A large body of literature has shown that gender is an act to be “done” or “performed” (Butler, 1988; West & Zimmerman, 1987), resulting in a gendered organization of society maintained by social interactions (Goffman, 1977; Ridgeway & Correll, 2004). Thus, the way that gender is expressed in these interactions, e.g., via femininity and masculinity, offers differential social power and status (Connell, 2020; Edwards, 2005; Hamilton et al., 2019). Despite this nuanced theoretical understanding of gender, measures of gender in empirical data have historically been coarse—often proxying the complexity of gender as a binary sex variable (male/female) (Ridgeway & Saperstein, 2024; Westbrook & Saperstein, 2015). This mismatch between theoretical understanding and empirical measurement has come to the fore as there have been changes to social meanings of gender. For instance, transgender, nonbinary, and otherwise noncisgender identification has become more prevalent (at least via disclosure on surveys) (Lagos, 2022); and, among cisgender individuals, there are some settings where women receive social advantages undertaking traditionally masculine roles and men undertaking traditionally feminine roles (e.g., Evans, 1997; Rosette & Tost, 2010). Despite these changes to the distribution and social meaning of gender, many censuses and surveys did not have the tools to capture gender more holistically (Bates et al., 2022; Pao et al., 2025).

Many studies have attempted to ameliorate this gap between measurement and construct by creating new measures of gender identity (Budnick et al., 2025); however, studies of gender expression have been less common (Magliozzi et al., 2016). Gender expression has been defined as a dimension of gender that is “how an individual signals their gender to others through behavior and appearance” (Bates et al., 2022, p. 4). This has often been operationalized as masculinity and femininity, wherein gender identities of “man” and “woman” often correlate with higher “masculinity” and “femininity,” respectively (Westbrook & Saperstein, 2015). To capture dimensions of masculinity and femininity, scholars have begun using gradational gender scales, which ask survey respondents to mark their self-rated masculinity and femininity, separately, on unipolar scales (Alexander et al., 2021; Cassino, 2020; Garbarski, 2023; Magliozzi et al., 2016). These scales have been taken up on population-level surveys like the U.S. General Social Survey and European Social Survey and targeted sub-population surveys like the U.S. Trans Survey.

At present, gradational gender scales have been most popular in the U.S. and Europe; however, this tool to study gender expression may be most useful in some of the places that currently do not have them—such as Japan. Japan is a particularly interesting case due to its persistent gender stratification, despite continued women’s workplace participation (Brinton & Oh, 2019; Holbrow, 2025; Nemoto, 2016), and many institutional conditions (such as educational systems) that hinder gender equality (Uchikoshi et al., 2025). Nonetheless, Japan is also a place of innovation for gender measurement—with recent studies adapting new measures (such as a three-step gender measure) to a Japanese local context (Hiramori & Kamano, 2020). Though there have been advances in measuring gender more holistically in Japan, to our knowledge, gradational gender scales have not been tested yet with a Japanese sample.

In the following, we translate gradational gender scales to Japanese and use them in two waves (October 2023 and February 2024) of an online longitudinal survey fielded by the Social Science Japan Data Archive (SSJDA) hosted at the Institute of Social Science, the University of Tokyo. We evaluate item nonresponse rates, response distributions, and within-respondent response (in)stability over time. These metrics highlight potential country-specific differences which pattern gradational gender scale responses and indicate the need for further research within Japan and other non-Western contexts when using these scales internationally.

The instrument that we used was based off Magliozzi, Saperstein, and Westbrook’s (2016) gradational gender scales. There are two sets of gradational scales proposed by Magliozzi et al.: the first are answered in response to the question, “In general, how do you see yourself?”. The second are answered in response to the question, “In general, how do most people see you?”. For both questions, there are two Likert scales, respectively labeled “Feminine” and “Masculine”, each containing 7-points (marked 0 to 6) with 0 being “not at all” and “very.” These scales have been validated and further tested in English with a U.S. sample (Garbarski, 2023) and have been translated internationally on the European Social Survey, Round 11 data. Other studies have previously used these scales to indicate differences by other sociodemographic characteristics or social outcomes, like health and political attitudes, showing the utility of these measures to understanding gender more holistically (Choi & Merlo, 2021; Hart et al., 2019; Pao, 2023; Pao & Roundy, 2025).

