Health data from the National Sample Survey shows an increase in morbidity in India over the years. However, given that the data is self-reported, it is difficult to ascertain whether this indeed reflects higher actual illness burden or an enhanced perception of morbidity. This column shows that reporting behaviour varies systematically with socio-demographic characteristics, and this can be used to disentangle perceived and actual morbidity.
A key challenge in the analysis and interpretation of health survey data, which often relies on self-reported illness and usage of healthcare services, is tackling differences in the ways that individuals understand and use the response options for a given question. The morbidity trend from the preliminary report of the recently released National Sample Survey Organisation’s (NSSO) 71st round (January to June 2014) of the National Sample Survey (NSS) titled “Key Indicators of Social Consumption in India: Health” is indicative of this problem yet again.
The 71st round is a part of a decennial series - the previous three rounds in the series were the 60th round (January to June 2004), the 52nd round (July 1995 to June 1996), and the 42nd round (July 1986 to June 1987). Given the dearth of national-level objective/measured health data1 in India (the last National Family Health Survey (NFHS-3) which has objective information on health was published about 10 years ago in 2005-06), this becomes an important source of data for health planning and tracking especially in the areas of morbidity and healthcare utilisation.
Looking at data from the 52nd, 60th and 71st rounds of the NSS, we can see that morbidity – as captured by the Proportion (per 1,000) of Ailing Persons (PAP) in last 15 days – has consistently increased for both rural and urban locations over the years. It is consistently higher in urban as compared to rural areas and the gap between the two has been widening.
Socio-demographic status and the perception of being ill
A crucial issue in this context is that the analysis of the morbidity profile of the respondents is entirely captured by subjective/self-reported health measures in the survey. In all of these NSSO surveys, every household in the sample is asked the following questions:
(i) Whether they were ill in the last 15 days and if so, the nature of illness?
(ii) Whether they were hospitalised in the last 365 days and if so, why?
Now one of the main problems with self-reported ailment is that the perception of being ´ill´ can vary systematically by cultural, regional and other socio-demographic factors including expectations for own health and disease awareness. For example, a person with simple cough and cold who belongs to a backward region is less likely to consider himself/herself ´ill´ and may not report illness in the survey, as compared to a person from a developed region.
Apart from rural/urban location, this reporting variation can happen due to differences in perception of illness, which may be influenced by age, gender, education, income, language and personal experience of illness. Someone who has ready access to health facilities may perceive illness and seek remedy more readily than someone in a less favourable environment. As a result, different groups are likely to interpret their health status within their own specific contexts and therefore use different reference points when they respond to the same question.
In such cases it is difficult to decide whether higher levels of reported illness from the NSS records indeed reflects higher real burden of illness or an enhanced perception of morbidity, as captured by self-reported responses of being ill in the reference period of last two weeks. Similarly, it is not possible to determine whether the actual burden of illness is indeed more in urban areas and increasing at a faster rate vis-à-vis rural areas, or if the difference is mostly due to the difference in perception of illness of the respondents, depending on their location. As we do not have corresponding objective health indicators, it may be problematic to attribute the increase in morbidity rate (as measured by the PAP) to actual illness burden. Hence, it is important to identify how much of the illness burden can be attributed to actual disease burden after taking into account the systematic differences in illness perception (and thus, reporting of it) by different socio-demographic subgroups.
A finding from the NSS that is potentially indicative of this problem is that states with higher literacy and better healthcare systems, like Kerala and Tamil Nadu, have been consistently found to have much higher PAPs than states with poor healthcare systems like Uttar Pradesh and Bihar. Similarly, the PAP also increases with higher income quintile as seen from the figures of the survey reports. A similar pattern is observed for reported hospitalisation: where states with lower Infant Mortality Rate (IMR) and Maternal Mortality Rate (MMR) have higher hospitalisation rates compared to worse-performing states.
Disentangling perceived and actual illness burden
In this context, I examine the pattern of reporting differences in self-reported health status from a nationally representative survey in India – the World Health Survey (WHS)-SAGE survey (Dasgupta 2014
). The data covers six states namely Maharashtra, Karnataka, West Bengal, Rajasthan, Uttar Pradesh and Assam from 2007 to 2009. The data collected includes self-reported assessments of health linked to anchoring ´vignettes´, which are hypothetical stories that describe the health problems/situations of third parties in several health domains. I look into the question of formalising the pattern of reporting variation of the respondents by their socio-demographic subgroups.
Typically in the vignette schedule, a hypothetical third-party scenario is described to the household survey respondents and they are asked to give a rating of the health situation of the person in the scenario. The assumption is that the respondents would rate the health situation of the hypothetical person just as they would have rated their own health. Since all the respondents are being asked the same question depicting the same health situation, any difference in rating of the given situation may be attributed to the difference in subjective reporting pattern. The systematic pattern of reporting variation can then be used to account for the reporting differences by socio-demographic subgroups, thus enabling greater comparability of the actual health status.
I find evidence of systematic differences in reporting behaviour that vary with demographic characteristics such as age, gender and education, and community characteristics such as location (rural/urban) and level of development in the state.
The results reiterate the previous point that difference in reported health status can largely be attributed to the differences in subjective health perceptions. The study finds an interesting pattern of reporting variation, which is that individuals from less developed states typically under-report illness. This is perhaps suggestive of the fact that socially disadvantaged individuals fail to perceive and report the presence of illness because an individual’s assessment of their health is directly contingent on their social experiences. It can be attributed to lower expectation for own health/higher tolerance for diseases where an individual may not see himself/herself as being unhealthy conditional on the health norm prevailing in one’s community. Hence, there is strong evidence on systematic reporting bias and the problem of comparability of morbidity profiles with self-reported data remains a serious issue. This has several important implications.
Implications for measurement of subjective indicators of well-being
First, one should be careful in making interpersonal comparisons of health status using self-reported health data. In the absence of objective data on health status (for example, clinical test records, which are costlier to undertake) it would be prudent to include an additional vignette schedule in the household questionnaire of the NSS that can be used to detect any bias arising from perceptions, thus enabling policymakers to judge how much of the difference in reported health condition can be attributed to real burden of illness and how much to subjective perception of illness. Additionally, my analysis lends support to the use of vignettes data to account for the subjective variation in health perception and to make possible greater comparability across distinct socio-demographic groups. The findings have potentially broader implications for measurement and assessment of subjective responses regarding complex variables that comes under the ambit of individual well-being (like that of happiness, empowerment, trust etc.) that are increasingly getting attention in social science research.
- Such data is based on clinically tested or measured illness levels of the sample households.