Gender differences in autobiographical memory: females latently express communality more than do males

ABSTRACT

Gender differences have been found in several aspects of autobiographical memory (i.e. personally experienced events). For example, previous studies have shown that females’ autobiographical memories contain more communal and emotional expressions than do males. However, an important question concerns whether these differences can be observed both in the manifest content (i.e. what is actually said) and in the latent content (i.e. the underlying meaning of what is said). In the present exploratory study, we extended the current knowledge concerning gender differences in autobiographical memory by investigating the manifestly expressed words, as well as the latently expressed words in autobiographical memory descriptions. We observed an overall gender difference in the latent content of the autobiographical memories. Furthermore, females latently described their memories in more communal terms than males did. No other gender differences were found. Our results indicate that females’ autobiographical memories are more communally oriented than male's.

KEYWORDS:

• Autobiographical memory
• gender differences
• LSA
• LIWC
• content analysis

Introduction

Several studies have established gender differences in autobiographical memory (AM) across both cultures and age cohorts. For example, females’ AMs are more communally oriented (e.g. American children, Buckner & Fivush, 1998; Polish middle-aged adults, Niedźwieńska, 2003) and contain more emotional expressions in comparison to males’ AMs (e.g. American, mainly, undergraduate students, Bauer, Stennes, & Haight, 2003; American, young and old adults, Rice & Pasupathi, 2010). Previous research has been oriented towards the manifestly expressed words (i.e. the actual words said) in AM descriptions. Knowledge concerning the latently expressed words in AMs (i.e. the underlying meaning in the expressed words) is lacking. The latently described in AM is of importance to investigate as latent language use is an essential aspect of communication (Landauer, 2007); as such, the underlying meaning in verbally shared AMs.

In the current study the assumption of manifest and latent language use was based upon the cognitive perspectives of fuzzy-trace theory (Brainerd & Reyna, 1990, 2004) and knowledge representation within the framework of latent semantic analyses (LSA) as a theory of meaning (e.g. Landauer, 2007; Landauer & Dumais, 1997). Fuzzy-trace theory (Brainerd & Reyna, 1990, 2004; Reyna, 2012) suggests the existence of two parallel forms of memory representations; apparent verbatim representations of information and meaning-based—gist—representations of information. As such, the former are mental representations of precise event details whereas, as Brainerd and Reyna (2004) delineates “gist traces, on the other hand, are representations of semantic, relational, and other elaborative information about a memory target” (p. 399), for example, “the word COLLIE participates in meanings such as 'dog', 'farm animal', and 'pet'” (p. 401). LSA theory states that the semantics of expressed words are revealed in the contextual usage of words in natural language, and as such, reveals the underlying meaning in a description Landauer (2007). For example, LSA would likely show that the word collie co-occurs/is mentioned in conjunction with words such as dog, farm animal, and pet, in comparison to less semantically related words, such as boat. Accordingly, we suggest that the expression of verbatim mental representations could be operationalised as the manifest content, that is, the actual word said (e.g. collie) and the gist mental representations as the latent content, that is the underlying meaning in the expressed words (e.g. a collie is a dog which herds sheep at a farm).

In the present study, the memories were described verbatim. However, we suggest that with the aid of differentiated content analysis methods, the verbatim memory information and the gist memory information can be analysed. Following this, we have analysed the data by means of two computerised text analysis methods: (i) Linguistic Inquiry and Word Count (LIWC; e.g. Pennebaker, Francis, & Booth, 2001) and (ii) and Latent Semantic Analysis (LSA; e.g. Landauer, 1999, 2007; Landauer & Dumais, 1997). LIWC is a frequency-based method where words in a text are matched to words in specific categories (i.e. emotional processes). We operationalised LIWC to measure the manifest content. LSA is an analysis method for quantifying the semantic content in terms of semantic similarity between words or sets of words. We operationalised LSA to measure the latent content.

Previous research investigating gender differences in AM has predominately measured the manifest content, using manual coding, and in most cases found that females describe their memories in more emotional and communal terms than do males (e.g. Bauer et al., 2003; Study 2; Fitzgerald, 2010; Niedźwieńska, 2003; Thorne & McLean, 2002). Only a few studies investigating AM content have used LIWC (e.g. Hong et al., 2015; Pasupathi, 2007; Rice & Pasupathi, 2010; Schryer, Ross, St. Jacques, Levine, & Fernandes, 2012; Tani, Smorti, & Peterson, 2015). Of these studies, only Tani et al. (2015) investigated gender differences per se and found that females’ and males’ memory descriptions (i.e. AMs related to friendship) contained as many emotional expressions. However, LIWC has been used extensively in other research domains. For example, Newman, Groom, Handelman, and Pennebaker (2008) found in a study investigating gender differences in language use, that females expressed a higher frequency of words relating to psychological processes (i.e. emotions and sensations) and social processes (e.g. words related to family, friends, and humans) than males did. To our knowledge, only three studies have used LSA to investigate the AM content (Garcia et al., 2015; Garcia & Sikström, 2013; Karlsson, Sikström, & Willander, 2013), and none of them focused on gender.

