India’s disability estimates: Limitations and way forward

BackgroundWith India preparing for the next decennial Census in 2021, we compared the disability estimates and data collection methodology between the Census 2011 and the most recent population-level survey for India and its states, to highlight the issues to be addressed to improve robustness of the disability estimates in the upcoming Census.MethodsData from the Census 2011 and from two complementary nationally representative household surveys that covered all Indian states with the same methodology and survey instruments–the District-Level Household Survey-4 (DLHS-4, 2012–2013) and the Annual Health Surveys (AHS three rounds, 2010–11, 2011–12 and 2012–13) were used. Data from DLHS-4 and AHS 2012–13 round were pooled to generate estimates for the year 2012–13. Data collection methodology between the sources was compared, including the review of definitions of each type of disability. The overall, mental, visual, hearing, speech, and movement disability rate (DR) per 100,000 population were compared between the sources for India and for each state, and the percent difference in the respective rates was calculated. We explored the reliability of these estimates comparing yearly data from the AHS for three successive rounds.ResultsSurvey data were collected through proxy reporting, however, it is not entirely clear whether the data were proxy- or self-reported or a mix of both in the Census. The overall DR was 25.1% higher in the Census (2,242; 95% CI 2,241–2,243) than the survey (1,791; 95% CI 1,786–1,797) per 100,000 population, with the state-level difference ranging from -64% in Tamil Nadu to 107% in Sikkim state. Despite both sources using nearly similar definitions for overall disability and disability by type, the difference in DR was 125.5%, 54.2%, -25.7%, -19.7%, and 21.9% for hearing, speech, mental, movement, and visual DR, respectively. At the state-level, the difference in disability-specific estimates ranged from -84% to 450%. The extent of variations in the disability-specific estimates in AHS successive rounds ranged from -25% to 929% at the state-level.ConclusionsThere is momentum globally towards building disability measurement that is consistent with the data required for monitoring of the Sustainable Development Goals to ensure robust estimation of disability. The current estimates from the Census and surveys seem much lower than would be expected at the population level. We make recommendations that India needs to take serious note of in order to improve the validity and reliability of India’s disability estimates.【背景】印度正筹备2021年十年一次人口普查（decennial Census），本研究对比了2011年人口普查与印度全国及各邦最新的人口层面调查的残疾估计数及数据收集方法，旨在明确为提升即将开展的人口普查残疾估计稳健性所需解决的问题。 【方法】本研究采用了2011年人口普查数据，以及两项采用统一方法与调查工具、覆盖印度所有邦的互补性全国代表性家庭调查数据：第四轮地区级家庭调查（District-Level Household Survey-4, DLHS-4，2012–2013）与年度健康调查（Annual Health Surveys, AHS，共三轮，2010–11、2011–12及2012–13）。研究合并DLHS-4与2012–2013轮次AHS的数据，生成2012–2013年的残疾估计值。研究对比了不同数据源的数据收集方法，包括对各类残疾定义的梳理。分别对比了印度全国及各邦每10万人口的总体、精神、视力、听力、言语及运动残疾率（Disability Rate, DR），并计算了对应残疾率的百分比差异。同时通过对比AHS三轮连续年度数据，探究了上述估计值的可靠性。 【结果】调查数据均通过代理报告收集，但人口普查中数据究竟为代理报告、自我报告还是二者混合，目前尚不明确。每10万人口的总体残疾率，人口普查结果为2242（95%置信区间(Confidence Interval, CI)：2241–2243），较调查结果的1791（95% CI：1786–1797）高出25.1%；各邦层面的差异范围为泰米尔纳德邦的-64%至锡金邦的107%。尽管两类数据源对总体残疾及各类别残疾的定义近乎一致，但听力、言语、精神、运动及视力残疾率的差异分别为125.5%、54.2%、-25.7%、-19.7%及21.9%。在邦级层面，特定类型残疾的估计值差异范围为-84%至450%。AHS连续轮次中，各邦特定类型残疾估计值的变异幅度范围为-25%至929%。 【结论】全球范围内，构建契合可持续发展目标（Sustainable Development Goals, SDGs）监测所需数据的残疾测量体系的趋势日益显著，以确保残疾估计的稳健性。当前人口普查与调查得到的残疾估计值，远低于人口层面的预期水平。本研究提出若干建议，印度需予以重视，以提升本国残疾估计值的效度与信度。