Sex Roles

, Volume 55, Issue 3, pp 259–266

Implicit and Explicit Occupational Gender Stereotypes

Original Article

DOI: 10.1007/s11199-006-9078-z

Cite this article as:
White, M.J. & White, G.B. Sex Roles (2006) 55: 259. doi:10.1007/s11199-006-9078-z


This study was designed to compare implicit and explicit occupational gender stereotypes for three occupations (engineer, accountant, and elementary school teacher). These occupations represented the end points and middle of a masculine–feminine continuum of explicit occupational gender stereotypes. Implicit stereotypes were assessed using the Implicit Association Test (IAT), which is believed to minimize self-presentational biases common with explicit measures of occupational gender stereotypes. IAT results for the most gender stereotyped occupations, engineer (masculine) and elementary school teacher (feminine), were comparable to explicit ratings. There was less agreement with less stereotyped comparisons. Results indicated that accounting was implicitly perceived as more masculine than explicit measures indicate, which calls into question reports of diminishing gender stereotyping for such occupations.


Occupational gender stereotypes Implicit stereotypes Stereotypes Implicit Association Test 

Popular beliefs have long held that because of their stereotyped traits and temperaments men and women are suited for different kinds of occupations. One of the earliest empirical examinations of these occupational gender stereotypes was conducted by Shinar (1975) who showed that college students thought that some occupations required masculine traits, while others required feminine traits. The method that Shinar (1975) and others (Beggs & Doolittle, 1993; White, Kruczek, Brown, & White, 1989) used to study occupational stereotypes is the traditional method of measuring stereotypes of all types. Indeed, it was first used by Katz and Braly (1935) in their very early work on national stereotypes. This approach treats stereotypes as a collection of traits or attributes that the respondent consciously and explicitly associates with members of different groups. Most conceptual treatments of stereotypes, and all popular accounts, have emphasized these explicit processes and their contents.

Persons acquire stereotypes, in part, through personal experience. But because stereotypes are part of the beliefs and shared assumptions that societies have about different types of people and groups, they are also part of the society’s collective knowledge. In order for a society to socialize its members, these stereotypes must be explicitly, even if subtlety, taught (Stangor & Shaller, 1996). Whether stereotypes are individual or cultural in origin, the emphasis on explicit beliefs is not surprising considering that the content of stereotypes has great intrinsic interest to both the person using the stereotype and the person targeted by it. Even when objectively wrong, stereotypes simplify social perception and serve as guidelines for social interaction.

It is increasingly clear that implicit processes are important in stereotyping. Greenwald and Banaji (1995, p. 15) have defined implicit stereotypes “as the introspectively unidentified (or inaccurately identified) traces of past experience that mediate attributions of qualities to members of a social category.” Implicit stereotypes and other implicit cognitive forms reflect the continuing influence of past experience and learned associations. They are the remaining influence of explicit beliefs that, although consciously abandoned or rejected, continue to influence cognition and perception. This influence is often beyond conscious control and may be invoked or primed by briefly presented stimuli (cf., Fazio, Sanbonmatsu, Powell, & Kardes, 1986). Even among those persons who explicitly disavow bias toward out-group members, appropriate priming may trigger implicit stereotyped judgments (Banaji, Hardin, & Rothman, 1993). Stereotypes may thus exist and continue to bias perceptions at an implicit level, even if they are not present at an explicit level (Kunda & Spencer, 2003).

A strategy for describing implicit stereotypes and other implicit cognitions is provided by the Implicit Association Test (IAT; Greenwald, McGhee, & Schwartz, 1998). The IAT assesses implicit stereotypes by measuring their underlying automatic associations with other concepts. This is done by first establishing the speed with which responses can be made to a computer-presented target-concept and an associated attribute. Although the IAT procedure will be explained later in greater detail, consider for now a response to the target-concept “nurse,” which has been paired with the attribute “female.” This is a commonly stereotyped association; responses should be made rapidly. This is because the strong association between “nurse” and “female” facilitates quick retrieval and cognitive processing. The IAT procedure then reverses the visual presentation such that the target-concept, “nurse” is paired with the attribute “male.” This is not the stereotypical association and should result in slower responses despite the respondent’s best conscious efforts. The prior implicit association or stereotype will interfere with the respondent’s answer. Further, the strength of this effect will be influenced by the strength of the pre-existing stereotype. If the stereotype is strong or well established, the effect will be larger. If it is weak, the effect will be smaller or non-existent because there is no prior association to overcome.