For international comparability, one of the authors, a native Japanese speaker, translated Magliozzi et al.’s instrument into Japanese—maintaining the scale points, question ordering, and format of the original English scales. As with U.S. and European precedents, all respondents, regardless of gender identity, would see both masculinity and femininity scales. Importantly, this translation of gradational gender scales was exploratory so diverged from some other known best practices, such as using double translation and team discussion processes (Behr & Shishido, 2016; Harkness, 2003). This translation used for gradational gender scales was the one that was approved by SSJDA for greater respondent clarity, but its slight divergences from standard gradational gender scales (namely, through the inclusion of the words “masculine” and “feminine” in the question prompt) may itself cause differences in response patterns. The translated version of the gradational gender scales can be seen in Figure 1. Note that these two questions were asked consecutively. Responses stemming from the first two scales are referred to as “self-rated femininity” and “self-rated masculinity,” since respondents are rating how they see themselves. Responses stemming from the latter two scales are called “reflected appraisal femininity” and “reflected appraisal masculinity,” since they are what respondents believe others see. Though we believe that these slight deviations in translation from the original scales increased clarity for native Japanese speakers, we encourage future studies that use different translation processes, and possibly other translations, to replicate our findings.

We draw on data from an online longitudinal survey conducted by the SSJDA at the Institute of Social Science, the University of Tokyo. These data and codebooks are available via application from the SSJDA website at the Institute of Social Science: https://ssjda.iss.u-tokyo.ac.jp/Direct/. All code used to analyze these data are available via the Open Science Foundation: https://osf.io/t7s98.

The Wave 1 of the SSJDA Panel survey was conducted in February and March 2021, targeting men and women aged 20 to 39 residing in Japan as of the end of December 2020. Participants were selected using a stratified two-stage random sampling method, where the country was first divided into 11 regions, and then each region was further stratified into five categories based on population size. The number of survey locations allocated to each stratum was proportional to its population size. After a total of 100 locations were randomly selected, 50 individuals from each location were randomly chosen based on the Basic Resident Register (Jūmin Kihon Daichō). Invitation letters were mailed to 5,000 respondents, and responses were collected via the web using an online survey service called Lime Survey. The number of valid responses to the SSJDA Panel Wave 1 was 1,329, resulting in a response rate of 26.6%. Respondents who completed the survey received a 500 Japanese yen gift card. In February 2022 when Wave 3 of the survey was fielded, additional survey participants were newly recruited. The target population consisted of men and women aged 21 to 40 residing in Japan as of the end of December 2021 to make their birth years equivalent to the Wave 1 participants. Using the same sampling method with an increased sample size (6,600), the number of valid respondents was 1,576, resulting in a response rate of 23.9%.

The authors received permission to field questions in the SSJDA Panel after applying through a call for proposals. The questions proposed by the authors were fielded in the SSJDA Panel in Waves 6 and 7, October 2023 and February 2024 respectively. Survey invitation letters were sent to 2,152 participants for the Wave 6 survey and to 2,112 participants for the Wave 7 survey after excluding individuals who had declined to participate in the follow-up survey. The number of valid responses was 1,528 and 1,603, respectively, resulting in response rates of 71.0% and 75.9%.

Two thirds of the respondents (1,024 respondents in Wave 6 and 1,091 respondents in Wave 7) were randomly assigned to the gradational gender scale questions. All respondents were given the option to skip questions. Respondents who saw the gradational gender scales were randomly assigned to see the measure at either the beginning or the end of the survey. The original purpose of this design was to examine the relationship between gradational scale measures and various life course outcomes (e.g., subjective health), and to test whether this relationship differs depending on whether the gradational gender questions are asked before or after the outcome questions. Because the allocation to the beginning and end conditions was random and balanced in both waves, we believe that the overall results—and the differences between waves 6 and 7—are not affected by our use of different survey arms. Out of the respondents who were asked the gradational gender responses, 992 respondents (1,046 respondents in Wave 7) remain after restricting the sample to those with complete information on key sociodemographic characteristics: gender,¹1

A small number of respondents selected “other” or did not provide a response to the gender question. Due to the limited sample size, this study focuses on respondents who identified as either men or women.

