Confidence Interval Methods for Coefficient Alpha on the Basis of Discrete, Ordinal Response Items: Which One, if Any, is the Best?

Estimación paramétrica de la confiabilidad y diferencias confiables

“…Cabe mencionar que se espera que el límite inferior del IC para el coeficiente a sea mayor que 0,70 3 . Mediante el programa ICalfa 5 se realizaron los cálculos de los IC (nivel de confianza de 95%) para cada a bajo el método Fisher 6 , obteniéndose lo siguiente: para Salud Física: 0,688-0,731; para Salud Psicológica: 0,730-0,768; para la dimensión Social: 0,603-0,656; y para la dimensión Ambiental: 0,741-0,778. Estos resultados indican que las dimensiones Salud Psicológica y Ambiental mantienen un nivel de confiabilidad poblacionalmente superior a 0,70, mientras que la dimensión Social puede alcanzar niveles debajo de 0,70, ya que tiene todo su IC debajo de 0,70.…”

Performance Evaluation of Confidence Intervals for Ordinal Coefficient Alpha

“…While several simulation studies have analyzed the performance of various methods for confidence intervals for coefficient alpha, (Cui & Li, 2012;Iacobucci & Duhachek, 2003;Padilla et al, 2012;Romano et al, 2011), no known published studies have analyzed confidence intervals for ordinal alpha. Historically, highly constrained assumptions have hindered researchers from calculating precise confidence intervals for coefficient alpha, much less ordinal alpha, up until recently (Duhachek & Iacobucci, 2004).…”

“…Confidence intervals may be calculated on most any statistic, including means, effect sizes (Thompson, 2002(Thompson, , 2006a, and reliability coefficients (Cumming & Fidler, 2009;Fan & Thompson, 2001;Romano et al, 2011). In order to compute a confidence interval and conduct statistical inference, information about the standard error and sampling distribution of a statistic are necessary.…”

“…The second objective is to examine the precision of ordinal coefficient alpha in terms of root mean square error (RMSE) and bias. There are several known factors that affect confidence intervals for reliability estimates including sample size, the number of items on an instrument, skewness of the responses, and correlation amongst the items (Cui & Li, 2012;Duhachek & Iacobucci, 2004;Romano, Kromrey, Owens, & Scott, 2011). The following research questions will guide this study: a) How do the number of response categories, sample size, number of items, and skewness impact the precision of coverage rates, interval width, and variance of interval width with respect to varied population parameters of ordinal coefficient alpha?…”

“…The transformed statistic can be appropriately computed within a normal theory framework for a confidence interval and transformed back into the original units (Romano et al, 2011 …”