An empirical comparison of methods for analyzing over-dispersed zero-inflated count data from stratified cluster randomized trials

Are the greatest professors those who publish the most? Aspects related to scientific research production of professors from Medicine programs at a Peruvian university during the Covid-19 pandemic

“…Although Are the greatest professors those who publish the most? Aspects related to... there are different solutions to this problem (see, for different fields, BORHAN et al;CHAI;BAILEY;YANG;SIMPSON;2010), a strategy similar to the one used by Fletcher et al (2005) was applied; however, instead of ordinary regression, gamma regression was used. This happened due to issues related to normality, data dispersion and by dealing with natural numbers versus decimals.…”

¿Los mejores profesores son los que publican más? Factores asociados a la producción científica en docentes del grado en medicina en una universidad peruana durante a pandemia de la Covid-19

“…Luego de una primera exploración de los datos, la gran cantidad de docentes sin publicaciones sesgó la muestra por datos con valor de cero (o zero-skewed data en la literatura anglosajona). Aunque existen soluciones diversas a este problema (véase, en para disciplinas distintas, BORHAN et al;CHAI;BAILEY;YANG;SIMPSON; 4-Los datos usados en la investigación pueden consultarse en Figshare: https://figshare.com/s/3a68b879d0e5394dded4 2010), se siguió una estrategia parecida a la utilizada por Fletcher et al (2005); no obstante, en vez de una regresión ordinaria se utilizó una regresión gamma. Esto sucedió debido a asuntos relacionados con la normalidad, dispersión de datos y el trabajo con números naturales frente a los decimales.…”

Performance of methods for analyzing continuous data from stratified cluster randomized trials – A simulation study

“…Moreover, a majority of the literature suggested that the stratification variable(s) used in the randomization should be adjusted for in the analysis [ 2 , 8 , [11] , [12] , [13] , [14] ], while a few studies argued in favour of non-adjustment [ 15 ]. Studies based on both stratified RCTs on individuals and stratified CRTs indicated that the failure to adjust for stratification leads to wider confidence intervals, larger p-values and a reduction in power [ [16] , [17] , [18] ]. Thus, ignoring the adjustment for both clustering and stratification may yield a misleading conclusion about the treatment effect, and society may miss out on the benefit of a treatment.…”

“…The authors recommended small sample corrections - degrees of freedom correction for mixed-effects method and standard error (SE) correction for GEE, for analyzing continuous data from CRTs with the small number of clusters, while van Breukelen and Candel [ 28 ] provided a mathematical explanation of their findings. Borhan et al [ 17 , 18 ] empirically compared several methods for analyzing continuous and count data from stratified CRTs using the data from the Mallick et al [ 5 ] and the ViD OS study [ 29 ], respectively. In these studies, the authors found that the overall conclusions in terms of statistical significance were similar for all methods.…”