Another motivation for performing this exploratory study was that gender differences in AM have never been investigated in a Swedish population. This is of particular interest because Sweden has been ranked in the global gender gap report (World Economic Forum, 2017) as one of the most gender equal societies in the world (no. 5 of 144 countries, see Table 1). Further, as illustrated in the World Value Survey (Inglehart et al., 2014) Sweden has been ranked (cf. 92 countries) as one of the countries with the least traditional cultural values (i.e. emphasis on the importance of religion, parent–child ties, deference to authority and traditional family values, cf. tolerance of foreigners, homosexual, and gender equality, rising demands for participation in decision-making in economic and political life).

Table 1. The global gender gap report 2017 (World Economic Forum, 2017). Gender equality index of countries described in the current study.

Country	Rank
Sweden	5
Denmark	14
Canada	16
Spain	24
Australia	35
Poland	39
USA	49
Italy	82
China	100
Qatar	130

As findings in one culture may not necessarily generalise to other cultures or subcultures (Ottsen & Berntsen, 2014), it is of interest to extend the literature concerning gender differences to a culture not previously studied and considered highly gender-equal and with low traditional values. Thus, our goal was to understand gender differences in the manifestly and the latently expressed words in AM descriptions with Swedish participants.

Given that the aim of the study was to address AM content from a broader perspective, the present study operationalised and focused on the content by means of: (i) overall content differences, (ii) agentic and communal expressions, and (iii) episodic details (i.e. emotional, cognitive and perceptual expressions). An additional aim was to investigate if the present study would replicate studies demonstrating gender differences in the number of retrieved specific AMs.

Communal and agentic expressions

It is commonly assumed that females are more communal than males and that males are more agentic than females (e.g. Bakan, 1966; Wiggins, 1991). The characteristics of agency include expressions of needs and pursuits of independence/ autonomy, individualism, mastery, achievement, self-assertion, power, and separation from others. On the other hand, communal characteristics include expressions of needs of and pursuits of interdependence/ being a part of a community (i.e. collectivism), cooperation, affiliation, intimacy, caring for and helping others, and maintenance of relationships (Abele & Wojciszke, 2014; Bakan, 1966; McAdams, Hoffman, Mansfield, & Day, 1996; Wiggins, 1991). In other words, a distinction is made between being more or less self-oriented or other-oriented.

Gender characteristics are hypothesised to be an outcome of gender socialisation in accordance with sociocultural norms and stereotypical gender role descriptions. More specifically, that gender norms and gender stereotypes are transferred to children by parents, the child-care system, teachers, peers and society (e.g. Bem, 1981; Bussey & Bandura, 1999; Carter, 2014; Leaper & Friedman, 2014; Maccoby, 2000). The experience of being an agentic or a communal individual has been proposed to influence several psychological functions and processes, such as self-perception, perception of others and social groups, and further to direct attitudes, motivation, behaviour, cognitive functions, interpersonal interactions, to form a part of the identity and characteristic adaption (Abele & Wojciszke, 2014; Bakan, 1966; Diehl, Owen, & Youngblade, 2004; Fivush, Bohanek, Zaman, & Grapin, 2012; McAdams, 2001; McAdams et al., 1996, 2006). Following this, gender differences in agentic and communal experiences should be reflected in females’ and males’ AMs.

Indeed, AM studies have demonstrated that females’ memory descriptions contain more communal expressions than males’ do (Buckner & Fivush, 1998, 2000; Ely, Melzi, Hadge, & McCabe, 1998; Niedźwieńska, 2003; Study 2, Thorne & McLean, 2002), and refer more to others than themselves (Study 2, Fitzgerald, 2010; Han, Leichtman, & Wang, 1998; Niedźwieńska, 2003; Wang, Leichtman, & White, 1998). Further, females rate the function of AMs higher in intimacy maintenance than males do (Pillemer, Wink, DiDonato, & Sanborn, 2003). However, there is a certain amount of conflicting evidence with some studies showing no difference between females and males in communal expressions in their AM content (McAdams et al., 1996; Pasupathi, 2007; Study 1, Thorne & McLean, 2002; Wang, 2001, 2004).

There is less support for the notion that males’ memory descriptions contain more agentic expressions than females’ do. On the contrary, in two studies, it has been found that females expressed more themes of agency in their AM's than males did (Fivush et al., 2012; Wang, 2001). Furthermore, no significant difference was found in two studies investigating gender differences in the number of agentic expressions in AMs (McAdams et al., 1996; Study 1, Thorne & McLean, 2002). To our knowledge, only two studies have shown some support for the notion that males’ AMs contain more agentic related themes (Niedźwieńska, 2003; Study 2, Thorne & McLean, 2002). Niedźwieńska (2003) showed that male participants scored higher in a sub-factor of agency (status/victory), but there was no gender difference in the total agentic score. Thorne and McLean (Study 2, 2002) found that males’ memory narratives contained more themes related to toughness (e.g. endurance in the face of adversity), which possibly could be considered to be related to mastery and achievement. We have not found any study showing that males refer more to others in their memory descriptions.

Episodic details and number of retrieved events

Females elaborate more when describing their memories than males do. Research has shown that females use (i) more episodic details as a composite measure of, for example, place, time, emotional, cognitive and perceptual details (Fivush et al., 2012; Han et al., 1998; MacDonald, Uesiliana, & Hayne, 2000; Wang, Hou, Tang, & Wiprovnick, 2011; cf. causal and insight language, Rice & Pasupathi, 2010), and (ii) especially, as a single measure, provide more emotional expressions in their memory descriptions than males do (e.g. Bauer et al., 2003; Buckner & Fivush, 1998; Niedźwieńska, 2003; Rice & Pasupathi, 2010; Ross & Holmberg, 1992; cf. Tani et al., 2015). In one study, Bauer et al. (2003) measured cognitive and perceptual expressions as separate categories and found no gender differences in either category. However, a statistical trend was found (p = .055), suggesting that females used more perceptual terms than males.