Explicit traits for occupational gender stereotyping are, by contrast, assessed using familiar Likert-type rating scales. The most common explicit traits measured by these scales involve whether an occupation should be considered masculine, neutral, or feminine. Nursing, for example, has been consistently rated as a “feminine” occupation (White et al., 1989). Essentially two explanations for these explicit stereotypes have been made. One is that certain jobs require personality traits more likely to be found in one gender. If, for example, a good nurse should be caring and women are perceived as more caring than men, then it follows that women would make better nurses than men (Spence & Helmreich, 1978). A second explanation involves which sex is more prevalent in the occupation. Despite increases in the number of men who are nurses, most nurses are women. Because women predominate in nursing, it confirms the observation, the stereotype, that nursing is an occupation best occupied by women (cf. Glick, Wilk, & Perreault, 1995).

Have the explicit occupational stereotypes that Shinar (1975) identified 30 years ago changed since her study? Results have been mixed. Stereotypes attached to some occupations appear to have become more gender-neutral. This is especially true of occupations where the ratio of male to female practitioners has become more balanced. Other occupations, usually those with skewed sex ratios, remain gender-typed (e.g., Beggs & Doolittle, 1993; Cejka & Eagly, 1999; White et al., 1989). Yet, methods used in these prior studies have all focused on explicit stereotypes that are prone to social desirability and self-presentational effects. Even those who consciously accept occupational gender stereotypes may hesitate to express them (cf. Yoder & Schleicher, 1996). These stereotypes are often socially unacceptable and, at least in the United States, associated with potentially illegal sex discrimination (e.g., Civil Rights Act of 1964, Title VII). Moreover, because implicit stereotypes may persist long after they are no longer explicitly accepted, occupational gender stereotyping may persist even among those who consciously disavow it. Researchers who use explicit stereotype measures may thus underestimate occupational gender stereotyping.

In the present study we considered implicit occupational stereotypes for three occupations (i.e., engineer, accountant, and elementary school teacher). These occupations represent the middle and end points of a masculine–feminine continuum of explicit occupational gender stereotypes identified by White et al. (1989) in their replication and extension of Shinar’s (1975) original work. In order to allow comparisons with implicit stereotypes, measures of explicit stereotypes for these occupations were also made. It was hypothesized that the occupation pair with the most pronounced difference in explicit gender stereotypes would have a larger IAT latency (i.e., a larger effect) and, accordingly, a stronger implicit stereotype than the other pairs.

Materials and Methods


A total of 156 students from two colleges within the university participated voluntarily. Most of the students (66 men, 55 women) were business majors. The rest (12 men, 23 women) studied education. The mean age was 21.8 (SD = 4.7). Students reported their ethnicity as follows: Caucasian American (85.9%), African American (9.6%), Asian American (1.3%), Native American (.6%), Hispanic American (1.9%), and Other (.6%). They indicated their class standing to be: freshman (3.2%), sophomore (25%), junior (41.7%), senior (26.3%), graduate (2.6%), and other (1.3%). Students earned extra credit, research participation points according to plans approved by their respective course instructors.


Implicit stereotypes

Three occupations with similar educational requirements formed the occupation target pairs for the IAT (i.e., engineer, accountant, and elementary school teacher). These occupations represented the first, third, and fifth quintiles of the distribution for gender typing reported by White et al. (1989). Engineer was rated as a masculine gender-typed occupation by these earlier respondents; elementary school teacher was feminine gender-typed. Even though accountant was in the middle or neutral quintile, ratings placed it toward the masculine range of occupations (3.39 on a 1–7 scale).

Stereotyped attributes of these occupations were identified in a preliminary study. Undergraduates (n = 62; 13 men, 47 women) identified the five words that most quickly came to mind when thinking about the occupations. Respondents were assured that there were no right or wrong answers. The highest ranked five words that were unique to each occupation were then used as attributes for the occupation (e.g., elementary school teacher: patience, creativity, children, caring, and storytime).

Five pairs of gender-typed names were also chosen. These names were among the most common 20 names (ten per gender) given to babies in the 1980s (Social Security Administration, n.d.). This decade was chosen because it was most likely to be the decade of birth for our participants. Where possible, names were chosen for similarity of origin and initial spelling (e.g., Michael—Michelle).

