Research Article |
|
Corresponding author: Aleksandra Ya. Burdyak ( burdyak-ay@ranepa.ru ) © 2026 Aleksandra Ya. Burdyak.
This is an open access article distributed under the terms of the Creative Commons Attribution License (CC BY 4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Citation:
Burdyak AYa (2026) Gender Differences in Housing Provision in Russia: An Age–Period–Cohort Decomposition, 2003–2023. Population and Economics 10(3): 82-104. https://doi.org/10.3897/popecon.10.e199193
|
The well‑being of women and men within households is a central concern of gender research in demography and economics, yet it has rarely been examined in relation to housing in Russia using large‑scale microdata. Over 2000–2024, Russia’s housing provision increased by half against a backdrop of household nuclearisation, a female longevity advantage, and uneven generational access to the post‑Soviet housing stock. Whether this expansion reflects macro‑level growth or generational replacement – and whether gender differentially shapes the housing trajectory – remains an open question.
We use Rosstat’s Household Budget Survey microdata for 2003–2023 (2.61 million observations) and apply an age–period–cohort (APC) decomposition with the Intrinsic Estimator (IE). The pseudo-panel comprises 354 cells across 19 five-year cohorts (1930–2024), 18 age groups (0–4, 5–9, …, 85+) and 21 survey years. The APC decomposition is estimated separately for the combined, male and female samples.
Under the IE identification, age is the largest component of the variance of the per-capita floor area (46%; 37% for men and 51% for women), followed by birth cohort (29%) and survey year (25%). The split between the components and the direction of the intergenerational gradient depend on the identifying restriction, whereas the gender difference in cohort and age effects is robust to it. The within-cohort improvement (+9.4 m2) is partly offset by a negative compositional shift (−1.8 m2): older, better-housed cohorts are being replaced by younger, worse-housed ones. The age profile reveals low housing provision at ages 30–44 and a peak at ages 65–69, when the household contracts to a childless couple for men and, more often, to a one-person household for women. For women, the swing between the two stages is nearly twice that for men. Mean housing provision is higher for women (+2.83 m2 in 2023). The gender gap is significantly positive in most cohorts born in 1935–1974 and significantly changes sign in the cohorts born in 1975–2014, where men are better housed; in the two youngest cohorts (2015–2019 and 2020–2024), observed in only part of the survey waves, the gap is not statistically significant.
The gender gap in housing provision largely reflects the demographically determined composition of households – for example, the prevalence of elderly women living alone (widows, elderly women who have never been married) – rather than differences in individual housing rights. The growth in the mean per-capita floor area over 20 years conceals opposing dynamics within and between cohorts, as well as the zigzag pattern of housing provision variation across the stages of the family life cycle.
age–period–cohort, gender gap, family life cycle, housing provision, cohort analysis, Intrinsic Estimator, pseudo-panel, Russia, shift-share decomposition
Over the last quarter of a century, housing provision for the Russian population has increased by a factor of 1.5 as a result of increased construction and wider access to housing, rising from a mean of 19.2 m2 per person in 2000 to 29.4 m2 in 2024 (
The gender dimension of housing adequacy in Russia is of particular interest, given that demographic factors – such as women’s higher life expectancy, which results in their predominance among widows and among older adults living alone – create systematic differences in the gender composition of households. International literature consistently highlights housing vulnerability of women, ranging from the challenges faced by single mothers (
As part of the journal’s special issue on gender studies in economics and demography, we examine the differences in housing trajectories between women and men as household members. This study aims to decompose the effects of age, cohort, and survey year on housing provision among the Russian population between 2003 and 2023, providing separate estimates for men and women. The study addresses three key research questions. Firstly, what is the relative importance of the three APC effects – biographical (age), macroeconomic (period) and generational (cohort) – in explaining the variation observed in housing provision? Secondly, what factors have contributed to the observed increase in the mean per-capita housing area over 20 years: improvements within cohorts or a structural shift resulting from generational replacement? Thirdly, are there significant gender differences in the APC effects, and at what ages and across which generations are they most pronounced?