Seidlitz and Diener (1998) proposed that females’ superiority in AM recall could be driven by the number of details encoded at the original experience, which in turn would generate more potential retrieval cues and subsequently enhance memory recall. In support of their hypothesis, they showed that females provided more detailed event descriptions at encoding and recalled more events than males. Other studies have also showed that females retrieve more memories than males do (Davis, 1999; Fujita, Diener, & Sandvik, 1991; Ottsen & Berntsen, 2014; Pillemer et al., 2003; Rubin & Berntsen, 2009; cf. Goddard, Pring, & Felmingham, 2005; Willander & Larsson, 2007).

Seidlitz and Diener (1998) suggested that the female propensity to “encode their experiences in greater detail than men may be because their self-concepts are more defined in terms that reflect the interpersonal context of their experiences, including the perceptions and reactions of others” (p. 269). Further, females’ “relatively greater detail of encoding may reflect their stronger tendencies for affiliation and acuity of perception of others emotions in social interactions” (p. 269). Thus, Seidlitz and Diener (1998) provided a theoretical account for the female tendency to describe their memories in more detail, retrieve more memories, and express more communality than males do.

Females' tendency to elaborate more about their experiences than males has also been suggested to be based on early childhood learning of how to talk about their experiences with their caregivers (Fivush, Haden, & Reese, 2006, p. 1580). In accordance with the socialisation theory, individual- culture- and gender differences have been explained by differences in parent–style conversation with the child about her/his experiences (for overviews see, Fivush et al., 2006; Fivush & Nelson, 2004; Fivush, 2011; Nelson & Fivush, 2004). A notion which has found support in several studies showing that children who have parents who elaborate more with their child about the child's experience and encourage them to participate in the conversation have children with better AM skills (e.g. Wang, 2006; cf. Bauer & Larkina, 2014; for overviews see, Fivush et al., 2006; Fivush & Nelson, 2004; Fivush, 2011). Concerning gender, parents tend to elaborate more about the child's experience with their daughters compared to their sons (e.g. Fivush, Berlin, Sales, Mennuti-Washburn, & Cassidy, 2003; Reese & Fivush, 1993). This is shown to be especially true about emotional aspects of events (e.g. Adams, Kuebli, Boyle, & Fivush, 1995; Aznar & Tenenbaum, 2015; interestingly, sad events, not happy, angry or fearful ones, Fivush, Brotman, Buckner, & Goodman, 2000). Furthermore, conversations about emotional experiences are placed in an interpersonal context to a greater extent with daughters than with sons (Fivush et al., 2000, 2003), that is, in line with a communal orientation. Correspondingly, girls participate more in memory conversations (Reese & Fivush, 1993), remember more events (Experiment 1, Davis, 1999), elaborate more (Fivush et al., 2003) especially about emotional aspects (e.g. at age six but not at age three, Adams et al., 1995; Bauer et al., 2003; Buckner & Fivush, 1998; Americans not Asians, Han et al., 1998, cf. Wang, 2004), and as mentioned earlier describe more communal aspects in their AMs (e.g. Buckner & Fivush, 2000) than boys do.

The hypothesis concerning parent–style conversation and how children's memories are expressed (e.g. Fivush et al., 2006) has found further support in studies investigating cultural differences (e.g. Wang, 2006). American parents elaborate more and talk more about the child's experience than Chinese parents, while Chinese parents encourage their children to not talk about themselves but instead focus on the collective (e.g. Wang, 2006; Wang, Leichtman, & Davies, 2000). Subsequently, the parental conversation style is reflected in the child's memory descriptions in a similar direction. As such, American childrens' (and adults’) memory descriptions contain more details, emotional expressions and tend to be self-oriented, while Chinese are sparser, more emotionally neutral and interdependently oriented (e.g. Han et al., 1998; Wang, 2001, 2004, 2006). Interestingly, in a cross-cultural study a gender difference was found in American children (i.e. girls elaborated more than boys), but not among Korean and Chinese children (Han et al., 1998). In contrast, Wang (2001) found a gender difference in adult Chinese but not among American adults, such as Chinese females’ AM descriptions were lengthier and more autonomously oriented than Chinese males’ AM descriptions. It has been discussed that child-rearing practices are similar between the genders in China (Han et al., 1998; Wang, 2001; Wang et al., 1998) or even that there is an emphasis on female competence in the Chinese culture (Wang, 2001). This line of research highlights the importance of further investigation concerning the interaction between socialisation, culture, and gender differences in AMs.

In summary, based on previous research, we wanted to explore gender differences in (i) the AM content overall, (ii) communal and agentic expressions, and (iii) episodic details (i.e. emotional, cognitive and perceptual expressions), especially emotional terms. We investigated these items in the manifestly, and the latently expressed words in AM descriptions. Further, we also studied differences in the number of retrieved events.

