Projection model

To produce projections of the population by sexual identity, we draw upon new advances in demographic modelling (Wilson et al., Reference Wilson, Temple and Lyons2021). Specifically, we develop a modified multi-state cohort-component projection model in which the adult population, aged 18 years and above, is divided into sexual identity categories, and the childhood population, aged 0–17, is projected separately and is not categorised by sexual identity. The model was designed according to the principles of movement population accounts (Rees, Reference Rees1984) and is based around a set of population accounting equations. The accounting equation for the childhood population can be expressed as:

(1)$$P_{s, a + 1}^{} ( {t + 1} ) = P_{s, a}^{} ( t ) -D_{s, a\to a + 1}^{} + I_{s, a\to a + 1}^{} -E_{s, a\to a + 1}^{} $$

while for the adult population the equivalent is:

(2)$$\eqalign{P_{s, a + 1}^i ( {t + 1} ) = & \,\,P_{s, a}^i ( t ) -D_{s, a\to a + 1}^i + I_{s, a\to a + 1}^i -E_{s, a\to a + 1}^i \cr & \quad + \mathop \sum \limits_{\,j, j\ne i} C_{s, a\to a + 1}^{\,ji} -\mathop \sum \limits_{\,j, j\ne i} C_{s, a\to a + 1}^{ij} } $$

where P = population; t = time; a = age; a → a + 1 = the change in a cohort's age from a to a + 1 between times t and t + 1; s = gender; i, j = sexual identities; D = deaths; I = immigration; E = emigration; and C = identification change between sexual identity categories.

The time reference for demographic components of change in the equations is the projection interval t to t + 1 but labels are omitted for clarity.

Immigration is projected directly as flows because it is less influenced by population sizes and more by policy settings. Deaths, emigration and identification change in both equations are projected by multiplying an occurrence/exposure rate by the population at risk. For example, emigration in Equation 2 is projected as:

(3)$$E_{s, a\to a + 1}^i = e_{s, a\to a + 1}^i \displaystyle{1 \over 2}( {P_{s, a}^i ( t ) + P_{s, a + 1}^i ( {t + 1} ) } ) $$

where e = emigration rate.

The population-at-risk of emigration is calculated as the mean of the start-of-interval population at the end-of-interval population. The latter is permitted on the right-hand side of the equation because an iterative calculation scheme is employed whereby the end-of-interval population is updated in successive iterations until convergence is achieved. For the newly born cohort the population-at-risk is defined differently; it is approximated as half the end-of-interval population aged 0, following Willekens and Drewe (Reference Willekens, Drewe, Heide and Willekens1984).

When cohorts in the childhood ages reach the age of 18, they are assumed to adopt a particular sexual identity:

(4)$$P_{s, 18}^i ( {t + 1} ) = P_{s, 18}^{} ( {t + 1} ) \; p_{s, 18}^i $$

where p = proportion of the population with a specified sexual identity.

This simplification of the more complex reality of developing a sexual identity over many years is necessary due to data limitations.

Births are projected by multiplying ASFRs by the female population at risk in the standard way:

(5)$$B = \mathop \sum \limits_a \left({b_a^{} \displaystyle{1 \over 2}[ {P_{\,f, a}^{} ( t ) + P_{\,f, a}^{} ( {t + 1} ) } ] } \right)\;$$

where B = births; b = ASFR; and f = females.

Births are then divided by sex using the sex ratio at birth of 105.5 males per 100 females. Births are not projected separately by sexual identity of women because this output is not required for the purposes of this study; it also avoids the challenge of indirectly estimating fertility rates by sexual identity.

Data sources

The 2016 populations used as the ‘jump-off’ populations for the projections were prepared by disaggregating to single-year age groups the broad age group population estimates by sexual identity created earlier by Wilson et al. (Reference Wilson, Temple, Lyons and Shalley2020). These estimates were sourced using data from three data collections from two representative national household surveys: the General Social Survey (GSS), and Waves 12 and 16 of the Household, Income and Labour Dynamics in Australia (HILDA) Survey. These data are adjusted to be consistent with the official Estimated Resident Population (ERP) of Australia. For further information, see Wilson et al. (Reference Wilson, Temple, Lyons and Shalley2020). Data on the Total Fertility Rate (TFR) and age-specific fertility rates (ASFRs) were obtained from various issues of the Births Australia publication on the Australian Bureau of Statistics (ABS) website (ABS, 2021a). Mortality data for Australia were downloaded from Deaths Australia publications (ABS, 2021b) and the Human Mortality Database (HMD, 2021). A series of historical life tables were calculated using these deaths and population estimates from the HMD and ABS (HMD, 2021; ABS, 2021c). International migration data were obtained from the ABS publication Migration Australia (ABS, 2021d) as well as from customised tables of single year of age immigration and emigration data obtained directly from the ABS. Data on changes to reported sexual identity over time were obtained from the HILDA Survey waves of 2012 and 2016. HILDA is a large panel survey covering about 17,000 private household residents each year (Watson and Wooden, Reference Watson and Wooden2021). Since 2012 it has asked questions on sexual identity every four years.

Projection assumptions

The above model was used to produce projections of Australia's LGBQ population from 2016 to 2041. The formulation of projection assumptions presented a considerable challenge due to the limited coverage and quality of data on sexual identity populations. Just two sexual identity categories in the adult ages, LGBQ and heterosexual, were selected for these projections. The initial jump-off populations for 2016 were created by using the population estimates by gender and broad age group prepared by Wilson et al. (Reference Wilson, Temple, Lyons and Shalley2020) and disaggregating them to single-year age groups. The disaggregation process involved linear interpolation of the proportions of the population identifying as LGBQ from broad ages to single-year age groups, multiplying these proportions by single-year ERPs, and then making final adjustments to ensure single-year age LGBQ populations summed over age to the original broad age groups. For the childhood ages of 0–17, the jump-off populations consisted of ERPs prepared by the ABS (2022).