During the post-Soviet period, the housing sector in the Russian Federation underwent several stages of institutional development, transitioning from a centralised, socialist system to a market-based one (
International studies highlight the vulnerability of women’s housing situation. In European countries
In the international literature, the relationship between demographic composition and housing consumption is examined through the lens of cohorts. A contextually similar study using German Socio-Economic Panel Survey data from 1984 to 2018 showed that, when controlling for income level, wealth, household size, and age, later cohorts consume significantly more per-capita floor area (
In the Russian context, the APC approach is appropriate due to the presence of demographic waves, such as the Soviet cohorts of varying sizes and the demographic trough of the 1990s. The classic application of APC models is the study of cohort differences in mortality (
The APC method is based on the pseudo-panel technique (
The linear relationship between the three variables – age, period and cohort – necessitates the imposition of additional explicit or implicit restrictions on the parameters and their relationships. It is well known that no method provides a perfect solution to the identification problem and that APC modelling results depend partly on model design choice, as the relative contributions of age, period and cohort effects vary according to the restrictions imposed. This supports the argument for viewing the results of APC modelling critically (
The Intrinsic Estimator (IE) uses a different restriction. From the set of solutions, the one that is orthogonal to the null space of the design matrix – that is, the one with the minimum Euclidean norm – is selected. This solution is provided by the Moore–Penrose pseudoinverse and the linear trend is distributed among three sets of coefficients. The IE is consistent in large samples and is widely used in demographic and sociological APC studies (
The choice between DP and IE is a matter of substantive interpretation. If the linear trend of improving housing conditions is interpreted as intergenerational, with each subsequent generation building more than the previous one, then the DP is preferable. However, if the trend is regarded as macroeconomic, with the housing stock growing simultaneously for all cohorts, then the IE is more appropriate. As Russia’s housing stock grew by almost 50 per cent between 2003 and 2023, while the mean per-capita floor area as per the Sample Survey of Household Budgets data increased by 28.7 per cent, the macro-trend of growth in housing provision is significant. Therefore, the DP’s assumption of a zero linear trend for this period appears implausible. This is precisely why we used the IE in this study.
In order to analyse medium-term growth in housing provision, it was broken down into two components below: growth in provision within cohorts and growth resulting from older cohorts being replaced by younger ones (the composition effect) (
The data source is Rosstat’s microdata from the Sample Survey of Household Budgets for 2003–2023 (21 waves). Rosstat has conducted the survey quarterly since 1996, using a sample of approximately 45,000–50,000 households per year. The data include individual characteristics for each family member. The target variable is the per-capita floor area, which is calculated by dividing the total floor area of the dwelling by the number of residents. This value is the same for all members of the household. A cohort is defined according to the year of birth of each household member. A single household may belong to several cohort cells simultaneously (for example, parents and children) and be included in both gender subsamples. Outliers are removed from the sample: an observation is excluded if the floor area divided by the number of rooms exceeds the median value by more than a factor of four.
As the Sample Survey of Household Budgets is a repeated cross-sectional survey rather than a longitudinal panel, the synthetic/pseudo-panel method (
The sample was split by gender for the gender breakdown, and the APC model was estimated overall for all individuals, as well as for each sex separately, using KVZV weighting
The final sample comprised 2,607,832 individual observations
The model is specified at the level of pseudo‑panel cohort cells:
, (1)
where ȳct – weighted mean per-capita floor area within a cohort cell c in the survey year t; μ – intercept; αa – age effect (a = t – c); πt – survey year effect; γc – cohort effect; εct – random error. Age, survey year, and cohort were coded using effect coding. Consequently, the sum of the coefficients within each of the three blocks is zero. Logarithmic transformation stabilised the variance, enabling the coefficients to be interpreted as percentage changes.
Estimation was carried out using the weighted ordinary least squares (OLS) method. Weights (nhh) were assigned based on the number of unique households in a cell, whereby several household members falling within the same cell (peers) were counted as a single observation. This weight was chosen because the per-capita floor area is identical for all members of the same household. Rosstat’s household sample weights (KVZV) were applied to the cell at the microdata aggregation stage, and nhh was used at the weighted OLS regression stage.
To address the identification problem, the Intrinsic Estimator was applied in its canonical parametrisation (
By default, the IE effects are expressed as deviations from the mean (mean = 0). To facilitate interpretation, the effects for each survey year were reparametrised with respect to the base year of 2003: , which alters neither the model fit (R2) nor the estimated age and cohort effects: the shift is absorbed by the intercept. A shift-share decomposition was applied to break down the observed change in mean provision into components attributable to growth within cohorts and changes in cohorts (
, (2)
where wc – proportion of cohort c in the population (weighted by KVZV), yc – mean housing provision for cohort c, Y – mean housing provision for the entire population (the sum of the products of the proportions and the mean values of the cohorts), Δ – difference between the final and initial years of observation, the overbar denotes the mean of the two end years. The first component is the composition effect and the second is the within-cohort effect. As both factors are symmetrically averaged, the sum of the two components is identically equal to the actual change in the mean. The decomposition was calculated for two ten-year intervals (2003→2013, 2013→2023) and in total for 2003→2023 for the entire population and separately for men and women.