Method

Participants

Forty-two undergraduate students at the Department of Psychology, Stockholm University (21 females, 21 males, mean age of 26.91 years [SD = 6.26], age range 19–42, 76% [the same for females and males] were between 20 and 30 years of age) participated in the study. The genders did not differ in age, t(40) = .92, p = .50, 95% CI: −3.75 to 4.16 (Mfemales = 27.01, SD = 6.88; Mmales = 26.81, SD = 5.75). The students gave written informed consent and participated in the study for course credits.

The present study is part of a larger project and two other studies focusing on the cueing effects have previously been published on this dataset (Karlsson et al., 2013; Willander, Sikström, & Karlsson, 2015), in which participants were randomly allocated to one of the four cue-conditions with 20 participants per condition. However, for the present study, the data were collapsed across cue-condition. The original data set consisted of 80 participants (59 females, 21 males). In order to achieve an even gender distribution in the present study, we randomly selected 21 females from the original group of 59 females. The sampling was performed so that the number of females equalled the number of males in the respective cue-condition (visual five females/five males; auditory four females/four males; olfactory five females/five males; multimodal seven females/seven males).

Materials

The stimulus material consisted of four cue-modalities, three unimodal sensory cues (visual, auditory, and olfactory), and one multimodal cue where the visual, auditory, and olfactory cues were presented simultaneously. The participants were presented with 15 cues in each condition. The cues represented contexts covering common events familiar to most people (see Appendix). We did not observe any significant differences between females’ and males’ number of retrieved memories with regard to context or sensory retrieval cue.

The presentation of the retrieval cues was as follows (exemplified by the bar context). In the visual condition the participants were presented with a chromatic picture of a glass of beer on a bar table; in the auditory condition with typical bar sounds (no music) as a beer being poured and glass colliding; in the olfactory condition with the smell of beer; and in the multimodal condition all the previously mentioned matching cues were presented simultaneously. The visual cues were presented on a 22-inch LCD computer screen. The auditory cues were presented using a pair of AKG 701 reference headphones attached to the same computer on which the visual cues were presented. The olfactory cues were presented in dark glass jars to prevent visual identification of the cue. The participants held the glass jars themselves and were instructed when to start sniffing.

Procedure

The participants were asked to retrieve any specific AM (i.e. a personally experienced event occurring at a single time point typically spanning seconds to hours) from any part of their lives evoked by the retrieval cue. The cues were presented in a randomised order that was unique to each participant. Thirty seconds was allowed for retrieval. In case no memory was retrieved within this time, the next cue was presented. In the case of successful retrieval, the participants wrote down a “title” of the memory. Next, they verbally described the memory in as much detail as possible in three minutes. These verbalised event descriptions were recorded with a digital voice recorder and afterward transcribed to text.

The analysis methods: LIWC and LSA

Linguistic inquiry and word count

LIWC measures the percentage of words pertaining to specific linguistic and psychological categories in relation to the total amount of words in a text (Pennebaker, Boyd, Jordan, & Blackburn, 2015; Pennebaker et al., 2001; Tausczik & Pennebaker, 2010). LIWC2001 contains 85 categories capturing different dimensions (see.www.liwc.net). Words such as “happy” and “sad” are examples of words listed in the category “emotional expressions”. In the LIWC method, the percentage of the number of words in a given description that is included in each category is calculated. For example, the AM description “I was very happy” would result in a LIWC score of 25% for the emotional category, because one of the four words (happy) is an emotional word.

To tap psychological dimensions relevant for the present study we used ten wordlists from the LIWC2001 version (Pennebaker et al., 2001) and two from the 2015 version (Pennebaker et al., 2015; see Table 2 for information regarding the included wordlists). In the Swedish translation, these wordlists contain between approximately 33–1445 word-stems. The reason for mainly using the 2001 version of LIWC was that only 2001 was available in Swedish translation, but it did not contain the wordlists “achievement” and “affiliation” as the 2015 version did. Because we were interested in the latter wordlists, these were translated and added.

Table 2. LIWC, wordlists used.

Category	Example words
Agency
Achievement^a	achievement, award, competence, goal, purpose, successful, victory
Power^b	power, demand, law, masterful, rebellious, ruler, victim, worthless
Communality
Affiliation^b	affiliation, ally, compromise, comradely, hug, reconcile, togetherness
Social processes^a	interact, affair, band, celebrate, people, meet, involve, help, share
Family^a	aunt, sister, brother, cousin, family, grandchild, mother, niece
Friends^a	buddy, comrade, fiancé, fellow, friend, guest, pal, roommate
Humans^a	adult, chick, guy, human, individual, kid, ladies, partner
References to others^a	everybody, he, her, him, our, she, someone, them, they, we
Communication^a	communicate, admit, express, inform, reply, say, suggest, talk, ask
Emotional processes^a	emotional, happy, angry, hateful, lively, sad, anxious, satisfied, joy
Cognitive processes^a	analyse, aware, because, decide, expect, guess, know, remember,
Perceptual processes^a	view, drank, ear, hear, look, itch, noose, observe, smell, squeeze

^aLIWC2001.

^bLIWC2015.

