The Big Two

Agency and Communion

Definitions

The Big Two are the most crucial psychological traits that characterize people.

• Agency
• refers to the will of a person of achieving goals
• “Refers to a person’s striving to be independent, to control one’s self” (Abele,Uchronski et al.)
• Communion
• refers to the will of a person to form meaningful relationships with others
• “Refers to a person’s striving to be part of a community, to establish close relationships with others, and to subordinate individual needs to the common good” (Abele,Uchronski et al.)

​

Why are we interested in them?

• they are the most frequent themes in �
• autobiographical memories (McAdams, Hoffman, Day, & Mansfield, 1996) �
• descriptions or evaluations of self and others (Abele & Bruckmüller, 2011; Wojciszke, 1994)�
• perception of groups s (Cuddy, Fiske, & Glick, 2008; Fiske, Cuddy, Glick, & Xu, 2002)

Nonetheless little effort has been devoted to investigating their linguistic manifestations

Previous Studies - self reports on psychological traits

Most of the previous studies have been done on self-report trait rating scales by the usage of adjectives such as

• active, dynamic, efficient, assertive, self-confident for agency
• helpful, understanding, reliable, likable, empathic, friendly for communion

Drawbacks of self-reported rating scales:

• people tend to answer in a manner that will be viewed favorably by others
• adjective are entirely chosen by the researcher

Previous Studies - Natural Language

Only a few studies have captured agency and communion in natural language, through the analysis of, for example,

• obituaries
• job advertisements
• letters of recommendation for academic positions

​

Previous Studies - Natural Language

Approach used so far

• usage of self-developed communal and agentic word lists
• competent judges to analyse the texts unders investigation

Drawbacks of such approaches

• the words list consisted mainly in adjectives, while recent studies showed that agency is better captured in verbs
• difficult to compare studies since each of them uses different words lists
• researchers conducted coding of the language data in a manual fashion, which is time consuming

A need for reliability and efficiency

It is imperative to develop a more objective, automated method for assessing and estimating levels of agency and communion in natural language use.

4 studies have been conducted to move in such direction.

Word Counting Approach in Language Analysis

The recent accessibility of large quantities of natural language data has prompted the development of computerized applications for text analysis.

Such methods of linguistic inquiry rely on the assumption that language is inevitably linked to human psychology and that psychological processes find their reflection in the words that people use.

The dictionary-based (or “word-counting”) approach identifies linguistic manifestations of psychological constructs, by coding and quantifying words in natural language samples that are known (or strongly believed) to be diagnostic of particular psychological constructs.

​

​

LIWC2015

• LIWC is a text analysis tool that has been developed from the end of the 90s�
• LIWC stands for Language Inquiry and Word Counts�
• it calculates the degree to which various categories of words are used in a text�
• it can process texts ranging from emails to speeches, poems and transcribed natural language�
• the last version of the software has been released in 2015 by the name LIWC2015

LIWC 2015 - How it works

LIWC2015 contains

• word categories: Groups of dictionary words that tap a particular domain (e.g., negative emotion words)
• a dictionary: set of around 6400 words,stems, and emoticons. �a word contained in the dictionary is referred to as a dictionary word

a word in the input file is referred to as a target word.

​

LIWC - How it works

Given a single file or a group of files, for each file LIWC2015 does the following:

1. reads one target word at a time.
2. As each target word is processed, the dictionary file is searched, looking for a dictionary match with the current target word.
3. If the target word is matched with a dictionary word, the appropriate word category scale (or scales) for that word is incremented.

​

First Study

• aim: develop agency and communion dictionaries (set of words) that capture both constructs in language

The process can be summarized in 3 steps

1. a committee of 6-8 trained judges generated lists of words that were evaluated to be psychologically relevant to agency and communion
2. the committee evaluated the Agency and Communion dictionaries’ psychometrics properties
3. 3. the committee re-evaluated the dictionaries by deciding which current words should be retained or removed, and which new words should be included in each dictionary

First Study

in short�

• the final agency dictionary consists of 192 words�
• the final communion dictionary consists of 184 words�

across 6 large, broadly representative corpora of texts, both dictionaries were found to possess strong psychometric properties on a par with or exceeding other mainstream psychological text analysis dictionaries.

​

Second Study

The second study tested the convergent and discriminant validity of the Agency and Communion dictionaries developed in Study 1.

Convergent Validity

• Convergent validity takes two measures that are supposed to be measuring the same construct and shows that they are related

Discriminant Validity

• Discriminant validity shows that two measures that are not supposed to be related are in fact unrelated

Second Study

The second study tested the convergent and discriminant validity of the Agency and Communion dictionaries developed in Study 1.

It did so by examining with selected LIWC2015 dictionary categories both

• the direct overlap �
• the semantic similarity

​

Second Study - Terminology

Direct Overlap�

• the overlap between each pair of two dictionaries is defined as the shared number of words that appear in both dictionaries, divided by the total number of words in both dictionaries (AKA Jaccard similarity)�
• the criterion for direct overlaps was binary, that is, words were either part of a dictionary or not

Second Study - Terminology

Semantic Similarity �

• Computed through Latent Semantic Analysis, namely computing a semantic multi-dimensional space in which vectors represent words and wordlists in relation to how they are used in natural language. �The cosine distance between two vectors estimates whether two dictionaries are similar or different �
• words are context dependent so it is needed to take into consideration the semantic of words �
• example: the LIWC ”Power” dictionary does not include words like ”bear”, but one can express power by saying ”strong as a bear”.