For the gender decomposition, the IE model was estimated independently on three samples: a combined sample of 2,225,549 unique households across all cells, a male sample of 1,109,970, and a female sample of 1,454,702. All three estimates used the same specification (354 cells) and the same OLS weight structure (nhh), which was calculated for the respective subsample and represents the number of unique households in a cell. The gender gap δ was calculated separately for age and cohort effects as the difference between the female and male IE coefficients (δ = W − M). The standard error (SE) of the difference was calculated as and serves as an upper-bound estimate, since the exact variance is equal to Var(δ) = Var(W) + Var(M) − 2 ∙ Cov(W,M), while the male and female estimates within the household are positively correlated (their per-capita floor area is identical). Calculating the standard error using bootstrapping clustered by household and unclustered (by individual) (500 replicates) showed that the analytical SE is approximately one and a half times greater than both estimated values (median ratio 0.63). Clustering by household affected the standard error in both directions; the median ratio of the clustered standard error to the unclustered standard error was 0.91. However, in two out of nineteen cohorts, the ratio exceeded one (1.05 and 1.08). The largest (most conservative) of the three estimates – the analytical one – was used: it exceeded the clustered estimate in all 19 cohorts, and its confidence intervals were wider. If the difference is significant according to these intervals, it is all the more significant according to either of the two bootstrap estimates.
Figure
Increase in housing provision by five-year cohorts for 2003–2023: left axis – absolute growth, m² per capita; right axis – relative growth, %. Source: Author’s calculations based on the Sample Survey of Household Budgets data.
Age-Period-Cohort (APC) modelling using the Intrinsic Estimator on the combined sample explains 99.1 per cent of the variation in the logarithm of the mean per-capita floor area across the pseudo-panel cells. The quality of the fit is also high for both men and women (R2 = 0.979 and 0.990, respectively). Age is the largest source of variation (45.6 per cent for all respondents, 37.3 per cent for men, and 50.6 per cent for women; Table
Decomposition of the explained variation in the APC model (IE): combined, men, women
| Source of variation | Combined | Men | Women |
| Age | 45.6% | 37.3% | 50.6% |
| Survey year (period) | 24.9% | 36.9% | 20.4% |
| Birth cohort | 29.5% | 25.8% | 29.0% |
| R2 | 0.991 | 0.979 | 0.990 |
Cohort effects decline from +0.20 for the 1930–1934 cohort, to −0.27 for the 2015–2019 cohort, and then to −0.22 for the 2020–2024 cohort (see Figure
Cohort effect of APC decomposition (IE) as a percentage of the mean. The three lines represent the combined, men and women respectively. The shaded area represents the 95 per cent confidence interval (CI). Conversion from log to percentage: (exp(β) − 1) × 100. Source: Author’s calculations.
The housing space provision for men and women is similar across cohorts and, on the whole, shows a downward trend. In cohorts born between 1935 and 1974, women have access to better housing (this difference is statistically significant in all cohorts except those born between 1965 and 1969), which is consistent with older women being more likely to live alone or in smaller households. In cohorts born between 1975 and 2014, however, women are more likely than men to live in families with children and share their housing space with other family members. In these cohorts, men’s housing provision is higher (the difference is statistically significant in all eight cohorts). In the two youngest cohorts, those born between 2015 and 2019 and between 2020 and 2024, which are observed in only 9 and 4 of the 21 survey waves, the point estimate is negative but not significantly different from zero.
The survey-year effects demonstrate a consistent increase from 0.000 in 2003 to +0.324 in 2023, indicating a +38 per cent improvement over 20 years (Figure
Survey year effect (IE), percentage compared with 2003. The three lines represent the combined, men and women respectively. The shaded area represents the 95 per cent CI. Source: Author’s calculations.
The age profile is markedly non‑linear, with a range of 0.39 log units (48%) in the combined sample (Figure
Age effect of APC decomposition (IE) as a percentage of the mean: three lines – combined, men, women. The shaded area represents the 95 per cent CI. Source: Author’s calculations.