To identify the concepts relevant for our investigation, we merged several LIWC categories. Specifically, we operationalised agency to be the use of words from the wordlist's achievement and power. Communality was operationalised to be the use of words from the wordlist's affiliation and social processes. The category social processes contain the subcategories family, friends, humans, references to others, communication and also some words related to social processes not included in the subcategories (i.e. non-communicative social interaction terms). The wordlist perceptual processes include visual, auditory, and tactile related words. Olfactory related words were added to the perceptual processes’ wordlist by the first author. All LIWC wordlists were translated from English to Swedish by the first author and are available upon request.

Latent semantic analysis

The basic assumption of LSA as a method is that the closeness of two semantic representations (i.e. words’ assigned vector values) in a semantic space reveals the underlying meaning in a description (e.g. Landauer, 1999, 2007; Landauer & Dumais, 1997).

Giving an extensive description of the LSA technique is out of the scope of the present article. For thorough descriptions see, for example, Garcia et al. (2015), Kjell, Kjell, Garcia, and Sikström (2018) and Landauer (2007). For a conceptual understanding, the semantic space will be briefly described, followed by a description of how the semantic representations and the similarity scores were generated, and finally, an illustrative example of semantic similarity.

The semantic space is a set of vectors constructed from a large set of words and their co-occurrence. The semantic space was constructed by using one of the largest corpora of text in Swedish, that is, the Swedish version of Google Ngrams (approximately 1 TB, 30 trillion words). All semantic analyses were made with the web-based LSA software Semantic Excel (www.semanticexcel.com), which has been developed by our research group. The software comes with the above-mentioned semantic space included.

The semantic representations (i.e. vectors) were generated by the LSA algorithm as follows. First, a semantic representation for each word included in the memory descriptions and each word included in the LIWC wordlists was generated. We used the words included in the LIWC wordlists to generate semantic representations to tap the same psychological and social processes analysed with the LIWC technique. Second, by summarising the semantic representations included in each participants’ memory descriptions and each wordlist, respectively, a semantic representation for each participant and each wordlist were generated.

These semantic representations were then used to calculate a semantic similarity score between the female and male participants’ semantic representations, respectively, and the wordlist's semantic representations, respectively, by measuring the cosine of the angle between these two semantic representations. For example, the angle of the cosine between the female participant's semantic representation and the semantic representation of the emotional processes’ wordlist was calculated. The resulting value (i.e. semantic similarity score) constitute a semantic similarity score, and it spans from +1 (identical) to −1 (maximal dissimilarity), where zero represents independence. To give an intuitive understanding of semantic similarity, consider the following example:

I remember an event when I was, due to a bet – in a boldness contest against a friend – out on thin ice covering a river, knowing the water below to be deep, cold and current. I heard the sound of moving ice. My legs were trembling, but I moved forward. I made it to the other side. I succeeded and won the bet! I felt proud.

In the paragraph, not one word regarding the gender of the competitor is mentioned, but if males are more agentic than females, the memory should have been described by a male. This is because the description includes words related to agency, as bet, boldness, competition, succeeded and proud, which all are words included in the LIWC wordlists achievement and power. Consequently, the presumed male participant's semantic representation of the memory description would be closer in the semantic space (i.e. more similar) to the semantic representation of the wordlists achievement and power, than the semantic representation of a female's memory description, which presumable would not contain those words to the same extent. In comparison, if females are more communal than males, a female's memory description would contain more words related to communality than a male's memory description. For example, “I remember, when I was out skating with a friend. She helped me do pirouettes. The next day I showed my family that I could do it. They complimented me. I felt proud”. The words in italics are included in the wordlists, affiliation, friends, family, and other references. Accordingly, the semantic representation of the female's memory description would be closer to the semantic representations of the wordlists affiliation, friends, family, and other references than a male's memory description, which presumable would not contain those words to the same extent.

In the present research, we used the AM semantic representations to predict gender differences in the semantic content overall (for details concerning semantic-numeric correlations see Kjell et al., 2018), and second, the semantic similarity scores to analyse by variance, the difference between the genders and each LIWC wordlists.

Results

All participants were able to retrieve specific AMs (M = 10.45, SD = 2.95, Range: 4–15 events). In total, the database of retrieved AMs consisted of 439 events. The mean amount of words used to describe the events was 445.62 (SD = 100.99; Range: 217–611 words).

The overall semantic content differs

A multiple linear logistic regression was conducted with the semantic representations (i.e. the word vectors) of the AMs as predictors and gender as the outcome variable. The coefficients of the regression analysis were used to predict gender. The number of semantic dimensions used in the prediction was optimised to a mean value of 8, using an N-leave-out procedure. The result was evaluated by the standard N-leave-out (N = 10) cross-validation procedure (Kohavi, 1995) and yielded a significant point-biserial correlation between gender and gender as predicted from the semantic representation, r(42) = .57, p < .0001, one-tailed (95% confidence interval .32–.74). Thus, gender differences were observed in the semantic content.

Communal and agentic expressions

Two separate MANCOVAs were performed; one based on the LIWC frequency scores and one based on the LSA semantic similarity scores. The LIWC MANCOVA included the nine categories as the dependent variables and gender as the independent variable. The MANCOVA also included the modality of the sensory cues as a covariate. No gender differences were found in communal or agentic expressions in any of the main or subcategory scores (F_1,39 = .89, p = .55, ${\eta }_p^2$ = .21, observed power = .39). However, the LSA MANCOVA with the nine categories as the dependent variables, gender as the independent and sensory cue modality as a covariate showed a significant gender difference (F_1,39 = 3.29, p = .006, ${\eta }_p^2$ = .50, observed power = .89). The sensory cue covariate was not significant (p > .05). Compared to the males’, the females’ semantic representation was closer to the semantic representations of the three categories family, friends, and humans. For LSA, no other significant differences were observed (p > .05). See Table 3 for the MANCOVAs, Table 4 for the LIWC descriptive statistics and Table 5 for the LSA descriptive statistics.