Broadly, the sample also looks similar between survey waves.⁴4

Of the 1,024 respondents asked to complete the gradational gender scale questions in Wave 6, 8.2% did not participate in the Wave 7 survey. Thus, while the attrition rate is not negligible, it remains relatively small.

The response rate on gradational gender scales from the overall sample was relatively low across the overall sample: 91.0% (92.7%) for self-rated femininity in Wave 6 (Wave 7); 86.6% (94.2%) for self-rated masculinity in Wave 6 (Wave 7); 90.4% (92.3%) for reflected appraisal femininity in Wave 6 (Wave 7); and 86.4% (93.8%) for reflected appraisal masculinity in Wave 6 (Wave 7). This is notable compared to the relatively higher response rates for other questions in the survey, such as labor force participation with a 99.1% response rate and parenthood status with a 99.8% response rate, both in Wave 6. This compares to nonresponse rates of more sensitive characteristics like annual income in the raw data (98.1%). Although the response rates for the gradational gender scales are not high, they are noticeably higher in Wave 7 than in Wave 6. This pattern may suggest that respondents become more comfortable with these items after repeated exposure and may better understand what the questions imply.

Item nonresponse was particularly high for the gender “atypical” scales—namely, femininity for men and masculinity for women. Table 2 shows item nonresponse for gradational gender scales by wave and by gender. Around 22 percent of the men (women) in the sample did not respond to the self-rated femininity (masculinity) in Wave 6, and around 23 percent of the men (women) in the sample did not respond to the reflected appraisal femininity (masculinity). In Wave 7, nonresponse decreased for men and women for both self-rated and reflected appraisal scales, but the decrease in nonresponse was steeper for women than for men. By Wave 7, 17 percent of men (10 percent of women) did not respond to their “gender-atypical” self-rated scale, and 18 percent of men (11 percent of women) did not respond to their “gender-atypical” reflected appraisal scale.

Notably, those who did and did not respond to the gradational gender scales varied by other sociodemographic characteristics. Table 3 uses Wave 6 data to show differences in responders and nonresponders. Men who responded to all gradational scales were 76.4% of the sample; men who responded to masculinity scales only were 21.1% of the sample; and other combinations (femininity only or nonresponse on both scales) were 2.5% of the sample. Women who responded to all gradational scales were 75.6% of the sample; women who responded to femininity scales only were 20.4% of the sample; and other combinations of scale nonresponse were 4.0% of the sample. Men (women) who responded to all gradational gender scales were substantially different than those who replied to only masculinity (femininity) scales.

Among both women and men, younger respondents are more likely to answer all scales, while older respondents are less likely to answer gender “atypical” scales. Similarly, college-educated men and women are more likely than non-graduates to answer all scales. Among men, non-employed respondents are more likely to answer all scales than their employed counterparts, whereas the opposite pattern is observed among women. This may reflect age differences, as non-employed men are more likely to be younger and still in school, while non-employed women may have left the workforce due to family-related events such as marriage or childbearing. Higher income is also associated with greater likelihood of responding to all scales, although this association is less pronounced than that for educational attainment. Finally, non-married respondents are more likely to answer all scales than married respondents. This pattern may be confounded by age.

Among responders, the distribution of femininity and masculinity is displayed in Figure 2. In the SSJDA sample the modal responses of gender-typical self-rated and reflected-appraisal scales are 3, the scale midpoint. For the gender-atypical scales, the distribution of responses for men and women are more dispersed—particularly for women’s description of their masculinity. Notably, this response distribution is very different from existing patterns seen in other country contexts. Appendix Table 1 displays the median masculinity and femininity response for men and women in the 2024 General Social Survey (GSS) in the U.S. and the 2023 European Social Survey (ESS) for countries which included the gradational gender scales. This table shows that across the U.S. and European contexts for adults in the same age range as the SSJDA (i.e., 20–39), the median response for self-rated gender-typical scales was generally a 5 or 6. Though the 2023 ESS did not include reflected appraisal measures, the 2024 GSS shows that U.S. respondents’ median gender-typical reflected appraisal was also a 6.