Separate estimates by gender reveal substantial differences in the amplitude of the age profile. The profile is notably steeper for women, with a dip in the family nest phase reaching −0.20 log units at 35 years of age (compared to −0.10 at 40 years of age for men). Meanwhile, the late peak occurs at 70 years of age, equalling +0.23 (compared to +0.15 at 60 years of age for men). The range between these two stages is 0.43 log units for women, compared to 0.25 for men (+54 per cent versus +28 per cent from trough to peak) – almost twice as large. The total amplitude of the profile is 0.44, compared to 0.30 log units. The mechanism behind the late peak also differs. For men, the peak coincides with the empty nest phase: the proportion of men living as a couple without children rises from 19 per cent among 45–49-year-olds to 65 per cent among 65–69-year-olds, while the proportion of men living alone in these age groups does not exceed seven per cent. For women, the empty nest phase occurs between the ages of 55 and 64. From the age of 65 onwards, living alone becomes the most common pattern: 40 per cent of women aged 65–69 and 45 per cent of women aged 70–74 live alone, compared with 30 per cent and 24 per cent who live as a couple without children (Sample Survey of Household Budgets, 2015–2023). Consequently, the peak for women occurs a decade later than for men (Figure
Figure
Trajectories for six selected cohorts by gender: mean per-capita floor area, m2 (solid line – women, dashed line – men). Source: Author’s calculations.
Between 2003 and 2023, the mean per-capita floor area across the 14 cohorts observed at the start and end of the twenty-year period increased by 7.56 m2 (see Table
Shift-share decomposition of change in mean housing provision: composition and within-cohort components, m2 per person
| Period | Sex | Total | Composition | Within-cohort |
| 2003–2023 | Combined | +7.56 | −1.79 | +9.35 |
| 2003–2023 | Men | +7.00 | −1.12 | +8.12 |
| 2003–2023 | Women | +7.95 | −2.14 | +10.10 |
| 2003–2013 | Combined | +4.03 | −1.21 | +5.23 |
| 2003–2013 | Men | +3.37 | −0.92 | +4.29 |
| 2003–2013 | Women | +4.53 | −1.31 | +5.84 |
| 2013–2023 | Combined | +2.78 | −1.42 | +4.20 |
| 2013–2023 | Men | +2.90 | −0.97 | +3.87 |
| 2013–2023 | Women | +2.69 | −1.67 | +4.36 |
A separate shift-share decomposition revealed a common pattern for both sexes: within-cohort growth dominates structural shift. For men, the total increase is 7.00 m2 (+8.12/ –1.12), and for women it is 7.95 m2 (+10.10/–2.14). Within-cohort growth is more pronounced among women, but the negative structural effect is also stronger: older women who are leaving the sample are characterised by a particularly high per-capita floor area (they are more likely to live alone), while young women entering the sample have a smaller floor area. A breakdown by the decades 2003–2013 and 2013–2023 shows a slowdown in progress. For women, the contribution of within-cohort growth fell from +5.84 m2 in the first decade to +4.36 in the second. For men, it fell from +4.29 to +3.87. As the decomposition is calculated for cohorts observed at both ends of each decade (15 and 16 cohorts compared to 14 for 2003–2023), the components of the sub-periods do not add up to the total for the entire period.
The mean per-capita floor area for women was higher than that for men in every year of the observation period: 19.35 m2 versus 17.07 m2 in 2003 (+2.28 m2) and 24.88 m2 versus 22.05 m2 in 2023 (+2.83 m2). To analyse the gender gap, the differences in age- and cohort-specific IE effects were calculated.
The gender gap in age effects is significant in most age groups and non-monotonic (see Figure
Gender gap (W − M) in the age effect, percentage points. The shaded area represents the 95 per cent CI. Source: Author’s calculations.
Figure
Gender gap (W − M) in the cohort effect, percentage points. The shaded area represents the 95 per cent CI. Source: Author’s calculations.