Table 3. LIWC and LSA, MANCOVA statistics for gender differences in agentic and communal expressions in AM descriptions (n = 42).

	LIWC				LSA
Categories	F (1,39)	p	${\eta }_p^2$	F (1,39)	p	${\eta }_p^2$
Agency
Achievement	.327	.571	.008	3.16	.083	.075
Power	.181	.673	.005	2.65	.112	.064
Communality
Affiliation	2.78	.103	.067	1.15	.291	.029
Social processes	3.23	.080	.076	1.31	.259	.033
Family	.333	.567	.008	5.88	.020	.131
Friends	.210	.649	.005	4.16	.048	.096
Humans	.012	.913	.000	5.71	.022	.128
Communication	2.64	.112	.063	.033	.856	.001
References to others	6.63	.014	.145	.790	.380	.020

Table 4. LIWC, percentage frequency scores for words pertaining to agency and communal expressions (n = 42).

	Females		Males
Categories	M (SD)	95% CI	M (SD)	95% CI
Agency
Achievement	.305 (.125)	[.20, .40]	.333 (.180)	[.30, .40]
Power	3.19 (1.04)	[2.8, 3.6]	3.31 (.734)	[2.9, 3.7]
Communion
Affiliation	2.79 (.923)	[2.4, 3.2]	2.33 (.845)	[1.9, 2.7]
Social processes	4.90 (.984)	[4.4, 5.4]	4.33 (1.05)	[3.9, 4.8]
Family	1.11 (.424)	[.90, 1.3]	1.21 (.540)	[1.0, 1.4]
Friends	.208 (.176)	[.10, .30]	.234 (.120)	[.20, .30]
Humans	1.11 (.476)	[.90, 1.4]	1.13 (.584)	[.90, 1.4]
Communication	.781 (.279)	[.70, .90]	.643 (.266)	[.50, .80]
Reference to others	5.22 (1.16)	[4.7, 5.7]	4.32 (1.10)	[3.8, 4.8]

Note: CI = confidence interval.

Table 5. LSA, semantic similarity scores pertaining to agentic and communal expressions (n = 42).

	Females		Males
Categories	M (SD)	95% CI	M (SD)	95% CI
Agency
Achievement	.089 (.006)	[.085, .092]	.920 (.007)	[.089, .095]
Power	.079 (.004)	[.077, .081]	.081 (.004)	[.079, .083]
Communion
Affiliation	.077 (.006)	[.074, 081]	.075 (.008)	[.072, .078]
Social processes	.187 (.008)	[.183, .191]	.184 (.010)	[.180, 188]
Communication	.125 (.007)	[.121, .128]	.124 (.008)	[.120, .128]
Family	.068 (.006)	[.065, .071]	.063 (.007)	[.060, .066]
Friends	.031 (.006)	[.029, .034]	.028 (.006)	[.025, .030]
Humans	.148 (.008)	[.145, .151]	.142 (.008)	[.139, .146]
Reference to others	.804 (.007)	[.800, .807]	.801 (.008)	[.798, .805]

Note: CI = confidence interval.

Episodic details and number of retrieved events

Emotional, cognitive and perceptual expressions

Two separate MANCOVAs were performed, one on LIWC percentage scores and another on the LSA semantic similarity scores. In both analyses, with the nine categories as dependent variable, gender as independent and sensory cue modality as a covariate, no gender differences were found (LIWC: F₁,₃₉ = .28, p = .84, ${\eta }_p^2$ = .02, observed power = .1; LSA: F₁,₃₉ = .85, p = .48, ${\eta }_p^2$ = .07, observed power = .22). The results for the MANCOVAs are presented in Table 6, the LIWC descriptive statistics in Table 7, and the LSA descriptive statistics in Table 8. Thus, females did not use more emotional, cognitive, or perceptual expressions in their AM descriptions than males did.

Table 6. LIWC and LSA, MANCOVA statistics for gender differences in emotional- cognitive- and perceptual expressions in AM descriptions (n = 42).

	LIWC			LSA
Categories	F (1,39)	p	${\eta }_p^2$	F (1,39)	p	${\eta }_p^2$
Emotional processes	.010	.923	.000	.835	.366	.021
Cognitive processes	.090	.766	.002	.917	.344	.023
Perceptual processes	.761	.388	.019	.920	.343	.023

Table 7. LIWC, percentage frequency scores for words pertaining to emotional- cognitive- and perceptual expressions (n = 42).

	Females		Males
Categories	Mean (SD)	95% CI	Mean (SD)	95% CI
Emotional processes	1.78 (.373)	[1.6, 2.0]	1.77 (.435)	[1.6, 1.9]
Cognitive processes	11.56 (1.94)	[10, 12]	11.38 (1.90)	[11, 12]
Perceptual processes	1.74 (.365)	[1.6. 1.9]	1.64 (.323)	[1.5, 1.8]

Note: CI = confidence interval.

