Gender and sex concepts 2023 Census methodology consultation – feedback summarises the feedback we received about the Gender and sex concepts 2023 Census methodology consultation.
Thank you to submitters who provided feedback. This summary of feedback is accompanied by a report on the final methodology for gender and sex at birth concepts.
Methodologies for filling gaps in gender and sex at birth concepts for the 2023 Census sets out how gaps in census responses will be managed for the 2023 Census.
Contents
Consultation process
Feedback received on questions
Submission content out of scope of consultation
Summary of methodology decisions
Consultation process
The 2023 Census introduced questions on gender, sexual identity, and variations on sex characteristics for the first time, and a change was made to how information was collected on sex. Consequently, new methodologies were required to fill in gaps for these variables.
Between December 2022 and January 2023, we asked for public feedback on our planned approach for using alternative data sources for the gender and sex concepts, to inform our methodology decisions. Submitters could complete an online survey, email a written submission, or attend a virtual technical user forum held on 15 December 2022.
We asked for public feedback on:
- the order and use of alternative data to fill missing information for gender and sex at birth
- whether we should impute for missing responses to the sexual identity and variations of sex characteristics questions, given the apparent lack of alternate data sources that we could use to fill missing responses
- potential admin data sources, especially those that clearly distinguish between the gender and sex at birth concepts, and those that could be a substitute for missing sexual identity and variations of sex characteristics responses.
2023 Census: Methodology for filling gaps for gender and sex concepts – Consultation technical paper: December 2022 provides more information.
Findings
We received and analysed 44 submissions comprising:
- 28 online form submissions
- 3 emails from organisations
- 9 emails from individuals
- oral feedback from 4 individuals who attended the online forum.
Although we present some numerical totals for questions, it should be noted that submissions may not be representative of the population. We asked respondents to put alternative data sources in order. We added up a score that gives higher weight to sources earlier in respondents’ lists, which indicated the more trusted or accepted sources. These scores and tallies of responses give a broad sense of patterns across submissions.
Around half of the submissions commented on issues that were out of scope of the consultation, including:
- having to fill out one of either gender or sex at birth on the online form (rather than gender, sex at birth, or both being compulsory)
- gender being the primary demographic variable that we will produce census information on, replacing the previous use of the sex variable (in line with the data standard). Although gender will be the default output variable, data on sex at birth will still be available in outputs.
Feedback received on questions
The feedback received informed final decisions about methodologies for the 2023 Census. The feedback was based on the methodology paper that reflected our thinking at that time. Our subsequent investigations have resulted in some modifications, most notably introducing an option for a combination of direct alternative data sources and statistical imputation, that were not discussed in the December 2022 paper.
The rest of this section summarises the responses received on key questions. We asked different questions about the gender and sex at birth questions.
Gender
Question: For gender, what is your preferred order for using the three methods – historical (2013 and 2018) census responses, admin data, and statistical imputation?
There were 26 responses that gave some preferences, while 18 responses did not address the question.
We weighted the 26 multiple responses (3 for first choice, 2 for second choice, 1 for third choice). Historical census responses had the highest weighted score, followed by admin data, with statistical imputation ranking third.
Two responses recommended leaving gaps in the data, which is not an option because of the decision already made to provide a complete dataset for gender.
Two responses did not want to use imputation.
Two responses recommended using 2023 Census sex at birth responses first, and one mentioned using gender from the 2023 Census dwelling form / household summary form.
One response was in favour of statistical imputation as the primary way of filling gaps because it results in similar data quality for people who are gender diverse, transgender, and cisgender, instead of “concentrating the vast majority of the error into a group that is already vulnerable and under-represented”. They did not want to use historical census responses and admin data, noting that individuals in those data sets were forced into a binary selection and therefore data will undercount non-cisgender populations.
As general principles for selecting admin data, submissions suggested favouring admin data sources with non-binary categories or prioritising admin sources that are most likely to be up to date.
Question: If we use historical census data for gender, should we use both 2013 and 2018 Census?
There were 22 usable responses to this question, with 17 responses indicating that historical data could be used, and four of these preferring the use of 2018 Census data first.
Five responses did not want to use historical data. Reasons given included:
- inconsistencies over time due to public campaigns of how to respond to the question or due to differing Stats NZ instructions on how to represent a divergence from expected sex or gender
- historical responses may not reflect a person’s current gender, because they have changed how they identify or because a historic form was completed by a parent or other household member who might not know or acknowledge a person’s gender
- gender and sex are different concepts that should be kept separate; the historical question was not explicit that it referred to sex, though many people probably interpreted it to be sex.
Sex at birth
The online questionnaire asked different questions for sex at birth than the questions in the consultation document. We have included a summary based on the questions posed in the online questionnaire.
Question: Should we be using historical census data for sex at birth?
We received 28 relevant responses to this question. Twenty responses said yes, historical data should be used.
One respondent said, “The use of historical census responses for sex at birth is as accurate as it is possible to obtain. It is likely that the small numbers of people who have changed their gender answered the sex question according to their gender identity in the 2013 and 2018 Census”.
Some thought that the historical questions would be okay for cisgender people but would not be accurate for trans people.
One respondent noted, “I’ve put gender on my census form since the ’90s – you won’t have my sex at birth for a way way time back”.
Question: If, after using admin data and/or historical census responses for sex at birth we still have gaps, should we fill these in using statistical imputation or leave gaps in the data?
For 15 respondents, the preference was to leave gaps, while nine respondents preferred imputation.