Robustness was evaluated in four distinct ways. Firstly, an alternative Deaton–Paxson APC identification was implemented in which the linear trend relates to age and cohort rather than the survey year. Secondly, an alternative parametrisation of the same minimum-norm IE solution was carried out with projection onto the space of the full set of dummy variables (indicator coding). Thirdly, the reference model was re-estimated using subsamples of people aged 18 and over and 25 and over (excluding children). Fourthly, a cluster bootstrap analysis was performed at the household level (500 replicates) to estimate the standard errors of the gender gap. The findings are presented in Table
Both the variance decomposition and the direction of the intergenerational gradient are sensitive to the choice of restriction. In the Deaton–Paxson identification, the proportion attributable to age rises to 78.5 per cent and the proportion attributable to cohort falls to 21.2 per cent, while the sign of the cohort gradient reverses in this model (r = −0.96 with IE estimates). By contrast, in the indicator coding, the cohort accounts for 62.1 per cent and age for 11.2 per cent. Therefore, the declining cohort profile (Figure
By contrast, the gender difference in cohort and age effects is robust to the choice of the APC model. Its sign is consistent across all 19 cohorts and 18 age groups. The correlation is 0.999 when the identification method is changed, and 0.9999 when the parametrisation is changed. Excluding children steepens the cohort gradient without altering its shape (the correlation with the reference specification is 0.999 for the 18+ subsample and 0.994 for the 25+ subsample). Cluster bootstrap analysis shows that, on average, accounting for intra-household dependence does not increase standard errors: the median ratio of cluster SE to non-cluster SE is 0.91, ranging from 0.76 to 1.08. In two cohorts (2015–2019 and 2020–2024), the ratio exceeds one. The bootstrap estimates are lower than the analytical estimate in all cohorts; therefore, the confidence intervals used in the paper remain conservative.
The key finding of this study is the gender difference in the cohort and age effects on housing provision. This difference remains consistent regardless of the identification method or parametrisation chosen. The declining cohort profile obtained in the reference IE specification of the APC model (from +0.20 to −0.27) does not imply that younger generations are worse off in absolute terms. Under this specification, the effects represent a cohort’s position relative to the overall trend while the survey-year effect (+38 per cent over 2003–2023) captures the bulk of the linear trend of macro-level improvement. The age profile (a dip between 30 and 44 years, peaking at 65–69 years) is consistent with that observed in Germany (
The consistently higher level of housing provision for women can be attributed mainly to the demographic composition of households rather than to differences in individual property rights. In the Sample Survey of Household Budgets, housing adequacy is measured at the household level, while individual per-capita floor area is an aggregate figure that is the same for all family members. The distribution of men and women across household types is asymmetrical, as calculated from the Sample Survey of Household Budgets between 2015 and 2023. Among those aged 65 and over, 41.9 per cent of women live alone compared to 8.6 per cent of men. Meanwhile, men are more likely to remain in complete married households (see Table
This finding is consistent with the results of a study by
The design of our study follows the methodology of
Shift-share decomposition reveals that the aggregate indicator, mean per-capita floor area, obscures actual trends. A policy focused on the aggregate may underestimate real progress within cohorts and structural challenges resulting from generational replacement. Among women, the increase in housing provision and the negative structural shift are both greater than among men. The gender-specific nature of the shift-share decomposition reflects demographic patterns: women are more likely to live alone in older age groups and older cohorts have a higher baseline level.
This study has several limitations. Firstly, the Sample Survey of Household Budgets measures floor area at the household level, meaning the per-capita figure is an aggregate figure identical for all family members. Gender differences arise from differences in the demographic composition of households rather than from differences in individual housing rights, on which the Sample Survey of Household Budgets provides no information. Furthermore, apart from the composition and size of the household at the time of the survey, no information is available on transitions between states, such as widowhood, relocation to live with relatives, or a change of residence. This study offers explanations for these transitions as hypotheses that are consistent with the observed household composition. Direct verification of these hypotheses requires panel data. Secondly, the cohorts born between 2005 and 2024 consist predominantly of children; therefore, their cohort effect does not reflect their own housing-related decisions. Testing on samples of adults aged 18+ and 25+ retains the shape of the cohort profile (r = 0.999 and 0.994, respectively, for the combined sample), while the gradient becomes steeper (see Table
Our findings reveal a specific pattern of housing vulnerability in Russia: while the mean floor area per elderly woman is nominally high, this does not represent housing adequacy. This group is characterised by the combination of having a large dwelling and low income, which may entail high maintenance costs, limited mobility, and financial stress; moreover, life satisfaction among older people living alone in Russia is closely linked to economic conditions (
This study is the first to apply the APC approach with Intrinsic Estimator identification to housing provision of the Russian population. Microdata from the 2003–2023 Sample Survey of Household Budgets were used to make separate estimates of the age–period–cohort effects for men and women. The following findings were obtained:
1. In the canonical IE identification, age accounts for the largest component of variance in housing provision (approximately 46 per cent across the entire sample, 37 per cent for men and 51 per cent for women). Birth cohort accounts for approximately 29 per cent, and the survey year for approximately 25 per cent. The variance decomposition across components and the direction of the intergenerational gradient depend on the restriction chosen and are not identifiable from the data. Older cohorts (1930–1974) exhibit a positive cohort effect, while younger cohorts (1985–2024) exhibit a negative one. The range is from +0.20 to −0.27 log units.