Explicit stereotypes

Participants made explicit ratings of the three target occupations (engineer, accountant, and elementary school teacher) using a 7-point Likert format scale (1 = masculine, 4 = neutral, 7 = feminine). This was the same scale used earlier by White et al. (1989).


Participants made their responses on an IBM-compatible (Pentium III processor) desktop computer running Inquisit software (Version 1.33) in a Windows 98 environment (Draine, 2003). To avoid having the explicit stereotype scales serve as primes for implicit responses, the program always presented the IAT first, followed by explicit stereotype scales. Participants were told to respond as quickly as they could to the IAT while making as few errors as possible.


An experimenter greeted participants and explained that the study examined associations between words and occupations. Participants learned that all responses would be made on a desktop computer. After giving their informed consent to participate, participants followed directions shown on the computer screen. The program first assessed implicit stereotypes, followed by explicit stereotypes, and demographic information.

The first step in creating the IAT scores involved having the participants discriminate between two occupation targets (e.g., engineer—elementary school teacher) and between the concepts associated with them. In one version, “Engineer” appeared on the left side of the computer screen and “Elementary School Teacher” appeared on the right. Centered below the two targets was a randomly selected concept that pilot testing had shown was associated with one of the targets. The student’s task was to press as rapidly as possible either the left (f) key if the concept was associated with the left appearing target or the right (j) key if the concept was associated with the right appearing target. For example, if “Blueprint” appeared, the correct response would be the left (f) key because it is associated with the target “Engineer.” Following the student’s correct response the next of ten trials commenced. If the response was in error, the word “Error” flashed on the screen for 400 ms after which the next trial began.

In a similar manner, participants discriminated between the attribute male or female. In one version, “Male” appeared on the left side of the screen and “Female” appeared on the right with one of ten randomly chosen names centered below. Assuming the first name was “Matthew,” the correct response would be the left (f) key. As in step one, correct responses were followed by the next trial; incorrect responses received an error message for 400 ms, followed by the next trial. The third step combined steps one and two such that a response key was shared. In the current example, either the phrase “Engineer or Male” appeared on the left and “Elementary School Teacher or Female” appeared on the right. Words from the previous two lists of concepts and attributes appeared centered below, but combined in random order for a total of 20 trials. Assuming that the word “Amanda” had been randomly chosen, a correct response would require pressing the right (j) key. So that error messages would not interfere with responses, none were presented in this step; the next trial always followed the participant’s response.

The fourth step reversed the positional association for male and female. On this step and in this example, the word “Female” appeared on the left of the screen and “Male” on the right for a total of ten trials. In contrast to step two, a correct response to a masculine name such as “John” would require pressing the right (j) key. Correct responses and errors were treated as in step one. The fifth step was similar to step three, yet included the target with the reversed attributes (e.g., engineer or female on the left). Once again, the individual concepts and attributes appeared in random sequence for a total of 20 trials. As with step three, no error messages were presented.

After participants had completed the IAT, they completed the Likert rating scales for the three target occupations: engineer, accountant, and elementary school teacher. Each occupation appeared individually on the screen. Participants indicated their ratings for each occupation by clicking the appropriate scale points with the computer mouse. Participants’ final tasks were to enter their sex, ethnicity, class standing, and age when prompted by the program. The last screen of the program contained a statement of our appreciation for their help. The experimenter debriefed the participants, thanked them, and gave them course credit for research participation.

Stereotypes are inferred from relative response speeds to the IAT’s tasks. The quicker responses that are anticipated to step three’s stereotypically congruent engineer–male pairs (and elementary school teacher–female pairs) than to step five’s stereotypically incongruent engineer–female pairs (and elementary school teacher–male pair) would imply that engineer–male is more strongly associated and readily retrieved than engineer–female. Responses to word pairs that are not congruent with existing associations require more time and cognitive effort than pairs that “fit” existing associations.

The IAT procedure we used resulted in 20 trials for the combined tasks. Although many researchers have used larger numbers of trials (e.g., 40 trials), our decision to do so was occasioned by the requirement that unique words generated by our pilot study participants be used for each occupation. Pre-testing had showed overlap across occupations when ten words were requested; hence, we requested five words from our participants. Greenwald et al. (1998) noted that IAT magnitudes were unchanged when as few as five exemplars were used per category. Nosek, Greenwald, and Banaji (2005) more recently have reported that IAT effects varied little with eight, four, or even two exemplars per category.

