Comparación de Tres Métodos de Análisis de Variables Binarias Comparación de Tres Métodos de Análisis de Variables Binarias

This study compares lineal models (ML), contingency tables using Chisquare (TC) and logistic regression (RL) on their efficiency to test hypothesis about differences of percentages between two treatments. Efficiency was defined as the probability of not commit error type I. A series of simulated cas...

Full description

Bibliographic Details
Main Author: Hugo H. Montaldo
Format: Article
Language:Spanish
Published: Universidad de Guanajuato 2012-02-01
Series:Acta Universitaria
Subjects:
Online Access:http://www.actauniversitaria.ugto.mx/index.php/acta/article/view/290
Description
Summary:This study compares lineal models (ML), contingency tables using Chisquare (TC) and logistic regression (RL) on their efficiency to test hypothesis about differences of percentages between two treatments. Efficiency was defined as the probability of not commit error type I. A series of simulated cases with 4 and 25 repetitions were analyzed. Considering the lower efficiency of ML, particularly with few replicates and the limitations of TC for the analysis of the binary variables, the results of this preliminary study show that in agreement with the theoretical expectations, the RL method, readily available in several tested programs of statistical software, should be prefered for the analysis of experiments with binary dependent variables.<br><span style="font-family: Times New Roman; font-size: small;"> </span><p class="MsoNormal" style="margin: 0cm 0cm 0pt; line-height: normal; mso-layout-grid-align: none;"><span style="font-family: "AGaramond-Regular","serif"; font-size: 9pt; mso-bidi-font-family: AGaramond-Regular;"><span style="font-family: Times New Roman;">El presente trabajo compara los modelos lineales (ML), las tablas de contingencia con Chi-cuadrado (TC) y la regresión logística (RL) en cuanto a su eficiencia para probar hipótesis de diferencias entre porcentajes de dos tratamientos. Se definió eficiencia como la probabilidad de no cometer error tipo I. Se analizó una serie de casos simulados con 4 y 25 repeticiones. En conclusión, considerando la menor eficiencia de los ML, particularmente con pocas repeticiones y las limitaciones de las TC para el análisis de variables binarias, los resultados de este estudio preliminar ilustran las expectativas teóricas de que el método de RL, actualmente disponible en varios programas evaluados de software estadístico, debería ser preferido para el análisis de experimentos con variables dependientes binarias.</span></span></p><span style="font-family: Times New Roman; font-size: small;"> </span>
ISSN:0188-6266