Fertility assumptions were prepared in terms of the TFR and the age profile of ASFRs. A long-run TFR of 1.65 was selected with a temporary decline over the first few years due to the impacts of COVID-19. The age profile of fertility was assumed to continue shifting gradually to older ages for the first decade of the projection horizon, and then remain constant.

Mortality assumptions were specified in terms of life expectancy at birth, with the same assumptions used for both sexual identity populations due to data limitations. Life expectancy was assumed to continue long-run improvements so that male life expectancy at birth would rise from 80.0 years in 2016–2017 (the first projection period from 1 July 2016 to 30 June 2017) to 86.3 years in 2040–2041, while female life expectancy would increase from 84.9 years in 2016–2017 to 88.8 years by 2040–2041.

Immigration and emigration projections were constrained to total net migration assumptions. Net international migration was assumed to be 225,000 per annum from 2025–2026 onwards, but with a big COVID-related drop in the short run. Due to the lack of migration data by sexual identity, immigration was divided between sexual identity populations according to population size, while the same set of national emigration rates was applied to both populations.

Two projection scenarios were created which differ only in their assumptions on identification change. Because of the limited data environment, we stress that both these scenarios are speculative and involve considerable amounts of judgement. A third scenario assuming zero identification change is also included for illustrative purposes.

For the no identification change scenario, we assumed that as cohorts turned age 18 the proportions adopting a sexual minority identity in the future would be the same as the sexual minority proportions for males and females aged 18–24 in the 2016 population estimates (Wilson et al., Reference Wilson, Temple, Lyons and Shalley2020). Older cohorts were assumed to retain their sexual minority proportions estimated for 2016. Specifically, each person is assumed to age through time with the same sexual identity they reported in 2016. In this scenario, the social and legal environment for sexual minority individuals does not progress any further despite the huge shift in social attitudes and introduction of legal protections associated with sexual orientation in Australia over the last few decades (Campbell et al., Reference Campbell, Perales and Baxter2021). Prejudice, discrimination and stigma – minority stress (Meyer, Reference Meyer2003) – remain constant into the future and result in no further increases in sexual orientation identification by cohort.

For the identification change scenario, we assumed that the proportion of the population assuming a sexual minority identity at age 18 would continue to increase, and that there would be changes to reported sexual identity at all adult ages. In this scenario, progress in social attitudes and legal protections continue, and people are more willing to describe themselves as having sexual minority identities. Younger cohorts experience their formative years and develop their sexual identities in more accepting social environments (Meyer et al., Reference Meyer, Wilson and O'Neill2021), and over time therefore a greater proportion feel comfortable adopting a sexual minority identity. This scenario assumes continuation of the trend observed in the United States of America where younger cohorts are much more likely to identify as LGBQ+ (Jones, Reference Jones2022). The proportion reporting a sexual minority identity at age 18 was assumed to gradually increase over time based on the trend observed between 2012 and 2016 up to a ceiling of 0.1. The limit of 0.1, although arbitrary, was imposed to prevent the proportions becoming very large by the end of the projection horizon due to the uncertainty of extrapolating a change measured over just a four-year period. We calculated identification change rates by age and gender based on reported change in sexual identity between 2012 and 2016 in the HILDA Survey. Due to the small sample size and jagged age schedules of rates, we applied smoothing to the rates using cubic splines (Wilson et al., Reference Wilson, Temple and Lyons2021). Because of the small sample size and uncertainty surrounding these identification change rates by age and gender, they were constrained to a total net identification change number during the calculation of the projections. This number acts as a plausibility constraint on the changes to and from the LGBQ population. The age- and sex-specific flows from the LGBQ population, and flows to it, are proportionally adjusted so that they match the total net identification change assumption. The net identification change assumption for the first year of the projections was derived from the annual average amount of change indicated by the HILDA Survey between 2012 and 2016. We assumed that net gains to the sexual minority population would fall gradually to zero over the next two decades as society becomes more accepting of the sexual diversity.

For illustrative purposes, we created an additional set of projections in which the proportions of the population identifying with a sexual minority identity by age and gender remain constant over time. We refer to this as the fixed propensity scenario. Specifically, the age–sex rates of identification remain constant implying that future cohorts of the older population are attributed with the sexual identification rates of the older population in 2016. Thus, changes to the sexual minority population occur from national population growth alone. This projection is included primarily to advise against using this simple approach, and to emphasise the importance of explicitly modelling the various demographic processes which affect population change (particularly identification change).

Sensitivity analyses

To illustrate the potential impact of identification change on the projections, we undertook a simple sensitivity analysis to demonstrate the effect of different rates of identification change to and from the sexual minority population. Annual age-invariant rates of heterosexual to sexual minority change from 0.0005 to 0.0050 were applied in several increments, and then equivalent rates of sexual minority to heterosexual change from 0.0125 to 0.1250 were applied, also in several increments. The maximum values of these rates approximate twice the annual average identification change rates apparent between 2012 and 2016 for the 40+ population in the HILDA Survey.

We also undertook a sensitivity analysis for mortality given the application of national mortality rates to the sexual minority population. We created projections with life expectancy at birth being varied for both males and females by up to five years either side of the national life expectancy assumption. The variation was applied throughout the projection horizon, and life expectancy was progressively increased and decreased in one-year increments. Life expectancy at five years above or below national life expectancy should be considered an extreme assumption, with a one- or two-year variation being more plausible.

Ethics approval for this project was granted by the Melbourne School of Population and Global Health Human Ethics Advisory Group.