Table 8. LSA, semantic similarity scores pertaining to emotional- cognitive- and perceptual expressions (n = 42).

	Females		Males
Categories	Mean (SD)	95% CI	Mean (SD)	95% CI
Emotional processes	.057 (.004)	[.055, .059]	.058 (.005)	[.056, .060]
Cognitive processes	.229 (.008)	[.224, .233]	.232 (.011)	[.227, .236]
Perceptual processes	.193 (.008)	[.189, .197]	.190 (.010)	[.187, .194]

Note: CI = confidence interval.

Number of retrieved events

An ANCOVA, with the number of retrieved events as the dependent variable, gender as between group variable and sensory cue modality as a covariate, showed no significant gender difference, F(39) = .07, p = .79, 2% CI: 9.55–11.36 (Mfemales = 10.57, SD = 2.96; Mmales = 10.33, SD = 3.01).

Discussion

The results in the present study showed an overall difference in the latently expressed words between females and males with regard to the AM content. Furthermore, the analyses showed that females latently expressed more aspects of communality in their AMs than the males did. No significant gender differences were observed in agentic themes, emotional, cognitive or perceptual expressions, in the manifest or latent content, or in the number of retrieved events.

Thus, the results partly replicated previous research by demonstrating that females latently described their AMs in more communal terms than males did (e.g. Niedźwieńska, 2003). Further, the results in the present study may be of particular interest, given that studies performed in societies, such as China, Poland, and the USA have found differences in the manifest content of AM, whereas our results in a Swedish context only demonstrated significant gender differences in the latent content. In the present research the verbatim experience of events was operationalised as the manifest content (i.e. the actual words said), and the gist experience of events as the latent content (i.e. the underlying meaning in the expressed words [Landauer, 2007]). According to fuzzy-trace theory (Brainerd & Reyna, 1990, 2004; Reyna, 2012), verbatim and gist information are roughly encoded, stored, and retrieved in parallel. Accordingly, we interpreted the results as demonstrating that females’ gist experience of events is more communally oriented than males are.

Communal and agentic expressions

Our results showed that a sample of Swedish young adult females and males differed in their expression of communality (i.e. family, friends, humans) in the AM content. However, we did not find any differences with regard to agency. Concerning communality, the results replicate previous findings in other societies (e.g. Poland, Niedźwieńska, 2003; USA, Thorne & McLean, 2002). One possible explanation could be that although gender equality has been in focus in Sweden since the 1970s (Bayne, 2009; Björklund, 2007) and social policies endorse both males and females to be equally represented in all domains of society, such as in positions of power and family care (Björklund, 2007; Gender Equality in Sweden, 2013–2018, Swedish Institute), there is still gender inequality. Females are to a larger extent performing typical long-established female chores, such as taking care of the household, children and elderly compared to males (Björklund, 2007; Gustafsson & Szebehely, 2001; Nordic Gender Equality in Figures, 2015). Furthermore, females tend to enrol in caring-oriented educations and work in healthcare to a larger extent than males (Statistics Sweden, 2016). Therefore, females’ tendency to describe more communal aspects in their AMs than males could be a manifestation of females’ greater involvement in caring oriented occupations.

However, the gender difference in the present study was found in the latently described and not as shown in previous research in the manifestly described (e.g. Poland, Niedźwieńska, 2003; USA, Thorne & McLean, 2002). Concerning the manifest content, it is possible that females’ and males’ verbatim communal experience is less gender-role divided in Sweden, than in other less gender equal societies with more traditional culture values (Inglehart et al., 2014; World Economic Forum, 2017). Additionally, in line with the implicit social cognition theory (Greenwald & Banaji, 1995), the communal latent language use could be an expression of conduct in accordance with implicit traditional communal gender norms, gender stereotypes and gender role identifications, which do not necessarily comply with societal gender norms and public gender equality policies (Bayne, 2009; Björklund, 2007; Leaper & Friedman, 2014). In any event, females’ gist of experience could be more communally oriented than males’ gist of experience, which the measure of semantic similarity could capture.

However, we do not argue that males’ mental representations do not contain communal themes of experiences. Instead, males’ reduced expression of communality when describing their AMs could, for example, be due to a prevalent masculinity norm. For example, compared to females it is less acceptable for males to diverge from traditional gender role prescriptions (e.g. Bayne, 2009; Leaper & Friedman, 2014; Lenton, Sedikides, & Bruder, 2009; Levant, 2011; Salaby & Frey, 1975, as cited in Bussey & Bandura, 1999). Conforming to the masculinity norm could thus prevent males from expressing communality and prevent caregivers from encouraging boys to be communal.

Concerning agency, the present study replicated previous research (McAdams et al., 1996; Study 1; Thorne & McLean, 2002) by showing a non-significant difference between the genders in an agentic orientation in the AM descriptions.

Episodic details and number of retrieved events

No differences were found in either the LIWC analyses or in the LSA analyses regarding emotional- cognitive and perceptual expressions, or the number of retrieved events. Thus, our results did not replicate previous research related to these dimensions (e.g. Bauer et al., 2003; Experiments 1–5, Davis, 1999; Pillemer et al., 2003; Wang et al., 2011). To our knowledge, all previous studies have been performed in countries that are not as gender equal as Sweden (see Table 1, World Economic Forum, 2017).