Design and Hypotheses

All participants responded to two pairs of target occupations and gender attributes. In the case of the preceding example, this would be: engineer + male & elementary school teacher + female, followed by engineer + female & elementary school teacher + male. Here, participants were first presented with an occupation and gender pair that was congruent with gender stereotypes, followed by an incongruent pair. Gender stereotype congruency order defined the first between subjects variable (i.e., congruent first, incongruent second or incongruent first, congruent second). The three pairs of target occupations served as the second between subjects variable (i.e., engineer–elementary school teacher, engineer–accountant, and accountant–elementary school teacher). No differences were expected for gender stereotype congruency order. In contrast, the occupation pair having the greatest differences in gender stereotypes (i.e., engineer–elementary school teacher, the first and fifth quintile) was expected to have a larger IAT value than the other two pairs which fall much closer on the occupation stereotype continuum (i.e., engineer–accountant, first and third quintile; accountant–elementary school teacher, third and fifth quintile).


The improved scoring algorithm recommended by Greenwald, Nosek, and Banaji (2003) was used to calculate D for each participant’s IAT responses. D is similar to Cohen’s (1992) effect size, d, in that the differences between IAT test steps or blocks are standardized by their pooled standard deviation. All responses in the two test blocks were considered for these calculations. Trials with latencies greater than 10,000 ms and participants with more than 10% of responses 300 ms or less were eliminated. Block means of the remaining trial response latencies and standard deviations for the pooled test block latencies were calculated. These means, plus 600 ms, replaced error latencies. Differences between block means, with error replacement, were then divided by the pooled standard deviation, without error replacement.

The resulting D values are reported in Table 1. These data are grouped by three target occupation comparisons (e.g., engineer vs. accountant). Each target occupation is further defined by the gender presentation order of the job target (e.g., male engineer vs. female accountant contrasted with female engineer vs. male accountant). The influence of these variables (target occupation pairs, gender stereotype congruency presentation order) was examined in a two-way ANOVA with D serving as the dependent variable. The main effect for target occupations was significant, F (2, 150) = 8.552, p<.001. As anticipated, the engineer–accountant IAT comparison was significantly smaller (M = 0.226, SD = 0.477) than the other two comparisons based on Tukey’s HSD post hoc test. The engineer–elementary school teacher (M = 0.602, SD = 0.422) and accountant–elementary school teacher comparisons (M = 0.494, SD = 0.526) did not differ from one another. There was no main effect for gender stereotype congruency presentation order nor was there a significant interaction effect.
Table 1

IAT D values for job and occupant gender comparisons.


IAT D values

Stereotyped job and occupant gender comparisons



Traditionally masculine job vs. traditionally feminine job

Male engineer & female elementary school teacher before



Female engineer & male elementary school teacher

Female engineer & male elementary school teacher before



Male engineer & female elementary school teacher




Traditionally masculine job vs. neutral job

Male engineer & female accountant before



Female engineer & male accountant

Female engineer & male accountant before



Male engineer & female accountant




Neutral job vs. traditionally feminine job

Male accountant & female elementary school teacher before



Female accountant & male elementary school teacher

Female accountant & male elementary school teacher before



Male accountant & female elementary school teacher




Note. Scores reflect latency differences between the two sequential comparisons; see text for details on calculation. Higher scores imply greater stereotype differences. IAT = Implicit Association Test. Each condition had 26 participants.

Explicit Stereotypes

Explicit ratings for the three target occupations manifest stereotyped perceptions and are shown in Table 2. On the 7-point rating scale (1 = masculine, 4 = neutral, 7 = feminine), mean ratings for engineer were the most masculine (2.3), accountant was rated as nearly neutral (3.6), and elementary school teacher was rated as the most feminine (5.6). It is possible to place these ratings in an historical context by using means and standard deviations from two other studies. Ratings were compared using z-tests and are also shown in Table 2. Shinar’s (1975) participants rated engineer as 1.9 on the 7-point scale, while participants of both White et al. (1989) and those from the current study rated it as approximately 2.3. Accountant was rated as 2.5 in 1975, 3.4 in 1989, and 3.6 in 2003. Each of these means is significantly different from each other. Ratings from the three studies for elementary school teacher do not statistically differ (5.6, 5.5, and 5.6, respectively).
Table 2

Comparison of explicit occupation sex stereotype ratings for target occupations.


