Robustez de los índices de ajuste del análisis factorial confirmatorio a los valores extremos.
Fecha
2020
Tipo
artículo original
Autores
Rojas Torres, Luis
Título de la revista
ISSN de la revista
Título del volumen
Editor
Resumen
El objetivo de este trabajo es evaluar la robustez de varios índices de ajuste del Análisis Factorial Confirmatorio (SRMR, RMSEA, TLI, CFI y GFI) a la presencia de valores extremos. Para el alcance del objetivo se planteó un estudio de simulaciones con 3× 4 × 2 condiciones: tamaño de muestra (100, 200 y 500), porcentaje de valores extremos (0%, 1%, 5% y 10%) y número de variables con valores extremos (1 y 2). Los conjuntos de datos base (0% de valores extremos) por tamaño de muestra, fueron simulados a partir de una distribución que se ajustaba a un AFC con tres factores correlacionados. Los conjuntos de datos con valores extremos se crearon a partir de la sustitución de observaciones en los conjuntos de datos base. Luego, en cada conjunto de datos se estimó un AFC con tres factores correlacionados. Se obtuvo que con los puntos de corte clásicos de los índices de ajuste, todos estos fueron robustos a la presencia de datos extremos en conjuntos de datos de 200 y 500 observaciones. Con 100 observaciones se obtienen índices robustos pero considerando puntos de corte ajustados a la estructura factorial y el tamaño muestral trabajado.
The goal of this work is to evaluate the robustness of several Confirmatory Factor Analysis fit indices (SRMR, RMSEA, TLI, CFI and GFI) to the precense of outliers. For this purpose, it was planed a simulation study with 3 × 4 × 2 conditions: sample size (100, 200 and 500), outliers percentage (0%, 1%, 5% and 10%) and number of variables with outliers (1 and 2). The baseline data sets (0% of outliers) by sample size were simulated from a distribution which fit to a CFA with three factors correlated. Data bases with outliers were created from substitution of observations in baseline data sets. Later, in every data base was estimated a CFA with three factors correlated. It was obtained that all indices with classical cutoffs were robust to outliers with sample sizes of 200 and 500. With 100 observations, it was obtained that fit indexes were robust to outliers, but considering cutoffs adjusted by the factorial structure and the sample size.
The goal of this work is to evaluate the robustness of several Confirmatory Factor Analysis fit indices (SRMR, RMSEA, TLI, CFI and GFI) to the precense of outliers. For this purpose, it was planed a simulation study with 3 × 4 × 2 conditions: sample size (100, 200 and 500), outliers percentage (0%, 1%, 5% and 10%) and number of variables with outliers (1 and 2). The baseline data sets (0% of outliers) by sample size were simulated from a distribution which fit to a CFA with three factors correlated. Data bases with outliers were created from substitution of observations in baseline data sets. Later, in every data base was estimated a CFA with three factors correlated. It was obtained that all indices with classical cutoffs were robust to outliers with sample sizes of 200 and 500. With 100 observations, it was obtained that fit indexes were robust to outliers, but considering cutoffs adjusted by the factorial structure and the sample size.
Descripción
Palabras clave
Valores extremos, Análisis factorial confirmatorio, Robustez, Estudio de simulación, Error Monte Carlo, Outliers, Confirmatory factor analysis, Robustness, Simulations study, Monte Carlo error