One possible account for this lack of difference in our Swedish sample could be that Sweden, according to the global gender gap report (World Economic Forum, 2017) shows a rather small educational gender gap. For example, pre-school and school educators are required to provide girls and boys with opportunities to behave and interact on the same terms (Gender Equality in Sweden, 2013–2018, Swedish Institute; Sex Education in the Swedish Curricula, 2016). Following the rationale of the details at encoding hypothesis (Seidlitz & Diener, 1998), in conjunction with the parental conversation style hypothesis (e.g. Fivush et al., 2006; Nelson & Fivush, 2004), we suggest that if girls and boys are encouraged to elaborate to the same degree with their caregivers about their experiences it should be reflected in the same number of event details remembered and memories retrieved.

In sum, we have suggested that the lack of gender differences in most variables could be due to societal implementations, fewer gender norms, gender stereotypes and gender role identifications in the Swedish in comparison to other cultures. However, in the present study, cultural values and gender equality references were mentioned to put the data in a relevant context. For the future it would be of interest to perform cross-cultural studies between societies more or less gender equal and with more or less traditional values (Inglehart et al., 2014; World Economic Forum, 2017). This is because, as shown in previous cross-cultural research, gender differences can be found in American but not Asian children's memory descriptions (Han et al., 1998). Further, as in the present study, research has found there to be no difference between the genders in agentic related themes in AM descriptions among graduate students (McAdams et al., 1996; Study 1, Thorne & McLean, 2002). This is in contrast to studies which have demonstrated that female teenage students describe their memories in more agentic terms in an American sample (Fivush et al., 2012) and a young adult Chine sample (Wang, 2001), while in a Polish middle-aged non-student sample, males expressed more agentic terms (Niedźwieńska, 2003). Thus, gender differences across societies and age-cohorts merit further investigation.

Methodological considerations

A potential limitation of the current study could be the sample size. With a larger sample, more minute differences between females and males may have been observed. However, it should be noted that the amount of data per individual is substantial, which in turn has enabled a more fine-grained gender comparison.

Concerning, the stimulus material, the cues were meant to target specific contexts common in most peoples’ lives. We aimed to develop stimulus material that was unbiased across the various cues. That is, one cue should not differ with regard to gender, age, and so on. Since we did not observe any significant gender differences across the cue conditions and contexts, the data were collapsed across the cue conditions and contexts. Although the null-hypothesis should not be interpreted as support for no differences, we still argue that these results provide some indirect support for our view that the material may be unbiased. However, with this said it would, of course, be of interest to further investigate this issue concerning society, as possible context-dependent gender differences could occur.

Future directions

It is interesting to note that LSA detected the observed gender differences. It would be of further interest to investigate the LSA technique in relation to other content analytic methods. We have argued that LSA can capture the underlying meaning in the expressed words. For the future, it would of interest to design a study to investigate this suggestion explicitly. If such comparison shows that LSA would indeed be more potent at capturing the underlying meaning compared to other content analytic methods, it could potentially open new directions for the method to be systematically applied in for example forensic and therapeutic practices. Another interesting application could be the building of a gender equality index based on LSA.

Conclusion

In the present study, we observed gender differences in the latently expressed AM content. Females latently expressed more aspects of communality in their memory descriptions than males did. In contrast, we did not observe any differences in the manifest content. Specifically, there was no difference in the latent or manifest content regarding agentic terms or episodic details, or in the number of retrieved events. We suggest that one explanation for these patterns of findings could be that females’ gist experiences of events are more communally oriented than males’.

Acknowledgment

Assistance provided by Ph.D. Ivo Todorov was greatly appreciated.

Disclosure statement

No potential conflict of interest was reported by the authors.

Additional information

Funding

This work was supported by Swedish Research Council: [Grant Number SRC grant #2008–2415].

Appendix

Table

Context	Pictures	Sounds	Odors
Bar	Bar table with a glass of beer	Typical bar sounds (no music), beer being poured, glass colliding	Beer
Cafe	Cafe table with a cup of coffee	Typical cafe sounds (no music), plates, spoons, cups colliding, coffee being poured into a cup	Coffee
Gas station	Car being filled with gasoline at a gas station	Typical gas station sounds, (e.g. gas pump)	Gasoline
Indoor swimming bath	Indoor swimming bath	Typical indoor swimming bath ambient sounds, water splashes, laughters	Chlorine
Camp fire	Camp fire	The sound of a burning camp fire	Burned-smell
Cigarette	Cigarette, cigarette smoke, ashtray	Lighting a cigarette, inhale/exhale sounds	Smoked cigarette
Cooking	Chopping onion	Chopping onion	Onion
Dentist	A dentist with a drill	Dentist drill and saliva suction	Eugenol
Eating an apple	Half-eaten apple	Biting and chewing an apple	Apple
Garden	Garden with flowers	Ambient garden sounds as bees	Flower
Harbour	Harbour by the sea with fishing boats	Boat, see and seebirds	Fish (Baltic herring)
Mopping a floor	Mop and bucket	Mop swirl, dipping a mop in a bucket with water	Soft soap
Paint a wall	Painting wood panels	Opening a jar of paint, stirring paint, brush strokes	Wall paint
Forest walk	Pine forest	A hike in the woods, wooden sticks breaking	Pine
Washing clothes	Washing machine	Washing machine in action	Washing powder