Source studies

Shinar (1975)

White et al. (1989)

Current study















Accountant (CPA)







Elementary school teacher







Note. Current data were collected in 2003. Shinar’s (1975) title for an occupation is in parentheses, if it differed from White et al. (1989) or the current study. Rating scale: 1 = masculine, 4 = neutral, 7 = feminine. Means not sharing the same superscript letter (x, y, or z) are significantly different, z, p < .05.

an = 177, bn = 176, cn = 175, dn = 174, en = 156, fn = 120. To avoid confusion with mean differences, these superscripts are found with standard deviations.

Implicit and Explicit Measures

Correlations among implicit and explicit measures are shown in Table 3. All of these correlations are based on difference scores. In the case of the implicit measures, these are the scores originally shown in Table 1. A positive value reflects a preference for the gender stereotypic comparison pair, e.g., male engineer and female elementary school teacher. Explicit scores reflect the absolute value of the difference between each of the three pairs on the masculinity–femininity scale. A higher score implies greater gender stereotyping for two occupations.
Table 3

Correlations between explicit and implicit stereotype measures.







 Explicit likert scale Difference scores


 1. Engineer–elementary school teachera



 2. Engineer–accountanta




 3. Accountant–elementary school teachera




 Implicit IAT scores

 Male engineer & female elementary school teacher vs. female engineer & male elementary school teacherb




 Male engineer & female accountant vs. female engineer & male accountantb




 Male accountant & female elementary school teacher vs. female accountant & male elementary school teacherb




aExplicit n = 156. bImplicit n = 52 for each of the three separate participant groups.

**p < .01.

Correlations among explicit scores indicate that participants who stereotyped engineers and elementary school teachers also stereotyped accountants and elementary school teachers, 0.76, p < .01. There was a similar positive correlation between scores on the engineer–elementary school teacher and the engineer–accountant comparisons, 0.46, p < .01. In contrast, stereotyping scores on the engineer–accountant comparison were inversely associated with stereotyping scores on the accountant–elementary school elementary school teacher comparison, −0.16, p < .01. As will be recalled, implicit scores for different occupation comparisons were drawn from different participant groups. Correlations across the different IAT comparisons were accordingly not possible. Nonetheless, Table 3 shows the correlations between each participant group’s implicit occupation comparison and the three explicit occupation pairs. Of particular interest are the three correlations between the IAT and explicit measures of the same occupation pairs. The correlation between the IAT engineer–elementary school teacher gender stereotype and the corresponding explicit comparison was 0.16, ns. The IAT and explicit engineer–accountant correlation was 0.28, p < .01. Finally, the IAT and explicit accountant–elementary school teacher correlation was 0.07, ns. None of the other comparisons were significant.


The occupation of accounting presents an interesting example of how assessment of implicit processes may add to understanding occupational gender stereotypes. Explicit ratings for this occupation have shown it to be increasingly and consistently rated as a “neutral” occupation (Beggs & Doolittle, 1993; White et al., 1989). Further, the number of women who are now accountants exceeds those who are men. Given these ratings and the high percentage of women accountants, one might assume that the 1970s stereotype of the male accountant would be gone. Implicit stereotype results suggest that this is not the case. Although the stereotype is not as pronounced as that for engineer, the smallest IAT effect occurred between engineers and accountants. As noted earlier, the smaller the effect, the smaller the stereotype difference that is implied.

In contrast, the largest IAT effect is in keeping with the explicit stereotype results: respondents were able to identify male engineers and female elementary school teachers more quickly than when the job occupants were of the other sex. These two occupations thus appear to be strongly gender stereotyped. Further, the nature of these stereotypes is in keeping with the explicit ratings. Engineering is stereotyped as a masculine occupation and elementary school teaching is stereotyped as a feminine occupation. The next largest IAT effect involved accountants and elementary school teachers. Responses to male accountants and female elementary school teachers were much faster than when the other sex was paired with the occupations. As will be recalled, these two effects were not significantly different from one another, but are different from the engineer vs. accountant effect. This suggests that engineers and accountants are more similar than are accountants and elementary school teachers.