2. The survey year effect (approximately 25% of the total variance) represents the main macro-trend: a steady increase in the per-capita floor area of 38% between 2003 and 2023. Gender differences in the survey-year effects are negligible (a maximum of 0.03 log units).
3. The age effect (approximately 46% of the total variance) follows a distinctly non-linear pattern: a trough between the ages of 30 and 44, a late peak between 65 and 69, and a subsequent decline towards the end of life. In women, the range between these stages is almost double that in men (0.43 versus 0.25 log units), and the peak occurs a decade later. In men, it coincides with the empty nest phenomenon, while in women it coincides with transitioning to living alone.
4. The increase in the mean housing provision of +7.56 m2 over 20 years is entirely due to improvements within cohorts (+9.35 m2), which are partly offset by a compositional shift resulting from generational replacement (−1.79 m2). Both of these components are more pronounced among women (see Table
5. As seen from Figure
6. Women consistently have a higher level of housing provision than men, with the gap between the two increasing from +2.28 m2 in 2003 to +2.83 m2 in 2023. Among cohorts born between 1975 and 2014, the gap significantly changes sign (−0.03…−0.07 log units), with young men enjoying a higher level of housing provision than their female peers. However, in the two youngest cohorts (2015–2019 and 2020–2024), the gap is non-significant, and the conclusion regarding the change in sign does not apply.
7. Gender differences primarily represent the different housing conditions experienced by men and women in households of various compositions, rather than differences in their individual housing rights or property resources. Housing policy measures must take these specific circumstances into account, particularly with regard to older women living alone.
The article was written on the basis of the RANEPA state assignment research programme.
Aleksandra Yaroslavovna Burdyak – Senior Researcher. Center “Institute of Social Analysis and Forecasting”, Institute of Applied Economic Research, Russian Presidential Academy of National Economy and Public Administration. Moscow, 119034, Russia. Email: burdyak-ay@ranepa.ru
| Cohort | Combined | M | W | W − M |
|---|---|---|---|---|
| 1930–1934 | +0.20 | +0.18 | +0.20 | +0.03 |
| 1935–1939 | +0.17 | +0.12 | +0.19 | +0.07 |
| 1940–1944 | +0.15 | +0.09 | +0.18 | +0.09 |
| 1945–1949 | +0.13 | +0.09 | +0.15 | +0.06 |
| 1950–1954 | +0.13 | +0.08 | +0.15 | +0.07 |
| 1955–1959 | +0.12 | +0.10 | +0.14 | +0.03 |
| 1960–1964 | +0.11 | +0.08 | +0.13 | +0.06 |
| 1965–1969 | +0.09 | +0.07 | +0.10 | +0.02 |
| 1970–1974 | +0.04 | +0.02 | +0.06 | +0.04 |
| 1975–1979 | +0.01 | +0.03 | −0.01 | −0.03 |
| 1980–1984 | −0.03 | −0.01 | −0.04 | −0.03 |
| 1985–1989 | −0.03 | +0.01 | −0.05 | −0.06 |
| 1990–1994 | −0.05 | −0.01 | −0.08 | −0.07 |
| 1995–1999 | −0.07 | −0.03 | −0.10 | −0.06 |
| 2000–2004 | −0.11 | −0.07 | −0.13 | −0.06 |
| 2005–2009 | −0.16 | −0.13 | −0.16 | −0.03 |
| 2010–2014 | −0.21 | −0.18 | −0.23 | −0.06 |
| 2015–2019 | −0.27 | −0.26 | −0.27 | −0.01 |
