Aportes metodológicos en la estimación de tamaños de muestra en estudios poblacionales de prevalencia

Abstract

This dissertation addresses the application of six statistical approaches used to estimate sample sizes in multinomial populations which correspond to: Angers (1974), Tortora (1978), Thompson (1987), Cochran (1977), Bromaghin (1993) and Fitzpatrick & Scott (1987), such approaches are widely discussed in the literature of statistical sampling but generated controversy when applying in health studies because they do not always allow combining costs, representation and adequate sample sizes for sampling scheme simple random sampling and complex populations where the design variable or study corresponds to a multinomial distribution type. Initially discusses how the use of a maximun variance when the design variable with k = 2 gives estimates of prevalence categories considering a P = 0. 50 for this sample size estimate without knowing previous values of this estimator which delivers biased estimates. Later theoretical populations were simulated for variables k = 3, 4, 5, 6 and 7 categories, generating 25 different populations of size N = 1,000,000 varying proportions according to different values for different categories. For these populations were extracted by simple random sampling, samples of different sizes were estimated using the six approaches mentioned above that considered different values of sampling errors, then the performance of these was assessed by: 1) sample size, 2) Level of real confidence, 3) average Estimator, 4) bias and 5) mean Square Error. The discussion then focuses on determining which delivery method best used sample sizes and estimates considering scenarios where the categories considered ranging from k = 3 to k = 7, finally proposes and discusses the use of measures of uncertainty or entropy Shannon to study the uncertainty associated with the estimated vectors using different methods.

Date of Award	17 Sept 2014
Original language	Spanish
Supervisor	Albert Navarro Gine (Director)