Twenty responses did not identifiably address the question.
Question: For sex at birth, what is your preferred order for using the selected methods?
We received 18 suggestions for ordering the methods. For the four methods – historical census, imputation, admin data, and leaving gaps – we weighted the multiple responses (4 for first choice, 3 for second choice, 2 for third choice, and 1 for fourth choice).
There was a clear preference for historical census data, then imputation, then admin data, and leaving gaps was the least preferred option.
Sexual identity and variation of sex characteristics questions
Question: Should we attempt to fill in missing sexual identity information using statistical imputation?
The technical paper noted that we do not have complete and accurate historical census data or admin data sources to fill the gaps or to understand the distribution of this variable and that imputation is likely to be of low quality. Respondents tended to agree, and only two responses recommended imputation. There were 21 responses that recommended leaving gaps. One person said that imputation is unlikely to be useful. The other 19 responses did not directly address the question.
One submission suggested doing the imputation and having that data available with census microdata (in the Integrated Data Infrastructure (IDI)), so researchers can choose whether or not to use it. Other responses warned that the stigma in reporting a non-heterosexual identity is likely to lead to non-response bias, undercounting sexual minority populations in a way that imputation may not fix. In the future there may be more alternative data sources that could help test the results of an imputation model.
Question: Should we attempt to fill in missing variations of sex characteristics information using statistical imputation?
As with the sexual identity question, we do not have alternative data sources so the only option would be to use imputation to address gaps in census information. Responses were similar to the sexual identity question, with most people recommending leaving gaps and only two people recommending imputation.
Feedback on alternative data sources
Question: Are there any admin data sources that clearly distinguish between the sex at birth and gender concepts?
Respondents who mentioned the quality of data sources think the quality of gender data collected by various government sources is improving over time, although systems may not accurately capture gender or transgender status. Three specific admin data sources were identified for gender.
- Ministry of Social Development – Information gathered for benefit purposes allows people to identify their gender.
- Ministry of Education – Although people can change details of their records, including gender, it has sometimes changed back to birth certificate data.
- National Health Index (NHI) – Several respondents who worked with health data suggested that the quality of this data may be inconsistent, but the data may have up-to-date gender information for many people.
For the sex at birth concept, one submission called birth records (from the Department of Internal Affairs) “very well aligned on the concept”.
Question: Are there any admin data sources that hold information on sexual identity and variations of sex characteristics?
Several responses suggest that health data may be useful for variations of sex characteristics. This information is not currently available in the IDI for use in the combined census.
Submission content out of scope of the consultation
This section provides some examples of feedback received that relates to decisions that have already been made or issues that are related but not within scope of the consultation. Many comments relate to the gender and sex at birth questions, or revisit issues discussed in the data standard.
Data standard for gender, sex, and variations of sex characteristics provides more information.
- There are sensitivities surrounding providing sex and gender information.
- There is a lack of general awareness of the difference between the word sex and the word gender.
- There was concern that questions deliberately left blank would subsequently be filled in by Stats NZ.
Note: For most variables in the 2023 Census, we intend to fill gaps to make variables complete. This will be the case for gender and sex at birth, but not for sexual identity, variations in sex characteristics, or the derived variables cisgender and transgender status and Rainbow LGBTIQ+ indicator.
- There was concern that Stats NZ has no control over the quality of admin data, and the confusion about how variables related to gender and sex have been collected particularly raises quality concerns.
Note: Stats NZ considers these factors in assessing whether admin data is usable.
- There was some concern about social license or ethical license to use admin data for this purpose.
- There was some scepticism of statistical imputation, calling it “guessing” or “making up” data. Others were concerned that actual census responses may not be representative of the true population so Stats NZ needs to be very careful with imputation as it could “likely propagate bias”.
- Many respondents mentioned the need to be transparent about using statistical imputation.
Note: For all variables output in the 2023 Census, Stats NZ will provide information on the breakdown of data sources used, and for customised data requested it will be possible to have flags and filter on the different data sources used for variables.
- Respondents care about data stewardship, including how variables are presented and accessed. For instance, one requests “strong data governance and privacy safeguards” for access to the sex and gender data.
Summary of methodology decisions
Stats NZ will use the following methods to fill missingness in the gender and sex at birth variables for the 2023 Census.
Sexual identity and variations of sex characteristics: Gaps for these variables will not be filled by alternative data sources.
Gender: If a census response does not include a usable value for this variable from the 2023 Census individual form, then the gender variable will, in order, be sourced from:
- the 2023 Census dwelling or household summary form
- from admin data sources, where the admin data and Stats NZ concepts align
- statistical imputation.
Sex at birth: If a census response does not include a usable value for this variable from the 2023 Census individual form, then the sex at birth variable will, in order, be sourced from:
- admin data sources where the admin data and Stats NZ concepts align
- historical census responses for sex
- wider admin data sources with different concepts of sex or gender
- statistical imputation.
While the statistical imputation of one variable will not be dependent on the other, gender and sex at birth are imputed simultaneously. If both need imputing, they will be imputed from the same record.
Decisions about the methods employed for gender and sex at birth variables have been informed by:
- internal Stats NZ technical review
- consultation with key stakeholders and experts in data about Rainbow communities
- public consultation process on methodologies for gender and sex concepts (this consultation)
- discussion with the Data Ethics Advisory Group.
This feedback will also support the appropriate presentation of the outputs from the 2023 Census and clear explanation of the variables for data users.
For more information
Email [email protected]
