用Excel畫出來(1)log-likelihood,(2)1st derivative,(3)2nd derivative等三條曲線.
check表7.1(p.165)之結果相同.
以上為第7章的form a的前5題答對,後5題答錯的情形(a=1.5; b分別為-2,-1.5,-1,-0.5,0,0,0.5,1,1.5,2)
http://tnc.k12.edu.tw/1002015816/manuscripts/rasch/Parallel%20Forms_3PL.xls

C7儲存格(如上列公式)
(1)產生200 persons各theta, (2)再產生各題的step 困難度: (3)RSM資料產生方式如上式,1-5的選項產生,第6列為各題的thresholds的難度ranges(各題相同),(4)如果是PCM則由亂數產生器直接產生即可,RSM則予調整至range相同(用c2=c5-c6即可違成).
除此之外,資料產生另以期望值做為來源,即a1-a6間與excel的rand()比較

利用Rasch分析的二點計分模式,研究答題反應的各種變化與特性,因此需要建立一套答題反應,依左方過程,可以完成fit資料的結構.
本資料只有第9題infit 及outfit MNSQ為1.23, 1.46, infit 及outfit STND分別為2.5及 3.0, 因此為快速產生資料模擬分析的方式...

本研究利用多向度Rasch模式分析健保每週在網路上揭露的346家醫院門急診與住院成長率的資料，探討：(1) 門急住診間的構面相關係數為何？(2)測量信度可提昇多少？(3) 各醫院層級與健保分局轄區間的成長力之比較為何？(4) 哪家醫院的成長力異常需予關注？研究發現，經刪除不符模式預期的門住診3週及急診7週的資料，其三構面的信度及相關係數分別界於(0.905 – 0.961)及(0.168 – 0.990 )，高於單向度測量的(0.84 – 0.85)及(0.166 – 0.98 )。健保分局除東區外，皆呈顯著性負成長。只有台北區、北區、中區分局、醫學中心、區域醫院及地區醫院，在門急住診三構面皆呈統計顯著性（p <.05）負成長。有幾家醫院的成長情形跟大多數醫院比較起來，相當異常，值得進亦步探討。此外，我們也詳細將醫院按地區和層級分類，探討不同地區和層級在門急住診的各季難度變化。

Many contests were annually held in hospitals, using summed scores directly to rank performances of examinees. In contrast to classical test theory, Rasch(1960) analysis with the software Facets(Linacre,2006) was implemented in this study, detecting judges fair or not in a contest selected the best nurse head, to examine whether items or judges meet the requirement of unidimensionality and then to estimate the abilities of examinees after removing unexpected items and unfair judges.
An annual contest selected the best nurse head was held at an academic medical center in spring 2006. Four nurse superintendents as judges appraised 10 candidates with force ranking scores, coding 9 for the most excellent and 0 the poorest, across 7 indicators. Five approaches were proceeded to (1) examine data fitting the Rasch model, (2) evaluate impacts while removing unexpected data, (3) distinguish changes before and after using the fitted data, (4) detect underlying latent characteristics of the awarded nurse head, and (5)summary results of this achievement contest.
The results showed (1) the item 7(whether fully execute superintendents’ occasional tasks and commands) and the judge 3 should be removed from the assessment lists due to misfitting Rasch model’s expectation. (2) the most excellent performance were shown to candidate 5 and 10 in management and education respectively, poorest to candidate 4 in communication and coordination, and data distortion scoring onto judge 3. (3)Leadership and education, statistically significantly differentiated by linear regression, were suggested to be much more emphasized for would-be candidates pursuing the title of the best nurse head in the next year.

"Objective Measurement in the Social Sciences" stream of the 2006 ACSPRI Conference, Dec. 11-13, Mon.-Wed., Sydney, Australia.
Convenor:Andrew Stephanou, Australian Council for Educational Research
http://www.acspri.org.au/conference2006//conf_topic.php?id=5