There are several methodological and theoretical reasons we may see this middle-of-the-scale response distribution in our Japanese sample. From a methodological standpoint, differences between the GSS and the SSJDA sample may be partially explained by the lack of follow-up question that alerted respondents of a missing response. The 2024 GSS used this follow-up prompt, which reduced gender-atypical nonresponse, whereas the SSJDA did not.⁵5

GSS methodological statements say respondents who attempted to skip the question were provided a “gentle ‘nudge’” to encourage responses. For more information, see the codebook here: https://gss.norc.org/content/dam/gss/get-documentation/pdf/codebook/GSS%202024%20Codebook%20R2.pdf

We then evaluated response stability within respondents who replied to scales in both waves and differences in nonresponse between waves. Figure 3 shows the distribution of the difference in Wave 6 and Wave 7 responses among respondents who replied to the same scale in both waves. Figure 3 shows that the distributions are relatively normal, centered around 0—indicating no change in identification between waves. Despite the modal response being 0 for all combinations of masculinity and femininity changes for men and women, Figure 3 nonetheless still shows a degree of fluidity for how the same respondents may approach these gradational gender scales across time.

To show this fluidity more clearly, Table 4 shows nonresponse rates for men and women separately within each wave. We find that there is substantial variation within respondents over time. The majority of men who responded with a 0 for their self-rated and reflected appraisal femininity in Wave 6 would give the same response in Wave 7. However, this is not true of women who initially gave a 0 on their gender-atypical scales: while the plurality of women gave a 0 again in Wave 7, we see more variation across response categories, which may reflect changes either substantively in identification over time or a methodological artefact. For instance, response instability may indicate that women may report (and experience) higher degrees of fluidity, instability, or ambivalence in gender expression. This may also (or instead) indicate ambiguity in the measure itself, due to (for instance) the vagueness of the response scale where only the end points at 0 and 6 are labelled. Nonetheless, many men and women who provided middle-of-the-scale responses in one wave would also report a middle-of-the-scale response in another wave—indicating that the scales may be capturing variability over time, but may have some degree of central tendency. These are novel findings, and we hope future studies may be able to replicate these findings in other Japanese samples and add more knowledge to what may drive changes over time.

In this study, we perform a preliminary evaluation of response distributions and nonresponse patterns for gradational gender scales using a young adult Japanese sample. Overall, our study shows how gradational gender measures may appear differently in a different, non-Western context. Using a sample of adults in a longitudinal study (SSJDA), we show that there is a relatively high degree of nonresponse to gradational gender scales, and this may be particularly elevated when respondents are first exposed to the scales. Among those that do respond to the gradational scales, the majority respond to both gender-typical and gender-atypical masculinity and femininity scales, followed by respondents who just respond to their gender-typical (i.e., masculinity for men, femininity for women) scales.

Responders and nonresponders varied on a variety of sociodemographic characteristics—indicating differential nonresponse for both men and women. Younger respondents, college-educated respondents, and higher income respondents are more likely to answer all scales. The response distribution for responders indicates a modal respond of 3 (the gradational scale midpoint) for gender-typical scales. The responses for gender-atypical scales tended to be more uniformly dispersed among responses at the lower-end of the scale among responders. This is different from other samples, like those in the U.S., which tend to have more extreme responses (e.g., 0s and 6s, the minimum and maximum of the scales).

Finally, due to the panel nature of the data, we could evaluate response stability over time (i.e., between waves). Though most respondents tended to respond identically on each scale (i.e., a change of 0 from wave to wave), the distributions appear quasi-normal, indicating a degree of fluctuation for the same respondents at different points in time. This is a novel intervention, since many surveys using gradational gender scales only collect cross-sectional data. Our data allows us to evaluate within-respondent variation, pointing to the need to study the cause of variability: asking gradational gender scales at multiple points in time for the same respondent may, on the one hand, reveal substantive changes in gender expression, or, on the other hand, indicate response instability due to methodological challenges in the measure itself. In sum, we hope that this translation of gender scales to a different population, which experiences a different gendered context, shows the need for further research on the use of gradational gender scales in Japan and other non-Western contexts to evaluate their comparability internationally.