| 2020–2024 | −0.22 | −0.18 | −0.24 | −0.07 |
Age effects (IE), log units. Selected ages. The W−M difference is significant at the 5 per cent level in all age groups except 45, 80, and 85 years old
| Age | Combined | M | W | W − M |
|---|---|---|---|---|
| 0 | −0.19 | −0.15 | −0.21 | −0.06 |
| 20 | −0.08 | −0.05 | −0.09 | −0.04 |
| 30 | −0.10 | −0.06 | −0.12 | −0.06 |
| 35 | −0.16 | −0.10 | −0.20 | −0.10 |
| 40 | −0.14 | −0.10 | −0.16 | −0.06 |
| 45 | −0.06 | −0.06 | −0.04 | +0.01 |
| 50 | +0.06 | +0.05 | +0.08 | +0.04 |
| 55 | +0.15 | +0.13 | +0.17 | +0.04 |
| 60 | +0.18 | +0.15 | +0.21 | +0.06 |
| 65 | +0.20 | +0.14 | +0.23 | +0.09 |
| 70 | +0.20 | +0.13 | +0.23 | +0.10 |
| 75 | +0.18 | +0.11 | +0.20 | +0.08 |
| 80 | +0.13 | +0.10 | +0.12 | +0.02 |
| 85 | +0.10 | +0.07 | +0.10 | +0.03 |
Household size and distribution by household type, and proportion of widowed individuals by sex and age group (% based on the Sample Survey of Household Budgets 2015–2023, weighted by KVZV)
| Age | Sex | Household size | Living alone | Childless couple | Couple with children | Single- parent family | Other type | Widowed individuals |
|---|---|---|---|---|---|---|---|---|
| 0–17 | W | 4.0 | 0.0 | 0.9 | 84.4 | 14.0 | 0.7 | – |
| 0–17 | M | 4.0 | 0.0 | 0.7 | 84.5 | 14.1 | 0.7 | – |
| 18–29 | W | 3.4 | 5.8 | 10.5 | 65.3 | 16.4 | 2.0 | 0.2 |
| 18–29 | M | 3.5 | 3.7 | 8.1 | 69.1 | 17.6 | 1.5 | 0.1 |
| 30–44 | W | 3.4 | 5.1 | 8.5 | 70.1 | 15.2 | 1.1 | 1.4 |
| 30–44 | M | 3.4 | 6.7 | 9.3 | 76.1 | 7.0 | 0.8 | 0.2 |
| 45–59 | W | 2.5 | 19.1 | 31.2 | 34.3 | 12.9 | 2.5 | 10.0 |
| 45–59 | M | 2.9 | 5.7 | 35.7 | 54.5 | 3.3 | 0.8 | 0.9 |
| 60–74 | W | 2.1 | 38.2 | 31.3 | 13.5 | 13.3 | 3.7 | 38.4 |
| 60–74 | M | 2.5 | 6.4 | 64.4 | 24.8 | 2.8 | 1.6 | 5.5 |
| 75+ | W | 2.2 | 41.6 | 18.7 | 13.4 | 15.8 | 10.5 | 76.8 |
| 75+ | M | 2.7 | 13.1 | 47.8 | 22.9 | 11.0 | 5.2 | 29.5 |
| 65+ (informational) | W | 2.1 | 41.9 | 24.6 | 12.8 | 14.7 | 6.2 | 56.9 |
| 65+ (informational) | M | 2.6 | 8.6 | 60.1 | 23.4 | 5.3 | 2.6 | 13.3 |
Proportions of people living in one- and two-person households by age and sex (%), based on the Sample Survey of Household Budgets (2003–2023), weighted by KVZV. Source: Author’s calculations.
Robustness checks for the APC estimates: explained variance and consistency with the baseline specification
| Specification | Shares of variance, % (age / survey year / cohort) | Consistency with the baseline specification |
| Baseline: IE, effect coding | 45.6 / 24.9 / 29.5 | – |
| Deaton–Paxson identification | 78.5 / 0.3 / 21.2 | cohort effects; r = −0.96; gender difference: sign for 19/19 cohorts and 18/18 ages, r = 0.9986 |
| Indicator coding | 11.2 / 26.7 / 62.1 | cohort effects; r = 0.996; gender difference: sign for 19/19 cohorts and 18/18 ages, r = 0.9999 |
| 18+ subsample | 45.2 / 32.5 / 22.3 | cohort effects; r = 0.999 |
| 25+ subsample | 40.6 / 35.5 / 23.9 | cohort effects; r = 0.994 |
| Cluster bootstrap, 500 replicates | – | gender difference SE: clustered to non-clustered – median 0.91 (0.76–1.08), ratio > 1 in 2 cohorts out of 19; both bootstrap estimates are below the analytical estimate in all cohorts |