Men | EU- SILC (2005–2017)* n=1 885 712 | ESS (2002–2016)† n=110 603 | Difference between surveys (2)−(1)‡ | |||

Δ inequality in 1 year | (1) | (2)§ | Average inequality¶ | (1) | Average inequality¶ | |

Prevalence difference (95% CI) | 0.28 (0.21 to 0.34) | 0.11 (0.03 to 0.17) | 12.05 (11.26 to 12.84) | −0.03 (−0.27 to 0.21) | 9.92 (9.25 to 11.95) | 0.14 (−0.14 to 0.41) |

Prevalence ratio (95% CI) | 0.026 (0.017 to 0.028) | 0.010 (0.004 to 0.015) | 1.650 (1.591 to 1.710) | −0.003 (−0.015 to 0.004) | 1.532 (1.421 to 1.701) | −0.013 (−0.030 to 0.004) |

Women | EU- SILC (2005–2017)*n=2 064 508 | ESS (2002–2016)† n=128 401 | Difference between surveys (2)−(1)‡ | |||

Δ inequality in 1 year | (1) | (2)§ | Average inequality¶ | (1) | Average inequality¶ | |

Prevalence difference (95% CI) | 0.26 (0.19 to 0.32) | 0.12 (0.04 to 0.18) | 12.53 (11.76 to 13.36) | 0.02 (−0.21 to 0.23) | 9.19 (7.13 to 11.05) | 0.10 (−0.19 to 0.38) |

Prevalence ratio (95% CI) | 0.016 (0.011 to 0.019) | 0.009 (0.004 to 0.012) | 1.563 (1.522 to 1.612) | 0.001 (−0.013 to 0.015) | 1.421 (1.30 to 1.54) | 0.008 (−0.010 to 0.025) |

Significant at the 95% level in bold.

Estimates are obtained from gender-stratified logistic models using microdata with the dichotomous GALI indicator as dependent variable:

(1) logit(GALI)=b0+b1(age)+b2(age)(age)+b3(education)+b4(year)+b5(year*education)+b6(country).

(2) logit(GALI)=b0+b1(age)+b2(age)(age)+b3(education)+b4(year)+b5(year*education)+b6(country)+b7(GALI comparability).

The annual change in prevalence by education is estimated by fitting the logistic regressions with all countries pooled (using the product of the population and survey weights in the regression), and after estimation, the command

*margins edu_3cat, dydx(year)*gives the average marginal (partial) effects. This means that the effects are calculated for each observation in the data and then averaged. The annual change in prevalence difference is estimated by subtracting the marginal (partial) effects of the low and high educated.The annual average change in prevalence ratio is estimated by predicting the prevalence of disability by education and year after fitting the logistic models using the

*margins*command and the prevalence ratios, and then calculating the average change over the period of study.*European Union Statistics on Income and Living Conditions (EU-SILC) annual microdata between 2005 and 2017. Member countries in EU-SILC use variants of the GALI question over time.

†European Social Survey (ESS) biannual microdata between 2002 and 2016. ESS uses the same version of the GALI question for all countries and years. This version omits the 6-month time frame of the standard GALI question.

‡Two-sample t-test between the EU-SILC and ESS coefficients.

§GALI comparability estimates include baseline model plus a three-level categorical variable related to phrasing (comparable, partially comparable, not comparable) for EU-SILC only.

¶Average age-standardised GALI prevalence over the corresponding period for each survey using the 2013 European Standard Population for all countries included in the sample.