Second Study - Results and Discussion

Agency

• for agency, the highest overlap and semantic similarity scores occurred between Achievement and Reward dictionaries �
• Agency was not associated with Work and Power as expected�
• Power includes words related to both ends of the power spectrum (e.g. “rich” and “poor”)�
• Work included words associated with communication (e.g. “collaboration” and “team”)�
• Being agentic does not necessarily mean being powerful (e.g. quitting smoking)

Second Study - Results and Discussion

Communion

• for Communion, the highest overlap was with Affiliation, followed by Friends, Social, and Family

​

In sum, the analysis of direct overlap and semantic similarity indicated that the proposed dictionaries of Agency and Communion both showed satisfactory convergent and discriminant validity.

​

​

Third Study

The third study tested the convergent validity of the proposed dictionaries by comparing agency and communion related word use, against human ratings of agency and communion for a list of professions.

In order to do so, the dataset collected by Fiske and Dupree (2014) has been used.

Third Study - Fiske and Dupree (2014)

Fiske and Dupree asked 116 American adults to evaluate a list of 42 common American professions on the dimensions of

• competence (referred to as agency)
• warmth (referred to as communion)

Third Study - Method

1. compute the semantic similarity scores between each profession listed by Fiske and Dupre and the Agency and Communion dictionaries developed in the previous studies�
2. compare these scores with the corresponding ratings from participants in the original studies�

A higher correspondence indicates that the use of agency and communion words as captured by our dictionaries matches the meaning of agency and communion as measured with scales.

​

Third Study - Semantic Similarity

How was the semantic similarity computed?

• Latent Semantic Analysis has been used to create a semantic representation. �
• The semantic space was created by using a media news corpus containing around 80000 Reuters news messages published in 1996 and 1997�
• as in Study 2, the semantic similarity has been measured between the Agency and Communion dictionaries and individual words from the list of selected professions.

​

Third Study - Results

• The relationship appeared to be relatively strong�
• the existing LIWC dictionaries theoretically related to agency (i.e., Cause, Achievement, Power, Reward, Work, and Insight) or to communion (i.e., Family, Friends, Social, and Affiliation) are linked to self-ratings of agency and communion to a lesser extent than the proposed dictionaries of agency and communion

​

Fourth Study

Study 4 aimed to apply the proposed Agency and Communion dictionaries to detect differences in how female-dominated and male-dominated jobs are advertised.

in line with previous studies we should have a confirmation that

• male-dominated jobs are described mostly by agentic words�
• female-dominated jobs are described mostly by communal words

​

Fourth Study - the dataset

• Data set of 20,000 job advertisements published on the most popular U.S. online job boards, namely Monster.com. �
• As the data set contained various job advertisements, job titles have been filtered to select male and female dominated job professions

​

Fourth Study - Method

1. First, compute the total number of words for every job description. �
2. Compute the number of agentic and communal words for every ad based on the Agentic and Communion dictionaries. �
3. Give to each advertisement agency and communion scores, respectively as the percentage of total agentic and communal words in each.

​

Fourth Study - Evaluation

The evaluation on the nature of job advertisements has been conducted through a mixed design ANOVA (analysis of variance).

Such analysis considered

• jobs (male-dominated, female-dominated) as a between-subjects factor

and

• wording (agency score, communion score) as a within-subjects factor

​

Fourth Study - Results

the results showed that

• for male-dominated occupations, agentic words were more likely than communal words�
• for female-dominated jobs, communal words were more likely than agentic words

as expected.

​

Fourth Study - Results

• goal: �predict agency and communion starting by the men percentage for a job�
• how: �regression analysis using the Maximum Likelihood Robust estimator in Mplus 7. �
• results: �The standardized regression coefficients indicate that percentage of men in the profession was positively related to agency scores and negatively related to communion scores. �There is though no causal relationship between between the big twos and gender-role congruency in job advertisement

​

Limitations of Word Count

• context insensitive�
• ignores irony and sarcasm�
• no analysis on the linguistic function of a word (e.g. “He is a strong man”, “He was walking in a strong wind”) �
• the linguistic function problem could be solved by using an open access algorithm that allows for determining the role of a word in the sentence (e.g., Stanford Natural Language Processing (NLP): Manning et al., 2014 �
• Nevertheless, experience teaches that such methodological refinements usually do not change the main results (see Boyd, 2017).

​

Alternative Approach

• the Agency and Communion dictionaries presented in this reading were only based on one definition (i.e., Diehl et al., 2004) provided to the Dictionary Creation Committee, although many conceptualizations of Big Two have been developed. �
• In order to gain more insights on this matter, the researchers created alternative dictionaries of agency and communion that are based on semantic associations with the Big Two (see Supplemental Study 1 in the SOM) �
• The same studies previously presented have been conducted on these new dictionaries �
• the constructs of agency and communion were captured less efficiently wrt to the results of the first approach

​

Alternative Approach - Considerations

By comparing these two approaches we learn that

• the language used to describe the concepts of agency and communion are not particularly effective at capturing the constructs as they are manifest in verbal behavior �
• a new measure is required for capturing agency and communion in verbal behavior �
• a new approach to assessing each construct may help us to better understand how they operate in daily life

​

Future Works

• examine whether people use more agentic words than communal words when describing themselves, and more communal words than agentic words when describing others, as explained in “Dual Perspective Model of Agency and Communion” (Abele & Wojciszke, 2014) �
• capture links between the Big Five personality traits and the Big Twos with natural language (Abele & Wojciszke, 2014) �
• verify whether the presented results are replicable using broader ML algorithms as NLP�
• combine the word-counting approach with semantic analysis to further verify the past finding on the role of valence in the perception of the Big Twos (Suitner & Maass, 2008)

​