R., Woodman, T., & Sedikides, C. (2018). Pass me the ball: Narcissism in performance settings. International Review of Sport and Exercise Psychology, 11(1), 190–213. doi: 10.1080/1750984X.2017.1290815 [DOI] [Google Scholar]
  45. Rohit, G. V., Bharadwaj, K. R., Hemanth, R., Pruthvi, B., & Kumar, M. M. (2020, August). Machine intelligence based personality prediction using social profile data. In 2020 Third International Conference on Smart Systems and Inventive Technology (ICSSIT) (pp. 1003–1008). IEEE. doi: 10.1109/ICSSIT48917.2020.9214175 [DOI] [Google Scholar]
  46. Sagi, O., & Rokach, L. (2018). Ensemble learning: A survey. Wiley Interdisciplinary Reviews: Data Mining and Knowledge Discovery, 8(4), e1249. doi: 10.1002/widm.1249 [DOI] [Google Scholar]
  47. Schimpff, A. (2019). “We” but not “me”: A sociolinguistic study of the speaker-exclusive first person plural pronoun “we”. Lifespans & Styles, 5(1), 1–15. http://journals.ed.ac.uk/lifespansstyles [Google Scholar]
  48. Schwartz, H. A., Eichstaedt, J. C., Kern, M. L., Dziurzynski, L., Ramones, S. M., Agrawal, M., Shah, A., Kosinski, M., Stillwell, D., Seligman, M. E. P., & Ungar, L. H. (2013). Personality, gender, and age in the language of social media: The open-vocabulary approach. PLoS One, 8(9), e73791. doi: 10.1371/journal.pone.0073791 [DOI] [PMC free article] [PubMed] [Google Scholar]
  49. Seidman, G., Shrout, P. E., & Zeigler-Hill, V. (2020). Untangling the associations that narcissistic admiration and narcissistic rivalry have with agency, communion, and romantic commitment. Journal of Research in Personality, 89, 104022. doi: 10.1016/j.jrp.2020.104022 [DOI] [Google Scholar]
  50. Stachl, C., Pargent, F., Hilbert, S., Harari, G. M., Schoedel, R., Vaid, S., Gosling, S. D., & Bühner, M. (2020). Personality research and assessment in the era of machine learning. European Journal of Personality, 34(5), 613–631. doi: 10.1002/per.2257 [DOI] [Google Scholar]
  51. Sumner, C., Byers, A., Boochever, R., & Park, G. J. (2012, December). Predicting dark triad personality traits from Twitter usage and a linguistic analysis of tweets. In 2012 11th International Conference on Machine Learning and Applications (Vol. 2, pp. 386–393). IEEE. [Google Scholar]
  52. Tackman, A. M., Baranski, E. N., Danvers, A. F., Sbarra, D. A., Raison, C. L., Moseley, S. A., Polsinelli, A. J., & Mehl, M. R. (2020). ‘Personality in its Natural Habitat’ revisited: A pooled, multi-sample examination of the relationships between the Big Five personality traits and daily behaviour and language use. European Journal of Personality, 34(5), 753–776. doi: 10.1002/per.2283 [DOI] [Google Scholar]
  53. Tausczik, Y. R., & Pennebaker, J. W. (2010). The psychological meaning of words: LIWC and computerized text analysis methods. Journal of Language and Social Psychology, 29(1), 24–54. doi: 10.1177/0261927X09351676 [DOI] [Google Scholar]
  54. von Humboldt, S., Ribeiro-Gonçalves, J. A., Costa, A., Low, G., & Leal, I. (2021). Sexual expression in old age: How older adults from different cultures express sexually? Sexuality Research and Social Policy, 18(2), 246–260. doi: 10.1007/s13178-020-00453-x [DOI] [Google Scholar]
  55. Weiss, B., & Miller, J.D. (2018). Distinguishing between grandiose narcissism, vulnerable narcissism, and narcissistic personality disorder. In Hermann A., Brunell A., & Foster J. (Eds.), Handbook of trait narcissism (pp. 3–13). Springer. doi: 10.1007/978-3-319-92171-6_1 [DOI] [Google Scholar]
  56. Wong, T. T., & Yeh, P. Y. (2020). Reliable accuracy estimates from k-fold cross validation. IEEE Transactions on Knowledge and Data Engineering, 32(8), 1586–1594. doi: 10.1109/tkde.2019.2912815 [DOI] [Google Scholar]
  57. Yarkoni, T. (2010). Personality in 100,000 words: A large-scale analysis of personality and word use among bloggers. Journal of Research in Personality, 44(3), 363–373. doi: 10.1016/j.jrp.2010.04.001 [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

gbad061_suppl_Supplementary_Material
Click here for additional data file. (329.3KB, pdf)

Articles from The Journals of Gerontology Series B: Psychological Sciences and Social Sciences are provided here courtesy of Oxford University Press

RESOURCES