IAT results are similar to explicit ratings in one sense: accountants fall between engineers and elementary school teachers in degree of gender stereotyping. They differ, though, in that the implicit results indicate that accountants and engineers are more similar than explicit measures would suggest. Accounting is implicitly perceived to be a masculine job. This is perhaps partly due to popular portrayals of accountants as men and the perceptual association of accounting with mathematics, an area stereotypically associated with men.

At least as important in explaining the divergence in implicit and explicit stereotypes of accountants are the numbers of women who now practice accounting contrasted with earlier times. When Shinar published her study in 1975 only 25% of accountants were women (Beggs & Doolittle, 1993), but now 59% of those engaged in accounting or auditing are women (U.S. Department of Labor, 2004). The prevalence of women in accounting is a relatively new phenomenon. The continuing implicit masculine stereotype of accounting may be, in part, due to perceptions associated with the earlier high numbers of male accountants (cf. Eagly & Steffen, 1984). This may occur in a manner suggested by Betsch, Plessner, Schwieren, and Gütig (2001). In their study, cognitively busy participants were presented with fluctuating values of corporate stock. Even though participants found it difficult to recall explicit information about the companies, they had formed implicit evaluative attitudes as a consequence of their exposure to the stock values. Stock values appear to have been stored as summed information. To the degree that stereotypes in general and implicit stereotypes in particular represent cognitive effort saving strategies for busy perceivers, new information may be incorporated into implicit stereotypes in the summative fashion suggested by their model. Only when enough new information has been added will the “account values” of implicit stereotypes change sufficiently for change in the stereotype to occur.

Although several studies have shown good reliability and validity for the IAT (Banse, Seise, & Zerbes, 2001; Greenwald & Nosek, 2001), correlations with demonstrably reliable and valid explicit measures have been only modest (Brunel, Tietje, & Greenwald, 2004; Cunningham, Preacher, & Banaji, 2001; Kawakami & Dovidio, 2001). Our results are similar. Why the two types of measures do not yield more similar results has been a matter of theoretical and research interest with several explanations having been proposed. One proposal is that implicit measures reflect spontaneous stereotype processes (e.g., nonverbal behavior), whereas explicit measures manifest deliberative processes associated with more overt, potentially prejudicial behavior (Dovidio, Kawakami, Johnson, Johnson, & Howard, 1997). Others include inaccuracy in explicit self-reports due to impression management considerations or due to errors of introspection (Brunel et al., 2004). Another point of view is that implicit responses represent affective responses and explicit responses include a larger cognitive component (Hofmann, Gawronski, Gschwendner, Le, & Schmitt, 2005).

Implicit occupation sex stereotypes may thus provide information about what a person feels about someone performing a job versus what they think or what they are willing to express. As an example of this, consider Yoder and Schleicher’s (1996) study where undergraduates socially distanced themselves and pejoratively evaluated nontraditionally employed women even though these same respondents did not express overt or explicit bias. As Greenwald and Nosek (2001) have argued, under some circumstances implicit measures may predict behavior better than explicit approaches. Implicit approaches may be especially useful in understanding why someone who expresses gender-neutral explicit stereotypes may still respond more positively to a person of one sex or the other in a particular job. A persistent implicit stereotype may account for some of this behavior.

Interpretation of this first application of the IAT to occupational gender stereotypes should be tempered by the nature of the participants and the occupation sample. Although the respondents were very similar to Shinar’s (1975) sample, non-college groups (e.g., older executives) might evidence different patterns of stereotyping. Secondly, a drawback of the IAT is that it is relatively time consuming for respondents in comparison to an explicit rating scale. This influenced our decision to examine only three occupations. Future research may wish to use a broader range of occupations, perhaps focusing on those jobs where changes in gender composition are rapidly occurring.


Rachel Blalock, Beau Isley, Michiko Iwasaki, and Vance Jackson collected data by serving as experimenters for this study. We appreciate their generous help.

Copyright information

© Springer Science+Business Media, LLC 2006

Authors and Affiliations

  1. 1.Department of Counseling PsychologyBall State UniversityMuncieUSA
  2. 2.Department of AccountingBall State UniversityMuncieUSA

Personalised recommendations