Intra-cluster correlations in socio-demographic variables and their implications: An analysis based on large-scale surveys in India. (March 2023)

• Record Type: Journal Article
• Authors: Dwivedi, Laxmi Kant; Mahaptra, Bidhubhusan; Bansal, Anjali; Gupta, Jitendra; Singh, Abhishek; Roy, T.K.
• Is Part Of: SSM - population health. Volume 21(2023)
• Publication Date: 2023-03
• Subjects: Clustering -- PSU -- NFHS -- Intra-class correlation coefficients -- ICC -- Multilevel analysis -- Religion -- Caste -- Use of family planning -- Anaemia -- Full immunization -- Precision loss -- Cluster size -- Sample size -- Effective sample size -- India; ICC: Intra-class Correlation Coefficient -- NFHS: National Family Health Survey -- PSU: Primary Sampling Unit -- CEB: Census Enumeration Block -- EPSEM: Equal Probability of Selection Method; Social medicine -- Periodicals; Medical anthropology -- Periodicals; Public health -- Periodicals; Psychology -- Periodicals; Medicine -- Periodicals; 362.105
• DOI: 10.1016/j.ssmph.2022.101317
• Languages: English
• ISSNs: 2352-8273

Abstract

Abstract: Individuals who share similar socio-economic and cultural characteristics also share similar health outcomes. Consequently, they have a propensity to cluster together, which results in positive intra-class correlation coefficients (ICCs) in their socio-demographic and behavioural characteristics. In this study, using data from four rounds of the National Family Health Survey (NFHS), we estimated the ICC for selected socio-demographic and behavioural characteristics in rural and urban areas of six states namely Assam, Gujarat, Kerala, Punjab, Uttar Pradesh, and West Bengal. The socio-demographic and behavioural characteristics included religion & caste of the household head, use of contraception & prevalence of anaemia among currently married women and coverage of full immunization services among children aged 12-23 months. ICC was computed at the level ofPrimary Sampling Units (PSUs), that is, villages in rural areas and census enumeration blocks in urban areas. Our research highlights high clustering in terms of religion and caste within PSUs in India. In NFHS-4, the ICCs for religion ranged from the lowest of 0.19 in rural areas of Kerala to the highest of 0.67 in urban areas of West Bengal. For the caste of the household head, the ICCs ranged from the lowest of 0.12 in the urban areas of Punjab to the highest of 0.46 in the rural areas of Assam. In most of the states selected for the study, the values of ICC were higher for the use of family planning methods than for full immunization. The value of ICC for use of contraception was highest for rural areas of Assam (0.15) followed by rural areas of Gujarat (0.13). A higher value of ICC has considerable implications for determining an effective sample size for large-scale surveys. Our findings agree with the fact that for a given cluster size, the higher the value of ICC, the higher is the loss in precision of the estimate. Knowing and taking into account ICCs can be extremely helpful in determining an effective sample size when designing a large-scale demographic and health survey to arrive at estimates of parameters with the desired precision. Highlights: The analysis suggests that intra-class correlation coefficients (ICC) varied from 0.03 to 0.71 for selected socio-demographic and behavioural variables, suggesting a high clustering in the populations living near each other in a PSU. Clustering was more pronounced in relation to the religion than the caste of the head of the household in all the selected states, especially in the rural areas. In most of the states selected for the study, the values of ICC were higher for the use of family planning methods than for full immunization. The findings agree with the fact that for a given cluster size, the higher the value of ICC, the higher is the loss in precision of the estimate. Measuring ICC can effectively help in deciding the optimum cluster size, which can reduce the bias in correctly estimating the health outcomes of the country.