Reglas de detección del funcionamiento diferencial del item. Estudio del efecto del tamaño de muestra en presencia de dif paralelo. = The differential item functioning testing rules. The effect of the sample size presence of parallel dif.
Tipo de material:
- A.01 J82j Raa
- ABSTRACT This study analyzes the potential effect of the sample size on the differential Item functioning (DIF) detection in presence of Parallel DIF. The false positives and correct identifications rates of two rules that combine the Breslow-Day tests, global (BD) and of trend (BDT), with the standard Mantel-Haenszel (MH) procedure, without the Bonferroni’s adjustement were analyzed. Comparisons were made with the rule proposed by Randall Penfield and the modified Mantel-Haenszel procedure. The responses to 75-items tests were simulated without impact with 100 repetitions. Reference and Focal groups were of equal size, 1,000 or 200. The presence of Parallel DIF was generated in 20 items with a marked increase of the difficulty parameter in the Focal group. The MH procedure was strongly affected by the sample size, with very inflated false positives rates for large samples. This effect moved to the four procedures. For both sample sizes, the rule based on BDT without the Bonferroni’s adjustement was the most powerful and the Penfield’s rule presented the false positives rates less inflated.
RESUMEN
En el presente estudio se analiza la posible incidencia del tamaño
de muestra en la detección del funcionamiento diferencial del ítem
(Differential Item Functioning, DIF) frente al DIF Paralelo. Se analizó la tasa de falsos positivos e identificaciones correctas de dos
reglas que combinan a las pruebas de Breslow-Day, global (BD) y
de la tendencia (BDT), con el procedimiento estándar de MantelHaenszel (MH), sin el ajuste de Bonferroni. Se efectuaron comparaciones con la regla propuesta por Randall Penfield y con el procedimiento de Mantel-Haenszel modificado. Las respuestas a tests
de 75 ítems fueron simuladas, sin impacto, con 100 repeticiones.
Los grupos de Referencia y Focal se tomaron de igual tamaño,
1,000 o 200. La presencia de DIF Paralelo se generó en 20 ítems
con un marcado incremento del parámetro de dificultad en el grupo
Focal. El procedimiento MH resultó muy afectado por el tamaño de
muestra, con tasas de falsos positivos muy infladas para las muestras grandes. Efecto que se trasladó a los cuatro procedimientos.
Para ambos tamaños muestrales, la regla basada en BDT sin el
ajuste de Bonferroni resultó la más potente y la regla de Penfield
presentó la tasa de falsos positivos menos inflada.
ABSTRACT
This study analyzes the potential effect of the sample size on the
differential Item functioning (DIF) detection in presence of Parallel
DIF. The false positives and correct identifications rates of two
rules that combine the Breslow-Day tests, global (BD) and of trend
(BDT), with the standard Mantel-Haenszel (MH) procedure, without the Bonferroni’s adjustement were analyzed. Comparisons
were made with the rule proposed by Randall Penfield and the
modified Mantel-Haenszel procedure. The responses to 75-items
tests were simulated without impact with 100 repetitions. Reference and Focal groups were of equal size, 1,000 or 200. The presence of Parallel DIF was generated in 20 items with a marked increase of the difficulty parameter in the Focal group. The MH procedure was strongly affected by the sample size, with very inflated
false positives rates for large samples. This effect moved to the
four procedures. For both sample sizes, the rule based on BDT
without the Bonferroni’s adjustement was the most powerful and
the Penfield’s rule presented the false positives rates less inflated.