Angoff's delta method revisited: improving DIF detection under small samples.

摘要: Most methods for detecting differential item functioning (DIF) are suitable when the sample sizes sufficiently large to validate null statistical distributions. There is no guarantee, however, that they will still perform adequately there few respondents in focal group or both reference and group. Angoff's delta plot a potentially useful alternative small-sample DIF investigation, but it suffers from an improper flagging criterion. The purpose of this paper improve classification rule under mild assumptions. This improvement yields modified with adjusted criterion small samples. A simulation study was conducted compare classical approach Mantel–Haenszel method. It concluded consistently less conservative more powerful than usual plot, also method as long at least one small.