Elsevier

Annals of Epidemiology

Volume 4, Issue 4, July 1994, Pages 295-301
Annals of Epidemiology

Original report
Estimation of design effects in cluster surveys

https://doi.org/10.1016/1047-2797(94)90085-XGet rights and content

Abstract

Cluster sampling can produce estimates of disease prevalence that are more variable than those from simple random sampling. This variance inflation or “design effect” depends on the prevalence of disease, the cluster sizes, and the magnitude of disease association within clusters. Design effects from prior surveys may not be appropriate for a planned survey if these components differ. We estimated within-cluster associations using pairwise odds ratios, which are more portable than design effects because they do not depend on the cluster sizes. Within-village pairwise odds ratios and design effects were estimated for fever and cough from four studies in Africa and Asia. Odds ratios ranged from 1.04 to 1.34 and 1.03 to 1.24, respectively. Design effects ranged from 2.35 to 6.80 for fever and 1.99 to 7.39 for cough. The design effect was more affected by cluster size and odds ratio than by variation in cluster size for a given sample size.

References (27)

  • A. Sommer

    Nutritional Blindness. Xerophthalmia and Keratomalacia

    (1982)
  • R. Bailey et al.

    Analysis of the household distribution of trachoma in a Gambian village using a Monte Carlo simulation procedure

    Int J Epidemiol.

    (1989)
  • J.M. Tielsch et al.

    The epidemiology of trachoma in southern Malawi

    Am J Trop Med Hyg.

    (1988)
  • Cited by (41)

    • Intra-cluster correlations in socio-demographic variables and their implications: An analysis based on large-scale surveys in India

      2023, SSM - Population Health
      Citation Excerpt :

      In case of high values of ICC, the assumption of a deff of 2 would be untenable for an EPSEM design. To adjust for the loss of precision in such a case, reducing the cluster size is the most effective method (Agarwal, Awasthi, & Walter, 2005; Gulliford et al., 1999; Katz, 1995; Katz & Zeger, 1994). It should also be noted that for a higher value of ICC, the loss of precision of the estimate will be more.

    • A study protocol for a cluster randomized pragmatic trial for comparing strategies for implementing primary HPV testing for routine cervical cancer screening in a large health care system

      2023, Contemporary Clinical Trials
      Citation Excerpt :

      Both assumptions are estimated using internal data of prior cervical cancer screening practice change at KPSC. The design effect was estimated by simulating cluster-randomizations in these prior data as the ratio of squared standard errors of the between-arm differences with vs. without cluster-level random effects. [51] All power calculations were conducted in PASS 14. [52]

    • From Agadez to Zinder: Estimating coverage of the MenAfriVac™ conjugate vaccine against meningococcal serogroup A in Niger, September 2010 - January 2012

      2013, Vaccine
      Citation Excerpt :

      We estimated general and regional coverage by aggregating the cluster-samples from the districts, using the size of the household and the target district population as survey weights [21]. We calculated DEFF to estimate the loss of precision due to cluster-sampling [22]. We fitted multivariate Poisson regression models with robust variance to assess the association between vaccination status and variables of interest.

    • Evaluating Imaging and Computer-aided Detection and Diagnosis Devices at the FDA

      2012, Academic Radiology
      Citation Excerpt :

      The results of a study by Cole et al (59) are consistent with these biases. More recently, some progress has been made to understand enrichment bias (60), recognizing the relationship to survey sampling and verification bias (61–65). What has been learned is that the direction and magnitude of enrichment bias depends on the correlation between the measurement used to drive the enrichment (eg, screen-film mammographic interpretations at enrollment) and the ROC ratings collected from the imaging devices being compared (eg, screen-film and full-field digital mammographic interpretations in the reader study) (60).

    • Intraclass correlation coefficient and outcome prevalence are associated in clustered binary data

      2005, Journal of Clinical Epidemiology
      Citation Excerpt :

      A relationship between ICC and prevalence was identified by Robertson and Lerner [25] in the context of studies of heritability. This association was exemplified by data presented by Feng and Grizzle [4] as well as by Mickey and Goodwin [9], Katz et al. [10,11], and Slymen and Hovell [12]. This relationship might be expected, because higher prevalence will be associated with greater variance, but at any given prevalence, assuming nonnegativity, there are few theoretical constraints on the value of an ICC.

    View all citing articles on Scopus

    This work was prepared under Cooperative Agreement DAN-0045 between the Office of Nutrition of the United States Agency for International Development (USAID) and The International Center for Epidemiologic and Preventive Ophthalmology (ICEPO), and was supported by National Institutes of Health grants S 10-RRO4060 and A125529 and Merck Biostatistics Department program grant. The surveys from which the data in this report were obtained were collaborative projects between ICEPO and the national partners in each country, funded by the Office of Nutrition, USAID (except where noted).

    Malawi: The Ministry of Health, Helen Keller International (HKI), and the International Eye Foundation.

    Zambia: The National Food and Nutrition Commission, Tropical Disease Research Centre, the Flying Doctor Service, and the Ministry of Health (funded by the International Development Research Center/Canada).

    Indonesia: The Directorate of Nutrition, Department of Health, and Helen Keller International.

    Nepal: The National Society for the Prevention of Blindness (Nepal Netra Jyoti Sangh).

    The authors wish to acknowledge the computing assistance of Vincent Carey.

    View full text